John A. Thorpe Elementary Topics in Differential Geometry Springer-Verlag New York Heidelberg BerlinJ. A. Thorpe Department of Mathematics State University of New York Stony Brook, New York 11794 USA Editorial Board P. R. Halmos Managing Editor Indiana University Department of Mathematics Bloomington, Indiana 47401 USA F. W. Gehring University of Michigan Department of Mathematics Ann Arbor, Michigan 48104 USA WTAE AMS Subject Classification: 53-01 With 126 Figures Library of Congress Cataloging in Publication Data Thorpe, John A Elementary topics in differential geometry. (Undergraduate texts in mathematics) Bibliography: p. Includes index. 1. Geometry, Differential. I. Title. QA641.T36 516'.36 78-23308 All rights reserved. No part of this book may be translated or reproduced in any form without written permission from Springer-Verlag. © 1979 by Springer-Verlag New York Inc. Printed in the United States of America. 987654321 ISBN 0-387-90357-7 Springer-Verlag New York ISBN 3-540-90357-7 Springer-Verlag Berlin HeidelbergTo my parents whose love, support, and encouragement over the years have to a large extent made the writing of this book possible.Preface In the past decade there has been a significant change in the freshman/ sophomore mathematics curriculum as taught at many, if not most, of our colleges. This has been brought about by the introduction of linear algebra into the curriculum at the sophomore level. The advantages of using linear algebra both in ‘the teaching of differential equations and in the teaching of multivariate calculus are by now. widely recognized. Several textbooks adopting this point of view are now available and have been widely adopted. Students completing the sophomore year now have a fair preliminary under- standing of spaces of many dimensions. It should be apparent that courses on the junior level should draw upon and reinforce the concepts and skills learned during the previous year. Unfortunately, in differential geometry. at least, this is usually not the case. Textbooks directed to students at this level generally restrict attention to 2-dimensional ‘surfaces in 3-space rather than to. surfaces of arbitrary dimension. Although most of the recent books do use linear algebra, it is only the algebra of R*. The student’s preliminary understanding of higher dimensions is not cultivated, This book develops the geometry of n-dimensional surfaces in (n + 1)- space. It is designed for a 1-semester differential geometry course at the junior-senior level. It draws significantly on the contemporary student's knowledge of linear algebra, multivariate calculus, and differential equations, thereby solidifying the student’s understanding of these subjects, Indeed, one of the reasons that a course in differential geometry is so valuable at this level is that it does turn out students with a thorough understanding of several variable calculus. Another reason that differential geometry regularly attracts students is that it contains ideas which are not only beautiful in themselves but are viiViii Preface basic for both advanced mathematics and theoretical physics. It has been the author’s experience that students taking his course have been more or less evenly divided between mathematics and physics majors. The approach adopted in this book, describing surfaces as solution sets of equations, seems to be especially attractive to physicists. The book considers from the outset the geometry of orientable hyper- surfaces in R"*', exhibited as inverse images of regular values of smooth functions. By considering only such hypersurfaces for the first half of the book, it is possible to move rapidly into interesting global geometry without getting hung up on the development of sophisticated machinery. Thus, for example, charts (coordinate patches) are not introduced until after the initial discussions of geodesics, parallelism, curvature, and convexity. When charts are introduced, it is as a tool for computation. However, they then lead the development naturally into the study of focal points and surfaces of arbitrary codimension. One of the advantages of treating the geometry of n-dimensions from the outset is that one can then illustrate each concept simultaneously in each of the low dimensions. Thus, for example, the student’s understanding of the Gauss map and its (spherical) image is aided by the possibility of studying 1-dimensional examples, where the spherical image is a subset of the unit circle. : The main tool used in developing the theory is that of the calculus of vector fields. This seems to be the most natural tool for studying differential geometry as well as the one most familiar to undergraduate students of mathematics and physics. Differential forms are not introduced until fairly late in the book, and then only as needed for use in integration. Students who have completed a good 2-year calculus sequence including linear algebra and differential equations should be adequately prepared to study this book. There are occasional places (e.g., in Chapter 13 on convexity) where some exposure to the ideas of mathematical analysis would be helpful, but not essential. There is probably more material here than can be covered comfortably in one semester except by students with unusually strong backgrounds. Chapters 1-12, 14, 15, 22, and 23 contain the core of basic material which should be covered in every course. Most instructors will probably also want to cover at least parts of Chapters 17, 19, and 24.Preface ix The interdependence of the chapters is as follows: A few concepts in the early part of Chapter 13 are used in later chapters but these may be studied, by those skipping Chapter 13, as needed. Like the author of any textbook, I owe a considerable debt to researchers and textbook writers who have preceded me and to teachers, colleagues, and students who have influenced me. While I cannot explicitly acknowledge all these, I must at least credit M. do Carmo and E. Lima whose paper, Isometric immersions with semi-definite second quadratic forms, Arch. Math. 20 (1969) 173-175, inspired the treatment of convex surfaces in Chapter 13, and S. S. Chern whose paper, A simple intrinsic proof of the Gauss-Bonnet formula for closed Riemannian manifolds, Ann. of Math. (2) 45 (1944) 747-752, inspired the treatment of the Gauss-Bonnet theorem in Chapter 21. In addition, special thanks are due to Wolfgang Meyer whose comments on the manuscript have been extremely helpful. Stony Brook, New York JOHN A. THORPE November, 1978Chapter 1 Graphs and Level Sets Chapter 2 Vector Fields Fhapter 3 The Tangent Space Fhapter 4 Surfaces Chapter 5 - ! . Vector Fields on Surfaces; Orientation Chapter 6 The Gauss Map pter 7 Geodesics Chapter 8 Parallel Transport Contents 13 23 31 38 45 xixii Chapter 9 rhe Weingarten Map Chapter 10 Curvature of Plane Curves Chapter 11 Pre Length and Line Integrals Chapter 12 Curvature of Surfaces Chapter 13 Convex Surfaces Chapter 14 / Parametrized Surfaces Chapter 15 7 rocal Equivalence of Surfaces and Parametrized Surfaces Chapter 16 Focal Points Chapter 17 Surface Area and Volume Chapter 18 _/ Minimal Surfaces Chapter 19 The Exponential Map Chapter 20 Surfaces with Boundary Chapter 21 The Gauss-Bonnet Theorem Chapter 22 Rigid Motions and Congruence Contents 53 62 68 82 95 108 121 132 139 156 163 177 210Contents Chapter 23 Isometries Chapter 24 Riemannian Metrics Bibliography Notational Index Subject Index xiii 220 231 245 247 249Associated with each real valued function of several real variables is a collec- tion of sets, called level sets, which are useful in‘studying qualitative proper- ties of the function. Given a function f: U + R, where U < R"*”, its level sets are the sets f~*(c) defined, for each real number c, by SMC) = (re ++ Xn1) © US ty «+5 Xnv1) = Oe The number cis called the height of the level set, and f ~ ‘(c) is called the level set at height c. Since f~*(c) is the solution set of the equation S(%1,.--. Xe) =e, the level set f~1{c)-is often described as “the set SL (Xt, +25 Xn) =O” The “level set” and “height” terminologies arise from the relation be- tween the level sets of a function and its graph. The graph of a function f: UR is the subset of R"*? defined by graph(f) = {(xts-r-2%eva) ERM? (ys oy Hert) EU and yer =S Ep <0 Sad For ¢ > 0, the level’set of f at height c is just the set of all points in the domain of f over which the graph is at distance ¢ (see Figure 1.1). For c < 0, the level set of fat height cis just the set of alf points in the domain of funder which the graph lies at distance ~c. For example, the level sets f~*(c) of the function f(x, .... Xn+4) =i + +++ 4+ x2, are empty for 'c <0, consist of a’single point ‘(the origin) if c = 0, and for c >"8 consist of two points if n= 0, circles centered at the origin with radius \/c ifn = 1, spheres centered at the origin with radius //c ifn = 2, etc (see Figures 1.1 and 1.2). %2 1 Graphs and Level Sets graphof yy Kx f(xy, Xp) = xp + x3 graph of A P(x) = xy XL S1(c) bx} = el f1 (0) [xp + x} =e] @:n=0 (b):n=1 Figure 1.1 The level sets f~*(c) (c > 0) for the function S (ty oes Xmas) SXF Ht XP -1 To £2) £13) a (3). 