Curves

and Surfaces
- ' SECOND EDITION

Sebastian Montiel
Antonio Ros

Graduate Studies
in Mathematics
Volume 69

American Mathematical Society

Real Sociedadi Matematica Espafiola
Curves and Surfaces
SECOND EDITION This is a text on the classical differential geometry of curves and surfaces from
a modern point of view. In recent years, there has been a marked increase in
the interest in the global aspects of this topic, which is the emphasis here. In
order to study the global properties of curves and surfaces, more sophisticated
tools are necessary than are usually found in texts on the same topic. Also,
topological questions arise and must be understood. Thus, for instance, there is
a treatment of the Gauss-Bonnet theorem and a discussion of the Euler charac-
teristic.The authors also cover Alexandrov's theorem on compact surfaces in R3
with constant mean curvature. The last chapter discusses the global geometry
of curves, including periodic space curves and the four-vertices theorem in the
general case where the curves are not necessarily convex.
Besides being an introduction to the lively subject of curves and surfaces, this
book can also be used as an entry to a wider study of differential geometry in
general. It is suitable for a first-year graduate course or an advanced under-
graduate course.

U
For additional information
and updates on this book, visit
www.ams.orglbool<pageslgsm-69

ISBN 978-0-8218-4763-3

American Mathematical Society

www.ams.org
91178082111847633
GSM/69.R Real Sociedad Matematica Espanola
www rsme.es
 376 Index

Region, 136 Cohn-Vossen, 218

Regular value, 36, 111 divergence, 158, 159, 184, 322
Reilly, 173 Euler, 82
Reparametrization, 8 Fary-Milnor, 345

Curves and Surfaces by arc length, 8

Riemann, 32, 239
Rigid motion, 7, 80, 144, 207
Fenchel, 342
four-vertices, 329, 348
Fubini, 146
Rinow, 206, 238 fundamental, 11, 18, 209
SECOND EDITION Rodrigues, 2, 68 Gauss, 160
Rustihauser, 312 Hadamard-Stoker, 175, 178
Hilbert, 91
Saint-Simon, 68 Hilbert-Liebmann, 94, 184
Saint-Venant, 2 Hopf-Rinow, 238
Samelson, 108, 118, 312 inverse function, 52
Sard, 112, 132, 311, 321 Jellett-Liebmann, 93, 185
Sebastian Montiel Schmidt, 311 Jordan curve, 320
Schwarz, 173 Jordan-Brouwer, 116
Antonio Ros Second fundamental form, 68, 76, 77, 207
Second variation of the length, 253
Minding, 241
Pappus, 163
Segre, 312 Rolle, 20
Senff, 2 Sard, 112, 132, 321
Separation sphere rigidity, 213
global, 116 turning tangent, 333
local, 108 Whitney-Grauenstein, 331
Translated by Sebastian Montiel Serret, 2 Theorema Egregium, 68, 205, 210
Sphere, 36, 45, 71, 79, 143, 145, 213, 225, Torsion, 2, 15
Translation Edited by Donald Babbitt 292
Steiner, 173
Torus, 37, 142, 283
Transversality
Stoker, 172, 175, 178, 273, 311 of curves, 22
Stokes, 135, 157, 311 of curves and surfaces, 109
Surface, 32, 34 of planes and surfaces, 113
convex, 174 of surfaces, 50, 51
of revolution, 55, 163
orientable, 70 Umbilical, 84, 85
Graduate Studies oriented, 70 Umlaufsatz, 333
parallel, 124, 145, 147
in Mathematics star-shaped, 57, 95, 177 Variation
strictly convex, 175 of a curve, 220
Volume 69 triply periodic, 56 of a surface, 193
umbilical, 84 Vertex of a curve, 346
with parallel s.f.f., 263 Volume, 147, 159
Surfaces
congruent, 207 Weierstrass, 311
diffeomorphic, 43 Whitney, 312, 331
isometric, 208 Wirtinger, 352
Wirtinger's inequality, 352
Tangent
American Mathematical Society line, 8, 319
TPn Providence, Rhode Island plane, 44
to a curve, 3
Theorem
Real Sociedad Matematica Espanola Alexandrov, 187
Bonnet, 181, 254
Madrid, Spain Brouwer, 288
Brouwer fix point, 163
Brouwer-Samelson, 118
Chern-Lashoff, 156
 Editorial Board of Graduate Studies in Mathematics
Index 375 David Saltman (Chair)
Walter Craig Nikolai Ivanov Steven G. Krantz

Isoperimetric inequality normal, 236 Editorial Committee of the Real Sociedad Matematica Espanola
in the plane, 326 tubular, 122
in the space, 195 of a curve, 321 Guillermo P. Curbera, Director
Newton, 1
Jackson, 312 Nirenberg, 173, 273
Luis Alias Alberto Elduque
Jacobi, 206 Normal Emilio Carrizosa Pablo Pedregal
Jacobian, 76, 277 line, 8, 319 Bernardo Cascales Rosa Maria Miro-Roig
absolute value, 57, 136 of a curve Javier Duoandikoetxea Juan Soler
on curves, 368 in the plane, 9
Jellett, 91, 172, 185 of a curve This work was originally published in Spanish by Proyecto Sur de Ediciones, S. L.
Jordan, 107, 116, 310, 320 in the space, 14 under the title Curvas Y Superficies © 1998.
of a surface, 46, 51 The present translation was created under license for the
Karcher, 173, 186 sections, 81 American Mathematical Society and is published by permission.
Kelvin, 135
Kneser, 312 Oresme, 1 Translated by Sebastian Montiel; translation edited by Donald Babbitt.
Kramer, 31 Orientation, 119 Figures by Joaqufn Perez; used with permission.
of curves, 319, 320
Lagrange, 205 of surfaces, 70 2000 Mathematics Subject Classification. Primary 53A04, 53A05, 53C40.
Lancret, 2, 22 preserving, 76
Laplace, 2 Oriented distance, 125
Laplacian, 351 Osculating
on 113, 165 circle, 21 For additional information and updates on this book, visit
on surfaces, 197 plane, 23 www.ams.org/bookpages/gsm-69
Lashoff, 156 Osserman, 313, 346
Legendre, 275 Oval, 328, 329
Leibniz, 1 Ovaloid, 172, 174
Length intersection with lines, 177 Library of Congress Cataloging-in-Publication Data
arc, 5, 8
Montiel, Sebastian, 1958-
of a curve, 7 Pappus, 311 [Curvas y superficies. English]
of a periodic curve, 315 Paraboloid, 37, 90, 174, 256 Curves and surfaces / Sebastian Montiel, Antonio Ros ; translated by Sebastian Montiel ;
Liebmann, 91, 172, 184, 205 hyperbolic, 80 translation edited by Donald Babbitt. - 2nd ed.
Liouville, 206 Parametrization, 35 p. cm. - (Graduate studies in mathematics ; v. 69)
Logarithmic spiral, 9 by curvature lines, 261 Includes bibliographical references and index.
change of, 39, 40 ISBN 978-0-8218-4763-3 (alk. paper)
Mobius strip, 71, 74 orthogonal, 96, 261 1. Curves on surfaces. 2. Geometry, Differential. 3. Submanifolds. I. Ros, A. (Antonio),
Malus, 2 Partition of unity, 143 1957- II. Babbitt, Donald G. III. Title.
Map Period QA643. M6613 2009
constant, 41 of a periodic curve, 314 516.3'62-dc22 2009008149
differentiable, 40 Poincare, 276, 295, 301
identity, 41 Point
inclusion, 41 critical, 48
Massey, 173, 273 elliptic, 83, 87
Copying and reprinting. Individual readers of this publication, and nonprofit libraries
acting for them, are permitted to make fair use of the material, such as to copy a chapter for use
Maxwell, 135, 311 hyperbolic, 83, 87 in teaching or research. Permission is granted to quote brief passages from this publication in
Measure zero, 126, 131 of double intersection, 115 reviews, provided the customary acknowledgment of the source is given.
on a surface, 142 of first contact, 115 Republication, systematic copying, or multiple reproduction of any material in this publication
Meusnier, 2, 68 parabolic, 83 is permitted only under license from the American Mathematical Society. Requests for such
Milnor, 312, 345 planar, 83 permission should be addressed to the Acquisitions Department, American Mathematical Society,
Minding, 172, 205, 241 umbilical, 83 201 Charles Street, Providence, Rhode Island 02904-2294, USA. Requests can also be made by
Minkowski, 172, 189, 311, 325 Poisson, 2 e-mail to reprint-permissionmams . org.
Monge, 1, 31, 67 Poncelet, 2 © 2005 by the American Mathematical Society.
Monodromy, 196 Principal curvatures, 78 Second Edition © 2009 by the American Mathematical Society. All rights reserved.
Monotone convergence, 148 Principal directions, 79, 84 The American Mathematical Society retains all rights
Mukhopadhyaya, 312 Projection, 125 except those granted to the United States Government.
Myers, 182 Ptolemy, 31 Printed in the United States of America.
Puiseux, 241
0 The paper used in this book is acid-free and falls within the guidelines
Neighbourhood established to ensure permanence and durability.
coordinate, 35 Quadric, 37, 80
Visit the AMS home page at http://www.ams.org/
10987654321 141312111009
 374 Index

Distance, 239 Four-petal rose, 313

Divergence, 157, 322 Fourier, 2
on surfaces, 184 Frenet, 2, 15
Domain dihedron, 9
inner, 118, 174 trihedron, 14
inner of a curve, 320 Fubini, 146
outer, 118 Function
outer of a curve, 320 differentiable, 40
regular, 159 distance, 41, 88
Dominated convergence, 148 height, 41, 85, 87, 301
Dupin, 2 implicit, 36, 49
integrable, 139
Efimov, 173 inverse, 58
Ellipsoid, 36, 37, 80, 174 on a sphere, 150
This book is dedicated to Eni and Juana, our wives, and to our sons and Equations support, 95, 177
Darboux, 240
daughters. We are greatly indebted to them for their encouragement and Frenet, 10 Gauss, 2, 31, 68, 157, 172, 203, 210, 232,
support, without which this book would have remained a set of teaching Euler, 1, 31, 67, 206, 312 275, 295, 301
notes and exercises. characteristic, 295, 301, 304 lemma, 232
Evolute, 21 map, 68, 76
Evolvent, 21 Geodesic, 1, 223
Existence and uniqueness ball, 232
theorem of, 17, 227 existence, 228
Exponential map, 230 homogeneity, 229
minimizes locally, 233
Fary, 312, 345 Germain, 68
Fenchel, 312, 342 Gluck, 312
Field Gradient, 162
conformal on a sphere, 292 on surfaces, 197
conservative, 351 Graph, 35, 53, 71
non-degenerate, 289, 291 Grauenstein, 312, 331
normal, 69, 119 Green, 157
on 112, 322 Gromov, 311
on 113, 157 Group of isometries, 209
on a surface, 69
tangent, 69, 257 Helein, 311
First fundamental form, 68, 207 Hadamard, 172, 175, 178, 311
First integral, 259 Hartman, 173, 273
Flow of a field, 258 Heintze, 173, 186
Formula Heintze-Karcher's inequality, 186
area, 152 Helicoid, 91, 239
Bertrand-Puiseux, 241 Helix, 4, 16, 19, 22
change of variables, 144 Herglotz, 205, 217, 312
Crofton, 343 Hessian, 86, 87
Diquet, 241 Hilbert, 91, 172
Gauss-Bonnet, 295, 301 Hopf, 206, 238, 312
Herglotz, 217 Hurwitz, 311
integration in polar coordinates, 147 Hyperboloid, 37, 44
length, 321
Poincare-Hopf, 301 Index
variation of area, 193 of a vector-field, 288
variation of length, 222 rotation, 329
variation of volume, 194 Integral
Formulas onSxR, 137
Frenet, 15 on a surface, 139
Green, 351 Isometry, 208
Minkowski, 184 local, 207
Index

Alexandrov, 172, 187 Crofton, 312, 343

Ampere, 2 Curvature
Arc length, 1 centre, 21
Archimedes, 161 Gauss, 68, 77
Area, 141 mean, 68, 77
of a curve, 1
Bartels, 2 in R3, 14
Bernoulli, D., 1 plane, 10
Bernoulli, John, 1, 206 radius, 1
Bernoulli, Joseph, 2 total, 330
Bertrand, 68 Curve, 3
Bianchi, 31 convex, 328
Binormal, 2, 14 coordinate, 35
line, 22 integral, 258
Blaschke, 172 knotted, 345
Bonnet, 182, 204, 276, 295, 301 on a surface, 42
Brouwer, 108, 116, 118, 161, 288, 310 p.b.a.l., 9
Brunn, 173, 189, 311, 325 parallel, 321
Brunn-Minkowski inequality, 189, 325 parametrized, 313
periodic, 314
regular, 8
Catenoid, 95, 239 simple, 38, 313, 318
Cauchy, 2, 32, 205 Cylinder, 37, 44, 71, 80, 95, 225
Chain rule, 48
Chasles, 68
Chern, 156 Darboux, 2, 68, 240
Circumscribed circle, 346 trihedron, 240
Clairaut, 2, 31, 240 Degree
Cohn-Vossen, 205, 218 between curves, 365
Comparison between surfaces, 277
of curves, 22 local, 281
of lines and surfaces, 111, 112 Dependence of parameters, 149
of planes and surfaces, 113 Descartes folium, 313
of surfaces, 96 Diffeomorphism, 43
Cone, 38, 39 local, 54
Congruence, 207 Differential, 46, 48
Connected component, 36 Diquet, 241

373
 372 Bibliography

[19] D. J. Struik, Geometria Diferencial Cldsica, Aguilar, 1970.

[20] J. A. Thorpe, Elementary Topics in Differential Geometry, Springer-Verlag, 1979.
[21] E. Vidal Abascal, Introduccidn a la Geometria Diferencial, Dossat, 1956.
[22] A. Wallace, Differential Topology, Benjamin, 1968.

Contents

Preface to the Second Edition xi

Preface to the English Edition xiii
Preface xv
Chapter 1. Plane and Space Curves 1

§1.1. Historical notes 1

§1.2. Curves. Arc length 2

§1.3. Regular curves and curves parametrized by arc length 8
§1.4. Local theory of plane curves 9
§1.5. Local theory of space curves 13
Exercises 19
Hints for solving the exercises 23

Chapter 2. Surfaces in Euclidean Space

§2.1. Historical notes
§2.2. Definition of surface
§2.3. Change of parameters
§2.4. Differentiable functions
§2.5. The tangent plane
§2.6. Differential of a differentiable map
Exercises
Hints for solving the exercises
Chapter 3. The Second Fundamental Form 67

vii
 Viii Contents

§3.1. Introduction and historical notes 67

§3.2. Normal fields. Orientation 69
§3.3. The Gauss map and the second fundamental form 76
§3.4. Normal sections 81
§3.5. The height function and the second fundamental form 85
Bibliography §3.6. Continuity of the curvatures
Exercises
88
94
Hints for solving the exercises 97

Chapter 4. Separation and Orientability 107

§4.1. Introduction 107
[1] M. Berger, B. Gostiaux, Differential Geometry: Manifolds, Curves and Surfaces, §4.2. Local separation 108
Springer-Verlag, 1988. §4.3. Surfaces, straight lines, and planes 111
[2] M. P. do Carmo, Differential Geometry of Curves and Surfaces, Prentice-Hall, §4.4. The Jordan-Brouwer separation theorem 116
1976.
§4.5. Tubular neighbourhoods 120
[3] L. A. Cordero, M. Fernandez, A. Gray, Geometria Diferencial de Curves y Super-
ficies, Addison-Wesley Iberoamericana, 1995. Exercises 125
[4] A. Goetz, Introduction to Differential Geometry, Addison-Wesley, 1970. Hints for solving the exercises 126
[5] H. Hadwiger, Vorlesungen caber Inhalt, Oberflache and Isoperimetrie, Springer- §4.6. Appendix: Proof of Sard's theorem 131
Verlag, 1957.
[6] H. Hadwiger, D. Ohmann, Brunn-Minkowskischer Satz and Isoperimetrie, Math. Chapter 5. Integration on Surfaces
Z., 66 (1956), 1-8.
§5.1. Introduction
[7] C. C. Hsiung, A First Course in Differential Geometry, Wiley-Interscience, 1981.
§5.2. Integrable functions and integration on S x R
[8] W. Klingenberg, Curso de Geometria Diferencial, Alhambra, 1978.
[9] D. Lehmann, C. Sacre, Geometrie et Topologie des Surfaces, Presses Universitaires
§5.3. Integrable functions and integration on surfaces
de France, 1982. §5.4. Formula for the change of variables
[10] R. S. Millman, G. D. Parker, Elements of Differential Geometry, Prentice-Hall, §5.5. Fubini's theorem and other properties
1977.
§5.6. Area formula
[11] J. W. Milnor, Topology from the Differentiable Viewpoint, University of Virginia
Press, 1965. §5.7. The divergence theorem
[12] S. Montiel, A. Ros, Curves y Superficies, Proyecto Sur de Ediciones, 1997, 1998. §5.8. The Brouwer fixed point theorem
[13] B. O'Neill, Elementary Differential Geometry, Academic Press, 1966. Exercises
[14] R. Osserman, The four-or-more vertex theorem, Amer. Math. Monthly, 92 (1985), Hints for solving the exercises
no. 5, 332-337.
[15] A. V. Pogorelov, Geometria Diferencial, Mir, 1977. Chapter 6. Global Extrinsic Geometry
[16] N. Prakash, Differential Geometry, Tata McGraw-Hill, 1981. §6.1. Introduction and historical notes
[17] M. Spivak, A Comprehensive Introduction to Differential Geometry, vols. 3 and §6.2. Positively curved surfaces
5, Publish or Perish, 1979.
§6.3. Minkowski formulas and ovaloids
[18] J. J. Stoker, Differential Geometry, Wiley-Interscience, 1969.
§6.4. The Alexandrov theorem
371 §6.5. The isoperimetric inequality
Contents ix
370 9. Global Geometry of Curves

Exercises 195
Proposition 9.74. Let b : ][8 -* S1 be a differentiable map. Then b is
Hints for solving the exercises 198
periodic with period A> 0 if and only if there exists an integer n E 7G such
Chapter 7. Intrinsic Geometry of Surfaces 203 that
fs+A
§7.1. Introduction 203 27rn = (Jaccb)(t)dt
§7.2. Rigid motions and isometries 206
J
S
for each s E R. In this case, the number n is the degree of the A-periodic
§7.3. Gauss's Theorema Egregium 210 map b.
§7.4. Rigidity of ovaloids 214
Proof. If b is A-periodic, it is enough to take n E Z as its degree as defined
§7.5. Geodesics 219 before and to keep in mind the last equality in the statement. Conversely,
§7.6. The exponential map 229 suppose that there exists n E Z such that
s+A

f
Exercises 239
Hints for solving the exercises 242
§7.7. Appendix: Some additional results of an intrinsic type 252 Then, defining a map b : ][8 -i ]E8 as in (9.4), we have that
Chapter 8. The Gauss-Bonnet Theorem 275 (s+A)-(s)=2irn
§8.1. Introduction 275 for all s E R. Thus, (s + A) = A)) =p(a(s)) _ q(s), that is, b is
§8.2. Degree of maps between compact surfaces 276 A-periodic.
§8.3. Degree and surfaces bounding the same domain 284
§8.4. The index of a field at an isolated zero 288
§8.5. The Gauss-Bonnet formula 295
§8.6. Exercise: The Euler characteristic is even 304
Exercises: Steps of the proof 305

Chapter 9. Global Geometry of Curves 309

§9.1. Introduction and historical notes 309
§9.2. Parametrized curves and simple curves 313
§9.3. Results already shown on surfaces 319
§9.4. Rotation index of plane curves 329
§9.5. Periodic space curves 338
§9.6. The four-vertices theorem 346
Exercises 350
Hints for solving the exercises 353
§9.7. Appendix: One-dimensional degree theory 365

Bibliography 371

Index 373
9.7. Appendix: One-dimensional degree theory 369

Conversely, if this last equality holds,

(9.3) =gds [1 - ()J _ -2[(',) + (')J

ds
Then, since kbI2 = II2 = 1, we take derivatives and obtain
=0
and, thus, q5' and '1' are proportional to Jq5 and J'1', respectively. Then
q' = det(b,
Substituting these equalities into (9.3) and using the fact that the two Ja-
cobians are equal, we get

and, thus, the distance kb - '/i is constant. If the two curves b and /'
coincided at a point, then they would be equal. But, in any case, we can
find a rotation A E 80(2) such that b and /' coincide at a given point of
[a, b]. We complete the proof by applying the above to the curves Aq5 and
b. O

We now proceed by supposing that b : [a, b] -* S1 is a differentiable

curve. Define another map b: [a, b] -* I[8 by the equality
s
(9.4) i(s) s=c+f(Jac)(t)dt,
where so E [a, b] is arbitrary and c E I[8 will be determined later. Thus
= p o b is a differentiable map from [a, b] to the circle satisfying
_ (cos ¢, sin Q)' _ ¢'(- sin , cos ).
Hence, since q5' = Jac b, we have that
Jac 'b = Jac b.
By Lemma 9.73, zb = p0 b and b differ by a rotation. But b(so) = p(c) _
(cos c, sin c). Thus, if we choose c E ]E8 so that p(c) _ (so), we arrive at the
equality. That is, (9.4) is an explicit expression for any lift of a differentiable
curve to the circle. From this expression, it follows that the degree of a closed
differentiable curve b is given by
a
fdeg = 2ir o
Jac - so
Jac ] =
2ir a
Jac .

Using this, we can prove the following result.


Preface to the Second
Edition
This second edition follows the plan and the arrangement of topics of the
first edition. Some proofs have been revised or simplified. A number of
typos and mistakes in the exercises and in the hints for solving them have
been corrected. We are very grateful to the publishing staff of the Ameri-
can Mathematical Society and of the Royal Spanish Mathematical Society,
especially to Edward Dunne and Guillermo Curbera, for their willingness
to undertake a second edition and for their support. We also thank Donald
Babbitt for his effort and considerable improvment of both the English and
the mathematics in this second edition.
Sebastian Montiel
Antonio Ros
10 February 2009
xi
368 9. Global Geometry of Curves
that p o b - b with an initial condition given, for example, at 0 E R. Thus,
each restriction bs : [s, s + A] -* ll8 is clearly a lift of the curve 3. Hence
deg s = Z [(s + A) - - [(5 + A) -
for each s E R. On the other hand, since b is A-periodic,
A)) _ (s + A) _ (s) =p(a(s))
and, thus, (s + A) - 7.(s) is a continuous function in $ taking only integer
values. Therefore, it is constant. Consequently, deg bs is independent of s.
Then, we define the degree of the A-periodic map b : I[8 -* S1 by
deg = deg3 = [(s + A) -
for any $ E R. Of course, the properties of this degree of periodic maps are
similar to those of closed curves. We list them in the following exercises.
Exercise 9.70. Let b, ib : ]E8 -i S1 be two A-periodic continuous maps. A
homotopy between them is, by definition, a continuous map F : ll8 x [0, 1] -i
S1 such that F(s, 0) _ (s), F(s,1) _ sb(s) for all s E ]E8, and $ E-+ F(s, t)
is an A-periodic map for all t E [0, 1]. Using a lift of the homotopy F to I[8,
show that deg ¢ = deg 'ib.
Exercise 9.71. Suppose that b, /' Il8 -* S1 are two A-periodic continuous
:
maps with the same degree. Prove that F = p o F is a homotopy between
them, where F is the map defined by F(s, t) _ (1 -
t) E I[8 x [0, 1]. Then, the homotopy is differentiable if and only if the
original maps are differentiable.
Exercise 9.72. Prove that the degree of periodic continuous maps to the
circle remains unchanged through rotations of ]E82 and changes sign when we
compose with plane symmetries.
Now suppose that the curve b : [a, b] -* S1 C ]E82 is differentiable; it is
enough to assume piecewise Ci. We define the Jacobian of b, in agreement
with Exercise (13) of Chapter 2, Exercise 3.15, and Section 5.2, as the
function Jac b: [a, b] -* I[8, given by
Jac b = q5') = (Jq5, q5'),
where J : ]E82 1182 is the 90 degree counterclockwise rotation of the plane.
Lemma 9.73. Let b, z/,' : [a, b] -* S1 be two differentiable curves. Then b
and /' differ by a rotation of I[82 if and only if Jac b = Jac /'.
Proof. The necessary condition is clear since, if A E 80(2) and /,' =
we have
Jac z/ = det('I', z/') = det(Ab, Ab') = det(b, b') = Jac b.
9.7. Appendix: One-dimensional degree theory 367

The property shown in Proposition 9.66 will allow us to define the degree
of a continuous curve into the unit sphere b: [a, b] -* Si, provided that it is
closed in the sense of having the same endpoints. Let b: [a, b] -* ]E8 be any
lift of b. Then, we have
p((a)) = b(a) = cb(b) =
and, thus, (a) and fi(b) differ by an integer number of turns. The integer
deg = -[(b) - (a)J E 7G

is called the degree of b. Perhaps the most remarkable property of this

degree is the invariance under homotopies with cb(a) _ fi(b), which is also
a consequence of Proposition 9.66 above. In fact, if F : [a, b] x [O, l] -* 51
is a continuous homotopy between two closed curves bo and cb1 defined on
[a, b] with the same origin bo(a) _ b1(a) E Si, we consider its lift F
[a, b] x [0,1] -* I[8 with an arbitrary initial condition. Then, F(s, 0) and
F(s,1) are lifts for the curves bo and i, respectively, and, moreover, F(a, t)
and F(b, t) are constant curves. Hence
deg o = Z [(b,0) - fr(a,0)] = Z [fr(b, 1) - fr(a, l)] = deg 1.
Exercise 9.67. Let b: [a, b] -* Si be a closed continuous curve in the circle,
and let G E 80(2) be a rotation of angle 8 E R. Prove that, if b is a lift of
b, then b +0 is a lift of G o b. Deduce that deg(G o b) - deg b. In addition,
prove that, if S E 0(2) is a symmetry, then deg(S o b) - - deg b.
Exercise 9.68. Let h : [c, d] -* [a, b] and b : [a, b] -* Si be a continuous
function and a continuous curve in the unit circle. Show that if b is a lift of
b, then ooh is a lift of ooh. Deduce from this that, if h is a homeomorphism,
then deg b, depending upon whether h(c) = a and h(d) = b
or h(c) =band h(d) = a.
Exercise 9.69. Let b, b: [a, b] - 1 be two continuous curves in the circle
such that (a) _ 'b(a) _ fi(b) _ 'b(b) and such that deg'b. Show
that, if b and b are lifts of b and b, respectively, with the same initial
condition, then F = p o F is a homotopy between them, where F(s, t) _
(1 - t) E [a, b] x [0, 1]. This homotopy is clearly
differentiable, if the original curves are differentiable.
From the above, it is not difficult to define the degree of a periodic
continuous map b:III - * 51 with period A> 0 as follows. For each s E III, we
obtain, by restriction, a closed curve S : [s, s+A] - Si in the circle, to which
we can associate, according to the discussion above, a degree deg S E Z.
Now then, it is not difficult to see, using Proposition 9.66 and an induction,
that there exists a unique continuous map c : III -* III lifting b, that is, such
 366 9. Global Geometry of Curves

a point ri E S1 such that a([ti, t2+1]) C 'j Then a(to) = a(a) = qo E V''° .
Since so E III is a point such that p(so) = qo, we have
So E p-1 (V ''o) = UT0

jEN
and thus so belongs to a unique U'0,0. We define ao : [to, ti] -* III by
Preface to the English N
ao = (p1jroj0)1 o a,

Edition which is continuous and satisfies &o(a) = so and p o ao = a on [to, t1]. The
other end of the lifted curve satisfies p(ao (t1)) = a(t1) E a([t1, t2]) C V''1.
Therefore,
&o (ti) E p-l (VT1) =
jEN
Thus, o (t1) belongs to a unique term of this union, which we denote by
Ujl. We define a1 : [t1, t2] -* S by
This book was begun after a course on the classical differential geometry )-1
o a,
a1 = (pIui
of surfaces was given by the authors over several years at the University of
Granada, Spain. The text benefits largely from comments from students as which is continuous as well and satisfies p o a1 = a and, thus, p(cx1(t1)) =
well as from their reaction to the different topics explained in the lectures. a (ti) = p(ao (t 1)) E V''1, which is the homeomorphic image of Uj' through
Since then, different parts of the text have been used as a guide text in p. Then
several undergraduate and graduate courses. The more that time passes, the &o(t1) =al(ts).

more we are persuaded that the study of the geometry of curves and surfaces We repeat this process finitely many times to obtain k continuous curves
should be an essential part of the basic training of each mathematician and, &o, &1> , &k-1 defined, respectively, on the successive subintervals of the

at the same time, it could likely be the best way to introduce all the students
concerned with differential geometry, both mathematicians and physicists or
engineers, to this field.
The book is based on our course, but it has also been completed by
including some other alternative subjects, giving on one hand a larger co-
herence to the text and on the other hand allowing the teacher to focus the
course in a variety of ways. Our aim has been to present some of the most
relevant global results of classical differential geometry, relative both to the
study of curves and that of surfaces.
This text is indeed an improved and updated English version of our
earlier Spanish book Curvas y Superficies [12], published by Proyecto Sur
de Ediciones, S. L., Granada, in 1997, and republished in 1998. We are
indebted to our colleague Francisco Urbano for many of the improvements
and corrections that this English version incorporates. We also recognize
a great debt to Joaquin Perez, another colleague in the Departmento de
xiii
partition and such that
p ° az - al[ti>ti+1] and ai(ti) = ai-i(tz)
for i = 0, ... , k - 1. Finally, we define a : [a, b] -i ll8 by
a(t) = ai(t) if t E [t,t2+1].
This a is clearly continuous by construction and satisfies p o a = a and
a(a) = sp.
Turning to uniqueness, suppose that there are two continuous curves
,6, 'y: [a, b] -* ]E8 with the required conditions. We denote by A the subset
of the interval [a, b] formed by those t such that ,6(t) = ry(t). It is trivial to
see that A is closed and non-empty since ,Q(a) _ -y(a) = so. We now show
that it is also open. In fact, let to E A. Then ,Q(to) = 'y(to) and p(,Q(to)) _
p(y(tp)) = a(to) E S1. Since p is a local homeomorphism, there exists an
open neighbourhood U of ,6(to) = 'y(to) in ]E8 and another neighbourhood V
of a(to) in S1 such that p is a homeomorphism of U to V. Since ,6 and ry are
continuous, there is a neighbourhood I of to such that ,6(I), 'y(I) C U. Then
Hence I C A and A is open. For maps defined on a rectangle, a
completely similar proof works. O
9.7. Appendix: One-dimensional degree theory 365
xiv Preface to the English Edition

Exercise (17): The length of such a curve must be greater than or equal
Geometria y Topologia of our university, for creating the sixty-four figures
to that of the unit circle, that is, 2ir and equality holds only for this circle.
In fact, if L < 2ir, integrating the inequality and taking the definition of accompanying the text.
rotation index into account, It is also a pleasure to thank Edward Dunne, Editor at the American
L
Mathematical Society, who proposed the possibility of translating our Span-
0 < i(a) = 2J k(s) ds < Z <1 ish text, giving us the opportunity of exposing our work to a much wider
audience. We also owe a great debt of gratitude to our Production Editor,
and thus i(a) is 0 or +1 and either of the two equalities forces the curvature Arlene O'Sean, for substantially improving the English and the presentation
to be constantly 0 or +1. of the text.
The authors gratefully acknowledge that, while this book was being
9.7. Appendix: One-dimensional degree theory written, they were supported in part by MEC-FEDER Spanish and EU
Grants MTM2004-00109 and MTM2004-02746, respectively.
In the study of the fundamental group of the circle Si, it is standard to
utilize the map p : ]E8 -* Si, given by the expression S. Montiel and A. Ros
p(s) _ (coss,sins), Vs E 118, Granada, 2005
which is differentiable as a map from I[8 to ]E82 and has the following funda-
mental property, which is characteristic of covering maps.
Proposition 9.66. Let a : [a, b] -* Si be a continuous curve in the circle,
and let so E ll8 such that p(so) = a(a). Then, there is a unique continuous
curve a : [a, b] -* ][8 such that
a(a) = so and p o a = a.
We say that & is a lift of a and that so is its initial condition at a. Moreover,
the same result is true for continuous maps defined on a rectangle [a, b] x
[c, d] C I[82, where the initial condition is now taken at the point (a, c).

Proof. It is clear that, for each q = (cos 8, sin 8) E 51, the open neighbour-
hood Vq = p((8 - ir, 8 + ii)) of q satisfies the property that p-1(V9) is the
disjoint union of the open subsets (0+ (2k -1)ir, 0+ (2k+ 1)ir), with
k E 7G, of R. Furthermore, the map p is a homeomorphism onto its image
when it is restricted to each of these intervals. Thus, the family {V}, with
q varying along Si, covers Si. Hence, if a : [a, b] -* 1 is a continuous curve
with a(a) = qo, we deduce that {a-1(V4)}, with q E Si, is a cover by open
subsets of the interval [a, b]. Let 6 > 0 be the Lebesgue number of this cover.
Then
A C [a, b],
a(A) C V9 for some q E S1.
diam A < S
Consider a partition {to = a < ti < < tk = b}, with norm less than
this S, and lift the curve a on each subinterval of the partition. We want to
assemble the diverse lifts, taking care of preserving the continuity of the final
curve. By the choice of the partition, there exists, for each i = 0, ... , k - 1,
 364 9. Global Geometry of Curves

for some function h of class Cl, which exists according to Exercise (12). A
direct calculation from this expression of f leads to the equation
z
f'(t)2 - (B'' A)2 f (t)2 = h'(t)2 sine B At - A)

Preface -}- dt (B_At2sinBAtcosBAt).

Integrating between A and B and using the fundamental theorem of calculus,
we get

A
f'(t)2 dt - (B ZA)2 A
f(t)2 dt =
A
h'(t)2 sine B A(t - A) dt > 0,

Over the past decades, there has been an outstanding increase in interest and equality holds if and only if h'(t) = 0 for each t E [A, B], that is, if
in aspects of global differential geometry, to the neglect of local differential h(t) =a for some a E ]E8.
geometry. We have wished, for some years, to present this point of view in a
text devoted to classical differential geometry, that is, to the study of curves Exercise (14): Let Rl be the line determined by the points a(0) and
and surfaces in ordinary Euclidean space. For this, we need to use more a(L/2), which are different since a is injective on [0, L). Hence, it is clear
sophisticated tools than those usually used in this type of book. Likewise, that a(0) and a(L/2) lie on Rl. Now, set
some topological questions arise that we must unavoidably pay attention to. a(L/2) - a(0)
We intend, at the same time, that our text might serve as an introduction 1
Ia(L/2) - a(0)1
to differential geometry and might be used as a guidebook for a year's study
of the differential geometry of curves and surfaces. For this reason, we find The function f : [0, L/2] -> ]E8, given by
additional difficulties of a pedagogical nature and relative to the previous f(s) _ (a(s + L/2) - a(0), el), Vs E [0, L/2],
topics that we have to assume a hypothetical reader knows about. These
different aspects that we want to consider in our study oppose each other on is continuous and satisfies
quite a number of occasions. We have written this text by taking the above f(0) _ I- a(0)I > 0 and f(L/2) _ -Ia(L/2) - a(0)I <0.
problems into account, and we think that it might supply a certain point of
equilibrium for all of them. Thus, there exists so E (0, L/2) such that the straight line R2 passing
To read this book, it is necessary to know the basics of linear algebra through a(so) and a(so + L/2) is perpendicular to Rl.
and to have some ease with the topology of Euclidean space and with the
calculus of two and three variables, together with an elementary study of Exercise (15): By our choice of the frame of the plane, we can apply the
the rudiments of the theory of ordinary differential equations. Besides these Wirtinger inequality of Exercise (13) to each of the four integrals into which
standard requirements, the reader should be acquainted with the theory of we have decomposed

f
L
Lebesgue integration. This novelty is one of the features of our approach. Ia(s)12 ds.
Even though students of our universities usually learn the theories of inte-
gration before starting the study of differential geometry, in the past, these Each of them is less than or equal to L2/4ir2 times the corresponding integral
theories have been used only in a superficial way in the introductory courses of the derivative. Grouping the four resulting integrals, we obtain
fL
Ia(s)12 ds < L Ia'(s)I ds =
J
xv as required.
Hints for solving the exercises 363
xvi Preface

and, thus, by the fundamental theorem of calculus,

((Vf)(p),v) =
f1

J
t (tX(tp),v) dt = (X(p),v)
to this subject. Instead, we will use integration as a powerful tool to obtain
geometrical results of a global nature.
The second basic difference in our approach, with respect to other books
for all v E I[82. Therefore X= V f. on the same subject, is the particular attention that we will pay, on several
occasions throughout the text, to some questions of a topological character.
Exercise (10): The first equality of Exercise (8) of this chapter, applied The understanding of such questions will appear sometimes as a goal in itself
to the case f = g, and keeping in mind that f vanishes along C and is and often as an essential step to obtaining results of a purely geometrical
harmonic, gives us type.

f 1vf12=o.
Thus, V f vanishes on 11, and hence f must be constant. Since it vanishes
A third remarkable characteristic of this course on classical differen-
tial geometry, this time of a pedagogical nature, although it also has some
consequences in the theoretical development of the text, is the authors' de-
along C, it has to be identically zero on 11. termination to use a language free of coordinates whenever possible. We
think that, in this way, we get a clearer statement of the results and their
proofs.
Exercise (12): We define a function h on the interval [A, B] by
We would like to thank Manuel Ritore for his work on drawing the
f(t)/sinBA(t_A) if t e (A, B), figures accompanying the text. We would also like to thank someone for
having typed it and for having decoded the numerous and endless error
messages of W[FX when it was processed, but, in truth, we cannot. Instead,
h(t) _ it is easy to recognize our indebtedness to the students of the third year of

/
mathematics at the University of Granada, who, for three years, brought
about, with their questions and misgivings, new proofs and points of view
relative to diverse problems of classical differential geometry. We are also
We see that h is continuous at the endpoints by using the fact that f(A) indebted to earlier textbooks on this subject, mainly to those cited in the
1(B) = 0 and L'Hopital's rule. Moreover, by definition, we have bibliography and in a very special way to the book Differential Geometry
of Curves and Surfaces ([2]) of M. P. do Carmo. With it, we passed from
f(t)=h(t)sinBA(t_A), `dt E [A, B]. ignorance to surprise with respect to the knowledge of differential geometry.
Last, we point out that, for the sake of self-containedness, we have in-
On the other hand, if t e (A, B), we have cluded answers to many of the exercises that we posed within the main text
f (t) sinB A (t - A) - B A f (t) cos B A (t - A) and in the list of exercises that we have included in each chapter. The ex-
ercises that we have chosen to answer are not necessarily the most difficult,
h (t) _ r but those more significant from our point of view. We have marked them
sine B-A(t - A)
with a vertical arrow, like T, at the beginning of the exercise.
Consequently, using L'Hopital's rule again, we have
. ,
lim h (t) =
B-A f (A) ,
and lim h (t) = -
B-A f (B), S. Montiel and A. Ros
Granada, 1997
and thus h is of class C1.

Exercise (13): Set

f(t) = h(t) sin B A(t - A), t e [A, B],
 362 9. Global Geometry of Curves

Chapter 1
to r. That is, f(s) > r for all s E I[8. Now utilizing the expression for the
area of S2 that we gave in previous exercises, we have
L
A(11) = -- I0 (c(s),N(s))ds -L.
If equality occurs, the function f(s) takes the constant value r. Differentiat-
Plane and Space ing, using the Frenet formulas, and recalling that the trace of a is an oval, we
obtain that (cr(s), T(s)) vanishes identically. From this, all the normal lines
Curves of a pass through the origin. We complete the exercise by using Exercise
(5) of Chapter 1.

Exercise (6): The necessary condition can be easily checked. In fact, in

the case where X is conservative, that is, X = V f for a certain differentiable
function f defined on R2, it is enough to use Exercise 9.30. Now suppose
that the vector field X defined on the whole of the plane satisfies div JX = 0.
1.1. Historical notes By definition, this means that
From the very beginning of calculus, LEIBNIz (1646-1716) and NEWTON ax 2 ax 1
(1643-1727) dealt with problems of plane curves. In 1684, LEIBNIz tried ax ay
to define the curvature of a plane curve in his Meditatio nova de natura on II82, and this is equivalent to
anguli contactus et osculi, although, in the 14th century, the bishop of the
French city of Lisieux, NICOLAS D'ORESME, had also tried to do it, at ((dX)(v),w) _ (v,(dX)(w))
least in a qualitative way, in his Tractatus de configurationibus qualitatum et for all v, w e I[82. We define a function f : I[82 -i ][8 by
motum. In fact, the first author to find the correct definition of curvature was
i
LEONHARD EULER (1707-1783), who initiated so-called intrinsic geometry
in 1736 by introducing the intrinsic coordinates of a plane curve: arc length
f(p) = f (x(tp),p) dt
0
and radius of curvature instead of curvature. The definition of curvature for each p e ][82. Since the map (t, p) F- (x(tp), p) is differentiable, the
as the rate of variation of the angle that the tangent line makes with a function f that we have just defined is also differentiable. Now, if p, v e II82,
given direction is due to the great mathematician himself. The theory of we have
closed curves with constant width was also initiated by Euler. In addition, 1
d d
he made, together with J OHANN and DANIEL BERNOULLI (1667-1748 and
1700-1782, respectively), contributions to the theory of surfaces. In fact,
((Vf)(p),v) - ds s=o
f(p+sv)= ds
(x(t(p + sv)), p + sv) dt
he was the first to characterize geodesics as solutions of certain differential
i
equations, and he showed that a material point that is constrained to move _ {((dx)(tv),p) + (x(tp),v)}dt.
on a surface and is not subjected to any other forces must move along one
of its geodesics. Using the hypothesis, we obtain
GASPARD MONGE (1746-1818), an engineer in the French army, cre- f1
ated the theory of space curves in 1771, motivated by some problems of ((Vf)(p), v) = J (t(dX)t(p) + X (tp), v) dt.
a practical nature military fortification. However, his first publications
On the other hand,

1 t (tx(tp),v) _ (x(tp),v) + t ((dX)tp(p), v) ,

Hints for solving the exercises 361 2 1. Plane and Space Curves

But, if p e SZ and v e IIBz, then, by definition, on the subject did not appear until 1785. MONGE had a clear precedent:
the Recherches sur les courbes a double courbure, which ALEXIS CLAIRAUT
(dcbt)(v) = v + t(dX)(v).
(1713-1765) wrote when he was only sixteen. In this treatise, CLAIRAUT had
From this, a simple calculation gives us already made a "famous" attempt to study the geometry of space curves.
det(dcbt)p = 1 + t(div X)(p) + tz det(dX)p, In 1795, MONGE published his results on space curves in a work that has
been looked upon as the first book on differential geometry: Applications de
and, now, integrating over St, we get l'arnalyse a la geometrie. The importance that this text attained can be un-
derstood by keeping in mind that its fifth edition appeared only thirty years
A(cbt(SZ)) = A(12) + t f div X + tz f Jac X. after the death of MONGE. Among his students, we find such famous math-
ematicians and physicists as LAPLACE, MEUSNIER, FOURIER, LANCRET,
Taking the derivative with respect to t at t = 0 completes the exercise.
AMPERE, MALUS, POISSON, DUPIN, PONCELET, and OLINDE RODRIGUES.
The analytical methods for studying curves, already introduced by EULER,
Exercise (2): We put the origin of the plane at the centre of the disc where were now applied by MONGE in exactly the same way as we do it today. In
C is included. Thus, we have J< r for all s E R. On the other hand, fact, it was AUGUSTIN CAUCHY (1789-1857) who was the first to define no-
we showed in the solution to Exercise 9.31 that tions such as curvature and torsion in a truly modern manner in his Lecons
fL
sur l'application du calcul infinitesimal a la geometrie.
A(11) = -- I (c(s),N(s))ds.
It was not until 1846 that the first complete compilation of results, writ-
From the Schwaxz inequality, we obtain ten in a more comprehensible manner than usual in this period, was pub-

A(11)< 2 IZL.
L lished as a treatise on space curves by BARRE DE SAINT-VENANT (1791-
1886). He was the creator of the term binormal. The unification of all previ-
ous efforts is due to F. FRENET and J. SERRET (1816-1868 and 1819-1885,
If equality occurred, among other things, we would have that a(s)I = r for respectively), who independently discovered the so-called Frenet-Serret for-
all s e R. That is, C would be contained in the circle of radius r and centre mulas around 1847 and 1851, respectively. However, these equations had
at the origin. Since they are both compact simple curves, C and this circle been written already in 1831 by the Estonian mathematician KARL E.
must coincide. SENFF (1810-1917), a student of M. BARTELS, who, in turn, had been a stu-
dent at the same time as GAUSS at Gottingen. The contributions by SENFF
Exercise (3): The same expression that we used in the solution to Exercise were probably ignored when they appeared because they did not use the lan-
(2) for the area of St allows us to write guage of linear algebra. The French authors De Saint-Venant, Frenet and
1/2 Serret belonged to the group of mathematicians associated with the Journal
1
A(St) < 2
L
(s) ds < 2
1
(o
\J
L
Ids I\/ de Mathematiques pores et appliquees, published without interruption since
1836, the year in which JOHANN BERNOULLI (1809-1882) founded it. Fi-
where the last inequality is nothing more than the Schwarz inequality in nally, we have to mention GASTON DARBOUX (1824-1917), who started the
LZ ([0, L]). If equality occurs, Icv is constant and we are done by Exercise theory of curves as we know it today, introducing, in particular, the notion
(2) above. of moving frame.

Exercise (4): The function f(s) _ -(a(s), N(s)) measures the distance
from the origin to the tangent line of the curve a at the time s. We fix
s e ][8 and let v be the segment of the line joining the origin with the
tangent line to a at s and that is perpendicular to it. As mentioned before,
the length of v is the value f(s). If x is the endpoint of v different from the
origin, Theorem 9.40 and the fact that a(][8) is an oval imply that x SZ
and, thus, x BT. Hence, the length of 0 has to be greater than or equal
1.2. Curves. Arc length
To start with, there are two ways of thinking about what one usually under-
stands intuitively as a curve. The first one is as a geometrical locus, that is,
as a set of points sharing a certain property. The second one is as the path
described by a moving particle. In this case, the coordinates of the curve
should be functions of a parameter which usually stands for time. For the
1.2. Curves. Arc length 3 360 9. Global Geometry of Curves

moment we will adopt this second approach because it allows us to more Exercise 9.48: Equality (9.1) allows us to compute the rotation index of
quickly use the techniques of calculus to describe the geometrical behaviour a, Q, and ry by integrating their curvatures multiplied by the length of the
of a curve. In spite of this, we are mainly interested in those properties velocity vector. To compute these integrals, we take Remark 1.15 into
depending only on its image set, that is, on its geometrical shape. This is account and get ka(t)Ia'(t)I = 1 and
why, in Chapters 2 (Example 2.18 in Section 2.2) and 9, we will again take 2 sin t sin 2t + cos t cos 2t (t)Vry,(t)I3 3- 2 sin t
up the relation between these two ways of thinking of curves. kQ (t)IQ(t)I -2 sinz t + 4 cost 2t
kry
5 - 4 sin t'
Definition 1.1. A differentiablei curve is a C°O map a : I - I[83 defined on for each t e R. Integrating these three functions between 0 and 2ir, we
an open, possibly unbounded, interval I of R. Referring to avector-valued obtain the required rotation indices. It is clear, from the first equality, that
function a, the word differentiable means that if i(a) = 1. The second function we integrate clearly satisfies
a(t) _ (x(t),y(t),z(t)), k(t + ir)I/3'(t + ir)I = -k(t)I/3'(t)I, Vt e R.

the three coordinate functions x,y,z are differentiable. The vector Thus, since
1 B-F2r

a'(t) = (x'(t),y'(t),z'(t)) 2(Q) - 2ir ,f kQ(t)IQ'(t)I dt,

whose components are the derivatives of the components of a is called the for each 8 E ][8, making the change of variable t H t + ir, we have
tangent vector, or velocity vector, of the curve a at t E I, or at the point -1J
a(t), even though this last expression is ambiguous. i(Q) =2f dt = dt = -2(Q),

We will say that the curve a : I - ][83 is a plane curve if there exists and, thus, i(,Q) = 0. As for the third curve, it is not difficult to see that
a plane P C I[83 such that a(I) C P. Since we are considering only those the function k.y Iry'I is increasing in sin t. Thus, its extremes are 5/3, which
properties of curves that are invariant under rigid motions, and we can is attained at t = 3ir/2, and 3, which is attained at t = ir/2. Therefore
always find a rigid motion of ][83 (see the comments after Proposition 1.6) 1
< -J
zir
5
taking the plane P to be the plane z = 0, we may restrict ourselves to the k.)(t)'y'(t) dt = i('y) :f; 3.
3
case a : I - ][83 where
Equality on the right-hand side would force k.yIry'I to be identically equal
a(t) = (x(t),y(t),O), to 3, which cannot happen. Hence, i(ry) is an integer in [5/3, 3). That is,
that is, to the case of differentiable maps a : I - I[82. 2 7) 2.

Example 1.2. A curve can have self-intersections (i.e., it does not neces- Exercise 9.64: The curve ry of Exercise 9.48 is regular, 2ir-periodic, does
sarily have to be injective), as the plane curve a : ][8 -f II82 given by
not have a simple trace, and its curvature satisfies
a(t) = (t3 -4t,t2 -4) (t) = 3 3 - 2 sin t
yt E I[8 '
shows (see Figure 1.1). ry (5 - 4 sin t)3/2'
according to this exercise. Then
Example 1.3. The image or trace of a curve can have cusps even though the k/ 24- sisinn tt)5/2
(t) = 3 cost,
curve is differentiable, for instance (see Figure 1.2), the curve a : III - R2 (5 - b't E I[8,

given by which vanishes only when t = ir/2 and t = 3ir/2.

a(t) _ (t3, t2).
One can check that a(R) is the graph of the function f : ][8 -> I[8 defined Exercise (1): By the change of variable theorem for the Lebesgue integral
by f(x) = x23, which is not differentiable at the origin. on ][8z, we have

1 From now on, differentiable will always mean of the C°° class.
A(cbt(1)) _ f Jac ¢t = fdet(dcbt)pd.
Hints for solving the exercises 359 4 1. Plane and Space Curves

positive on (52, Si + L). Thus, if a is a point of the line joining a(sl) and
a(s2) and v is a direction vector of this line, we see by Exercise 9.3 that the
function k'(a - a, v) does not change sign and is non-zero on a non-empty
open set. Thus, the previous integral cannot vanish. This is a contradiction
since k' has to vanish on each of the intervals (Si, 52) and (s2, sl + L).

Exercise 9.45: It is clear that, if s e ][8,

/3(s + B) = a(f (s + B)) = a(f (s) + A) = a(f (s)) _ ,Q(s)
and thus Q is B-periodic since f is a diffeomorphism and it is impossible
to find other periods smaller than B. On the other hand, from the relation
,Q = a o f, we obtain by differentiation that ,Q' = f'(cv! o f). Consequently, Figure 1.1. Curve with a self-intersection
the tangent map of /3 is
a?of
f
Thus T ([o,B] = Ta ° f([o,B] Since f : [0, B] -* [0, A] is a diffeomorphism, we
have
i(,Q) = degTp = degTal[o,A] = degTa = i(a),
where we have used Exercise 9.68.

Exercise 9.46: Let F : R x [0,1] -* 1R2 be the homotopy between cxo and
cxl. We know that, if we fix t e [0, 1], the derivative of s e IIS H F(s, t) is
non-zero. We then define H : R x [0,1] -* Si by
Figure 1.2. Curve with a cusp
d
F(s, t)
H(s, t) = ds se R, t e [0,1],
d Example 1.4. Even if a curve is injective, it does not have to be a homeo-
F(s, t) morphism onto its image. This is the case (see Figure 1.3) for the folium of
ds
which is continuous and A-periodic in the s-variable, since F is. Now, ap- Descartes a : (-1, oo) - R2, given by
plying Exercise 9.70, we have that 3t 3t
3t2
a(t) _
l1 + t3' l1 +t
i(ao) = deg degH0 = degHl = deg
IcoI IciI - a(al)
and we are done. Example 1.5. The following curves will play a prominent role from now
on because of their properties:
Exercise 9.47: Let b : I[82 -i II82 be a rigid motion or a homothety. Then, a(t)=pt+q,
if a : I[8 II82 is a periodic regular curve, we have ,(3(t) = c + r (cos T, sin T),
t bt
y(t) = a cos t , a sin
a2 -F b2 a2 + b2a2
' + b2)'
where A is the linear part of b, if b is a motion, or the identity map, if it is where p, q, c E ][82, r > 0, and a, b 0. They are, respectively, straight lines,
a homothety. Appealing to Exercise 9.72, we are done. circles, and circular helices. (See Figure 1.4.)
1.2. Curves. Arc length 5 358 9. Global Geometry of Curves

for each s, u e I, it means that the distance from a(s) to the tangent line of
a at u coincides with the distance from a(u) to the tangent line of a at s.

Exercise 9.41: The curve cx : III -* R2, given by

c(t) = (acost,bsint), dt E III,
is a regular 2ir-periodic curve whose trace is the ellipse C. Hence, using
Remark 1.15,
k(t)_ ab
0, b't E I[8.
(a2 cos2 t + b2 sinz t)3/2
Figure 1.3. Folium of Descartes Since C is diffeomorphic to a circle, it is a compact connected simple curve
with positive curvature.

Exercise 9.42: The parabola C is a connected simple curve closed in ][82

since it is the graph of a differentiable function defined on the real line
I[8. Moreover, a : Il8 I[82 given by a(t) _ (t2/2p, t) is clearly a regular
parametrization for it. Using the same formula as in Exercise 9.1, it follows
that
k(t)=- ____ VteR,
p2 +t2
and the parabola has non-vanishing curvature everywhere and, thus, it is
positive for the opposite orientation.