7 “1 7Q) a1 3 £71) 7a) sta) fa) t*O) aa [9 | weee ot -2 -1 0 1 2 2X1 (a):n=0 (b):n =1 (:n=2 Figure 1.2 Level sets for the function f (x1, ..., Xn+1) = x3 +1 + x344. For n = 1, level sets are (at least for non-constant differentiable functions) generally curves in R?. These curves play the same roles as contour lines on a topographic map. If we think of the graph of fas a land, with local maxima representing mountain peaks and local minima representing valley bottoms, then we can construct a topographic map of this land by projecting orthog- onally onto R. Then all points on any given level curve f~*(c) correspond to points on the land which are at exactly height c above “sea level” (x3 = 0). Just as contour maps provide an accurate picture of the topography of a land, so does a knowledge of the level sets and their heights accurately portray the graph of a function. For functions f: R? R, study of the level curves can facilitate the sketching of the graph of f. For functions f: R° > R,1 Graphs and Level Sets 3 the graph lies in R‘, prohibiting sketches and leaving the level sets as the best tools for studying the behavior of the function. One way of visualizing the graph of a function f: U + R, U c R?, given its level sets, is as follows. Think of a plane, parallel to the (x,, x2)}-plane, moving vertically. When it reaches height c this plane, x, = c, cuts the graph of f in the translate to this plane of the level set f~ *(c). As the plane moves, these sets generate the graph of f (see Figure 1.3). @) Figure 1.3 Level sets and graphs of functions f: R? > R. The label on each level set indicates its height. (a}f(x1,-x2) = —x7 +x}. (b) A function with two local minima. The same principle can be used to help visualize level sets of functions f:U>R, where Uc R®. Each plane x; = constant will cut the level set $7*(c) (c fixed) in some subset, usually a curve. Letting the plane move, by changing the selected value of the x,-coordinate, these subsets will generate the level set f~1(c) (see Figure 1.4). v1 Graphs and Level Sets x=0 1 7 4 1 y2yy2 ad Xp = XQ FART y= xy=-1 fx =0 xg ex3-4 Yxsexd-1 [ead (b) Figure 1.4 Level sets in R*, as generated by intersections with the planes x4 = constant. (a) x? + x3 + x3 = 1. (b) x? — x3 +. x3 = 0. EXERCISES In Exercises 1.1-1.4 sketch typical level curves and the graph of each function. LA. f(x, x2) = x.1 Graphs and Level Sets 5 12. f(%1 X2) = x1 — 2. 13. fp, x2) = xf — ad, . 1.4. f(x, x2) = 3r® — 8r6 + 6r* where r? = x} + x3. [Hint: Find and identify the critical points of f as a function of r.] In exercises 1.5~1.9 sketch the level sets f ~ *(c), for n = 0, 1, and 2, of each - function at the heights indicated. 15. 16. 17. 18. 1.9, S%1, X25 00 SL (Xa 25 0 S61, X25 00 Sy Xa, S(%n X25 0 > Xn+1) = Xnv13 = 1,0, 1,2. > Xavi) = Oxt + xd te txts c=0, 1,4 + Xing) =X — XE — xe = — 1,0, 1,2. + Xanga) = xP —x¥— 0 — Xe -1,0, 1. > Xne1) = xP + x9/4 +0 + x2, l(nt 1; c= 1. 1.10. Show that the graph of Mf any function f: R" > R is a level set for some function FRI +R.Vector Fields The tool which will allow us to study the geometry of level sets is the calculus of vector fields. In this chapter we develop some of the basic ideas. A vector at a point p € R"*" is a pair v = (p, v) where v € R"**, Geomet- tically, think of v as the vector v translated so that its tail is at p rather than at the origin (Figure 2.1). The vectors at p form a vector space R’,*? of 2 v (pa, ¥) Figure 2.1 A vector at p. dimension n+ 1, with addition defined by (p, v) + (p, w)=(p, v+ w) (Figure 2.2) and scalar multiplication by c(p, v)=(p, cv). The set {(p, 01), «+5 (P, tn 1)} is a basis for R%*! where {v1, ..., U,+1} is any basis for R"*1, The set of all vectors at all points of R"** can be identified (as a set) with the Cartesian product R"*! x R"*+! = R?"*?, However note that our tule of addition does not permit the addition of vectors at different points of R'!, Given two vectors (p, v) and (p, w) at p, their dot product is defined, using the standard dot product on R"*?, by (p, v) - (p, w) = v + w. When (p, v) and 62 Vector Fields Figure 2.2 Addition of vectors at p. (p, w) € R3, pe R%, the cross product is also defined, using the standard cross product on R?, by (p, 2) x (Pw) = (p, » * w). Using the ‘dot product, the ‘length |/v| of a vector v = (p, v) at p.and the angle 6 betwee two Wectote SOs a and w = (p, w) are defined by Sly = yt? cos =v: w/|yi wl O<00 such that qeU whenever ja — pll R"*!, where J is some open interval in R. By smoothness of such a function is meant that a is of the form a(t) = (x,(t), ..., Xn+1(t)) where each x; is a smooth real valued function on J. The velocity vector at time t (t € 1) of the parametrized curve «: J > R"** is the vector at a(t) defined by w= (a0, Fo) = (x St... 240), This vector is tangent to the curve a at a(t) (see Figure 2.4.). If«(t) represents “ a(t) 1 &(t) Figure 2.4 Velocity vector of a parametrized curve in R?. for each t the position at time t of a particle moving in R"*! then a(t) represents the velocity of this particle at time t. A parametrized curve a: I + R"*? is said to be an integral curve of the vector field X on the open set U in R"*? if a(t)e U and a(t) = X(«(c)) for all te J. Thus « has the property that its velocity vector at each point of the curve coincides with the value of the vector field at that point (see Figure 2.5). Theorem. Let X be a smooth vector field on an open set U < R"*! and let pe U. Then there exists an open interval I containing 0 and an integral curve a: I> U of X such that (i) (0) = p.2 Vector Fields . 9 Figure 2:5 An integral curve of a vector field. (ii) If B: TU is any other integral curve of X with B(0) = p, then I < I and B(t) = a(t) for all t eT. Remark. The integral curve « is called the maximal integral curve of X through p, or simply the integral curve of X through p. Proor. This theorem is a reformulation of the fundamental existence and uniqueness theorem for solutions of systems of first order differential equa- tions. X, being a smooth vector field on U, has the form X(P) = (P, X10), ---s Xn+i(P)) where the X;: UR are smooth functions on U. A parametrized curve a: 1+ R"*! has the fofm a(t) = (10), --- Xnva(t)) where the x,: 1 + R are smooth functions on J. The velocity of « is ait) (ate, 84 Ries at) = (00, $0... Se} The requirement that « be an integral curve of X says that «(t) = X(a(t)), or 6 Ss (6) = Kile os Xeeal) () Ba (em Xe sllOh oes Mee lOh10 2 Vector Fields This is a system of n+ 1 first order ordinary differential equations in n+1 unknowns. By the existence theorem for the solutions of such equations,+ there exists an open interval J, about 0 and a set x;:1; > R of smooth functions satisfying this system subject to the initial conditions x(0) = p; for ie {1,...,n+ 1}, where p= (p,,..., Pa+1) Setting B,(t) = (x4(t), ---, X,+1(t)) for this choice of functions yields an integral curve Bi: 1,3 U of X with B,(0) = p. By the uniqueness theorem for the solutions of first order ordinary differ- ential equations,+ if X;:1,—R is another set of functions satisfying the system (E) together with the initial conditions %,(0) = p;,, then X,(t) = x,(t) for all t € I, 7 1. In other words, if 82: 1, > U is another integral curve of X with B,(0) = p then B,(t) = B,(t) for all te I, 7 Iz. It follows from this that there is a unique maximal integral curve « of X with «(0) =p (its domain is the union of the domains of all integral curves of X which map 0 to p) and that if 8: I U is any other integral curve of X with f(0) = p then B is simply the restriction of a to the smaller interval I. o ExampLe. Let X be the vector field X(p) = (p, X(p)) where X(x,, x2) = (—x2, x1) (Figure 2.3.(c)). A parametrized curve a(t) = (x,(¢), x2(¢)) is an integral curve of X if and only if the functions x,(t) and x,(t) satisfy the differential equations The general solution of this pair of equations is X,(t) = C, cos t+ C, sint X2(t) = Cy sin t— Cz cos t. Thus the integral curve of X through the point (1, 0) (with x,(0) = 1 and X2(0) = 0) is a(t) = (cos ¢, sin t), whereas the integral curve through an arbitrary point (a, b) (with x,(0) = a4 and x,(0) = b) is B(t) = (a cos t — b sin t, a sin t + b cos t) (see Figure 2.6). +See eg W. Hurewicz, Lectures on Ordinary Differential Equations, Cambridge, Mass.: MLT. Press (1958), p. 28.2 Vector Fields il 4X + _ integral curve through (1, 1) \ integral curve through (1, 0) Figure 2.6 Integral curves of the vector field X(x1, x2) = (1, X2, —X2, x1). EXERCISES 2.1. Sketch the following vector fields on R?: X(p) = (p, X(p)) where (a) X(p) = © 1) (4) X(x1, x2) = (2, ¥1) (b) X@)= ~p (©) X(%1 x2) = (—2x2, $41). 22. 2.3, 2.4. 25. 2.6. (©) X(%1, x2) = (x2,.