Exercise 9.44: The function f : ][8 -* ][8, given by

Figure 1.4. Circular helix
The first intrinsic notion for curves that we will study is arc length. Let
a : I -* l[83 be a curve and let [a, b] C I be a compact interval. We want
to measure the length of the piece a([a, b]) of the curve. A first intuitive
approach is given by the length of a polygonal line inscribed in the curve.
Precisely, let P = {to = a < tl < <t o = b} be a partition of the interval
[a, b]. We denote by
f(s) = k(s)(a(s) - a, v), Vs E ]E8,
is L-periodic and, using the Frenet formulas, we have
f'(s) = k'(s) (a(s) - a, v) + (N(s),v), Vs E I[8.
Integrating between 0 and L and taking into account the fundamental the-
orem of calculus for the integral on the left-hand side and Exercise 9.28 for
the second summand on the right-hand side, we find the required equality

Ld(a, P) _ I-
n

z=1
for any a, v e R.
To prove the assertion of the four-vertices theorem for ovals, we can
suppose that the curvature k of a is not constant on any subinterval of
the length of that polygonal line, where ui is the Euclidean length of any [0, L]. Then, there exist two numbers sl < 52 in [0, L) where k attains
vector u e ][83. Then we can prove its extreme values since k is continuous and L-periodic. Suppose that the
function k' does not vanish on the interval (Si, 52). In this case, if this
Proposition 1.6. If a : I -> ][83 is a curve and [a, b] C I, theme for every function k' changed sign on the other interval (52, sl + L), it would have
e > 0 there exists 8 > 0 such that at least two zeroes on it, since it has the same sign in a neighbourhood on
iPi <5 I La(a, P) - f b
ia'(t)I dtl <, the right of 52 and in a neighbourhood on the left of sl + L. Then, we can
assume that k' is, for example, strictly positive on (51,52) and that it is not
Hints for solving the exercises 357 6 1. Plane and Space Curves

by the equality of mixed derivatives. On the other hand,

aivJv
9 f - 8xa (-gf) + -y (9fx) = det (9y) - det(Vf, V9)
Exercise 9.31: It suffices to realize that Jac F = det(VFl, OF2). From
Exercise 9.30 and the divergence Theorem 9.23,

JacF=-JF2(a(s))((JOFI)(a(s)),N(s))ds,
L
J12 o
where a : I[8 -i I[82 is an L-periodic curve p.b.a.l. whose trace is C. Since
FC = IC, if we denote by x and y the components of a, we have Figure 1.5. Inscribed polygonal

Jac F = - L y(s)x(s) ds = - Z JoL(y(s)x (S) - x(s)y (S)) ds,

J J where JPJ = maxl<i<n Itz - ti-1J
where the last equality follows after applying the fundamental theorem of
calculus. But Proof. We define the function f : I3 --> l[8 by
y(s)x(s) - x(s)y'(s) _ (cr(s), N(s)), Vs E I[8. f(t1,t2,t3) = x'(tl)2 + y'(t2)2 + zI(ts)2,
Hence which is clearly continuous on I3 and therefore uniformly continuous on the

f Jac F = -2
c''
(c(s),N(s)) ds. compact set [a, b]3 C I3. Thus, for each e > 0, there is a a > 0 such that, if

== I<
(t1,t2,t3),(ti,t,t) E [a,b]3,
J
On the other hand, consider the identity field X : SZ -* 1R2, taking each
p e SZ to its own position X (p) = p. Then (dX)p = I1R2 for all p e SZ and, tl -tip < S
thus, div X = 2. Applying the divergence Theorem 9.23 to this field, we t2 - t2I <6 b a
obtain It3 - t31 <S
fL
2A(11) _ - (cr(s), N(s)) ds On the other hand, by using Cauchy's mean value theorem for the functions
J0 x, y, z on the interval [ti_ 1, ti], we have
and we are done.
I a\t2-1/ I f(f3, IZ, S2/ (t2 - t2-1)
Exercise 9.38: Differentiating alternately with respect to s and u in the for some yi,S2 j e [t_1, ti]. Hence - -
z

equality (cr(s) - a(u), N(s) + N(u)) = 0 and using the Frenet equations, we
n
have
(k(s) - k(u))(T(s),T(u)) = 0, b's,u E I.
La(cx, rD ) t2-1).
i=1
Differentiating again with respect to s, we obtain
Moreover,
k'(s)(T(s),T(u)) + (k(s) - k(u))k(s)(N(s),T(u)) = 0,
for all s, u e I. Putting s = u e I in the previous equality, we deduce that
k' is identically zero, that is, a has constant curvature. We are done by
f b Ia'(t)I dt =
d 2=1 f t2
j i
Ia'(t)I dt

Exercise 1.19. and by applying the mean value theorem to each of the integrals on the right
side, we have
As for the geometrical interpretation of the condition, since we can
rewrite it as 6 C
(cr(s) -em(u), N(u)) _ (c(u) - a(s), N(s)),
Ia'(t)I dt = - t_1) = -
a i=1 i=1
1.2. Curves. Arc length 7 356 9. Global Geometry of Curves

wherei e [ti_i, ti]. Now, if IPI < S, we have t2 - t2_1 < 5 for i = 1, ... , n. A is also open. Therefore A = J and thus /3 = a,j. Hence C = /3(J) = a(J)
As LZ, I2, 6i,' E [t_ i, ti], we also have I j2 - i I <5, I 'y2 - Z I <5, I6 - I <
2 v and, consequently, J = II8.
and thus

- fa Exercise 9.19: The parallel curve at distance t is the image of at : I[8 -i I[82
6
La(a, P) Ia'(t)I dt - f(e,e,e))(t - defined by at(s) = a(s) + tN(s), for all s e I[8, and where N is the inner
i=1
normal of a(][8). By the Frenet equations,
n n
ad(s) _ (1 - tk(s)) N(s), Vs E 1[8,
f(L9i,7i,5i) - 1(e., ea, e:)I(t: -t1) < b E a
(t2 - t2-1)
i=1 with 1 - tk(s) > 0 for all s E I[8, with t small enough. Consequently,
and this concludes the proof.
This result suggests that we define La(a) = fa Ia'(t)I dt to be the length
of a from a to b-or from a(a) to a(b). An important property of the length
L(at) _
fL(1-tk(s))ds=L-tJk(s)ds=L-2it,
0

of a curve that we have just defined is its invariance under rigid motions of
Euclidean geometry. In fact, this invariance property will appear repeatedly
throughout this text, and it will be a proof that the concepts that will be
defined are really geometrical notions.
Definition 1.7 (Rigid motion). Recall that a rigid motion b : I[8n -p I[8n
of Euclidean space I[8n is an isometry of ][8n viewed as a metric space with
the usual Euclidean distance. These isometries are necessarily affine maps
whose associated linear maps are orthogonal, i.e.,
q5(x)=Ax+b, VxR'2,
where A is a matrix in the orthogonal group O(n), that is, a matrix of order
n such that AAt = In, and b e 1[8n is a vector. Depending on whether
det A = fl, we say that the rigid motion b is either direct or inverse. In
the first exercise below we ask the reader to prove the invariance of length
under rigid motions which we referred to before.
Exercise 1.8. Let a : I -p ][83 be a curve and let M : I[83 -p ][83 be a rigid
motion. Prove that La(a) = La(M o a). That is, rigid motions preserve the
length of curves.
Exercise 1.9. T Let a : I -p ][83 be a curve and [a, b] C I. Prove that
Ia(a) - a(b)I <L(a).
In other words, straight lines are the shortest curves joining two given points.
Exercise 1.10. T Let a : I -+ ][83 be a curve. Show that
La(a) =sup{La(a, P) I P is a partition of [a, b] }.
Exercise 1.11. Let a : I -p ][83 be a curve such that Ia'(t)I = 1 for each
t e I. What is the relation between the parameter t and the length L(a),
where a e I? If a curve has this property, we will say that it is parametrized
by the arc length (p.b.a.l.).
where the last equality follows from the turning tangent Theorem 9.50. Fur-
thermore, by Remark 1.15,
k(S)
kt(S) det(«t(s),at(s)) _
1 - tk(s)'
for s e III and t small.
Exercise 9.20: Let a E II82 be arbitrary. If a : ][8 -* I[82 is an L-periodic
curve p.b.a.l. whose trace is C, the periodic function f : ][8 -* II8 given by
f(s) _ I- a12, s e I[8, must attain its maximum at some so e [0, L).
Let r = f(so) > 0, and let C be the circle centred at a with radius r. Then
C C D, where D is the closed disc determined by the circle. Hence, the non-
bounded connected set I[82 - D is contained in I[82 - C. Then, the domain SZ
determined by C is contained in D. It suffices now to apply Exercise (16) of
Chapter 1 to the curves C and C, taking into account that the inner normals
of C and C coincide at a(so).
Exercise 9.28: If SZ is the inner domain determined by a(I[8), consider the
constant vector field on St with value a e I[83 that has vanishing divergence,
and apply the divergence theorem (Theorem 9.23). Otherwise, notice that
N(s) = JT(s) =Jay(s) _ (Jc(s))', Vs E I[8,
since J is linear. It is now enough to apply the fundamental theorem of
calculus on [0, L], keeping in mind that a is an L-periodic function.
Exercise 9.30: The components of the field V f are the partial derivatives
f and fy. Then JOf = (-ft,, and, thus,
a a
div JD = (-f)+ fx = 0,
Hints for solving the exercises 355
for all s e J, and, since wand 3 are p.b.a.l., fp(s) = ±1, and thus, since f'
is continuous, f is of the form f(s) = s+c or f(s) _ -s+c, for all s e J and
some c e ]E8, as we wanted. Therefore, it is a question of showing that f is
differentiable. We can do it exactly as in Theorem 2.26 by proving that each
point of C has a neighbourhood on which one of the projections p to the
coordinate axes is bijective. Thus, if we restrict f to this neighbourhood,
f = (p o a)-1 0 (p o,Q) is a composition of differentiable functions.
After the previous discussion, an orientation on anon-compact con-
nected simple curve C will be a choice of a curve p.b.a.l. a : I -* I[83 that
is a homeomorphism onto C. Two of these curves will determine the same
orientation when they differ by a translation and they will determine the
opposite orientation if, in addition, they differ by a symmetry relative to
the origin. If C is compact, by Theorem 9.10, there exists a curve p.b.a.l.
a : I[8 -* II83 with period, say, L > 0. If ,3 is another curve of this type
with another period M > 0, since a restricted to (0, L) and Q restricted
to (0, M) are homeomorphisms onto C minus a point, by the above, then
these two intervals must have the same length. Hence L = M and, thus,
it makes sense to talk about the length of C with either ,Q(s) = a(s) or
/3(s) = a(-s + L) for each s e (0, L). In the first case, we say that a and ,3
determine the same orientation of C, and in the second case, the opposite
orientation.
Exercise 9.13: According to Theorem 9.10, we know that there exists a
curve p.b.a.l. cx : I C III -* 1R3 that is a homeomorphism between I and C.
If I were not the whole of R, then there would be some s e Bdry I - I. Let
{sn}nEr be a sequence of points of I converging to s. We have
Ic(sn) - &(5m)I =
fSs
n
a'(s) ds :s;; Is - sml
and, thus, {a(sfl)}nEN is a Cauchy sequence. Hence, it converges to a point of
I[83, which must belong to C = a(I), since it is closed. Thus a(sn) _
a(x) for some x e I. Since a is a homeomorphism, s = sn = x e I,
and we have a contradiction. Thus, the homeomorphism a is defined on I[8.
Now let /3: J C I[8 -* I[83 be any curve p.b.a.l. whose image is C. Suppose
that, after a translation and a symmetry relative to the origin if necessary,
0 E J, a(0) _ /3(0), and &(0) =,(3'(0). Then, the set
A = {s e J I a(s) _ ,Q(s) and o/(s) _ /3'(s)}
is a non-empty closed subset of J. Since the curve ,Q is also a local homeo-
morphism by the inverse function theorem applied to the composition of ,Q
with at least one of the projections of I[83, we have, if we use Exercise 9.1, that
8 1. Plane and Space Curves
Given two open intervals I and J in ][8, a diffeomorphism q5: 1 -p I, and
a curve a : I -p I[83, we may consider a new curve /3: J -p ][83 defined by
the composition ,6 = a o q5. We will say that ,6 is a reparametrization of a
because it has the same trace as a. The following exercise establishes that
the arc length is an intrinsic notion, since it only depends on the trace of
the curve and not on the precise parametrization being worked with.
Exercise 1.12. T Let q5: J -p I be a difFeomorphism and let a : I -> l[83 be
a curve. Given [a, b] C J with q5([a, b]) _ [c, d], prove that La(aoq5) = L(a).
1.3. Regular curves and curves parametrized by arc length
A standard method in differential geometry consists of substituting for a
given differentiable object, a curve in our case, the best linear object ap-
proaching it near a certain point.
Given a curve a : I --> ][83 and t e I, among all the straight lines in
][83 passing through a(t), the one having direction ap(t) is seemingly the
closest one to the curve at that point. Of course, this only makes sense
when a'(t) 0; see Example 1.3 in the previous section. In this case, we
will say that this straight line is the tangent line of a at t-or at a(t). A
curve is said to be regular when its tangent vector does not vanish anywhere,
i.e., when its tangent line is awell-defined object at each of its points. If
this happens, we make the following definition: we say that the normal line
of a regular plane curve at a given point is the line through this point that
is perpendicular to the corresponding tangent line.
Let a : I -> l[83 be any curve and to e I. We define the arc length
function from to, which we denote by S : I -> ][8, by
S(t) = Lo(a) =
f o
Ia'(u)I du.
Since the function u e I H Ia'(u)I is, in general, only continuous, the
function S is only Cl and
S'(t) = Ia'(t)I.
If we assume the curve a is regular, then by the inverse function theorem
S is a differentiable increasing open function. Then, if we put J = S(I),
the function S : I -+ J is a difFeomorphism between two open intervals.
Let b : J -* I be the inverse difFeomorphism and let /3: J -+ I[83 be the
reparametrization of a given by ,6 = a o q5. Then this new curve satisfies
= a'((s))'(s)
= _____
and thus 1= 1 for all s E J; see Exercise 1.11 above. It follows from
the above that any regular curve admits a reparametrization by arc length.
1.4. Local theory of plane curves
Figure 1.6. Frenet dihedron
From now on, unless we explicitly state otherwise, we will only consider
curves parametrized by arc length, which we abbreviate as p.b.a.l.
Exercise 1.13. Consider the logarithmic spiral a : ][8 - ][82 given by
a(t) _ (ae6t cos t, ae6t sin t)
with a > 0, b < 0. Compute the arc length function S : I[8 -p ][8 for to
corresponding to to e R. R,eparametrize this curve by arc length and study
its trace.
1.4. Local theory of plane curves
In many geometrical problems, the choice of a reference system that is
adapted to the situation that one is dealing with will simplify their res-
olution remarkably. To illustrate this, we will associate an orthonormal
frame of I[82 to every point along a plane curve p.b.a.l. whose evolution will
determine its more important properties.
Let a : I -p I[82 be a curve p.b.a.l. From now on, we will denote its
tangent vector a'(s) by T(s), which, in this case, is a unit vector. Thus
T : I - ][82 is a differentiable vector-valued function with IT(s)I = 1 for
each s e I. There are only two unit vectors in ][82 perpendicular to T(s).
We choose the vector N(s) = JT(s), where J ][82 -> ][82 is the ninety
degree counterclockwise rotation. We will call N(s) the normal vector of
a at s -or at a(s). With this choice, N : I -> I[82 is differentiable and
satisfies IN(s)I = 1, (T(s),N(s)) = 0, and det(T(s), N(s)) = 1 for every
s e I, where (,) is the usual scalar product and det is the determinant or
mixed product. Thus, we have constructed, for each s e I, an orthonormal
basis of ][82 that is positively oriented. This basis {T(s), N(s)} will be called
the oriented Frenet dihedron of the curve a at s.
9 354 9. Global Geometry of Curves
at least four connected components. In fact, a takes the interval (0, ir/2) to
the first quadrant since
sin 2t > 0,
tE(0,2) (cos t, sin t) is in the first quadrant.
Also, since
ir
t E (Z , ) sin 2t < 0,
(cost, sin t)is in the second quadrant,
the map a takes the interval (ir/2, ir) to the fourth quadrant. Reasoning in
this way, we see that a maps (ir, 3ir/2) to the third quadrant and (3ir/2, 2ir)
to the second quadrant. Since a is clearly 2ir-periodic, we have that R -
{(0, 0)} has four connected components that are the images through a of
these four intervals. Moreover,
lima(t) = lim a(t) = lim a(t) _ (0, 0),
t-+o t-+sir/z
and consequently there are points of C as close to the origin as we want in
each of the four components of R - {(0, 0)}.
Exercise 9.3: In fact, for each t e (-1, +oo), we have
_ (3(1-2t3) 3t(2-t3)'\
(l+t ' (l+t)3
which, obviously, can never vanish. Furthermore, if t, t' E (-1, +oo) and
a(t) =a(te), it is clear that t = 0 if and only if t' = 0. Suppose that t and
t' are not equal to zero. Then x(t) = x(t') and y(t) = y(t'), where x and y
are the components of a. Hence
x(t) x(t)
and, thus, a is injective. However, the folium J' = a((-1, +oo)) can-
not be a simple curve since (0,0) = a(0) E J' has a basis of discon-
nected neighbourhoods. Indeed, since a is continuous, a(0) _ (0, 0), and
a(t) _ (0, 0), it follows that if B is a ball of I[82 centred at the origin
to which, for example, the point a(1) (0,0) does not belong, then there ex-
ist E> 0 and r > 0 such that U = a((-s, s)) C B and V = a((r, +oo)) C B.
Notice that U and V are connected and disjoint, since a is injective and
(-e, e) and (r, +oo) are disjoint. If not, the union of these two intervals
would imply that (-e, +oo) and a(1) E B. Since this occurs for each ball
with radius less than the radius of B with centre at the origin, we are done.
Exercise 9.11: Since a and ,Q are bijective, we can consider the composition
f = c' o/3, which is a bijective map of J to I. Hence ,Q = ao f . Suppose that
we have already shown that f is differentiable. Then f'(s)&(f(s)) _
 Hints for solving the exercises 353 10 1. Plane and Space Curves

one has A way of measuring how a curve bends is to observe how the basis
JL {T(s), N(s)} associated to each of its points varies when we move along the
I
L (c(s),ei)2 ds + I L (c(s),e2)2 ds curve. This change can be described in terms of the derivatives T'(s) and
NP(s) of the two vector functions in that basis. Taking derivatives of the
sp+ 2 sp+L equalities
(c(s),ei) 2 ds + (c(s),ei)2 ds IT(s)12 = IN(s)12 = 1 and (T(s),N(s)) = 0,
Sp Sp -{- 2
we obtain that
/'L L
+J 2 (c(s),e2)2 ds + JL (c(s),e2)2 ds. (T'(s),T(s)) _ (N'(s),N(s)) = 0 and (T'(s),N(s))+(T(s),N'(s)) = 0.
0 2
Hence, the vector T'(s) has the same direction as N(s). Thus TM(s) _
Using this equality and the Wirtinger inequality (Exercise (13) above), k(s)N(s) for some k(s) E I[8, and this occurs for every s e I. Moreover, by
prove that definition,
pL
Irv(.c112 ric <
L3 k(s) - (T'(s),N(s))
Jo
and so the function k : I -> I[8 is differentiable. The real number k(s) is
This inequality, together with Exercise (3), shows the isoperimetric in- called the curvature of the curve a at s e I. We then have
equality. (This is a simplification of the Hurwitz proof.)
T'(s) = k(s)N(s) and NP(s) _ -k(s)T(s),
(16) Let a be a periodic curve with simple trace and length L, let St be its
inner domain and let f : St ][8 be a differentiable function. Prove that which are the Frenet equations of the curve a.

f =
-J 0
(df)3 (N(s)) ds.
(17) T What can you say about the length of a periodic parametrized plane
curve whose curvature k satisfies 0 < k < 1?
Hints for solving
the exercises
Exercise 9.1: Let R be the line passing through p E Il83 with direction
v E 2. Then, the affine map a ][8 -* R given by a(t) = p + tv is a
regular parametrization covering the whole of the line R. Let C be the
circle centred at a E I[82 with radius r > 0. Then, for each B E II8, the
map ry : (0,0 + 2ir) -* C given by 'y(t) = a + (r cos t, r sin t) is a regular
parametrization of C covering C - {(r cos B, r sin 0)}. We let B run through
][8 and cover the whole of C.
Exercise 9.2: Let R. = a(][8) be the four-petal rose. We know that the
origin is in R. since a(0) _ (0, 0). If R. were a simple curve, the origin would
have a neighbourhood in R. homeomorphic to II8. This is impossible since we
are going to show that each punctured neighbourhood of the origin in R has
Remark 1.14. What follows is the first justification for the name of curva-
ture that we have given to the function k. We have Ik(s)I _ IT'(s)I _ Ia"(s)I,
which is the length of the acceleration vector of a. Since Ia'(s)I = 1 for all
s e I, this acceleration is only centripetal and not tangential. Then Ik(s)I
is responsible for changing the direction of a at a(s). On the other hand,
k(s) _ (T'(s), N(s)) _ (a"(s), Ja'(s)) = det(a'(s), a"(s)), and then the sign
of k(s) tells us about the orientation of the basis formed by the velocity and
the acceleration of the curve. That is, if k(s) > 0, then the curve changes
its direction counterclockwise, and if k(s) < 0, it does so clockwise. This is
an important difference when compared to the case of space curves that we
will study in the following section, since an everywhere positive curvature
will be defined for them.
Remark 1.15. Let a : I -p R2 be a regular curve, not necessarily p.b.a.l.
We know that a can be reparametrized by arc length; that is, there is another
interval J and a diffeomorphism f : J - I such that /3 = a o f is a curve
p.b.a.l. having the same trace as a. We define the curvature of a at time
t e I by the equality
ka(t) = k(f'(t)).
Keeping in mind that fp(s) = 1/Ia'(f(s)) I for each s E J, we easily see that
ka(t) = la,(t)l3 det(a (t),a "(t))
1.4. Local theory of plane curves 11 352 9. Global Geometry of Curves

or, alternatively, (11) Let St C Il82 be an open set and let f : SZ -* I[8 be a differentiable function.
Show that the following assertions are equivalent.
ka(t) = det(T(t),T'(t)), b't E I, f is harmonic.
where T(t) = a'(t)/Ia'(t) is the unit vector tangent of a at the instant t E I. f satisfies the mean value property; that is, for each po E St and
each ball Br(p0) centred at po with radius r > 0 contained in St,
Example 1.16. Straight lines a : I[8 -p ][82, a(s) = as+b, a, b E ][82, al = 1, one has
have identically zero curvature, since a"(s) = 0 for every s E R.
Example 1.17. Let a : ][8 -p ][82, with a(s) = c + r (cos T, sin T), be the f(o) - r2 f (p) dp
counterclockwise circle with centre c e l[82 and radius r > 0. We have
5\
T = a(s) _ -r1 (cos r,s sin r) ,
For each po E SZ and each circle a centred at po with radius r > 0
contained in S1, one has

N(s) = JT(s) _ (- cos S, - sin S)

T'
f
,
r 2irr

and thus k(s) = 1/r for all s e R. Analogously, clockwise circles have
(df)a(s) (a(s) - po) ds = 0.
constant negative curvature -1/r.
(12) T Let A < B be two real numbers and f [A, B] -* II8 a function of
Exercise 1.18. Let a : I -p I[82 be a curve p.b.a.l. and let M : ][82 -p I[82 be class C2 such that f(A) = 1(B) = 0. Show that there exists a function
a rigid motion. If we denote the curve M o a by ,6, prove that h : [A, B] -* I[8 of class Cl such that
k«(s) for all s e I, if M is direct,
-k«(s) for all s e I, if M is inverse. 1(t) = h(t) sin B
7T
A (t - A)
Exercise 1.19. T Let a : I -p I[82 be a curve p.b.a.l. Prove that a is a
segment of a straight line or an arc of a circle if and only if the curvature of for each t E [A, B].
a is constant.
(13) T Wirtinger's inequality. Let f [A, B] -* ][8 be a function of class C2
Exercise 1.20. T Let a : I -p ][82 be a curve p.b.a.l. If there is a differentiable such that 1(A) = 1(B) = 0. Then
function 8 : I -p ][8 such that 8(s) is the angle that the tangent line of a at
s makes with a fixed direction, show that 8'(s) = fk(s).
(f(t))dt
This last exercise shows that the curvature of a plane curve determines, JA (B-A)2JA
up to a sign, the angle that its tangent vector at each point makes with
a fixed given direction. In this sense, the curvature is a genuine intrinsic and equality occurs if and only if 1(t) = asin for some a E I[8.
quantity for plane curves p.b.a.l. In fact, we will prove, in the following
result, that the curvature suffices, up to a direct rigid motion, to determine (14) T Let a : II8 ][82 be an L-periodic curve p.b.a.l. with simple trace.
a curve completely. Prove that there exists so E (0, L/2) and two perpendicular lines Ri and
R2 in the plane such that a(0), a(L/2) E Rl and a(so), a(so+L/2) E R2.
Theorem 1.21 (Fundamental theorem of the local theory of plane curves).
(15) T Under the hypotheses of the previous exercise and choosing the affine
Let ko : I -> ][8 be a differentiable function defied on an open interval I C R.
frame of the plane
Then, there exists a place curve a : I -p ][82 p. b.a.l. such that ka(s) = ko(s)
for every s e I, where ka is the curvature function of a. Moreover, if
,(3 : I -> ][82 is another plane curve p.b.a.l. with kg(s) = ko(s), for all s E I,
a(2) - a(0) also + 2) -a(so)
these there exists a direct rigid motion M : ][82 -> ][82 such that ,6 = M 0 a.
O = Rl n R2, el = e2
I-c(so)I'
 Exercises 351 12 1. Plane and Space Curves

and that equality occurs if and only if a is a circle centred at the origin. Proof. We define a function 0 : I -p III by
(4) T Let a be a periodic curve p.b.a.l. with length L, whose trace is an s

oval and let St be its inner domain. Suppose that St contains the disc 0(s) = ko (u) du,
so
BT = {p E R2I II <r}. Prove the following statements.
- (N(s), a(s)) > r for each s E ][8. where so e I is arbitrary. Then 0 is differentiable and, from Exercise 1.20
A(St) > (r/2)L and equality is attained if and only if a is the circle
above, must be, up to a constant, the angle that the tangent line of the
centred at the origin with radius r. curve that we want to construct makes with a fixed direction. This is why
we define a : I -p R2 by
(5) Let A C II82, X, Y be vector fields defined on A, let f : A -* I[8 be a
differentiable function, and a, b E ][8. Prove the following equalities. «(S) _ (fcos0(u)dufsin0(u)du)
Z
div(aX + bY) = a divX + b divY and div(f X) = f divX + (Vf, X).
for any 51, 82 E I. The map a is clearly differentiable, and for s E I,
(6) T Let X be a vector field defined on the whole of the plane. Show that as(s) _ (cos0(s),sin0(s)).
X is a conservative field if and only if div JX = 0.
(7) Let A C ][82, and let f, g : A -p II8 be two differentiable functions. We Thus Ia'(s)I = 1 for each s e I and, hence, a is a curve p.b.a.l. Its Frenet
define the LaPlacian of f by dihedron is

a2 a2
T(s) _ (cos0(s),sin0(s)), N(s) = JT(s) _ (-sin0(s),cos0(s)).
On the other hand, we have
Prove that L f = div V f and O(f g) = g0 f + fLg +2 (Vf, fig). T'(s) _ (-8'(s) sin 0(s), 8'(s) cos6(s))
(8) Let a be an L-periodic curve p.b.a.l. whose trace is a simple curve and and, by the definitions of curvature and of the function 8,
let St be its inner domain. If f, g : St -* ][8 are differentiable functions,
show that ka(s) = (T'(s), N(s)) = 0'(s) = ko(s)
L as we had anticipated.
Jcz
fg + J (Vf, 09) _ -f f((s)) (dg)a(s) (N(s)) ds,
Now suppose that /3: I , ][82 is another plane curve p.b.a.l. with kg(s) _
ko(s) for all s e I. Pick any so e I. Since the Frenet dihedra of a and ,3 at

Jst
fg - /s sp

{Ta(so),Na(so)} and {T(so),N(so)}

If(c(s))(dg)a(s)(N(s)) ds + I L g (a(s))(df)«(s) (N(s)) ds. are two positively oriented orthonormal bases of l[82, there exists a unique
orthogonal matrix A E 80(2) such that
These identities are known as Green's formulas.
ATa(sO) = Z'Q(SO) and AN«(SO) = NQ(SO)
(9) If St C II82 is an open set and f : St -* ][8 is a differentiable function, we
say that f is harmonic when E f = 0 on ft Prove that the functions Let b = 16(so) - Aa(so) E ][82 and M : ][82 -> ][82 be the direct rigid motion
determined by A and b, that is,
g : II82 -][8, g(p) _ (a, p) + b with a E ][82, b E II8,
Mx = Ax -I- b for every x E l[82.
h : ][82 -{0} h(p) =log IpI2 We will now show that the plane curve y I -p ][82 p.b.a.l. defined by
are harmonic. y = M o a coincides with ,Q. We have, in fact, that
(10) T Let St be the inner domain determined by a compact connected simple y(so) = Aa(so) + b = /3(5o),
plane curve C and let f : St -* II8 be a differentiable function such that T(so) = ATa(so) = T(so),
O f = 0 on St and f1c = 0. Prove that f = 0 on S2. Nry(so) = JT.y(so) = JTQ(so) = Na(so).
1.5. Local theory of space curves 13 350 9. Global Geometry of Curves

Moreover, from Exercise 1.18 above,

k.y(s) = ka(s) = ko(s) for each s E I.
If we define f : I - I[8 by

f(s) = 2 [ITh(s) - INQ(S) -

N7(S)I2]
then f(so) = 0 with
-
f'(s) _ (Tb(s) - Ty(s),TQ(s) - T.y(s)) + (N,(s) Ny(s), NQ(s) - N.y(s)).
Using the Frenet equations and the fact that kQ = ky = ko, we have f'(s) = 0
for each s e I. Therefore f vanishes everywhere and hence
TQ(s) - T.y(s) = 0 for each s E I, Figure 9.15. Cases of a simple curve and a curve with self-intersections

and then Exercise 9.65. Consider the curve a : ][8 -* ][82, given by
,(j(s) - y(s) = constant e l[82.
But ,3(5o) = ry(so) and we conclude that /3= y = M o a. O c(t) = (t,t2), Vt E IR.

Exercise 1.22. Let I -p ][82 be two plane curves p.b.a.l. such that Show that the trace of a is a simple curve but also that a has a unique
ka(s) _ -kg(s) for every s e I. Show that there is an inverse rigid motion vertex.
M : l[82 -p II82 such that /3= M o a.
Exercise 1.23. If a : I -> I[82 is a curve p.b.a.l. defined on an open interval
of ][8 containing the origin and symmetric relative to it, we define another Exercises
curve /3: I -p ][82 by ,Q(s) = a(-s) for each s e I. Prove that Q is p.b.a.l.
and that kg(s) _ -ka(-s) for each s E I. (1) T Let SZ C I[82 be a bounded open set and let X be a vector field defined
Exercise 1.24. T Let a E I[8+, and let a : (-a, a) -p ][82 be a curve p.b.a.l. on ft. Suppose that there exists e > 0 such that, for tI < e, the map
with ka(s) = ka(-s) for each s e (-a, a). Prove that the trace of a is ct - cbt(1), given by cbt(p) = p + tX (p), is a diffeomorphism. Prove
symmetric relative to the normal line of a at 0. that
Exercise 1.25. T Let a E ][8+, and let a : (-a, a) -> ][82 be a curve p.b.a.l. fdiv X.
dt I

with ka(s) _ -ka(-s) for each s e (-a, a). Prove that the trace of a is Taking this equality into account and interpreting X as the field of
symmetric relative to the point a(0). velocities of a fluid with constant density, give a heuristic proof of the
divergence theorem for the plane.
1.5. Local theory of space curves
(2) T Let C be a compact connected simple plane curve with length L and
As we did for plane curves, we will associate to any point of a space curve let SZ be its inner domain. If C is contained in a closed disc of radius
a positively oriented orthonormal basis of three-space, whose evolution will r > 0, prove that
give us its geometrical features when we move along the curve. A(S2) < 2 L
Let a : I -p R3 be a curve p.b.a.l. Its tangent vector a'(s) is a unit vector
and equality holds if and only if C is a circle.
which, as in the case of plane curves, will be denoted from now on by T (s).
Then, one has that the vector-valued function T : I -p R3 is differentiable (3) T Let a be an L-periodic curve p.b.a.l. whose trace is a simple curve
and satisfies I= 1 for every s e I. Taking derivatives in this last and let 1 2 be its inner domain. Show that
1
equality, we obtain
(T'(s),T(s)) = 0 for any s E I.
2A(1) ; (fo
L l 2 L2
9.6. The four-vertices theorem 349
Figure 9.14. Circle circumscribed to C
point r E I' such that k(r) < 1/R. In fact, assume without loss of generality
that S is centred at the origin and the points p and q are on some vertical
straight line V ; see Figure 9.15. Since this line V cuts S transversely and
thus C at the points p and q, there must be points of C on the right-hand
and on the left-hand sides of V; see, for instance, Exercise (17) at the end
of Chapter 1.
We will now consider the component I' of C with endpoints p and q that,
in each neighbourhood of p, has points on the right-hand side of V. Since I',
together with the two disjoint half-lines of V starting from p and q, forms a
curve without self-intersections separating the plane into two components
which we will naturally call right and left regions we have that the inner
normal of C, restricted to I', points toward the left region. We move the
circle S to the left up to the last position S' in which it intersects F. There
are only two alternative cases: either I' is contained in S and S' = S or I' is
not contained in S, S' touches I' at a non-endpoint, and the two curves are
tangent at this point. In both cases, I' and S' are tangent at a non-endpoint
of I' and the inner normal of I' at this point is the inner normal of S'. We
deduce, by comparison, that the curvature of I' at this contact point is less
than or equal to 1/R. Thus, the curvature of I' attains its minimum at a
non-endpoint. (Recall that the curvature of C at p and q was greater than
or equal to 1/R.) The same argument gives us another local minimum of
the curvature of C in the interior of the other arc determined by p and q.
Since in each arc of C determined by two local minima there has to be at
least a local maximum, we infer that the curvature of C has at least four
different local extremes. This completes the proof of the theorem.
Exercise 9.64. T Find an example of a periodic regular parametrized plane
curve whose trace is not a simple curve and for which the four-vertices
theorem is false.
14 1. Plane and Space Curves
Figure 1.7. Frenet trihedron
That is, TM(s) is perpendicular to T(s). We define the curvature of a at s
by k(s) _ IT'(s)I. Note that, in this situation, k(s) > 0 for each s E I,
which was not always the case for plane curves. In general, by definition,
the curvature function k : I -* ][8 is only continuous. Assume in what follows
that the curve a satisfies k(s) > 0 for each s E I. Then the function k will
be differentiable and also the vector TM(s) will not vanish anywhere. Thus
it makes sense to consider the vector
TM(s) _ 1 ,
N(S) k(s)T(s).
We will call this vector the normal vector of a at s and, from previous
results, we know that the map N : I -* ][83 is differentiable and satisfies
IN(s)I = 1 and (T(s), N(s)) = 0 for each s E I. Thus, in order to obtain a
positively oriented orthonormal basis of ][83, it suffices to add a third vector
B(s) = T(s) A N(s),
where n denotes the vector product of Euclidean three-space, which we will
call the binormal vector of the curve a at s. Of course, the map B : I -* Il83
is also differentiable.
At each point a(s) of the curve that we are studying, we have built
a positively oriented orthonormal basis {T(s), N(s), B(s)} which will be
called the Frenet trihedron of the curve a at s. It depends differentiably on
s. This construction is possible only when the curve a : I -* ][83 p.b.a.l. has
curvature k(s) _ T'(s) strictly positive everywhere.
We now investigate the variation of the Frenet trihedron along the curve
a. As in the case of plane curves, we will do this by considering the deriva-
tives of the three vector-valued functions T, N, and B. It follows, by the
definition of curvature, that
T'(s) = k(s)N(s)
1.5. Local theory of space curves 15
for each s E I. On the other hand, taking derivatives of the equality B(s) _
T(s) A N(s), which defines the binormal vector, we have
B'(s) = T'(s) A N(s) + T(s) A N'(s) = T(s) A N'(s)
since the vectors T'(s) and N(s) have the same direction. Thus
(B'(s),T(s)) = det(T(s), N'(s),T(s)) = 0
for each s E I. Moreover, since B(s)12 = 1 everywhere and taking deriva-
tives again, we obtain
(B'(s), B(S)) = o.
That is, the vector Ba(s) has no components either in the direction of T(s)
or in that of B(s). So, Ba(s) must have the same direction as N(s) and we
set
B'(s) = 'r(s)N(s),
where T(s) E ][8 for each s e I. Since T(s) _ (B'(s),N(s)), the function
T : I -* Il8 is differentiable. The real number T(s) is called the torsion of the
curve a at s. Finally, it remains to study the derivative NP(s) of the normal
vector. Taking derivatives of the equalities IN(s)12 = 1, (N(s), T(s)) = 0,
and (N(s), B(s)) = 0, we have that
(N'(s), N(s)) = 0,
(N'(s),T(s)) _ - (N(s),T'(s)) _ -k(s),
(N'(s), B(s)) _ - (N(s), B'(s)) _
and thus
N'(s) = -k(s)T(s) - 'r(s)B(s)
for each s E I. The relations that we have just deduced,
( T'(s) = k(s)N(s),
(1.1) N'(s) _ -k(s)T(s) - T(s)B(s),
I. B'(s)='r(s)N(s),
are the Serret-Frenet equations of the curve a.
Remark 1.26. A curve a : I -* Il83 p.b.a.l. contained in a plane can be
thought of alternatively as a plane curve or as a space curve. The curvature
of a viewed as a space curve is nothing but the absolute value of its curvature
when considered as a plane curve.
Example 1.27. A straight line a : Il8 -* ][83, a(s) = sa + b, with a, b E ][83
and al = 1, considered as a space curve, has zero curvature, since T(s) = a
for each s E Il8, and thus k(s) _ lT'(s)l = 0.
348 9. Global Geometry of Curves
Figure 9.13. E 1 S,, = 0
region St that is the complement in the disc of a neighbourhood of S1 with
radius s. We see that this region is included in a circle with radius less than
R-with centre on the x-axis to the right of the origin. This contradicts the
fact that R is the radius of the circumscribing circle to the set E. D
Corollary 9.62. Let E be a compact subset of R2 with more than one point,
and let C be its circumscribing circle. The intersection E fl C has at least
two points. Moreover, if it has exactly two, they are antipodal.
Finally, we are able to give the four-or more-vertices theorem in its
strongest version due to Osserman.
Theorem 9.63 (Four-vertices theorem). Let C be a compact connected sim-
ple curve of the plane R2. Then C has at least four vertices. Moreover, if
it has exactly four, then C touches its circumscribing circle at exactly two
antipodal points.
Proof. We know that C is a compact subset of R2 with more than one
point and, thus, by Proposition 9.60, it has a circumscribing circle S whose
radius we denote by R > 0. By Corollary 9.62, we know that C fl S has
at least two points. Suppose that p E C fl s is an arbitrary point in this
intersection. Exercise (1) of Chapter 1 tells us that the curves C and S
are tangent at p and, since the inner domain determined by C (see the
Jordan curve Theorem 9.14) is contained in the circle S, the normal vectors
corresponding to the positive orientations of C and S coincide at p. We can,
then, apply the result on the comparison of plane curves of Exercise (16)
of Chapter 1 and deduce that k(p) 7 11F1, where k is the curvature of the
curve C; see Figure 9.14.
Suppose now that p and q are two different points in C fl S and let IF
be any of the two connected components of C - {p, q}. We show, and for
this it is fundamental that C is simple, that either r C S or there exists a
9.6. The four-vertices theorem 347 16 1. Plane and Space Curves

In particular, F is continuous. Furthermore, if C(p, F(p)) is the circle cen- Example 1.28. Let a I -* ][83 be a plane curve p.b.a.l. Suppose that
tred at p with radius F(p), we have that E C C(p, F(p)), since the plane where the trace of the curve lies is P = {x E ][83 I (a, x) = b},
x-pF(p), VxEE, with a E ][83, al = 1, and b E R. Suppose also that, as a space curve, it
has positive curvature. Then its torsion and its Frenet trihedron are well-
where we deliberately conflate circles and discs, and thus defined objects. Since (T(s), a) = 0 and (T'(s), a) = 0, the tangent and
C(p, F(p)) E CE b'p E 1182. normal vectors at any point of a are perpendicular to the vector a. Then,
B - fa and T(s) = 0 for each s E I; see Exercise 1.32 below.
Also, if C E CE has radius R> 0, we can find a point xo E E with ixo -p1 _
F(p) and hence Example 1.29. As a consequence of the discussions above, a circle with
F(p) = Ixo -p1 R. radius r > 0 has, as a space curve, constant curvature 1/r-independent of
Thus, C(p, F(p)) is the least circle centred at p E II82 containing E. the rotation sense-and torsion identically zero.
Now let xo E E be such that F(0) = maxXEE lxi _ Ixol, and let p E Il82 Example 1.30. In Example 1.5 of Section 1.2, we defined the circular helix
be such that Il > 2lxol Then, for each x E E we have c:R -* R3 by
lp-xI Il - lxi
2lxol - Ixol = F(O).
a(s) = (acos a2 +
s
, a sin
s bs
a2
Taking the maximum relative to x, we get b2+ b2' a2 + b2 /
F(p) ? F(0) if II ? Ro = 2lxoI > 0. with a, b 0, which is a curve p.b.a.l. Taking derivatives, one has that
1 s s
By continuity, F attains its minimum in the closed ball Bp (0) at some T (s) = (_a sin a cos a2 + b2' b '
a2 +
b2'

point po. Therefore, a2 + b2

F(p) if Il
F'(po) $ J
Ro,
T'(S) a2 + b2 I cos a2+ b2 ,sin a2 + b2' 0 I .

F(0)F(p) if
Hence, the curvature of the circular helix is
Hence, F has an absolute minimum at the point po. That is, the circle
C = C(po, F(po)) is a circle in the plane containing E and it has minimum k(s) = IT' (s)l = a2
°l b2'
radius among all the circles including it. +
that is, a non-null constant. Next, consider its Frenet trihedron and its
If there were two different circles Cl and CZ with minimum radius F(po)
torsion. In fact, its normal vector is
containing E, then we would have E C Cl fl C2. But the intersection of
two circles of the same radius and different centres is always contained in
another circle C3 with smaller radius. Thus, we would deduce that E C C3i
N(s)=- T'(s) = la cos a2s b2 sin a2s b2 0

which is impossible. D and its binormal vector will be given by

Another fundamental property of circles circumscribing compact sets is, B(s) = T (s) n N(s) = lal -b sin a2s b2 b cos a2s b2 a
besides its uniqueness, the way in which they touch the set. a a+ b2

Proposition 9.61. Let E be a compact subset of the plane with more than Then, by differentiating with respect to s, we have
one point and let C be its circumscribing circle. Then, there are points of B'(s) = (a2 a 2 bcos S , -bsin S 0= -b N(s).
+ b)
b2,

Eon each closed half-circle of C. al a2 +

a2 + b2 a2 + b2
From this equality, we deduce that the torsion of the circular helix is
Proof. Suppose that C is centred at the origin and that its radius is R> 0.
Consider the two closed half-circles Si and 52 determined by the y-axis and -b
a2
such that E n Si = 0; see Figure 9.13. Since E and S1 are disjoint compact + b2'
subsets, we have that dist (E, Si) = e > 0. Then E is contained in the closed that is, it is also constant.
1.5. Local theory of space curves 17 346 9. Global Geometry of Curves

Exercise 1.31. Let cx : I -* R3 be a curve p.b.a.l. Prove that cx is a segment
of a straight line if and only if the curvature of cx vanishes everywhere.
Exercise 1.32. T Show that a curve a : I -* ][83 p.b.a.l. with k(s) > 0 for
each s e I is plane if and only if its torsion vanishes everywhere.
Exercise 1.33. Let cx : I -* R3 be a curve p.b.a.l. and let M : -* R3 be
a rigid motion. If we set /3 = M o cx, show that
T«(s) for each s E I, if M is direct,
kg(s) - ka(s) and 'r (s) -
-Ta(s) for each s E I, if M is
closed segment of ]R3 joining these two points at height r. We can construct
an injective continuous map from a closed disc of R2 with diameter r2 - r1
to ]R3 taking each horizontal segment of the disc to each of those of 0r. This
contradicts the fact that C was assumed to be a knot. U
9.6. The four-vertices theorem
The four-vertices theorem asserts that a compact connected simple plane
curve C has at least four vertices. We use the term vertex of C for any point
a(s) E C for which the curvature satisfies k'(s) = 0, where c : IR -*
inverse.

is any periodic regular parametrization of C. It is easy to be convinced,

Exercise 1.34. Let a : I -* ][83 be a curve p.b.a.l. with positive curvature
defined on an interval I C Il8 symmetric relative to the origin. We define
another curve /3: I -* Il83 by ,Q(s) = a(-s), for each s E I. Show that ,Q is a
curve p.b.a.l. and that kg(s) = k«(-s) and 'r(s) _ -T«(-s) for each s E I.
In the same way that a plane curve p.b.a.l. is determined, up to a direct
rigid motion, by its curvature, we will be able to reconstruct, up to a rigid
motion of R3, a curve cx : I -* R3 p.b.a.l. with positive curvature from its
curvature and its torsion. The existence proof that we gave in the funda-
mental Theorem 1.21 of the local theory of plane curves was a constructive
proof. In the case of space curves, we will use instead the existence theorem
for first order linear ordinary differential equations applied to a particular
equation whose matrix of coefficients is determined by the curvature and
the torsion functions of the curve. Before beginning the proof, it is conve-
nient to write the F Tenet equations (1.1) that we have obtained for a space
curve p.b.a.l. in a vectorial form. In fact, if {T (s), N(s), B(s)} is the tenet
trihedron of the curve cx : I -* R3 p.b.a.l. with k(s) > 0, for each s E I, we
define a map from I to R9 by
T(S)
5F-3 N(s) b's E I,
B(s)
according to the notion of curvature in Remark 1.15, that this definition of
a vertex does not depend upon the parametrization utilized. For instance,
it is not difficult to check that an ellipse, which is not a circle, has exactly
four vertices at its intersections with its axes. A first version of this theorem
appeared as early as 1909 as we mentioned in the introduction of this
chapter for convex curves; see Exercise (4) of this chapter. In this context,
the theorem provides a sufficient condition for a positive periodic function to
be the curvature of a simple curve. Here we will prove the theorem for simple
curves not necessarily convex following an idea due to Osserman. Consider
a circle circumscribing the curve and count the points of contact with the
curve. Between each two contact points, there are necessarily local extremes
of the curvature and, thus, vertices, in such a way that a curve with exactly
four vertices can have only two contact points with the circumscribing circle.
We will start by studying the circle [circumscribed to] circumscribing any
compact subset of the plane.
Proposition 9.60. Let E be a compact subset of ]R2 with more than one
point. Let CE be the family of all the circles of the plane such that the
corresponding discs contain E. There exists in CE a unique circle C with
the least radius that will be called the circle circumscribing E.
Proof. Let F : * be the function given by

where we are considering T(s), N(s), B(s) as column vectors in Il83. Then F(p) = maEx lx -pI, dp E R2,
which is awell-defined function since E is compact. If, for some p, we had
T'(s) k(s)N(s) that F(p) = 0, then E C {p}, which is impossible by the hypothesis. On
N'(s) = (
-k(s)T(s) - 'r(s)B(s) the other hand, if p, q E II82, then
B'(s) J T(s)N(s)
Ix-qI CF(p)+Ip-4'I and Ix-pi F(q)+p-q
03 k(s)I3 03 T(s)
for each x E E. Taking maxima, we obtain
-k(s)I3 03 -T(s)I3 N(s)
03 T(S)13 03 B(S) Ip-qi, dpi q E II82.
9.5. Periodic space curves 345 18 1. Plane and Space Curves

Crofton's formula will allow us to show for a periodic curve whose trace
is a simple curve, that to be knotted in ]R3, it is necessary that the curve
bend at least twice the amount that we would expect for a general curve; see
Figure 9.12. Before stating the corresponding result, proved first by Fary
and Milnor, we need to make clear what we mean when we say that a curve
is knotted. Let C be a compact connected simple curve in ]R3. We will
say that C is knotted or that it is a non-trivial knot when we cannot find
any injective continuous map from the closed unit disc of j2 to TIS3 whose
restriction to the frontier circle is a homeomorphism onto C. Otherwise, we
will say that the curve C is unknotted.
Theorem 9.59 (Fary-Milnor's theorem). Let a : TIC -* TIS3 be an A-periodic
regular curve with positive curvature everywhere such that its trace is a sim-
ple curve C. Then, if C is a knot, we have
Thus, the map
T(S)
sH N(s)
B(s)
is a solution to a first order linear differential equation. This property of the
Frenet trihedron of a curve alone implies the following result.
Theorem 1.35 (Fundamental theorem of the local theory of space curves).
Let I C ][8 be an open interval and let ko, To : I -* Il8 be two differentiable
functions with ko(s) > 0, for each s E I. Then there exists a curve a : I -*
][83 p. b.a.l. such that k«(s) = ko(s) and Ta(s) = To(s) for each s E I, where
k« and Ta are the curvature and torsion functions of a. Furthermore, a is
unique up to a direct rigid motion of Euclidean space Il83.

f
A
Proof. Consider the following first order linear differential equation
k(s) ds > 4ir,
(1.2) x'(s) = Ao(s)x(s),
where k is the curvature function of a.
where
Proof. Assume without loss of generality that a is p.b.a.l. Since the cur- 03 kO(s)I3 03
vature k of a does not vanish, the tangent curve T : TIC -* S2 is regular and Ao(s) _ -ko(S)Ia 03 -TO(S)Ia
A-periodic. We can then apply Crofton's formula to it and obtain 03 TO(S)I3 03

if
f
A
We take a E ][89 such that the vectors
k(s) ds = Lo (T) = 4 2 nT(a) da,
J to = (air a2 as) no = (a4,a5,a6), and bo = (a7,a8,a9)
with nT defined as in the statement of Crofton's theorem. Suppose that the
conclusion of the theorem were not true. In this case, we could find a point form a positively oriented orthonormal basis of Il83. Let f : I -* ][89 be a
a E §2 such that nT(a) <4. That is, the function solution of (1.2) with initial condition f(so) = a and sp E I arbitrary. Then,
if we define t, n, b : I -* ][83 by
s E [0,A) H-* (T(s),a) _ (-y,a)'(s)
t = (fi, f2, f), n = (f4, fs, f6), and b = (f7, f8, f),
has at most three zeroes, or, in other words, the function
we have
s E [0,A) H-* h(s) _ (-y,a)(s)
t'(S) _ ko(S)n(S),
has at most three critical points. Since h is continuous and A-periodic, it n'(s) _ -ko(s)t(s) - To(s)b(s),
attains its maximum and its minimum in [0, A). If they were attained at the b'(s) = T(5)fl(5).
same time, then they would coincide and, hence, a would be a plane curve Let M(s) be the matrix whose entries are the scalar products of the vector-
and, thus, since its curvature does not vanish, it would be an oval. In this valued functions t(s), n(s), and b(s), that is,
case, its trace C could not be a knot, by part A of Theorem 9.40. Hence, the
maximum and the minimum are attained at different times and take different It(s)12 (t(s),n(s)) (t(s),b(s))
values r2 > r1. Then, for each r E (r1, r2), the function h - r has exactly M(s) _ (n(s),t(s)) In(s)12 (n(s),b(s))
two zeroes, since, otherwise, the Cauchy mean value theorem would supply (b(s),t(s)) (b(s),n(s)) Ib(s)12
two different critical points of h also different from the absolute maximum From this we see that M(s) satisfies the differential equation
and minimum. Since the curve has a simple trace, each plane of R3 whose
height is between r1 and r2 cuts C at exactly two points. Let dr be the (1.3) Ms(s) = A(s)M(s) - M(s)A(s),
Exercises 19
where
0 ko(s) 0
A(s) _ -ko(s) 0 -TO(s)
0 To(s) 0
Furthermore, M satisfies the initial condition M(so) = I3. But the matrix
function that is identically I3 also satisfies the same equation (1.3) with the
same initial condition. Thus, M - 13 and {t(s), n(s), b(s)} is an orthonormal
basis of l[83 for each s E I. As a consequence,
det(t(s),n(s),b(s)) =f1
for each s E I. On the other hand, since det (t(so), n(so), b(so)) = 1, by
continuity, the basis formed by the vectors {t(s), n(s), b(s)} is positively
oriented for every s E I.
Now, define a : I -* I[83 by
a(s) = ft(u)du, b's E I.
o
Then a is differentiable and as(s) = t(s), and thus Ic'(s)I _ It(s)I = 1 and
a is a curve p.b.a.l. Moreover, TM(s) = t'(s) = ko(s)n(s); that is, ka(s) _
IT'(s)I = ko(s) and N(s) = n(s). Hence B(s) = T(s) A N(s) = t(s) An(s) _
b(s) and Ba(s) =ba(s) = To(s)n(s) = To(s)N(s), and finally Ta(s) = To(s).
The uniqueness can be proved in a way completely analogous to that of
the fundamental Theorem 1.21 of the local theory of plane curves.
Exercise 1.36. Let I -* Il83 be two plane curves p.b.a.l. with ka(s) _
kg(s) > 0 and Ta(s) _ -TQ(s) for each s E I. Prove that there exists an
inverse rigid motion M : ][83 -* ][83 such that ,Q = M o a.
Exercise 1.37. Let a : I -* Il83 be a curve p.b.a.l. with k(s) > 0 for every
s E I. Prove that a is an arc of a circular helix or an arc of a circle if and
only if both the curvature and the torsion of a are constant.
Exercises
(1) Let a : I _* Il82 be a regular plane curve and a E ][82 - a(I). If there
exists to E I such that f - al > N- al for each t E I, prove that
the straight line joining the point a with a(to) is the normal line of a at
tp. The same is true if we reverse the inequality.
(2) T Let a : I _* ][82 be a regular plane curve and let R be a straight line in
l[82. If one can find a number to E I such that the distance from a(t) to
R is greater than or equal to the distance from a(to) to R, for all t e I,
344 9. Global Geometry of Curves
Figure 9.12. Knotted and unknotted curves
Therefore, using Fubini's Theorem, we have
rA r 2zr
1 IJac¢`=J J
Thus, it is enough to show that the functions n(o,1), defined in Chapter
5 to give the area formula, and ny, defined in the statement that we want
to prove, coincide except on a subset of measure zero in S2. To do this, it
suffices to show that
n(O,1)(a) _ n-r(a)
for each a E 52, which is a regular value for 0 (see Sard's Theorem 4.33)
and such that (y(0), a) # 0--since, with this restriction, we eliminate only
a great circle. Then take a point a E 52 with these characteristics. In
this situation, we know that the set 0-1(f al) is finite and that, taking the
definition of 0 and the fact that (-y(0), a) 0 into account,
n(q, l)(a) _ #1(8,t) E [0, A) x (0, 27r) I (y(s), a) = 0}.
It only remains to show that, for the last set on the right-hand side whose
cardinal number we want to compute, the projection on the first component
is bijective. This is true since, for each s E IR, the map
t E [0, 27r) i--> O(s, t) = cos t T(s) + sin t T(s) A y(s)
is injective.
Remark 9.58. From Crofton's formula, we can produce a proof of the
inequality provided in Fenchel's Theorem that will only be valid for curves
with positive curvature everywhere. In fact, it is enough to realize that, if
the curve is plane, the desired inequality is already shown in Proposition 9.21
and, if not, then nT > 2 since T is not contained in any open hemisphere
and this implies, by the periodicity of T, that, for each a E S2, we have
(T, a) = 0 at least two times in [0, A).
9.5. Periodic space curves
Proof. The discussion before the statement of the theorem implies that we
can apply Proposition 9.54 to the curve T of the unit tangent vectors of a.
Thus, since we have seen that the trace of T is not included in any open
hemisphere of 2, the length of T has to be greater than or equal to 2ir, that
is,
2 < Lo (T) _ A IT'(s)Ids=j A k(s) ds
0 0
and, if equality occurs, then the trace of T is the union of two great half-
circles. In this case, it will be contained in a closed hemisphere and thus a
is a plane curve whose curvature, as a plane curve, does not change its sign,
according to part B of Proposition 9.21. Moreover, the rotation index of a
is fl, since it has total curvature 2ir. Hence, we may apply Remark 9.53
and conclude that a is convex. O (6) T Show that a curve a : I -* ][82 p.b.a.l. is a segment of a straight line or
We will next state and prove an integral formula due to Crofton that
measures the total number of intersections that a periodic regular curve,
whose trace is in the unit sphere, has with its great circles. After stating
this, one might guess that the corresponding proof must be closely related
to Proposition 9.54.
Theorem 9.57 (Crofton's formula). Let 'y : ][8 , §2 be an A-periodic regular
curve. Then, the function n.y : S2 -* N U {oo}, given by
r(a) _ #{s E [O,A) I (y(s),a) = 0}, b'a
which measures the number of points, with their multiplicity, intersecting
each great circle of the sphere, is integrable and
fS2 n( a) da =4 Lo (ry)
Proof. We represent by O the open rectangle of Il82 given by (0, A) x (0, 2ir)
and let q5: O -> 2 be the differentiable map given by
(s, t) = cost T(s) + sin t T(s) n ry(s),
where (s, t) E O and T(s) is the unit tangent vector of the curve y at time s,
that is, T(s) _ -y'(s)/Iry'(s)I, which is awell-defined map since ry is regular.
We now apply the area formula of Theorem 5.28 to the map b and to the
constant function f O -> I[8 equal to +1 and deduce that the function
is integrable on O and that
IJacI 1)
fo = 1S2
Keeping in mind the definition of the absolute value of the Jacobian given
in Section 5.2 or in the exercises cited there, we have
IJaccbl(s,t) _ Idet(cbs,cbt,cb)I(s,t) _ Icostl V"()I
343 20 1. Plane and Space Curves
and such that a(to) R, then show that the tangent line of a at to is
parallel to R.
(3) T Let a I -* ][82 be a regular plane curve. Let [a, b] C I be such
that a(a) a(b). Prove that there exists some to E (a, b) such that
the tangent line of a at to is parallel to the segment of the straight line
joining a(a) with a(b). (This is a generalization of Rolle's theorem of
elementary calculus.)
(4) Prove that a curve a : I -* Il82 p.b.a.l. is a segment of a straight line if
and only if all its tangent lines are concurrent.
(5) T Prove that a curve a : I -* ][82 p.b.a.l. is an arc of a circle if and only
if all its normal lines pass through a given point.
an arc of a circle if and only if all its tangent lines are equidistant from
a given point.
(7) T Given a curve a : I -* Il82 p.b.a.l., prove that all the normal lines of a
are equidistant from a point if and only if there are a, b E l[8 such that
1
as + b
for each s E I. (See Exercise (11) below.)
(8) Let a : I -* l[82 be a curve p.b.a.l. and let so E I. We define the function
f : I -* l[8 by
f(s) = (cr(s) - c(so), N(so)),
measuring the oriented distance from the point a(s) to the tangent line
of a at so. Prove that f(so) = 0, f'(so) = 0, and f"(so) = k(so). Deduce
from this the following statements:
. If k(so) > 0, then there is a neighbourhood J of so in I such that
a(J) belongs to the closed half-plane determined by the tangent
line of a at so and the vector N(so).
. If there exists a neighbourhood J of sO in I such that a(J) is in the
closed half-plane determined by the tangent line of a at so and the
normal vector N(so), then k(so) > 0.
Obtain analogous results for the reverse inequalities.
(9) Let a : I -* Il82 be a curve p.b.a.l. and so E I with k(so) > 0. Let
as = a(so) + AN(so), A E l[8 - {0}, be a point in the normal line of a at
so. Define the function f,, : I -* ][8 by
IA(s) _ I- aaI2,
which is the square of the distance from the points of the curve to the
point a. Show that fa(so) _ A2, f(so) = 0, and f,(so).= 2(1-Ak(so)).
As a consequence, deduce the following:
 Exercises 21 342 9. Global Geometry of Curves

If A < 1/k(so), then there exists a neighbourhood J of so in I such Since, moreover, d(m,-y(a)) < ir/2 since 'y(a) and 'y(b) are not antipodal, we
that a(J) is outside of the circle with centre as and radius Al. conclude that
If A> 1/k(s0), then there exists a neighbourhood J of so in I such d(m,'y(s)) < , Vs e TE;
that a(J) is inside of the circle of centre as and radius A.
We see in this way that the circle centred at a(so)+(1/k(so))N(so) with that is, the trace of ry is contained in the open hemisphere whose pole is the
radius 1/k(so) is the best circle approaching the curve near a(so). This point m. O
circle will be called the osculating circle; its radius, the curvature radices;
and its centre, the curvature centre. The curve whose trace consists of Remark 9.55. Proposition 9.54 is still valid for Lipschitz curves since we
all the curvature centres of a, /3(s) = a(s) + 1 /k(s)N(s), provided that only required this type of regularity in Theorem 7.42. In fact, for regular
k(s) > 0, will be called the evolute of a. curves, we could have given an easier proof and the second alternative of
part B would not be admissible, unless the two arcs were in the same great
(10) T Let a : I ][82 be a curve p.b.a.l. with k(s) > 0 for all s E I. Prove circle. The important point is that it can be applied to curves that are not
that all the osculating circles of a meet at a given point if and only if a necessarily regular. This will permit us to prove Fenchel's theorem for any
is an arc of a circle. regular curve in TE without the usual assumption that the curve has positive
(11) T Let a : I -* l[82 be a curve p.b.a.l. with k(s) > 0 for all s E I. Prove curvature everywhere.
that all the normal lines of a are equidistant from a fixed point if and
only if the evolute of a is a circle. In such a case we say that a is an Now suppose that a : Ilk -* TE is an A-periodic, A > 0, regular curve.
evolvent of that circle. Then, the tangent map T : III -* TE that takes each s E IR to the unit
(12) Suppose that a plane curve a : I II82 p.b.a.l. has positive and non- tangent vector of a at s, that is,
decreasing curvature. Let /3: I l[82 be its evolute. Prove that
T(s) b's E II8,
1 1 Ia'(s)I'
is a differentiable map of I[8 to S2 that is A-periodic as well. Moreover, note
for anyaEland sElwiths>a. that, for each a
(13) T Let a : I -* l[82 be a curve p.b.a.l. with positive and non-decreasing fA
f
A
curvature and with ,Q as its evolute. If a E I, prove that (a'(s), a) ds = I (a(s), a)' ds = 0
0

Ia(s) - /3(a)l
k()
for each s E I with s > a, that is, a([a, oo) fl7) is inside of the osculating
circle of a at a.
(14) T Let a : I ][82 be a curve p.b.a.l. and let a E Il82 - a(I). If there
exists so E I such that la(s) - al < la(so) - at for all s E I, show that
1 curve, in any closed hemisphere. From this information, we can state the
lc(so) - al
(15) Let a : Il8 Il82 a plane curve p.b.a.l. such that a(II8) is included in a
closed disc of radius r > 0 and lk(s)I < 1/r for each s E R. Prove the
following.
. The function f(s) _ la(s) - a is the centre of the disc,
is bounded from above and satisfies f"(s) > 0 for all s E R.
Any differentiable function f : Il8 -* Il8 that is both bounded from
above and satisfies f" > 0 is necessarily a constant function.
and, hence, the trace of T cannot be contained in the open hemisphere with
pole a and, if it were contained in the corresponding closed hemisphere, the
function (T, a) = (a', a)/ I a' I would not change its sign and would be, in this
way, identically zero, and, consequently, a would be a plane curve. That is,
if a is a periodic regular curve, the trace of its unit tangent map T on the
unit sphere is not contained in any open hemisphere and, if it is not a plane
following result concerning the total curvature of periodic regular curves in
3.
Theorem 9.56 (Fenchel's Theorem). If a : Il8 -> Il83 is an A-periodic regular
curve, its total curvature satisfies
A
k(t) dt > 2ir
L
and equality is attained if and only if a is a convex plane curve.
9.5. Periodic space curves
to a point '-y(d), with d e [x, A). Then, -yI[t,d] is also an arc of a great circle
with length less than ir; see Figure 9.11. We now repeat the same argument
replacing -y(0) = '-y(A) by '-y(t) = ry(t + A) and we find y e (t, t + A) such
that
L) = Lr() = L =
Notice that it must necessarily occur that y > d. If the pair t, y is not the
required pair a, b either, then there exists r E (t, y) U (y, t + A) playing the
role that t played before. Suppose, without loss of generality, that r is in
(t, y). Consequently, the traces of and are arcs of different great
circles. Thus, we have no alternative but '-y(r) _ -y(d). As before, the arc of
the great circle joining 'y(r) = 'y(d) and 'y(y) can be extended, inside of the
trace of y, until a last point '-y(e) for some e E [y, c + A], in such a manner
that the arc of the great circle joining ry(d) and 'y(e) also has length less
than ir. If e = c+ A, we would have a spherical triangle ry(c), ry(t), '-y(d) with
perimeter 2ir and this would force its three sides to be parts of the same great
circle. Therefore, e <c + A. Then, we can choose a number a E (e, c + A)
and repeat the above discussion a third time, this time replacing -y(0) by
y(a). Thus, we find b e (a, a + A) such that
Ly) = L(y) = = of a at s. Suppose that a(I) lies in a sphere and that all its binormal
Consequently b > t + A since otherwise,
L(y) > L(y) = Ly) =
For the same reason, b < x + A. mom what we already know about the
trace of -y, neither the trace of nor the trace of y 6,a+A] can be the
union of two arcs of great circles. Therefore,
d(y(a), ry(s)) + d(ry(s), ry(b)) c ir
for all s E II8, as required.
We also know that the points -y(a) and y(b) are not antipodal and, then,
as in the proof of part A, we can consider its middle point m Suppose
that a point ry(s), s E I[8, of the trace of ry is on the great circle with pole m,
that is, d(m, ry(s)) _ ir/2, and let f be the 180 degree rotation around the
point m. Taking Exercise (11) of Chapter 7 into account, we can write
r = 2d(m, ry(s)) = d(m, ry(s)) + d(m, f (ry(s))) = d(ry(s), f (y(s))).
But, since m is the middle point of y(a) and ry(b),
< d(ry(s), ry(b)) + d(ry(b), f('y(s))) = d(ry(s), y(b)) + d(y(a), y(s)) < ir,
where we have used our particular choice of a and b. This contradiction im-
plies that the continuous function s H d(m, y(s)) never takes the value ir/2.
341 22 1. Plane and Space Curves
Deduce that a is a circle centred at a with radius r > 0.
(16) T Comparison of two curves at a point. Let a and /3 be two regular plane
curves defined, respectively, on two intervals of Il8 containing the origin.
Suppose that a(0) _ ,3(0) = p and that a'(0) _ ,13(0), or alternatively
Na(0) = NQ(0), that is, a and /3 are tangent at p and have the same
orientation. We say that a is over ,Q at p if there is a neighbourhood of
0 E Il8 where (a - a(0), N«(0)) > (/3 -,Q(0), NQ(0)). Prove the following.
If a is over /3 at p, then k«(0) > kQ(0).
. If k«(0) > kQ(0), then a is over /3 at p.
(17) T Two regular plane curves a and ,13, defined, respectively, on two in-
tervals including the origin, intersect transversely at p = a(0) _ ,3(0) if
a'(0) and /3'(0) are not parallel. Show that, if this happens, the traces
of at and ,Q have non-empty intersection, for t sufficiently small, where
at is the curve defined by at(u) = a(u) + ta, a E l[82, lal = 1.
(18) Let a : I -* ][83 a curve p.b.a.l. with positive curvature. Then a is an
arc of a circle if and only if it has constant curvature and its trace is
contained in a sphere.
(19) T If a : I -* l[83 is a curve p.b.a.l. with positive curvature and s E I, we
call the line passing through a(s) with direction B(s) the binormal line
lines are tangent to this sphere. Show that a is an arc of a great circle.
(20) T Let a : I -* ][83 be a curve p.b.a.l. with positive curvature. If the trace
of a is contained in a sphere and it has constant torsion a, prove that
there exist b, c E Il8 such that
k(5)
b cos as + c sin as
(21) A curve a : I -* ][83 p.b.a.l. and with positive curvature is said to be a
helix when all of its tangent lines make a constant angle with a given
direction. Show that a is a helix if and only if there exists a E Il8 such
that T(s) = ak(s) for each s E I. (This result is known as Lancret's
theorem.)
(22) T Let a : I -* ][83 be a curve p.b.a.l. with positive curvature such that
k'(s) 0 and T(s) 0 for each s E I. Prove that the trace of a is
contained in a sphere of radius r > 0 if and only if
1 k'(s)2 2
k(5)2 + = r .
k(s)4'r(s)2
(23) T Let a : I -* Il83 be a curve p.b.a.l. with positive curvature and con-
tained in a sphere of radius r > 0. If so E I, show that the following
assertions are equivalent:
k'(so) = 0,
 Hints for solving the exercises 23 340 9. Global Geometry of Curves

. k(so) = hr or 'r(so) = 0.
(24) T We use the term osculating plane of a curve a : I -* ][83 p.b.a.l.-
with positive curvature-at s E I for the plane passing through a(s)
with normal vector B(s). Let so E I. If there is a plane P passing
through a(so), including the tangent line of a at so and such that, for
any neighbourhood J of so in I, a(J) intersects the two open half-spaces
determined by P, show that P coincides with the osculating plane of a
at so

(25) Let a : I -* ][83 be a curve p.b.a.l. with positive curvature. Show that
the two following assertions are equivalent:
All the osculating planes of a are concurrent. Figure 9.11. The trace of -y
The curve a is a plane curve.
(26) Let a : I -* Il83 be a curve p.b.a.l. with positive curvature. Prove that Suppose, instead, that y does not pass through any pair of antipodal
there exists a curve w : I -* l[83 such that the Frenet equations of a can points. We will prove that, in this case, it is possible to find two numbers a
be written as and b such that a G b <a + A, La(y) = Lb+A(ry) _ ir, and
T'(s) = w(s) n T(s), N'(s) = w(s) n N(s), and B'(s) = w(s) n B(s). d(ry(a), ry(s)) + d(ry(s), ry(b)) < ir, Vs E ][8.

The vector w(s) is called the angular velocity of a at s. To show this, let us start by taking, as in the previous case, some x E (0, A)
such that
(27) Under the hypotheses of Exercise (26), show that the angular velocity
w is constant if and only if a is an arc of a circular helix. L) L)
= = 2 = r.
If the pair 0, x does not serve as the required pair a, b, then there exists
t e [0, A] such that
Hints for solving ir <d('y(O),'-y(t)) + d('y(t),'y(x)).
the exercises Notice that t e (0, x) U (x, A) since -y(0) and '-y(x) are not antipodal. We will
assume, without loss of generality, that t e (0, x). On the other hand, since
Exercise 1.9: It is sufficient to take into account that, if f : I -* R3 is a d('Y(O),'Y(t)) + d('Y(t),'Y(x)) Lo('Y) + Lt ('Y) = Lo('Y) _
continuous function, then
we see that ry, restricted to [0, t] and to [t, x], minimizes the length between
rb
its endpoints. Again Theorem 7.42 tells us that the trace of ry on these
fi intervals is contained in two arcs of great circles that must be different.
Otherwise, since the sum of its lengths does not exceed ir, we would have,
(the Schwarz inequality). Now apply it to the case f = a'. To complete the
from Exercise (11) of Chapter 7, that
exercise, apply the fundamental theorem of calculus.
d('Y(O),'Y(x)) = d('Y(O),'Y(t)) + d('Y(t),'Y(x)) _
Exercise 1.10: From Proposition 1.6, it only remains to show that which is impossible. Thus, the arc of a great circle joining 'y(O) and y(t) and
b belonging to the trace of ry ends at ry(t), but it could possibly start before
L (a, P) L(a) =
f « (t)I dt 'y(0). Suppose that, in fact, it starts at 'y(c) for some c < 0. Hence, ryI[c,t] is
an arc of a great circle that cannot be extended inside of the trace of y and
for each partition P = {to = a < ti < < t= b} of [a, b]. This that must have length less than ir since 'y(c) and 'y(t) cannot be antipodal.
follows by applying Exercise 1.9 to the curve cx on each subinterval [t2_1, t2], In the same way, the arc of the great circle joining y(t) and -y(x) inside of
i=1,...,n. the trace of -y could not end at the point y(x). Assume that this goes up
9.5. Periodic space curves 339 24 1. Plane and Space Curves

and, thus, we have Exercise 1.12: By the chain rule,

2d(m,'Y(s)) < d(ry(s),'Y(x)) -}- d(ry(x), f('y(s))). I(aocb)'(t)I = Ia'(cb(t))Ikb'(t)I.
But f is an involutive isometry, and hence Since is a diffeomorphism, we have either b' > 0 or b' <0. Now use the
theorem for the change of variables for integrals of functions of one variable.
d('y(x), f('y(s))) = d(f('y(x)), 'y(s)) = d('y(O), 'y(s)),
since m is the middle point of '-y(O) and ry(x). Combining the previous Exercise 1.19: The sufficient condition can be shown in this way: if the
inequalities, we deduce that curvature is identically zero, then the Frenet equations imply that a is a
segment of a straight line. On the other hand, if it is a non-null constant
2d(m,'Y(s)) d('Y(O),'Y(S)) + d('Y(s),'Y(x)) L) = 2 ko, one can prove that the function f : I -* II8, given by
We have proved, then, that f(s) = a(s) +
d(m,'y(s)) < 2 == d(m,'y(s)) 4, Vs E I[8. is constant by differentiating with respect to s.

Since, by hypothesis, L/4 < ir/2, it follows that, for each s E Il8,
either d(m,'y(s)) < 4 < 2 or d(m,'y(s)) > 2.
Since the function s H d(m, ry(s)) is continuous and
d(m,'Y(O)) = 2d('Y(O),'Y(x)) 4 normal line of a at a(0). Define another curve /3: (-a, a) -* ][82 by
only the first alternative is admissible. Therefore
d(m,'y(s)) <, Vs e TE;
that is, the trace of ry is contained in the open hemisphere whose pole is the
point m, as we wanted to prove.
Part B. We now modify our assumption and assume that L = L( 'y) _
2ir. Suppose, first, that, in the trace of -y, there are two antipodal points
of 2. For instance, imagine that y(0) and -y(x) are antipodal for some
x E (0, A). Then, utilizing Exercise (22) of Chapter 7 again, we obtain
r = d(ry(0), (x)) < Lo (ry) and r = d(ry(x), (A)) < L (ry).
Adding these two inequalities, we have
2ir < Lo Y) + L( 'y) = Lo Y) = 2ir,
and so the curve ry, restricted to the subintervals [0, x] and [x, A], minimizes
the length between its endpoints. We apply Theorem 7.42 and deduce that
the trace of ry on each of these intervals coincides with the trace of a geodesic
of length ir, that is, with a great half-circle. Then, in this case, the trace
of ry is the union of two great half-circles with the same endpoints, which is
one of the two possibilities expected in B.
Exercise 1.20: Differentiate the equality
cos B(s) _ (T(s),a),
where a e ][82 is a given unit vector. Notice that, then, (N(s), a) = f sin B(s).
Exercise 1.24: Let : ][82 - ][82 be the plane symmetry whose axis is the
,(i(s) _ q(a(-s)) for each s e (-a, a).
Using Exercises 1.22 and 1.23, check that kp = ka. Also, since /3(0) _
a(0) and ,6'(0) = a'(0), we complete the exercise by using the fundamental
Theorem 1.21.
Exercise 1.25: This can be solved in a way similar to Exercise 1.24 above
if one considers that is the central symmetry of IEg2 with respect to the
point a(0) and if one bears in mind that this symmetry is a direct rigid
motion and not an inverse motion as in the case of axial symmetries.
Exercise 1.32: From Example 1.28 it only remains to consider the sufficient
condition. If T - 0, then the Frenet equations imply that B(s) = a, for all
s e I and some unit vector a e ][83. Thus
(a(s),a)' = (T(s),a) = 0.
Then (a(s), a) = b, for some b e R.
Exercise (2): If a E R and u e Il82 is a unit vector normal to the straight
line R, then the function f : I -* Il8, given by
f(t) = (a(t) -a,u),
Hints for solving the exercises 25
measures the (oriented) distance from the point a(t) to the line R. Since
a(to) R, there is a neighbourhood of to in I in which f does not vanish.
Hence, f measures, up to a sign, the distance from the points of (the trace
of) a to R. Then f has a minimum at to e I and thus f'(to) = 0.
Exercise (3): It is enough to apply Rolle's theorem to the function f
[a, b] -* l[8, given by
f(t) = det(a(t), a(a) - a(b)).
Exercise (5): The necessary condition is clear. Suppose that all the normal
lines of a pass through a E I[82. Then the function f : I -* ][8, given by
f(s) = la(S) - a12,
has derivative f'(s) = 2(a(s) - a,T(s)). But a(s) and a are points of the
normal line of a at a(s). Then f' vanishes everywhere because a(s) - a is
normal.
Exercise (6): Assume that a e I[82 is the point from which all the tangent
lines of a are equidistant. If s e I, then the tangent line to a at s passes
through a(s), and is perpendicular to N(s). Then
(a(s) - a, N(s)) = c
for all s e I and for some c e R. Taking derivatives and using the Frenet
equations, one sees that, for each s e I, either k(s) = 0 or 1 + ck(s) = 0.
Since k is continuous, either k is identically zero or identically -1/c.
Exercise (7): The normal line to a at s e I passes through a(s) and the
vector T(s) is perpendicular to it. Then, if all the normal lines of a are
equidistant from a e I[82, there exists some c e ][8 with
(a(s) - a,T(s)) = c for each s e l.
Differentiating with respect to s, we have
1 + k(s)(a(s) - a, N(s)) = 0.
Taking derivatives again, we obtain
k'(s)(a(s) - a, N(s)) - k(s)2(a(s) - a,T(s)) = 0.
Multiplying by k(s) and using the above equalities, we deduce that
k'(s) + ck(s)3 = 0,
or (1/k2)" = 0 as we were expecting. Conversely, if the equality of Exercise
(7) holds, it is obvious that k does not vanish anywhere. Furthermore,
338 9. Global Geometry of Curves
A periodic regular parametrized plane curve with rotation
index fl is convex if and only if its curvature does not
change sign.
9.5. Periodic space curves
In this section, we will continue the study that we carried out on the total
curvature of periodic plane curves in the previous section. We now wish to
consider the total curvature of skew curves, that is, of curves whose trace
is not, in general, contained in any plane of We will also see that this
global geometrical quantity associated to the curve includes some topological
information. For plane curves, this total curvature controlled the regular
homotopy class. Here, it is also related to the manner in which the curve
embeds in three-space. We will start with a property of periodic curves
whose trace is not too long and that is contained in the unit sphere, which
will be applied later to the tangent curve of any periodic regular curve.
Proposition 9.54. Let 'y : III §2 be a periodic parametrized, but not nec-
essarily regular, spherical curve with length L> 0. The following assertions
are true.
A: If L < 2ir, then the trace of y is included in some open hemisphere
of the sphere 2.
B: If L = 2ir, then either the trace of ry is contained in some open
hemisphere or it is the union of two great half-circles with the same
endpoints.
Proof. Part A. If A > 0 is the period of ry and S : [0, A] -* I[8 is its arc
length function with 8(0) = 0, we have that S(A) = L. Thus, since S is
continuous, there must exist an x E (0, A) such that
5(x) = L) = = L).
Hence, if d is the intrinsic distance (consult Exercise (5) of Chapter 7) on
the unit sphere, we have
d('y(O),'y(x)) L('y) <ir,
and, consequently (see Exercise (22) of Chapter 7) the points y(0) and 'y(x)
are not antipodal. Then, according to Exercise (23) of Chapter 7, we have
its middle point m E S2 at our disposal. Now let s E [0, A) be such that
the corresponding point of the curve satisfies d(m, ry(s)) < ir/2. Then, if we
represent by f the 180 degrees rotation around the point m of S2, the point
m is also the middle point of y(s) and f('y(s)). Hence
d(ry(s), f('y(s))) = 2d(m, ry(s))
9.4. Rotation index of plane curves 337 26 1. Plane and Space Curves

A third consequence is the characterization of convex simple plane curves (1/k2)' = 2(1/k)(1/k)' = 2c for some c e R. Let f : I -* I[82 be the function
in terms of its curvature. defined by
Corollary 9.52. Let a : I[8 Il82 be a periodic curve p. b. a.l. whose trace f(s) = a(s) + N(s) - cT(s), Vs E I.
C = a(I[8) is a simple carve, and let k be its curvature. Then C is convex if
Differentiating and using the Frenet equations, we see that
and only if k > 0.

Proof. The condition on the curvature is necessary, as we already pointed f'(s) = ()'N(s) - ck(s)N(s) =0.
out in Definition 9.39, because of Exercise (8) of Chapter 1. Conversely,
Therefore, there is a e Il82 such that
suppose that k(s) > 0 for all s E I[8. We will see in (9.4) of Appendix 9.7
that a(s) - a = cT(s) - N(s).
s
k(s)
a' (s) = 1 (t) dt
so Thus (a(s) - a,T(s)) = c for each s e I.
is a lift of a', for every so E III. Hence, since k > 0, this lift is non-decreasing.
Moreover, by the turning tangent Theorem 9.50, Exercise (10): If a is an arc of a circle, its evolute is the curve constantly
passing through its centre and, thus, the osculating circle at any point is
1 = deg a' = 1 [&'(sO + L) - a '(so)] = a '(so + L) just that one containing a. Hence a and any of the osculating circles have
that is, a' takes the interval [so, so + L) to the interval [0, 2ir), where p : III -* an infinite number of common points. Conversely, let a e R2 be a common
1 is injective. Thus, our hypotheses on the curve imply that, if there exist point of all the osculating circles of a. We have
sl, s2 E [0,L) such that sl <52 with a'(sl) = a'(s2), then a' is constant on
the interval [Si, s2]. On the other hand, if C is not convex, then there is a la(s) + -N(s) - a
point a(so) E C with so E [0, L) with the property that its affine tangent line
for all s e I. Differentiating, we obtain
separates the points of C. As a consequence, the height function f : III -* TE,
given by
f(s) _ (a(s) - a(so), N(so)), k(s) )
reaches its maximum and its minimum at different points and at points Then, on the open set where (1/k)' 0, one has that (a-a, NJ = 0; that is,
different from a (so) , namely, at a (s 1) and at a (s2) , respectively, where all the tangent lines pass through a. From Exercise (4), the curvature would
sl, s2 E [0,L) and so Si, so s2, and sl s2. Since sl and s2 are critical need to vanish on that open set, and this is impossible. Thus (1/k)' = 0,
points for f, we have f'(si) = f' (s2) = 0; that is, the vectors a' (sl) and and so k is a positive constant.
a' (s2) are either perpendicular to N (so) or they are proportional to a' (so) .

Therefore, two of the three tangent vectors of a at so, sl, and s2 coincide.
From the property that we proved at the beginning for the curve a, a' should
be constant on a closed interval whose endpoints are two of three numbers
so, s 1, and 2 This would lead, on this interval, to the height function f
being constant, and this contradicts the choice that we made of the three
points of the curve. Thus, the curve C must be convex. D
Exercise (11): Show that the tangent line of the evolute ,6, at s e I, is
just the normal line of a at s e I, and use Exercise (6) and the fact that
the curvature of a is positive everywhere. Another way to solve it is the
following: prove that the curvature of the evolute /j of a satisfies

Remark 9.53. If we carefully go over the above proof, we will observe the and then apply Exercise (7).
fact that the trace of the curve a is a simple curve, that is, that a has
no self-intersections, has only been invoked when we used that the rotation
Exercise (13): For each s e I, we have
index of (the positively oriented) a is 1. For this reason, we have actually
proved the following assertion. la(s) - /3(a)I la(s) - /3(s)l + /3(s) - /3(a)l.
Hints for solving the exercises 27 336 9. Global Geometry of Curves

Using the definition of Q and taking into account Exercise 1.9, we deduce and the loop ,Q, also taking values on the unit circle, given by
that a' (0) ifs=0,
Q(S) - Q(a)I <Z'a(/3)
We finish by considering Exercise (12). a(2s)
ifs E (0, L/2),
i

Exercise (14): The function f : I -* Il8, given by

f(s) = a(s) - al2,
is differentiable and has a local maximum at so E I. It only remains to
compute f"(s) and to set f"(so) < 0.

Exercise (16): Define two functions f and g on the interval of common
definition of the two curves by
f(s) _ (a(s) - p, u) and g(s) _ (/3(s) - p, u),
where u is the common normal at s = 0. From our hypothesis, the function
f - g has a local minimum at s = 0. To finish the exercise, check that
(1 - g)"(O) = k(0) - k(0).
Exercise (17): Let I be an interval of ][8 containing 0 on which two curves
a and /3 are defined. Consider the map F : I x I[8 -* ][82, given by
F(u, t) = at(u) - ,(i(u) = a(u) - /j(u) + ta.
Then F(0, 0) = 0 and Fu(0, 0) = a'(0) - ,6'(0) 0. We can, then, apply
the implicit function theorem and deduce that there exists an interval K
including the origin and a differentiable function f K -* ][8 such that
f(0) = 0 and F(f (t), t) = 0 for all t e K. In other words,
at(f (t)) _ 13(f (t)) for all t e K.
Hence, the curves at and ,6 meet at each t e K.
Exercise (19): We know that al2 = 1 and that the only root of the
second degree polynomial p(t) _ a + tBl2 - 1 is t = 0. Then (a, B) - 0.
Differentiating al2 = 1, we obtain (a, T) - 0. Differentiating again, we have
l+k(a, N) - 0. Since a is a unit vector everywhere and (a, T) _ (a, B) = 0,
we have that (a, NJ is identically -1. Thus the curvature k is constantly
equal to 1. On the other hand, differentiating the expression (a, B) = 0, we
see that the torsion T vanishes and so the curve is a plane curve.
a'(0) ifs=L.
Moreover, H(0, t) = F(0, 0) = a'(0) and H(L, t) = F(L, L) = a'(L) = a'(0)
for each t e [0, 1]. Then, the definitions of the rotation index, degree of a
periodic map to 1, and the invariance of the degree under homotopies (see
Appendix 9.7) imply the equality
i(a) = dega' = degaf[o,L] = deg-y = deg/3.
Thus, the proof will be complete if we show that the degree of the loop Q is 1.
To show this, let f : [0, L] -> ][8 be the lift of Q such that f(0) = 0; recall that
,6(0) = a'(0) _ (1, 0). Hence, we obtain that f(s) < ir for 0 <s <L/2 since
on this interval p(f (s)) _ ,3(s) = a(2s)/Ia(2s) I and a has the trace over the
x-axis; that is, its second component is always greater than or equal to zero.
This, together with the fact that p(f (L/2)) _ ,3(L/2) _ -a'(0) _ (-1, 0),
implies that f(L/2) _ ir. Reasoning in a similar way on the interval (L/2, L)
and, taking into account that Q takes exactly the opposite values on this
interval, we get that f(L) = 2ir. Hence
i(a) = deg/3 = {f(L) - f(0)} =1,
as we wanted to prove.
An obvious consequence of the turning tangent theorem, combined with
the Whitney-Grauenstein Theorem 9.49, is that each compact simple plane
curve is regularly homotopic to a circle. Another result is an improvement
of Proposition 9.21, where we can now characterize when equality holds.
Corollary 9.51. Let a : III IIS2 be a periodic curve p. b. a.l. with length
L > 0 whose trace is a simple curve. If k is its curvature function, then

f
L
Ik(s)I ds > 2ir
Exercise (20): Differentiating al2 = 1 twice, we get
1 and equality is attained if and only if k(s) > 0 for all s e ][8. (Recall that we
(a,T)=0 and are assuming positive orientation.)
9.4. Rotation index of plane curves 335 28 1. Plane and Space Curves

Hence
1
() = k(a, T) + r(a, B) = a(a, B).
Differentiating one more time, we obtain
1 " 1
=ar(a,N)=-a,
(j)
and we are finished.

Exercise (22): If the trace of a lies in a sphere of radius r > 0, then

I al2 = r2. Differentiating this expression three times, we have
Figure 9.10. The map h
(a, T) = 0, 1 + k(a, N) = 0, and k'(a, NJ - kT(a, B) = 0.
Therefore
1 a a

and consequently r2 - lal2 - \a' T/2 + \a' N/2 + \a' B/2 - (k) + (Tk2)
Conversely, if this last equality occurs, we take derivatives and get, using
the fact that k' 0,
ri(iV'
f
1

aV(0) dr
lim F(sl, s2) = -cV(O).

f
.1
(Si,s2)-+(o,L)
aV(0) dr
This equality is just what we need to prove that ,6' = 0, where /3: I -* ][83
is the curve given by

Hence, F is also continuous at the vertex (0, L) of the triangle T.

Now consider a continuous map h : [0, L] x [0, 1] -* T such that, as Thus, there exists a e R3 such that
indicated in Figure 9.10,
a-a=--N+-
k
-k1 1
T
1 '
B,

which obviously implies

h(s, 0) _ (s,s) if s E [0, L],

(0,2s) if sE [0,L/2},
Ia_a12 - (1)2 ri ()/]2=r2.
J V
(2s-L,L) if s E [L/2, L] Exercise (23): Some previous exercises should have already convinced us
that, if the trace of a lies in a sphere, we have
with h(0, t) _ (0,0) and h(L, t) _ (L,L) for each t e [0, 1]. mom this, we 1 + k(a, NJ = 0 and k'(a, N) - Tk(a, B) = 0.
see that the continuous map H = F o h : [0, L] x [0, 1] -> 1 is a homotopy From the first equation we deduce that
between the loop ry, given by
1 = -k(a, N) kial = rk.
Hence, k(so) = 1/r if and only if a(so) _ -N(so), and we finish by using
s E [0,L] F--- ry(s) = H(s, 0) = a'(s) E Si the second equality.
 334 9. Global Geometry of Curves
Hints for solving the exercises 29

by
Exercise (24): Assume that the plane P is {p e IEg3 I (p, a) + b = 0}, where
a E ][83 is a unit vector and b E R. Consider the function f : I -* Il83 defined a(s2) - a(sl)
by f(s) _ (a(s), a) + b. Then if s1 <52 and (51,52) # (0, L),
Ia(s2) - a(si)I
f(so) = a, f'(so) _ (T(so),a) = 0, and f"(so) _ k(SO) (N(SO),a). F(si, 52) = a'(t) if sl = 82 =
Since f has neither a local maximum nor a local minimum at so, we have
(N(so),a) = 0, that is, a = fB(so). -a'(O) if (81,82) = (O,L),
which is awell-defined map since a is injective on [0, L). Let us check that
this map F is continuous. It suffices to do this along the hypotenuse of T
and at the opposite vertex (0, L). In fact, if (Si, 82) E [0, L] x [0, L],
(9.2)

a(sz)-a(sl) = JO1 d a(sl--r(s2-si)) dr = (82-Si) J1 a'(sl+r(s2-si)) dr.

Hence, if s1 52, we obtain that

f1
J a'(sl + r(s2 - sl))dr
F(sl,S2) _ 1

a'(sl + r(52 - Si)) dr

f
Since the integral appearing in this expression depends continuously on s1
and s2, we have
pl
as(s) dr
lim F(s1, s2) _ J0 1
= as(s),
(SZ,S2) +(S,S)
as(s) dr
0

for each s E [0, U, and, consequently, F is continuous along the hypotenuse

of T. On the other hand, since a is L-periodic,
a(52 - L) - a(sl)
F'(Si, 82) -
I

when (Si, s2) E T and sl <82. Applying equality (9.2) and observing that
s2 - L - sl <0 if (Si, 82) L (0, L), we deduce that
pl
a'(sl +r(s2 - L - sl)) dr
JO
F(Sl,S2) - - i
a'(sl + r(s2 - L - sl)) dr
IJo
9.4. Rotation index of plane curves 333

cE(O) = (0,0) a'(O) = (1,0)

Figure 9.9. The curve a

Now then, since 1(s, t) is constant in s, it follows that both ao and a1 are
constant functions. As a consequence, the tangent maps ao and a1 of the
two curves are constant. Integrating the tangent maps, we would deduce
that these two curves are lines, which would contradict the fact that they are
periodic. Therefore, the curves of the homotopy that we have constructed
are all regular curves. U

The following theorem computes the rotation index for those regular
parametrized plane curves whose trace is a simple curve, that is, whose
trace has no self-intersections.
Theorem 9.50 (Umlaufsatz: turning tangent theorem). Let a : R - 1R2 be
an L-periodic, L> 0, curve p.b.a.l. which is injective on [0, L), i. e., whose
trace is a simple curve. Then, the rotation index of a is either +1 or -1
and

f
L
k(s) ds = 2ir
if k is the curvature of a and a is positively oriented.

Proof. We can assume, after applying a rigid motion of ][82 to the curve a,
that its trace C = a(I8) belongs to the closed half-plane {(x, y) E ][82 y > 0},
that the origin (0,0) lies on C, and that the straight line y = 0 is the affine
tangent line to the curve a at this point; see Figure 9.9. This process
does not modify the rotation index, according to Exercise 9.47. Also using
Exercise 9.45, we can assume that a is p.b.a.l. and that a(0) _ (0,0) and
a'(0) _ (1, 0). The reparametrizations that we may need to make would
change, at most, the sign of i(a).
Having made these assumptions on the curve, we define a map F : T ->
1, where T is the triangle
T= {(sl,s2) E ][82 I 0 < s1 < 82 < L},
 332 9. Global Geometry of Curves

Chapter 2
arising from another periodic curve. Thus, by the definition of the rotation
index, the tangent maps To = c and Ti = a1 are L-periodic and have
the same degree. Using Exercise 9.71 in Appendix 9.7, we check that ao
and a1 are homotopic as L-periodic curves taking values in the circle. Let
F = p o F R x [0, N
1] - S1 be Nthe homotopy
N N
that we have constructed,
Where F = (1 - t) ao + tat and ao and a1 are arbitrary lifts of ao and a1,
Surfaces in .Euclidean respectively. Define H : R x [0,1] - R2 by

Space
PS
H(s, t) = F(u, t) du - L L F(u, t) du, b'(s, t) E ][8 x [0, 1].
J0 J
This map is continuous and satisfies
fs S fL
H(s, 0) = F(u, 0) du - L F(u, 0) du = ao(s) - ao(0),
J J
for all s E II8, and, analogously,
2.1. Historical notes H(s, 1) = al(s) - ai(0), Vs ER.
As happened in many other branches of mathematics, KARL FRIEDRICH Moreover, since F is L-periodic in its first variable, we have
GAUSS (1777-1855) was the first mathematician to undertake a systematic s+L /L
f
Es
study of surfaces in Euclidean space. With his work, he initiated what is now H(s+L,t)-H(s,t) = F(u, t) du - F(u, t) du - F(u, t) du
0 0
called differential geometry (this name is due to L. BIANCHI). GAUSS was,
in the period from 1821 to 1848, a scientific adviser to the governments of
J J
s+L
Hannover and Denmark with the goal of a complete geodetic study of their
territories. Working on this study led him to think about many problems f J
F(u, t) du -

This shows that ao - ao (0) and ai - al (0) are homotopic as L-periodic

0
L F(u, t) du = 0.

arising from the precise determination of a piece of Earth's surface. These

problems could be translated, in a mathematical way, into some questions
about so-called skew surfaces', which were not an entirely new subject since
they had already been considered by EULER, CLAIRAUT, and MONGE. The
task of these geometers who preceded GAUSS focused mainly on the prob-
lem of determining when a given surface could be developed, that is, when
it could be mapped, preserving distances or other geometric characteristics,
onto a plane. This problem had its origin, in some sense, in the ancient
world: stereographic projection was already described in the Geography of
PTOLEMY (c. 150). After Ptolemy, there was nothing new until the appear-
ance of the Mercator projection by G. KRAMER (1512-1594), who knew
that it was a conformal projection. Shortly after the beginnings of calcu-
lus, EULER initiated the theory of surfaces in his Introductio in analysin
infinitorum, published in 1748, and in his Recherches sur la courbure des
1 These are surfaces that are ruled but not developable.
31
curves. From this, it is not difficult to see that ao and al are homotopic
as well. It remains to show that each L-periodic curve of the homotopy is
regular. In fact,
d 1 L
H(s, t) = F(s, t) - - F(u, t) du,
ds L j0
for each s e R and each t E [0, 1]. If, for some pair (s, t), the above deriv-
ative vanishes, since F takes its values in S1, the Schwarz inequality would
force u F--* F(u, t) to be constant, that is, one of the homotopy curves would
be constant. We show that this would lead to a contradiction.
N
From the
uniqueness shown in Proposition 9.66, the curve u F--* F (u, t) would also be
constant and furthermore, by the invariance of degree under homotopies, we
would have deg ao = deg a1 = 0. Thus, c and a1 would be L-periodic func-
tions and, thus, would attain their minima. Suppose that, reparametrizing
one of them, if necessary, by means of a translation, the two curves reach
their minima at s = 0. Then, since t e [0, 1], we have
N N
fr(s, t) = (1- t)ao(s) + tat(s) > 1(0, t).
9.4. Rotation index of plane curves 331
Figure 9.8. Index two curve
when considered as a 2ir-periodic curve, is 1 and that the rotation indices
of the curves
,3(t) _ (cost, sin 2t), `dt E 1[8,
'y(t) _ ((1 - 2 sin t) cost, (1 - 2 sin t) sin t + 2 ), t E ][8,
whose traces appear in Figures 9.7 and 9.8, are 0 and 2, respectively.
The topological content of the rotation index of periodic regular plane
curves is manifestly contained in the following theorem.
Theorem 9.49 (Whitney-Grauenstein's theorem). Let ao, a1 Ilk -p 1R2
be two periodic regular curves. Then, their rotation indices coincide, i. e.,
i(ao) = i(a1) if and only if, up to increasing reparametrizations, ao and a1
are regularly homotopic, that is, there exists a homotopy between the periodic
curves, suitably reparametrized, through regular curves.
Proof. The equality of rotation indices is necessary because of Exercises 9.45
and 9.46, since the existence of a regular homotopy between the curves im-
plies the existence of a homotopy between the respective tangent maps.
Conversely, suppose that i(ao) = i(a1). If we define 3 : Ilk 1R2 by
(A1\
/3(s) = ai A0s
where A0 and Al are the periods of ao and a1, respectively, we have that
3 is an increasing reparametrization of a1 that is Ao-periodic. By Exer-
cise 9.45, its rotation index is the same as that of a1. This shows that,
without loss of generality, we may assume that the two periods coincide,
i.e., that A0 = A1. We may also suppose that both curves have the same
length L > 0 because of Exercise 9.47, which asserts that homothetic curves
have the same rotation index and the fact that two homothetic curves are
always homotopic. Therefore, we may suppose that both are p. b. a.l. since
the reparametrization by the arc length is an increasing reparametrization
32 2. Surfaces in Euclidean Space
surfaces, published in 1767, where, in addition to studying the problem of
the relation between surfaces and planes, he introduced the so-called normal
sections and a first approach to the notion of the curvature of a surface.
Taking into account this earlier work and the fact that in 1795 M0NGE
had also published the first synopsis of the theory of surfaces in Applications
de l'Analyse a la Geometrie, it was GAUSS who first studied surfaces in a
general way in his Disquisitiones generales circa superficies curves, which ap-
peared in 1828. To get an idea of the situation in the study of surfaces when
he began his work, we should point out that only two years earlier, in 1826,
CAUCHY had rigorously proved the existence of the tangent plane at each
point of a surface. In this treatise of GAUSS, among other things, the funda-
mental role of the curvature later called Gauss curvature is already clear,
and glimpses of the distinctions between local/global or intrinsic/extrinsic
aspects of the study of surfaces can already be seen. GAUSS was also the first
geometer who realized that thinking of surfaces as objects endowed locally
with two coordinates is much more convenient for studying some problems
than viewing them as subsets of space whose three coordinates satisfy a
given relation, or even as boundaries of solid bodies.
One year before the death of GAUSS, in 1854, BERNARD RIEMANN
(1826-1866) presented his Habilitation Schrift Uber die Hypothesen welche
der Geometrie zu Grunde Liegen, thus initiating the second great period
of differential geometry by going from the notion of a surface to that of a
Riemann manifold, which has been so fruitful in several fields of physics and
mathematics.
2.2. Definition of surface
In Chapter 1 we applied calculus to study curves. Our explanations were
centred on concepts tangent line, length, curvature, etc. arising from ge-
ometrical intuition, but they involved calculus to be formulated in a rigorous
manner. Now we propose studying, with the same methods, another impor-
tant family of geometric objects: surfaces. One of the main objectives of this
chapter is to transfer the usual analysis on open subsets of the Euclidean
plane differentiability, differential of a function, etc. to surfaces.
The first problem that we have to address is that of defining surfaces.
In Figures 2.1 and 2.2, one can see some examples of objects that could, at
first glance, deserve to be called surfaces. The definition of a surface that
we will choose, and that of course is not the only one possible, is determined
mainly by the tool that we will use to study these objects: calculus. Thus
our theory will include neither Example 5 of Figure 2.2 nor any of the
polygons, so important in geometry, because they have, in a more or less
intuitive sense, a lack of differentiability. We will also throw out Example
2.2. Definition of surface 33
Figure 2.1. Examples 1, 2, 3
lip
Figure 2.2. Examples 4, 5, 6
Figure 2.3. Examples 7, 8, 9
4 of Figure 2.2 because it has a strange point (which one?), Example 6 of
Figure 2.2, because it has an edge-even though it is defined by means of a
differentiable equation-and Example 7 of Figure 2.3, because it intersects
itself. The remaining figures, which bend smoothly in Euclidean space, have
no corners or edges, and do not intersect themselves, are examples of the
type of surfaces that we want to study. All of them share a property that
we choose as our first approach to the final definition.
330 9. Global Geometry of Curves
Figure 9.7. Eight curve
where k is the curvature of the regular curve a, as defined in Remark 1.15.
Therefore
(9.1) 2iri(a) = JoA k(t) Ia'(t) Idt = L k(s) ds.
This formula closely resembles the Gauss-Bonnet J formula proved in part C
of Theorem 8.38 for compact surfaces in 1[83. As in the case of a surface, the
right-hand side of this equality is usually called the total curvature of the
curve a and we will see in what follows that the left-hand side also has a
topological meaning. This equation could have been inferred, in a heuristic
way, from Exercise 1.20.
In the following exercises, which one can easily solve from the results of
Appendix 9.7, we will list most of the remarkable properties of the rotation
index of a periodic regular plane curve.
Exercise 9.45. T (Invariance under increasing reparametrizations.) If f
][8 -> ][8 is an increasing diffeomorphism such that f(s+B) = f(s)+A for each
s E 1[8 and a : ][8 ll82 is an A-periodic regular curve, show that Q = a o f
is a B-periodic regular curve and that i(,(3) = i(a). If f is decreasing, then
i(/3) = -i(a).
Exercise 9.46. T Two A-periodic, A > 0, regular curves ao, al : ][8 -> ][82
are said to be regularly homotopic if there exists a continuous homotopy be-
tween them through periodic regular curves with the same period A. Prove
that, if ao and ai are regularly homotopic, their rotation indices coincide.
Exercise 9.47. T Prove that the rotation index of periodic regular plane
curves is invariant under direct rigid motions and homotheties of ][82 and
that it changes sign under inverse rigid motions.
Exercise 9.48. r Show that the rotation index of the circle
a(t) _ (rcost,rsint), b't ER,
9.4. Rotation index of plane curves 329 34 2. Surfaces in Euclidean Space

Exercise 9.41. T Prove that the ellipse

2 2
C = {(x,y) E 1[82 I a2 + b2 = 1}
with half-axes a > b> 0 is an oval.
Exercise 9.42. T Show that the parabola
C = {(x, y) E I[82 y2 = 2px}
with p> 0 is a simple curve with positive curvature everywhere and is closed
in X82.

Exercise 9.43. Show that, if C is an oval and R is a straight line, we have

R f1 C = 0, or R f1 C = {p} and R is tangent to C at p, or R f1 C consists of
exactly two points; see Exercise (6) of Chapter 6 and Remark 6.3.
Exercise 9.44. T The four-vertices theorem for ovals. Let a : R - R2 be
a periodic curve p.b.a.l. of length L > 0 and curvature function k. Show,
using the Frenet formulas, that
Figure 2.4. Bending a piece of a plane
Definition 2.1 (Intuitive definition of surface). A surface is a subset of
1R3 such that each of its points has a neighbourhood similar to a piece of
a plane that bends smoothly and without self-intersections when bent in
three-space.

f
L
k'(s)(a(s) - a,v) ds = 0
for all a, v E IR2. Use this and the previous exercise to prove that, if the
trace of a is an oval, then there exist at least four points in [0, L) where the
derivative of k vanishes.
9.4. Rotation index of plane curves
Let a : R 1R2 be a periodic regular parametrized plane curve with period
A> 0. Then the map T : Ilk -p §1 C 1R2 taking t E Ilk to the unit tangent
vector of the curve a at the time t, that is,
T(t) =
is also a differentiable map that is A-periodic. We can, then, talk about
its degree see Appendix 9.7 and compare with Definition 8.1 deg T E Z
which is, roughly speaking, the oriented number of times that the tangent
vector of the curve a turns around the circle as we go along the trace of
the curve exactly once. For this reason, we will call this integer the rotation
index of the curve a and we will denote it by
i(a) = deg T = degl.
aI
Since T is differentiable, this degree defining the rotation index can be cal-
culated, by Proposition 9.74, by using
Jac T = det (T, T') = k I a' I ,
Notice that, since we will consider surfaces as subsets, we move away
from the point of view that we chose for studying curves. In order to arrive
at a formal definition of a surface, we need to rewrite the idea of a plane
region bending smoothly in the space into more rigorous language. To do
this, we will argue by analogy with the one-dimensional case. Based on what
we learned in Chapter 1, we may replace the expression line segment which
bends smoothly and without self-intersections by the following:
An open interval I and a map a : I - k3 that is a differ-
entiable injective map with a'(t) 0 for all t E I.
Thus, analogously, we might translate the expression piece of a plane which
bends smoothly and without self-intersections in the following way:
A non-empty open set U C Ii 2 and a a differentiable injec-
tive map X : U -> R3 whose differential (dX )q : R2 -} R3
is injective for all q E U.
We now introduce the fundamental notion of this chapter.
Definition 2.2 (Surface). We will say that a non-empty set S C 13 is
a surface if, for each p E S, there exist an open set U C 1R2, an open
neighbourhood V of p in 5, and a differentiable map X : U -> 13 such that
the following hold.
is X(U) = V.
ii: X : U - -3 V is a homeomorphism.
iii: (dX)q : IE82 -> 13 is injective for all q E U.
2.2. Definition of surface 35
Remark 2.3. We can write X in terms of its components as X (u, v) =
(x(u, v), y(u, v), z(u, v)). Then, saying that X is differentiable is equivalent
to saying that the three functions x, y, and z are differentiable.
Remark 2.4. If X is differentiable and satisfies condition i, then X : U - V
is continuous and surjective. Thus, to check ii, it is enough to see that X is
injective and that X1 is continuous. Condition ii is, in fact, a somewhat
technical requirement. For the moment, it suffices to note that this condition
forces X to be injective.
Remark 2.5. The fact that (dX)q is injective for all q E U is equivalent to
requiring that the vectors
X(q)
ax and
ax
- 8u 9)
be linearly independent.
Remark 2.6. The maps X appearing in Definition 2.2 will be called param-
etrizations of the surface S, and their variables u, v will be called local coordi-
nates of S. The curves obtained from X by fixing one of its variables are the
so-called coordinates curves. These parametrizations should be thought of as
away of introducing coordinates on a region of the surface or, equivalently,
a way of drawing a map of that region. This is why the term coordinate
neighbourhood is a common name for the image of such a paxametrization.
Example 2.7 (Planes). Set S = {(x, y, z) E I[83 ax + by + cz = d} with
(a,b,c) L (0,0,0). If c L 0, we can put S = {(x,y,z) E I[83 z = Ax +
By + C}. We define a differentiable map X II82 -p II83 by X (u, v) _
(u, v, Au + By + C). Then X (I[82) = S, X is a homeomorphism whose
inverse map X-1 : S -p I[82 is given by X-1(x, y, z) _ (x, y), and its partial
derivatives Xu = (1,0, A) and X _ (0,1, B) are linearly independent at
each point. Notice that this surface is covered by a single paxametrization.
This does not happen, in general, for a given surface.
Example 2.8 (Graphs). Let U C I[82 be an open set, and let f : U - I[8 be
a differentiable map. The graph of f is the following subset of I[83:
S = {(x,y,z) E ][83I (x,y) E U, z = f(x,y)}.
In this situation, the map X : U - S, given by X (u, v) _ (u, v, f(u, v)), is
a parametrization with X (U) = S. Therefore, S is a surface that is covered
again by only one coordinate neighbourhood.
Example 2.9 (Open subsets). If S is a surface and Si C S is a non-
empty open set, then Si is also a surface. To check this, it is enough to
observe that, if X : U - S is a parametrization of S with X (U) fl Si Ql,
then X X -1(Sl) -p Si is a parametrization of Sl. As a consequence,
328 9. Global Geometry of Curves
9.3.5. Positively curved simple curves. In Chapter 6, we studied some
of the features of closed or compact surfaces with positive Gauss curvature
and their connection to the theory of convex sets. With these precedents in
mind, it is natural to use the term ovals for the compact connected simple
plane curves whose curvature does not vanish anywhere or, equivalently, that
has positive curvature everywhere; recall that we are considering positive
orientations and see Exercise 9.20. The concept of a convex curve is with
respect to the above notion of an oval.
Definition 9.39 (Convex plane curve). A simple plane curve will be called
convex when it lies entirely on the closed half-plane determined by the affine
tangent line at each of its points. Also, it will be called strictly convex when
it is convex and touches each of its affine tangent lines at only a point.
In this sense, Exercise (8) of Chapter 1 says that an oval is always locally
strictly convex and that each convex curve has non-negative curvature, with
respect to the positive orientation. Note that it also makes sense to talk
about convex curves in the context of parametrized curves, with this same
definition.
As in the case of surfaces, we will be able to obtain results, not only on
compact curves, but on curves closed in Euclidean space. If C is a connected
simple curve closed in R2 with positive curvature everywhere, we will denote
by 11 the domain of R2 -c to which the normal vector points, and we will call
it its inner domain. Some of the results that we have established concerning
closed positively curved surfaces in the Hadamard-Stoker Theorems 6.1 and
6.5 and in the Bonnet Theorem 6.8 can also be proved, using the same
methods, for simple plane curves with positive curvature. We summarize
them in the following theorem.
Theorem 9.40. Let C be a connected simple plane curve with positive cur-
vature everywhere that is closed as a subset of 1[82. Then the following hold.
A: Its inner domain 11 is convex.
B: C is strictly convex.
C: If C is compact with length L> 0 and a : ][8 -> 1[82 is a parametriza-
tion b. a.l. of C, then the normal vector N : 1[8 -> S1 defines an
L-periodic map injective on [0, L).
D: If C is not compact and a : 1[8 -> ][82 is a parametrization b.a.l. of
C, then the normal vector N : ][8 -> S1 is injective and C is a graph
over an open interval of a straight line of ll82.
E: If k(p) > 0, then C is an oval.
9.3. Results already shown on surfaces 327 36 2. Surfaces in Euclidean Space

By Example 9.27, we have that each connected component of a surface is also a surface, because connected
components of a locally connected space are open.
f
L
f(p,s)ds=2ir, b'p E SZt. Example 2.10. Let S C I[83 be such that S = USi, where each of the
, then S is a surface.
SZ is an open subset of S. If Sz is a surface for all i E
Integrating over Stt and applying Fubini's theorem, we have
L Example 2.11 (Spheres). Let S2 = {p E I[83I ll2 = 1} be the sphere centred
2irA(Stt) _ f(p,s)dpds, 0 < t < s. at the origin with unit radius. Denote by U = {(u, v) E 1[82 u2 + v2 < 1}
o sit
the unit disc of the Euclidean plane and by f U -p I[8 the differentiable
Utilizing Exercise 9.29 for the curve at with a = a(s) and b = N(s), this function defined by
equality becomes
L L (at(u) - a(s), N(s)) (at(u) - a(s), Ja't(u)) f(u, v) = 1 - u2 - v2, for all (u, v) E U.
2irA Stt) - _ duds. For a E we consider the open neighbourhood of a in 2 given by Va = {p E
Jo Jo I at(u) - a(s)2 §21 (p, a) > 0}. Choosing the coordinate system of I[83 so that the vector a
Notice that, using the Schwaxz inequality, the integrand on the right-hand has components (0,0, 1), we have that VQ is the graph of the function f and,
side has absolute value bounded by 1+ e max k Taking limits as t -> 0, the
i.
from Example 2.8, Va is a surface. Since S2 = UaS2 Vd, using Example 2.10,
required equality follows from the dominated convergence theorem applied we see that the sphere S2 is a surface.
to the two sides of the equality. O
Example 2.12 (Diffeomorphic images). Let 01, 02 C ][83 be two open sets
The following exercises will be useful for completing the proof of the and q: Ol -p 02 a diffeomorphism. If Si C Ol is a surface, then S2 = q(Sl)
isoperimetric inequality in the plane as a consequence of only the divergence is also a surface; see Exercise 2.45. A proof of this fact consists simply in
theorem. realizing that, if X U - Si is a paxametrization of Si, then the map
o X : U -p S2 is a parametrization of S2.
Exercise 9.37. Let u, v, w E ][82. Prove that
Consequently, any ellipsoid is a surface, since there is an affine transfor-
-4 (u, v) (u, w) (u, v - w)2 mation of I[83 taking it into 2. In particular, spheres of arbitrary centre and
and with equality holding if and only if u is perpendicular to v -I- w. radius are surfaces.
Exercise 9.38. T Let a : I - ][82 be a plane curve p.b.a.l. Suppose that, Example 2.13 (Inverse images of a regular value). Let O C I[83 be an open
for each s, u E I, we have (a(s) - a(u), N(s) -F N(u)) = 0. Prove that a set, f O -p I[8 a differentiable map, and a E 1[8. We will say that a is
is a segment of a straight line or an arc of a circle. Interpret the above a regular value of f if, for each p e O with 1(p) = a, one has (df)p L 0
hypothesis geometrically. or, equivalently, (Vf) (p) L 0. We will need the following classical result of
calculus:
From the integral formula obtained in Lemma 9.36 for the area of the
Theorem 2.14 (Implicit function theorem). Let O be an open set of I[83,
domain Sl, along with Exercise 9.37 above, the inequality
p = (xO, yo, zo) E O, a E I[8, and let f : O -p I[8 be a differentiable function.
L
t<1
1L
(a(u) - a(s), N(u) - N(s))2 Suppose that f(p) = a and f() _ (Df/Dz)(p) L 0. Then there exist an
2irA S) - 4 Jo Jo a(u)-a(s)12
duds
open neighbourhood U of (xO, yo) in I[82, an open neighbourhood V of zo in I[8,
follows and equality is attained, according to Exercise 9.38, if and only if a and a differentiable function g : U -p V such that U x V C O, g(xo, Yo) = zo,

4 /'L
J
fL Iduds
is a circle. Using the Schwarz inequality, we have and
{p E U x VI f() = a} _ {(x,y,g(x,y)) E 1[83 (x, y) E U}.
In other words, if f(p) 0, then we can solve the equation f(x, y, z) = a
for the unknown z near p. A direct application is
L
ZJ
J
(1 - (N(u),N(s))) duds = 2L2, Proposition 2.15. Let a E 1[8 be a regular valve of the differentiable function
f : O -p I[8, where O is an open set of I[83. Then, if S = f({a}) is non-
where we have also taken Exercise 9.28 into account to get the last equality. empty, it is a surface.
2.2. Definition of surface 326 9. Global Geometry of Curves
37

Proof. Take p = (xO, yo, zo) E S. Then f(p) = a and (df)P L 0. Relabeling
the axes, if necessary, we can assume that f() L 0. Using the same
notation as in Theorem 2.14, we have that S fl (U x V) is a neighbourhood
of p in S, which is the graph of a differentiable function. We complete the
proof by taking Examples 2.8 and 2.10 into account.
Example 2.16 (Quadrics). Let A be a symmetric matrix of order four. We
consider the associated quadric
This theorem, as mentioned before, can be utilized to answer the isoperi-
metric question for the plane. In this way, we obtain the following result.
Theorem 9.34 (Plane isoperimetric inequality). Let C be a compact con-
nected simple curve of length L and let SZ be its inner domain. Then
4irA(St) < LZ
and equality occurs if and only if C is a circle.

S = {p E ll83I(1, pt)A (p ) =0}.
Suppose that S is non-empty and that, for each p E I[83, we have (1, pt)A L 0;
that is, all the singular points of the quadric lie in the plane at infinity. We
next show that S is a surface in this case. For this, define a differentiable
function f : II83 -p II8 by
Proof. This proof is the same as the proof of Theorem 6.25 unless the
following is true. Once the inequality is established for each curve C sat-
isfying our hypotheses, we are able to conclude that any curve reaching
the minimum isoperimetric quotient must have constant curvature. Then,
to complete the proof, we appeal to Exercise 1.20 instead of Alexandrov's
Theorem 6.17.

f(p) _ (l,pt)A I
p 1 In the case of the plane, many interesting proofs of the isoperimetric
inequality have been given. This is probably because this problem has been
For any p, v E ][83, we have one of the most famous ones in geometry, going back to its ancient begin-
0 1
nings, which even involves Greek and Phoenician myths, and its relatively
(df)(v) = 2(1,pt)A recent solution. In Exercises (12)-(15) in this chapter, we will outline a
\ V J
proof by Hurwitz that does not utilize the Brunn-Minkowski inequality. In
If (df)p = 0 for some p E S, we deduce that (1, pt)A = (A, 0) with A E I[8. the following remark, we will sketch another proof having the charm of being
Also, since f(p) = 0, we see that A = 0, which would imply that (1, pt)A = 0, a mere consequence of the divergence theorem.
which is a contradiction. Hence, S is a surface, as Example 2.13 shows.
Remember that the real quadrics having all their singular points in the Remark 9.35. We will begin by giving an integral expression for the area
plane at infinity are just ellipsoids, hyperboloids, paraboloids, and cylinders. of the inner domain determined in the plane by a compact connected simple
curve.
Example 2.17 (Torus of revolution). Let S1(r) be a circle of radius r > 0.
If we rotate this circle around a straight line in the same plane that it Lemma 9.36. Let a be a periodic plane curve p. b. a. 1. with length L whose
belongs to, at distance a, a > r, from the centre of the circle, we generate trace is a simple curve and let 11 be its inner domain. Then
a geometrical locus that we will call the torus of revolution. If we place the
coordinate axes of I[83 so that the circle S' (r) lies in the plane x = 0 with 2irA(St) _ -
(a(u) - '(s), n'(w)) (a(s) a(u), N(s))
duds.
Ia(u) - a(s)12
centre at the point (0, a, 0) and if the rotation axis is the line x = 0, y = 0,
then the torus of revolution becomes Proof. In fact, let s > 0 sufficiently small so that we can consider the
S = {(x,y,z) E 1[83 ( x2 + y2 - a)2 + z2 = r2}. parallel curves-following Exercise 9.19-at(s) = a(s) + tN(s) with s E ll8
and ti < e. Then, each at is a periodic parametrized curve whose trace
Consider the differentiable function f I[83 - {z-axis} -p I[8 defined by is a simple curve and is such that at(s) _ (1 - tk(s))T(s), where k is the
-
f(x, y, z) _ (/'2 -I- y2 a)2 -I- z2. Then S = f'({r2}). Furthermore, curvature function of a. Let Stt be the inner domain determined by at.
Since at(1R) C S2, we have lit C S2. We consider the continuous function
Elf 2X(/X2 + y2 - a) Elf 2y (x2 + y2 - a) af
xz + z ay - x2 + 2 and az = 2z. f : Stt x [0, L] -> ][8 defined by
ax '

/(p,s) _
(p-a(s),N(s))
Hence, for each p E S, one has (df) L 0, and so S is a surface.
IP - a(s)I2
9.3. Results already shown on surfaces 325 38 2. Surfaces in Euclidean Space

for each p E SZ. Show that

(p-a,b) _ /'A (a(t) - a, b) (a(t) - a,
Js Ip - o dt,

dp Ia(t) -
using the divergence theorem and Remark 9.25.
Exercise 9.30. T Let A C ][82, and let f, g A - ll8 be differentiable
functions. Prove that div JO f = 0 and that div gJVf = det(0 f , Vg).
Exercise 9.31. T Let C be a compact connected simple plane curve and let
St be its inner domain. Let F = (Fl, F2) : S2 - ll82 be a differentiable map
such that Fc = Ic. Show that

f
Jac F = A(St).
Exercise 9.32. Show from the previous exercise that, if 11 is the inner
domain determined by a compact connected simple plane curve C, there
exist no differentiable maps F : SZ - 1R2 With F c = lc and F(1) C C.
That is, there are no differentiable retractions from 11 to its boundary.
Exercises 9.30, 9.31, and 9.32 above, all pertaining to the divergence The-
orem 9.23, provide a proof of the two-dimensional version of the Brouwer
fixed point theorem, along the lines of the three-dimensional case in Theo-
rem 5.41.
9.3.4. The isoperimetric inequality. In the case of three-dimensional
Euclidean space, the solution of the isoperimetric problem was postponed
until Alexandrov's Theorem 6.17 was proved. The reason was that the
equality in the Brunn-Minkowski inequality of Theorem 6.22 was not signif-
icant and we saw that surfaces attaining the minimum of the isoperimetric
quotient have constant mean curvature since they are critical points of the
area with respect to preserving volume variations. The corresponding ques-
tion to be solved now is: among all compact connected simple plane curves
with prescribed length L > 0, which determines an inner domain with the
greatest area? The solution to this problem-and to others of an analogous
nature-is given by the isoperimetric inequality for the plane. As a first step
to stating it, we point out that the same proof as in Chapter 6, with suitable
minor modifications, will give us the plane version of the Brunn-Minkowski
inequality.
Theorem 9.33 (Brunn-Minkowski's inequality in the plane). Let A and B
be two bounded open subsets of the Euclidean plane ][82. Then
(areaA)112 + (areaB)1"2 < (area (A + B))1!2,
Figure 2.5. Torw of revolution
Example 2.18 (Simple curves). We will use the name simple curve for
a one-dimensional object of the Euclidean space I[ 3-or of the plane R 2-
analogous to a surface- That is, a non-empty subset C of 1R3 will be a simple
curve when each of its points p E C possesses an open neighbourhood V
in C for which there exists a regular curve a : I -> R3 defined on an open
interval I of the real line mapping I homeomorphically onto a(I) = V. Each
of these regular curves a will be called, analogously to the case of surfaces,
a parametrization of the simple curve C. By following what we have learned
in Chapter 1, we can choose such parametrizations to be curves p.b.a.l. We
will prove the following in Theorem 9.10:
Any connected component of a simple curve in 1R3 or 1R2 is
homeomorphic either to a circle or to R, depending upon
whether it is or is not compact.
Moreover, the entire curve C is the image of a curve a : I C III ---> 1R3
p.b.a.l. In the first case, a can be chosen to be periodic, and in the second
case, it can be chosen to be a homeomorphism from I onto C. In fact,
straight lines and circles are the simplest examples of simple curves.
Exercise 2.19. 1 Prove that the cone C = {(x, y, z) (=- 1831 z2 = x2 + y2} is
not a surface.
Exercise 2.20. Show that compact surfaces and compact simple curves
have finitely many connected components.
2.3. Change of parameters
If a point p of a given surface belongs to the ranges of two different parametri-
zations, we would have two different systems of coordinates near this point.