->1) Find and sketch the gradient field of each of the following functions: (2) (1, x2) = x1 + x2 (b) fea, x2) = xf + 3 (©) f(a, x2) = 4 — 2d (4) f(a, x2) = (*E — x2)/4. The divergence of a smooth vector field X on Uc R"*}, X(p) = (P, X1(P), ---» Xnva()) for pe U, is the function div X: UR defined by div X = )72} (@X;/dx,). Find the divergence of each of the vector fields in Exercises 2.1 and 2.2. Explain why an integral curve of a vector field cannot cross itself as does the parametrized curve in Figure 2.4. Find the integral curve through p = (1, 1) of each of the vector fields in Exer- cise 2.1. Find the integral curve through p = (a, b) of each of the vector fields in Exer- cise 2.1. A smooth vector field X on an open set U of R"*" is said to be complete if for each p € U the maximal integral curve of X through p has domain equal to R. Determine which of the following vector fields are complete: (a) X(x1, x2) = (1 2, 1, 0) U = R? (b) X(x1, x2) = (x1, x2, 1, 0), U = R? — {(0, 0). (c) X(x1, x2) = (x1, x2, —X2, x1), U = R? — {(0, 0)}. (4) X(x1, 2) = (a, x2, 1 + x7, 0), U = RP.12 2.8. 29. 2.10. 211 2.12. 2 Vector Fields Let U be an open set in R"*?, let p € U, and let X be a smooth vector field on U. Let a: 1 U be the maximal integral curve of X through p. Show that if B: T+ U is any integral curve of X, with B(to) = p for some to € T, then f(t) = a(t — to) for all t € T. [Hint: Verify that if B is defined by B(e) = A(t + to) then B is an integral curve of X with B(0) = p] Let U be an open set in R**! and let X be a smooth vector field on U. Suppose a: IU is an integral curve of X with a(0) = a(to) for some to € I, to #0. Show that « is periodic; ie., show that a(t + to) = a(t) for all t such that both ¢ and t + to € I. [Hint: See Exercise 2.8.] Consider the vector field X(x1, x2) = (x1, x2, 1,0) on R?, For te R and p R?, let g,(p) = a(t) where a, is the maximal integral curve of X through p. (a) Show that, for each t, g, is a one to one transformation from R? onto itself. Geometrically, what does this transformation do? (b) Show that o = identity Puta =P ° Pn forallty,t2¢R 9-.=9;' forallteR. [Thus t+ 9, is a homomorphism from the additive group of real numbers into the group of one to one transformations of the plane.] Repeat Exercise 2.10 for the vector fields (a) X(x1, x2) = (x1 X25 —X25 1) (b) X(x1, x2) = (1 2, X15 X2) (©) X (x1, x2) = (X1» %2» X25 ¥1)- Let X be any smooth vector field on U, U open in R"*!. Let g,(p) = a(t) where a, is the maximal integral curve of X through p. Use the uniqueness of integral curves to show that 9,,(1,(P)) = @r,+:.(P) and @-,(p) = 9; *(p) for allt, ¢1, and tz for which all terms are defined. [¢, is called the local 1-parameter group associated to X.]The Tangent Space Let f: U - R be a smooth function, where U c R"*" is an open set, letce R be such that f~1(c) is non-empty, and let p ¢ f~*(c). A vector at p is said to be tangent to the level set f ~ *(c) if it is a velocity vector of a parametrized curve in R"*! whose image is contained in f~ '(c) (see Figure 3.1). v FO @:n=1 Figure 3.1 Tangent vectors to level sets. Lemma. The gradient of f at p € f ~*(c) is orthogonal to all vectors tangent to I> '(c) at p. Proor. Each vector tangent to f~‘(c) at p is of the form &(to) for some parametrized curve a: I> R"*! with a(to) = p and Image a < f~*(c), But Image a < f~ *(c) implies f(«(t)) = c for all t € I so, by the chain rule, Om FF aia) = Wilelto))" Ha) =O) Heo = 1314 3 The Tangent Space If Vf (p) = 0, this lemma says nothing. But if Vf (p) # 0, it says that the set of all vectors tangent to f~'(c) at p is contained in the n-dimensional vector subspace [Vf(p)}! of R%*! consisting of all vectors orthogonal to Vf (p). A point p € R"*? such that Vf(p) # 0 is called a regular point of f. Theorem. Let U be an open set in R"*! and let f: U > R be smooth. Let p ¢ U be a regular point of f, and let c = f(p). Then the set of all vectors tangent to 7 *(c) at p is equal to [Vf (p)]*. Proor. That every vector tangent to f~*(c) at p is contained in [Vf(p)]* was proven as the lemma above. Thus it suffices to show that, if v = (p, v) € [Vf (p)}, then v=4(0) for some parametrized curve a with Image a < f~*(c). To construct «, consider the constant vector field X on U defined by X(q) = (q, v). From X we can construct another vector field Y by sub- tracting from X the component of X along Vf: X(q)- Wf(@) =X(q) -— Gp . Y(q) = X(q) iw@le Vf (q): The vector field Y has domain the open subset of U where Vf # 0. Since p is a regular point of f, p is in the domain of Y. Moreover, since X(p) = v € [vf (p)}:, ¥(p) = X(p). Thus we have obtained a smooth vector field Y such that ¥(q) L Vf(q) for all q e domain (Y), and Y(p) = v. Now let a be an integral curve of Y through p. Then a(0)=p, &(0) = ¥(a(0)) = ¥(p) = X(p) = v and 4 fate) = Vo (a(t) + ele) = VE (a(e)) » Y(a(r)) = 0 chain since a is since Y 1 Vf tule an integral curve of Y for all t € domain a, so that f («(t)) = constant. Since f («(0)) = f (p) = c, this means that Image « < f~'(c), as required. oOo Thus we see that at each regular point p on a level set f~ *(c) of asmooth function there is a well defined tangent space consisting of all velocity vectors at p of all parametrized curves in f~*(c) passing through p, and this tangent space is precisely [Vf(p)]* (see Figure 3.2). EXERCISES 3.1. Sketch the level sets f~(—1), f71(0), and f~"(1) for f(xy, .-.5 Xee1) = x7 + +++ + x? — x2, 1; = 1, 2, Which points p of these level sets fail to have tangent spaces equal to [Vf(p)}!?3 The Tangent Space 15 tangent space at p tangent space vs (p) uP sc) to (a):n=1 (b); n =2 Figure 3.2 Tangent space at a typical point of the level set f~'(1), where 3.2. 3.3, 3.4, 3.5. 2 Xnea) = xP tt Xe LO. Show by example that (a) The set of vectors tangent at a point p of a level set need not in general be a vector subspace of R3*?. (b) The set of vectors tangent at a point p of a level set might be all of Rp**. Sketch the level set f~1(0) and typical values Vf(p) of the vector field Vf for pef 10), when (a) f(xi, x2) = x} + x9 —-1 (b) f (x1, x2) = xf - 33-1 (c) f(a, x2) = x3 — x3 (4) F (1, %2) = 1 — 23. Let f; U-+ R be a smooth function, where Uc R"*! is an open set, and let a: 1 U be a parametrized curve. Show that f « a is constant (i.e., the image of « is contained in a level. set of f) if and only if « is everywhere orthogonal to the gradient of f (ie., if and only if a(t) 1 Vf(a(t)) for all t € 1). Let f: U+ R be a smooth function and let a: I + U be an integral curve of Vf. (a) Show that (d/de)(f a)(t) = Vf (a(e))|]? for all ¢ € I. (b) Show that for each to € I, the function fis increasing faster along a at a(to) then along any other curve passing through a(to) with the same speed (ie., show that if B: 1+ U is such that (so) = a(t) for some spe I and 1B(so)l] = [le(¢o)I| then (d/de)( fo a)(to) = (d/de)(F B\to))-Surfaces A surface of dimension n, or n-surface, in R"** is a non-empty subset S of R"*! of the form S = f~‘(c) where f: U> R, U open in R"*?, is a smooth function with the property that Vf(p) # 0 for all p ¢ S. A 1-surface in R? is also called a plane curve. A 2-surface in R? is usually called simply a surface. An n-surface in R"*! is often called a hypersurface, especially when n > 2. By the theorem of the previous chapter, each n-surface S has at each point p Sa tangent space which is an n-dimensional vector subspace of the space R%** of all vectors at p. This tangent space will be denoted by S,. It is important to notice that this tangent space S, depends only on the set S and is independent of the function f which is used to define S. Indeed, S, is characterized as the set of all vectors at p which can be obtained as velocity vectors of parametrized curves in R"*! with images lying completely in S.Iff is any smooth function such that S = f~*(c) for some c € R and Vf(p) #0 for all p € S (by definition of n-surface, there must exist one such a function ; in fact there are many such functions for each n-surface S) then S, may also be described as [Vf (p)}. EXAMPLE 1. The unit n-sphere x? +++: + x2,, = 1 is the level set f~*(1) where f(x,,..., Xp+i) =x} + °° + x241 (Figure 3.2). It is an n-surface be- cause Vf (x1, +++) Xn+1) = (X15 +++) Xne1s 2X4 «++» 2Xn41) 18 Not Zero unless (X1, «+5 Xn+1) = (0, ..., 0) so in particular Vf(p) # 0 for pe f~ 1(1), [Warn- ing! Beware that for a vector (p, v) € R%,* ' to be zero it is only necessary that p= 05 thus (x4, ..., Xpeas 2X45 -++s 2Xn41) = O implies 2x, = +++ = 2x,41 = 0 $0 (x1, ..., X,+1) = 0.] When n = 1, the unit n-sphere is the unit circle. Exampce 2. For 0#(a;,...,d)+1)€R"*’ and beR, the n-plane 4X1, +77 + dy41Xp41 = is the level set f~'(b) where f (x1, ..., Xn+1) = 164 Surfaces 17 G,X, +7°* + yy Xn41- It is an n-surface for each b € R since VS (X15 +29 Xn) = Xs vey Met ts By o+y Oats) is never zero. A 1-plane is usually called a line in R?, a 2-plane is usually called simply a plane in R3, and an n-plane for n > 2 is sometimes called a hyperplane in R"*'. Two different values of b with the same value of (a, ..., +1) define parallel n-planes (see Figure 4.1). Vs) = (Pp, - 1,- 2) ve (p)= (2, - 1, - 2, - 3) (b):n=2 Figure 4.1 Parallel n-planes f~*(b), b= —2, —1, 0, 1, where S (1s +001 Sut) = = a = 2% — 3x3 0 = (N+ Dee EXAMPLE 3. Let f: U— R be a smooth function on U, U open in R". The graph of f | graph(f) = {(x1, ---5 Xne1) © RT: Xnes =S (Kay os Xa} is an n-surface in R"*" since graph (f) = g™ '(0) where G(% a5 +249 Xn 1) = Xn —S ay +++ Xn)18 4 Surfaces and Vg(x1, ---5 Xp41) = (X15 +) Xmen, — O01, ---, —O/Oxq, 1) is never zero. . EXAMPLE 4. Let S be an (n — 1)-surface in R", given by S =f~ '(c), where f;U->R (U open in R") is such that Vf(p) #0 for all pe f~*(c). Let g: Uy > R, where U, = U x R= {(xq, «..5 Xn41) € RYE: (x4, «--) Xn) € US, be defined by 0% 15 «+9 Xt) =L as +++ Xn): Then g~ +(c) is an n-surface in R"** because a g Vober ee Raerd (tiene teens Gores Geo} and f/0x,,..., 8f/@xX, cannot be simultaneously zero when G(X, «+s Xnv1) =S (X15 «0 Xn) = because Vf(x1,.--,%_) #0 whenever (X1, «+5 Xa) Ef *(c). The n-surface g~ *(c) is called the cylinder over S (see Figure 4.2). Ax AX eld) sa x2 > > Xx, xy O-sphere L-sphere a (unit circle) in (a): n=0 (b):n = 1 Figure 4.2. The cylinder g~*(1) over the n-sphere: (xy, -.-, Xn+1) = 3 +-7° + x2 EXAMPLE 5. Let C be a curve in R? which lies above the x,-axis. Thus C=f~*(c) for some f: U+R with Vf(p) #0 for all pe C, where U is contained in the upper half plane x, >0. Define S=g™‘(c) where g: U x ROR by g(x1, x2, x3) =f (x1, (x3 + x3)'/7). Then S is a 2-surface (Exercise 4.7). Each point p = (a, b) € C generates a circle of points of S, namely the circle in the plane x,=a consisting of those points (x1, X2, X3) € R? such that x, =a, x3 + x} = b?. S is called the surface of revolution obtained by rotating the curve C about the x,-axis (see Figure 43).4 Surfaces 19 %3 Figure 4.3 The surface of revolution S obtained by rotating the curve C about the Xy-axis. Theorem. Let S be an n-surface in R"*', S =f~1(c) where f: U > R is such that Vf (q) # 0 for all q € S. Suppose g: U > R is a smooth function and p< S is an extreme point of g on S; i.e., either g(q) < g(p) for all q € S or g(q) = g(p) for all q € S. Then there exists areal number A such that Vg(p) = AVf (p). (The number A is called a Lagrange multiplier.) Proor, The tangent space to S at p is S, =[Vf(p)]'. Hence S} is the 1- dimensional subspace of R%*! spanned by Vf(p). It follows, then, that Vo(p) = AVS (p) for some J. € Rif (and only if) Vg(p) € S}; i.e. if (and only if) Va(p) - v = 0 for all v e S,. But each ve S, is of the form v = &(to) for some parametrized curve a: I» S and to € I with a(t) = p. Since p = a(t) is an extreme point of g on S, to is an extreme point of g © a on J. Hence 0 = (fo a) (to) = Valalto)) * &(to) = Valp)-¥ for all ve S, and so Vg(p) = AVf(p) for some A, as required. o Remark. IfS is compact (closed and boundedt) then every smooth function g: UR attains a maximum on S and a minimum on S. The above theorem can then be used to locate candidates for these extreme points. If S is not compact, there may be no extrema. EXxamPLe. Let S be the unit circle x? + x3 = 1 and define g: R?> R by 9(X1, X2) = ax} + 2bx, x2 + cx} where a,b,c eR (see Figure 4.4). Then S=f-i(1) where f(y, xs) = x3 + x4, VE (1 ¥2) = (> Xa» 2X4 2X2), | and Vals» X2) = (1, X2, 2axy + 2bx2, 2bxy + 2exa), +S is closed if R**! — S is open; S is bounded if there exists MR such that |[pi| 0). The four points where these curves are tangent to the unit circle S are the extreme points of g on S. so Vg(p) = AVf (p) for p = (x1, x2) € S if and only if ax, + 2bx2 = 2Ax, bx, + cx, = 2Ax2 or (5 <)(:)-) Thus the extreme points of g on S are eigenvectors of the symmetric matrix (¢ ¢). Note that if x 2 ax? + 2bx4Xx_ + cx} = teal *\e) 2 is an eigenvector of (§ +) then gen’ bo - caxs(*) = Ax} + x3)=4 so the eigenvalue A is just g(p), where p = (x,, x2). Since a2 x 2 matrix has only two eigenvalues, these eigenvalues are the maximum and minimum values of g on the compact set S.4 Surfaces 7 21 EXERCISES 41. 42. 43. 4.4. 45. 46. 47. 48. 49, 4.10. 4.12. For what values of c is the level set f~*(c) an n-surface, where (a) fr, «25 Mega) = XE Ho tte () fa oy Meta) = xb to te (©) fas +++) Knot) = 1X20 at Hh Show that the cylinder x? + x3 = 1 in R® can be represented as a level set of each of the following functions:. (a) f(r, x2, %3) = x3 + x3 (b) f(x, x2, ¥3) = — xt — x3 (©) F(a» Xa» %3) = xt + 2x} + sin(xf + 29). Show that if an n-surface S is represented both as f~'(c) and as g~ *(d) where Vf (p) # 0 and Vg(p) # 0 for all p € S, then for each p € S, Vf(p) = AVg(p) for some real number A # 0. Sketch the graph of the function f: R? + R given by f (x1, x2) = x} — 3x7x2. [Hine: First find the level set f ~ (0). In what region of the plane is f > 0? Where is f < 02] The 2-surface graph (f) is called a monkey saddle. (Why?) Sketch the cylinders f~*(0) where (a) f(a, x2) = x1 (b) f(a, Xa» X3) = 1 — x8 (c) F(a X25 X3) = (Xf /4) + (9/9) - 1 Sketch the cylinder over the graph of f(x) = sin x. Verify that a surface of revolution (Example 5) is a 2-surface. Sketch the surface of revolution obtained by rotating C about the x,-axis, where C is the curve (a) x2=1 — (cylinder) (b) —x3+x3=1,x2>0 (1-sheeted hyperboloid) (c) x3+(%2-2)?=1 (torus) Show that the set S of all unit vectors at all points of R? forms a 3-surface in R*. [Hint: (x1, x2, x3, x4) € S if and only if x3 + x2 = 1] Let S = f~1(c) be a 2-surface in R° which lies in the half space x3 > 0. Find a function g: U + R (U open in R*) such that g~ *(c) is the 3-surface obtained by rotating the 2-surface S$ about the (x;, x2}-plane. . Let a, b,c € R be such that ac — b? > 0. Show that the maximum and mini- mum values of the function g(x1, x2) = x7 + x3 on the ellipse ax} + 2bx, x2 + cx} ='1 are of the form 1/A, and 1/4; where A, and 22 are the eigenvalues of the matrix (§ °). Show that the maximum and minimum values of the function (X15 --+5 Xne1) = LEG21 ayxrx; on the unit msphere x} +-+-+x241= 1, where (a,;) is a symmetric n x n matrix of real numbers, are eigenvalues of the matrix (a,,).22 4 Surfaces 4.13. Show that if S is an n-surface in R"*!, g: R"*! > R is a smooth function, and pe S is an extreme point of g on S, then the tangent space to the level set of g through p is equal to S,, the tangent space to S at p, provided Vg(p) # 0 (sce Figure 4.4). 4.14, Let S be an n-surface in R"*! and let po € R"**, po ¢ S. Show that the shortest line segment from po to S (if one exists) is perpendicular to S; i.e., show that if pS is such that ||po — p||? < ||po — ql? for all gS then (p, po — p) 1 Sp. [Hint: Use the Lagrange multiplier theorem.] Show also that the same conclu- sion holds for the longest line segment from po to S (if one exists). 4,15. R* may be viewed as the set of all 2 x 2 matrices with real entries by identifying the 4-tuple (x1, x2, x3, x4) with the matrix M1 X2 x3 x4) The subset consisting of those matrices with determinant equal to 1 forms a group under matrix multiplication, called the special linear group SL(2). Show that SL(2) is a 3-surface in R*. 4.16. (a) Show that the tangent space SL(2), to SL(2) (Exercise 4.15) at p = (} 9) can be identified with the set of all 2 x 2 matrices of trace zero by showing that SL(2), = |( la ‘yh a +anol [Hint: Show first that if _ a) a) = (SM) is a parametrized curve in SL(2) with a(to)=( $) then (dx,/de) (to) + (dxa/dt)(to) = 0. Then use a dimension argument] (b) What is the tangent space to SL(2) at g= (7 1)? 4.17. (a) Show that the set SL(3) ofall 3 x 3 real matrices with determinant equal to 1 is an 8-surface in R®. (b) What is the tangent space to SL(3) at oro moo > 1 p=]0 0Vector Fields on Surfaces; Orientation A vector field X on an n-surface S < R"*? is a function which assigns to each point p in S a vector X(p).€ R%,*! at p. If X(p) is tangent to S (i.e., X(p) € S,) for each pe S, X is said to be a tangent vector field on S. If X(p) is orthogonal to S (ie. X(p) € St) for each p € S, X is said to be a normal vector field on S (see Figure 5.1). (a) (b) Figure 5.1 Vector fields on the 1-sphere: (a) a tangent vector field, (b) a normal vector field. As usual, we shall work almost exclusively with functions and vector fields which are smooth. A function g: S— R*, where S is an n-surface in R"*+, is smooth if it is the restriction to S of a smooth function g: V > R* defined on some open set V in R"*? containing S. Similarly, a vector field X on S is smooth if it is the restriction to S of a smooth vector field defined on some open set containing S. Thus, X is smooth if and only if X: S> R"*" is smooth, where X(p) = (p, X(p)) for all p € S. 2324 5 Vector Fields on Surfaces; Orientation The following theorem extends to n-surfaces the theorem of Chapter 2 on the existence and uniqueness of integral curves. Theorem 1. Let S be an n-surface in R"*', let X be a smooth tangent vector field on S, and let p € S. Then there exists an open interval I containing 0 and a parametrized curve «: I S such that (i) (0) =p (ii) &(t) = X(a(t)) for all te T (iii) If B: T> S is any other parametrized curve in S satisfying (i) and (ii), then 11 and B(t) = a(t) for all t eT. A parametrized