Our purpose now is to find the properties of the map taking coordinates of
where the sum of subsets of ll82 is understood in the sense of Section 6.5. one parametrization into another.
2.3. Change of parameters 39 324 9. Global Geometry of Curves

Lemma 2.21. Let S be a surface and let X : U -p S be a parametrization Example 9.27. If a is a curve p.b.a.l. With length L > 0 that positively
whose image contains the point p. Let po E U be such that X (po) = p. Then orients a compact connected simple curve C and 11 is the inner domain of
there exists an open neighbourhood V of po and an orthogonal projection C, the vector field
I[82 onto one of the three coordinate planes of I[83 such that W = a
I[83
(is open in I[82 and such that 7r o X : V - W is a diffeomorphism.
X (p) - -
1p

is differentiable on SZ if a 12. One easily checks that

Proof. We explicitly write the components of the parametrization X as
X (u, v) _ (x(u,v),y(u,v),z(u,v)). (dX)(v)(p-a
Then the matrix
xu xv for all v E ][82. Hence, X has null divergence everywhere. As a consequence,
(dX)0 = yu yv (P0)
L (a(s)- a, N(S)) ds = 0, b'a
zu zv J
0
has rank two and, thus, has a non-null minor of order two. We will assume
that this minor is formed just by the first two rows of the matrix. Let Now suppose that a e 11. Since 11 is open, there exists s > 0 such that
7r : be the orthogonal projection onto the plane xy, which is given the closed ball B6 (a) with centre a and radius s is contained in 11. The set
by 7r (x, y, z) = (x, y). Then the composite map 7r o X is a differentiable map SZ' = 11 - B6 (a) is a bounded connected open subset of the plane. More-
of U to 1R2 and over, its boundary is a (Ilk) U S (a). Thus, 1' satisfies the requirements of
d( r o X)PO - (xu
Yu
xv I (p0)
yv /
Remark 9.24. Since the vector field X defined earlier is differentiable on SZ'
and has divergence zero, we have
is regular. Applying the inverse function theorem, we find a set V C U
verifying the required assertion. 0= L (a(s) - a' N(S)) ds + (/3(s) - a, NQ(S)) ds,
Jo I
2 o 1
2

Remark 2.22. Observe that Y = X o (7r o X)-1 : W -p S is a parametriza-
tion of S that covers p and that r o Y is the identity map on W. That is,
one can always cover a given point on a surface by a paxametrization that is
the inverse map of an orthogonal projection onto a coordinate plane. From
that, we deduce immediately the following result.
Proposition 2.23. Let S be a surface and take a point p E S. We can find
a parametrization of S near p whose image is the graph of a differentiable
function defined on one of the three coordinate planes of ll83.
Exercise 2.24. T Demonstrate that the one-sheeted cone defined by C =
{(x,y,z) E I[83i z2 = x2 + y2, z > 0} is not a surface.
Exercise 2.25. T Let S be a surface and let p be one of its points. Show
where ,(j(s) = a + e (sin(s/s), cos(s/e)) is the circle centred at a with radius
e oriented so that the normal vector Np(s) _ (1/s)(/3(s) - a) points to 52'.
Computing the second integral explicitly, we get
1L (a- a(s),N(s))
ds = 2ir, Va Eli.
0 la - alsll`
Notice that this equality is a two-dimensional version of the well-known
Gauss theorem of electrostatics that we have already seen in Remark 5.34.
Exercise 9.28. T Let a be an L-periodic plane curve p.b.a.l. whose trace is
a simple curve, and let 11 be its inner domain. Apply the divergence theorem
to prove that

f (N(s),a)ds=0,
L
that there are an open set O C 1[83 such that p E O and a differentiable map b'a E ][82.
f : O -p I[8 having 0 as a regular value in such a way. that S n O = f'({O}).
Exercise 9.29. Let a be a regular periodic plane curve with period A > 0,
We will now investigate the following situation. Let S be a surface and let
not necessarily parametrized by the arc length, whose trace is a simple curve,
XZ : Ui - S, i = 1, 2, be two parametrizations of it such that O = X1(U1) fl
and let SZ be its inner domain. If a SZ and b E ][82, let X be the differentiable
XZ(UZ) is non-empty. Then the map h = X2 1 o Xl : Xi 1(O) -p XZ 1(O) is
vector field defined on SZ by X(p) _ (p - a, b) (p - a)llp - a12. Prove that
a homeomorphism taking coordinates with respect to Xl into coordinates
with respect to X2. We will say that h is a change of parameters or a change
of coordinates.
(div X)(p) _ `Ip _a2
9.3. Results already shown on surfaces 323 40 2. Surfaces in Euclidean Space

where a : I[8 -> ][82 is a positively oriented periodic curve p. b. a.l. with period
L> 0 whose trace is the simple curve C.
Remark 9.24. We have given the divergence theorem for domains St deter-
mined by a compact connected simple curve in the plane. In fact, the proof
is also valid for those bounded connected open subsets of the plane whose
boundary consists of finitely many disjoint simple curves of this type. That
is (compare with Remark 5.33), we have the following statement.
If SZ is a bounded connected open subset of l[82 such that
Bdry SZ = U=1 Ci, where each CZ is a compact connected
simple curve, and X is a vector field defined on SZ, then
Figure 2.6. Change of parameters
Li

J
div X =
-J
i10
(X(cj(s)),N(s)) ds,
Theorem 2.26. Changes of parameters are difeomorphism s.
where each curve ai p.b.a.l. orients Ci in such a way that
its normal vector Ni(s) points to SZ for all s E [0, LZ]. Proof. Since the change of parameters h is a bijective map and its inverse
Remark 9.25. Let a ][8 -* I[82 be a periodic parametrized curve with map is another change of parameters, to show the theorem, it is sufficient
period A > 0, not necessarily p.b.a.l., whose trace is a simple curve. We to see that h is differentiable at each point of its domain X1 1(O). Let
know that, if b : ][8 -> I[8 is the inverse diffeomorphism of the arc length qi E Xz 1(O) with X, (qi) = p E S, i = 1, 2. Applying Lemma 2.21., we find
function S of a, the curve ,Q = a curve p.b.a.l. with length L = S(A). an open set V C X2 1(O) and a projection ir, which we may assume to be
If S2 is the inner domain determined by ,3(][8)-or by a(I[8)-and X is a field onto the plane xy, such that 7r o X2 : V -r (7r o X2) (V) is a diffeomorphism.
defined on S2, we have, by the divergence theorem, Then h-1(V) is an open neighbourhood of q1 and, in this neighbourhood,
f fL h = (ir o X2) - l o it o X1. Thus, h is differentiable on h -'(V) since it is the
I divX=- Io (X(/3(s)),N,(s))ds. composition of the following differentiable maps: X1 : h-1(V) C 1R2 R3,
7r :R3 R and (7r o X2)-1
: (1r o X2)(V) C R2 , V C 1R2.
Now, taking into account that ,Q(s) = a(cb(s)) and that Na(s) = J13'(s) _
and applying the change of variables theorem to the integral
on the right-hand side for the diffeomorphism b : (0, L) -* (0, A), we see 2.4. Differentiable functions
that f fA We said at the beginning of this chapter that one of our goals is to transfer
div X=- (X(c(t)),JcV(t))dt, calculus from Euclidean open sets into surfaces. In particular, we want to
Jo
J of the divergence theorem when we use an arbitrary
which is the expression make clear what it means for a function defined on a surface to be differen-
parametrization for the given simple curve. tiable. The idea is simply to use parametrizations to reduce the concept of
differentiability on a surface to that of differentiability on R2.
Example 9.26. Let a : Il8 -> ][82 be a periodic curve p.b.a.l. with length L>
0, whose trace is a simple curve and let SZ be the inner domain determined Definition 2.27 (Differentiable map). Let S be a surface and let 0 C R'
by a(][8). Consider the vector field given by X(p) = p, for all p E S2, that be an open set.
is clearly differentiable. Since (dX)P = I1R2 for all p, we have that div X is A: A function f : S -+ lR''b is said to be differentiable if, for any
always 2. Applying the divergence theorem, we obtain that parametrization X : U --- S of S, the composition f o X : U -} R?z
is differentiable.
(c(s),N(s)) ds,
B: A map f : 0 -} S will be differentiable if i f : 0 -> 1R3 is
where A(St) is the area of SZ. differentiable, where i : S - R3 is the inclusion map.
2.4. Differentiable functions 41 322 9. Global Geometry of Curves

C: If 51 is another surface, we will say that a map f S -p S1 is Proposition 9.21. Let a : ][8 ][82 be a periodic plane curve p. b. a.l. with
differentiable when i1 o f : S - R3 is differentiable in the sense of length L> 0. If k : ][8 -> ][8 is its curvature function, then the following hold.
A, where i1 represents the inclusion map of Si in
f
L

Exercise 2.28. Show that one can deduce from A, B, and C of Defini-
tion 2.27 that a differentiable map is necessarily continuous.
Exercise 2.29. Let f, g : S -p IIIm be differentiable vector-valued functions
and let A E III. Prove that the functions f -I- g, f - g, A f : S IIIm, and
(f, g) : S - III are also differentiable. If m = 1 and g(p) L 0 for all p E S,
check that f/g: S - R is also differentiable.
In what follows, we will introduce the differentiable functions defined
on or taking their values in surfaces that will appear frequently later in the
book.
Example 2.30 (Restrictions). Let O C III3 be an open set and let F
O -p Rm be a differentiable map in the sense of calculus. If S C O is a
surface, we immediately see that the map f : S IIIm defined by f = FMS is
differentiable in the sense of Definition 2.27 A. In particular, the inclusion
i : S -p R3, the identity map fps : S -p S, and the constant maps defined
on S are differentiable. In the following examples we will introduce other
important functions on surfaces that are also differentiable since they are
restrictions of differentiable maps.
Example 2.31 (Height function). Let P be a plane of III3 passing through
the point p0 with normal vector a E III3 - {0}. If S is any surface, then the
A k+(s) ds > 2ir, where k+ = max{0, k}.
L
Bf Ik(s)I ds > 2ir and, if equality holds, then the curvature of the
curve a is non-negative.
With regard to this statement, it is worth noting that there are peri-
odic plane curves with non-negative curvature everywhere with respect to
the positive orientation for which the equality in B does not happen, for
example, the curve y in Exercise 9.48.
9.3.3. The divergence theorem. As in the case of compact simple curves
we can also obtain, as a consequence of the length formula, the corresponding
divergence theorem, this time referring to integrals over the plane domains
in R2 determined by this class of curves. To do this, we make, as in the three-
dimensional case, the following definition (compare with Definition 5.30) :
Definition 9.22 (Divergence of a vector field). Let A C R2 and let X
A - R2 be a vector field defined on A. The divergence of X is the function
div X : A -f R given by
(div X) (p) =trace (dX).
Then,
2

function h : S -p III, defined by (div X) (p) _ ((dX)(e), e),

i=1
h(p) _ (p - poi a), by E S, where {ei, e2} is any orthonormal basis of I[82. If we choose el = (1, 0), e2 =
is differentiable because it is the restriction to S of a differentiable function (0, 1), then
defined on the whole space I[83. When al = 1, h measures the (oriented) (iax1()ax2 ()a (p)\

(dX) (ei) _ y2
a s (p)) and (dX) (e2) _ (a yl ,
height from the points of S to the plane P.
Example 2.32 (Square of the distance). If po is any point of I[83 and S is where X = (X1,X2). Therefore,
a surface, then the function f : S -p I[8, given by
div X = ax +
axl axe
f(p)=Ip-pol2, by E s, ay
and, in particular, div x is a differentiable function.
gives the square of the Euclidean distance from the points of S to po. For
the same reason as in Example 2.31 above, f is differentiable. Theorem 9.23 (Divergence theorem). Let C be a compact connected simple
curve in ][82. Suppose that SZ is the inner domain determined by C and that
Example 2.33 (Distance function). Take po E 1[83 -5, where S is a surface. x is a differentiable vector field defined on ft Then
Then the function f : S -p I[8, given by f fL
div x=-I (X (a(s)), N(s)) ds,
1(P) = I p - po I , dp E s, J o
9.3. Results already shown on surfaces 321
T(s)
Figure 9.6. Positive orientation
orientation to C. If not, we will consider the new curve 3 : IR -> 1R2 given
by /3(s) = a(-s). Then 0 is parametrized by are length, is L-periodic, and
/3(R) = C. Since the Frenet dihedra of the two curves are related by
Tp(s) = -Ta(-s) and Np(s) = -Na(-s),
we have that Q is positively oriented instead. Consequently, a compact con-
nected simple curve can always be reparametrized so that it is positively
oriented. From now on, we will always assume that simple curves homeo-
morphic to a circle are oriented in this way, unless otherwise specified.
Exercise 9.18. Following the guidelines of Section 4.5, define tubular neigh-
bourhoods and show their existence for compact plane simple curves. After
that, define the notion of a parallel curve, by analogy to that of the parallel
surface, as seen in Remark 4.29.
Exercise 9.19. ? If a : R - JR2 is a periodic curve p.b.a.l. with period
L > 0 and whose trace is a simple curve C, compute the curvature and the
length of its parallel curves in terms of those of C.
Exercise 9.20. T If C is a compact connected simple plane curve, show
that there are points with positive curvature with respect to the positive
orientation.
9.3.2. Sard's theorem, the length formula, and some consequences.
As in the case of surfaces, in order to prove the Jordan curve Theorem 9.14,
we need a version of Sard's theorem for differentiable maps of IR to R. Of
course, this one-dimensional version exists and an analogue to the proof
given in Chapter 4 remains valid for this case. As a consequence, we will
also have a corresponding area formula-the length formula, in this case. A
direct consequence of this formula is the following analogue to the Chern-
Lashoff Theorem 5.29 proved in Chapter 5.
42 2. Surfaces in Euclidean Space
measures the Euclidean distance from the points of the surface S to the point
po and is differentiable since it is the restriction to S of the differentiable
function p H p - poI defined on the open set II83 - {Po} including S.
Example 2.34 (Square of the distance to a straight line). Let R be a
straight line of I[83 through the point po with the unit vector a as a direction
vector. The function f : S -p I[8, given by
f(p) = IP -ioI2 - (p-po,a)2 = I
for each p E S, measuring the square of the distance from the points of S to
the line R, is differentiable.
Example 2.35 (Differentiable curve on a surface). If a I -p I[83 is a
differentiable curve in I[83 whose trace belongs to a surface S, then the map
a : I - S is differentiable according to Definition 2.27, part B. We say that
a : I -p S is a (differentiable) curve on the surface S.
The following important result shows that to use Definition 2.27, part
A, of a differentiable function of a surface into a Euclidean space, we do not
need to think about all the parametrizations of S.
Lemma 2.36. Let S be a surface and let f : S -p I[8'n be a function defined
on S. If, for each p E S, there exists a parametrization Xp : Up -p S such
that p E X(U) and such that f o XP : U- I[8"'z is differentiable, then f is
differentiable in the sense of Definition 2.27.
Proof. Take any parametrization X : U -p S of S. We have to prove that
f o X : U I[8'n is differentiable. We now show that this composition is
differentiable at any point q e U. By hypothesis, for the point p = X (q) E S,
there is a parametrization Xp : U-p S such that p E X(U) and f o Xp
Up I[8"` is differentiable. Let O = X (U) fl X(U), which is a non-empty
set, since it contains p. Then X -1(O) is an open neighbourhood of q and,
moreover, we know that the change of parameters h : X -1(O) -p X;1 (O) is
a diffeomorphism. Since we have f o X = (f o Xp) o h, on X-1(O), we are
done. O
Exercise 2.37. Let X : U -p S be a parametrization of a surface S. Prove
that the maps X : U -p S and X -1 : X (U) -p U C are differentiable,
according to Definition 2.27.
A property that we should expect, if our definitions of differentiability
are adequate, is that compositions of differentiable maps are differentiable.
In fact, we have
Theorem 2.38. Suppose that f : Si -p S2 and g : S2 -p S3 are differentiable
maps, where Si, S2, S3 are either surfaces or open sets of Euclidean spaces.
Then the composite map g o 1 : Si - S3 is also differentiable.
2.4. Differentiable functions 43
Proof. If 82 is a Euclidean open set, the result is a direct consequence of
Definition 2.27. If, instead, 82 is a surface, it suffices to show it in the case
where Si = O is an open subset of IISn and 83 = IIgm is a Euclidean space.
The remaining cases can be derived from this one and from our definitions.
Let q E O and let X : U -p S2 be a parametrization covering p = f(q)
such that 7r o X = ICU, where 7r : II83 IIg2 is the orthogonal projection
onto some coordinate plane; see Remark 2.22. Hence f -1(X (U)) is an open
neighbourhood of q in which
gof = (goX)o7rof,
and this is a composition of differentiable maps between Euclidean open
sets. Then, g o f is differentiable at each point q E O and, thus, we obtain
the required result.
Example 2.39. Let a : I -p 51 be a differentiable curve on a surface S1
see Example 2.35 and let f : Si -p 82 be a differentiable map of Si into
a surface or Euclidean open set 82. Then f o a : I -p 82 is a differentiable
curve on S2.
Example 2.40. If f : S - IIgm is a differentiable map defined on a surface
S and Si C S is open, then fis1 : S1 II8'n is differentiable.
Example 2.41. Suppose that a surface S is a union S = UzEI SZ, where
each SZ is open. If f : S -p II8m is a map such that each fs2 : SZ -p II8m is
differentiable, then f is differentiable.
Example 2.42. Let U C j2 be an open set of the plane. By identifying
IIg2 with a plane in R3, we can think of IjS2 as a surface and, so, U is one as
well. Take a function f : U - IIgm. Then, f is differentiable in the sense of
Definition 2.27 as well as in the usual sense.
Once we have the notion of a differentiable map between surfaces, we
can define the concept of differentiable equality for these figures. In fact, the
notion of a continuous map between topological spaces determines that of a
homeomorphism between them. In a similar way, a diff eomorphism between
two surfaces Si and S2 will be, by definition, a differentiable bijective map
q : Si -p S2 whose inverse : 82 -p Si is also differentiable. For in-
map-1
stance, the identity map I S : S -p S on a given surface is a diffeomorphism,
and, if : Si -p S2 is a diffeomorphism between surfaces, then its inverse
map-1
: S2 -p Si is a diffeomorphism as well. Two surfaces will be said
to be diffeomorphic when there exists a diffeomorphism between them. In
this case, we immediately see that they are homeomorphic in fact, a deep
result of differential topology asserts that two homeomorphic surfaces must
necessarily be diffeomorphic.
320 9. Global Geometry of Curves
kinds of curves, some of the results, suitably adapted, that we proved for
closed and compact surfaces of three-space remain true and, moreover, their
proofs are basically the same.
9.3.1. Jordan curve theorem. For example, the Jordan-Brouwer The-
orem 4.16, which reveals the separation property of surfaces closed in R3,
has as an analogue the so-called Jordan curve theorem in our case, the
Jordan differentiable curve theorem, to be precise. This result comes histor-
ically and perhaps logically earlier than the Jordan-Brouwer Theorem, as
we mentioned in the introduction to this chapter. There are various proofs
that are suitable for the level of our book, but, since our proof of the Jordan-
Brouwer Theorem 4.16 is valid only by replacing the surface closed in R3 by
the simple curve closed in R2, we will consider that the following theorems
have been proved.
Theorem 9.14 (Jordan curve theorem). Let C be a connected simple plane
curve that is closed as a subset of R2. Then R2 -C has exactly two connected
components with common boundary C.
Remark 9.15. The Euclidean space R2 is locally connected and R2 - C is
open. Hence, its connected components are open in R2. Furthermore, since
R2 is locally arc-wise connected, R2 - C is locally arc-wise connected and
its connected and arc-wise connected components coincide.
Remark 9.16 (The inner domain determined by a compact connected sim-
ple curve). If C is compact, there is a ball B centred at the origin such
that C C B. Thus, the connected set R2 - B is contained in R2 - C
and, consequently, either R2 - B C SZ+ or - B C SZ-, where SZ+, St-
are the two domains mentioned in Theorem 9.14. Suppose the latter is
true. Then, the connected component ft is not bounded and, furthermore,
SZ+ U C = - ft C B. Thus, SZ+ must be bounded. The connected com-
ponent that is bounded will be called the inner domain determined by C
and will usually be denoted by Q. The non-bounded connected component
will be called the outer domain. Note that SZ = SZ U C is also bounded and,
thus, it is compact.
Remark 9.17 (Orientation of a compact simple curve). Given a compact
connected simple curve C, let SZ be the inner domain determined by C.
We know, by Theorem 9.10, that there exists a periodic curve p.b.a.l. c
R -* R2, of period L> 0, such that ca(R) = C. One can show that there
are only two possibilities: either the normal vector N(s) of the curve c at
a(s), as defined in Section 1.4, points to the inner domain Q -in the sense
of Remark 4.22 for all s E R or N(s) points to the outer domain for all
s E R. If the first alternative occurs, we will say that c supplies a positive
9.3. Results already shown on surfaces 319
for the translation S= T-1 o R. That is, y: K-* U U V C C extends
uniquely to a periodic curve whose image must be C. In this case, therefore,
either directly or using Proposition 9.7, we have that C is homeomorphic
to a circle. Thus, we have shown that, if there are two parametrizations of
C in situation a, then C is a Jordan curve that is the trace of a periodic
parametrized curve. Similar reasoning for case b would only have given that
there is an interval K of R strictly containing I and a parametrization b.a.l.
y: K-* C whose image is U U V.
Suppose that C is not a Jordan curve and let a : I -* C be a parametriza-
tion b. a.l. of C that is maximal in the sense that it cannot be extended to
another parametrization defined on an interval properly containing I. Sup-
pose also that ca(I) is a proper open subset of C. Then, there exists at least
one point p in the boundary of C that is not in C. Since C is a simple curve,
there is a parametrization b. a.l. /3: J -* C covering p. Thus, /3(J) = V is
an open neighbourhood of p cutting U = ca(I), since p is in the closure of
U. Moreover, V is not contained in U since p U and U is not included in
V by maximality. Hence, the discussion above can be applied to these two
parametrizations. Since C is not a Jordan curve, they have to be in case b.
Thus, according to the discussion above, this parametrization would not be
maximal, which contradicts our supposition above. Consequently, we have
c (I) = C for such a maximal parametrization and C is topologically the
interval I or, equivalently, R. U Definition 2.47 (Tangent plane). Let S be a surface and let p E S be a
Exercise 9.11. T Prove that, if C is a connected simple curve, any two
curves a and /3 p.b.a.l., defined on two intervals I and J, respectively, which
are homeomorphisms onto the curve C, must satisfy either /3(s) = ca(s + c)
or /3(s) = a(-s + c) for some c E R. Show that we can talk about the
orientation of simple curves and that, if C is compact, it makes sense to talk
about its length.
Exercise 9.12. Define the concepts of the tangent line at a point, the nor-
mal line at a point, and the curvature of a simple curve, the latter depending
on the orientation.
Exercise 9.13. TLet C be anon-compact connected simple curve that is
closed as a subset of Euclidean space. Show that the domain of any curve
p.b.a.l. whose trace is C must be the whole of R.
9.3. Results already shown on surfaces
Once we have established the relation between the one-dimensional objects
analogous to surfaces, namely, simple curves in R2 and R3 and parametrized
curves studied in Chapter 1, we will now focus mainly on the geometrical
properties of simple curves that are closed as subsets of R2 or R3. For these
44 2. Surfaces in Euclidean Space
Exercise 2.43. If X : U -* S is a parametrization of a surface, show that
X : U -* X (U) is a diffeomorphism.
Exercise 2.44. Let f : Si -p 52 and g : 52 -p 5 3 be two diffeomorphisms
between surfaces. Prove that g o f Si -p S3 is another diffeomorphism.
In fact, the set of all the diffeomorphisms of a surface S is a group for the
composition.
Exercise 2.45. Let 01, 02 C 1ft3 be Euclidean open sets and let q5: 01 -p 02
be a diffeomorphism. If S C 01 is a surface, we know that q5(S) is another
surface; see Example 2.12. Show that b: S - (S) is a diffeomorphism.
Exercise 2.46. T Prove that the following pairs of surfaces are diffeomor-
phic: a plane and an elliptic paraboloid; a one-sheeted hyperboloid and an
elliptic cylinder.
2.5. The tangent plane
We want to define, at any point of a given surface, the best linear object
approaching the surface near that point. Remember that this technique
played a fundamental role in studying plane and space curves. We will
construct that object by using the notion of the tangent vector to a curve.
point. We will say that the vector v E I[83 is tangent to S at p if we can find
a curve a : (-E, s) -* S, s> 0, such that a(0) = p and a'(0) = v. The set
consisting of all the vectors tangent to S at the point p will be represented
by TPS.
Lemma 2.48. Let S be a surface, p E S, and X : U -* S a parametrization
of S with p E X(U). Then
TS = (dX)x-1()(R2).
Proof. Take w e ][82. Consider the line segment /3: (-s, s) -* U, for a small
s, given by ,6(t) = q + tw, where q = X (p). The curve ,6 satisfies Q(0) = q
and 3'(0) = w. Thus, if we define a (-s, s) -* S as the composition
a = X o Q, we have that a(0) = p and a'(0) _ (X o X3)'(0) _ (dX)q(w).
Hence, (dX)q(1ft2) C TPS, since a takes its values in S.
Conversely, take a vector v tangent to S at p. By definition, there is a
curve a : (-E, s) -* S such that a(0) = p and a'(0) = v. By taking s small
enough, we can assume, by the continuity of a, that its trace is contained
in X(U). We define a curve on U by Q = X-1 o a. We have that Q(0) = q
and a = X o/3. Thus, v = a'(0) _ (X o X3)'(0) _ (dX)q(/3'(O)); that is, any
vector tangent to S at p lies in the image of (dX)q.
2.5. The tangent plane 45
Figure 2.7. Tangent plane
Remark 2.49. As a first consequence of Lemma 2.48, the set TpS formed
by all the vectors tangent to a surface at a given point is a linear plane of
1R3 which will be called the tangent plane of the surface S at the point p.
Remark 2.50. If X : U -r S is a parametrization of S that covers p and
q = X-1(p), then, from Lemma 2.48, we see that (dX)q(R2) does not depend
on X and that {XZ6(q), X41(q)} is a basis of TpS.
Example 2.51 (Inverse images). If 0 C IR is open, f : 0 --+ R is a
differentiable function, and a is a regular value of f belonging to its image,
we know-see Example 2.13-that S = f ({a}) is a surface. For all p E S,
we see that
TpS = ker((df )p : 1R3 -} IIP).
In fact, if v E TpS, then there is a curve a : (-e, e) -+ S such that a(0) = p
and a'(0) = v. Hence, (f o a) (t) = a for all t and, by differentiating at t = 0,
(df)p(v) _ (f o a)'(0) = 0. Thus, v is in the kernel of (df )p. Since TpS and
ker(df )p are linear subspaces of dimension two and one of them includes the
other one, they have to coincide, as we wanted.
Example 2.52 (Planes). Let P = {p E 1R3 (p - po, a) = 0}, with po, a E R3
and jal = 1, i.e., the plane that passes through po and is normal to a. If we
define f : R3 -r R by f (p) = (p - po, a), then we have that, for each p E P,
TpP = ker(df)p = {v E R31 (v, a) = 0}; that is, the tangent plane of Pat
any point coincides with its direction plane.
Example 2.53 (Spheres). Let S2(r) _ {p E R31 Ip-po 12 = r2 } be the sphere
with centre po and radius r > 0. Then S2 (r) = f -1({r2}) for the function
f : 1R3 -. 1R given by f (p) = Ip - poI2 and, thus, TpS2(r) = ker(df )p = {v E
R3 (p - po, v) = 0}. Thus, TS2(r) is the linear plane orthogonal to the
position vector p - po relative to po.
318 9. Global Geometry of Curves
of the segment o were on parallel sides of I x J, the projection of v onto one
of the two intervals I or J would be surjective. Therefore U = a(I) would
be contained in V = 16(J) or vice-versa, which contradicts our hypotheses.
This proves B.
These two properties imply that X consists of at most two segments
and, in this case, the two segments have the same slope. That is, the only
two possible cases, up to a sign for the slope of the segments and a possible
interchange of I and J, can be seen in Figure 9.5. The fact that there
are only two alternatives for two parametrizations p.b.a.l. of a simple curve
whose images intersect without containing one another leads to the following
classification result that we have already used in Chapter 4 to prove the
Jordan-Brouwer Theorem 4.16; see the proof of Lemma 4.17.
Theorem 9.10 (Topological classification of simple curves). If a simple
curve of R2 or of R3 is connected, then either it is homeomorphic to III
and, in this case, it can be entirely covered by a parametrization, or it is
homeomorphic to a circle and, in this case, it is the trace of a periodic
regular parametrized curve.
Proof. Let a I -> C and ,6 J -* C be two parametrizations b.a.l.
of the connected simple curve C, defined on two open intervals of I[8 and
let U and V be their images, which are open in C. Suppose that U and V
intersect but that neither of them is included in the other. Then, we are in a
situation geometrically equivalent to one of the two cases a or b represented
in Figure 9.5. In case a, the set a-1(U fl V) has two connected components
Il and I2 and /3_i (U fl V) also has two connected components Jl and J2
such that f : Il U I2 -> Jl U J2 coincides when restricted, for example, to I2
with a translation T : ][8 -* ][8 in such a way that R(I2) = Jl. Then
K = I U T-1(J)
is an open interval of ][8 that strictly contains the interval I and we define a
surjective curve
on it by the expression
a(s) if s E I,
(,(3 oT)(s) ifs E T-1(J),
which is well defined since a = ,Q o f - ,Q o T on I fl T1 (J) = 12 and, thus,
is differentiable and p.b.a.l. Moreover, the interval K includes two disjoint
open subintervals Il C I C K and T-1(J2) C T-1(J) C K such that
ryIrl = ('y o and S(Il) = T-1(J2)
9.2. Parametrized curves and simple curves 317
7 J
J
I I
Figure 9.5. The set X, cases a and b
subsets of ][8 has to be an open interval contained in I or in J. Furthermore,
the change of parameters
f =13-1 o a: a-1(U n V) C I - /3-1(U n V) C J
is a diffeomorphism-recall Theorem 2.26-that satisfies a = ,6 o f on
a-1(U fl V) C I. Therefore, by the chain rule,
a(s) = f'(s),(3'(f (s)), Vs E a-1(U fl V).
Since a and ,6 are curves p.b.a.l., we have that If'(s)I = 1; that is, either
f' - 1 or f' - -1 on each connected component of a-1(U fl V). In other
words, f is, up to a constant, either the identity map or its opposite on each
of these connected components.
We consider the subset X of the rectangle I x J of ][82 defined by
X= {(si, s2) E I x J I a(sl) =/3(52)}.
This set X is clearly closed in I x J since a and ,Q are continuous. We now
state some other properties of this set.
Lemma 9.9. The following assertions concerning the set X defined above
are true.
A: X is a graph over a union of open subintervals of I and is also a
graph over a union of subintervals of J which consists of segments
with slope +1 or slope -1.
B: Each segment of X has its endpoints on non-parallel sides of Ix J.
Proof. In fact, assertion A follows if we observe that (Si, s2) E X C I x J
is equivalent to s2 = (,6-1 o a)(sl) = f(si), that is, X is the graph of the
bijective map f. Let v be one of the segments with slope f1 of X. Since
X is a closed subset of I x J, if at least one of the endpoints of o were not
on a side of the rectangle I x J, this endpoint would be in v and, thus, its
projection on I would not be an open interval. This contradiction implies
that the endpoints of v lie on the sides of I x J. Finally, if the two endpoints
46 2. Surfaces in Euclidean Space
Example 2.54 (Open subsets). If Si is an open subset of a surface S and
p E Si, then TpSI = TPS.
Given a point p of a surface S, we will call the straight line passing
through p that is perpendicular to TS the normal line of S at p. By this
definition, all the normal lines of a plane are parallel, and all the normal
lines of a sphere meet at its centre.
Exercise 2.55. Let S = {p E ][83I Ia)2 = r2}, with al = 1 and r > 0,
be a right cylinder of radius r whose axis is the line passing through the
origin with direction a. Prove that TS = {v e I[831 (p, v) - (p, a) (a, v) = 0}.
Conclude that all the normal lines of S cut the axis perpendicularly.
2.6. Differential of a differentiable map
The notion of differential of a differentiable map defined on an open subset
of Rn extends, in a natural way, to differentiable maps defined on surfaces.
Furthermore, the classical properties of the differential of calculus will also
be true in this new situation.
Definition 2.56. Let S be a surface and let f : S - II8tm be a differentiable
vector-valued function. For a point p e S, we define the differential of f at
p, which we will denote by (df), as follows: Given v e TS, choose a curve
a : (-s, s) -* S such that a(0) = p and a'(0) = v, and then we put
d
(df)(v) = (foa)(t) = (foa)'(O).
dt ' t=o
Lemma 2.57. (df)p : TS -f ][8tm is swell-defined map, that is, (df)(v)
does not depend on the chosen curve and, moreover, it is a linear map.
Proof. Let X : U -* S be a parametrization with p E X (U). We know that
TS = (dX)q(R2), where q = X-1(p), and so (dX)q : II82 -* TS is a linear
isomorphism. Taking s small enough, we may assume that a(-s, e) C X (U).
The curve X-1 o a satisfies (X-1 o a)(0) = q. Since X o (X-1 o a) -
a, differentiating at t = 0, we have (dX)q [(X-1 o a)'(0)] = v, that is,
(X-1 o a)'(0) _ (dX)1(v). Then, using the chain rule, we get
(foa)'(O)= - (fox)o(X1 o a)
t=o
= d(,f o X)9\(X-1 o a)'(O)) = d(,f o X)q o (dX)1(v),
which proves the equality
(df)p = d(f o X)q o (dX)',
from which our assertion follows. 0
2.6. Differential of a differentiable map 47 316 9. Global Geometry of Curves

Example 2.58. Let S be a surface contained in an open set O C ][83 and let
F : O -* ][8"'' be a differentiable map. Denote by f : S -* II8tm the restriction
f =Fps. Given p E S and v E TPS, for each curve a : (-E, s) -* S such that
a(0) = p and a'(0) = v, we have
(df)(v) _ (f o a)'(0) _ (F o a)'(0) _ (dF)(v).
Hence, (df)p is the restriction to TS of (dF)P ; I[g3 -+ Tftm
Example 2.59. Let S be a surface, p E S, v E TPS, and let a : (-E, s) -* S
be a curve such that a(0) = p and a'(0) = v.
A: (Height function) If po E I[83 and a E I[83 - {0}, then the differential
Figure 9.4. Parametrizations of a simple curve
of the height function h : S - Il8 given by h(p) _ (p - po, a) is

(dh)(v) - dt It_o (a(t) - poi a) _ (a'(O),a) _ (v,a).
B: (Square of the distance function) If po E ][83, then the function
f : S -* I[8, which is the square of the distance to the point po, has
differential
(df)(v) - dt t=o a(t) -pole = 2 (a'(0), a(0) - po) = 2 (v,p - po)
I
C: (Distance function) If po E ][83 - S, then the differential of the
distance function f : S -* II8 to the point po is given by
a homeomorphism onto its image since it is an injective identification map.
Therefore, all the maps
a: (t,t+L) -*C, `dtER,
form a family of homeomorphisms whose images cover C endowing it with
a structure of a simple curve as defined in Example 2.18. 0
Remark 9.8. In general, we call any subset of R3 which is homeomorphic
to a circle a Jordan curve. Then, the previous result shows that the trace
of a periodic curve without self-intersections is a Jordan curve.

(v,p-po) Exercise 9.4 and Proposition 9.7 above assert that, in at least two sit-
(df)(v) - dt
la(t) -poi _ uations, the trace of a regular parametrized curve is a simple curve when
t=o I p - po
the curve is a homeomorphism onto its image and when it is L-periodic and
D: (Square of the distance to a straight line) For each po, a E ][83
injective on the subintervals with length less than L. In the first case, the
with al = 1, the differential of the function f : S -* II8 defined by
resulting simple curve is topologically equal to R and, in the second case, it
1(P) = Ip -pole - (P - poi a)2 is given by is homeomorphic to a circle, that is, it is a Jordan curve. We will now see
-
(df)p(v) = 2 [(v,p -po) (p - poi a) (v, a)] that these two alternatives are the only possibilities for a connected simple
curve in R3.
Example 2.60. We compute here the differential for inclusions, constant
maps, and restrictions. In fact, let C C R3 be a connected simple curve, and let

A: (Inclusions) If S is a surface and i : S -* 1ft3 is the inclusion map, and f3:J-*VCC

then
(di)(v) = v Vv E TS,
that is, (di)P is the inclusion of TS into I[83.
B: (Constants) The differential of a constant function is, at any point,
the identically zero linear map.
C: (Restrictions) If Si C S is an open set in a surface S and f : S
][8tm is a differentiable map, then d(f181 )p = (df)P for any p e S.
be two parametrizations of C. That is, U and V are open subsets of C,
I and J are open intervals of R, and a and /3 are two regular curves
we can assume them to be p.b.a.l. without loss of generality which are
also homeomorphisms. Assume also that U f1 V 0 and that neither U is
contained in V nor V is contained in U; see Figure 9.4. Then U f1 V is a
non-empty open subset of U and of V. Thus, c 1(U f1 V) is a nonempty
open subset of the interval I and /3-1(U f1 V) is a non-empty open subset
of the interval J. Hence, each connected component of any of these open
9.2. Parametrized curves and simple curves 315 48 2. Surfaces in Euclidean Space

Then:

S(t + A) lI A Io'(u)Idu+JAIcV(u)Idu,
+A
Until now, we have extended differentials of differentiable maps between
Euclidean open sets to differentiable vector-valued functions defined on sur-
faces. We continue by considering the case of differentiable maps f : O -* S,
where O C II87 is an open set and S is a surface. If p E O and v E ][8", then
and, since a is A-periodic, the curve a : (-s, s) -* O given by a(t) = p+tv with s small enough satisfies
a(0) = p and a'(0) = v. Hence, f o a is a curve on S with (f o a)(0) = f(p)
S(t + A) =
J
A Idu + J t Idu = L + S(t), and (f o a)'(0) _ (df)(v). Otherwise, the image of (df)p : I[8n -* ][83 lies
in the linear plane Tf()S. From now on, we will deal with the restriction
where we have set (df)P : T f(p)S. Analogously, if f Si - 82 is a differentiable map
A
between surfaces, then the image of the differential of f at p, viewed as a
map from Si to ][83, is contained in the plane Tf(p)S2. In this situation, when
This number will be called the length of the curve a. Thus, if b : ][8 -> ][8 is we talk about the differential of f at p E Si, we will mean the restriction
the inverse diffeomorphism of S and /3 = a o b is the reparametrization of a (df)n : TPSl -* Tf(p)S2
by arc length, we obtain that By Theorem 2.38, we know that the composition of differentiable maps,
in the sense of Definition 2.27, is also differentiable. The following result
/3(s + L) = a(s + L)) = (q5(s) + A) = ,(3(s),
generalizes a classical property of calculus for the differential of a composite
and thus ,6 is L-periodic. Moreover, since A is the least period of a, then L map.
is the least period of ,6, provided that aI[o,A) is injective. As a consequence,
Theorem 2.61 (Chain rule). Let f Si - 82 and g : S2 -* S3 be two
from now on, when we say that a : ][8 -> ][83 is an L-periodic parametrized
differentiable maps, where Si, 82i S3 stand for either surfaces or Euclidean
curve, it will be understood that it is parametrized by arc length and that
open sets. Given p E Si, we have
L> 0 is its length, which coincides with its period.
Proposition 9.7. Let a : I[8 -* I[83 be an L-periodic parametrized curve, d(9 o f), - (dg)f() o (df).
L > 0, such that its trace has no self-intersections, that is, such that a is Proof. First, let us point out that the maps (df)P TpSI -3 T f()82,
injective on [0, L). Then the trace of a is a simple carve homeomorphic to (dg)f() ' Tf(p)82 * T(sof)(n)'Ss, and d(9 o f), : Tpsl -* T(9of)(P)S3, where
a circle.
T9Si = when Si is an open subset of ][8n, have domains and ranges which
allow us to make all the required compositions.
Proof. We have C = a(][8) = a([0, L]) since a is L-periodic. On the other
hand, the map a : [0, L] -* C is surjective, continuous, and closed-any If v E TPSI, pick a curve a (-s, s) -* Si satisfying a(0) = p and
continuous map defined on a compact space taking values in a Hausdorff a'(0) = v. Then foa is a curve on S2 with (foa)(O) = f(p) and (foa)'(O) _
space is closed. Hence, it is an identification map. Let R« be the equivalence (df)(v). Thus,
relation induced by a on [0, L] and given by
d(9 ° f)(v) _ [(9 ° f) ° a]' (0) _ [9 ° (f o a)]' (0) _ (dg)f() ((df)(v)).
siRs2 == c(s) = c(s2).
Then, the map a : [O, L]/R« -* a(][8) given by ([s}) = a(s) is a homeo-
morphism. We now observe, since a is injective on [0, L) and L-periodic,
Exercise 2.62. Let q5: Si -
S2 be a diffeomorphism between surfaces and
p E Si. Prove that (dq5)p : TpSi -* Tf(p)S2 is a linear isomorphism and that
that
sl = 82,
(dq5)1 - (dq5_().
s1Ras2 == s1 = 0, s2 = L, Given a differentiable function f S - ][8 defined on a surface and a
s1 =L,s2 =0 point p E S, we will say that p is a critical point of f when (df)P = 0.
and, thus, C is homeomorphic to the quotient of [0, L] where we identify the Proposition 2.63 below gathers together some of the properties of differen-
endpoints 0 and L, that is, to a circle. But, in addition, a : (0, L) -* C is tiable functions relative to their critical points.
2.6. Differential of a differentiable snap 49 314 9. Global Geometry of Curves

Proposition 2.63. Let S be a surface. The following two assertions are

true:
A: If f : S ][8tm is differentiable, S is connected, and (df)P = 0 for
all p E S, then f is constant.
B: If f : S -* ][8 is differentiable and at p E S there is a local extreme
of f, then p is a critical point of f.

Proof. Part A. Take a E f(s). Then the set A = {p E S f(p) = a} is

a non-empty closed subset of S. We check that it is also open. Choose
p E A and let X: U-* S be a parametrization such that p e X (U) for a
connected U. For each q E U we have, from Theorem 2.38, d(f o X)9 =
(df)x(q) o (dX)q = 0. Hence, from elementary calculus, f o X is constant on
U and, thus, f = (f o X) o X-1 is constant on X (U). Since f(p) = a and
p E X (U), we have X (U) C A; that is, A is open. Since S is connected we
deduce that A = S, i.e., f is constant.
Part B. Let p be a point of S where f attains a local extreme. If v E TS
and a : (-e, e) -* S is a curve such that a(0) = p and a'(0) = v, then the
differentiable map t H (f o a) (t) has a local extreme at t = 0. Consequently,
from elementary calculus, (df)(v) _ (f o a)'(0) = 0. Thus, p is a critical
point off. O for all t E R. Thus, if a is periodic, then there exists A E ][8+ such that
Example 2.64 (Implicit functions, transversality, and intersection of sur-
faces). The notion of the differential of a differentiable function defined on a
surface and the chain rule that we have proved allow us to obtain an implicit
function theorem in the context of the theory of surfaces.
Theorem 2.65. Let S be a surface, f S -* II8 a differentiable function,
p E S, and a E ][8. Suppose that f(p) = a and (df)P 0, that is, p is not a
critical point of f. Then there exist an open neighbourhood V of p in S, a
real number s> 0, and an injective regular curve a : (-e, s) -* ][83 that is a
homeomorphism onto its image, such that
a(0) = p and f({a}) f1 V = a(-s, s).
Thus, if a belongs to the image of f, then the set f -1({a}) is a simple curve,
in the sense of Example 2.18.
Proof. Let U be an open subset of II82 containing the origin and let X
U - S be a parametrization of S with X(0, 0) = p. We define g : U -* II8 by
g = foX; thus, g is differentiable with g(0, 0) = f(X(0, 0)) = f(p) = a and,
by the chain rule, (dg)(o,o) _ (df)o(dX)(o,o). Since (dX)(o,o) is injective and
(df)P 0, we have that (dg)(o,o) 0, that is, (ga,
that gv(0, 0)
0) (0, 0). Assume
0. Applying the implicit function theorem of calculus (see
Figure 9.3. Folium of Descartes
is regular and injective but its trace (see Figure 9.3), which is called the
folium of Descartes, is not a simple curve.
Exercise 9.4. If a : I -+ ][83 is a regular parametrized curve that is also a
homeomorphism onto its image a(I), prove that a(I) is a simple curve.
Now, let a : ][8 -* I[83 be a regular parametrized curve that is periodic as a
map, that is, such that there exists a real number a E ][8+ with a(t+a) = a(t)
a(t -{- A) = a(t) for all t E ][8, where A is the smallest positive number
satisfying this condition. We will say that A is the period of a and that a is
an A-periodic curve. To see that such an A exists, we argue by contradiction
as follows. If there is a sequence {a}EN of positive numbers such that
a(t -I- a) = a(t) and lim,0 ate, = 0 for some t E ][8, then
a(t + an) - a(t)
ap(t) = lim
n.-'°° n
=0,
which is not true since a is regular.
Example 9.5. The circle a(s) = c+r (cos(s/r), sins/r)) with centre c E ][82
and radius r > 0 is a periodic curve p.b.a.l. with period 2irr.
Example 9.6. Other examples of periodic curves with period 2ir having self-
intersections are, for instance, the curves a(t) =sin 2t(cos t, sin t), which we
previously called the four-petal rose, and ,3(t) _ (cos t, sin 2t); see Figure 9.7.
Let a : R -* R3 be an A-periodic parametrized curve, with A> 0, and
let S : R -* III be its arc length function, given by
ft
S(t) =Lo(a) =
J0
Idu.
9.2. Parametrized curves and simple curves 313
Figure 9.2. Four-petal rose
and we will prove it in its complete generality, following an idea due to R.
OSSERMAN in [14].
9.2. Parametrized curves and simple curves
We began Chapter 1 with the local theory of curves. There we defined the
notion of a curve-see Definition 1.1 as a differentiable map of an interval
of IIS either to IIS2 or to IIS3, depending on whether the curve is plane or not.
In spite of this definition, the properties that we studied and the concepts,
such as curvature or torsion, that we introduced were of a geometrical nature;
that is, they depended only upon the image or trace of the curve. We will
refer to them now as parametrized curves. Later, in Chapter 2, after the
definition of a surface in IIS3, we defined a simple curve in IIS2 or in IIS3 as a
one-dimensional analogue of a surface; see Example 2.18.
We should point out that the traces of parametrized curves, even when
regular, and simple curves are closely related geometrical objects, but are
not identical. This is more or less clear if we consider, for example, that the
first ones can have self-intersection points and the second ones cannot. The
following exercises also shed light upon this point.
Exercise 9.1. T Prove that lines and circles in IIS3 are simple curves.
Exercise 9.2. T We call a four-petal rose the parametrized curve a : IIS -k
given by
a(t) = (sin 2t cos t, sin 2t sin t)
for each t e I. Prove that its trace (see Figure 9.2) is not a simple curve.
Exercise 9.3. j Show that the parametrized curve a : (-1, +oo) -> 1[82,
given by
_ 3t 3t2 1
a(t)
1t3' 1+t3)' bt E I[8,
50 2. Surfaces in Euclidean Space
Theorem 2.14), we have the existence of two real numbers s, d > 0 and a
differentiable function h : (-s, s) -* (-S, S) such that
(-s,s) x (-6,6) C U, h(0) = 0,
and, moreover,
(2.1)
{(u, v) E (-e, s) x (-6,6)Jg(u,v) =a} _ {E ][82 Iu E (-s, E)}.
Keeping this in mind, consider the neighbourhood V = X((-s, e) x (-8, S))
of (0,0) and the curve a : (-E, e) -* ][83 given by a(t) = X (t, h(t)), which
is a homeomorphism onto its image since it is the composition of a graph
and X. Then a(0) = X(0, h(0)) = p, ap(t) _ (dX)(t,h(t))(l, h'(t)), which is
non-null since (dX)(t,h(t)) is injective. Furthermore, if we apply the map X
to equality (2.1), we obtain
{_ {a(t) E ][83 It E(-e, s)},
that is, f -1({a}) fl V is the trace a. D
An immediate geometrical application of this implicit function theorem
applies to the intersection of surfaces near a common point. If Si and 82
are two surfaces and p E Si n 82 is an intersection point, we will say that Si
and S2 are tangent at p when the corresponding tangent planes TpS1 and
TpS2 coincide. Otherwise, that is, when TpS1 TpS2, we will say that the
two surfaces intersect transversely or that they are transverse at p. In this
last situation, we may assert that, locally, the intersection is the trace of a
curve.
Proposition 2.66. If the surfaces Si and 82 intersect transversely at a
point p, then there are an open neighbourhood V of p in 1f3, an open interval
I C It, and a regular curve a : I -* 1f3, which is a homeomorphism onto its
image, such that a(I) = V n Si n 82.
Proof. From Exercise 2.25, we know that there exist an open neighbour-
hood O of p in ][83 and a differentiable function g : O -* II8 such that 0 is a
regular value of g and such that S2 f10 = g-1({0}). We define f : SiflO -* ][8
by f = Which is differentiable on the surface Si fl O containing the
point p. Moreover, f(p) = g(p) = 0 and (df)p = (dg)181. If p were a
critical point of f, then we would have TPSi C ker(dg)p = TPS2i see Exam-
ple 2.51. But this is impossible since Si and 82 meet transversely. It is now
enough to apply Theorem 2.65 above. O
In order to arrive at a global understanding of the intersection of sur-
faces, we will need a global version of the definition of transversality. We
will say that two surfaces Si and 82 intersect transversely when they are
2.6. Differential of a differentiable map 51
transverse at each common point. As a consequence of this definition and
Proposition 2.66, we have the following result.
Theorem 2.67. The intersection of two surfaces meeting transversely is
either empty or a simple curve-with possibly more than one connected com-
ponent.
Example 2.68 (Diagonalization of a symmetric matrix). Consider a sym-
metric matrix A of order three. For each v, w E ][83, it is clear that (Av, w) _
(v, Aw). We are going to show that the matrix A admits an orthonormal
basis consisting of eigenvectors. For this, we define a differentiable function
f : S2 -* II8 by f(p) _ (Ap, p), for all p E S2. We first will show that the
following assertion is true.
A point p E S2 is a critical point of f if and only if Ap =
f(p)p.
In fact, the differential of f at the point p e S2 is given by
(df)p(v) = 2 (Ap, v) by E {v E ][831 (p, v) = 0}.
Hence, (df)p = 0 if and only if Ap = Ap with A E ][8. Taking the scalar
product with p, we have A = f(p).
From this, if f is constant, then A is a scalar multiple of the identity
matrix and the result that we seek is trivial. If f is not constant, then
there exist pl, p2 E 2 at which f attains its maximum and its minimum,
respectively. Then, Api = f(pi)pi, Ape = f(p)p, and f(i)
f(p2).
Hence,
(f(pi) - f(P2)) (P1,p2) _ (f(P1)P1,P2) - (pi, f(p2)p2)
_ (Ap1,p2) - (pi,Ap2) = 0.
Thus, pl and p2 are orthogonal. Let p3 E 2 be such that {pi,p2, p3} is an
orthonormal basis of I[83. It suffices to show that p3 is an eigenvector of A.
In fact, for i = 1, 2,
(Ap3, pi) _ (p3, Apz) = f(p) (p3, pi) = 0.
Thus, Ap3 is proportional to p3.
Example 2.69 (Normal lines). Let S be a surface, po E ][83, and let f
S - I[8 denote the function square of the distance to the point po. Taking
Example 2.59 into account, we can deduce the following.
A point p E S is a critical point off if and only if the
normal line of S at p passes through the point po.
If S is compact, then the function f has at least a critical point. Hence, the
following is true.
312 9. Global Geometry of Curves
things go in a way similar to the Gauss-Bonnet Theorem, topological infor-
mation about the curve. But the-intrinsic--topology of the curve is known
in advance by the classification that we made at the beginning of this chap-
ter: the curve is homeomorphic to the circle. We will present a theorem by
WHITNEY and GRAUENSTEIN which states that this integer refers to the ex-
trinsic topology of the curve. In fact, it characterizes the regular homotopy
classes of periodic curves. A curve of this type can be deformed into an-
other one, avoiding the formation of cusps, only when their rotation indices
coincide. We will also show that, when the curve has no self-intersections,
its rotation index is ±1, a result which, in some sense, was already cited by
EULER, although it was correctly announced and understood for the first
time by the first researcher in degree theory, H. HOPF.
When we studied the local theory of curves in Chapter 1, an important
difference between the curvature functions of plane and space curves was that
for the first ones we could define an oriented curvature and for the second
ones we could define only a non-negative curvature. For this reason, the
total curvature of curves in ii looks more like the total absolute curvature
of plane curves. We knew that 2ir was a lower bound for this integral. We
will prove FENCHEL's Theorem (1929), which says that this bound remains
valid for periodic curves in R3 and that the equality characterizes the plane
convex curves. We will give a proof due to B. SEGRE (1934), also discovered
independently by H. RUSTIHAUSER and H. SAMELSON. Later, in 1948,
K. BORSUK generalized this proof to periodic curves in Euclidean space of
arbitrary dimension and conjectured that the lower bound 2ir should become
47r if the curve is a non-trivial knot, that is, if the curve is not the boundary
of a disc embedded in R3. The proof of this assertion was first obtained by I.
FARM and J.W. MILNOR in 1949 and 1950, respectively. The proof that we
will present is based on an integral formula due to M. W. CROFTON, which
will be a more or less immediate consequence of the area formula given in
Chapter 5.
We will finish this chapter by proving a global theorem about the simple
plane curves that are compact and connected: the so-called four vertices
theorem, which establishes the existence of at least four local extremes for the
curvature function of a curve of this type. It was first shown for ovals by the
Indian geometer H. MUKHOPADHYAYA in 1909 and rediscovered in 1912 by
A. KNESER. Since then, it has been one of the most popular theorems of the
global theory of curves. There is quite a number of proofs for it, for example,
by G. HERGLOTZ in 1930 and by S. B. JACKSON in 1944. In addition,
H. CLUCK (1966) showed that the theorem provides, in fact, a sufficient
condition for a positive periodic function to be the curvature function of
an oval. The theorem is indeed true without the convexity assumption
9.1. Introduction and historical notes 311
1 to integrals along the curve C. In the introduction to Chapter 5, we
have already talked about the statement and the first proofs of this theorem
which are due to STOKES and MAXWELL. In order for the proof that we
gave there to remain valid, it will be enough to have at our disposal the
existence of a one-dimensional version of the SARD'S theorem. Verifying
this point will allow us to establish some results about the total absolute
value of the curvature for periodic plane curves.
We continue to discuss the results that we proved for surfaces and that,
in some way, also occur in the global theory of curves and which were first
solved in this context. For instance, we have the isoperimetric question, that
is: Among all compact simple curves with a given length, which one encloses
the greatest area? This problem was raised as early as the writings of the
Greek mathematician PAPPUS (4th century B.C.), who asserted that the
solution was the circle. PAPPUS wrote a complete treatise on isoperimetric
problems, starting with the following motivation: why have bees chosen the
hexagonal form for the cells of their honeycomb? It was only after twenty-
three centuries that the first rigorous proof of the isoperimetric inequality
in the plane was published in the book on the calculus of variations by
WEIERSTRASS in 1870. Other later proofs are due to HURWITZ (1902),
using Fourier expansions, and to SCHMIDT (1939), which appears in almost
all the texts on the differential geometry of curves. Recently GROMOV [1]
gave a proof showing the isoperimetric inequality in any dimension as a
consequence of the divergence theorem. For us, the proof given in Chapter 6
for the three-dimensional case, which was based on the BRUNN-MINKOWSKI
inequality, remains valid in the two-dimensional case. We will sketch in some
of the exercises a version of the proof by HURWITZ and another proof using
only the divergence theorem, which is different from the GROMOV proof,
and which has been outlined independently by F. HELEIN and the authors.
The last global results, which have an analogue in the theory of surfaces,
correspond to the theorems of HADAMARD and STOKER given in Chapter 6.
There, we dealt with ovaloids. Here, we will consider ovals, that is, compact
connected simple curves in the plane with positive curvature everywhere.
In our study of surfaces in IIS3, it was clear that different integrals of
elementary functions of the Gauss and mean curvatures, such as
K dA, IKI dA, and H2 dA
is is s
have important geometrical properties of a global nature. In the case of plane
curves, all the local information is contained in the-oriented--curvature
function. We will prove that the integral of this curvature along a periodic
curve is an integer multiple of 2ir the length of the unit circle. The integer
appearing here, called the rotation index of the curve, should contain, if
52 2. Surfaces in Euclidean Space
From and point of ][83 we may draw a normal line to a
given compact surface.
If S is connected and all its points are critical for f, then f must be constant
and, thus, S is included in a sphere with centre po. Accordingly, we have
the following.
If all the normal lines of a connected surface meet at a
given point, then the surface is contained in a sphere cen-
tred at this point.
By arguing in a similar way, that is, by looking carefully for a convenient
function defined on the given surface and applying Proposition 2.63, we may
solve the following exercises.
Exercise 2.70. 1 Let S be a surface and a E ][83, aJ = 1. Show the following.
If S is compact, there exists a point of S whose normal line is
parallel to the vector a.
If S is connected and all its normal lines are parallel to the vector
a, then S is contained in some plane orthogonal to a.
Exercise 2.71. T Let S be a surface and let R be a straight line of I[83.
Prove the following.
If S is compact, then there is a point of S whose normal line inter-
sects the line R perpendicularly.
If S is connected and all its normal lines intersect R orthogonally,
then S is a subset of a right circular cylinder with axis R.
Exercise 2.72. T Let S be a compact surface. Show that there exists a
straight line cutting S perpendicularly at (at least) two points.
Exercise 2.73. T Let S be a surface that is closed as a subset of ][83. If
a e 183, prove that the distance function to the point a reaches its infimum
at some point of S, that is, f has a minimum.
Exercise 2.74. T Assume that S is entirely on one side of some plane P.
Show that S and P are tangent at each point of S f1 P.
One of the more useful results of calculus is the inverse function theorem
relative to differentiable maps between Euclidean open sets. We now show
that it is also true for differentiable maps defined on surfaces.
Theorem 2.75 (Inverse function theorem). Let f Si -f S2 be a differ-
entiable map between surfaces and let p E S. If (df)p : TPSi -* T f(p)S2
is a linear isomorphism, then there exist an open neighbourhood Vi of p in
Si and an open neighbourhood V2 of f(p) in S2 such that f(Vi) = V2 and
fv1 : Vi - V2 is a diffeomorphism.
2.6. Differential of a differentiable map 53
Proof. Let Xi : Ui -* Si, with i = 1, 2, be two parametrizations such that
p e Xi(Ul), f(p) E X2(U2), and f(X1(U1)) C X2(U2). Let qi E Ui, i = 1, 2,
be points such that Xi(gl) = p and X2(g2) = f(p). The map X2 1 010 Xi
Ul - U2 is differentiable and d(X2 1 o f o Xl)gl = (dX2)1 0 (df)P o (dXl)gl
is a linear isomorphism since it is a composition of isomorphisms. We now
are able to apply the inverse function theorem of calculus. From this it
follows that there are open neighbourhoods Wi C Ui with i = 1,2 of q2 such
that (X1 o f o Xi)(Wl) = W2 and such that X2 1 o f o Xi : Wl - W2 is
a diffeomorphism. Consider the open subsets V = X(W) of Si. Vi is an
open neighbourhood of p in Si and V2 is an open neighbourhood of f(p) in
82. Furthermore, f(V1) = VZ and fv1 = X2 0 (X1 of o Xl) o Xi 1 : Vi - VZ
is a diffeomorphism since it is a composition of diffeomorphisms. D
As an immediate illustration of the usefulness of this theorem, we are
going to obtain an improvement of Proposition 2.23. We proved there that
any surface is locally a graph over at least one of the coordinate planes. Our
goal now is to show that, near any of its points, the surface can be viewed
as a graph over the tangent plane at that point. This property will prove to
be very fruitful in what follows.
Corollary 2.76. Let po be any point of a surface S and let Pp be the (affine)
tangent plane at this point, that is, the plane passing through pp with direc-
tion plane TPoS. Then there are open neighbourhoods U and V of the point
po in Po and S, respectively, and a differentiable function h : U -* ][8 such
that V is the graph of h.
Proof. Take a unit vector a E 1ft3 perpendicular to the tangent plane TpoS.
Denote by f : S -* 1ft3 the orthogonal projection over Po, given by
f(p)=p-(p-po,a)a, b'p E S.
Then f(o) = po and f(p) - po L a for p E S; that is, f(S) C Po and
thus f : S -* Po is a differentiable map between the surfaces S and Po. A
straightforward computation from Definition 2.56 gives us
(df)(v) = v - (v,a)a, Vv E TPS.
Hence (df)0 : TS -* TpoPo = TpoS is the identity map. By the previous
inverse function theorem, we get an open neighbourhood V of po in S and
another one, U, of f(o) = po in Po so that f(V) = U and fv : V -* U is a
diffeomorphism. Defining h : U -* II8 by
d9 E U,
h(9) _ (f-1(9') -po,a),
completes the proof. 0 divergence theorem will relate the integrals of some functions defined on
310 9. Global Geometry of Curves
Figure 9.1. Which are the inner points relative to this carve?
Furthermore, their proofs can be translated from one case to the other, but
it is always easier to transfer results about surfaces to results about curves.
This is why, once the problem of the connection between the two possible
definitions of a curve is settled, we can proceed to show which results of
the theory of surfaces can be transferred to the theory of curves, along with
their proofs, except for minor changes.
We will first recover the JORDAN curve theorem, which asserts the fol-
lowing: a compact connected simple curve C of the plane, that is, home-
omorphic to the circle, divides the plane into exactly two domains, one of
them bounded. This means that, in the presence of the curve C, the remain-
ing points of the plane separate into two classes: inner and outer points, in
such a way that any two points in the same class can be joined by a curve
that does not intersect C and any curve joining two points in two different
classes must meet C. This result is obviously true for curves such as circles
or ellipses, but is less obvious depending upon how complicated the curve
is; see Figure 9.1.
This theorem was first announced by CAMILLE JORDAN (1838-1922) in
his Cours d'Analyse, and the complexity of the proof of such an intuitively
obvious assertion surprised quite a number of mathematicians at that time.
The surprise grew when some of them observed that the proof by JORDAN
suffered from some gaps that demanded considerable effort to avoid them.
The difficulty of the proof lies in the generality of the notion of the compact
simple curve as a subset of the plane that is homeomorphic to the circle.
The proof for the case of surfaces of the JORDAN-BROUWER theorem that
we gave in Chapter 4 is also valid for differentiable curves.
Once we have the JORDAN curve theorem, we are able to talk about
the inner domain f determined by C when C is a compact and connected
simple plane curve. As in the higher dimensional case, the corresponding

Chapter 9
Global Geometry
of Curves
9.1. Introduction and historical notes
We finish our study of classical differential geometry by coming back to the
topic of Chapter 1: geometric properties of curves. Curves were defined as
differentiable maps from a real interval to a Euclidean plane or space. This
definition allowed us to quickly examine the easiest geometric properties
that can be obtained using methods of differential calculus. We already
pointed out in Chapter 1 that we use the term geometric properties only for
those properties depending on the image of the map, that is, on the trace
of the curve.
After our first exposure to curves, we concentrated on the study of sur-
faces, which, on one hand, form the central block of classical differential
geometry and, on the other hand, show more clearly the differences between
extrinsic/intrinsic and local/global properties. We immediately observed an
asymmetry between the definitions of a curve, as a map, and of a surface,
as a set of points. This is why, in Chapter 2, after discussing the concept
of a surface, we introduced the idea of a simple curve as an analogous one-
dimensional object. At the beginning of this chapter, we will deal with the
relation between the two classes of curves: parametrized and simple. We
will show that the simple curves are just the traces of some parametrized
curves and, moreover, if they are connected, they are topologically equal to
either the real line or to the circle.
Another reason for leaving the study of global properties of curves until
the end of the book is that some of these properties are shared with surfaces.
309
54 2. Surfaces in Euclidean Space
Let f Sl -p S2 be a differentiable map between surfaces. We will
say that f is a local diffeomorphism if, for each p e Si, there is an open
neighbourhood Vi of p in Si and an open neighbourhood VZ of f(p) in SZ
such that f(Vi) = V2 and f : Vi - V2 is a diffeomorphism. Notice that a
bijective local diffeomorphism is a diffeomorphism.
Proposition 2.77. Let Si and 52 be two surfaces and let f : Si - S2 be
a differentiable map between them. Then the following three statements are
tree:
A: f is a local diffeomorphism if and only if (df)P is an isomorphism
for each p e Si.
B: If f is a local diffeomorphism, then f is an open map.
C: If Si is a subset of S2, then Si is open in S2.
Proof. Part A. If f is a local diffeomorphism, then we know from Exer-
cise 2.62 that for each p e Si, (df)P is an isomorphism. Conversely, if (df)P
is regular for all p e Si, then the inverse function theorem (Theorem 2.75),
applied at each p, implies that f is a local diffeomorphism.
Part B. Let V C Si be an open set and p e V. By definition, there
are open sets V C Sz with p e Vi, f(p) E V2 such that f : Vi -> V2 is a
homeomorphism. In particular, f(Vi fl V) is an open neighbourhood of f(p)
in SZ included in f(V). Hence, f(V) is an open subset of S2.
Part C. The inclusion i : Si -> S2 is a local diffeomorphism-see Exam-
ple 2.60-and hence, using part B, it is an open map. Thus, Si = i(Sl) is
an open subset of S2. O
Example 2.78. As an immediate consequence of the above result, we can
modify the assertions made in Example 2.69 and in Exercises 2.70 and 2.71.
If all the normal lines of a connected surface S pass through a point
po, then S is an open subset of a sphere centred at po. If, moreover,
S is compact, then S is a sphere with centre po.
If all the normal lines of a connected surface S are parallel, then S
is an open subset of a plane.
If all the normal lines of a connected surface S cut a straight line R
perpendicularly, then S is an open subset of a right circular cylinder
with axis R.
Example 2.79. Let S be a surface and po S. Consider the differentiable
function f : S -p S2, defined by
p - po
by E S.
Ip - poly
Exercises 55
308 S. The Gauss-Bonnet Theorem

I-
For v E TPS, choose a curve a : (-e, e) -> S with a(0) = p and a'(0) = v. for each differentiable function h defined on the sphere such that
Then
(df)(v)
d a(t) -P0 _ 1 (p-po,v) fh(p)dp=0.
(2.2)
- dt t-p I-pol (P-Po).
Ip-pol v Deduce from this that
If p - po E TPS, we have
deg (p) dp
fS2
(df)(p-po)= 1
(_0)_IP0I2IppoI

Ip-po f is any differentiable function defined on the sphere.

that is, p - po E ker(df )P. On the other hand, if v E TS - {0} satisfies (12) Using functions with arbitrarily small support on the sphere and the
(df)(v) = 0, using (2.2), facts that the function deg(¢, -) is locally constant and takes its values
in 7L, where b : S -k S2 is a differentiable map and S is an oriented
Ip-poI2 compact surface, prove that
p-po - (p-po,v)v

and, hence, p - pp E TpS.
Thus, we have just proved that (df)p has anon-trivial kernel if and only
if p - po E TPS. It results from part A of Proposition 2.77 that f is a local
diffeomorphism if and only if one cannot draw a line tangent to S from the
point pp.
deg q5 = deg(q5,p)
for each p e 2 that is a regular value of q5.
(13) Prove that the degree of a differentiable map from an oriented compact
surface to the unit sphere has to be an integer. As a consequence, the
Euler characteristic of such a surface of IL83 is even.