curve a: I > S satisfying condition (ii) is called an inte- gral curve of the tangent vector field X. The unique « satisfying conditions (i}(iii) is the maximal integral curve of X through p € S. Proor. Since X is smooth, there exists an open set V containing S and a smooth vector field X on V such that X(q) = X(q) for all qe S.Letf: U>R and c € R be such that S = f~ '(c) and Vf(q) # 0 for all gS. Let W= {qe Un V: Vf(q) + 9}. Then W is an open set containing S, and both X and fare defined on W. Let Y be the vector field on W, everywhere tangent to the level sets of f, defined by ¥(q) = X(q) — (R(a)- Vf(@/IVF@I?)VF (4). Note that Y(q) = X(q) for all q S. Let «: I> W be the maximal integral curve of Y through p. Then « actually maps I into S because (fe ay(t) = Vf (a(t) + a(t) = VF(e(e) - Y(a(2)) = 0, and f © a(0) = f(p) = c, sof > a(t) = c for all t € I. Conditions (i) and (ii) are clearly satisfied, and condition (iii) is satisfied because any B: TS satisfy- ing (i) and (ii) is also an integral curve of the vector field Y on W so the theorem of Chapter 2 applies. o Corollary. Let S = f ~'(c) be an n-surface in R"*', where f: U > Ris such that Vf (q) # 0 for all q € S, and let X be a smooth vector field on U whose restric- tion to S is a tangent vector field on S. If «: I> U is any integral curve of X such that a(to) € S for some ty € I, then a(t) € S for allt € I. Proor. Suppose «(t) ¢ S for some t € I, t > to. Let t, denote the greatest lower bound of the set {te I:t> to and a(t) ¢ S} Then f(a(t))=c for tp to. The proof for t < to is similar. oO A subset S of R"*! is said to be connected if for each pair p, q of points in S there is a continuous map «: [a, b] > S, from some closed interval [a, b] into S, such that a(a) = p and a(b) = q. Thus S is connected if each pair of points in S can be joined by a continuous, but not necessarily smooth, curve which lies completely in S. Note, for example, that the n-sphere (Figure 5.2) is connected if and only if n > 1 (Exercise 5.1). X x3 1 s° é s s? x LS : Pp 4™ % % (a): n=0 (b):n=1 ():n=2 Figure 5.2 The n-sphere x} + --- + x2,, = 1 is connected if and only if n > 1. In this book, we shall deal almost exclusively with connected n-surfaces. As we shall see in Chapter 15 (see Exercise 15.13), given any n-surface S and any p ¢ S, the subset of S consisting of all points of S which can be joined to p by a continuous curve in S is itself an n-surface, and it is connected. Hence we can study S by studying separately each of these “connected compon- ents” of S, \ Theorem 2. Let S < R"*? be a connected n-surface in R"* '. Then there exist on S exactly two smooth unit normal vector fields N, and N,, and N2(p)= —N,(p) for all pe S. Proor. Let f: U -+ R and c € R be such that S = f~ *(c) and Vf(p) # 0 for all pe 5S. Then the vector field N, on S defined by Wp) | Nip)=727-7> PES Tork clearly has the required properties, as does the vector field N, defined by N,(p) = —N,(p) for all pe S. To show that these are the only two such vector fields, suppose N, were26 5 Vector Fields on Surfaces; Orientation another. Then, for each p € S, N3(p) must be a multiple of N,(p) since both lie in the 1-dimensional subspace S} c R’,*!. Thus Na(p) = gP)NiP) where g: S— R is a smooth function on S (g(p) = Na(p) + Ni(p) for p € S). Since N,(p) and N,(p) are both unit vectors, g(p) = +1 for each pe S. Finally, since g is smooth and S is connected, g must be constant on S (see Exercise 5.2). Thus either N, = N, or N; = N2. Oo ‘A smooth unit normal vector field on an n-surface S in R"*'-is called an orientation on S. According to the theorem just proved, each connected n-surface in R"*! has exactly two orientations. An n-surface together with a choice of orientation is called an oriented n-surface. Remark. There are subsets of R"* ' which most people would agree should be called n-surfaces but on which there exist no orientations. An example is the Mobius band B, the surface in R° obtained by taking a rectangular strip of paper, twisting one end through 180°, and taping the ends together (see Figure 5.3). That there is no smooth unit normal vector field on B can be Figure 5.3. The Mdbius band. seen by picking a unit normal vector at some point on the central circle and trying to extend it continuously to a unit normal vector field along this circle. After going around the circle once, the normal vector is pointing in the opposite direction! Since there is no smooth unit normal vector field on B, B cannot be expressed as a level set f~*(c) of some smooth function ff: UR with Vf(p) 0 for all p € S, and hence B is nota 2-surface accord-5 Vector Fields on Surfaces; Orientation 27 ing to our definition. B is an example of an “ unorientable 2-surface”. Until Chapter 14, we shall consider only “orientable” n-surfaces in R"*'. A unit vector in R%*! (pe R"**) is called a direction at p: Thus an orientation on an n-surface S in R"*! is, by definition, a smooth choice of normal direction at each point of S. On a plane curve, an orientation can be used to define a tangent direction at each point of the curve. The positive tangent direction at the point p of the oriented plane curve C is the direction obtained by rotating the orientation normal direction at p through an angle of —1/2, where the direction of Positive rotation is counterclockwise (see Figure 5.4). @ (b) Figure 5.4 Orientation on a plane curve: (a) the chosen normal direction at each point determines (b) a choice of tangent direction at each point. On a 2-surface in R?, an orientation can be used to define a direction of rotation in the tangent space at each point of the surface. Given 6 € R, the positive 0-rotation at the point p of the oriented 2-surface S is the linear transformation Ry: S,>S, defined by Rg(v) = (cos Ov + (sin 6)N(p) x V where N(p) is the orientation normal direction at p. R, is usually described as the “right-handed rotation about N(p) through the angle 6” (see Figure 5.5). S ee =NO)x¥ Figure 5.5 Orientation on the 2-sphere: at each point the chosen normal direction determines a sense of positive rotation in the tangent space. The satellite figure is an enlarged view of one tangent space.28 5 Vector Fields on Surfaces; Orientation On a 3-surface in R*, an orientation can be used to define a sense of “handedness” in the tangent space at each point of the surface. Given an oriented 3-surface S and a point pe S, an ordered orthonormal basis {e;, €2, €3} for the tangent space S, to S at p is said to be right-handed if the determinant a1 det | ©? 3 N(P) is positive, where N(p) = (p, N(p)) is the orientation normal direction at p and e; = (p, e;) for ic {1, 2, 3}; the basis is left-handed if the determinant is negative. On an n-surface in R"*1 (n arbitrary), an orientation can be used to partition the collection of all ordered bases for each tangent space into two subsets, those consistent with the orientation and those inconsistent with the orientation. An ordered basis {v,, ..., v,} (not necessarily orthonormal) for the tangent space S, at the point p of the oriented n-surface S is said to be consistent with the orientation N on S if the determinant vy det vy N(p) is positive; the basis is inconsistent with N if the determinant is negative. Here, as usual, v; = (p, v;) and N(p) = (p, N(p)). EXERCISES 5.1. Show that the unit n-sphere x? + --* + x24, = 1 is connected if n > 1. 5.2. Show that if S is a connected n-surface in R"*! and g: S— R is smooth and takes on only the values +1 and —1, then g is constant. [Hint: Let p € S. For q€S, let «: [a, b] + S be continuous and such that «(a) = p, a(b) = q. Use the intermediate value theorem on the composition g o «.] 5.3. Show by example that if S is not connected then Theorem 2 of this section fails. 5.4. Show that the two orientations on the n-sphere x} + °-* + x24 =r? of radius 7 > 0 are given by Ny(p) = (p, p/r) and Na(p) = (p, —p/r). 5.5. R" may be viewed as the n-surface x,+, = 0 in R"*!. Let N be the orientation on R"c R"*! defined by N(p) = (p, 0,...,0, 1) for each p € R”. (This N is called the natural orientation on R".) Show that, given this orientation for each n, (a) the positive tangent direction at p € R' is the direction (p, 1, 0), (b) the positive 0-rotation in R2, p € R?, is counterclockwise rotation through the angle 0, and| 5 Vector Fields on Surfaces; Orientation : 29 5.6. 5.7. 5.8. 5.9. (c) the ordered orthonormal basis {(p, 1, 0, 0, 0) (p, 0, 1, 0, 0), (p, 0, 0, 1, 0)} for R3, p € R°, is right-handed. Let C be an oriented plane curve and let v be a nonzero vector tangent to C at pe C. Show that the basis {v} for C, is consistent with the orientation on C if and only if the positive tangent direction at p is v/||v|. [Hint: Let 6 denote the angle measured counterclockwise from (p, 1,0) to the orientation direction N(p), so that N(p) = (p, cos @, sin 6). Express both v and the positive tangent direction at p in terms of 6.) Recall that the cross product v x w of two vectors v = (p, 0;, v2, v3) and w= (p, 1, W2, Ws) in R2 (p € R®) is defined by VX w= (p, 0203 — V3W2, D9W1 — D1W3, D1 W2 — 024). (a) Show that v x w is orthogonal to both v and w and that ||v x wi = Iv|| | w]| sin 6, where @ = cos~*(v - w/||y|| ||w]|) is the angle between v and w. (b) Show that if u = (p, us, uz, us) then uy uy us us (vx w)=ve(wxu)=we(uxv)=|o, 02 05). Ww, Wa Ws (c) Show that the only vector x in R} such that u- x is equal to the determi- nant above (part b) for all we R} is x =v x w. Let S be an oriented 2-surface in R* and let {v, w} be an ordered basis for the tangent space S, to S at pe S. Show that the consistency of {v, w} with the orientation N of S is equivalent with each of the following conditions: (a) N(p):(v x w)>0 (b) w/||w|| = Ro{v/|vl]) for some 6 with 0 < @ (Ww x a) = Ww (ux ¥) (vii) v x w is orthogonal to both v and w .30 5 Vector Fields on Surfaces; Orientation (viii) u+(v x w) =0 if and only if {u, v, w) is linearly dependent (ix) An ordered orthonormal basis {e;, €2, €3} for S, is right-handed if and only if e3 (e; x €2) > 0. 5.10. Let S be an oriented n-surface in R"*', with orientation N, and let p € S. (a) Show that an ordered basis for S, is inconsistent with N if and only if it is consistent with —N. (b) Suppose {v,, ..., ¥_} is an ordered basis for S, which is consistent with N and suppose {W,, ...,W,} is another ordered basis for S,. Show that {w, .... Wa} is also consistent with N if and only if the matrix (a;,), where w; = Dy 4y¥), has positive determinant. [Hint: Complete each basis to a basis for 3+! by adjoining N(p). What is the relationship between (a;;) and the two matrices which determine the consistency of the given bases with NYThe Gauss Map An oriented n-surface in R"*! is more than just an n-surface S, it is an n-surface S together with a smooth unit normal vector field N on S. The function N: S > R"*? associated with the vector field N by N(p) = (p, N(p)), pe S, actually maps S into the unit n-sphere $"< R"*, since ||N(p)|| = 1 for all p € S. Thus, associated to each oriented n-surface S is a smooth map N: S—S*, called the Gauss map. N may be thought of as the map which assigns to each point p € S the point in R"*! obtained by “translating” the unit normal vector N(p) to the origin (see Figure 6.1). Figure 6.1 The Gauss map of a 1-surface in R?. The image of the Gauss map, N(S) = {q € S": q = N(p) for some p € S} is called the spherical image of the oriented n-surface S (see Figure 6.2). 3132 6 The Gauss Map 4X x N(S) Ss (a):n=1 AX *2 N(S) x1 A Ss (b):n =2 Figure 6.2 The spherical image of one sheet of a 2-sheeted hyperboloid x} — x3 — s+ —x2,,=4, x,>0, oriented by N= —Vf/||Vf|| where f(x1,--- 2 2 2 x1 x2 Xnt ie The spherical image of an oriented n-surface S records the set of direc- tions which occur as normal directions to S. Hence its size is a measure of how much the surface curves around in R"*'. For an n-plane, which doesn’t curve around at all, the spherical image is a single point. If an n-surface is compact (closed and bounded) then it must curve all the way around: the spherical image will be all of S". Although we do not yet have enough machinery to prove this theorem in full generality, we can already prove an important special case, namely the case in which S is a level set of a smooth function defined on all of R"*!. Theorem. Let S be a compact connected oriented n-surface in R"*' exhibited as a level set f ~*(c) of a smooth function f: R"** + R with Vf (p) # 0 for all pe S. Then the Gauss map maps S onto the unit sphere S".6 The Gauss Map 33 Proor. The idea of the proof is as follows. Given v € S", consider the n-plane v'. By moving this n-plane far enough in the v-direction, it will have null intersection with S. Bringing it back in until it just touches S at some point p, it will be tangent there (see Figure 6.3). Hence at this point, N(p) = +v. If Figure 6.3 The Gauss map of a compact oriented n-surface is onto. N(p) = —», then N(q)=v where q is obtained similarly, by moving the n-plane in from the opposite direction. More precisely, consider the function g: R"*?—+ R defined by g(p) = pov; ie, g(xy, ---5 Xn¢a) = 1X1 +77 + Opt 1 Xng1 Where v= (ay, «5 dye) The level sets of g are the n-planes parallel to v‘. Since S is compact, the restriction to S of the function g attains its maximum and its minimum, say at p and q respectively. By the Lagrange multiplier theorem (Chapter 4), (p, ») = Volp) = AVS (P) = AIWF )IN@) for some A € R. Hence v and N(p) are multiples of one another. Since both have unit length, it follows that N(p) = +v. Similarly, N(q) = +v. It remains only to check that N(q) # N(p). For this, it suffices to con- struct a continuous function «: [a, b] + R"*', differentiable at a and b, such that (i) a(a) =p, a(b)=q, &a)= ® »), &(b)=(q,0) and (ii) a(t) ¢ S fora «(a) =f © a(b) = c, this would imply that there exist t; and t, between a and b with f° a(t,) > candf > a(t) 1. Then the function a defined by a,(t) (0 0 since (g © a)(t) = Va(a,(t)) - a:(t) =v-v>Osog is increasing along «,, and the maximum value of g on S is attained at a,(0) = p; (2) a,(t) ¢ S for t > 0 since (g © «)'(t) = v - (—v) <0 so g is decreasing along «,, and the minimum value of g on S is attained at a,(0) = q; and (3) a3(t) ¢ S for t € [b, ba] since @3(t) € S, and S; \S=@. oOo Remark. Some insight into the general case of this theorem can be gained6 The Gauss Map 35 from the following intuitive argument. Suppose S is compact and connected. Then S divides R"*! into two parts, a part inside S and a part outside S. Let S = f~1(c) where f: U +R. Then, for small enough « > 0, the level set f7*(c + e) will be an n-surface S, gotten by pushing each point of S.a short distance out from S along Vf (see Figure 6.5). =f Ya (ete) s,s ete) s=f(c) S_=f7}(c-e) ~ — Figure 6.5 Given a compact connected n-surface S = f~*(c), the nearby level sets f7 "ce —«) and f~(c + e) are slightly inside and slightly outside S. (Possibly, f~ *(c + ¢) might also contain some points far away from S but we can ignore such points in the present argument.) Similarly, for small enough > 0, the level set f~*(c — e) will be an n-surface S_ on the other side of S, gotten by pushing each point of S a short distance out along — Vf. Denoting by V the set of points between S_ and S,, by V, the set of points in R"*! — V which lie on the same side of S as S,, and by V_ the set of points in R"*! — V which lie on the other side of S, we can define 7’: R"*! + R by S(p) forpeV Fp)={ete forpev, ce—e forpeV_. Then 7 is continuous on R"*, smooth on the open set V about S, and F-\(c) = S. The above proof can now be applied, with f replaced everywhere by f, to show that the Gauss map is onto. EXERCISES In Exercises 6.1-6.5, describe the spherical image, when n = 1 and when n=2, of the given n-surface, oriented by Vf/||Vf|| where f is the function defined by the left hand side of each equation. 6.1. The cylinder x3 +++: +.x2,, =1. | 62. The cone —x? + x3 +--+ + x24: =0, x1 >0. 63. The sphere x} + x3 +°** + x24, =7? (r > 0). 6.4. The paraboloid —x, + x3 +°7° + x24. =0.36 6.5. 6.6. 6.7. 68. 6.9. 6 The Gauss Map The 1-sheeted hyperboloid —(x2/a?) + x3 +++ +x241=1 (a> 0). What happens to the spherical image when a> co? When a0? Show that the spherical image of an n-surface with orientation N is the reflection through the origin of the spherical image of the same n-surface with orientation —N. Let a = (a4, ..., an+1) € R"*!, a #0. Show that the spherical image of an n- surface S is contained in the n-plane a,x; +*** + dq+1%q+1 = 0 if and only if for every p € S there is an open interval J about 0 such that p + ta € S for all te I. [Hint: For the “only if” part, apply the corollary to Theorem 1, Chapter 5, to the constant vector field X(q) = (q, a).] Show that if the spherical image of a connected n-surface S is a single point then S is part or all of an n-plane. [Hint: First show, by applying the corollary to Theorem 1, Chapter 5, to the constant vector fields W(q) = (q, w), where w 1 v, {v} = N(S), that if Bis an open ball which is contained in U and pe S n Bthen Ho BCS where H is the n-plane {x € R"*!: x - p= p- v}. Then show that if a: [a, b] > S is continuous and a(t) € B for t; 0 if and only if Vf(a(t2)) + &(t2) < 0.) (b) Show that if S is compact and a goes to oo in both directions [ie Fim,..— lla(¢)|| = lim, +. {la(t)| = 00] then a crosses S an even number of times. 