Exercises
(1) T Prove that any surface has an empty interior when thought of as a
subset of ll83.
(2) T Look for an example of a surface which is neither a closed subset of
1[83 nor a proper subset of a larger connected surface.
(3) T Let S be a surface satisfying (S) = S for any rotation b of the space
I[83 whose axis R does not meet S. We will say that S is a surface of
revolution. Prove that, for this type of surface, all the planes containing
the rotation axis R intersect S transversely. If P is such a plane, show
that S fl P is a curve that generates S. Give local parametrizations of
S reflecting this fact.
(4) Let S be a compact surface, and let there be a differentiable function
f : S -> I[8 with at most three critical points. Show that S is connected.
(5) Take a compact surface S and a differentiable function f : S -> ll8 defined
on it. Estimate the number of connected components of S in terms of
the number of critical points of f.
(6) T Let Si and S2 be two compact surfaces that do not intersect. Prove
that there is a straight line that cuts at some point of Sl and at another
point of 52 perpendicularly.
 Exercises: Steps of the proof 307 56 2. Surfaces in Euclidean Space

(7) Given a differentiable tangent field V defined on an open set U C 2 of

the unit sphere, we define a map JV : U -* IE83 by
JV (p) = V (p) n p, b'p E U.
Prove that JV is another tangent field on U. If V = O f , where f is a
differentiable function defined on U, show that
div JVf = 0.
(8) Let h : -> I[8 be a differentiable function with

fh(p)dp=0,
and let a e S2. Show that there exists a function H : - {a, -a} -> IE8
constant along the meridians of the sphere joining awith -a and such
that
(VH)(p) = rh(p) a n p, 'dp e - {a, -a},
where rh is the function defined in Exercise (6).
(9) Prove that, for each differentiable function h : 2 - I[8, with
fh(P)dP=0
and each a e 2, there exist two differentiable tangent fields V and W
defined, respectively, on 2 - {-a} and on 2 - {a} and a differentiable
function H : 2 - {a, -a} -> IE8 such that
div V = h1§2_{_a}, div W = h1S2_{Q}, and JVH = (V - W)1S2_{a,-a}
(10) Let h : 2 -> 1[8 be a differentiable function. Show that the following
statements are equivalent.
a: The integral of h over 2 is zero.
b: There exists a differentiable tangent vector field T on the sphere
such that div T = h.
(Hint for a = b: Fix a e 2 and define T to be the vector field V of the
previous exercise in a neighbourhood A of a and to be W + JVH on
the complement of a neighbourhood B of a such that B C A, where the
field W is given in the previous exercise and the function H is defined
as f H, with H given in the previous exercise and f 2 -> IL8 any
differentiable function with f - 1 in a neighbourhood of A and f - 0
in a neighbourhood C of -a.)
(11) Combining Exercises (2) and (10), show that, if q5: S -> 2 is a differen-
tiable map from an oriented compact surface and the unit sphere, one
has
deg(¢,p)h(p) dp = 0,
J2
Figure 2.8. Triply periodic surfaces
e,2
T Let S= { (x, y, z) C R3 1 + + ez2 = a} with a 7 3. Prove that
e&,a
(7)
S is a surface. Find a diffeomorphism between S and the sphere S2.
(8) A surface is said to be triply periodic when it is invariant under three
linearly independent translations of 113. This type of surface plays an
important role in crystallography. Prove that the sets P, D, and c,
given, respectively, by the equations
cosx+cosy +Cosz=0,
cos(x--y--z)+cos(x+y-z) -cos(x+y+z) = 0,
sin(x + y) + sin(x - y) + sin(y + z)
+sin(y -- z) + sin(x + z) + sin(x - z) = 0
are examples of triply periodic surfaces. They are called, respectively,
primitive, diamond, and gyroid. Show also that, for these three ex-
amples, there are rigid motions of k3 interchanging the two regions
determined by the surface.
(9) I Let 0 : S -> 1R3 be a differentiable map defined on a surface that
satisfies
(d4 )p : TpS - - 1R is injective for all p E S.
q: S ---r O(S) is a homeomorphism.
 Exercises 57 306 8. The Gauss-Bonnet Theorem

Prove that (S) is a surface and that b : S -> (S) is a diffeomor- c: If is the conformal field on the sphere defined in Example 8.26,
phism. If S is compact, show that the second requirement is equivalent one has
to requiring that b be injective.
(10) T Take a positive differentiable function f : S2 -> I[8 on the unit sphere.
(p) _ (d)[aT(p)J - aT (p))
Show that S(f) _ {f(p)p E I[83 p E S2} is a compact surface and that for each t e ][8 and each a, p e S2.
b : S2 -* S(f) given by ¢(p) = f(p)p, for p e S2, is a diffeomorphism. d: For each pair of points a, p e S2, one has
(11) T Let S be a compact connected surface. We say that S is star-shaped q5( -a)
with respect to the origin if 0 S and one cannot draw a straight line
q5( a) = a, _ -a,
from 0 that is tangent to S. Prove that S is star-shaped with respect to ifp#-a.
q5( p) = a
the origin if and only if it is one of the surfaces S(f) of Exercise (10).
(4) Given a differentiable function h : -* IE8 on the unit sphere, show that
(12) j Let S be a connected surface and suppose that each of its points has
an open neighbourhood in S contained either in a plane or in a sphere. the integrals
Prove that the entire surface is contained either in a plane or in a sphere. fh() = f(ho)(p)(Jac)(p)dt,
(13) Let f : Si -* 52 be a differentiable map between surfaces. If p e Si and
{el, e2} is an orthonormal basis of TPS, we define the absolute vale of 0
the Jacobian of f at p as 9h(p) = f(hoc)(p)(Jacc)(p)dt
IJac f - lA (df)p(e2)I. define two differentiable functions f h and gh on the open sets 2 - {-a}
and 2 - {a}, respectively, and that, moreover, we have
a: Prove that this definition does not depend on the orthonormal basis
div (f h aT) _ -h and div (gh aT) = h,
chosen.
b: Prove that IJac f L: 0 if and only if (df)p is an isomorphism. for each a E 2.
(14) Let S be a surface and let I C I[8 be an open interval. A map F : S x I (5) Suppose that a e S2 and that a : ][8 -> 2 is a 2ir-periodic curve p.b.a.l.
][8m is said to be differentiable if, for any parametrization X : U -* S of whose image is the equator of the sphere perpendicular to the vector a.
S, the composition F o (X x idI) : U x I -* ][8m is differentiable in the We define X : 1[82 -* 2 by X (s, t) _ q5t (a(s)). Prove that X(][2) _
usual sense. In this situation, prove the analogue to Lemma 2.36 for all - {a, -a} and that X restricted to any open subset of the plane of
p E S. the form (a, a + 2ir) x IL8 is a parametrization of the sphere.
(15) Let F, G : S x I -> I[8"/z be two differentiable maps and let A be a real (6) Given a differentiable function h 2 IE8 on the sphere, consider
number. Prove that F+G, F-G, AF : S x I -* I[8"/z are also differentiable another differentiable function rh : - {a, -a} -k IEB, given by
and that (F, G) : S x I -> ][8 is a differentiable function. When m = 1
and G(p, t) L 0 for all (p, t) E S x I, show that F/G : S x I -> I[8 is J (h o fit )(p)(Jac fit ) (p) dt.
differentiable.
If a : IIS - 52 is a 2ir-periodic curve p.b.a.l. whose image is the equator of
(16) If F : S x I -* I[8m is differentiable, we define, for each t e I, Ft : S - I[8"/z 2 perpendicular to the vector a, use the parametrization in the exercise
by Ft(p) = F(p, t) for all p e S. Prove that Ft is differentiable for all above to show that
t e I. Analogously, if p e S, we can define a curve ap : I -> ][8by 2ir
ap(t) = F(p, t) for each t e I. Prove that ap is a differentiable curve for (rho a) (s) ds = L h(p) dp.
al1pES. 0

(17) j Consider a differentiable map F S x I- ][8m. For each p e S A s a consequence, r o a has a 2ir-periodic primitive if and only if the
and each t e I, we can define the differential of F at (p, t) as the map integral of h over the sphere vanishes.
Exercises: Steps of the proof 305 58 2. Surfaces in Euclidean Space

(dF)(,t) : TS x R -* Rm, given by

Exercises: Steps
of the proof
(1) Let S and S' be surfaces oriented by Gauss maps N and N', respectively,
and let b : S -> S' be a differentiable map between them. Given a
differentiable tangent field W on S', we define, for each p e S, a vector
V(p) E TS by
det(V(p), N(p), v) = det((w ° ¢)(p), (N' ° )(p),
for all v e TPS. Prove that the tangent field V defined on S in this way
is differentiable by showing that, if X : U -* S is a parametrization of
S defined on an open subset U of j2, the following equalities are true:
det(VoX,NoX,Xu) =det(WoqoX,N'oqoX,(qoX)u),
det(VoX,NoX,Xu) =det(W oq5oX,N'oq5oX,(q5oX)v).
Take the partial derivatives of these equations, alternatively with respect
to v and to u, subtract, and finally obtain
div V = (Jacq5)[(divW) o
q5].
d
(dF)(,t)(v, a) = F(a(s), t + sa), Vv E TS, a e R,
s=o
where a : (-E, E) -* S is any curve with a(0) = p and a'(0) = v. Prove
that (dF)(,t) is a well-defined linear map.
(18) Let O C Rn be an open set, S a surface, and I an open interval of R.
A map G : O -* S x I is called differentiable when the composite maps
pi oG and p2 oG are differentiable, where pi : S x I -p S and P2 : S x I -* I
are the projections. Show that G : O -* S x I is differentiable if and
only if j o G : O - R4 is differentiable, where j : S x I -* R3 x R _ R4
is the inclusion map.
(19) T (Inverse function theorem) Let F : S x I -* R3 be a differentiable map
and (po, to) E S x I such that (dF)(0t0) : 10S x R -* R3 is a linear
isomorphism. Prove that there exist an open neighbourhood V of po in
S, a real number E> 0, and an open neighbourhood W of F(po, to) in
R3 such that F (V x (to - E, to + E)) C W and
iVx(to-e,to+e) : V x (to - E, to + E) -* W
is a diffeomorphism.
(20) T Suppose two surfaces Si and S2 cut each other transversely at a point
p e Si f1 S2. Prove that, if t e R is small enough, then cbt(Si) f1 S2 L: 0,

(2) Combining Exercise (1), Exercise 8.10, and the divergence theorem for where qt is the translation of R3 given by ¢5t(p) = p+ta, for each p e R3,
surfaces, prove that if q5: S -* S'is a differentiable map between two and where a e R3, al = 1.
oriented compact surfaces, then (21) Prove that the union of a set of horizontal planes in R3 is a surface if
and only if the set of their heights is a discrete set in the z-axis.
deg(q5, q) (div W) (q) dq = 0,
sI
for each differentiable tangent vector field W on S'. Hints for solving
(3) For each t e IIS and each vector a e S2, define a differentiable map the exercises
S2 -k 2 by
p + [(cosht - 1) (p, a) + sinh t] a 2
t a(p) _ cosh t + (p, a) sinh t
Vp E S . Exercise 2.19: If the cone C were a surface, since p = (0,0,0) E C,
it would imply that there would be a parametrization of C whose image
Prove that the following assertions are true. would include the point p. By restricting its domain, we could find an open
a: t o = 0 t = c -t, for all s, t e IIS and a e 2. Hence, each neighbourhood V of p in C mapped homeomorphically (by means of this
qt is a diffeomorphism of the sphere and qo is the identity map. parametrization) onto an open ball of R2. If we remove any point of this
b: The Jacobian of the diffeomorphism qt is, given by ball, the resulting punctured ball is connected as well. Thus, we obtain a
1
contradiction by noting that V - {p} is not connected. Indeed, if P is the
" wv 'r" vr' plane z = 0 and P+ and P- are the two open half-spaces into which it
(cosht+ (p,a)sinht) divides R3, since
foreach tEIlt, a,peS2. CnP= {(0,0,0)},
Hints for solving the exercises
we infer that V f1 P+ and V f1 P- are two non-trivial disjoint open subsets
of V.
Exercise 2.24: If the one-sheeted cone C were a surface, since p = (0,0,0) E
C, Proposition 2.23 would imply that there would exist an open neighbour-
hood V of p in Cthat is a graph of a differentiable function h on some of the
three coordinates planes. Since the orthogonal projections onto the planes
x = 0 and y = 0 are not one-to-one when restricted to C, V has to be a
graph over the plane z = 0. But, since in fact
C={(x,y,z)E][831z=+ x2+y2},
that is, C is the graph over z = 0 of the function f(x, y) _ + x2 + y2,
the functions h and f must coincide on the domain of h, which is an open
neighbourhood of the origin in ][82. Since f is not differentiable at (0, 0), we
obtain a contradiction.
Exercise 2.25: According to Proposition 2.23, there are an open subset
V of S with p e V, an open set U C R2, and a differentiable function
h : U -* R such that V is, up to a rearrangement of the coordinates x, y, z,
the graph of h, that is,
V = { (x, y, z) E R3 I (x, y) e U, z = h(x, y) }.
Then, put O = U x R C R3, which is an open set, and f : O -* R, the
function defined by
f(x,y,z) = z- h(x,y).
As fz - 1, zero is a regular value of f and it is easily seen that S fl O =
f ({0}).
Exercise 2.46: Any elliptic paraboloid is aflznely equivalent to the parab-
oloid of revolution
P _ {(x, y, z) E R3 I z = x2 + y2}.
Since each aflzne transformation of R3 is a diffeomorphism, according to
Exercise 2.45, any elliptic paraboloid is diffeomorphic to P. By analogous
reasoning, any two planes of R3 are diffeomorphic. Thus, it suffices to see
that P is diffeomorphic to the plane P with equation z = 0. We define
f:P-*Pby
f(x,y,z) _ (x,y,x2+y2).
This map is differentiable as a consequence of Example 2.30. Furthermore,
the same argument shows that the map g : P -* P, given by
g(x, y, z) _ (x,y,0),
59 304 8. The Gauss-Bonnet Theorem
Aft
-00
k) AWO
a
I
Figure 8.8. A surface S,,
and smoothing the unions, as in Figure 8.8. The height function corre-
sponding to the vector a in this figure has exactly two extremes-which are
global-and 2k saddle points. Hence, X(Sk) = 2(1 - k), according to the
Gauss-Bonnet-Poincare Theorem and Proposition 8.34. In this way, we have
surfaces whose Euler characteristic is any non-positive integer. In fact, any
compact surface of 1R3 is a topological sphere, a torus, or is homeomorphic
to some of these Sk. The proof of this assertion is beyond the scope of this
book, but a consequence of this is that the only possible Euler characteris-
tics have already appeared in our examples. In Section 8.6, we will sketch,
in a list of exercises, a proof of the fact that a compact surface of JR3 always
has an even Euler characteristic. By the way, other interesting facts will
appear.
8.6. Exercise: The Euler characteristic is even
If S is a compact surface of Euclidean space I3 and N : S --> 52 is a Gauss
map, we know that
X(S) = 2 deg N,
according to the Gauss-Bonnet-Poincare Theorem. We will now give some
exercises that will lead to the following result.
Theorem 8.42. The degree of a differentiable map q : S _ 52 from a
compact surface to the unit sphere is always an integer. In fact, deg 0 =
deg( i, q) for each q E R C S2 that is a regular value of 0. As a consequence,
the Euler characteristic of a compact surface is always an even integer.
This result is also true for differentiable maps taking values not neces-
sarily on the unit sphere but on any other compact surface.
8.5. The Gauss-Bonnet formula 303
Figure 8.7. Two toroidal surfaces
surface must remain invariant. For instance, the Gauss curvatures of the
two surfaces in Figure 8.6 take considerably different values; note that one
of them has positive constant curvature and the other one possesses some
regions of negative curvature. In spite of this, they both must have total
Gauss curvature equal to 47r.
Remark 8.39. We usually say that a surface S of 113 is a topological sphere
if it is diffeomorphic to the unit sphere S2. For example, ovaloids are topo-
logical spheres, by the Hadamard-Stoker Theorem 6.5, although, of course,
there are topological spheres with negative Gauss curvature regions. The
Gauss-Bonnet-Poincare Theorem implies that the Euler characteristic X(S)
of such a topological sphere is just 2. Therefore, its total Gauss curvature
has to be exactly 4ir and the sum of the indices at the zeroes of a tangent
field with isolated zeroes has to be exactly 2. In particular, on all these
surfaces each tangent field must have at least one zero, as seen for ovaloids
in Corollary 8.32.
Remark 8.40. A surface in 13 is called a torus, or a toroidal surface, if it
is diffeomorphic to a torus of revolution as was introduced in Example 2.17.
For example, Figure 8.7 shows two samples of toroidal surfaces with quite
different properties. Example 8.27 and the Gauss-Bonnet-Poincare Theorem
lead to the conclusion that the Euler characteristic x(S) of a torus in R3
has to be 0. Therefore, its total Gauss curvature must also vanish. This
means that the negatively curved and the positively curved regions must
compensate for each other. Finally, each tangent field on a torus, with
isolated zeroes, is such that the sum of indices is zero, as we checked for the
fields described in Examples 8.27 and 8.36 on the torus of revolution.
Remark 8.41. It is not difficult to see that there are compact surfaces
in JR3 with Euler characteristic different from 0 and 2. One can construct
a compact surface Sk, with k E N, k > 1, by glueing k tori of revolution
60 2. Surfaces in Euclidean Space
is differentiable as well and is the inverse map of f. Regarding the one-
sheeted hyperboloid and the elliptic cylinder, arguing as before, one should
show that the surface
x={(x,y,z)
is difFeomorphic to that given by
C = {(x,y,z) E I[83 1x2 + y2 = 1}.
Exercise 2.70: Consider the height function h S -* R relative to the
plane passing through the origin with normal direction a. In Example 2.59
we computed its differential and saw that p e S is critical for h if and only if
a is perpendicular to the tangent plane IS. If S is compact, by continuity,
h must have at least one extreme point po . This point will be critical for h,
according to part B of Proposition 2.63. Then a 1 Tp0S and the normal line
of S at po is parallel to the vector a. On the other hand, if all the normal
lines of S are parallel to a, then all the points of S are critical for h. By part
A of Proposition 2.63, h will be constant, that is, (p, a) = c for all p e S
and some c e R. Thus
Sc P = {p e R3(p,a) = c},
and this P is clearly a plane perpendicular to the vector a.
Exercise 2.71: We work here in a way similar to Exercise 2.70, changing
the height function to the square of the distance function to the given line
R, whose differential was computed in Example 2.59.
Exercise 2.72: Consider the function f : S x S -* ][8, given by
f(p,q) = Ip-q12, dpi q E S.
Since S is compact, there is a pair (a, b) E S x S where f reaches its max-
imum. It follows that a L b; otherwise, f would vanish identically and S
would reduce to a point, giving a contradiction to the definition of a surface.
Let R be the line through a and b. It is clear that R cuts S at the points a
and b. We show that R is perpendicular to S at these points. By the same
choice of a and b, the two differentiable functions g, h : S -* ][8, given by
g(p)=p-b2, h(p) _ Ia - pI2, dp E S,
attain their maxima at a and b, respectively. We now use part B of Proposi-
tion 2.63 and the computation of the differential of the square of the distance
function made in Example 2.59.
Hints for solving the exercises 61 302 8. The Gauss--Bonnet Theorem

Exercise 2.73: The function f : S -> ][8, which measures the distance to
the point a e 1[83, given by f(p) _ p - al, for all p E S, is anon-negative
continuous function. Hence there exists
r = inf f(p) = inf Ip - al > 0.
pES pES

By definition of infimum, we can find a sequence {Pn}nEN of points of S such

that
lim f(p) = r.
n-4oo
Since it is convergent, the sequence {f(Pfl)}flEN is bounded; that is, there is
a number C E R+ such that Iin - al < C for all n e N. Then the sequence
{Pn}nEN is inside of a sphere centred at a. Therefore, there is a subsequence
{Pflk }kEN converging to a point p e R3. Since the surface S is closed, we
have p e S, and, since f is continuous,
r = lim
k-+oo
f(Pnk) = f(P)
and, thus, the infimum is attained at p.
Exercise 2.74: Let Po be a point of S f1 P and assume that a e R3 is a
unit vector normal to P. Consider the height function to the plane P. By
hypothesis, this function is either non-negative or non-positive and, thus,
it has an extreme at Po. Using Proposition 2.63 and Example 2.59, we get
a 1 Tpo S. Thus Tpo S = P.
Exercise (1): We first check that no points of S are interior. Indeed, if
p e S, from Exercise 2.25, there is an open subset O of I[83 containing p and
a function f : O -* ][8 such that, up to a rearrangement of x, y, z, f(p) L 0
and f'({O}) = S fl O. Define F : O -* ][83 as follows:
F(x, y, z) _ (x, y, f(x, y, z)), (x, y, z) E O.
Next, applying the inverse function theorem to the map F, prove that there
are two open subsets V and W of II83 such that p e V C O and F(p) _
(pl,p2, 0) E W, F1v : V -* W is a diffeomorphism and is such that
F(VnS)=WnP,
where P is the plane with equation z = 0. Then p is not an interior point of
S since, otherwise, F(p) _ (Pi,P2, 0) would be interior to the plane P. This
is impossible since planes have an empty interior in Euclidean three-space.
(See Lemma 4.1.)
Exercise (2): Let S be the right cylinder built over the trace of a loga-
rithmic spiral contained in the plane z = 0 as in Exercise 1.13. The map
Figure 8.6. Sphere and deformed sphere
Proof. Given the compact surface 5, we know from Proposition 8.37 that
there exists a differentiable tangent field. W on the surface that is non-
degenerate. Proposition 8.25 implies that
x(S) _ E i(W, p)
pES
is an integer and, as a consequence of the Gauss-Bonnet Theorem 8.29 and
Corollary 8.30, it only depends on the surface S and satisfies requirements
B and C. To prove assertion A, suppose that S1 and 52 are two compact
surfaces such that there is a diffeomorphism 0 : Sz -> 52. We know that
x(Si) = E i(W1,p) and x(S2) = i(W2,p),
pESi pES2
where Wl and W2 are two non-degenerate tangent fields on S1 and S2,
respectively. Now consider the vector field W41 defined on 52 from W1 and
. Proposition 8.28 asserts that W'o is, like W2, a non-degenerate tangent
field on the surface 52 and that, moreover,
i(Wo, q) i(Wj,p) = x(SI).
gES2 pESI
Applying statement B, which was already proved, to the surface S2, we have
i(Wi, q) i(W2, 4) = x(S2),
gES2 pESa
and, thus, A is true.
One of the most surprising and beautiful consequences of this Gauss-
Bonnet-Poincaxe Theorem is that, when a compact surface is deformed, for
example, if we dent a sphere, the values taken by the curvatures, say, the
Gauss curvature, can be remarkably modified on those regions affected by
the deformation. However, the total amount of Gauss curvature on the
8.5. The Gauss-Bonnet formula 301
Proposition 8.37. Let S be a compact surface. There exists a subset S of
measure zero of the unit sphere S2 such that, for each a E S2-S, the gradient
Oha of the height function ha, defined on S by p e S H ha(p) _ (p, a), is a
non-degenerate tangent field. Moreover, the sum of the indices at its zeroes
satisfies
(Vha,p) = deg(N, a) -}- deg(N, -a);
PES
that is, it is equal to the number of elliptic points of S with tangent plane
perpendicular to the vector a minus the number of hyperbolic points of S
with tangent plane perpendicular to a.
Proof. Let N : S -p 2 be a Gauss map for the surface S. We define the
subset s C as the complement of the set consisting of the points of the
sphere that are simultaneously regular values for the maps N and -N. By
Sard's Theorem and Remark 4.7, this set indeed has measure zero. Now
let a be any point of S2 - s and consider its corresponding height function
ha on S. Its critical points are just the points of S with tangent plane
perpendicular to the vector a, that is, the points lying in N' ({a, -a}).
Since a and -a are regular values for N, by our choice, the Jacobian of N
does not vanish at these points. But this Jacobian, according to Remark 8.3,
is the Gauss curvature, which is also the determinant of the Hessian of the
height function ha, by Proposition 3.37. Now, from Proposition 8.34, the
vector field Oha is non-degenerate and its indices at each zero are either +1
or -1, depending on the sign of the Gauss curvature at each of them. 0 a be a unit vector in the direction of R, and let b, c e I[83 be two vectors such
From this result, we are able to deduce a stronger version of the global
Gauss-Bonnet formula than we obtained in Theorem 8.29.
Theorem 8.38 (The Gauss-Bonnet-Poincare Theorem). For each compact
surface S in Euclidean space Il83, there exists an integer X(S), which we will
call its Euler characteristic, with the following properties.
A: If Sl and S2 are diffeomorphic, then X(Sl) = X(S2).
B: Any differentiable tangent field V on S, with only isolated zeroes,
satisfies
i(V,p) = x(S).
pE5
C: If K is the Gauss curvature function of S, we have
f
K(p) dp = 2irX(S),
and, thus, 2 deg N = X(S), if N is a Gauss map of S.
62 2. Surfaces in Euclidean Space
X : R2 -* S C R3, defined by
X (u, v) = (e_ucosu,e_usinu,v), (u,v) E R2,
is a bijective differentiable map and its differential is easily seen to be in-
jective at each point. To see finally that X is a parametrization and that,
therefore, S is a surface, it is enough to prove that, if R represents the z-axis,
then S f1 R = 0 and the restriction to S of the map Y - R -* R2, given
by
Y x z= (log 1
/x2 +y
z
is a continuous inverse map for X. Finally, S is not a closed subset of R3
since it can easily be seen that S = S U R. (Compare the proof that S
is a surface with Exercise (9) below and see also Exercise (7) at the end
of Chapter 3.) Suppose that there exists a connected surface S' such that
S S'. Take a point p in the boundary of S in S'. Since S is closed in
- R, we have that p e R. Given any vector v e R3, we can find a sequence
of points pn E S and a sequence of vectors vn e TS such that limp n = p
n
and lim vn = v. This is clear when v is either vertical or horizontal and
n
consequently for any vector v e R3 by linearity. Then the tangent plane
IS' is not well defined, which is a contradiction.
Exercise (3): Let p E S fl P, where P is any plane containing the rotation
axis R, which we will suppose to pass through the origin (and so p L: 0). Let
that a, b, c is a positively oriented orthonormal basis of three-space. Then
the curve a : 1[8 -> II83, given by
a(t) _ (p,a)a+ ((p,b)cost+ (p,c)sint)b+ (-(p,b)sint+ (p,c) cost) c,
for all t e I[8, has a trace included in S since a(t) _ t(p), where t : I[83 -> ][83
is the rotation of angle t around R. Moreover,
a(O)=p and a (0) _ (p, c) b - (p, b) c = p n a.
Thus p n a E TPS. If S and P do not intersect transversely at p, then
Tp5 = TpP = P (see Example 2.52). Hence we would have that p n a E P.
But also a e R C P and p e P. This is impossible since P is a plane. Thus
S cuts transversely each plane containing the straight line R. On the other
hand, if P and Q are planes containing R, then there is a rotation with axis
R interchanging them. Then, if Pf1 S = 0, we would have QflS = Ql. Since S
cannot be empty, S intersects each plane P including R. From Theorem 2.67,
each intersection S fl P is a simple curve that clearly generates by means of
rotations around R the whole of the surface S. Finally, let Po be the plane
passing through the origin perpendicular to the vector c and let a : I -* Po
Hints for solving the exercises 63 300 8. The Gauss-Bonnet Theorem

be a regular curve defined on an open interval of R whose trace is contained

in the simple curve S f1 Po. We will have
a(t) = x(t) a + y(t) b, b't e l,
for some functions x and y. It can be shown that X : I x (8, 8 + 27r) -* I[83,
given by
X (u, v) = x(u) a + y(u) cos v b - y(u) sin v c, (u,v) e I x (8, B + 27r),
is a parametrization of S, for all B E R.

Exercise (6): This exercise can be solved like Exercise 2.72. The function
f : Sl x S2 -> ][8 given by f(p, q) _ p - q12, for each p e Sl and q E 82i is Figure 8.5. Two height functions on a torus
continuous on a compact set and it is positive since Sl fl S2 = Ql. Take a pair
(a, b) E Si x S2 where f reaches some extremum and see, as in Exercise 2.72,
Example 8.35. Consider the height function f : S2 -} III corresponding to
that the line joining a and b satisfies the requirement.
the unit vector a that is given by
Exercise (7): Consider the differentiable function f : ll83 -> ll8, defined by f (p) = (p, a), `dp E IR3.
eX2 ez2 It is clear, using for instance Example 2.59 and the definition of gradient,
f (x, y, z) = + eye + , (x, y, z) E
that the gradient of this height function f is just the conformal field aT

f
whose gradient, given by
z= (2xeX2 , 2yey2, 2zez2 )

vanishes only at the origin. Since f(O, 0,0) = 3 < a, this number a is a
regular value of f. Therefore, by Proposition 2.15, either S is empty or it is
a surface that does not contain the origin. For each (x, y, z) E R3-{(0, 0, 0) },
studied in Example 8.26. We saw there that it was a non-degenerate field
with two zeroes of index +1; that is, f has exactly two critical points that
are extremes, specifically, the global maximum and the global minimum.
Example 8.36. In Example 8.27, we gave an example of a tangent vector

yt
field on the torus of revolution S, given by the equation ( x2 + y2 - a)2 +
we define a function h : R - R by z2 = r2 with 0 < r < a, with no zeroes. On such a torus S, consider the
et2X2 2,
h(t) = f(tx,ty,tz) = +e "2 +e e R. two height functions
Since h(0) = 3 and h(t) tends to infinity as t -* +oo, there is a point of S f (x, y, z) = z and g(x, y, z) = x, 'd(x, y, z) E S.
in each half-line starting at the origin. Moreover, since It is easy to see, for instance looking at Figure 8.5, that the field f vanishes
y2et2y2
h'(t) = 2t(x2et2x2 + + z2et2z2) > 0 if t > 0, along the two circles where the planes z = a and z = -a touch the torus S.
Then f is not a vector field with isolated zeroes. However, the field g is
in such a half-line there will be exactly one point of S. Therefore, this shows non-degenerate on S since g has exactly four critical points
that S is a surface and that the central projection b : S -p S2, defined in
Example 2.79, is bijective. On the other hand, if (x, y, z) E S, then the (-a-r,0,0), (-a+r,0,0), (a-r,0,0), and (a+r,0,0)
position vector of this point of the surface does not belong to the tangent and, among them, the first one is the global minimum of the function g, the
plane T() S; otherwise we would have, according to Example 2.51, fourth one is its global maximum, and the two remaining points are saddle
x2ex2 y2ey2 points. We point out that the sum of the indices of g at these four zeroes,
0 = (df)(,,)((x, y, z)) = + + z2ez2,
according to Proposition 8.34, is zero, as predicted by Corollary 8.33.
which is not true. Taking Example 2.79 into account, we deduce that
(dcb)() is an isomorphism for each (x, y, z) E S. By Proposition 2.77, With these preliminaries and taking Example 3.36 into account, we have
b is a local diffeomorphism that, since it is bijective, is a diffeomorphism. the following corollary to Sard's Theorem.
8.5. The Gauss-Bonnet formula 299
Proposition 8.34. Suppose that S is a surface in R3 and that f : S -p ][8
is a differentiable function defined on it. The following assertions are true.
A: The gradient field V f has its zeroes just at the critical points of
the function f.
B: Let p e S be a zero of the gradient O f . Then the gradient is
non-degenerate at p if and only if the Hessian (d2f)P of f at this
critical point is anon-degenerate quadratic form on the plane TPS.
C: The index of V f at a critical point p of the function f with non-
degenerate Hessian is +1 if f has a local extreme at this point p
and is -1 if f has a saddle point at p.
Proof. Part A. For p e S and v e TS, we have
((Vf)(p),v) _ (df)p(v).
So the first assertion is obvious.
Part B. Suppose that, indeed, the point p e S is a zero of the field
O f , that is, a critical point of the function f. If v e TS, we take a curve
a (-S, S) -> S such that a(0) = p and such that a'(0) = v. By the
definition of gradient, we have
((Vf)(a(t)), a '(t)) _ (df)a(t)(a'(t)) - °a)(t)
for all t e (-6, 6). Taking derivatives at the time t = 0, not forgetting that
(Vf)(p) = 0, we obtain
2
(foa)(t) = ((dVf)(v),v).
t=o
The left-hand side of this equality is nothing more than the Hessian of the
function f at the critical point p, as defined in Definition 3.34, applied to
the vector v, that is,
((dVf)(v),v) _
at each zero p of V f and each v e TPS. Therefore, the differential (dVf)p
is regular if and only if (d2f)p is anon-degenerate quadratic form.
Part C. If p e S is a critical point of f with non-degenerate Hessian,
then Proposition 8.25 implies, in this case, that
+1 if det(d2 f )P > 0,
i(Vf,p) =
-1 if det(d2 f )p < 0,
and we complete the proof by applying Proposition 3.35. U gEVp=gEVgflVp=Sq=Sp=Sa,
64 2. Surfaces in Euclidean Space
Exercise (9): Prove that, if X : U -j S is a parametrization of S, then
b o X : U -f (S) is a paxametrization. Such parametrizations cover (S),
which will be, for this reason, a surface. It is also clear that b : S -j cb(S)
is bijective. Furthermore, it is a diffeomorphism since, by Proposition 2.77,
part A, and our first assumption, ¢ is a local diffeomorphism. Finally, if S
is compact, since b is continuous, it has to be a closed map. Since it is also
a continuous bijective map from S onto cb(S), it must be a diffeomorphism.
Exercise (10): Under these assumptions, show that the map b: §2
given by
(p)=f(p)p, dp E S2
satisfies the necessary conditions of Exercise (9) so that it can be applied.
Exercise (11): It is clear that 0 5(f). If, from ¢(p) = f(p) p e 5(f), we
could draw a tangent line passing through the origin, then we would have
pe (d¢)p(TpS2), that is,
p= (dcb)(v) = (df)(v)p+f(p)v
for some v e TpS2. Taking the scalar product by p and taking into ac-
count that (p, v) = 0, we would deduce that v = 0, and this would be a
contradiction.
Exercise (12): For each p e S there exists an open subset Vp of S such
that p e Vp, and there exists either a sphere or a plane Sp such that Vp C S.
If, for p, q e S, the corresponding neighbourhoods meet, we have
O7 VpfVq C SfSpfSq.
But if Sp Sq, then the intersection Sp f1 Sq is either empty, or it has only
a point, or it is a circle, or finally it is a straight line. Since Vp f1 Vq is
a non-empty open subset of S, using, for instance, a parametrization, we
could get a subset of Sp f1 Sq homeomorphic to an open ball of R2. This is
not possible since the order of the connection of the points is not the same.
Then
Since S 0, for a e S the set
A={pES1Sp=Sa}
is a non-empty subset of S. If p e A, then Vp C Sp = Sa and, thus,

Hints for solving the exercises 65
and Vp C A. Hence A is an open subset of S. On the other hand, if {Pn}nEN
is a sequence of points of A converging to a point p e S, then there exists
NeNsuch that
n>N=pneVp.
Then pN E V. f1 VPN and, as a consequence, Sp = SPN = Sa. Thus p e A
and A is a closed subset of S. Since the surface S is connected, S = A and
so S is included either in a plane or in a sphere Sa.
Exercise (17): A proof similar to that of Lemma 2.48 can be given. Let
(p, t) E S x I, and let X : U -j S be a parametrization covering p. If e > 0
is small enough, then a(-s, E) C X (U). Let /3: (-s, e) -* S x I be the map
given by ,Q(s) _ (cr(s), t -f- sa), whose image is in X (U) x I. We have
F o fQ - [Fo (Xx idl)] o (X-1 x idI) o /3.
From this, according to the definition and applying the chain rule, we have
-
(dF)(p,t) d[F o (X x idl)](e,t)[d(X x
where q = X-i(p). This equality proves the assertion.
Exercise (19): Let X : U -f S be a parametrization of the surface S such
that X (qo) = po for a certain qo e U. The composition
G=Fo(X xidl) : UxI CR2 xR-*R3
is differentiable and
(dG)(qo,to) _ (dF)(0,t0) ° d (X x idI) (go,to) .
Thus the differential on the left side is an isomorphism. Using the inverse
function theorem, there are an open neighbourhood V' of qo contained in U,
a number E > 0, and an open neighbourhood W of G(qo, to) in R3 such that
G(V' x (to - E, to + E)) = W and G : V' x (to - E, to + E) - W
is a diffeomorphism. Then V = X (V'), E, and W are the required objects.
Exercise (20): Let X and Y be two parametrizations of Si and of 52i re-
spectively, both of them defined on the same open subset U of ][82 containing
the origin and such that X(0, 0) = Y(0, 0) = p. Define F : U x ]l8 -f II83 by
F(u, v, t) = X (u, v) - Y(u, v) - ta.
We are done by the implicit function theorem (Theorem 2.14), as in Exercise
(17) at the end of Chapter 1.
298 8. The Gauss-Bonnet Theorem
is a homotopy between b0 = -N and = (V -F pN)/IV + Conse-
quently, since N is the inner normal field of S, we have from Remark 8.3
and Exercise 8.17 that
deg IXI Is_E - - deg XI ISE = if
4 K(p) dp
Substituting this into (8.5), we deduce that
k
X
K(p) dp = deg
s
To complete the proof it suffices to recall Definition 8.24 of the index of a
tangent field on a surface at an isolated zero and the fact that, by Proposi-
tion 8.25, this index vanishes at the points of S where V does not. O
We have a first immediate and important consequence of the Gauss-
Bonnet Theorem.
Corollary 8.30. All the differentiable tangent fields on a compact surface
of II83 with isolated zeroes have the same sum of the indices at these zeroes.
For example, if the surface is an ovaloid, in particular, a sphere S2(r) of
arbitrary radius r > 0, we saw in Remark 6.6 that its total Gauss curvature
is exactly 4ir. From this, we get the following consequence.
Corollary 8.31. The sum of the indices at the zeroes of a differentiable
tangent field that has only isolated zeroes on an ovaloid is exactly 2.
This was the situation that we saw in Example 8.26 for the conformal
fields on a sphere. As a consequence of this, we can generalize the well-known
result by Brouwer that we already have seen in Theorem 8.20.
Corollary 8.32. Each differentiable tangent vector field has at least one
zero on an ovaloid.
Finally, recall that in Example 8.27 we exhibited a non-vanishing dif-
ferentiable tangent vector field on the torus of revolution. From this, the
following statement can be proved.
Corollary 8.33. The sum of the indices of a differentiable tangent field
with only isolated zeroes on a torus of revolution is zero.
In order to strengthen the Gauss-Bonnet Theorem 8.29, we will consider
a particular type of tangent vector fields on surfaces: those vector fields that
are the gradient, as defined in Exercise (17) of Chapter 6, of differentiable
functions defined on them.
8.5. The Gauss-Bonnet formula 297

points al, ... , ak where V does. As a consequence, the vector field X does
not vanish on the regular domain 1 that we constructed in R3. Then, in
this situation, we can apply Theorem 8.11 to the map
X .- 2

Consequently, we have

deg
x -I- deg
X
-I- deg
X
=
Ise

where, in order to compute these degrees, we have to put the inner orien-
tations relative to 1; see Figure 8.4 on S_ and on the spheres S2 (a),
i = 1, ... , k. But, if we consider the same equation for the outer orientations
that each of them has as a compact connected surface, we obtain
k
(8.5) deg IXI Is_E - deg IXI ISE _ deg IXI ISZ(az)
Now then, from the definition of the vector field X, it follows that, for each
p e (-2e, 2e),
XIgp =V oFp 1+pNoFp 1 = (V +pN) oFp 1,
where we already know that Fp : S - S,o is a diffeomorphism preserving
orientation. For this reason, taking Remark 8.6 into account, we conclude
that
V - pN
deg X
X sP
= deg
I V + PNI
for each p y 0 with -2E < p < 2E. Finally, the invariance of the degree
under homotopies, which we saw in Remark 8.14, implies

V + pN
deg N when p> 0,
deg IV
+ pNI deg(-N) when p < 0,
since it is not difficult to see that, if p> 0, q: S x [0,1/p] - S2, given by
N+tV bpES, 0<t< 1,
p
IN +tVI'
is a homotopy between ¢o = N and = (V -I- pNI, and that,
analogously, if p < 0, then /i : S x [1/p, 0] -> S2, given by
N+tV ypES, 1/p<t<0,
IN +tVI
 296 8. The Gauss-Bonnet Theorem

Chapter 3

The Second
Fundamental Form
Figure 8.4. The regular domain S1

for some k e N. We choose, using Theorem 4.26, a real number e > 0

3.1. Introduction and historical notes
Having defined the most convenient notion of a surface for our purposes and
having developed the associated differential calculus, we will now proceed
to the study of the geometry of surfaces which, initially, will be from a
local point of view. As we did in the case of curves with curvature and
torsion, we will define certain functions or other types of geometric objects
which, in some sense, control their shape. The first mathematicians who
made significant contributions to this subject seem to have been EULER
and MoNGE. The latter carried on intense activity as a teacher at the
military school of Mezieres and, after the Revolution,1 as a founder of the
famous Ecole Polytechnique. In fact, MoNGE was its director during the
entire Napoleonic empire. The members of the various schools of French
geometers that arose as a result of his teaching thought of a surface near a
given point as a family of curves passing through this point. In this way, in
order to know how the surface bends in space, they focused on studying the
geometric behaviour of all of these curves.
EULER had already noticed that the curvatures of all the plane curves
resulting from intersections of a given surface with all the planes containing
its normal line at a point are not arbitrary, but, in fact, only two of these
sections are needed to know the curvatures of all of them. More precisely,
one of MoNGE's students during his period at Mezieres, JEAN BAPTISTE
'Interestingly, he happened to be presiding over the Jacobin government on the day that
Louis XVI was executed.
67
such that N26(S) is a tubular neighbourhood of S and, thus, the open set
SZ2e = SZ_ - 11, where 1L and SZ are the inner domains determined by
the parallel surfaces S_ and respectively, is a regular domain whose
boundary is the union of these two parallel surfaces. In other words, we
have
SZ2 = F(S x (-e, e))
where F : S x (-2e, 2e) - N2(S) is the diffeomorphism introduced in (4.1),
corresponding to the normal field N inner to S. We now take an open ball
B (a) of IIS3 centred at the zero ai of V, for i = 1,... ,k, with radius small
enough so that
B(ai) fl B(ad) = Ql when i yj
and also so that
B(a2) C SZ2 for i = 1, ... ,k.
We then have the situation of Figure 8.4. Therefore, according to Re-
mark 5.33, the open subset of IIS3, given by
=26-(B(a1)U''UB(ak)),
is a regular domain contained in the tubular neighbourhood N2(S) whose
boundary is the disjoint union
where each S2(a) is the sphere with centre ai, i = 1, ... , k, which is the
boundary of B(ai).
We now consider, as in Definition 8.24, the vector field X on the tubular
neighbourhood N2E(S) of the surface S, given by
X (F(p, t)) = X (p + tN(p)) = V (p) -}- tN(p), b'(p, t) E S x (-2e, 2e).
Since F is a diffeomorphism, X is awell-defined differentiable map. Fur-
thermore, since V (p) E TS for each p e S, X vanishes exactly at the same
8.5. The Gauss-Bonnet formula 295
From this and Proposition 8.25, it follows that
=2(V,p)
and we are done. 0 tangent plane. This is the second fundamental form of the surface at the
8.5. The Gauss-Bonnet formula
The global version of the celebrated Gauss-Bonnet Theorem asserts that the
total Gauss curvature of a compact surface of R3 is exactly 2ir times a certain
integer associated to the surface: its Euler characteristic. This integer has a
topological nature and can be assigned to the surface in different ways, for
instance, using triangulations or by means of the homology groups of the
surface or through its fundamental group. In any case, two diffeomorphic
surfaces have the same Euler characteristic. Furthermore, if two compact
surfaces of IIS3 share this number, then they are diffeomorphic. A detailed
study of all these features of the Euler characteristic of a surface, and even
its rigorous definition, go beyond both the aim remember that this is a
text on differential geometry and the requisite tools required for reading
this book.
But, fortunately, there is another approach to the Euler characteristic.
In 1885, Poincare showed that the Euler characteristic of a compact surface
also appears when one sums up the indices of a tangent field on the surface
having only isolated zeroes. In 1926 this was generalized to higher dimen-
sions by H. Hopf, after some earlier attempts by Brouwer and Hadamard.
The version of the Gauss-Bonnet Theorem that we will state and prove will
combine the two above-mentioned results and will connect the total Gauss
curvature of a compact surface in R3 with the sum of the indices at the
zeroes of a tangent field of this class on the surface. This will allow us to
see that the total Gauss curvatures of two diffeomorphic surfaces have to
coincide.
Theorem 8.29 (First version of the Gauss-Bonnet Theorem). Let S be a
compact surface in Euclidean space ll83 and let V be a differentiable tangent
field defined on S that has only isolated zeroes. Then
fK(p)dp = 2ir i(V, p),
PES
where K is the Gauss curvature function on the surface S.
Proof. Since the two sides of the required equality are additive with respect
to the connected components of the surface S, we may suppose, as usual,
that the surface is connected. Also, since S is compact, the number of
isolated zeroes of the field V is finite. We will denote them by {ai,.. . , ak },
68 3. The Second Fundamental Form
MEUSNIER (1754-1793), pointed out that all these curvatures are exactly
in modern language the values taken by a certain quadratic form on the
given point, so named by way of contrast with the first fundamental form,
which is nothing more than the restriction of the Euclidean metric to this
tangent plane.
As we pointed out at the beginning of Chapter 2, GAUSS, following some
earlier papers of EULER, was the first geometer who went beyond the re-
sults of the French school when he pointed out that it is more convenient to
think of surfaces as two-dimensional objects, that is, as pieces of the plane
subjected to some deformations. In his book on surfaces cited earlier, he re-
trieves and makes complete sense of the two most famous objects associated
with a surface: the Gauss map, which EULER had already used, and the
total curvature or Gauss curvature, introduced but not totally exploited by
OLINDE RODRIGUES, a passionate follower of the SAINT-SIMON doctrines,
who had been a student of MONGE in the Ecole Polytechnique. In fact, the
statement that this Gauss curvature does not depend on the way in which
the surface bends in R3, but only on its first fundamental form that is,
it could be computed by two-dimensional geometers unable to fly over the
surface is the content of the famous Theorema Egregium that one can find
in the Disquisitiones. GAUSS stressed the importance of total curvature,
once its intrinsic nature had been shown, although this importance was not
widely appreciated at that time. One of the mathematicians who argued
against him by preferring another function that also measures the bending
of the surface was SOPHIE GERMAIN (1776-1831), who had been led to de-
fine the so-called mean curvature-an extrinsic curvature depending strongly
on the way in which the surface is immersed in R3 by the study of some
elasticity problems. Even if one recognizes the fundamental importance of
the discovery of the intrinsic character of the Gauss curvature and its influ-
ence in the later birth of Riemannian geometry, from the point of view of
the geometry of submanifolds, it can be argued that SOPHIE GERMAIN was
right in some sense since, at least, the literature about the mean curvature is
somewhat larger than the corresponding literature on the Gauss curvature.
In this short review of the history of the geometry of surfaces, we should
again mention G. DARBOUX (1842-1917), successor of CHASLES as professor
of Geometrie Superieure at La Sorbonne and of BERTRAND as secretary
of the Academie des Sciences. Among his numerous accomplishments, we
mention the Lecons sur la theorie generale des surfaces, a true compilation
of all that was known about surfaces up to that time and one of the best
written books on geometry.
3.2. Normal fields. Orientation 69 294 8. The Gauss-Bonnet Theorem

ll82; see Proposition 2.23. Then X' _ o X : U -> S'is a parametrization of

Figure 3.1. Normal fields
3.2. Normal fields. Orientation
In Chapter 2 we defined, at each point of a given surface of R3, the corre-
sponding tangent plane. We know that the variation of the tangent vector-
or of the normal vector-led us, in the case of plane curves, to the concept
of curvature and to the consideration of their local geometry. We will now
investigate how the tangent plane of a surface varies when we move from a
point to points close to it. This task is considerably easier if we focus on
how the normal direction to the surface changes.
Definition 3.1 (Field, tangent field, and normal field). Let S be a surface
of Euclidean space R. In general, a--differentiable-vector field on S will
be, by definition, a differentiable map V : S -> R'. If the values taken by V
at each point of the surface belong to the tangent plane of S at this point,
that is, if V(p) E TpS for all p E S, we will say that V is a tangent field to
the surface. If, instead, they are orthogonal to it at each point, that is, if
V (p) 1 TpS for all p E S, we will say that V is a normal field to the surface;
see Figure 3.1. If, moreover, IV (p) = 1 at each point p of 5, V will be said
to be a unit normal field defined on the surface S and we will represent it
usually by N.
It is clear that, at any point p E S, there are exactly two unit vectors
of 113 perpendicular to the tangent plane TpS. The following result tells us
that one can determine continuously, even smoothly, a unit vector normal
at each point, provided that we restrict ourselves to a small piece of the
surface.
Lemma 3.2. Let S be a surface and let X : U -> JR3 be a parametrization
of S. Then there is a unit normal field defined on the open set V = X (U).
S' and, moreover,
(V o X') (u, v) = (dq)x() [(V o X) (u, v)],
for each (u, v) E U. Since (V o X) (u, v) E TX(, )S, we have, using our
choice of X, that
(v o x)(u, v) _ (dX)(,)(W(u, v)), d(u, v) E u,
where the map W = (W', W2) : U -p X82 can be written as
(8.3) W=iroVoX
and, thus, it is clearly differentiable. Thus, the expression above for V o X'
can be transformed into
(8.4) (V o X')(u, v) = d(q o X)(,)(W(u, v)),
for each (u, v) E U. This shows that V o X' is differentiable on U since
d(qoX)(, ) is a linear map from II82 to ll83 depending smoothly on (u, v) E U.
It is also clear by the definition of the field V that, if p e S, then
0 if and only if V(p) = 0.
This is so since (dq)p is an isomorphism. Now suppose that p e S is a zero
of V and that, thus, q(p) is a zero of V. In this case, we already know that
(dV)p is an endomorphism of TS and that (dV() is an endomorphism
of We now study the relation between these two linear maps. To
do this, put p = X (a, b) for some (a, b) E U. Then, from (8.3), we have
W (a, b) = 0. Taking this into account, we take partial derivatives of (8.4)
and obtain
d(v o x')(a,b) = d(q o x)(a,b) o (dW)(a,b).
On the other hand, if we apply the chain rule to (8.3), we have
(dW)(a,b) _ ir o (dV)p o (dX)(a,b).
Putting this equality into the previous one, we see that
d(V o X')(a,b) _ (dq)p ° (dV)P ° (dX)(a,b).
Recalling now that X' _ o X and applying the chain rule to the left-hand
side, we finally have that
(dV)() o (dq)P = (dq)p o (dV)p
or, equivalently,
(dV)() = (dq) 0 (dV) 0 (dq5)'.
An immediate consequence is that V is anon-degenerate tangent field at p
if and only if V is non-degenerate at q5(p) and, moreover,
det(dV)p.
8.4. The index of a field at an isolated zero 293
Figure 8.3. Field without zeroes on a torus of revolution
is a Gauss map for the torus, for each (x, y, z) E S. From this, we check
without difficulty that the differentiable field V : S -> 1R3, defined on the
torus by
1 xz
V (X' y, z) = yz a x2 + y ,
r x2 -+y 2 x2 -+y 2
is tangent to S at each point. In fact, V provides at any point of the torus a
unit vector that is tangent to the meridian passing through this point. We
have, then, an example of a tangent field on the torus of revolution that
does not vanish anywhere. This is in clear contrast to the case of the sphere
where, according to the Brouwer Theorem 8.20, this situation cannot occur.
A fundamental property of non-degenerate tangent fields is that they
can be transferred from one surface to another by a diffeomorphism, and
this process will not change the indices at their zeroes.
Proposition 8.28. Let 0 : S -> S' be a diffeomorphism between two surfaces
in i3 and suppose we have a non-degenerate tangent field V on S. Then,
the map VO : S' -r 1R3, given by
(V" o q')(p) _ (dcb)p(V (p)), Vp E S,
defines a tangent field on S' that is also non-degenerate. Moreover, O(p) E S'
is a zero of VO if and only if p E S is a zero of V and, in this case, we have
i(VO, gy(p)) = i(V, p).
Proof. It is clear, by definition, that V' is a map from S' to IR such that
VO'(q) E TqS', Vq E S'.
We next show that V O is differentiable. Suppose that X : U -> S is a
parametrization of S defined on an open subset U of JR2 whose inverse X-1 :
X (U) C 1R3 -> U C Ii is the restriction of a coordinate projection 7r :
70 3. The Second Fundamental Form
Proof. At each q e U, the vectors Xu(q), form a basis of the tangent
plane TX(9)S. Therefore, if q E U, the vector NX (q) of II83, given by
Xw(q) A Xv(q) _ Xu A X
NX q) - IX(q) A Xv(q) 9'1,
Xu A
is a unit vector normal to the surface at the point X(q). Thus, we can define
a differentiable map
NX : U _ R3
such that
NX (q) I 7'x(q) S and I NX (q) I2 =1
for all q e U. Finally, we denote by N : V = X (U) -j ]l83 the composite
map N = NX o X. Then, N is the required unit normal field.
It seems intuitively clear that a unit field normal to a surface determines
one side of it. We assert now that a connected surface has at most two sides.
Lemma 3.3. If S is a connected surface and Ni and N2 are two unit normal
fields on S, then either Nl = N2 or Nl = -N2.
Proof. Indeed, if p e S, we have that N1 (p) and N2(p) are two unit vectors
perpendicular to the same plane TS. Hence, either they coincide or they
are opposite. Then S = A U B, where A = {p e S I Ni (p) = N2 (p) } and
B = {p e S I Ni (p) _ -N2 (p) }, with A and B disjoint subsets of S that are
closed since Ni and N2 are continuous. Since S is connected, we have that
either A = S and B = Q1 or A = Q1 and B = S.
Definition 3.4 (Orientable surfaces). If, on a given surface S, there is a
unit normal field defined on the whole of the surface, we will say that S is
orientable. Each unit normal field on an orientable surface S will be called
an orientation of S. Then, Lemma 3.2 asserts that any surface is locally
orientable and, when a surface is connected and orientable, Lemma 3.3 states
that there are exactly two different orientations on it. The orientable surface
S will be said to be oriented when a precise orientation has been chosen on
it.
Example 3.5 (Planes are orientable). Planes are orientable as a direct
consequence of Lemma 3.2. Also, one can explicitly construct a unit normal
field on a given plane P = {p e ][83 I (p - po, a) = 0} that passes through the
point po with unit normal vector a. In fact, remember that, if p e P, then
TP={v E R3(v,a) = 0}.
Thus N : P -f ][83 given by N(p) = a for all p E P is a unit normal field
defined on P.
3.2. Normal fields. Orientation 71 292 8. The Gauss-Bonnet Theorem

Example 3.6 (Spheres are orientable). Consider the sphere S2(r) with cen-
tre po and radius r. We know that, if p e S2(r), the corresponding tangent
plane is the orthogonal complement of the vector p - po. Thus, the map
N : S2(r) -j ][83, given by
N(p) = (p - po), Vp

is a unit normal field defined on the sphere-the so-called outer normal field.
Example 3.7 (Graphs are orientable). This assertion follows Lemma 3.2
since it allows us to construct an explicit unit normal field on a given graph.
Let S be a surface that is the graph of the differentiable function f : U - ]l8
defined on an open subset U of ]182. We know that X : U -f ]183, given by Figure 8.2. Conformal field on a sphere
X (u, v) _ (u, v, f(u, v)), is a paxametrization that covers the whole of S. that is, the sum of the indices of the zeroes of V is an integer.
Therefore, N = NX o X-1 : S -* II83, with
NX _ Xv _ Example 8.26 (Conformal fields on a sphere). Given a unit vector a E
IXuAXvI
1

1+ID.fI
(-f, fv, 1), we define a differentiable map aT on the unit sphere by
is a unit normal field defined on S. p E S2 di aT (p) = a - (a, p)p.
Example 3.8 (Inverse images are orientable). In fact, if O is an open subset In particular, (aT (p), p) = 0 for each p E S2; that is, aT is a tangent field on
of ][83, f : O -f II8 is a differentiable function, a e ][8 is a regular value of f, the unit sphere S2. This vector field will be called, for reasons that we will
and S = f1({a}) 0 is the corresponding surface, we have seen that, at not explain here, a conformal field. It is clear that the conformal field aT
each point p e S, corresponding to the unit vector a has exactly two zeroes at the antipodal
points a and -a of S2. On the other hand, if p E S2 and v E TTS2, we have
TS = ker(df )P = l91 E R3 I ((Vf), v) _ O},
and, thus, V f18 = (ffy, fz) is a normal field defined on all S. Since a -
(daT)P(v) _ -(a, v)p (a, p)v.
is a regular value, V f does not vanish at the points of S. Then the map Thus, putting p = a and p = -a in this expression, we have
N : S -f II83, given by (daT)a(v) _ -v and (daT)_a(v) = v, Vv E TaS2 = T_aS2.
N= 1 Of1s = 1 fx fy fz) At these two zeroes, the corresponding linear maps (daT)a and (daT)_a pre-
Ivu I ,f + fy " fz serve the orientation of the plane TaS2 = T_aS2. Hence, each conformal field
aT is a non-degenerate field on the sphere and, according to Proposition 8.25,
is a unit normal field on S.
As a simple example, take O = ]l83 and f(x, y, z) = x2 + y2. Then i(aT, a) = 1 and i(aT, -a) = 1.
S = f({r2}), r > 0, is a right circular cylinder of radius r whose axis is Finally, we see that the sum of the indices of the zeroes of the field aT on
the z-axis. From the above, N(x, y, z) _ (1/r)(x, y, 0) is a unit normal field the sphere is exactly 2.
on this cylinder.
Example 8.27 (A non-vanishing tangent field on a torus). Consider again
Even though we proved that any surface is locally orientable and all the torus of revolution
the previous examples are orientable surfaces, we now want to show that x2+y2-a)2+z2=r2},
there exist non-orientable surfaces. To do this, we will examine in detail an
S={(x,y,z) EIR3I(
explicit example: the Mobius strip; see Figure 3.2. Take a circle C centred where 0 < r < a. Since this surface is given implicitly, one can see, for
at the origin with radius 2 inside the plane of equation z = 0 in R3 and a line example, using Example 3.7, that
segment L with length 2 whose centre is the point (0,2,0) E C and whose
N(x, y, z) = x2 + y2
(x(x2+y2_a),y(x2+y2_a),zy2)
direction is given by the vector (0, 0,1) . Let S be the set of the points of r
8.4. The index of a field at an isolated zero 291 72 3. The Second Fundamental Form

As a consequence of what we learned above, we see that this definition

does not depend, as in the case of the index of vector fields defined on open
domains of Il83, either on U or on r. Also, we immediately have a version
of Lemma 8.22 for surfaces. Now suppose that the point a E S is actually
a zero of V, that is, V (a) = 0. Then, if u E TaS and a : (-S, S) -* S is a
curve on the surface with a(0) = a and a'(0) _ u, we have

0= dt I ((N o a)(t), (V o a)(t)) _ (N(a), (dV)a()).

t=o
Hence,
(dV)a(TaS) C TaS,
and we can think of the map (dV)a as an endomorphism of the tangent
plane TaS. We will say that the tangent field V is non-degenerate at the
point a when the map (dV)a is regular, that is, when det(dV)a 0. We
show that this is equivalent to the field XV being non-degenerate at a in
the sense that we gave to this property for vector fields on R3. Since, by Figure 3.2. Mobius strip
definition,
XV B(a)f1S = VBr(a)lS
we have that R swept out by the segment L while its center runs uniformly along C. It

(dXV)alTds - (dV)a. rotates around its centre in the plane determined by its position vector and
the z-axis in such a way that, when the centre has gone back to the original
In addition, since
position, L has completed only a half-revolution.
Xv(a + tN(a)) = V (a) + tN(a)
We first show that this subset S of R', described in a rather cinematic
for t small, we take derivatives with respect to t at t = 0 and obtain way, is a surface. For this, note that if u E R denotes the angle made by
(dXv)a(N(a)) = N(a). the position vector of the centre of the turning segment L with the original
position, then the centre L has components given by
In conclusion, we have det (dX)a = det(dV)d and, thus, V is non-degenerate
at a if and only if Xv is non-degenerate. Therefore, we have the following c(u) = (2 sin u, 2 cos u, 0).
result.
Proposition 8.25. Let V be a differentiable tangent vector field on a surface When the centre of L arrives at the position c(u), L has turned around
S and let p E S. If V(p) 0, then i(V,p) = 0. If, on the other hand, its centre at an angle of u/2. In this way, the direction vector d(u) of the
V(p) = 0, then (dV) is an endomorphisrri of the tangent plane TPS. If the segment L(u) is determined by applying to the original direction (0, 0,1)
field V is non-degenerate at p, that is, if this endomorphism is regular, p is first a rotation of angle u/2 around the y-axis and, then, another rotation
an isolated zero of V and, moreover, of angle u around the z-axis. Therefore
+1 if det(dV) > 0,
i(V, p) _ cos u sin u 0 1 0 0 0
-1 if det(dV) <0. d(u) = - sin u cos u 0 0 cos 2- sin 2 0
0 0 1 0 sin 2 COS 2 1
Therefore, if the surface S is compact and V is non-degenerate, it has finitely
many zeroes and -sin usin u
z(V, p) E 7L; - cos u sin
pES cos 2
3.2. Normal fields. Orientation 73 290 8. The Gauss-Bonnet Theorem

Consequently, the points of ][83 where the moving segment L lies when its into an integral over the unit sphere by applying it to the diffeomorphism
centre is at c(u), after having turned an angle of u, are exactly of S2 to S(a) given by p H a + rp. We thus have
c(u) + vd(u) _ (2sinu, 2 cos u, 0) + v(- sin u sin 2 , - cos u sin 2 , cos 2 ), 2

where v e (-1, 1). From these considerations, we define a map F : II82 -f I[83
Z(X' a) 4ir z X(a - rp)J3 det((dX)a+Tp(ei), (dX)a+rp(e2), X dp

by for each r > 0 small enough. But, since X vanishes at a and is non-
degenerate at this point,
(3.1) F(u, v) _ ((2 - v sin 2) sin u, (2 - v sin 2) cos u, v cos 2).
X(a+rp)
It is clear that F is differentiable and, according to what we have just seen, lim (dX)a(p) 0, by E 2 .
r- O r
the subset S satisfies
From this, we infer that, as r goes to zero, the integrand above converges
S = F([0, 2ir] x (-1, 1)) = F(][8 x (-1,1)). pointwise to the continuous function
The last equality is deduced from the fact that F(u + 2ir, -v) = F(u, v),
as one can easily check by definition (3.1). Furthermore, F is clearly 4ir- det((dX)a(ei), (dX)a(e2), (dX)a(p)) = .
pE2HI I

periodic in the variable u.