6.10. Let S be a compact n-surface in R"**. A point p € R"** — Sis outside S if there exists a continuous map a:([0,0)—+R"*'—S such that a(0)=p and lim, }a(¢)|| = 00. Let ©(S) denote the set of all points outside S. (a) Show that if p: [a, b] + R"*? — S is continuous and f(a) € 0(S) then f(t) € O(S) for all t € {a, b]. (b) Show that 0(S) is a connected open subset of R"*!,Geodesics Geodesics are curves in n-surfaces which play the same role as do straight lines in R”. Before formulating a precise definition, we must introduce the process of differentiation of vector fields and functions defined along pa- rametrized curves. In order to allow the possibility that such vector fields and functions may take on different values at a point where a parametrized curve crosses itself, it is convenient to regard these fields and functions to be defined on the parameter interval rather than on the image of the curve. A vector field X along the parametrized curve a: I + R"*' is a function which assigns to each t € J a vector X(t) at a(t); ie., X(t) € Raj)’ for allt € I. A function f along « is simply a function f: J > R. Thus, for example, the velocity & of the parametrized curve «: [+ R"*? is a vector field along « (Figure 7.1); its length ||é||: I > R, defined by ||4||(t) = ||é(t)|| for allt € J, isa function along «. ||| is called the speed of a. & (1) & (4) Figure 7.1 The velocity field along a parametrized curve a. Note that a(t,) = a(t2) does not imply that &(¢1) = &(t2). 387 Geodesics 39 Vector fields and functions along parametrized curves frequently occur as restrictions. Thus if X is a vector field on U, where U is an open subset of R"*? containing Image a, then X © a is a vector field along «. Similarly f > « is a function along a whenever f: U + R, where U > Image a. Each vector field X along « is of the form X(t) = (2(0), Xa(0), «+» XnvalE)) where each component X, is a function along «. X is smooth ifeach X;: I> R is smooth. The derivative of a smooth vector field X along a is the vector field X along « defined by x()= (at, S09... Mert) X(t) measures the rate of change of the vector part (X,(¢), ..., X,+1(¢)) of X(t) along a. Thus, for example, the acceleration & of a parametrized curve is the vector field along « obtained by differentiating the velocity field [& = (&)] (see Figure 7.2). aX, a (to) a(t) Figure 7.2 The acceleration (to) is the derivative at to of the velocity vector field &. It is easy to check (Exercise 7.4) that differentiation of vector fields along parametrized curves has the following properties. For X and Y smooth vector fields along the parametrized curve «: J > R"*+ and fa smooth func- tion along a, (i) K+Y¥Y)=X+¥ (i) (X) = fX+fX (ii) (K- YY =K-Y¥+X-¥ where X + Y, fX, and X - ¥ are defined along « by (X + ¥)(t) = X(0) + ¥(t) | (FX)e) = FOX) (K+ YN) = XQ) + YO) for all te I.40 7 Geodesics A geodesic in an n-surface Sc R"*! is a parametrized curve «: 1+ S whose acceleration is everywhere orthogonal to S; that is, &(t) € Sz) for all t € I. Thus a geodesic is a curve in S which always goes “straight ahead ” in the surface. Its acceleration serves only to keep it in the surface. It has no component of acceleration tangent to the surface. Note that geodesics have constant speed, because a(t) eS.) and G(t) € Sip for allt eT ‘ones that 4 bade? = 4 (0) 4) = 240) -5(0) = 0. EXAMPLE 1. If an n-surface S contains a straight line segment a(t) = p + tv (t € I) then that segment is a geodesic in S. Indeed, &(t) = 0 for allt € Iso in particular &(t) 1 Sj) for all t € I. EXAMPLE 2. For each a, b, c, de R, the parametrized curve a(t) = (cos(at + b), sin(at + b), ct + d) is a geodesic in the cylinder x} + x3 = Lin R®, because &(t) = (a(t), —a? cos(at + b), —a? sin(at + b), 0) = +a?N(a(t)) for all t € R (see Figure 7.3). Image x $*3 Image 2 %3 % x2 2 X2 x x1 xy (a) (b) (c) Figure 7.3 Geodesics a(t) = (cos(at +b), sin(at +b), ct +d) in the cylinder x7 +x3= 1. (a)a=0, (b) c=0, (c) a#0,c #0. EXAMPLE 3. For each pair of orthogonal unit vectors {e,, e2} in R® and each ae R, the great circle (or point if a = 0) a(t) = (cos at)e, + (sin at)e, isa geodesic in the 2-sphere x} +x3+x3=1 in R%, because &(r) = (a(t), — aa(t)) = +a?N(a(t)) for all t R (see Figure 7.4). Intuitively, it seems clear that given any point p in an n-surface S and any initial velocity v at p (v € S,) there should be a geodesic in S passing through7 Geodesics 4 3 Image « Figure 7.4 Great circles are geodesics in the 2-sphere. Pp with initial velocity v. After all, a racing car travelling on S, passing through p with velocity v, should be able to continue travelling “straight ahead” on S at'constant speed ||v||, thereby tracing out a geodesic in S. The following theorem shows that this is the case, and that the geodesic with these properties is essentially unique. Theorem. Let S be an n-surface in R"**, let p € S, and let v € S,. Then there exists an open interval I containing 0 and a geodesic a: I + S such that (i) (0) = p and &(0) =v. (ii) If B: TS is any other geodesic in S with B(0) = p and B(0) =v, then Tc] and B(t) = a(t) for all t eT. Remark. The geodesic a, is called the maximal geodesic in S passing through p with initial velocity v. Proor. S = f~1(c) for some c € R and some smooth function f: U > R (U open in R"**) with Vf(p) # 0 for all p € S. Since Vf (p) # 0 for all p in some open set containing S, we may assume (by shrinking U if necessary) that Vf (p) # 0 for all p € U. Set N = Vf/||Vf ||. By definition, a parametrized curve «: I - S is a geodesic of S ifand only if its acceleration is everywhere perpendicular to S; that is, if and only if a(t) is a multiple of N(«(¢)) for all t € I: H(t) = g(QN(o(2)) for all t ¢ I, where g: I> R. Taking the dot product of both sides of this equation with N(a(t)) we find g=%-Noa=(k-Noas—a-Nea =-&-Noa, | since & + N oa =0, Thus «: J S is a geodesic if and only if it satisfies the differential equation (6) &+(&+Nea)(Noa)=0.42 7 Geodesics This is a second order differential equation in «. (If we write a(t) = (x1(t), .-., Xn+1(t)), this vector differential equation becomes the system of second order differential equations dx, "a ON; dx, dx, TET De Neos Bae Ge Oe oo Mae Ge Ge = where the N; (j € {1, ..., 1 + 1}) are the components of N.) By the existence theorem for the solutions of such equations,+ there exists an open interval J, about 0 and a solution B,: J, > U of this differential equation satisfying the initial conditions B,(0) = p and B,(0) = v (that is, satisfying x,(0) = p; and (dx,/dt)(0) =v, for ie{l,....n+1}, where p=(pi,...,Pr+1) and Vv = (p, 04, ..., Un41))- Moreover, this solution is unique in the sense that if B: 1, U is another solution of (G), with B,(0) = p and B,(0) =v, then B,(t) = B2(t) for all t € 1, 7 I,. It follows that there exists a maximal open interval J (J is the union of the domains of all solutions to (G) which map 0 into p and have initial velocity v) and a unique solution #: I> U of (G6) satisfying «(0) = p and 4(0) = v. Furthermore, if B: I > U is any solution of (G) with (0) = p and B(0) = v then Tc J and f is the restriction of « to 7. To complete the proof, it remains only to show that the solution « to (G) is actually a curve in S. For, if so, it must be a geodesic since it satisfies the geodesic equation (G), and the rest of the theorem follows from the uni- queness statements above. To see that « is in fact a curve in S, note first that for every solution a: 1+ U of (G), «Na =0. Indeed, «Noa! =%-Noata+Nea=0 0 by (G), so & + N is constant along «, and («+N a)(0)=v-+N(p)=0 since v € S, and N(p) L S,. It follows then that (F< a) (0) = Vf (ee)) = at) = VF (aLe))[IN(@e)) + fe) = 0 for all t ¢ I so fo a is constant, and f(«(0)) = f(p) = c so f(«(t)) = for all te J; that is, Image « c f~*(c) = S. o It follows from the theorem just proved that each maximal geodesic on the unit 2-sphere in R* (Example 3) is either a great circle (parametrized by a constant speed parametrization) or is constant (a(t) = p for all t, some p) since such a curve can be found through each point p with any given initial velocity. Similarly, each maximal geodesic on the cylinder x? + x3 = 1 in R® (Example 2) is either a vertical line, a horizontal circle, a helix (spiral), or is constant. + See eg. W. Hurewicz, Lectures on Ordinary Differential Equations, Cambridge, Mass.: MIT Press (1958), pp. 32-33. See also Exercise 9.15.7 Geodesics 43 EXERCISES TA 72. 73. 74. 75. 16. 77, 78. Find the velocity, the acceleration, and the speed of each of the following parametrized curves: (a) a(t) =( 2) 08 t, sin t) (cos 32, sin 3t) 308 f, sin t, t) (cos t, sin ¢, 2 cos t, 2 sin t). Show that if a: J R"*! is a parametrized curve with constant speed then &(t) L &(t) for all t € I, Let a: 1+ R'*! be a parametrized curve with &(t) # 0 for all t € I. Show that there exists a unit speed reparametrization of a; i.c., show that there exists an interval J and a smooth function h: J + I (onto) such that h’ > 0 and such that B =a hhas unit speed. [Hint: Set h = s~* where s(t) = J}, ||é(¢)|| dt for some toe I] Let X and Y be smooth vector fields along the parametrized curve a: 1 + R"** and let f: I+ R be a smooth function along «. Verify that (a) KiY)=X+¥ (b) (PX) = FX + fX () (KY =X-¥4X-¥. Let S denote the cylinder x? + x3 = r? of radius r > 0 in R®. Show that a is a geodesic of S if and only if « is of the form a(t) = (r cos(at + b), r sin(at + b), ct + d) for some a, b, c,d eR. Show that a parametrized curve a in the unit n-sphere x} + ++" + xn+1 = lisa geodesic if and only if it is of the form a(t) = (cos at)e, + (sin at)ez for some orthogonal pair of unit vectors {e1, €2} in R"*! and some a € R. (For a #0, these curves are “great circles” on the n-sphere.) Show that if a: I > S is a geodesic in an n-surface S and if B = a ° his a repar- ametrization of a(h: T+ I) then f is a geodesic in S if and only if there exist a,b € R with a> 0 such that h(t) = at + b for all te I. Let C be a plane curve in the upper half plane x2 > 0 and let S be the surface of revolution obtained by rotating C about the x,-axis (see Example 5, Chapter 4). Let a: I+ C, a(t) = (x(t), x2(t)), be a constant speed parametrized curve inc. For each 0 € R, define ay: 1+ S by a(t) = (x(t), x2(t) cos 8, x2(t) sin 8) and, for each € I, define ,: R + S by the same formula: Bi(0) = (x1(t), x2(t) cos 8, x2(t) sin 8).7 Geodesics 2 meridians parallels profile curve C Figure 7.5 Geodesics on a surface of revolution: all meridians are geodesics; a parallel is a geodesic if it cuts the “ profile curve” C at a point where the slope of C is 79. 