By using the continuous dependence on parameters, we obtain
Now we denote the closed rectangle [0, 2ir] x [-1, 1] in ][82 by R and
let K = F(R). Then F : R -j K is an identification map because it is i(X,a) _ det 4d X)d dp
continuous, surjective, and closed, the latter since it is defined on a compact JS2 (dX)a(p)I3
set. Let RF be the equivalence relation induced on R by the map F. If The conclusion follows by recalling that
(u, v), (u', v') E R, then we have F(u, v) = F(u', v') if and only if
dp 4ir
(2- v sin 2) sin u = (2 - v' sin 2) sin u', J 2 I(dX)a(p)13 I

(3.2) (2 - v sin 2) cos u = (2 - v' sin 2) cos

v cos 2 = v'cos 2 .
We next consider differentiable tangent fields defined on surfaces.
Squaring the first two equalities and adding them, we obtain that
Definition 8.24 (Index of a tangent field on a surface at an isolated zero).
(2_vsin)2 = (2_vIsin)2. Let S be a surface oriented by a Gauss map N : S -* 2 and let V be a
Now, since lvi < 1 and 0 < sin 2 < 1 since 0 < u/2 < 7r, we have lv sin 2 < 1 tangent field on S. In other words, V is a differentiable map from S to
and, hence, 2 - sin 2 > 0. An analogous result can be obtained for the pair ll83 such that (V (p), N(p)) = 0 for all p E S. A point a E S is called a-
(u', v') and, thus, possible-isolated zero of V when there a neighbourhood U of a in S such
u I. u' that V does not vanish on U - {a}. If this is the case, since the open ball
2 - v sin - = 2 - v sin 2 . Br(a) centred at a with radius r sufficiently small is contained in a tubular
2
Therefore, (u, v)RF(u', v') if and only if neighbourhood of U, we can define a new vector field X on the open subset
.u ,
v sin 2 =v sin 2
u'
.
and
u ,
v cos - =v cos - .
u'
Br(a) of 1[83 by
2 2 Xv(F(p, t)) = Xv(p + tN(p)) = V (p) + tN(p), ep E u n Br(a),
Squaring the two equalities again and adding, we have lvi _ Iv'i. If this where F is the map defined in (4.1). With this definition we see that X is
common absolute value vanishes, going back to (3.2), we have that sin u = differentiable and that X does not vanish on Br(a), except maybe at the
sin u' and that cos u = cos u' and, thus, u and u' differ by an integral number point a. Thus, a is a-possible-isolated zero of X. We define the index
of complete turns. Recalling that u, u' E [0, 2ir], we conclude that u = u', of the tangent field V at the point a by
or u = 0 and u' = 2ir, or u = 2ir and u' = 0. However, if lvi _ lv'i 0,
then either v = v' 0 or v' _ -v 0. In the first case, using (3.3), we i(V, a) = j(X, a).
8.4. The index of a field at an isolated zero 289
small radius r will be contained in the punctured neighbourhood U - {a},
we can define the index of the field X at a as the number
X have that u/2 and u'/2 have opposite sines and cosines. Therefore, either
i(X, a) = deg ER.
XJ
I S (a)
It is important to note that by Remark 8.13, the right-hand side does
not depend on the radius r and that, since we have not shown that the
degree has to be an integer, the index of a vector field at an isolated zero is,
for us, a real number. We will see, however, that under some assumptions,
it is not difficult to prove that this index is indeed an integer.
Lemma 8.22. If the vector field X does not vanish at the point a of its
domain, then i(X, a) = 0.
Proof. In fact, in this case, the map X/IX I extends differentiably to a closed
ball centred at a and we are done by Definition 8.21 and Corollary 8.12. O
Another important case where we can prove that the index of a vector
field X at an isolated zero is an integer is the following. Suppose that at
a E O, X (a) = 0. We say that X is anon-degenerate vector field at the
point a when the linear map
(dX)a : - open subset of S, and x a = Fi Ua is a homeomorphism between Ua C
is regular or, equivalently, when det (dX )a = (Jac X) (a) 0. We should
point out that, as a consequence of the inverse function theorem, if the field
X is non-degenerate at the point a, then a is an isolated zero of X. In
general, we will say that the vector field X itself is non-degenerate when it
is non-degenerate at each of its zeroes. In this case, the discussion above
shows that its set of zeroes is discrete.
Lemma 8.23. Let X be anon-degenerate vector field. Then the index
i(X, a) of the field X at each of its zeroes a E O is either +1 or -1, de-
pending on whether (dX)a preserves or reverses orientation in ll83, that is,
depending on whether det (dX )d is positive or negative.
Proof. In fact, by Definition 8.21 of the index of a field at an isolated zero
and Remark 8.4, we have that
Z(X' a) det((dX)(e1),(dX)(e2),X (p))dpi
4ir *(a) I X (p)I3
where, for each p E ST(a), the vectors el, e2 E TS(a) form an orthonormal
basis positively oriented with respect to the outer normal. We now use the
change of variables formula to transform the integral on the right-hand side
74 3. The Second Fundamental Form
have that the angles u/2 and u'/2 have the same sines and the same cosines.
Hence u = u' since u/2, u'/2 E [0, 7r]. In the second case, again by (3.3), we
u = 0 and u' = 27r or u = 27r and u' = 0. Thus, we may summarize all these
calculations in the following expression:
(u,v) =
(U,V)RF(U',V') ==
{ uu=27r,u'=0,v'=-v.
= 0,u' = 27r,v' = -v,
Consequently, F induces a homeomorphism F from the quotient R/RF to
K C IiJ3. This quotient space R/RF is usually called a compact Mobius
strip. Hence K is topologically a compact Mobius strip. The restriction
of F to R' = [0, 27r] x (-1, 1) gives us a homeomorphism between R'/RF
and S = F([0, 27r] x (-1, 1)), which is an open subset of K. Thus S is
topologically an open Mobius strip.
The same computations that we have just made above show that S =
F([a, a + 271] x (-1, 1)) for each a E III, that F is an identification map
between [a, a + 271] x (-1, 1) and S, and that, on the open subset Ua =
(a, a + 271) x (-1, 1) of II 2 , it is injective. In conclusion, v c = F (U) is an
Va C S. Therefore, the Mobius strip S will be a surface, according to our
and
Definition 2.2, if we check that each x a has an injective differential at each
point of its domain or, equivalently, if (dF)() is injective at each (u, v) E
Ilk x (-1, 1). Moreover, in this case, the map x a will be a parametrization.
In fact, it follows from the very definition (3.1) that
_ /(2 - v sin 2) cos u - 2 v sin u cos Z ,
- (2 - v sin 2) sin u - 2 v cos u cos 2 , - 2 v sin 2
u u u
Fv = - sin u sin - , - cos u sin - , cos -
2 2 2
It follows from this that
IFAFvI2 = (2 - v sin Z )2 + 4v2 >0
for all (u, v) E R2. Thus, S is a surface and each X", with a E III, is a
parametrization of S.
We now show that our Mobius strip S is an example of a non-orientable
surface. In order to prove that, it is convenient to consider and solve the
following two exercises.
3.2. Normal fields. Orientation 75
Exercise 3.9. T Let P be a property that can be attributed to the unit
vectors normal to a given surface S in such a manner that, for each unit
vector v normal to the surface at a point p, either v or its opposite -v
satisfies P and, moreover, the two cannot happen simultaneously. Suppose
also that, for each v obeying P, there exist an open neighbourhood V of p
in S and a differentiable unit normal field N defined on V such that N(q)
satisfies P for each q E V. Show that S is an orientable surface.
Exercise 3.10. T Let a : [a, 6] -* S be a continuous curve on a surface
S, and let n E I[be a unit vector normal to S at the point a(a) of the
curve. Prove that there is a unique continuous map N : [a, b] -* IE83 such
that N(a) = n and
IN(t)12 = 1 and N(t) 1 Ta(t)S, `dt E [a, b].
Prove that, if a is a closed curve and S is orientable, then N(b) = n.
Conversely, if, for each closed curve a, the corresponding normal field N
along a with N(a) = n satisfies N(b) = n, then S is an orientable surface.
Consider the curve a : [0, 27r] -* S on the Mobius strip given by a(t) _
F(t, 0), which is clearly differentiable and closed because of the periodicity of
F. In addition, the map N : [0, 27r] -* Il83, given by N(t) _ 0), is
a differentiable normal field along a, and it is obvious that N(0) and N(27r)
are opposite each other since F(u + 27r, -v) = F(u, v) and thus
F(0,0) = F(27r, 0) and 0) _ 0).
Hence, taking Exercise 3.10 into account, S is anon-orientable surface.
Remark 3.11. The proof that we gave to show that the open Mobius strip
is a non-orientable surface consists geometrically in proving that, if we travel
along the central circle of the strip-the trace of the centre of the moving
segment-starting on one side-remember that it always makes sense to talk
about a side locally-when we complete a turn, we will find ourselves on the
opposite side.
Remark 3.12. In Chapter 4, we will prove the important Theorem 4.21,
asserting that any compact surface of ][83 is orientable. In fact, this result is
true if we weaken the hypothesis to suppose that the surface is only closed
as a subset of Euclidean space.
Exercise 3.13. T Suppose that asurface S is aunion S = Si U S2i where
Si and S2 are two orientable surfaces such that Si fl SZ is connected. Prove
that S is also orientable.
Exercise 3.14. T Let S, Si, S2 be three surfaces such that S = Si U S2, Si
and S2 are connected, and Si n S2 has exactly two connected components,
namely A and B. If Si and S2 can be oriented so that the orientations
288 8. The Gauss-Bonnet Theorem
Exercise 8.19. Let S be a compact connected surface and let St be its inner
domain. If a E 1[83 - S, define a differentiable map qa : S -* S2 by
d(p - Ip
-ale
p E S.
Prove that deg Q = 0 if a S2 and that deg qd = 1 when a E St. (Use
Remark 5.34.)
An interesting application of the invariance of the degree under homo-
topies is the following well-known result on tangent fields defined on spheres,
which is due to Brouwer.
Theorem 8.20 (Brouwer's hairy ball theorem). A tangent field on the unit
sphere 2 of ll83 necessarily vanishes at some point.
Proof. Suppose, on the contrary, that W is a differentiable tangent vector
field on 2 with no zeroes. Then V = W/ W is a differentiable tangent
vector field of unit vectors defined on 2, that is, a differentiable map V
52 such that (V(p) , p) = 0, for each p E 2 Using this V, we define a
map
x [0,ir] __ 52
by
q(p, 0) = p cos 0+ V (p) sin 0.
This map is differentiable and, clearly, supplies a homotopy between qo =
and q _ -Is2. From Remark 8.14, we deduce that
deg deg (-
However, Exercise 8.17 above or, equivalently, Definition 8.1, implies that
these degrees are +1 and -1, respectively. This is a contradiction, leading
to the required conclusion.
8.4. The index of a field at an isolated zero
The degree of a differentiable map between compact surfaces, which mea-
sures a kind of mean of the number of preimages, will permit us to introduce
the notion of the index of a vector field at an isolated zero of the field. We
will do this first for fields defined on II and, later, for tangent fields defined
on a given surface.
Definition 8.21 (Index of a vector field at an isolated zero). Let O C
be an open set and let X : O -* R3 be a differentiable vector field defined on
0. Given a point a E O, we will say that a is a possible isolated zero of
X if there exists a neighbourhood U of a in O such that X does not vanish
on U- {a}. In this case, since the sphere S2(a) centred at a with sufficiently
8.3. Degree and surfaces bounding the same domain 287 76 3. The Second Fundamental Form

where p E S, t E [0, e] , and F : S x (-2E, 2e) -* N2 (S), as defined in (4.1), induced on A coincide and those induced on B are opposite, then S is a
is a difeomorphism. For this reason, b is a well-defined differentiable map. non-orientable surface. Show that this is the case for the Mobius strip.
Using Remark 8.13, we have
Exercise 3.15. T Let f : Si -f S2 be a local diffeomorphism between two
deg b1s = deg b1g ,
surfaces Si and SZ with SZ orientable. Let N2 be a unit normal field on S2.
We define a map Ni : Si -f I[83 as follows: if p E Si, we put
where we have chosen the inner orientations on S and SE. Now then, it
follows from the definition of the map /' that aAb
Nl(p) - IaAbI'
V)IS7a and s_b where a, b is a basis of TPSI satisfying
Since F : S -* S is a difeomorphism that preserves orientation see, for N2(f(p))) > 0.
instance, Example 5.17 we deduce from Remark 8.6 that
Show that Ni is a unit normal field on Si and that, consequently, Si is also
deg qa = deg (qb o F') = deg Ib orientable.
That is, two differentiable maps that are homotopic through a differentiable Exercise 3.16. Consider two diffeomorphic surfaces Si and S2. Show that
homotopy have the same degree. Sl is orientable if and only if S2 is orientable.
Exercise 8.15. Let S be a compact connected surface. Suppose that its Exercise 3.17. If S is an oriented surface, N is the corresponding unit
Gauss map N : S -* S2 extends to a differentiable map of S2 to S2, where S2 normal field, and p E S is a point of S, then we will say that a basis {a, 6}
is its inner domain. Show that of the tangent plane TS is positively oriented when det(a, b, N(p)) > 0.
Otherwise, we will say that it is negatively oriented. If Si and SZ are two
K(p) dp = 0. oriented surfaces, we will say that a local diffeomorphism f Si - S2
Is
preserves orientation if its differential at each point of Si takes positively
Exercise 8.16. Let S be a compact surface and let V : S x [a, b] -* ][83 be oriented bases on Si into positively oriented bases on S2. We define a
a nowhere vanishing differentiable map. Prove that function Jac f Si -* Ithat is called the Jacobian of f-compare with
Exercise (13) at the end of Chapter 2-by the equation
deg Va
I
- deg I V6I '
(Jacf)(p) = det((df )P(ei), (df)(e2),N2(f(p))),
with respect to any orientations taken on S and on 2. where {ei, e2} is a positively oriented orthonormal basis of TPSl. Prove that,
Exercise 8.17. Let q: S -* 2 be a differentiable map from an oriented if Si and SZ are connected, f preserves orientation if and only if its Jacobian
compact surface to the unit sphere. Show that if A is an orthogonal matrix, is positive everywhere.
then deg(A o q) - f deg q, depending on det A = fl.
3.3. The Gauss map and the second fundamental form
Exercise 8.18. Check that the function
For the remainder of this chapter, we will assume all the surfaces to be
f(x,y,z)=('+y2_a)2+z2_r2, 0 < r < a, orientable. In any case, according to Lemma 3.2, all the results that we will
which is differentiable on ll83 - {z-axis} and which was used in Example 2.17 obtain can be applied to small enough pieces of an arbitrary surface. Now,
to define the torus of revolution S = f -1({0}), has no critical points on the let S be a surface and let N be a unit normal field on S. Since IN(p)12 = 1
open set for all p E S, if S2 represents, as usual, the sphere centred at the origin of ][83
with radius 1, we have N(S) C 2 and, consequently, each unit normal field
N on S can be thought of as a differentiable map N : S -p S2 of the surface
which is the domain determined by S. From this, deduce that the Gauss into the unit sphere 2. The map that takes each point on the surface to
map of S can be smoothly extended to a map from S2 to the unit sphere and a unit vector orthogonal to the surface at this point will be called a Gauss
confirm the assertion in Exercise 8.9 about the degree of this Gauss map. map on S.
3.3. The Gauss map and the second fundamental form
By analogy with the case of curves, we expect that the variation of the
Gauss map N on the surface S, when we move in a neighbourhood of a given
point, will give us some information about the shape of the surface near this
point. In this case, the variation will no longer be controlled by a unique
function-like the curvature in the case of curves-but by the differential
(dN)p : TS -* Tnr(p)S2. But-see Example 2.53-the tangent plane to the
sphere S2 at the point N(p) is the orthogonal complement of the vector
N(p), that is, TN(P)S2 = TPS. It follows that the differential of a Gauss map
at each point of a surface is an endomorphism of the tangent plane at this
point. Furthermore, an endomorphism of atwo-dimensional vector space
has only two associated invariants, namely, its determinant and its trace.
We are led in a natural way to consider the following two functions defined
on the surface:
(3.4) K = det dN )p
K(p) C
H(p) = -trace (dN),
and H p E 5,
2
which are, respectively, the Gauss curvature and the mean curvature of the
surface at this point. It is clear that, since we are in a two-dimensional
context, the Gauss curvature K does not change its sign when we reverse
the orientation of the surface and, so, it can also be globally defined on
non-orientable surfaces. However, the mean curvature H depends on the
orientation chosen on the surface and changes its sign when we reverse it.
Gradually, in this chapter and in the following ones, we will give different
geometrical meanings to these two functions defined on surfaces in IiJ3.
Since we have at our disposal the endomorphism (dN)p on TS, p E S
and the Euclidean scalar product on the tangent spaces TS, we can define
a bilinear form O'p:
vp:TpSxTpS-*lEB, pES,
vp(v, w) _ -((dN)p(v), w), v, w E TPS.
This is the so-called second fundamental form of the surface S at the point
p, in terms of which we can rewrite (3.4) as
(3.5) K(p) = det vP and H(p) = Z trace vp, p E S.
This seems to imply that there is a first fundamental form at each point
of the surface although we have not yet defined it explicitly. Indeed, it is
nothing more than the restriction to each tangent plane of the Euclidean
scalar product of the space I[83.
Lemma 3.18. The second fundamental form vp of a surface S at any of
its points p E S is a symmetric bilinear form on the corresponding tangent
plane TPS. This can also be expressed by saying that the differential (dN)P
77 286 8. The Gauss-Bonnet Theorem
where each NZ, i = 1, ... ,k, is the Gauss map of the surface SZ inner with
respect to the domain S2. But, if p E SZ and {ei, e2} C is an orthonor-
mal basis positively oriented with respect to NZ, for i = 1, ... ,k, it follows
from (8.2) that
(X, NZ)(p) = (X(p), ei A e2) = ei, e2)
= det((N' o q)(p)> (dq)(e), (dq5)(e)) _ (Jacq1g)(p).
Finally, we obtain the equality
Jac sl + + f Jac ISk = 0;
Js
that is, the sum of the degrees of the restrictions with i = 1, ... , k,
vanishes, which is what we wanted to prove.. O
Corollary 8.12. If a differentiable map q: S -* S' between compact con-
nected surfaces extends smoothly to the closure of the inner domain deter-
mined by S, then deg q5 = 0.
Proof. It suffices to apply Theorem 8.11 above to the inner domain S2 de-
termined by the compact connected surface S.
Remark 8.13. Let Si be a compact connected surface with S2i the corre-
sponding inner domain, and suppose that S2 is another surface of the same
type such that S2 C i . Let S22 be the inner domain of this second surface.
We know, by Remark 5.33, that S2 = SZi - SZ2 is a regular domain with
boundary Si U S2. If q : S2 -* S' is a differentiable map taking values on a
compact surface S', we have
deg q51Sl = deg q5152
provided we choose simultaneously either the inner or the outer orientations
on Si and S2.
Remark 8.14 (Homotopy invariance). Suppose that S and S' are two ori-
ented compact surfaces of R3 and that
is a differentiable map, where a and b are two real numbers with a < b.
Take > 0 small enough so that N2 (S) is a tubular neighbourhood of S,
as defined in Definition 4.24. If S2 is the inner domain of S and SZ the
domain of the inner parallel surface S at distance E -see Remark 4.28
then f = S2 - SZ is a regular domain in R3 with boundary of = S U SE. We
define a map b : f - S' by
(F(p, t)) _ (p + tN(p)) _ (p, a + (b - a)),
8.3. Degree and surfaces bounding the same domain 285 78 3. The Second Fundamental Form

All the summands on the right-hand side containing partial derivatives of of a Gauss map at each point is a self-adjoint endomorphism on the tangent
second order cancel each other by the equality of mixed derivative; thus, plane at this point.
divX = qy, qz) cby).
Proof. Let U be an open subset of and let X : U -p S be a parametriza-
Now then, if p E St, we have that tion of S. Then, as we already know, at each point (u, v) E U, the vec-
tors Xu (u, v) and Xv (u, v) lie in the tangent plane Tx(u,) S, and the vector
gy(p) _ (dq)(O, 1, 0) E T()S', (N o X) (u, v) is perpendicular to this plane. We obtain from this the fol-
lowing equalities of functions defined on U C
(NoX,X)=O and (NoX,X)=O.
(N' o c)x(p) - d(N' o )(1, 0, 0) _ 0, 0)] E Differentiating with respect to the v-variable in the first equality and with
since (dN')() is an endomorphism of Therefore, the three vectors respect to u in the second one and subtracting the two resulting expressions,
we get
(N' ° q( p), q, (p), and q5z (p)
((NoX),X) = ((NoX),X),
lie in a plane. Thus, where we have taken into account the equality of the mixed derivatives
det((N' o )x y, z) = 0 Xuv = Xvu Then, by the chain rule (Theorem 2.61),
.

on f2. By similar reasoning,

det((N' o q5)y, cbz) = det((N' o q5)z, x, qy) = 0,
and, thus, X is a vector field with vanishing divergence.
On the other hand, if {u, u2, u3} is the canonical basis of R3, the defi-
nition of the vector field X implies that
ui, u2) _ (X, u3) = det(N' o q qx, qy),
ul, u3) _ u2) = det(N' o fix, qz),
Uz, us) _ (X, u1) = det(N' o b qy qz)
(N o X)u(u, v) = d(N o X)()(1, 0)
_ (dN)x(,) [(dX)(,) (1,0)] _ (dN)x(,) [Xu(u, v)]
for all (u, v) E U and, similarly,
(No v) _ (dN)x()[Xv(u,v)].
Thus, by the definition of the second fundamental form, we see that
oX(q)(Xu(9), Xv(4')) =°'x(q)(Xv(R), Xu(R))
for all q E U. Since the {Xu(q), form a basis of TX(y)S, the bilinear
form 0X(q) is symmetric.

Consequently, for each p E S2, the two skew-symmetric bilinear forms One of the most important features of a self-adjoint endomorphism on a
(u,v) E XH u, v), Euclidean vector space is the fact that it can be diagonalized. In this way,
as a consequence of Lemma 3.18, we may establish the following definition.
(u,v) E 1183 X H det((N' o q5)(p), (dq)(u), (dq)(v)) For each point p of the surface S, the endomorphism -(dN)p has two real
eigenvalues that we will denote by kl (p) and k2 (p) and that we will suppose
coincide on the canonical basis ul, u2i u3. Hence, they both must be equal, to be ordered so that kl (p) < k2 (p). They are called the principal curva-
that is, tures of S at the point p and, since they are the roots of the characteristic
(8.2) det((N' o )(p), polynomial of -(dN)p, the functions kl and k2 defined on the surface must
satisfy
for all p E S2 and all u, v E ll83.
Now, we apply the divergence Theorem 5.31 to the vector field X whose K=kkH= 1 =12,
k (i+k2), nad kZ2- 2Hk2 +K _ 0,
2
2 1, 2.
divergence vanishes on the regular domain S2 whose boundary is Si U U Sk
Since the discriminant of this second degree equation has to be non-negative
and obtain so that two real roots exist, we have
f(XNi)+...+f(XNk) (3.6) K(p) < H(p)2, Vp E S,
3.3. The Gauss map and the second fundamental form 79 284 8. The Gauss-Bonnet Theorem

and equality occurs at a given point if and only if the two principal cur-
vatures coincide at this point. Associated to these two eigenvalues of the
endomorphism -(dN)p we have the corresponding eigenspaces, which are
two orthogonal lines when kl (p) k2 (p) . In this case, we will call the cor-
responding directions of these two tangent lines the principal directions of
Sat the point p. (See Remark 3.29 below.)
Example 3.19 (Planes). In Example 3.5, we saw that, if P is a plane in
I[83 with unit normal vector a, then a Gauss map for P is the constant map
N - a. Hence (dN)p = 0 and so vP = 0 for each p E P. In other words,
planes have null second fundamental form at each point. As an immediate
consequence of (3.4) and (3.5), we have that the Gauss and mean curvatures
of a plane are identically zero. The same is true for the corresponding
principal curvatures.
Example 3.20 (Spheres). If S2(r) is a sphere with centre po E I[and
radius r> 0, we know from Example 3.6 that its outer Gauss map is given
by
N(p) = r (p - po), b'p E S2(r).
Then, it follows from the definition of a differential that
(dN)(v) = rv, Vv E TpS2(r),
and thus
(v,w) _ -r (v, w), Vv, w E TpS2(r).
Therefore, at each point of a sphere, the first and second fundamental forms
are proportional. Moreover, using definitions (3.4) and (3.5) again, we obtain
K(p) = r2 and H(p) _ -r, p E S2(r).
That is, the Gauss curvature of a sphere is constant and its value is the
inverse of the square of the radius, while the mean curvature of a sphere
relative to its outer normal is constant and its value is exactly the opposite
of the inverse of the radius. Furthermore, the principal curvatures coincide
at each point and equal -1/r.
Example 3.21 (Cylinder). We also determined, in Example 3.8, an outer
Gauss map N for the right circular cylinder S with radius r > 0 and with
revolution axis the coordinate z-axis, which was described by the implicit
equation x2 + y2 = r2. In fact, we saw that N(x, y, z) _ (1/r)(x, y, 0) for all
(x,y,z) ES. Thus
(dN)(,,)(vl, v2, v3) = 1 (v1, v2, 0
r ax
where 0 < r < a, that is, the torus of revolution of Example 2.17. Show
that, if K is its Gauss curvature and (x, y, z) E S, then
K (x, y, z) > 0 if and only if x2 + y2 > a2,
K(x, y, z) <0 if and only if x2 + y2 < a2,
and that the image of the set of the points of S with vanishing Gauss cur-
vature through a Gauss map N of S consists only of the north and south
poles of the unit sphere. Show also that each point of the sphere, except
these two poles, has exactly two preimages through N, one of them with
positive Gauss curvature and the other one with negative Gauss curvature.
From this, determine the function deg(N, -) defined on S2 - {north, south}
and compute deg N.
Exercise 8.10. Let q : S -* S' be a differentiable map. Deduce, from the
area formula in Theorem 5.28 and Definition 8.1 of deg(q, -), that
(10 )(p)(Jac q5)(P)dP= f f(q) deg(, 9)d9,
Is '
for each integrable function f : S' -* Ilk.
8.3. Degree and surfaces bounding the same domain
The main property of the degree of a map between oriented compact surfaces
in R3 is contained in the following theorem. From this property, the well-
known invariance under homotopies as well as our first version of the Gauss-
Bonnet Theorem will follow.
Theorem 8.11. Let S2 C R3 be a regular domain whose boundary 9 =
Sl U U S is a union of k disjoint compact connected surfaces. If S' is
another compact surface and q : S2 -* S' is a differentiable map, then
deg g1sl + ... + deg q1s =0,
where we have taken any orientation of S' and, simultaneously on all the Si,
i = 1, ... , k, either the inner orientation or the outer orientation relative to
the domain ft
Proof. Choose any Gauss map N' : S' S2 on the surface S' and the
corresponding orientation and define a differentiable vector field X on S2
by
-
X _ (det(N' O q cb, q5z), det (N' o b, q), det (N' o b, cb )) ,
where the subscripts x, y, z stand for partial derivatives. Then, recalling the
definition of the divergence of a field see Definition 5.30 we have
a a a
div X = det(N det(N
ay az
8.2. Degree of maps between compact surfaces 283 80 3. The Second Fundamental Form

and, consequently,
1
O(,,) ((V1, v2, v3), (Wi, W2, w3)) = -- (viwl + v2W2)
r
for each (x, y, z) E S and each (vi, v2, v3), (wi, W2, W3) E It is not
difficult to deduce from this that the Gauss and mean curvatures of the
cylinder S are both constant and take, respectively, the values

K(x,y,z) =0 and H(x,y,z) = -1, (x,y,z) ES.

Figure 8.1. Neighbourhood of q where deg(q., -) is constant
contradiction since, by continuity, q(a) = q, but a has to be different from
thep1,...,Pk
In conclusion, if we set Wi = Ui n 4-1(V), with i = 1, ... , k, we have
shown (see Figure 8.1) that, for each q E S' - O(N), there exists an open
neighbourhood V, which can be taken to be connected, such that
0T1(V)=W1U...UWk,
2r
The corresponding principal curvatures are k1 - -(1/r) and k2 - 0, whose
respective principal directions are those of the circle of radius r and of the
line parallel to the axis passing through the given point.
Exercise 3.22. Consider as a surface S the hyperbolic paraboloid given by
S = {(x, y, z) E IE83 z = x2 - y2}.
Study the second fundamental form of S at the point (0,0,0) and show that
it is not asemi-definite bilinear form. In other words, show that its Gauss
curvature is negative at this point.

where the Wi, i = 1, ... , k, are disjoint connected open subsets of S and Exercise 3.23. T Show that an orientable connected surface whose Gauss
where each 01W, is a diffeomorphism from W, to V. Hence, each element of and mean curvatures are identically zero must be an open subset of a plane.
V has exactly k preimages through q5-just as the point q has-and at all Exercise 3.24. T Let S be the regular quadric-see Example 2.16-given
the points of each Wi the map preserves or reverses orientation simultane- implicitly by the second degree equation
ously since its Jacobian does not change sign. This proves that deg(4, -) is
constant on V. 0 = (1,pt)A ( 1) = (Bp,p) + 2(b,p) + c, pE
Remark 8.8. Proposition 8.7 describes the real number deg 0 associated
to a differentiable map between oriented compact surfaces as the mean of where A is a symmetric matrix of order four, B is anon-null symmetric
the number of preimages through 0, counted with positive or negative sign matrix of order three, 6 E ][83, and c E Il83. Show that
depending on whether the orientation is preserved. Thus, it is the mean of
an integer-valued function that is locally constant. On the other hand, all N(p) = ::, dp E s,
the computations that we have made until now have supplied only integer
degrees. In fact, as one might already suspect, the degree of such a map ¢ is a Gauss map defined on S. As a consequence, check that the corresponding
is always an integer, and the function deg(ci, -) is not only locally constant second fundamental form is given by
on R, but also it coincides identically with the degree of 0. In any case, we
will not need to know that much to obtain our version of the Gauss-Bonnet (v,v) _ - 1 IBp+bIV, v), p E S, v E S.
Theorem. We will only need to establish the result when the second surface
is the unit sphere in order to get an improvement of this version. Therefore, an ellipsoid has positive Gauss curvature at each of its points.

Exercise 8.9. Consider the surface of 113, given by Exercise 3.25 (Invariance under rigid motions). Let S be an orientable
surface and let i5 : I[-* IE83 be the rigid motion given by q5(p) = Ap + b,
S = {(x, y, z) E R3 I x2 + y2 - a)2 + z2 = r2}, where A E 0(3) and b E If N is a Gauss map for the surface S, prove
3.4. Normal sections 81 282 8. The Gauss-Bonnet Theorem

on R, where the integrable function was introduced in Section 5.6.

Likewise, it follows that
deg(¢, q) = sgJac)(q), bq E R,
where the sign function of the Jacobian, denoted by sgJac, is given by
+1 if (Jac)(p) >0,
p E S H sgJac(p) _ -1 if (Jacq)(p) <0,
0 if (Jac ) (p) = 0.
Figure 3.3. Normal sections Consequently, if we apply the area formula-the version for surfaces con-
tained in Theorem 5.28-to the map b : S -* S' and the function sgJac
that N' = A N a 0-1 is a Gauss map for the image surface S' = O(S). S -* ][8, we then obtain
Conclude that
(dN').O(p) = A o (dN)p o A-', fs deg(q, 4) dq =
Js
sgJac(p) dp
Now, notice that the integrand on the right-hand side is nothing more than
0'4,(p)((dc)p(v), (dcb)p(w)) = Aw) = ap(v, w)
the function Jac b. This, together with Definition 8.1, gives the required
a'(P)(Av,

for each p E S, v, w E TS, where a and a' stand, respectively, for the second
fundamental forms of S and S. Also, find the relation between the Gauss
and mean curvatures of S and S'.
3.4. Normal sections
We will work at a fixed point p of an orientable surface S on which we have
chosen a Gauss map N : S - S2. For each direction v E TPS tangent to S
at p, we consider the plane Pq, passing through the point p and having as a
direction plane the plane spanned in 1R3 by the vectors N(p) and V. In this
way, {Py I v E TJS} is the pencil of planes whose axis is the normal line of
the surface S at the point p. Each plane in this pencil cuts the surface at the
point p transversely. Indeed, p E S fl P,, for all v E TpS and, if the equality
TpS = TpP = span IN (p), v} holds, then we would have N(p) E TpS, which
is impossible. Applying Theorem 2.67 to this particular situation, we see
that, for each v E TpS, the intersection S fl P., near the point p, is the
trace of a regular curve. This intersection is usually known as the normal
section of the surface S at the point p corresponding to the direction v; see
Figure 3.3.
Proposition 3.26. Let S be an orientable surface, N a Gauss map, and
ap the second fundamental form at p E S. Then, if v E TpS is a unit
vector tangent to S at p, the curvature of the section S fl P, at the point p,
considered as the image of a plane curve with velocity v and where {v, N(p)}
determines a positive orientation for P,,, is just ap(v, v).
equality. Thus, it remains to show that deg is constant on each con-
nected component of R, which is an open subset of S' since R = S' - q5(Al),
where Al C S is the set of the points of S where the Jacobian of b vanishes.
In other words, it remains to show that deg(, -) is locally constant. To
prove this, take q E S' - q5(Al) to be a regular value of b and let
b({q}) = {pi,. .. , pk}
be the set of its preimages. Since (Jac )(p) 4 0 for i = 1, ... , k, each linear
map
(d)p:TpS_4TqS', i = 1, ... , k,
is regular. Using Theorem 2.75 (inverse function theorem), we can find open
neighborhoods, which we can choose to be disjoint, Ul,... , Uj of pl, ... , pk
in S and V1,.. . , Vk of q in S', respectively, such that each restriction Uz
is a diffeomorphism from UZ to V2, i = 1, ... ,k. Then, there exists an open
subset V of S' such that
and
If not, there would exist a sequence {fl}flEN of points of S' converging to q
and a sequence {an}nEN of points of S such that
q(an)=qn and nEN.
Since S - (U1 U U Uj) is a closed subset of S, it is compact. Thus,
a subsequence of {an}nEN, which we will continue to denote by the same
symbol, would converge to a point a E S - (U1 U U Uj). This leads to a
8.2. Degree of maps between compact surfaces 281 82 3. The Second Fundamental Form

if the diffeomorphism b reverses the orientation. Proof. As mentioned earlier, there exists a real number by > 0 and a
curve av : (-6v, 6v) -* II 3which we may assume to be p.b.a.l. such that
Now that we have stated the elementary properties of the degree intro- av (0) = p and av (-6v, 6v) C S n P. Hence, reversing the orientation of a
duced in Definition 8.1 for differentiable maps between oriented compact if necessary, we will also suppose that a(0) = v. Since a( t) E Tai (t) S, we
surfaces, we are going to obtain a geometrical interpretation of this degree have
as a sort of mean value of the number of preimages of a point under the Vt E (-6v, 6v).
(av(t), (No av)(t)) = 0,
map. To do this, we will use the area formula of Theorem 5.24. In fact,
given a map ¢ : S -* S' of this type, let q E S' be one of its regular values, Taking the derivative with respect to t at the time t = 0, we obtain
that is-see Definition 4.5-a point such that (a(O), N(p)) _ -(a'v(0), (N ° av)'(0)) _ -(v, (dN)(v)).
p E q5'({q}) == 4 0. Now it is enough to remember the definitions of the curvature of a plane
We know by Theorem 2.75 (the inverse function theorem) that the set of curve and of the second fundamental form. D
the preimages of q through b is finite. We define the local degree of the map Remark 3.27. This geometrical interpretation of the second fundamental
b at the point q as the integer given by form can be viewed as a reformulation in modern language of the so-called
-f-1 if 0, Euler theorem, generalized later by Meusnier, that is, the discovery of the
deg(qi,4) fact that the curvatures of the infinitely many normal sections of a surface
PEA-1 ({4}) -1 if (Jac)(p) <0. at a given point are not a random set of numbers, but are organized as
the values taken by a second degree polynomial function in two variables,
Hence, this local degree of ¢ at q defines a map by
that is, of a quadratic form. In fact, the most common formulation of this
q E S' regular value of b F-* q) E 7L result is that kv = (v, e 1 ) 2k1(p) + (v, e2 ) 2 k2 (p), where kv is the curvature
on the subset R. of S' of regular values of b, which we know, by Sard's of the normal section at p E S corresponding to the direction v and where
theorem 4.33, has a complement of measure zero. Notice that, grosso modo, ki(p), k2 (p) are the principal curvatures at this point. This can be easily
q) measures the number of preimages of q under ¢ taking into account inferred from Proposition 3.26 above.
the local behaviour of b relative to the orientation and that, moreover, in Remark 3.28. Proposition 3.26 interprets the values taken by the second
general, it is impossible to extend the definition of to the whole fundamental form in terms of curvatures of normal sections. This inter-
of S', since the non-regular values can have infinitely many preimages. The pretation, together with the comparison result of plane curves contained in
relation between the degree of b and this local degree is established in the Exercise (8) at the end of Chapter 1, allows us to give a geometrical meaning
following result. to the sign that the second fundamental form takes for a given direction.
Proposition 8.7. Let b S -* S" be a differentiable map between two We have, in fact, the following.
oriented compact surfaces and let R be the subset of 5' consisting of the If v E TS is a tangent direction at a point p of an oriented surface
regular values of ¢ whose complement is a set of measure zero. Then the S, with Gauss map N, and o( v, v) > 0, then the corresponding
function normal section SnPv lies locally on the side of the affine tangent
deg(¢, -) :RCS' -> 7L, plane toward which the normal vector N(p) points and, moreover,
defined above, is constant on each connected component of the open subset it touches this plane only at p.
R of S' and is integrable on S'. Moreover, If the normal section SnPv of a surface S at a point p corresponding
to a tangent direction v E TS has a neighbourhood of the point
deg q = A(S) J deg(q, 4) d4 p lying on the side of the tangent plane toward which the normal
vector N(p) points, then cr (v, v) > 0.
Proof. It is clear that the function is integrable over S' since it Of course, the analogous assertions for the case of a negative sign are also
follows from its definition that valid. It is a matter of elementary linear algebra that the sign of a quadratic
I -)I :c 1) form defined on a real vector plane is mainly controlled by its determinant.
3.4. Normal sections 83 280 8. The Gauss-Bonnet Theorem

can again apply the divergence theorem (Theorem 5.31) to the same vector
field X on the regular domain SZ' = St - BE-see Remark 5.33-and get
f(XN) _ - f E
dpi

where N is the inner unit normal field. Substituting the precise value of X
and taking (8.1) into account, we have
V
deg-=±
VI 4 2 lApi
sdetA
E
1
3 dpi

and the sign is either positive or negative depending upon whether the ori-
Figure 3.4. Hyperbolic point entation fixed on S is the inner or the outer orientation. We now apply the
change of variables formula to the diffeomorphism
In the particular case of the second fundamental form o'p of S at the point pES2EpESE
p, this determinant is, according to (3.5), the Gauss curvature K(p). When and transform the integral on the right-hand side into an integral over the
this curvature is positive, crp must be definite-positive or negative. This unit sphere, that is,
is why the points of a surface with positive Gauss curvature will be called
elliptic points. We now give a consequence of what we pointed out earlier. deg VI f det A f 1
lApi3 dp.

All the normal sections at an elliptic point of a surface lie The sign depends again on whether the orientation of S is either the inner
on the same side of the tangent plane at this point. or the outer orientation. Suppose that we choose the inner orientation.
Notice that this assertion is not enough to infer that some neighbourhood Then, from Exercise (4) of Chapter 5, which computes the integral on the
right-hand side, it follows that
of the given point is entirely on a side of the tangent plane. However, this
is also true, as will be proved in Proposition 3.38. On the other hand, when / +1 if det A > 0,
K(p) < 0 at a point p E S, the second fundamental form o-p is not semi- d eg I p E S r-
definite, that is, it is a Lorentz scalar product on TpS. Then, the points Api E 2 / -1 if det A < 0,
of a surface with negative Gauss curvature will be called hyperbolic points. which is true when 0 E ft
In this case, the two principal curvatures of S at p have different signs:
Remark 8.6 (Degree and composition with diffeomorphisms). Consider
ki(p) < 0 and k2 (p) > 0 and, thus, the normal sections corresponding to
two differentiable maps
the respective principal directions lie on different sides of the tangent plane.
Therefore we obtain the following statement, which will also be improved in q:S-*S' and 'b:S'-*S"
the next section: between oriented compact surfaces, and suppose that the first of them, b, is
Each neighbourhood of a hyperbolic point of a surface in- a diffeomorphism preserving the orientation; see Exercise 3.17. Thus, since
tersects the two open half-spaces determined by the tangent the Jacobian of a map between surfaces is multiplicative compare with
plane at this point; see Figure 3.4. Exercise 5.2 we have

The points p E S of the surface where the Gauss curvature vanishes will Jac ('u' o b) = [(Jac) ° cbllJaccbl.
be called parabolic when up h 0 and planar when ffp = 0. As a consequence of Definition 8.1, we obtain

Remark 3.29. Keeping in mind that the eigenvalues of a bilinear symmetric o)= f[(Jac ) ° ] iJac l= Jac i = deg ,
form-or of a quadratic form-on a two-dimensional Euclidean vector space J
can be obtained as the maximum and the minimum values taken by the where we have again used the change of variables formula. Analogously,
form on the circle of unit vectors, the principal curvatures and directions of -deg'b
8.2. Degree of maps between compact surfaces 279 84 3. The Second Fundamental Form

where el, e2 E TS form a positively oriented orthonormal basis. As a the surface S at a point p E S, defined in Section 3.3, can be characterized
consequence, the degree of the normalized field associated to V is as follows:

deg
If k1 (p) = k2(p), then all the normal sections of S at p
det((dV)p(ei), (dV)(e2), V(p)) dp
Js IV(p)I3 have the same curvature. Otherwise, the principal direc-
Remark 8.5. A particular case included in Remark 8.4 is when the vector tions of S at p correspond to the two normal sections of
maximum and minimum curvature and, thus, they are or-
field V on the surface S is given by
thogonal.
V (p) = Ap, `dp E S, For this reason, the points of a surface where the two principal curvatures
where A is a regular 3 x 3 matrix or, equivalently, a linear automorphism of coincide or, equivalently, according to (3.6), where the equality K = H2
Suppose that 0 S. In this case, the differential of V coincides with holds, are called umbilical points. This is clearly equivalent to the fact that
A at each point and, if p E S and el, e2 E TS form a positively oriented the second fundamental form is proportional to the first fundamental form
orthonormal basis, we have the Euclidean scalar product at the point, and that the differential of the
Gauss map at this point is a similarity. In short, from an umbilical point,
det(Ael, Ae2i Ap) _ (p,N(p))detA, the surface is seen to bend in the same way along any direction. When all
where N is the Gauss map compatible with the orientation chosen on S. the points of a surface S share this property, we will say that S is a totally
Therefore, umbilical surface. Examples 3.19 and 3.20 show that planes and spheres and,
hence, any open subset of them are totally umbilical surfaces. The following
V detA P (p,N(p)) result says that, in fact, we cannot expect to find more examples.
(8.1) deg VI 47r Js dpi
Api3
Theorem 3.30 (Classification of totally umbilical surfaces). The only con-
and this is true whenever V does not vanish on S, that is, when 0 S. nected surfaces of R3 that are totally umbilical are the open subsets of planes
We now define a new vector field X on the open subset If83 - {0} by and spheres.
X(x) b'x E l[83 - {0}, Proof. Let U be an open connected subset of I[82 and let X : U -i S be
IAI' a parametrization of the surface S. Because of Lemma 3.2, we have at our
which is clearly differentiable. Moreover, disposal a local Gauss map N on the open set X(U) C S. In the proof of
(IAxI5 Lemma 3.18 we have seen that
IAxl3v - 3 ` v>x
(dN)x(,) [Xu(u, v)] = (N o X)(u, v),
for each x E ][83 - {0} and each v E ][83. Thus, X is a vector field with
vanishing divergence on its entire domain. If the origin does not lie in (dN)x(,)[XV(u,v)] = (NoX)v(u,v)
the inner domain St determined by the surface S, then the vector field X for each (u, v) E U. Since S is totally umbilical, the differential (dN)x(,) is
is differentiable on S2 and we can apply the divergence Theorem 5.31 to proportional to the identity map on the plane Tx(u,v) S. Then, this implies
conclude that that there exists a function f : U -* R such that
(X,N) =0. (dN)x(,) = f(u, v) ITX(u,v)S, `d(u, v) E U.
JS
Then, from (8.1) and the definition of X, we have Combining the last three equations, we have

deg pESH Ap
ES 2
=0 if0
-
SL
(3.7) (NoX)=fXu and (NoX)=fX.
App
From the first of the equalities in (3.7), and after scalar multiplication by
If, instead, the origin of ]R3 lies in 1, we choose E > 0 so that the ball B6 the nowhere-vanishing vector-valued function Xu defined on the open set U,
centred at the origin, along with its closure, is contained in ft Then, we we see that the function f is differentiable. Once we have checked this point,
3.5. The height function and the second fundamental form 85
it will allow us to take partial derivatives of the two equalities relative to v
and u respectively. We thus obtain
(NoX)= fXu + f Xuv and (NoX)vu = fXv + .f Xvu
Finally, applying the fact that mixed second derivatives are equal in our
situation, we obtain the following equality of functions defined on U:
fXv = fv Xu
Since Xu(u, v), v) are linearly independent at each point of U, it fol-
lows that fu = f - 0. Since we have taken U to be connected, it follows
that the function f is constant. If this constant is zero, we deduce from
(3.7) that N is constant on X (U) and thus-see Exercise 2.70-X(U) is
contained in a plane of Il83. Suppose then that f is anon-null constant
function. Using (3.7) again, we deduce that X - (1/f)(N o X) - c for some
c E I[83. Therefore
1 surface and the orientation corresponding to the outer normal field on the
IX-c12 = I f (N o x) I2 = f2,
that is, X (U) is contained in the sphere with centre c and radius 1/Ill.
Summarizing, we have proved that, under our hypotheses, the surface S is
covered by open connected subsets, each of which is contained either in a
plane or in a sphere. Since the surface is connected, we conclude the proof
by appealing to Exercise (12) at the end of Chapter 2.
Corollary 3.31. The only connected totally umbilical surfaces that are
closed as subsets of X83 are planes and spheres.
Exercise 3.32. T Suppose that a surface S and a plane P are tangent along
the trace of a regular curve. Show that all the points of this curve are either
parabolic or planar points of S.
Exercise 3.33. T If a surface contains a straight line, then the surface has
non-positive Gauss curvature at all the points of this line.
3.5. The height function and the second fundamental form
In Section 3.4, we obtained an interpretation of the second fundamental form
by considering the surface as consisting locally of a pencil of regular curves
passing through a given point. Now, we will change our point of view and
think of the surface as a two-dimensional object. In this way, we will see
another useful geometrical interpretation of this quadratic form associated
to each point of the surface.
Definition 3.34 (Hessian at a critical point). Suppose that S is a surface
and that f : 5 -* IE8 is a differentiable function on S. If p E S is a critical
278 8. The Gauss-Bonnet Theorem
by Theorem 5.14 (the change of variables formula). Similarly, if b is a
diffeomorphism that reverses the orientation, then deg b = -1.
Remark 8.3. Consider a Gauss map N : S -* S2 of a compact surface S.
When we talk about the degree of N, it is natural to consider the orientation
on S determined by N itself. Once we have fixed this orientation, if p E S
and {ei, e2} C TS is a positively oriented orthonormal basis consisting of
principal directions of S at p, then we have
(Jac N)(p) = det((dN)P(el), (dN)(e2), N(p))
= ki(p)k2(p) det(ei,e2,N(p)) = K(p)
That is, the previous computation shows that the Jacobian of the Gauss
map-either of the two possible choices-is the Gauss curvature, provided
that we consider the orientation compatible with this normal field on the
sphere; see Exercise (5) in Chapter 3 and Remark 6.6. This is one of the
reasons why we are choosing the outer orientation for compact surfaces
unless otherwise indicated. As a consequence,
degN = 1 K (p)d p
S
that is, the degree of the Gauss map of a compact surface is its total Gauss
curvature divided by the area of the unit sphere. For the present, the above
formula is not significant, since the degree of N is nothing more than a real
number without any specific property.
Remark 8.4. Suppose that S is a compact surface in R3 and V : S -* R3 is
a nowhere vanishing differentiable map, or, in other words, a differentiable
vector field, not necessarily tangent to the surface see Definition 3.1
without zeroes. Clearly the map
V 2
is a differentiable map between compact surfaces and, thus, it makes sense
to talk about its degree. To compute it, take p E S and v E TpS. We have
(d) =
wj(dV)p(v)+multipleofV(p).
Consequently,
(Jac Vj)() - IV(p)I3 (dV)(e2), V(p)),
8.2. Degree of maps between compact surfaces 277
situation, we have already defined see Exercise 3.17 the Jacobian for a
given differentiable map b: S -p S' as the function
Jacq:S -*
defined by:
(Jac )(p) = (d)(e2),
where p E S and {ei, e2} is a positively oriented orthonormal basis of TS,
that is, such that
det(el, e2i N(p)) = 1.
It is worth noting that this function would not be awell-defined object if
the surfaces S and S' were not oriented and that it clearly changes sign
when one of the orientations of S or S' is reversed. Notice, however, that
its absolute value
(dq5)p(ei) A (dq5)p(e2)I, Vp E 5,
does not depend on the orientations chosen and, thus, it can also be defined
for non-orientable surfaces. In fact, we defined this absolute value of the
Jacobian for a given differentiable map between surfaces in Exercise (13) of
Chapter 2, and it played a central role in the theory of integration developed
in Chapter 5.
Definition 8.1 (Degree of a map between compact surfaces). If : S -* S'
is a differentiable map between compact oriented surfaces, we define the
degree of the map b to be the real number
deg = A(S') fJaccb.
This degree is clearly additive relative to the connected components of S.
This is why we will assume that both S and S' are connected. Moreover, it
is also immediate that, if one of the two orientations on S or S'is reversed,
this degree changes sign.
In the following remarks, we will gather some of the more important
properties and particular cases relative to the degree of maps between com-
pact surfaces. Throughout this chapter, unless otherwise stated and for
reasons which will become clear later, each compact connected surface will
be oriented by means of its outer Gauss map.
Remark 8.2. If the map b is a diffeomorphism, then, either from Defini-
tion 8.1 or by Exercise 3.17, its Jacobian Jac b has a constant sign, either
positive or negative according to whether it preserves or reverses the orien-
tation. If b preserves the orientation, we have
de=5') / i
11=1,
is,
86 3. The Second Fundamental Form
point of f, as introduced in Section 2.6, then we define the Hessian of f at
the point p as a map (d2f)P of TS into lI8 defined by:
(d2f)(v)=
2
(fo)(t), v E TPS,
dt t=o
where a : (-e, s) - S is a curve on the surface with a(0) = p and a'(0) = v.
Proposition 3.35. The map (d2f)p : TS - ][8 is well-defined by the pre-
vious expression at any critical point p E S of the function f, that is, this
expression does not depend on the choice of the curve. Furthermore, the
following properties hold:
A: (d2f)p is a quadratic form on the tangent plane TPS.
B: If the function f has a local maximum-respectively, minimum-
at the point p, then the map (d2f)P is semi-definite negative-
respectively, positive.
C: If (d2f)p is anegative-definite quadratic form -respectively, posi-
tive definite-then f has an isolated local maximum-respectively,
minimum-at p.
Proof. Choose a parametrization X U -> S with p = X (q) and q =
(a, b) E U. Taking s > 0 sufficiently small, we may assume that a(-s, e) C
X(U). If we represent the composite curve X-loa by j3 and write explicitly
its components ,Q(t) _ (u(t), v(t)), then we have
(f o = (f o X) o (X' o )(t) = (f o X)(u(t), v(t)).
Computing the second derivative with respect to t at t = 0 and applying
the chain rule, we have
(.f o a X)(a, b)u'(0)2 + 2(f o X)(a, b)u'(o)v'(o)
(f a X)(a,
+(.f a X)(a, b)v'(0)2 + (.f 0 X)(a,
On the other hand,
(f 0 X)u(a, b) =d(f 0 X)(a,b)(1, 0) _ (df)p[(dX)(a,b)(1, 0)] =0
since p is a critical point of the function f. Similarly, (f o X)(a, b) = 0.
Hence,
(f 0 cE)"(O) =
(.f 0 X)(a, b)u'(0)2 + z(.f 0 X)(a, b)u'(o)v'(o) + (f X)(a, b)v'(0)2.
Now, it suffices to realize that (u'(0), v'(0)) = Q'(0) _ (dX')(v) and, thus,
we can conclude that the right-hand side of this equality depends only on
v and does not depend on the curve a and that it is a quadratic form in v.
This proves assertion A of the proposition.
3.5. The height function and the second fundamental form 87
To establish B, suppose that f has a local maximum at p, that v E TS,
and that a : (-s, e) -i S is a curve with a(0) = p and a'(0) = v. Thus, the
function focE has a local maximum at 0 and, hence, (d2f)(v) _ (f oa)"(0) <
0, that is, (d2f)P is negative semi-definite.
As for assertion C, the calculations we made to establish A imply that,
if (d2f)p is negative definite at the critical point p, then q = (a, b) is a critical
point of foX: U C IE82 - IE8 and its Hessian matrix at this point is negative-
definite. Using elementary calculus, the proof can be easily completed. O
Example 3.36 (Height function). Let h : S -> lI8 be the height function
relative to the plane that passes through po E 1183 with unit normal vector
a. Let p E S be a critical point of h. We proved in Example 2.59 that this
is equivalent to a 1 TS and, thus, we will suppose that N(p) = a for a
suitable local Gauss map N. If v E TS and a is a curve on S with velocity
v, then we have
2
(d2h)p(v) ((t) - poi a) _ (a '(O), a) _ (&'(O), N(p))
=d2 I
t=o
By Proposition 3.26 we infer that
v)
Note that a point of a surface is always critical for the height function relative
to the tangent plane at this point. This allows us to assert the following
result.
Proposition 3.37. The second fundamental form vpo of the surface S at
the point po with respect to the choice of normal field N(po) at this point
is nothing more than the Hessian (d2h)0 of the height function h(p) _
(p - a, N(po)), p E S, a E IE83, relative to any plane parallel to the tangent
plane at this point.
Now, if we combine Propositions 3.35 and 3.37, we obtain the following
information about elliptic and hyperbolic points of a surface, which improves
what we obtained in Remark 3.28.
Proposition 3.38. Let S be a surface, K its Gauss curvature function, and
p E S. Then the following hold.
A: If K(p) > 0, that is, if p is an elliptic point, then there exists
a neighbourhood of p in S lying on one of the sides of the affine
tangent plane of S at p. Moreover, p is the only contact point, in
this neighbourhood, between the surface and the plane.
B: If K(p) <0, that is, if p is a hyperbolic point, in each neighbour-
hood of p in S, then we find points of the surface on both open
half-spaces determined by the affine tangent plane of S at p.
276 8. The Gauss-Bonnet Theorem
is equal to the product of its spherical excess ai + a2 + it by the square
of its radius. Later, in 1848, BONNET generalized GAUSS'S result to any
simply connected region of a surface, thus initiating what is now known as
the Gauss-Bonnet formula, although here it still was of a local nature.
We will only consider the global version of the theorem since it better
matches the goals of our book and is more relevant from the point of view of
modern geometry. Moreover, since we are assuming that the reader only has
the knowledge of topology found in an introductory course, we cannot prove
that the total Gauss curvature of a compact surface is equal to 2ir times its
Euler characteristic where the latter arises typically from triangulations or
homology groups or the fundamental group of the surface, etc. That is, we
will not assume that the reader is familiar with this particular topological
invariant and its properties. Thus, what we will show is that total GAUSS
curvature is 2ir times a certain integer which, at first, will be the sum of
the indices of a tangent field on the surface with isolated zeroes, that is, a
hybrid of the Gauss-Bonnet formula and the famous results by POINCARE
and H. Hopp. As a consequence, this integer will be an invariant of the
diffeomorphisms class of the surface and will be called, of course, the Euler
characteristic. Next, we will prove that this integer is twice the degree of
the Gauss map N, that is, twice the number of preimages counted with
their orientations of a point in S2 that is a regular value of N. Hence, it
will always be an even number, as we will prove at the end of the chapter.
We hope that the above discussion convinces the reader that this theo-
rem deserves a chapter all by itself. It clearly reveals the spirit of modern
geometry, which is to connect several quantities in strange and exciting ways
that, at first, might appear to be completely unrelated: the total Gauss cur-
vature of a compact surface, the number of zeroes of a tangent field on the
surface, and the degree of the Gauss map. In fact, the result is not only
very elegant, as GAUSS claimed about his local version of the theorem, but
also, from a modern point of view, it is more egregium than the Theorema
Egregium that we studied in Chapter 7. It is the illustrious precursor of
a series of results relating the global topological structure of surfaces, or
more generally, Riemann manifolds, to the properties of their curvatures.
Furthermore, like the Gauss-Bonnet Theorem, most of the results of global
differential geometry have something to do with some integral formula or,
using more modern language, with some cohomological object naturally as-
sociated with the differential calculus of the surface.
8.2. Degree of maps between compact surfaces
Let S and S' be two surfaces in Euclidean space R3 that we will assume to
be oriented by means of two Gauss maps N and N', respectively. In this

Chapter 8
The Gauss-Bonnet
Theorem
8.1. Introduction
The problems that we dealt with in Chapters 6 and 7 should have given the
reader enough evidence to convince her/him that global differential geometry
is one of the most attractive branches of mathematics. In this discipline,
we can see how closely general, algebraic, and differential topology interact
with analysis. One of the most beautiful and famous examples of this topic
is the Gauss-Bonnet Theorem or formula which relates the integral of the
Gauss curvature over a compact surface, i.e., its total Gauss curvature, with
an integer naturally associated to it, i.e., its Euler characteristic. It will turn
out that this integer is the same for any two diffeomorphic surfaces.
The first version of this result -a local version at the time was obtained
by GAUSS in his Disquisitiones. There, he discovered a relation between the
integral of the Gauss curvature over a geodesic triangle and the sum of the
angles at its vertices. We state it precisely:
The total Gauss curvature of a geodesic triangle is equal
to the excess over ir of the sum of its angles.
GAUSS commented on this theorem as follows: This theorem, if I am not
wrong, should be counted among the most elegant ones of the theory of skew
surfaces. Indeed, this theorem of GAUSS generalized another well-known
theorem of LEGENDRE in spherical trigonometry: if we denote by al, a2,
and a3 the three angles at the vertices of a spherical triangle-whose edges
are by definition arcs of great circles-then the area of a spherical triangle
275
88 3. The Second Fundamental Form
Example 3.39 (Square of the distance). We consider the function f mea-
suring the square of the distance from each point of S to the point po E IE83,
that is, f(p) _ Ip - po I2, p E S. We have already seen in Example 2.69 that
p is a critical point of f if and only if po belongs to the normal line of S at
p. Suppose that this is the case, that is,
po = p + AN(p) for some A E ll8,
where N is a Gauss map for S. Now take v E TS and let a be a curve in
S with velocity v. Then
2
IcE(t) -P012
dt2 t=o
= 21v12 + 2(a"(0),p -po) = 21v12 - 2A(aV'(0),N(p))
Again using Proposition 3.26, we have
- Aap(v,v)],
(d2f)p(v) = 2[1v12 po = p+AN(p)
We can infer from this some interesting results similar to what we obtained
in the previous example for the height function. Two of them are proposed
to the reader in the following exercises, and the remaining ones are in the
exercises at the end of this chapter.
Exercise 3.40. T Show that there are no compact surfaces that have neg-
ative Gauss curvature everywhere.
Exercise 3.41. T Let S be a surface and p E S. Prove that p is an elliptic
point if and only if there exists a point po E Il83 such that p is a local
maximum of the square of the distance to the point po.
Exercise 3.42. T Show that there are no compact surfaces that have non-
positive Gauss curvature everywhere.
3.6. Continuity of the curvatures
On each of the tangent planes of the surface, we have defined a quadratic
form, namely the second fundamental form, which contains, as we have
begun to see, non-trivial information about the geometry of the surface
near the point. In the case of curves, this role was played by the curvature
and the torsion functions. They both appeared naturally as differentiable
functions. However, until now, we have not said anything about the way
in which the second fundamental form or the curvatures arising from it
depend on the points of the surface. We proceed in this section with an
examination of this dependence and some of its consequences.
3.6. Continuity of the curvatures 89
274 7. Intrinsic Geometry of Surfaces

Proposition 3.43. The Gauss and mean curvatures of any surface are
differentiable functions. The principal curvatures are continuous functions
that are differentiable on the open subset consisting of the non-umbilical
points.
Proof. Let S be the given surface and let X : U -* S be a parametrization
of S. In Lemma 3.2, we constructed a Gauss map N = Nx o X' on
the open subset X (U), where Nx = Xu A XV/IXu A We work at
an arbitrary point of U without representing it explicitly. Thus, since the
vectors {X, X} form a basis of the plane TS, the first fundamental form
(i.e., the scalar product), the second fundamental form, and the differential
of the Gauss map can be represented with respect to this basis by means
of symmetric matrices of order two. We will denote them by M, E, and A
respectively. Writing the corresponding entries, we have
case, the vectors v and w would have the same direction and, consequently,
since the vectors (df)('u) and (df)(y) would be in the direction of R_
1(x), they would be proportional. Since (df)z is an isomorphism, u and
y would be proportional as well, and this is a contradiction. In short, the
vector y is parallel to u and thus
f(R(x, u)) = f(R(x, y)) = Rf(),
which is the required equality (7.27) shown to be true for all x E P.
Now, take p, q E S such that Rp 4 RQ and let Rl and RZ be two parallel
lines in the plane P such that
f(R1)=R and f(R2)=Rq,
whose existence follows from (7.27), which we have just seen, and the fact
that f is surjective. Using Exercise 7.63 above, we have

M= EF E= IXI2, F= (X,X), G= Xv2.

dist (f (x), RQ) < dist (x, R2), b'x E Rl.
(3.8)
FG ,
u u I l

Since Rl and R2 are parallel, dist (x, R2) = dist (R1, RZ), and hence
The functions E, F, G : U C IlS2 -* R are obviously differentiable. As for the dist (z, RQ) < dist (Rl, R2), `dz E R.
second fundamental form, Therefore, Rp and Ry are two straight lines of ][83 that do not intersect
(3.9) and such that the distance from the points of one of them to the other one
= o f), f 9
e=a-x (X,u X),
u f = ax (X, Xv),
u 9 = ax (Xv, Xv). is bounded. Hence, they are parallel. Thus, there exists a vector v E S2
such that, for each p E S, the straight line R(p, v) is entirely contained in
S. Consequently, if Q is a plane perpendicular to v, then Q and S cut
We will show that these three new functions e, f, g : U C IlS2 -* R are
transversely, and their intersection, which is non-empty, is a simple curve-
also differentiable. This is a consequence of the definition of the second
see Example 2.64-that is closed in Q since S is closed in ][83.
fundamental form and of the proof of Lemma 3.18. In fact,

e = ax(Xu, Xu) = -((dN)xXu, Xu) = -((N o X)u, Xu) = (NoX,X),

since (N o X, Xu) - 0. Therefore, if we make the same calculation for f and
g, we obtain

e (Nx, Xuu) = 1 (Xu A Xv, Xuu)

IXu n Xv I
1
det(Xu, Xv, Xuu),
Xu n XvI
f = (Ny Xuv} = 1
IXu n Xv I
Xu nX Xuv)
(3.10) 1
det(Xu, Xv, Xuz,),
Xu AXv l
9
( Nx Xvv) =
,
1 Xu A X Xvv)
etv
( v,
IXu n Xv I

IXu
1
Xvl
d Xu,XXvv),
( ,
7.7. Appendix: Some additional results of an intrinsic type 273 90 3. The Second Fundamental Form

Theorem 7.64 (Hartman-Nirenberg, Stoker, Massey). Let S be a surface where the expressions on the right-hand sides are clearly differentiable func-
closed in Il83 with zero Gauss curvature everywhere. Then there exists a tions on U. Finally, the matrix A representing the endomorphism (dN)X in
simple curve C closed in a plane P of Il83 such that the basis {X, that we are working with is
5= U(p,a),
pEC
(3.11) A=
-M-'E
_ - (E
F,
F
G) -1 (fg) .

where a is a unit vector normal to P and R(p, a) denotes the straight line
passing through p with direction a. That is, S is a right cylinder over a
Thus, since trace (dN)X = -trace A and det(dN)X = det A, we have by
definition (3.4) that
simple curve closed in a plane.
K o X= eg-f2 2 eG F2 f F
Proof. We may consider separately each connected component of S, or, (3.12)
EG - F,2
and H o X-
equivalently, suppose from now on that S itself is connected. If S is a plane,
where the six functions E, F, G and e, f, g are given by (3.8), (3.9), and
it is clear that S is a right cylinder over any of its lines. Thus, assume that
(3.10), and thus they are differentiable with EG - F2 = IXu A 0.
S is not a plane. By Lemma 7.59, we know that, for each p e S, there exists
Since this holds true for any parametrization X, the functions K and H are
a unique straight line Rp such that p e Rp C S and, by Proposition 7.60,
differentiable on S. Concerning the principal curvatures kl, k2 : S -* I[8, we
there exists a plane P C R3 passing through the origin and a local isometry
know by the definition preceding inequality (3.6) that they can be expressed
f : P -* S that is surjective; take P = Tpo and f = exppo for any p0 E S.
in terms of K and H as:
We will show first that one can find a non-null vector u e P such that
(7.27) f(R(s, u)) = R f(), bx E P, kZ=H± H2-K, i=1,2,
where R(x, u) is the line in P passing through x with direction u. In fact, from which it is easy to complete the proof. 0
let p = f(0) E S and denote by v a unit vector in the direction of the line Remark 3.44. This proof provides us with the two equalities in (3.12),
Rp C S. We then have v e TS. We choose which allow us to easily compute the Gauss and mean curvatures and the
u = (df)1(v) E P. principal curvatures of a surface on the open subset covered by a parametriza-
tion in terms of functions arising from the parametrization itself and its first-
This u is a unit vector since (df)o is a linear isometry. Moreover, since and second-order partial derivatives. Note that on a large number of sur-
R(0, u) is a geodesic of P-see Example 7.26-its image under f must be, faces, there is at least one parametrization whose image is dense on the
by Theorem 7.25, a geodesic of S passing through f(0) = p with velocity at surface. In this case, although the surface itself might not be entirely cov-
this point (df)o(u) = v. But this geodesic has to be, from the uniqueness of ered by the parametrization, we will still be able to compute these curvatures
Theorem 7.30, the line R f(o) = R since, according to Exercise (15) of this using this parametrization since the curvatures are continuous.
chapter, any straight line contained in a surface is a geodesic. That is,
Example 3.45 (Paraboloid). Let S be the elliptic paraboloid given by
f(R(0,u)) = = Rf(o),
which is the required relation (7.27) for the origin 0 E P. Now, let x e P S = {(x,y,z) E I[83 12z = x2 + y2}.
be an arbitrary point of the plane P and let R f() be the straight line This surface is the graph of the function f : I[82 -* I[8 given by f(x, y) _
corresponding to its image f(s) E S. Denote by w a unit vector in the (1/2)(x2 + y2). Then S itself is covered by the following parametrization
direction of this line, and let y denote the vector (df);1 (w) E P. Suppose (see Examples 2.8 and 3.7):
that y and u have different directions. Then, the two lines R(x, y) and
R(0, u) of the plane P intersect at a point z. But, reasoning exactly as we X (u, v) _ (u, v, 2 (u2 + v2)), (u, v) E I[82.
did for R(0, u), it turns out that The first and second partial derivatives of X are
f(R(x,y)) = Rf(). Xu(u, v) _ (1,0,u) and v) _ (0,1, v),
Hence we have a point f(z) in R f() fl R f(o) and this is not possible, by the
uniqueness of Lemma 7.59, unless R f() = R f(o) = R. If the latter were the Xuu(u, v) = (0,0, 1), v) = (0,0,0), and v) = (0,0, 1).
3.6. Continuity of the curvatures 91 272 7. Intrinsic Geometry of Surfaces

Thus, the coefficients of the first and second fundamental forms, given in where we have used the first equality of (7.26). Finally, Lemma 7.11 implies
(3.8) and (3.10), are 2

E(u, v) = 1 +u2, F(u, v) = uv, G(u, v) = 1+v2, £ic8i = o,

1 1
f(u, v) = 0,

I_
e(u, v) = g(u, v) = since the surface has vanishing Gauss curvature. As a consequence, IC8 I is
1 + u2 + v2' 1 + u2 + v2 an affine function in t that satisfies
It follows, using (3.12), that
2+u2 +v2 =0.
K(X (u, v)) _ and H(X (u, v)) = 2
(1+u2+v2)2 (1+2 +v2) 2 Thus It) = a(s)t for some function a(s) and all t e R. But
Finally, if (x, y, z) E S, we have
1 + z2 Gs(s, t) _ Sexppta(s) = t(dexpp)t«(s)a'(s)
K(x, y, z) = 1 and H(x, y, z) =
(1+2z 2 ) 2 (1 +2x2)2 and, from this, it follows that, for any t,
Example 3.46 (Helicoid). The helicoid is defined as the surface S = X (IE82), a(s) = I(dexpp)ta(s)a'(s)I.
where X : Il82 -* ][83 is the map
Putting t = 0, since (dexp)oP is the identity map, we have
X (u, v) _ (vcosu,vsinu, au), a E IE8*, (u,v) E IE82.
It can be easily shown that S is indeed a surface and that X is a parametriza- a(s) _ (dexpP)opa'(s)I _ Ia'(s)I = 1,
tion that covers it entirely. Its first and second partial derivatives are since the curve a was taken to be p.b.a.l. Finally,
Xu(u, v) _ (-vsinu,vcosu,a) and v) _ (cosu,sinu,0),

X(26,v) _ (-sinu,cosu,0),
(dexp)v
I
- I_ 1G31(0,1) = 1.
Xuu(u, v) _ (-vcosu, -v sin u, 0), and X_ (0,0,0). This equality establishes the property that we were looking for.
Repeating the process followed in Example 3.45, we see that Exercise 7.61. Revise Exercise (17) in the list of exercises in this chapter,
2 keeping in mind the above proof.
K(X(u,v)) _ +v2 and H(X(u,v)) = 0.
Exercise 7.62. If S is a closed surface in R3 with negative Gauss curvature
everywhere, show that, for each p e S, the map expp : TS -* S diminishes
The fact that the principal curvatures of a surface depend continuously the length of curves.
on its points will allow us to prove two of the first results of global geometry
of surfaces in this book. They are known as Liebmann's theorem (1899) and Exercise 7.63. Let P be a plane, S a surface, and f : P -* S a local
Jellett's theorem (1853), and both are characterizations of the sphere among isometry. Prove that
all the compact surfaces in terms of the behaviour of their curvatures. We If(x)-f(y)I 5; k'YI
will obtain them as consequences of a theorem due to Hilbert (1945), for
which we will provide a simpler proof than the usual ones. for each pair x, y E P.

Theorem 3.47 (Hilbert's theorem). Let S be an oriented surface and let
kl <k2 : S - IE8 be the corresponding principal curvature functions. If, at a
point p E S, the following three conditions hold simultaneously:
A: the Gauss curvature at p is positive,
B: kl has a local minimum at p,
C: k2 has a local maximum at p,
We are now ready to state and prove the result that classifies all flat
surfaces closed in R3. This result was shown for the first time in 1959, as
a part of a much wider theorem, by Hartman and Nirenberg and later, in
1961 and 1962, by Stoker and Massey, respectively. These are surprisingly
late dates, which can be understood only because the geometers of earlier
centuries studied only the local version of the problem that is, they did
not assume the surface was closed in Euclidean space.
7.7. Appendix: Some additional results of an intrinsic type 271 92 3. The Second Fundamental Form

In order to complete our study of the shape of the closed flat surfaces
in ][83, it will be useful to introduce a property of the exponential map that
distinguishes surfaces with vanishing Gauss curvature.
Proposition 7.60. If S is a surface closed in Euclidean space with zero
Gauss curvature everywhere, then the exponential map exp: TS -> S is a
local isometry at any point p e S.
Proof. Since S is closed in I[83, Corollary 7.41 says that if p e S, expp is de-
fined on the entire tangent plane TS. Moreover, from Gauss's Lemma 7.36,
in order to prove that expis a local isometry, it only remains to check that
then p is an umbilical point.
Proof. We take the origin of R3 as our point p E S and choose the coordi-
nate axes so that
el = (1,0,0), e2 = (0,1,0), and N(p) _ (0,0,1),
where N is a Gauss map of S and el, e2 E TS are principal directions at
the point p. In this affine reference frame, the affine tangent plane of S
at p = (0,0,0) is the coordinate plane z = 0. By Proposition 2.23, we can
parametrize S near p as a graph over this plane z = 0. To be exact, there is a
parametrization X of Ssuch that X (u, v) _ (u, v, f(u, v)), where f : U -> IE8

i_ wi
if v, w e TS and (v,w) = 0, then

We can also assume that w # Op and that v O. The latter is due to

is a differentiable function on the open subset U of Il82 with (0,0) E U and
X(0, 0) = p. By our choice of p, we have
f(0,0)=0, f(0,0)=0, and f(0,0)=0.
the fact that (dexp)oP is the identity map, as seen in (7.13). Then, let This can be written in terms of X as
a : ]l8 -> TS be a parametrization by the arc length of the circle centred at
Op E TS with radius lvi such that a(0) = v and a'(0) = w/iwl. Consider X(0, 0) _ (0,0,0), X(0, 0) _ (1,0,0) = el, and X(0, 0) _ (0, 1,0) = e2.
the differentiable map G : ]l82 -* S, given by On the other hand, using the expressions in (3.10), which provide the coef-
G(s,t) = expta(s). ficients of the matrix of the second fundamental form ax of S at any point
X covered by the parametrization, we have
Either by a direct computation or by looking at the proof of Gauss's Lemma
7.36, we see that the equalities ,fuu ,fuv Ivy

(G8,G)=0 and IGti2=ivi2

e=
1 +fu+fv 1i+f+fv
_ and
1 + fu + fv
9

are true on Il82. Taking partial derivatives relative to the two variables, we
Since the vectors Xu, Xat the point (0,0) E U are principal directions, we
have
obtain
(C88, C) + (C8, Cr8) = 0, (Ctt, C) = 0, f(0,0) _ ki(p), f(0,0) = 0, and f(0,0) = k2(p).

(G8, Gt) = 0.
Let a and j3 be the curves on the open subset of S covered by X given by
(Gst, Gt) + (Gs, Gtt) = 0,
Combining these expressions and denoting by the superscript T the compo- a(u) = X (u, 0) and ,3(v) = X (0, v),
nent tangent to the surface, we see that and let El and E2 be the curves in I[83 defined as
(7.26) C=0 0 and G tiles.
Xv(u,o) and El(v) _ Xu(o,v)I Xu(o,v).
Hence, since E2(u) - IXv(u,o)I
a IGsI = (C,C8) According to these definitions, we have that
at iGsi' El(v) E and E2(u) E T«(u)S
we have, differentiating again with respect to t,
and, thus, we may consider the following two real functions hl and h2 of one
_ ((C),C8)+(C,C8) _ (C,C8)2 variable, which can be computed, from (3.10), in terms of f
8t2 Gs 1C81 ICS'3
and, since, from (7.26), (C, C5)2 = tI2, we deduce that hi(v) _ aa(v)(Ei(v), E'i(v)) - C1 +f)/1+ fu + fv (0'v)'
((C), Gs) _ ((C) t - (C)5, Gs) (u'°).
ate -
a2 iGrsl _
Icsl3
-
cs'3 h2(u) _ a«(u)(EZ(u),EZ(u)) _
(1 +f)/i + fu + f2
3.6. Continuity of the curvatures 93 270 7. Intrinsic Geometry of Surfaces

Therefore, by hypothesis C, It is obvious that we cannot expect to generalize this result to each
point of a flat closed surface. Indeed, through each point of a plane we find
h2(0) _ cr(e2,e2) _ k2(p) ? k2(a(u)) ? a«(u)(E2(u),E2(u)) = h2(u) infinitely many straight lines entirely contained in the surface. But, in fact,
and, analogously, using B, we will see that this is the only example with this behaviour.
hi(0) _ op(ei,ei) = ki(p) ki(13(v)) < ap(v)(El(v),Ei(v)) = hl(v) Lemma 7.59. Let S be a flat connected surface closed as a subset of I[83
That is, the function hl has a local minimum at 0 and h2 has a local maxi- different from a plane. Then, through each point p e S there is a unique
mum at the same point. Thus, straight line Rthat is entirely contained in the surface.
(3.13) h2 (0) < 0 < hi(0). Proof. After Lemma 7.58, it only remains to show that the assertion is
Differentiating with respect to u directly in the expression defining h2, we also true for the points of int j C S. We know that int j is an orientable
arrive at surface whose Gauss and mean curvatures vanish, that is, with trivial second
fundamental form at each point. Thus, Exercise 3.23 assures us that each
h2(u) _ [_(1+f22fvfvu(1+f+ffvv of its connected components is contained in a plane. Then, let C be a
2 (1 + f2)-1(1 + fu + fy 2 (2ffuu + 2 fvfvv)fvv connected component of int J and P the plane in IR3 including it. Thus
+ C, being a surface, is a non-empty open subset of the plane P, different
and, hence, differentiating again at u = 0, we obtain that from the entire plane P. Indeed, if C = P, then P = C C int J C S and,
since S is connected, one would have that P = S, but this is not allowed
/4(0) = -f(0,0)f(0,0) + under our assumptions. Once we have established that C is an open subset
In an analogous way, we get of the plane P different from P itself, we can assume that the topological
boundary Bdry pC of C in P is non-empty. Choose a point q e Bdry C.
h(0) = -f(0, 0)f(0, 0) + 0). Then, there is a sequence {m}mEN of points of C C int J converging to q.
Thus, from (3.13), we obtain Since S is closed, q e S and, since J is closed in S, we have that q e J n P.
If q were an interior point of 2', it would be in a connected component C'
o)fvv(o, o)(fvv(o, o) - f(0,0)) < o. of int J n P different from C. But, since q e P and C' has to be contained
Substituting the values that we have computed from these derivatives into in some plane in R3, we deduce that C' C P. In conclusion, we would have
the above, we finally have two non-empty open disjoint subsets C and C' of a plane with q e C' n C,
K(p)(k2(p) - ki(p)) 0.
which is impossible. Hence

Since A asserts that K(p) > 0, it follows that p is umbilical. O gEBdrypC=geBdry,FfP.

Corollary 3.48 (The Jellett-Liebmann theorem). A compact connected sur-
face with positive Gauss curvature everywhere and constant mean curvature
is necessarily a sphere.
Proof. Let c E R be such that H - c. We now show that c is necessarily
a non-null constant. Indeed, if c = 0, then k1 = -k2 and K = -ki < 0,
which gives a contradiction to the hypotheses. This implies that the surface
S be orientable. In fact, this follows from Exercise 3.9. Having established
the orientability of the surface, we consider the principal curvature functions
k1 < k2 corresponding to a chosen orientation of S. By Proposition 3.43,
they are continuous on S. Since S is compact, there is a point a E S where
k1 attains its minimum. But k2 = 2H - k1 = 2c - k1, and so k2 attains its
maximum value at this same point, say a, on S. On the other hand, the
On the other hand, the affine tangent plane of the surface S at the point q is
P itself, since Tqn S = Tqn int J = Tqn P for each n e N. Using Lemma 7.58,
there is a unique line Rq such that q e Rq C S which, for this reason, is
tangent to the surface at q. This implies that Rq C P and, moreover, we
know that Rq C Bdry F. Thus, for each q e Bdry pC, there is a unique
line Rq such that q e Rq C Bdry ,'F fl P. All these Rq with q e Bdry pC
are parallel since they are disjoint and coplanar. Then each Rq is entirely
contained in Bdry pC, because, if p e C and r e Rq and if the segment [p, r[
is not contained in C, then there would be a point q' E [p, r[ (Bdry C.
From this, it would follow that C C P - Rq' and it would be impossible
that q e Bdry C. Consequently, C must be either a half-plane of P or a
strip between two parallel lines. In any case, there exists a unique line in S
through each point of C. U
7.7. Appendix: Some additional results of an intrinsic type 269
points of the surface and of the flat points to which the parabolic points
accumulate.
Lemma 7.58. Let S be a closed flat surface in R. Let P be the open subset
of S consisting of all its parabolic points and let.' be the closed subset formed
by its flat points. For each p e P, there exists a unique straight line Rp such
that p e Rp C S. Moreover
pEP==RpCP,
p e Bdry P = Bdry J' == Rp C Bdry P.
Proof. Note that P is an open subset of S since S is closed and, hence,
orientable. Thus, if H is its mean curvature for a given orientation, the set
P consists of the points where H 0. Consequently, J' is closed and we
have that
S=PUJ'=PUBdryJ'UintJ'
are disjoint unions and thus Bdry ,% ' = Bdry P.
We first suppose that p e P is a parabolic point of the surface. By
Lemma 7.57, there is a unique straight line Rp of R3 such that p e Rp, and
the connected component Cp of Rp n P containing p is an open interval of Rp
where 1/H is an affine function of the arc length. Let Dp be the connected
component of Rp n S containing p. Clearly, Dp is an interval in Rp including
C. If this inclusion is not an equality, Cp cannot be the entire line Rp and,
hence, Cp has an endpoint q that, since Cp is open, lies in Dp - Cp C 2'.
Therefore, there exists an endpoint q of the interval Cp where H(q) = 0. This
contradicts the fact that 1/H is an affine function of the arc length of Cp
and q is a finite point. It follows that Dp = Cp is the connected component
of Rp n S containing p, which is, as already mentioned, an open interval of
the line R. Since we are assuming the surface S to be closed in R3, the
intersection Rp n S is a closed subset of the line Rp, and the same is true
for each of its connected components. Then Cp is a closed subset of R. By
connectedness, it follows that Cp = Rp, that is, Rp C P C S, as we wished
to establish. The uniqueness of the line Rp among those passing through p
and that are contained in S was already a consequence of Lemma 7.57.
We now take a point p e Bdry P C 2', that is, a flat point that can be
described as the limit of a sequence {Pn}nEN of parabolic points. Consider
the sequence of lines Rpn C S as discussed above. A subsequence of this
sequence, which will be denoted in the same way, converges to a line R of IR3;
see Exercise 4.12. Then, since S is closed, we have p e R C S. It remains to
see that this line R belongs to Bdry P = J n P. Each point of _R is the limit
of a sequence of points in the lines Rpn C P and, hence, R C P. Moreover,
if there were a point q e R n P, then we would have p E R = Rq C P, which
is impossible. We conclude, then, that R C Bdry P. 0 that this surface is a sphere.
94 3. The Second Fundamental Form
Gauss curvature of S is positive everywhere. From Hilbert's Theorem 3.47,
the point a is umbilical. Therefore, for any other point p E S,
k2 (p) k2 (a) _ k1 (a) ki(p).
Consequently k1 (p) = k2(p), that is, S is a totally umbilical surface. We
complete the proof by applying Corollary 3.31.
Corollary 3.49 (The Hilbert-Liebmann theorem). The only compact con-
nected surfaces with constant Gauss curvature are spheres.
Proof. If c e R is the constant value taken by the Gauss curvature K of
the surface S, we see that c> 0 by applying Exercise 3.42. Thus the mean
curvature H of S, relative to any local orientation, cannot vanish anywhere,
as (3.6) shows. Similar reasoning to what we have used in the proof of
previous results assures us that the surface S must be orientable. We have
again two global principal curvature functions k1 and k2 on the surface and
we can repeat, step by step, the proof of the Jellett-Liebmann Corollary 3.48,
taking into account that the maximum of k2 has to be reached at the same
point where k1 reaches its minimum, since the product of both functions is
nothing more than K - c. E1
Exercise 3.50. T Show that a compact connected surface with positive
Gauss curvature everywhere and such that the quotient H/K is constant
has to be a sphere.
Exercise 3.51. Prove that a compact connected surface with positive Gauss
curvature everywhere and with one of its principal curvatures constant is
necessarily a sphere.
Exercises
(1) T Let S be a surface and let p e S be a point. Show that there are
infinitely many points in the normal line of S at p for which the corre-
sponding distance functions have a local minimum at p.
(2) T Let S be a compact surface contained in a closed ball of ]E83 with
radius r> 0. Prove that there exists at least one point p E S such that
K(p) > 1/r2 and H(p) > 1/r.
(3) T Suppose that a compact connected surface S has the following prop-
erty: for each po E Il83 - A, where A is a countable set, the square of the
distance function at the point po has at most two critical points. Show
 Exercises 95 268 7. Intrinsic Geometry of Surfaces

(4) T Let S be an orientable surface and let N : S -p S2 be a Gauss map.
The support function of S based on the point po E II83 is defined as:
f(p) _ (p - P0, N (p)), by E S.
Show that, if S is connected, then f has constant sign if and only if S
is star-shaped with respect to the point pp; see Exercise (11) at the end
of Chapter 2. Prove also that, if f is constant and there exists a point
p e S with non-zero Gauss curvature, then S is an open subset of a
sphere centred at po.
(5) If N : S - S2 is the Gauss map of an oriented surface, check that the
Jacobian of N and the Gauss curvature function coincide at each point.
On the unit sphere S2, we take the orientation corresponding to the
outer normal field. (See Exercise 3.17 and Remark 8.3.)
(6) Prove that the Gauss map of a compact oriented surface is a local dif-
feomorphism if and only if its Gauss curvature is positive everywhere.
(See Exercise (13) at the end of Chapter 2.)
(7) Let P C ]E83 be a plane passing through the origin and let a : I - P be
a regular curve that is a homeomorphism onto its image. If a E ]f83 is a
unit vector normal to P and X : I x 1[8 -p 1[83 is defined by
X(s,t) = a(s) +ta,
had q as an endpoint, then there would be a curve Q : [0, p) -* Rfl S, for
some p> 0, with ,6(0) = q and ,6'(0) = v 0. Then, differentiating N o ,Q
on the right at 0, we would have that vy(v, v) = 0 and, hence, v would be
a direction for Ry. This would imply that R= Rq and, since q is in the
interior of Ry fl S, it would also be in the interior of Rp fl S and we would
have a connected set greater than Cp in Rfl S. This is impossible and,
thus, Cp is an interval of Rp without endpoints, that is, an open interval.
To complete the proof, it only remains to study the mean curvature
function H on each straight line Rp, p e S. By uniqueness, if q e CC
Rfl S, one has that R9 = Rand so Cp = Cy. Therefore, each point
q e Cp can be covered by a parametrization X : (-b, S) x (-s, e) -* S of S
by the curvature lines with X(0, 0) = q and such that a(u) = X (u, 0) is a
parametrization of Cq = Cp in a neighbourhood of q. For this X, equation
(7.22) and all equalities below it are valid. Differentiating (7.22) with respect
to u and (7.23) with respect to v and taking into account that 2 H = k2, we
get
Huu Xv + 2Hu Xuv + H Xuuv =0,
X uuv = gv Xu + g X uv
Substituting, we see that
H2 Hu
2HUUXv - 4 Xv + 2HgvXu - HgXv = 0.

show that S = X (I x 1[8) is a surface with identically zero Gauss curva-

H
ture. Compute its mean curvature at each point. Deduce that, if C is a Therefore, extracting only the Xv-component, we have
simple curve in the plane P, the set (7.24) HHuu - 2H2 - 2gHu = 0,
S={p+taIpECandtEIIB} where the function g is determined by (7.23) and, thus,
is a surface with identically zero Gauss curvature everywhere. It is (7.2.ril yn - (IXI2)u
2IXuI2
IXulu
IXuI
usually called the right cylinder with directrix C. (See Subsection 7.7.4.)
(8) Use the methods of Section 3.6 to compute the Gauss and mean cur- On the other hand, if s is the arc length function along C, relative to any
vatures of the torus of revolution described in Example 2.17. Find its choice of origin, we know that
elliptic, hyperbolic, and parabolic points.
(9) T Prove that the only surfaces of revolution with identically zero mean
du-la'I-IXul.

curvature that are closed as subsets of II83 are catenoids, i.e., surfaces of Substituting this expression in (7.24) and in (7.25), we see that
revolution generated by a catenary, and planes. d'l
d2 1
=0.
(10) Study the surfaces of revolution with constant Gauss curvature. ds2 H
(11) T Determine the umbilical points of the ellipsoid of the equation Therefore, the inverse function of the mean curvature H is affine with respect
to the arc length along Cp C Rp fl S. O
x2 y2 z2
2 + b2 + c2 li
a When the flat surface S of Lemma 7.57 is assumed to be closed, we can
where 0 < a < b < c. obtain stronger results. In fact, we will get good control on the parabolic
7.7. Appendix: Some additional results of an intrinsic type 267 96 3. The Second Fundamental Form

satisfies the requirements of the lemma. In fact, let X : (-b, b) x (-E, E) -* S (12) T Let X be a parametrization of a surface S such that F = (X,
be a parametrization of S by the curvature lines, whose existence is assured 0. This called an orthogonal parametrization. Prove that the Gauss
by Example 7.54, with X(0, 0) = p. Then: curvature K of S satisfies
NX = 0, NX = -k2 Xv, and (X,X)=0. KoX= - 2\/EG [ ( EG)
Differentiating with respect to v in the first equality and with respect to u + ( EG) u]
in the second one, we get (13) Let S be an orientable connected surface with constant principal curva-
tures. If S has an elliptic point, then it is an open subset of a sphere.
(7.22) (k2)X+k2xuv =0.
After scalar multiplication by Xu, we obtain
(14) T Consider a surface S and a dilatation b : Il83 -Il83. Compare the
second fundamental forms and the curvatures of S and of the image
Xuv Xu) = 0. surface S' _ (S).
Hence (15) T Repeat the previous exercise assuming that b is no longer a Euclidean
(X, Xv) = (Xv, Xv)u - (Xv, Xuv) = 0. dilatation but is the inversion b : ]E83 - {0} - ]E83 - {0}, given by
Furthermore, again using that NX = 0, we have pER3-{0}.
\Xuu, NX / _ \Xu, NX }u - \Xu, Nu = 0.
(16) Let S be an orientable surface and let N : S -f S2 be a Gauss map.
Consequently, the vector Xuu is, at each point of the domain of X, propor- Define F,.: S - ll83 by
tional to X. That is,
Fr (p) =p+rN(p), by e S,
(7.23) Xuu = gXu,
where g : (-6, 6) x (-E, E) -* IR is some differentiable function.
where r > 0. Suppose that Sr = Fr(S) is a surface and that Fr : S - Sr
is a difFeomorphism (see Remark 4.29). Show the following.
We now define a differentiable curve a : (-b, b) -* S on the surface by If p e S and el, e2 E TPS are principal directions, then (dFr)p(ej) _
a(u) = X (u, 0), Vu e (-b, b). (1-rki(p))ez, 1-rkz(p) 0, i = 1, 2, and 1 -2rH(p) +r2K(p) 0.
TpS = for each p e S and, hence, there is a Gauss map N'
This curve is regular and clearly satisfies
for Sr such that N' o F,. = N.
a(0) = p, a' (u) = Xu (u, 0), and a" (0) = X( u, 0). The-affine-normal line of S at p e S coincides with the-affine-
Then, from (7.23), we have that the acceleration a" of a is proportional to normal line of Sr at F,.(p).
the velocity a', at each point. Thus, the trace of a((-6, b)) is contained in . If el, e2 E TpS are principal directions at p e S with principal
a line. Indeed, in the straight line passing through a(0) = p with direction curvatures k1 (p) and k2(p), respectively, then el and e2 are also
a'(0) = Xu(0, 0)IIe1 E TS, we have principal directions of Sr at Fr (p) with principal directions
a((-6,6)) C RpfS. ki(F'T(p)) _ i = 1,2.
1 - rkz(p)'
Since a((-b, b)) is a neighbourhood of p in Rp, the same is true for Rp n S,
as required. . Prove that the Gauss curvature K' and the mean curvature H' of
Turning to uniqueness, if R were another line in R3 such that p e R S,. are given by
and R n S were a neighbourhood of p in R, then each vector v 0 in the K' 0F
Fr =
K
and H' o Fr =
H-rK
direction of R would satisfy op(v, v) = 0. But the only isotropic direction of 1 - 2rH + r2K 1 - 2rH + r2K'
the quadratic form p on TS is the principal direction e1. Hence, v and el If S has constant mean curvature H - 1/2r, show that K(p) 0
should be proportional, and thus R = R. for all p e S and that Sr has constant Gauss curvature K' - 1/r2.
Let Cp be the connected component of Rp n S containing the point p. (17) T Comparison of two surfaces at a point. Let Sl and S2 be two orientable
Since Cp is a connected subset of Rp containing p in its interior, we know surfaces that are tangent at a common point p e Si fl S2i and let Nl
that Cp is an interval of the line Rp with p in its interior. If this interval and N2 be the corresponding Gauss maps such that N1 (p) = NZ(p). We
 Hints for solving the exercises 97
express Si and 82, near the point p, as two graphs of functions fi and
12 defined on the common tangent plane. We say that Si is over S2 at p
if there exists a neighbourhood of the origin of this plane where fi > 12.
If Si is over S2 at p, show that vl (v, v) > v2 (v, v) for each v E
TPSI = In particular, Hl(p) > HZ(p).
If vl (v, v) > v2 (v, v). for each v 0, prove that Si is over S2 at p.
(Compare this with Exercise (16) at the end of Chapter 1.)
(18) T Let S be a graph over a disc in 1[82 with radius r> 0. If S has mean
curvature H > a at each point for some real number a, show that ar < 1.
(19) Prove that if an orientable surface has constant principal curvatures,
then either each of its points is umbilical or the Gauss curvature is non-
positive everywhere.
Hints for solving
the exercises
Exercise 3.9: It is enough to define G : S -f S2 C R3, mapping each point
p e S to the only unit vector G(p) normal to TS obeying P. This map is
well-defined by the first hypothesis about P and is differentiable because, by
the second hypothesis, it coincides locally with a differentiable vector field.
Exercise 3.10: Let A be the set of c e [a, b] such that, on [a, c], one can
find a continuous map N taking values in R3 with N(a) = n, I= 1,
and N( t) 1 Ta(t) S for all t e [a, c] . Since a E A, A is a non-empty set.
Furthermore, it is an open subset of [a, b] . In fact, if c e A, denote by V an
open neighbourhood of a (c) in S where a Gauss map N : V - 2 is defined.
This can be done because of Lemma 3.2. By the continuity of a, if c < b,
there exists > 0 such that a ((c - 26, c + 26)) C V. Moreover, changing N
to its opposite, if necessary, we may assume that N (a (c)) = N( c). . Define
N+6: [a, c + ] -f R3 by
J ()if t [a,c],
ift E [c,c+6],
l (Noa)(t)
which is continuous because of the choice of N. Thus [a, c + e] C A and,
thus, A is an open subset. We now show that it is also closed. To do this, let
{be a sequence in A converging to a number c e [a, b]. We take an
open neighbourhood V of a(c) as in the previous case. We know that there
is a k e I`N such that, if n > k, then a([cn, c]) C V. Changing the sign of the
Gauss map N on V, if necessary, we suppose that Nck (ck) = N(a(ck)). We
266 7. Intrinsic Geometry of Surfaces
Figure 7.6. Flat surface
open piece of paper folded up smoothly creates a surface with vanishing
Gauss curvature everywhere; that is, according to the notation that we in-
troduced in Remark 3.28, it only has parabolic or flat points. We will refer
to this situation by saying that the surface is flat. If the piece of paper,
which we use to build a flat surface by folding up, is not the whole of the
infinite sheet, which is always the realistic case, then it is not true that
the resulting surface has to be an open subset of a right cylinder. One can
smoothly deform a piece of paper in a way such as is shown in Figure 7.6,
without obtaining a piece of a cylinder. This is why it is surprising that, if
we add the hypothesis that the surface is closed or, equivalently, that the
piece of paper is infinite, we necessarily get a right cylinder.
In spite of what we have said above, it is always true that, even locally,
any flat surface contains plenty of segments of straight lines. This is the
subject of the following lemma, at least when the surface has no flat points,
but all of whose points are parabolic.
Lemma 7.57. Let S be an oriented flat surface whose mean curvature H
is positive everywhere. Then, for each point p E S, there exists a unique
straight line Rp of 11 3 such that p E R1, and Rp fl S is a neighbourhood of
p in Rp. Moreover, the connected component of R1, f1 S containing p is an
open interval of Rp on which 11H is an affine function of the arc length.
Proof. With respect to the orientation we have chosen, we have that H > 0
everywhere. Since K = 0, the principal curvatures k1 and k2 of S must
satisfy
0=k1 <k2.
Hence, there are no umbilical points on S and, thus, the principal curva-
tures are differentiable functions everywhere on S and we have, at each point,
exactly two orthogonal principal directions corresponding to the two curva-
tures. Given a point p E S, let Rp be the line passing through p and having
as its direction the principal direction el E TBS. We now show that this line
7.7. Appendix: Some additional results of an intrinsic type 265
where we have used (7.20) and the first equation of (7.21). Furthermore,
the second equations of both (7.21) and (7.18) allow us to infer that
Cv =Xv-(Xv,a)a+
1
NVX =0.
Thus, the map G is also constant on the open set U. That is,
X - (X,a)a+ NX = c,
for some c e ]183. If we perform scalar multiplication by the vector a on this
last equation and take the third equality of (7.21) into account, we see that
(a, c) = 0. In short, we have found a unit vector a and another vector c of
I[83 perpendicular to the first one such that
X - (X, a)a - c = _Nx.
2 Then, by uniqueness, N o a = Na. Thus
Taking lengths in this equality, we finally arrive at
lx - c2 - (X - c,a)2 =
In other words, the distance from the points of X (U) C S to the line passing
through c in the direction of a is constant and equal to 1/k2. This means that
the open subset X (U) of S is contained in a right circular cylinder of radius
1/k2 whose axis is the straight line that we referred to before. Thus, we have
proved that each point of the surface has an open neighbourhood included
in a right circular cylinder of radius 1/k2. Since the surface is connected,
this cylinder is the same for all its points and the proof is complete.
If we recall now that any surface that is closed as a subset of ]l83 is ori-
entable, as seen in Exercise (6) of Chapter 4, the following characterization
follows immediately from the classification Theorem 7.55 that we have just
proved.
Corollary 7.56. The only connected surfaces closed as subsets of Il83 having
constant principal curvatures are planes, spheres, and right circular cylin-
ders.
7.7.4. Flat surfaces. In this subsection, we will undertake the main sub-
ject of this Appendix: the study of closed surfaces in R3 with vanishing
Gauss curvature everywhere. Up to now, all the examples of this type of
surface that we have encountered have been the so-called right cylinders over
plane curves see Exercise (7) of Chapter 3. In fact, the result that we study
next will confirm that they are the unique examples. These right cylinders
over simple plane curves can be viewed as the physical result of folding up
a piece of paper with infinite extension winding up infinitely many times
if the curve is compact. Indeed, by Gauss's Theorema Egregium 7.10, any
98 3. The Second Fundamental Form
then define N: [a, c] -f II83 by
Nc(t -
I (t)if t
(Noa)(t) if
[ae,cj],which
is continuous by the choice of N on V. Hence c e A and A is a
closed subset of [a, b]. Since [a, b] is connected, we conclude that A = [a, b]
and, thus, there exists a continuous map N : [a, b] -f 1[83 with the required
properties. If there were another N' with the same properties, then (N, N')
[a, b] -f 1[8 would be continuous and would take only the values 1 or -1. Since
this product takes the value 1 at a, we infer that N' = N.
Suppose now that the curve a is closed, i.e., that a(a) = a(b) and
suppose also that S is orientable. Take a Gauss map N on S with N(a(a)) _
n and let Na be the normal field along a with initial condition N« (a) = n.
Na(b) =N(a(b)) = N(a(a)) = n.
To show the converse, we may assume that S is connected and, hence, arc-
wise connected, by working on one of its components. Choose po E S and a
unit n e 1[83 normal to S at po. Define N : S -f S2 as
N(p) = N«(1),
where a : [0, 1] -f S is a continuous curve such that a(0) = po and a(1) = p
and where Na is the normal field along a with initial condition n. This
definition does not depend on the choice of the curve a. Indeed, if ,0
[0,1] - S is another curve of this type and Nis the corresponding field,
and if N(1) _ -Na(1), then N.y = (-N) -1 * Na would be the continuous
normal field along ry = ,0-1 * a, which is a closed curve with 'y(O) = po. But
Nry(0) = N«(0) = n and Nry(1) _ -(N)-1(1) _ -N(0) _ -n,
which is impossible by our hypotheses. It remains to see that this map N is
differentiable. Suppose that V is an open connected subset of S where we
have defined a Gauss map G (see Lemma 3.2). Then G o a is a continuous
normal field along a, for each continuous curve a whose trace is contained
in V. The reasoning we used to show that Nwas awell-defined map shows
now that either G = N,v or G = -N,v, and we are done.
Exercise 3.13: Both surfaces Sl and S2 are open subsets of S by Propo-
sition 2.77. Let Nl and N2 be Gauss maps for Sl and S2, respectively. By
Lemma 3.3, Nl and N2 either coincide or are opposite on Sl n S2. Suppose
the first, reversing the sign of some of them if necessary. Then it suffices to
define N : S - 2 by N,sl = Nl and N,s2 = N2.
Hints for solving the exercises 99 264 7. Intrinsic Geometry of Surfaces

Exercise 3.14: Let Nl and N2 be Gauss maps for Si and S2, respectively, recall that K is the product of the principal curvatures k1 and k2. Hence,
and suppose that S is orientable. Choose a Gauss map N for S. Applying re-orienting the surface if necessary, we can assume that
Lemma 3.3, we know that N,sl and Nl are either equal or opposite. Change k1=0<k2.
N to -N if necessary so that N,sl = N1. Then Nl sl ns2 = N1s1ns2 . Since
S2 is also connected, using Lemma 3.3 again, we have that N,s2 and N2 From this, the first equality in (7.18) becomes
either coincide or are opposite. Hence
(7.20) Nu = 0
Nl I s1 ns2 = NI s1 ns2 = N2 I s1 ns2 or Nl I s1 ns2 = NI s1 ns2 = - N2 I s1 ns2
on the open set U.We now see that all the computations that we made earlier
which is impossible since Sl n S2 = Au B, N1 IA = N2 IA, and N1 1B = -N2 1B . lead to a proof that the tangent field is constant on U. In fact,
To see that this is the case for the Mobius strip, we use the notation (7.20) implies that
of Section 3.2 where it was described. We set Si = F([0, 7r/2) x (-1, 1) U
/XU =0.
(37r/2, 27r] x (-1, 1)) and S2 = F((7r/4, 77r/4) x (-1,1)). Both of the two
open subsets are orientable because they are covered by a parametrization. KIIXuII,.'NXl - IXuI'Nu)

The remaining properties can be easily checked. Moreover, from the third equality of (7.18) and equality (7.19), it follows
that
Exercise 3.15: Nl is well-defined in this way as a consequence of the fact
/(\xu Xu
,Xv - IXU
,Xvu =0.
that, if p e Si and {a, b}, {c, d} are two bases of T,S1 and M is the matrix (XUI lu \IXuI /
Finally
of the change of basis, then
det((df)p(c), (df)(d), NZ(f (p))) = det M det((df )(a), (df)(b), N2(f (p))).
fX\ X\ =0,
=IxUIK),)
\
We next show that Nl is differentiable. Let X : U - S be a parametrization since Xu/ Xu is a unit field. Since the {X, Xv, NX } are a basis of R3 at
on a connected open subset of 1[82. Then since: (1) {X, is a basis of each point of U, the above three equations are equivalent to
the corresponding tangent plane at each point of U; (2) we have
(df)x(Xu) _ (f o X)u and (df)x(Xv) _ (f 0 X); = 0 on the open set U.
IIXuII,.
and (3)(df)x is an isomorphism, we deduce that But, on the other hand, we deduce from (7.19) that
det((df)x(Xu), (df)x(Xv), Na o f o X) (XU
Xul
= Xuv
XU
- (Xuv,
X
XU)
X=
u O.
is anon-vanishing continuous function on U. We change, if necessary, the v I

parametrization X to Y(u, v) = X (v, u) so that this sign can always be Therefore, since the open set U is connected, there exists a vector a E S2
taken to be positive. Thus, according to the definition of Nl, we have such that
XuAXv Xu
N1oX= (7.21) = a and, thus, (XV, a) = 0 and (N, a) =0.
IXu A Xvl
and hence Nl is differentiable. Now consider the differentiable map G : U -* R3, given by

Exercise 3.23: Definitions (3.4) and (3.5) say that, under our hypotheses, G=X- (X,a)a+ Nx
the second fundamental form of the surface corresponding to any Gauss
The partial derivative of G with respect to u is
map N is identically zero. Thus, the differential of N vanishes at any point
of the surface. We conclude the exercise by applying Proposition 2.63 and G =X - X as 1 NX=
Exercise 2.70. 2
7.7. Appendix: Some additional results of an intrinsic type 263
Theorem 7.55 (Classification of surfaces with parallel second fundamen-
tal form). An orientable connected surface whose principal curvatures are
constant, or, equivalently, whose Gauss and mean curvatures are constant,
is necessarily an open subset of a plane, of a sphere, or of a right circular
cylinder.
Proof. Let k1 and k2 be the two principal curvatures, assumed to be con-
stant, of a connected surface S oriented by means of a Gauss map N. If
k1 = k2, then the surface is totally umbilical and, by Theorem 3.30, it must
be an open subset of a plane or of a sphere. Thus, we may assume that
the two constants k1 and k2 are different. Hence, by Example 7.54, for each
point of S, there exists a parametrization by curvature lines whose image
covers this point. Let X : U -+ S be a parametrization of this type defined
on a connected open subset U of R2. We have
(dN)x(X) = (N o X )u = Nu = -k1Xu,
(7.18) (dN)x(X) = (N o X )v = Nv = -k2Xv,
(X,X) = 0,
as equalities between functions defined on U. Differentiating the first equal-
ity of (7.18) with respect to v and the second one with respect to u, we
obtain
N v = -k1Xuv and N u = -k2Xvu.
By elementary calculus, the two mixed partial derivatives are equal and,
since we are supposing that the principal curvatures are different, we have
(7.19) Xuv = 0
on U. We now take partial derivatives relative to v in the third equality of
(7.18) and use the equation above. Thus, we have
(X,X) = 0
on U. Therefore, if the superscript T indicates, as usual, the component
tangent to the surface, we can write
(X)T = h Xv,
for some differentiable function h : U -* R. Thus
(CX v)u - (X), Xu) = (hXv + hXuv, Xu) = 0,
and, taking Lemma 7.11 into account, we conclude that
KoX O.
Consequently, under our hypotheses, the surface S, provided that it is not
totally umbilical, is flat; that is, it has vanishing Gauss curvature. Let us
100 3. The Second Fundamental Form
Exercise 3.24: The quadric S is the inverse image of 0 of the function f
defined on the whole of 1[83 by
f(p) _ (Bp,p) + 2(b,p) + c, p E 1[83.
It follows from this that
(Vf)=Bp+b, `dp E 1[83,
which does not vanish on S since S is regular; see Example 2.16. It only
remains to apply Example 3.8 together with the definition of the second
fundamental form and to remember that S is an ellipsoid if and only if the
matrix B is definite.
Exercise 3.32: The tangent plane of S along that curve is constant and so
any local Gauss map is constant when restricted to the points of the curve.
Therefore, the tangent vector at each point of the curve belongs to the kernel
of the differential of the Gauss map and, thus, this map has an identically
null determinant.
Exercise 3.33: Let I be an open interval of ]f8 and a : I -f S a segment
of a straight line p.b.a.l. contained in an open subset of S where we have
defined a Gauss map N. Since ((N o a)(s), as(s)) = 0 for each s e I, we
have
a'(s)) _ -((dN)a(s)(a'(s)), as(s)) _ ((Noa)(s),cV'(s)) = 0,
since the acceleration of the line vanishes. Then, the quadratic form v«(S)
is not definite for any s E I.
Exercise 3.40: Choose any plane. By compactness, there is a global ex-
tremum on the surface for the height function relative to this plane. Apply-
ing Propositions 3.35 and 3.37, the second fundamental form of the surface
at this point is positive semi-definite and, hence, its determinant is non-
negative.
Exercise 3.41: A point p e S is elliptic when its Gauss curvature is posi-
tive, that is, when the second fundamental form at this point QP is definite.
Choosing a suitable local orientation, we assert that p is elliptic if and only
if there is a local Gauss map N relative to which the second fundamen-
tal form is positive definite or, equivalently, there exists C E Il8+ such that
vp(v, v) > C1v12 for each v e TpS. But the latter, according to Example 3.39
and Proposition 3.35, is equivalent to the fact that p is a local maximum of
the square of the distance function to the point po = p + (1/C)N(p).
Hints for solving the exercises
Exercise 3.42: We choose p0 E R3 arbitrarily and, by compactness, we get
a global maximum on S for the square of the distance function to this point.
Applying Exercise 3.41, we see this point is elliptic.
Exercise 3.50: Since K > 0 we see from (3.6) that H2 > 0. Using Ex-
ercise 3.9, we see that S is orientable and, for a given orientation, there
is some c e R such that H = cK. Then k1 + k2 = 2ck1k2 and, thus,
11(1 - 2ck2) = -k2; since k2 > k1 > 0, we obtain
k2
k1 =
2ck2 - 1-
That is, k1 is a decreasing function of k2 and, hence, k1 will attain its
minimum at the points where k2 attains its maximum. We conclude the
exercise by appealing to Hilbert's Theorem 3.47 and Corollaries 3.48 and
3.49.
Exercise (1): Given p e S, we have
Iap(v,v)I 1k2(p)I}1v12
for all v e TpS. If we represent the above maximum by C E 1[8, then, for
A E II8 sufficiently small, we have that 1 - AC > 0. On the other hand,
reexamining Example 3.39, we know that p is a critical point of the function
fA that measures the square of the distance to the point p + AN(p),
A E II8, of its normal line. For these A we have
2[1v12 - IAIC],
and thus, by Proposition 3.35, p is a local minimum.
Exercise (2): Let p e S be a point where the square of the distance to a,
the centre of the ball of radius r in which S is contained, attains a maximum.
Then, according to Proposition 3.35 and Example 3.39, a is on the normal
line of S at p and
1 < (a - p, N(p)) ki(p), i = 1, 2.
Keeping in mind that (a - p, N(p))2 = a - r2, the exercise follows by
multiplying or adding the two inequalities.
Exercise (3): Let p e S be any point. This point is critical for fA, the
square of the distance function to as = p + AN(p), for all A E R. Since
S is compact, there exist global extrema for this function. By hypothesis,
p must be one of these points for all A E ]E8, except for maybe a countable
number of them. Therefore (d2fA) is asemi-definite form, according to
101 262 7. Intrinsic Geometry of Surfaces
that is, that Xu(q) is in the principal direction corresponding to kl, then
since X (q) is in the perpendicular direction and kl (p) L k2 (p), we would
obtain
(dN)p(X(q)) L -k1(p)Xv(4),
and thus, relabeling the u and v variables, if necessary, we can assume that
(dN)p(X(q)) L -ki(p)Xu(q),
and, hence, that
L -k2(p)Xv(9')
Therefore, on an open set smaller than the open set U where the parametriza-
tion X is defined, we can suppose, by continuity, that
W1 = NX u +(1£1 o X)Xu = -AXXu + (k1 o X)Xu 0,
Wa = NX v + (k2 o X )Xv = -AXXv + (k2oX)Xv LO
hold on U. Since the endomorphism A_ -(dN)satisfies
(A - 0
at each point, we have that
AXWI = (k2 o X )Wl and AXWZ = (kl o X )W2,
that is, the differentiable fields Vl = Wl o X-1 and VZ = W2 o X-1 defined
on the open set X (U) give at each point a pair of non-trivial vectors in
the principal directions corresponding to k2 and kl, respectively. Therefore,
they are also perpendicular. Applying Theorem 7.52 to this pair of fields,
we obtain the following result.
If p e S is anon-umbilical point of a surface, there exists
a parametrization X : U -* S of the surface S such that
p e X(U) and the vectors {Xu(q), at each point q
of U, are principal directions at X (q) E S.
We will call this type of parametrization a parametrization by the curvature
lines, since, in general, a regular curve on a surface is called a curvature
line when its tangent vector is an eigenvector of the differential of the Gauss
map at each time.
A first application of the existence of these classes of special paxametriza-
tions is the following local classification result, which works for those ori-
entable surfaces of ][83 whose-Gauss, mean, principal-curvatures are con-
stant. They are usually called surfaces with parallel second fundamental
form.
7.7. Appendix: Some additional results of an intrinsic type 261
and f2 be the corresponding functions. We next define a differentiable map
G : W --* on the open subset W = W1 n W2 of S containing p, by
G(9') _ (f2(q),fl(q)), bq E W C S.
Then, ker (dG)q = ker (df2)q fl ker (dfi)q. Since Vi and V2 are linearly inde-
pendent at the point p, we have that (dG): TS -p II82 is an isomorphism.
By the inverse function theorem, we can find an open neighbourhood V of p
in S and an open subset U of I[82 such that G : V -p U is a diffeomorphism.
The required parametrization is nothing more than X = G-1 : U -p V C S,
since, with this definition, we get
f2(X(u, v)) = u and fi(X(u,v)) = v, b(u, v) E U.
Taking partial derivatives with respect to v and u, we have
(df2)x(X)=O and (df1)x(X)=O,
which completes the proof since the kernels of (df2)q and of (dfi)q at each
point of W are spanned by V2(q) and V1 (q), respectively, as seen in Propo-
sition 7.51.
Example 7.53 (Orthogonal parametrizations). Let Y U -* S be any
parametrization of the surface S. Then, the partial derivatives Yu and Y at
each point of U give us a basis of the corresponding tangent plane. Thus,
the maps Vl, VZ : X (U) -> II83, given by
Vi = Yu o Y-1 and V2 = [-IYuI2Y + (Yu, o Y-1,
are clearly differentiable and determine tangent fields on the open subset
Y(U) of S that are non-null and orthogonal at each point. Applying Theo-
rem 7.52, there exists a parametrization covering an open subset of S inside
Y(U) whose parametric lines have the same traces as the integral curves of
Vi and V2. In conclusion, we have the following result.
Given a point p of a surface S, there exists a parametriza-
tion X U -> S such that p e X (U) and the function
(X, Xv) vanishes identically on U.
This type of parametrization will be called an orthogonal parametrization.
Example 7.54 (Parametrizations by curvature lines). Let S be an oriented
surface with no umbilical points; see Remark 3.29. By Proposition 3.43, the
principal curvatures kl and k2 of S are differentiable functions. Let p E S,
and let X U -* S be an orthogonal parametrization of S defined on an
open subset U of ll82 such that p = X (q) for some q e U. If we had that
(dN)(Xu(q)) = -ki(p)X(q),
102 3. The Second Fundamental Form
Proposition 3.35, for all these A. Using Example 3.39, we have that 1-ski,
i = 1, 2, have the same sign. That is,
0 < (1 - Akl(p))(1 - Ak2(p)) = 1 - 2AH(p) + AZK(p),
which is valid for all A E II8 except for maybe a countable subset. Then, the
discriminant of this second-degree polynomial has to be non-positive. There-
fore H(p)2 - K(p) < 0 and, from (3.6), the point p is umbilical. Since this
is true for each p E S, one finishes the exercise by applying Corollary 3.31.
Exercise (4): The first part is clear if we recall the definition of star-shaped;
see Exercise (11) at the end of Chapter 2. If the support function based on
po is constant and equal to c e II8, its differential, given by
(df)(v) _ (p - po, (dN)(v)), Vv E TS,
vanishes at each p E S. Thus
kZ(p)(p - poi ei) _ 0, z = 1, 2,
for each p e S where el, e2 E TS are principal directions. Consequently
K(p) = 0 or p - pp = cN(p),
for each p e S. As a consequence, the points of the open set where the
Gauss curvature does not vanish are in a sphere with centre po and radius
I c > 0. Since K is continuous, S is connected, and there are points where
K 0 and K can only take the values 0 and 1/c2, one has that there are
no points with null Gauss curvature on S, which concludes the exercise.
Exercise (9): We know from Exercise (3) at the end of Chapter 2 that
X (u, v) = x(u) a + y(u) cos v b +y(u) sin v c, (u,v) E I x J,
is a parametrization of the surface, where J is any interval of Il8 of length
2ir, the vectors {a, b, c} form a positively oriented orthonormal basis, and
a : I -f ][83, given by
a(u) =x(u)b+y(u)c, y >0,
is a local parametrization of the generating curve of S. From these parame-
trizations we may compute the coefficients of the first and second fundamen-
tal forms and the mean curvature H at the points of S covered by them, as
in (3.12). We obtain in this way that the mean curvature of S is identically
zero if and only if the components of a satisfy
(dx d2y dy d2x dx (dx 2 (dy 2
y
du due du due) du du) + du) = 0.
We now work on the open subset of I where dx/du 0. On it, there exists
the inverse function of x and we can reparametrize a so that y is a function
Hints for solving the exercises 103 260 7. Intrinsic Geometry of Surfaces

of x or, equivalently, so that a is a graph. Dividing the above equality by Now differentiating with respect to t at an arbitrary time, we have
(dx/du)3, it follows that
G(t's) t) = V(G(t, s)).
(*) -
yy" - 1 (y')2 = 0,
where ' represents differentiation with respect to the new variable x. From In particular, since V (p) and w are linearly independent, the map (dG)(o,o)
this equality it follows, first, that y" > 0 and, second, taking derivatives has a trivial kernel. Thus, in this situation, we can apply the inverse function
again, that yy"' - y'y" = 0, or, equivalently, that (log y)' = (log y")'. We theorem and find an open neighbourhood U of the origin in ll82 and an open
deduce from this that neighbourhood W C E of p in S such that G: U -p W is a diffeomorphism.
y"=Ky, K>0. We define the required function f : W -p II8 by f = pl o G-1, where pl is
All the solutions to this second-order linear equation are of the form the first component of the projection of I[82. Then
y = A cosh V"kx + a sink \/kx, A,1 e R. (df)G(s,t) o (dC)(5,t) = (dpl)(5,t) = P1,
But, if we impose the condition that such a solution obeys (*), we have for each (s, t) E U. Thus (df)G(s,t) L 0 at each point of U; that is, f has
-
KA2 Kµ2 = 1, and so no critical points in G(U) = W. Now let a : I -* W C S be an integral
curve of the field V defined on an open interval I of the real line. If we put
y 1 cosh(Kx " L), KER, LEI ,
g - p2 o G-1, then we can write
which is a catenary. Since the generating curve must be connected, either a(u) = G(f (u),g(u)), Vu E I,
it coincides entirely with a catenary or we have that dx/du vanishes every-
where. In this case, the generating curve is an open half-line perpendicular where 1(u) and g(u) denote f(a(te)) and g(a(u)), respectively. Taking
to the rotation axis and the surface is a plane. derivatives with respect to u we obtain
a'(u) = f'(u)G5(f(u),g(u)) + 9'(u)Gt(f (u),9(u))
Exercise (11): The gradient of the quadratic form defining the ellipsoid S
supplies us with anon-null normal vector at each point. That is, there is a Using the facts that a is an integral curve of V and that
non-vanishing differentiable function h defined on the ellipsoid and a Gauss Ct(f(u),g(u)) = V(C(f(u),g(u))),
map N such that
x y z we see that
h(x, y, z)N(x, y, z) = a2' b2' c2 '
V(G(f(u),9(u))) = f'(u)G5(f(u),g(u))+9 (u)V(f(u),9(u)),
for all p = (x, y, z) E S. If v = (A, B, C) E TPS, one has that
for all u e I. Since Gs and Gt are independent at each point of U, we
(dh)(v)N(p) + h(p)(dN)P(v) _ (a ' b ' c) have that f'(u) _ (f o a)'(u) vanishes everywhere on I. Therefore, f o a
is constant, as we wanted. The final observation in the statement follows
The point p is umbilical if and only if the linear map (dN)p is a multiple of immediately.
the identity map. This happens if and only if after scalar multiplication by
N(p)nv the member of the left-hand side vanishes. That is, p = (x, y, z) E S Theorem 7.52. Let Vl and V2 be two tangent fields on a surface S. Suppose
is umbilical if and only if that, at some point p e S, the values taken by the two fields V1 (p) and V2(p)
A B C are linearly independent. Then, one can find a parametrization X : U -+ S
a a a of the surface, defined on an open subset U of ][82, such that p e X (U) and
B =0, if 2 A + b2 B + C = 0. such that the vectors Xu and Xv are proportional to Vl o X and V2 o X; that
A
x
a2
y
b2
C
z
c2
a is, the parametric lines of X are reparametrizations of integral curves of Vl
and V2, respectively.
The determinant on the left-hand side is a quadratic form in the A, B, C
variables that vanishes on a plane. This forces it to be degenerate, and Proof. Let W1 and W2 be the two open subsets of S provided by Propo-
thus, the determinant of the corresponding symmetric matrix must vanish. sition 7.51 applied to each of the fields V1 and V2, respectively, and let fi
7.7. Appendix: Some additional results of an intrinsic type 259 104 3. The Second Fundamental Form