7.10. TAL 7.12. zero. The curves a are called meridians of S, and the circles 8, are parallels of S (see Figure 7.5). (a) Show that meridians and parallels always meet orthogonally; ic., do(t)- B,(0) = 0 for all te 1, ER. (b) Show that each meridian og is a geodesic of S. [Hint: Note that {c(¢), p,(0)} spans S, , where p = a,(t). Hence it suffices to check that d(t) is perpendicu- lar to both a(t) and B,(8).] (c) Show that a parallel B, is a geodesic of S if and only if the slope x2(t)/x1(t) of the tangent line to C at a(t) is zero. Let S be an n-surface in R'*}, let v € Sp, p € S,and let a: + S be the maximal geodesic in S with initial velocity v. Show that the maximal geodesic f in S with initial velocity cv (c € R) is given by the formula A(t) = a(ct). Let S be an n-surface in R"*', let pe S, let ve S,, and let a: 1+ S be the maximal geodesic in S passing through p with velocity v. Show that if B: T+ S is any geodesic in S with (to) = p and B(to) = v for some to € I then f(t) = a(t — to) for allt € I. Let S be an n-surface in R**# and let B: I+ S be a geodesic in S with (to) = B(O) and B(to) = B(0) for some to € 1, to + 0. Show that f is periodic by show- ing that B(t + to) = B(t) for all ¢ such that both t and ¢ + to € I. [Hint: Use Exercise 7.10.] An n-surface S in R"** is said to be geodesically complete if every maximal geodesic in S has domain R. Which of the following n-surfaces are geodesically complete? (a) The n-sphere x? +°++ + x24. = 1. (b) The n-sphere with the north pole deleted: x} + -*° + x21 =1, Xnt1 #1 (c) The cone x + x3 — x3 =0, x3 >0 in B®. (d) The cylinder x? + x3 = 1 in R®*. (e) The cylinder in R* with a straight line deleted: x? + x3 = 1, x, 41. tiveParallel Transport A vector field X along a parametrized curve a: JS in an n-surface S is tangent to S along a if X(t) € Su for all t € I. The derivative X of such a vector field is, however, generally not tangent to S. We can, nevertheless, ‘obtain a vector field tangent to S by projecting X(t) orthogonally onto S, for each te J (see Figure 8.1). This process of differentiating and then x N(a(t)) V Sa Figure 8.1 The covariant derivative X’(t) is the orthogonal projection onto the tangent space of the ordinary derivative X(t). projecting onto the tangent space to S defines an operation with the same properties as differentiation, except that now differentiation of vector fields tangent to S yields vector fields tangent to S. This operation is called covar- iant differentiation. : Let S be an n-surface in: R"*', let a: I> S be a.parametrized curve in S, and let X be a smooth vector field tangent to S along «. The covariant derivative of X is the vector field X’ tangent to S along a defined by X(t) = X() — [XO » N(@(t)ING()), pe 45 ‘46 8 Parallel Transport where N is an orientation on S. Note that X(t) is independent of the choice of N since replacing N by —N has no effect on the above formula. It is easy to check that covariant differentiation has the following proper- ties: for X and Y smooth vector fields tangent to S along a parameterized curve a: J + S and fa smooth function along a, (i) (X+YyY=xX'+Y (ii) (fX) =f'X + fx’ (iii) (X-Yy=X'-Y4+X-Y. These properties follow immediately from the corresponding properties of ordinary differentiation. For example, the following computation verifies (iii): (KX: Y/=X-¥4+X-¥ = [K+ (K+ Nog)Noa]-Y+X-[¥+(V-Nea)Noa] =X'-Y+X-Y, since N is perpendicular to S and X and Y are tangent to S. Intuitively, the covariant derivative X’ measures the rate of change of X along « as seen from the surface S (by ignoring the component of X normal to S). Note that a parametrized curve a: I — S is a geodesic in S ifand only if its covariant acceleration (a) is zero along a. The covariant derivative leads naturally to a concept of parallelism on an n-surface. In R"*!, vectors v = (p, v) € RS! and w = (q, w) € Ry*? are said to be Euclidean parallel if v = w (see Figure 8.2(a)). A vector field X along a a | (a) (b) Figure 8.2 Euclidean parallelism in R?: (a) parallel vectors; (b) a parallel vector field. parametrized curve «: I + R"*? is Euclidean parallel if X(t) = X(tz) forall - ty, t2 € I, where X(t) = (a(t), X(t) for t eI (see Figure 8.2(b)). Thus X is Euclidean parallel along « if and only if X = 0. Given an n-surface S$ in R"*! and a parametrized curve a: 1—>S, a smooth vector field X tangent to S along « is said to be Levi-Civita parallel, or simply parallel, if X’ = 0. Intuitively, X is parallel along « if X is a constant vector field along a, as seen from S.8 Parallel Transport 47 Note that Levi-Civita parallelism has the following properties: (i) If X is parallel along a, then X has constant length, since divi _ a ogy we GIMP = GX) = 2K X=. (ii) If X and Y are two parallel vector fields along a, then X + Y is constant along «, since (K+ Yy =X'+¥+X-+Y'=0. (iii) If X and Y are parallel along a, then the angle cos” '(X + Y/||X|| ||Y||) between X and Y is constant along «, since X + Y, |[X||, and ||Y|| are each constant along a. (iv) If X and Y are parallel along a then so are X + Y and cX, for allc € R. (v) The velocity vector field along a parametrized curve « in S is parallel if and only if « is a geodesic. Theorem 1. Let S be an n-surface in R"**, let a: I + S be a parametrized curve in S, let to € I, and let v € Sq). Then there exists a unique vector field V, tangent to S along «, which is parallel and has V(to) = v. Proor. We require a vector field V tangent to S along « satisfying V’ = 0. But : VW=V-(V:Noa)Noa =V-[(V+Noa)!—V-Néa]Noa =V+(V-Noa)Nou so V’ = 0 if and only if V satisfies the differential equation (P) V+(V°Noa)Noa=0. This is a first order differential equation in V. (If we write V(t) = (a(t), V,(t), ..-» Va+s(t)), this vector differential equation becomes the system of first order differential equations net aY Sv, oa\iN, ay =0 oo where the N, (j € {1, ..., m + 1}) are the components of N.) By the existence and uniqueness theorem for solutions of first order differential equations, there exists a unique vector field V along a satisfying equation (P) together with the initial condition V(t) =v (that is, satisfying Vi(to) =v, for ie{l,...,.n+1}, where v= (a(to),04,...,0,+1)). The existence and uniqueness theorem does not guarantee, however, that V is tangent to S along a.48 8 Parallel Transport To see that V is indeed tangent to S, simply note that, by (P), (V-Noa/=V-NoatV-Noa =[-(V-Néa)Noa]*Noa+V*Nia =-V:Neat+V-Nia=0, so V - N © ais constant along « and, since (V - N © a)(to) = v * N(a(to)) = 0, this constant must be zero. Finally, this vector field V, tangent to S along a, is parallel because it satisfies equation (P). a Remark. We have implicitly assumed in the above proof that the solution V of (P) satisfying V(t) = v is defined on the whole interval J and not just on some smaller interval containing to . That this is indeed the case can be seen from the following argument. Suppose Ic] is the maximal interval on which there exists a solution V of (P) satisfying V(t) = v. If 1 # I, there exists an endpoint b of T with be J. Let {t} be a sequence in I with lim,..,. t; = b. Since ||V|| is constant on J, ||V(t,)|| = ||v|| for all i, so the sequence {V(t,)} of vector parts of {V(z;)} takes values in a compact set, the sphere of radius ||v|| about the origin in R"*’. It follows that {V(t,)} must have a convergent subsequence {V(t,,)}. Let w = lim,..,, V(t;,), and let W be a solution of (P), on some interval J containing b, with W(b) = (a(b), w). Then W — V is also a solution of (P), on I > J, and in particular ||W — V|| is constant on I 4 J. But Jim |W) — VUe)l = Jw —w] =0 so |W — V|| =0 on J 1 T; that is, W = V on J 9 I. Hence the vector field on I u I which is equal to V on J and to W on I extends V to a solution of (P) on an interval larger than I, contradicting the maximality of I. Hence I =1, as claimed. Corollary. Let S be a 2-surface in R° and let a: I + S be a geodesic in S with & #0. Then a vector field X tangent to S along « is parallel along « if and only if both ||X|| and the angle between X and & are constant along « (see Figure 8.3). Figure 8.3 Levi-Civita parallel vector fields along geodesics in the 2-sphere.