Therefore, under our hypotheses, e(p) _ +oo for each p E S. In other This implies that x = 0 or y = 0 or z = 0. Examining each possibility, we
words, the following is true. see that the only valid choice is y = 0. From this, we deduce that there are
The integral curves of a bounded tangent field on a closed four umbilical points determined by the equations
surface are defined on all of R. In particular, this occurs
for any tangent field if S is a compact surface.
y = 0, C2 (C2 - b2)x2 = a2 (b2 - a2)z2.
As a consequence, in any of these situations, the flow F of the field will be
defined on the entire product S x R. Exercise (12): We will use the notation established in Section 3.6. From
equalities (3.10) and (3.4), we have, since F = 0, that
7.7.3. Special parametrizations. Even though the very definition of a
surface required the notion of a parametrization, we have pointed out through- (KoX)EG = e9-.f2 = (X, NX)(Xvv, NX) -
out this text that the results that we are interested in do not depend
IXuv12
on these parametrizations. Until now, almost every time we used such Xv> - + IXu I2,
parametrizations, the choice of the parametrization did not matter. How- where T represents the part tangent to the surface. Now, we must express all
ever, we have also seen that, to solve some problems, it was convenient to the summands on the right-hand side in terms of the non-trivial coefficients
utilize parametrizations with special properties see, for example, Hilbert's E and G and of their derivatives. In fact, since F = 0, we have
theorem (Theorem 3.47) and the Herglotz integral formula in Proposition 7.15.
We will now show that, under some assumptions, there are parametrizations (Xuu Xvv - IXI2 = (Xuu (Xuv
particularly useful on the surfaces. To do this, we will first obtain the so-
called first integrals for the integral curves of a tangent field on a surface, -(Xu, - -2E - G.
that is, we will prove that these integral curves are, locally, level curves for
some suitable functions defined on the surface. This will occur, of course,
Analogously, the fact that F = (X, X) vanishes implies that
where the integral curves are not constant, that is, far away from points lEu lEv
where the field vanishes.
Xuu=-
T
Xu-- Xv2E
2G

xT= -lE, Xu + 2G
Proposition 7.51 (First integrals of tangent fields). Let V be a tangent 1Gu
field on a surface S, and let p e S be a point where V (p) # 0. Then - Xv2E

there exist an open neighbourhood W of p in S and a differentiable function

f : W -+ ll8 without critical points such that, if a : I -+ W is an integral Xvv 1G u
T = --
Xu +
1G%,
curve of V defined on an open interval I of ll8, then f o a is constant. That 2 E 2 G
is, the kernel of (df)q is exactly the line spanned by V (q) for each q e W. We have, then, all the necessary data to compute the right-hand side of the
first equality, and one can then see that
Proof. Using Remark 7.49, we know that the flow q5 of the tangent field V
is defined on E x (-e, s), for some e > 0, where E is an open neighbourhood 1 1 1E2 1Gu 1EuGu 1EG
of p in S. On the other hand, let u e TES be a vector that, together with 2 2 4E 4G 4 E +4 G
V (p), forms a basis of the tangent plane TES, and let '-y: (-S, S) -+ E C S, and now a straightforward calculation leads to the required equation.
S> 0, be a regular curve with y(O) = p and -y'(0) _ u. We now define a
differentiable map G : (-S, S) x (-e, e) -+ S by
Exercise (14): A dilatation q with centre a E I[83 and ratio µ E I[8* is given
G(s, t) _ q5(ry(s), t), b(s, t) E (-S, S) x by
Then, G(0, 0) _ 1('(O), 0) = p and differentiating with respect to s at s = 0 gy(p) = a + µ(p - a), by E 1[83.
and t = 0, we obtain
Thus, according to Example 2.58, if p E S and v E TS, its differential is
G8(O,0) = -I q5(y(s),O) = - ry(s) = w.
3=0 s=o (dq)(v) = µv,
Hints for solving the exercises 105
258 7. Intrinsic Geometry of Surfaces

and hence Ti(p) S' = (dq)(TS) = TpS, for each p E S. Therefore, if N is a Theorem 7.48 (Integral curves and the flow of a tangent field). Let V be a
local Gauss map for S, N' = N o -' is a local Gauss map for S'. Applying differentiable tangent field on a surface S of Euclidean space. Given p E S,
the chain rule, we obtain there exist a number 0 < e(p) < +oo and a curve (-E(p), gy(p)) -p S
such that
(dN), p E 5, aP(0) = p and aP = V o
and hence _ (1/µ)vp. It follows that H' o - (1/µ)H and K' o - which is maximal under these conditions. Moreover, the set
(1/µ2)x.
F={(p,t) eSxRIItI <E(p)}
is an open subset of S x ][8, and the map
Exercise (15): The differential of this inversion with centre at the origin
is given by
defined by F(p, t) = a(t) is differentiable. The curves aare called the
p E S, v E TS. integral curves of the field V and the map F is called the flow of the field V.
If N is a local Gauss map for S, we see that the map N' defined by the Remark 7.49. For all p e S, we have 0 and, thus, (p, 0) E 2' for any
equality p e S. Since the subset 2' of S x ll8 is open, there exist an open subset W
(N,p)
(N' ° )(p) =N(p) - 2 p pEs of S containing p and a number S > 0 such that W x (-S, 8) C 2. That is,
the flow F of the field V is defined on at least a set of the form W x (-8, S)
is a local Gauss map for S', since it gives a unit vector field perpendicular with S > 0, where W is an open neighbourhood of p in S. This holds for all
to each (dq)p(Tp5). From this, we can deduce a relation between p E S.
o (dq)p and (dN). It is not difficult to check now that if p E S
and ei E TS, i = 1, 2, is a principal direction associated to the principal Remark 7.50. Suppose that the surface S is closed in ][83 and that the field
curvature k2(p), then e2 = (d)(e) is a principal direction of S' and its V that we are dealing with is bounded, i.e.,
associated principal curvature k' (gy(p)) is related to the previous one by V b'p E S,

II
k((p)) = pI2k(p) + 2(N(p),p). for some positive constant M. Let a : (-s, ) -+ S be an integral curve of
Therefore V that is maximal relative to the initial condition a(0) = p e S and such
that 0 < e < +oo. We take a sequence {tfl}flEN of numbers in the domain

t
(H' 0 cb)(p) = pI2H(p) + 2(N(p),p),
of a such that limto = e. For each pair of natural numbers n and in,
(K' 0 /)(p) = IpI4K(p) + 4IpI2H(p) + 4(N(p),p)2.
we have
t a'(t) dt) < Ia'(t)Idt.
Exercise (17): Jn tn
Thus, since Ia'(t)I _ IV(a(t))I < M,
The function h = fi - f2 measures the difference of heights relative to the
common tangent plane at p E Si fl 52i which can be assumed to be given Ia(tn) - a(ttn)I M Its - tml.
by the equation z = 0. This function has, by hypothesis, a minimum at Then, the sequence of images {a(te)}flEN is a Cauchy sequence that must
the origin. Hence, its Hessian matrix is positive semi-definite. But, if we converge to some point p e I[83. This limit has to be on S since it is closed.
examine the proof of Hilbert's Theorem 3.47, the Hessian matrices of fi Furthermore, since V is a continuous map, we obtain
and f2 at the origin are the matrices of the second fundamental forms
of Si and of SZ at p relative to the basis of the tangent plane formed by lim a'(tn) = lim V(a(ts)) = V(q).
the vectors (1, 0,0) and (0, 1, 0). We finish the exercise by noting that
Then a could be prolonged on the right-hand side beyond the interval (-s, e)
the Hessian matrix of h is nothing more than the difference of those of
as the integral curve of V passing through q. In the same way, it could
fi and f2.
be extended on the left-hand side and we would obtain a contradiction.
7.7. Appendix: Some additional results of an intrinsic type 257
106 3. The Second Fundamental Form

4. The tangent field V may be extended to a vector-field V on the open

subset N(U) of three-space where it is constant on the normal segments of
the surface; that is, V : N(S) -* ][83 is defined by
V=Vo
as we did in (6.1) to prove the divergence theorem (Theorem 6.10) on a com-
pact surface. We now consider the following system of first-order differential
equations on N(U):
(7.17) x' = V (x).
The theorem of existence, uniqueness, and dependence on the initial condi-
tions gives, for each point a e N(U), a real number 0 < E(a) < +oo and a
differentiable curve as : (-E(a), s(a)) -* N(U) such that
In this case, the hypothesis implies that h = fi - f2 has a critical point
at the origin with positive definite Hessian matrix. Therefore, it has a
local minimum there.
Exercise (18): If a < 0, then the result is clear. Suppose that a > 0.
Take 0 < R < r, and let SR be a sphere with radius R that has constant
mean curvature 1/R with respect to its inner normal. Assume that S is
a graph over an open disc with radius r centred at the origin of the plane
z = 0. Drop the sphere SR onto S in the two possible ways and obtain,
using Exercise (20) at the end of Chapter 2, two points of tangency between
S and S. At least at one of them, say p, the normal of S coincides with
the inner normal of S. Now use Exercise (17) to obtain

aa(0) = a and a(t) = V(aa(t)), b't e (-E(a),s(a)). a < H(p) <

This is called the integral curve of the vector-field V and is maximal under Then aR < 1 for all R < r.
these conditions. Moreover, if we put
D= {(a, t) E N(U) x RI ItI < E(a)},
then this set is open in N(U) x ][8 C ][83 x I[8, and the map
F:D-*N(U)
defined by
F(a, t) = aa(t), b'(a, t) E D,
is differentiable.
Take a point p e U C N(U) as an initial condition for (7.17) and let
a = a be the corresponding solution, i.e., integral curve, of V defined on
(-E(p), gy(p)). We denote by f the composition r o a : (-E(a), s(a)) - R.
Then f(0) = r(a(0)) = r(p) = 0 and, using the properties given in Exercise
(1) of Chapter 4,
f'(t) = (dr)(t)(a'(t)) _ ((Noir)(a(t)),a'(t))
_ ((N o ir)(a(t)),V(a(t))) _ ((N o 7r)(a(t)), (V o ir)(a(t))) = 0,
for all t e (-E(p), e(p)). Hence, f = r o a is identically zero, i.e., the trace
of the curve a is contained in the surface S and, thus,
a'(t) =V(a(t)) _ (V o 7r)(a(t)) =V(a(t)).
Conversely, each curve on the surface satisfying this last equality is an inte-
gral curve of the field V on the open subset N(U) of II83. Keeping all this in
mind and working as in the proof of Theorem 7.30, we can infer the following
result.

Chapter 4
Separation and
Orient ability
4.1. Introduction
The geometry of a surface near a point usually does not tell us anything at all
about its shape near other distant points. The local geometrical behaviour
is governed by such things as the Gauss map, the second fundamental form,
and its associated functions: Gauss and mean curvatures and principal cur-
vatures. In spite of this, we have seen in the Jellett and Liebmann results,
i.e., Corollaries 3.48 and 3.49, that a particular behaviour of some of the
curvatures, together with a global condition, such as compactness, which
prevents the surface from being a proper open subset of a larger surface,
can determine the whole surface geometrically and topologically.
To have adequate tools in order to carry out the global study of a surface
in Euclidean space, we start in this chapter and continue in the following
chapter to develop some results of an analytical and a topological nature
mainly for compact surfaces. We remark that compactness, or the weaker
hypothesis of closedness, is a natural condition to impose in order to avoid
surfaces that are contained as a proper subset of a larger one.
Our first goal will be to prove that a compact connected surface separates
Euclidean space into exactly two domains. That is, we want to establish
a three-dimensional version of the famous Jordan curve theorem whose
history and proof will be commented on in the introduction to Chapter 9
by giving a proof that is also valid in the two-dimensional case. This result
is usually known as the JORDAN-BROUWER theorem, in honour of CAMILLE
JORDAN (1838-1922), who stated it the first time for curves, as a more or
107
256 7. Intrinsic Geometry of Surfaces
Figure 7.5. Elliptic paraboloid
note that this upper bound is valid, not only for the extrinsic diameter of the
surface, that is, the diameter resulting from using the Euclidean distance,
but also for the intrinsic diameter that appears if we use the distance defined
on the surface in Exercise (5) of this chapter.
Remark 7.47. For a closed surface, it is impossible to deduce compactness
from the mere positivity of the Gauss curvature K. Indeed, let us recall the
elliptic paraboloid S, given by
S- {(x,y,z) ER312z=x2+y2},
which, since it is the graph of the differentiable function
R2
(x, V) E '-' 1 (x2 + y2),
is a closed surface diffeomorphic to the plane. According to Example 3.45,
we have that the Gauss curvature of S obeys
K(x,y,z) = (1+2z2)2' d(x,y,z) E S.
Thus, K is positive everywhere. However this paraboloid is not compact.
Of course, infPES K(p) = 0 since lim,z.+, K(x, y, z) = 0.
7.7.2. Tangent fields and integral curves. Let us consider a differen-
tiable tangent field V on a given, not necessarily closed, surface S. That is,
according to Definition 3.1, V : S - R 3 is a differentiable map such that
V (p) E 7TS, Vp E S.
We will work, as in the proof of Theorem 7.30 on the existence and unique-
ness of geodesics, on orientable open subsets U of S, and we denote by
N(U) the tubular neighbourhoods constructed on them when this is pos-
sible. Thus, on each of the N(U), we have a projection onto S and of an
oriented distance with the properties pointed out in Exercise (1) of Chapter
7.7. Appendix: Some additional results of an intrinsic type 255 108 4. Separation and Orientability

function H is positive. Then, we can see, by Exercise 3.9, that the surface
S is orientable. On the other hand, let p, q e S be any two points of the
surface. From the Hopf-Rinow Theorem 7.43, which can be applied since
S is closed, we have a geodesic p.b.a.l. -y : [0, 2} -+ S that minimizes the
length with 7(0) = p and -y(2) = q, where 2 is the length of ry. We apply
Lemma 7.44 to this geodesic and conclude that
(K o ry)(t).f (t)2] dt < e[f'(t)2 - k2.f (t)2} dt,
where F is the variation of ry given by
F(t, s) = exp,y(t)s f (t)B(t).
Moreover, if the function f [0, 2] -+ ll8 vanishes at the two ends of the
interval, then all the longitudinal curves of the variation FS : [0, 2] -+ S join
the points p and q since
F8(0) = expy(o)07(0) -p and F8(t) = exp7(e)07(e)
For this reason, the corresponding length function LF has a minimum at
s = 0. Thus L'F(0) > 0 and consequently
less trivial assertion, and of the Dutch mathematician L. E. J. BROUWER,
one of the founders of algebraic topology, who proved a general version in
1912. We will provide a proof based on an elementary version of the Z2-
degree theory. In addition, this proof will also enable us to consider in detail
the problem of the intersection of lines and surfaces. It will be followed by
a proof of the fact that any compact surface in R3 is orientable, another
result due to BROUWER for abstract surfaces. We will present the better-
known proof due to H. SAMELSON in 1969. We will conclude the chapter
with a proof of the existence of tubular neighborhoods, which we will use
extensively in what follows.
4.2. Local separation
The first objective of this chapter is to see that, in the case of a compact or
even a closed surface, the points of the space are clearly divided into two
- q.
classes. As a first approach, let us check that any surface satisfies a local
version of this property.
Lemma 4.1. Let p E S be a point in a surface. There exists a connected
open neighbourhood W of p in R3 such that W nS is connected and W -S has

f
e
[f'(t)2 - k2f(t)2] dt 0 exactly two connected components, either one having W n S as its boundary
in W.
for all differentiable functions f : [0, 2] -+ 1[8 such that f(0) = fQ!) = 0. Now
choose Proof. We know that the surface S near any of its points p E S can be
f(t) = sin t bt e described as an inverse image of a regular value of a function; see Exer-
which clearly obeys these conditions. Then cise 2.25. That is, there exists an open neighbourhood O of p in II83 and a
differentiable function f : O -* II8 having 0 as a regular value and such that
£12 O fl S = f'({O}). Note that p is not a critical point for f. Thus (df)P 0
0< cos2 t - k2 sin2 tJ dt =
o L
22 and we may assume, for instance, that f(p) 0. Define a differentiable
Therefore map G : O -* II83 by
G(x, y, z) _ (x,y,f(x,y,z)), (x,y,z) E O.
This is true for each pair of points p, q e S and thus It can easily be seen that G(p) _ (pl,p2, 0) and that (dG)p : ]L83 -* I[83 is
7r 7r
diam S < _ regular. By the inverse function theorem, there are an open neighbourhood
infS K(p) W C O of p in ]L83 and an open ball B centred at (pi,p2, 0) such that
As a consequence, S is a bounded subset of 1[83 and, since it is closed, it is G(W) = B and G : W -* B is a difFeomorphism. Therefore, W fl S is an
compact. open neighbourhood of p in S diffeomorphic to G(W fl S) = B fl P, where
P is the plane of ][83 with equation z = 0. The required statement follows
Remark 7.46. The upper bound provided by Bonnet's Theorem 7.45 for immediately. O
the diameter of a closed surface with Gauss curvature greater than or equal
to a positive constant cannot be improved. In fact, this is shown by the Remark 4.2. Lemma 4.1 above implies that, up to a diffeomorphism be-
spheres S2(r) with arbitrary radius r> 0. In this case, the Gauss curvature tween open subsets of R3, any point of a surface possesses a neighbourhood
is constant and its value is 1/r2. Hence k2 = infPES K(p) = 1/r2 and the that can be imagined as the equatorial disc of a Euclidean ball. The follow-
diameter of a sphere with radius r in ][83 is just ir/r. It is also important to ing is a consequence of this.
4.2. Local separation 109 254 7. Intrinsic Geometry of Surfaces

for each t e [a, b} and each s e R. Therefore

s (t,0) = ry (ry(t), f (t)B(t), 0) = f(t)B(t).
Taking derivatives with respect to t and again using the Darboux equations
in Exercise (10), we obtain
2

S N (7.16) ata (t,0) = f'(t)B(t) + f(t)a(t)(71(t),B(t))N(t),

2

as2
(t, o) _ 711(7(t)> f(t)B(t), o) _ .f (t)2a(t)(B(t), B(t)) N(t)
Figure 4.1. Local separation Then, it follows from (7.16) that
2
a2F
A surface separates Euclidean space locally into two com- asat
(t, o) = f'(t)2 + f(t)2a(t)(7'(t),B(t))2.
ponents. For this reason, we can distinguish, locally, two Also from (7.15) and (7.16), we get
sides on the surface.
aF 33F aa IaaF 3a2sF 2\ /3a2tF 3a2sF 2\
Remark 4.3. Another more or less immediate consequence of Lemma 4.1 at ' as2at ) (t, °) - t t' 2' (t, o)
is that, locally, the problem of determining the intersection of a simple
curve-or the trace of a regular curve-with a surface can be reduced, up to _ -f (t)2 oy(t)(B(t),B(t)) ory(t)('Y (t),'Y (t))
a diffeomorphism, to studying the case where the surface is the plane z = 0. Finally, again using (7.15) and (7.16), it follows that
For this, we will first define the concept of transversality between curves and
surfaces corresponding to what we considered in Example 2.64 for a pair of 8F 82F asat\

surfaces-and, for curves, in Exercise (17) at the end of Chapter 1. Let S
be a surface and let a : I -). R' be a regular curve defined on an interval
I of R intersecting at a point p = a(so) E S, with so E I. We will say
that the curve a and the surface S cut transversely at p when the tangent
vector a'(so) of the curve is not tangent to the surface S at p = a(so). From
this we can deduce a corresponding concept of tranverse intersection at a
common point of a surface S and a simple curve C; see Example 2.18. We
will say that a surface S and a simple curve C of R3 are transverse when
they intersect transversely at any point of s n C. We will first establish the
following property:
If S is a surface and C is a simple curve intersecting S
transversely in IRS, then the intersection SnC is a discrete
subset of 1R3. Moreover, in each neighbourhood of a point
in the intersection, there exist points of the curve on each
side of the surface-in the sense that we discussed above.
In fact, since diffeomorphisms of JR3 trivially preserve transversality, we
have already pointed out that it is sufficient to consider the case where
S = P is the plane of equation z = 0. In this case, if a : (-e, e) -> 1R3 is a
regular curve with a(O) = p E P that is transverse to P at p, we have that,
(t,0) = o.
at '
Substituting the three last equalities in (7.14), we obtain
f6
[f' (t)2 - f (t) {(t) ( (t), 'Y1(t))ay(t)(B(t), B(t))
Ja
-01(t)('y(t), B(t))2}} dt.
Now, it only remains to point out that the {-ye(t), B(t)} form an orthonormal
basis of the tangent plane Ty(t)S for each t e [a, b] and to take the definition
(3.5) of the Gauss curvature into account.
Theorem 7.45 (Bonnet's theorem). Let S be a surface closed as a subset of
Euclidean space II83 such that its Gauss curvature satisfies infPES K(p) > 0.
Then S is an ovaloid whose diameter, as a subset of ll83, satisfies
diam S <
infPES K(p)
Proof. We represent the infimum of the Gauss curvature K of the surface
by k2 > 0. Then, as a consequence of (3.6), the mean curvature of S, relative
to any local normal field, satisfies H2 > K > k2 > 0. At each point of the
surface we choose the local determination of the unit normal field so that the
7.7. Appendix: Some additional results of an intrinsic type 253 110 4. Separation and Orientability

to be compact, the fact that it is positively curved far away from zero
forces it to become compact, as seen in Theorem 6.8. We will now sharpen
Bonnet's result a bit. To do this, we will need the following lemma.
Lemma 7.44. Let ry [a, b] -* S be a geodesic p. b. a.l. on an orientable
:

surface closed as a subset of Euclidean space. The variation F : [a, b] x III S

S, given by
F(t, s) = exp,y(t)s f (t)B(t), b'(t, s) E [a, b] x ][8,
where B(t) = ry'(t) n (N o ry)(t) and N is a Gauss map of S, is a well-
defined map for all differentiable functions f : [a, b] -* ][8 and, moreover, its
Figure 4.2. &ansverse intersection
associated length function LF satisfies
b
if
L(0) [f'(t)2 - (K o )(t)f(t)2] dt.
= a a(t) _ (X (t), y(t), z(t)), t E (-e, e),
Proof. Since S is closed, we can apply Corollary 7.41 and assume that the then z(0) = 0 and z'(0) 0 0. This implies that there is an interval centred
exponential map of S is defined on the whole of TS. We then define a map at t = 0 in which z(t) vanishes only at t = 0 and in which there are points
with z(t) > 0 and with z(t) < 0, that is, points above and under the plane
F:[a,b]xR-pS P, which proves the above statement.
by the relation A direct consequence of elementary topology and the existence of this
F(t, s) = exp,y(t)s f (t)B(t), `d(t, s) E [a, b] x ][8, class of neighbourhoods for any point of a surface is the fact that the com-
plement in 1R3 of a closed connected surface S either is connected or has two
where B : [a, b] - ][83 is given by connected components.
B(t) _ -y (t) n N(t) and N(t) _ (No -y)(t), Proposition 4.4. Let S be a closed connected surface. Then R3 - S has
where f : [a, b] -* ][8 is any differentiable function and N is a Gauss map of at most two connected components. Moreover, each connected component of
R3 - S has S as its boundary.
S. Thus, this F is a variation of the geodesic -y whose variational field is
f(t)B(t), where B is the third vector of the Darboux trihedron associated
to ry; see Exercise (10) in the previous section. If LF : ][8 -* R is the length
Proof. Let C be a connected component of 1R3 - S. Our first observation
function of this variation, we know that, by definition,
is that, since S is closed, 1R3 -- S is locally connected and, thus, each C is

('
open in 1R3. We also have Bdry C # 0. Otherwise, we would have that
b
a2F 2 + aF 83F aF a2F 2 C = int C U Bdry C = int C = C, and thus C would be simultaneously an
(7.i4) L'1(O) _
a asat at ' as2at
I - at asat
fit, o at. open and closed subset of 1R3, which would imply that C = R', which is
impossible. We now put
But, for this specific variation, since F(t,0) = exp,y(t)(O,y(t)) = y(t) and lR -S=CUC',
taking Exercise (10) into account, we have where C is the union of all the connected components of ]R3-S different from
2
C, and thus it will be an open subset of R3. It follows that C U S = R3 - C'
(7.15) (t,0) = '(t) and ate (t,0) _ 'Y "(t) =
is closed and hence C C C U S. Thus, we have
at

where v is the second fundamental form of S relative to the Gauss map N.
On the other hand, recalling the definition of the exponential map and the
homogeneity of geodesics,
F(t, s) = eXp7(t)sf (t)B(t) ='Y('Y(t), S.f (t)B(t), 1) =
BdryC=C - CC S.
Therefore, Bdry C is a non-empty closed subset of S. We will now see that
it is an open subset of S as well. Indeed, if p E Bdry C C S, we know from
f(t)B(t), s), Lemma 4.1 that there exists an open neighbourhood W of p in JR3 such that
4.3. Surfaces, straight lines, and planes 111
W - S has exactly two connected components, namely C1 and C2, whose
common boundary relative to W is S f1 W. Since W - S C - S, we
have that C1 and C2 are contained in different connected components of
- S. In particular, each CZ, i = 1, 2, is contained in C or, alternatively,
does not cut C. If the latter occurs for i = 1 and i = 2, then we would
have C n W = 0, and consequently p C. This contradicts the fact that
p E Bdry C. Therefore, at least one of the sets CZ is contained in C, for
example, C1 C C. Then
W n S= Bdry C1 n W C C1 C C,
and, thus,
W n S C Bdry C.
This proves that the point p is interior to Bdry C and, thus, that Bdry C
is open in S. Since we are assuming that the surface S is connected, we
have Bdry C = S, as mentioned in the last statement that we want to
prove. Furthermore, the above reasoning also proves that if C is a connected
component of R3 - S and W is the neighbourhood given by Lemma 4.1 for
an arbitrary point p E S, then C contains at least one of the two connected
components C1 or C2 of W - S. Since these components are disjoint, there
cannot exist more than two. 0
4.3. Surfaces, straight lines, and planes
In this section, we will take an additional step in order to understand the
form in which a closed surface embeds in R3. Indeed, we will see that the
complement of such a surface is never connected. To do this, we will be
interested in some aspects of relative positions of straight lines and surfaces
and of planes and surfaces in Euclidean space.
First, since straight lines are the easiest examples of simple curves, what
we said in Remark 4.3 can be directly applied to studying the transverse
intersection of a straight line and a surface. In fact, if a line and a surface
meet transversely, then the intersection is a discrete set and, as near as one
gets to any point of this intersection, one can find points of the line on each
side of the surface in the local sense of Lemma 4.1.
We will show that there are situations where we can guarantee the ex-
istence of lines cutting a given surface transversely. For this, we need to
control the set of regular values of differentiable maps between surfaces,
which we will define now by analogy with Example 2.13.
Definition 4.5. Let Si and S2 be either surfaces in R3 or open subsets of R2
and let f : S1 -p S2 be a differentiable map between them. A point q E 52
is said to be a regular value of f when, for each p E f ({q}), we have that
252 7. Intrinsic Geometry of Surfaces
for some real numbers A and ,u. Hence, S is complete when C is compact
or when, even being non-compact, it has infinite length. To complete the
exercise, it suffices to present a simple plane curve with infinite length that
is not closed in its plane, for example, a spiral asymptotic to a circle.
Exercise (23): We know, from Exercise (11) above, that there is a unique
arc of a great circle joining p and q whose length is d(p, q). Let ry : [0, d(p, q)]
S2 be a parametrization of ry b.a.l. Let m = ry(d(p, q)/2) E 2. If ¢ is the
rotation of ir radians around m, the composition ¢ o -y is another geodesic
with
°'Y)(d(p, q)/2) = m and (cb o 'Y)'(d(p, 4)/2) _ -'Y'(d(p, q)/2).
By uniqueness, it follows that
( ° 7) (t) = 7(-t + d(p, q)), bt e [0, d(p, q)]
Then, ¢(p) = q and (q) = p. Moreover, ry : [0, d(p, q)/2] 52 is a geodesic
p.b.a.l. joining p and m and its length is less than ir. Therefore
d(p, m) = Lo(Ps)l2 'Y = Z d(p, 4)
An analogous argument applies for d(q, m).
7.7. Appendix: Some additional results of an intrinsic type
In the final section of this chapter, we will discuss some important results of
the intrinsic geometry of surfaces, such as the classification of surfaces with
parallel second fundamental form with constant principal curvatures and
the Hartmann-Nirenberg-Massey theorem, whose statement and proof do
not require any new tools, but which require a more involved use of the
tools that we have been using up to now. We include them in this separate
appendix to emphasize that they are not necessary for getting, on a first
reading, a wide and non-superficial idea about the subject of the geometry
of surfaces.
7.7.1. Positively curved surfaces. We will start by taking up one of
the results that gave rise to the so-called differential geometry in the large,
originally im Grossen, a German translation of the Latin expression in toto
due to Gauss. These results pertain to the global aspects that reveal in a
very clear way the relationship between the curvature and topology of certain
surfaces. We will deal with positively curved surfaces. In fact, we already
know that, if the surface is compact, its topology is necessarily spherical,
which is a part of the content of the Hadamard-Stoker Theorem 6.5. This
result shows how the sign of the curvature influences the global shape of the
surface. But it is still more surprising that, without assuming the surface
Hints for solving the exercises 251 1 12 4. Separation and Orientabili ty

Finally, the limit of the third derivative can be computed by applying the
recurrence given by the second-order equation. In fact,
. d3
1 m 3 Lt = - lim 0(K o F) (t, s) I (dexpp)t«(S) a'(s) I ds = -27rK(p).
t odt t o
To compute the required limit, it only remains to apply L'Hopital's rule three
consecutive times, taking into account the above equalities. For the second
Figure 4.3. 2'raresverse straight line

ILt dt.
limit, we can argue in a similar way, keeping in mind that, by Proposition 5.8,
t 2ir t
At=J (df )p is an isomorphism between TpS1 and Tg52 or equivalently-according
0 J0 o to Exercise (13) at the end of Chapter 2-when
To conclude, consider the equality by E f -' (f q)).
IJacfI(p) 0,
4irAt - Lt _ 1 girt - Lt girt + Lt
r 24
t - 3
3
art 3 art
- 12 Atart4
- irt2
For example, all the points of 52 that do not belong to the image off are
regular values in the sense that we have just defined. The following result
Now, take limits in this expression as t -* 0, using the previous computations
assures us, among other things, that most of the points of S2 are regular
and the fact that
values of the differentiable map f.
. 2'rt+Lt 2'r+Lt
lim = lim = 4, Theorem 4.6 (Weak version of Sard's theorem). Let f : Si -> S2 be a
t-+o in t-o ir
by L'Hopital's rule. Therefore, differentiable map, where S, and S2 are either surfaces or open subsets of
I R'. Then the set of regular values off is dense in S2. In particular, there
41-At - Lt
inn = K(p). always exist regular values of f.
t-o rt24
This means that, near an elliptic point, the geodesic circles satisfy an isoperi- Proof. The proof of the stronger assertion, usually known as Sard's theo-
metric inequality opposite to that shown for the plane. However, near a rem, will be given in Appendix 4.6 at the end of this chapter.
hyperbolic point we have a strict inequality in the same sense as that found
Remark 4.7. The same proof that we will give for this result will also
in the plane.
prove that, given a finite-or even countable-number of maps, f l ,-- . , f,
Exercise (20): From Exercise (15), we know that the straight lines of defined either on surfaces or on open subsets of 1R2, taking values in the
S parametrized in an affine way are geodesics of S. Any other geodesic same surface or open subset of 1R2, say S, then the set of the points of S
has to be, then, by uniqueness, transverse to all the lines that it meets. that are simultaneously regular values for fl,... , fn is dense in S.
Then, it admits aone-to-one projection onto the generating curve C. Let
Example 4.8 (Existence of transverse half-lines). Let S be a surface and
a : ][8 -* C C P be a parametrization b.a.l. (see Example 2.18), where P is
let po 0 S be an exterior point. In Example 2.79, we studied the central
the plane passing through the origin and perpendicular to a e ][83, al = 1.
projection map f : S -> S2 given by
Any geodesic ry of S is of the form
ny(s) =a(s)+f(s)a, f(p) - dp E S.
Ip - POI,
for some differentiable function f. Moreover, ry will be defined exactly where We computed its differential and saw that, if p E S, (df )p is an isomorphism
f is. But then if and only if the line passing through po and p is transverse to the surface
'y" (s)
= k (s)rt(s) + f"(s) a, at p. Therefore, a unit vector a E 52 is a regular value of this f when the
where n and k are the unit normal of a in P and its curvature, respectively. half-line starting at po in the direction of a cuts the surface transversely.
Hence y is geodesic if and only if /'(s) 1 T(8)S, that is, if and only if Thus, the following assertion is a consequence of Sard's Theorem 4.6:
1" - 0. Therefore From each point of 1R3 - S one can draw a half-line inter-
'y(s) = a(s) + (As + /) a, Vs e R, secting S transversely with direction arbitrarily close to a
4.3. Surfaces, straight lines, and planes 113 250 7. Intrinsic Geometry of Surfaces

given direction. Taking Remark 4.3 into account, the in-

tersection of these half-lines and S are discrete sets and,
thus, finite sets if S is compact.
Example 4.9 (Transverse planes). Let S be an arbitrary surface and let
P be a plane of R3 perpendicular to a unit vector a E S2. According to
to show that
I_
(df)exppv is an isometry as well on TS. By Gauss's Lemma 7.36, it suffices

for v, w e TS and v', w' E Tp'S' such that

0 < IvI_ IwI< E and (v,w)=(v',w')=O.
Lemma 3.2, we can cover the surface with countably many open orientable Using the parametrization Fand the corresponding one F' for S', as con-
connected subsets. We consider on each of them the two possible Gauss structed in Exercise (17) above with ta(s) = v and a'(s) = w, the equality
maps; see Lemma 3.3. We get in this way a countable family of differentiable that we are looking for will then be true if we show that Is) I does not
maps, all of them taking values in the sphere S2. From Remark 4.7, there is a depend on s and its dependence on t is determined by the constant value c
unit vector b E S2, arbitrarily close to a, which is a regular value for all these of the Gauss curvature K. Exercise (17) implies that IF8 is a solution of
maps. This means that all the points of the surface S whose tangent planes
are perpendicular to the vector b are regular points of these local Gauss x"(t)+cx(t)=O, 0<t<e,
maps. In particular, by the inverse function theorem (Theorem 2.75), these for each s. But
points form a discrete subset of S and, thus, they are at most countable. As limFs(t, s) = lim(dexpp)t«(s)a'(s) = 0,
a consequence, we can choose a plane P', arbitrarily close to P, in the pencil t-+o t-+o
of the planes perpendicular to the vector b, which avoids this countable set. where, in order to see that (dexp)(8)a'(s) depends continuously on t, one
Therefore, at the points of S n P', the tangent plane of S is not the direction can refer to Exercise (8). On the other hand, this same expression tells us
plane of P'. That is, S and P' intersect transversely. We summarize all of that
this in the following statement: lim 1F3(t, s) _ (dexp)oa'(s) = a(s),
t->o t
Given a surface S and a plane P in Euclidean space and, thus, there exists Fts(0, s) =as(s). Therefore
one can always find a plane P' arbitrarily close to P -in
the sense, for instance, of the unit normal vectors and the t o 1S)It = Ia'(s)I.
distances to the origin cutting S transversely. Hence, we obtain

We will now go more deeply into transversality of lines and surfaces by

showing that, in a suitable sense, transversality is an open condition. In fact,
we want to prove that, if a line and a surface intersect transversely at a point,
I_ Iif c> 0, IwIt if c = 0,

IwI sinh /t if c < 0.

then, when we move the line just slightly, transversality is preserved. In We are done since we can find analogous expressions for Fs(t, s) by replacing
order to say explicitly what just slightly means, we will now discuss sequences w by b(w)I = Iw
and limits of lines in Euclidean space. We say that a sequence of lines
{Rn}nEIY of R3 converges to a line R when there are affine parametrizations Exercise (19): It is clear that, again using the parametrization Ffrom the
an, a : Ilk -* R3 for each Rn and for R such that above exercises, that
2ir

lim an(t) = a(t), Vt E R. Lt=J IF8(t,s)Ids, 0<t<e,

0
This definition and subsequent properties extend immediately to sequences and, thus, Lt = 0. Moreover,
of half-lines and to sequences of segments. The following exercises, which 2ir

can be easily solved, emphasize the main properties of limits of straight to t oJo 1Fs(t' S)Itds = 2 Ia'(s)I ds = 2ir.
J
dtLt

lines.
The limit of the second derivative is
Exercise 4.10. Consider a sequence {R}eN of lines in R3 and a line R.
If pn E Rn and vn E R3 is a direction vector of Rn, for all n E N, and p E R t odt2Lt-t
s)Ittds=_
s)IFs(t, s)I ds=0.
Hints for solving the exercises 249
Thus, 1/312 has a critical point at O In general, if a parallel is a geodesic,
its distance to the axis of revolution is critical, or the tangent plane of
S along the parallel is parallel to the axis of revolution. The converse is
straightforward.
Exercise (14): Let y be a geodesic p.b.a.l. on the surface of revolution S
whose axis is R, the line through the origin with direction a E R3, lal = 1.
Since 'y" is normal to S at the point y, from the arguments in the solution
to Exercise (13), 'y" is perpendicular to ry A a and thus
det ('y", 'y, a) = 0.
Hence, it follows that
det(ry , ry, a) = c,
where c e R is a constant. This equality can also be written as
c=(,Aa) = )IAaI,
which is the Clairaut relation, since the scalar product on the left-hand side
is the cosine of the angle made by ry with the parallel passing through y
and the length on the right-hand side is the distance from ry to the axis of
revolution.
Exercise (17): Since expp : BE(0) -* BE(p) is a diffeomorphism, F is a
parametrization of S, restricted to (0, e) x I. Furthermore,
Ft(t, s) _ (dexp)(8)a(s) and Fs(t, s) = t(dexpp)t«(s)a'(s).
In this way, Gauss's Lemma 7.36 tells us that F is an orthogonal parametrizar
tion, that is, that (Ft, FS) = 0 and that, moreover, E _ IFtI2 = 1. We can
then apply Exercise (12) at the end of Chapter 3 and obtain
KoF=- 1 (G\t
2/G G t that S f1 R+ is a finite set, namely
where G = F3 2. Thus
IFsItt+(KOF)IF8I =0,
as required. This expression is valid on the domain of the polar coordinates
(t, s) of that is, on {Op}.
Exercise (18): We have to show that (df)q is a linear isometry for each
q e B6(p). But we know, by definition and the chain rule, that
(df)exppv o (dexp)v = o.
If v = this equality is nothing more than (df)p = i5 and, so, (df) is a
linear isometry. Then, we have to prove for each v e {0} that
114 4. Separation and Orientabili ty
and v e R3 is a direction vector of R, prove that limn 0 Rn = R if
lira pn = p and lira vn = v.
n- too
Exercise 4.11. T Let {Rn}nEN be a sequence of lines of R3 converging to
a line R. If pn e Rn and vn E R3 is a direction vector of Rn, for all n e N,
and the limits
lim pn = p and lim vn = v 0
n- too
exist, show that p E R and that v is a direction vector of R.
Exercise 4.12. T If {Rn}nEN is a sequence of lines all of which intersect a
given compact set of I[83, prove that there exists a subsequence converging
to some line.
Exercise 4.13. T Let {Pn}nEN be a sequence of points of a surface S con-
verging to a point p e S, and let {Rn}neN be a sequence of lines converging
to a line R. If each Rn is tangent to S at the point pn, prove that R is
tangent to S at the point p.
We will now see, as promised, that transversality of lines and surfaces is
an open condition. This fact will imply that, if the surface is compact, the
number of points of a transverse intersection is preserved when one moves
the line just slightly.
Proposition 4.14. Let R+ be a half-line, whose origin is in R3 - S, that
intersects a compact surface S transversely. If {R}neN is a sequence of
half-lines starting at points of R3 - S converging to R+, then there exists a
number N E N such that, if n > N, R,+z cuts S transversely and the number
of points of the sets R,+z f1 S and R+ fl S coincide.
Proof. Transversality of the intersection and compactness of S guarantee
Sf1 R+ _ {p1,...,pk}.
Suppose that the assertion about the sequence {R}eN is false. Then we
can assume, passing to a subsequence of the original one, that the sequence
{R}eN of half-lines starting from points of R3 - S and converging to R+
is formed by half-lines satisfying at least one of the following conditions:
A: R,+z is tangent to S at some point.
B: The number of points of R,+z f1 S is greater than the number of
points of R+ f1 S.
C: The number of points of R,+z f1 S is smaller than the number of
points of R+ f1 S.
4.3. Surfaces, straight lines, and planes 115
We will show that any of the above alternatives leads to a contradiction. If
A occurred, then there would be a sequence of points pn E S n Rn such that
Rn would be tangent to S at each pn, for all n e N. Since S is compact,
we can assume that this sequence converges to a point p E S, which also
belongs to R by Exercise 4.11. Then, using Exercise 4.13, the line R would
be tangent to S at p. This is a contradiction to the transversality hypothesis.
Next, if B occurred, since any sequence of points of the form pn E S n Rn
has to converge by the compactness of S and Exercise 4.11 for some of the
p1i ... , pk E S n R, then we could find two of these sequences with different
terms converging to the same limit point in S n R. We will call such a point
a point of double intersection of S and R. We will see later that this implies
that S and R are tangent at this point and so we would again arrive at a
contradiction. Finally, if C were true, then some of the points p1, ... , pk of
S n R could not be limit points of a sequence pn E S n R. Then we would
find a neighbourhood of p in S that would not intersect any of the R. We
will call such a point a point of first contact between S and R. We will also
see below that this would require R to be tangent to S at this point. In
conclusion, none of the three possibilities A, B, or C can be true, provided
we have proved the following result. D
Lemma 4.15. Let S be a surface, let R be a line of R3, and let p be a point
in S n R. We say that p is a point of first contact between S and R if there
is a sequence of lines {Rn}neN converging to R and a neighbourhood V of p
in S such that Rn n V = 0 for all n e N. We say that p is a point of double
intersection of S and R if there are two sequences {Pn}nN and {q}j of
points of S converging to p such that pn qn for all n e N and the sequence
of lines Rn determined by the pairs (pn, qn) converges to R. In both of these
cases, R is tangent to S at the point p.
Proof. Suppose that p satisfies the first case. Using Lemma 4.1, we can find
an open connected neighbourhood W of p in R3 such that W n S C V and
W - S has exactly two connected components C1 and C2. Let B C W be
an open ball of R3 centred at p. Since each Rn n B is connected, Rn cannot
intersect C1 n B and C2 n B at the same time. We extract a subsequence
formed of lines that cut only one of the two open sets, say C1 n B. By
continuity, the limit R cannot intersect C2 n B. Hence, R touches S at p
and it is locally on a side of S. Thus R is tangent to S at p from Remark 4.3.
On the other hand, if we were in the second situation, we may assume,
using rigid motions, that p is the origin and R is the z-axis in R3. If R were
not tangent to S at p, by reasoning similar to what we used in Corollary 2.76,
then a suitable neighbourhood of p in S could be described as the graph of a
function f defined on a disc D centred at the origin of the plane R2 such that
f(O, 0) = 0 and such that 0 is a regular value. Let p'n and q'n be the points
248 7. Intrinsic Geometry of Surfaces
by any natural number, where one of them is right-handed and the other
one is left-handed.
Exercise (12): Let p e S be a point and v e TES any unit vector. The
geodesic ry with ry(0) = p and -y'(0) = v lies in a plane with unit normal, for
example, a e Il83. Then ('y'(t), a) = 0 and ('y"(t), a) = 0 for each t. A direct
computation, or Exercise (10), permits us to rewrite the last equality as
7y(t) ('y'(t), 'y'(t))((N o 'y)(t), a) = 0,
where N is a local Gauss map for S and a is its second fundamental form.
By continuity, if vp(v, v) 0, we have ((No'y)(t), a) = 0. Hence, ry'n (Nory)
is always a or -a. Thus
0 = [ry'A(Nory)]' =y'A(Noy)',
since ry is geodesic. At t = 0, this equality tells us that (dN)(v) liv, or that
u) = 0, for each u e TS such that (u, v) = 0. Therefore, the two
quadratic forms v and (,)are proportional. Thus, S is totally umbilical.
Exercise (13): In Exercise (3) of Chapter 2, we saw that if S is a surface
of revolution whose axis is the straight line R passing through the origin
with direction given by the unit vector a e R3, we have, for each p e S, that
p A a E TS is a non-null vector tangent to the parallel passing through p.
Hence, the normal of S at p belongs to the plane spanned by p and a. Thus,
if a : I -* S is a meridian p.b.a.l., we have a' 1 a A a or, in another form,
det(a', a, a) = 0
along I. Differentiating, we obtain that det (a", a, a) = 0, or, equivalently,
a" is perpendicular to a n a at any time. But a" is also perpendicular to
the unit vector a'. Since a' and a n a are perpendicular, they generate Ta S.
Therefore, a" 1 TS and thus a is a geodesic. Suppose now that a : I -* S
is a parallel. The trace of a is, in this case, a circle of radius, say, r > 0, in
a plane perpendicular to the vector a. Let us provisionally put the origin at
the point of intersection of this plane and the axis R. Then, if a is p.b.a.l.,
a" = - (1/r) a and, thus, for a suitable choice on the normal of S, we have
1
No a = --a.
r
Let p be a point in the trace of a. We know that iI = r, N(p) =
and (p, a) = 0. Consequently, if /3: J -p S is a meridian with /(so) = p for
some so e J, we have
(/3'(so),N(/3(so)))
0
r (/ (so) , p)
r ( (So) (So)
Hints for solving the exercises 247
and the corresponding equalities for g hold. Thus
(10 X )v (un, vn) = (g o X )v (un, vn).
(f o X )u (un, vn) = (g o X )u (un, vn),
Since the partial derivatives of f o X and of g o X are continuous, we take
limits in the previous equalities and we are done.
Exercise (9): From the solution to Exercise (8), the set
A = {p e S Jf (p) = g(p) and (df)_ (dg)}
is a closed subset of S that we are supposing to be non-empty. Let e> 0 be
a number such that BE(p) is a geodesic ball. By Theorem 7.25, if v E TES
and t e (-e(p,v),e(p,v)), we have
f('y(p,v,t)) _ 'Y(f (p),(df)(v), t) and 9('Y(p,v,t)) _ 'Y(9(p),(d9)(v), t).
Therefore, if v e TS with lvi < E,
f(expv) = f('y(p, v, 1)) = 9('Y(p, v, 1)) = 9(eXApv)
and, thus, f and g coincide on BE(p). Hence, their differentials also coincide
on the points of this open set, which implies that A is an open subset of S.
Since the surface is connected, A = S and f = g.
Exercise (11): In Examples 7.26, 7.27, and 7.28, we have seen that the
traces of geodesics in a plane, in a sphere, and in a right circular cylinder
are, respectively, straight lines, great circles, and circular helices (including
generating lines and their orthogonal circles). Then, in the case of the plane,
there is exactly one geodesic joining two different points: the corresponding
segment. In the case of the sphere, if the two points are not antipodal, they
determine a unique great circle and, thus, there are exactly two geodesics
joining them: the two segments determined by the two points in this circle,
one of them with length less than it and the other one with length greater
than ir (provided that the sphere has unit radius). Then, by the Hopf-Rinow
Theorem 7.43, a segment of the great circle of length less than ir minimizes
the length between its endpoints. If the two points are antipodal, there are
infinitely many great circles passing through them. Each of them determines
two geodesics with length ir. In the case of the cylinder, it is a question of
showing how many circular helices join two given different points. Cutting
the cylinder along a generating line and unwinding (this process is a local
isometry, according to Exercise 7.4 and, hence, maps geodesics to geodesics,
by Theorem 7.25), the problem is reduced to finding finite sequences of
parallel lines at a constant distance joining two given points in a plane strip.
If these two points lie in the same circle, there are two geodesics: the two
arcs of this circle. In any other case, there are countably many geodesics:
two helices whose pitch is the difference of the height of the points divided
116 4. Separation and Orientability
of D that are projections over z = 0 of pn and qn converging to p. Then p'n
and q'n would converge to the origin and we could take as a direction vector
of the line Rn joining pn and qn the vector
_ qn - pn qn - p'' f(q) - f(p)
I'- 'I
__
n
I'- Pn'I
fn qn Pn '-in Pn
Taking a subsequence, we can assume that the sequence of the projections
vn of {Vfl}flEN converges to a vector v' E R3, which would have unit length.
Then, the sequence of z-components converges to (df)(oo) (v') by the mean
value theorem. But from Exercise 4.11, we see that this vector is a direction
vector of R, that is, of the z-axis. This is a contradiction, showing that R
is tangent to S at p.
4.4. The Jordan-Brouwer separation theorem
With regard to the definition of a surface, when we mentioned what is under-
stood by a simple curve in Euclidean space see Example 2.18 we asserted
that such a curve must topologically be a circle, provided that it is com-
pact and connected. In fact, we will give the proof of this fact in Chapter
9. Once we have accepted this result, we should intuitively expect that, as
happens with the circle, any compact connected simple curve divides the
plane into exactly two connected components. This assertion is the subject
of the Jordan curve theorem. With respect to surfaces, the situation is not
exactly the same, because, even in the compact case, they do not have a
fixed topological structure: there are spheres, tori, surfaces with two holes,
etc. However, we will show that an analogue to the Jordan curve theorem
remains true. It is widely known as the Jordan-Brouwer theorem. Compact
connected surfaces, whatever their topologies are, separate Euclidean space
into exactly two pieces. More generally, the same can be said for surfaces
that are closed as subsets of R3. The following proof for the case of surfaces
is also valid for curves, as we will point out in Chapter 9, although in this
case it could be significantly simplified.
Theorem 4.16 (The Jordan-Brouwer separation theorem). Let S be a con-
nected surface that is closed as a subset of R3. Then R3 - S has exactly two
connected components whose common boundary is S.
Proof. We will give the proof in the case when the surface S is compact
(for the closed non-compact case, see Exercise (6) at the end of this chap-
ter). Keeping Proposition 4.4 in mind, it suffices to prove that R3 - S is
not connected. Now, since S is assumed to be compact, as shown in Exam-
ple 4.8, from each point in R3 - S one can draw a half-line that intersects
S transversely and, hence, at a finite set of points. Call 1 and 1k', respec-
tively, the subsets of R3 - S from which one can draw a half-line cutting
4.4. The Jordan-Brouwer separation theorem 117 246 7. Intrinsic Geometry of Surfaces

and, thus, taking the infimum in a, we have

d(p,q)>Ip-qI
This inequality implies that, if d(p, q) = 0 for some p, q e S, then p = q.
Hence, d is a distance. But, in addition, it implies that the metric space
(S, d) is complete if S is closed in R3. In fact, each Cauchy sequence in S
must be a Cauchy sequence in R3 as well and, thus, it converges. Since S is
closed, the limit has to be in S. Finally, if f : S - S is a diffeomorphism,
it is clear that
F f (p),f (q) = f ° Fp,q, dp, q e S.
Figure 4.4. Angular region and simple curve If, moreover, f is an isometry, f preserves the lengths of curves. Thus
transversely at an even or an odd number of points. It is obvious that we d(f (p), f(q)) = inf L(f o'Y) = inf L(ry) = d(p, q)
Yerp,q Yerp,q
have R3 - S = 1 U 1' from the above. An important consequence of Propo-
for p, q e S, and we are done.
sition 4.14 is the fact that both 1 and f' are open subsets of R3. Since S is
compact, one can easily find half-lines that do not intersect it. Thus, f' is a
Exercise (6): Let us check first that vt is definite at any point ft(p), for
non-empty set. But 1 is not empty since, if p e S is a point of the surface each t e [0, 1]. In fact,
where some height function attains its minimum, it is straightforward to
find a half-line contained in the normal line of S at p that intersects S only
at p. Therefore, we have expressed the complement R3 - S as a union of two
detQ ft(P) - Kt(ft(p)) _ (Kt ° ft)() = K(p) > 0,
where we have used the Theorema Egregium 7.10. Now, let X : U -> S be
non-empty open sets of R3. We will now see that they are disjoint. Suppose, a parametrization with X(0, 0) = p. Each Xt = ft o X is a parametrization
on the contrary, that there are a point x S and two half-lines L1 and L2 of St with Xt(0, 0) = ft(p) and each Xt is obtained by fixing t in the map
starting at x cutting S transversely and such that L1 f1 S is a set with an G : U x [0,1] -* ][83, given by
odd number of points and L2 f1 S is a set with an even number of points. Let
P be the plane determined by the two half-lines. By Example 4.9, we have G(u, v, t) = f(X(u,v),t), b'(u, v) E U b't e [0, 1].
a plane P', close to P, intersecting S transversely. Hence, by Theorem 2.67, As a consequence, if we have chosen X so that the inner normal of S is given
their intersection is a simple curve C, which is necessarily a closed, and by det(Xu, Xv, N o X) > 0, we have, using the expressions in (3.10), that if
thus a compact, subset of S. Again by Proposition 4.14, we find in P' two w = Xu(0, 0) E TS,
half-lines L1 and L'2 starting at a point x' that still cut S transversely at an
odd number and an even number of points, respectively. Then (L1 U L2) n S aft(P) «dit)p(w), (df)(w)) = IGu ^ Gvl Gv, Guu)(0, 0, t)
has an odd number of points. But (L1 U L) f1 S = C f1 (L1 U L), and we
finally arrive at a contradiction to the following result about the intersection The point is that this expression is continuous in t and, since it cannot
of simple curves and angles in the plane. vanish because v ft is definite, it does not change sign. But, when t = 0,
this sign is positive. This completes the exercise.
Lemma 4.17. Let I' = Ll U L2 be the union of two different half-lines of
][83 with common origin at a point p, and let C be a compact simple curve Exercise (8): Let {Pn}nN be a sequence of points of S such that
in the plane P determined by I' that does not contain p and that intersects f(p) = and (dg),
Ll and L2 transversely. Then C fl I' has an even number of points. for each n E ICY, converging to a point p e S. Since f and g are continuous,
Proof. F divides the plane P into exactly two connected components, say it is clear that f(p) = g(p). Let X U -> S be a parametrization with
01 and 02. But we know that each connected component C' of C is home- X (u, v) = p for some (u, v) E U. We can assume that pn = X (u, vn) with
omorphic to a circle; see Example 2.18 and Theorem 9.10. We move along (un, vn) E U for each n e N. Then
C' with the orientation determined by any homeomorphism of this type. By (.f ° vn (.f ° X)v(un,vn) _
Hints for solving the exercises 245 118 4. Separation and Orientability

Exercise (2): We know that the helicoid 7- is covered by the image of a transversality, C' n F has a finite number of points, and each time that C'
parametrization X : R2 -* h, given by cuts F, it passes from one component Oi, i = 1, 2, to the other one. This
X (u, v) = (vcosu,vsinu,u), `d(u, v) E implies that if we start to walk along C' from a point of 01, each time that
we come back to the same component 01, we have crossed over F an even
and that the map Y : R2 -* C, given by number of times. In particular, this can be applied to a complete turnaround
Y (u, v) = (coshvcosu,coshvsinu,v), C'. Thus C' n F has an even number of points. Since C has finitely many
restricted to each open interval of the form (0, 0 + 2ir) x R, with 0 E III, is a connected components, we deduce that the same is true for C n F.
parametrization of the catenoid C. We define a map i5 : -p C by Remark 4.18. Since Euclidean space R3 is locally connected and S is a
=YoFoX-1, closed subset of it, the connected components of R3 - S are open subsets of
J2 is the map defined by R3. On the other hand, since Euclidean space is locally arcwise-connected
where F : R2
and R3 - S is an open subset, its connected and arcwise-connected compo-
F(u, v) = (u,argsinhv), nents coincide. Therefore, Theorem 4.16 also shows that a closed connected
which is clearly differentiable. Differentiating with respect to u and v in the surface separates R3 into two arcwise-connected components.
equality i5 o X = Y o F, we have Remark 4.19. The proof of the Jordan-Brouwer Theorem 4.16 also pro-
(dq)X(,)(-v sin u, v cos u,1) = 1 + v2 (- sin u, cos u, 0), vides a geometrical criterion for deciding whether a given point in R3 - S
is in either domain determined by S. This criterion consists of counting the
1 parity of the number of points in the intersection of the surface and any
(d)X(,)(cos u, sin u, 0) = 2 (v cos u, v sin u, 1).
half-line starting at the point and cutting the surface transversely.
1-v
Since Xu = (-vsinu,vcosu,1) and Xv = (cosu,sinu,0) form a basis of Remark 4.20 (Inner domain determined by a compact connected surface).
Txfl, this shows that (d)x(,) preserves the scalar product of vectors. If B is a ball containing the compact surface S, then its complement R3 - B
is a connected subset of R3 - S. Hence, it belongs to one of the two connected
Exercise (5): If p, q e S, there is one-to-one correspondence between I'p,q components C1 and C2. Suppose that R3 - B C C2. Then C2 is not bounded
and I'9ip, given as follows: if a : [a, b] -* S is a piecewise differentiable curve and C1 U S = - C2 C B and, thus, C1 is bounded. Therefore, only one
with a(a) = p and a(b) = q, we consider another curve a-1 [a, b] -* S of the two connected components in which a compact connected surface
defined by a-1(t) = a(-t + b + a). Since divides the space is bounded. We will call it the inner domain or simply
L(a1) = L(a), the domain determined by S and we will usually denote it by SZ. Clearly,
ci = SZUBdry SZ = SZUS is, from the above, a compact set. It is also clear that
we have d(p, q) = d(q, p). That is, d is symmetric. It is also clearly non- any half-line drawn from a point outside the ball B does not intersect the
negative. We next show that d satisfies the triangular inequality. Let p, q, r e surface. Taking Remark 4.19 into account, the points of the outer domain
S, a E I'p,q and ,Q E I'q,,.. Since lengths of the curves remain invariant under - ci are characterized as the origins of transverse half-lines intersecting
reparametrizations, we can consider that the two curves a and ,0 are defined the surface at an even number of points. Consequently, the points of the
on [0,11. Then a * /3: [0,1] -* S belongs to I'p,r and, thus, inner domain SZ are characterized as the initial points of transverse half-lines
d(p, r) Loa * Q) = L(a) + Lo(/3) cutting the surface an odd number of times.
Since this happens for every a e I'p,q and ,Q E I'q,T, taking the infimum on As a first remarkable consequence of the separation property stated in
the right-hand side, we get the Jordan-Brouwer theorem, we have the following result, originally due
d(p, r) < d(p, q) +d(q, r), b'p, q, r e S. to Brouwer in 1912, already mentioned when we defined the orientation of
In order to see that, in fact, d is a distance, we will relate it to the Euclidean surfaces in Remark 3.12.
distance. If p, q e S, let a E I'p,q. Using Exercise 1.9, Theorem 4.21 (Brouwer-Samelson's theorem). Any compact surface in Eu-
L(a) Ip-qi clidean space is orientable.
4.4. The Jordan-B.rouwer separation theorem 119 244 7. Intrinsic Geometry of Surfaces

Exercise 7.38: Given p E S2, if u E TpS2 is a unit vector, Example 7.27

implies that
y(p, u, t) =p cos t + u sin t, `dt e R.
This is why, if v e {0}, it follows from the homogeneity of geodesics
that
lVvi
-y(p, v, t) = p cost + sin t, `dt E R.

Thus, according to the definition,

Figure 4.5. Inner domain and inner normal field

expP v =
Il if v
1pcosJvi+--sinv 0,
p if v = Op.

Proof. It is sufficient to prove the result for each connected component Now, it only remains to check that the map i5 : S2 - {p, -p} -* given
of the surface. We will assume, then, that S is a compact connected sur- by
face. Consider the inner domain 1 determined by the Jordan-Brouwer The-
orem 4.16. If v E R3 is a unit vector normal to the surface at a point p E S, 9' = arc cos (p, 4)
- (p, 9')2 -
(p,q)p)
(9'

we will say that v points inward, or v is inner, when there is a neighbour-

hood of p in the half-line starting at p with direction v which is contained
in Q. We will show that this property of being inner, referring to non-null
vectors normal to the surface, obeys the conditions required in Exercise 3.9.
In fact, if a : ilk -+1R3 is the line
a(t) = p + tv,
we have that a cuts S at a(O) = p transversely, because a'(0) = v
TBS. Hence-use Remark 4.3-there exists r > 0 such that a(-e, £) fl S =
{p}, and there are points in the image of a on both sides of the surface.
Consequently, either a(0, e) C 1 or a(-E, 0) C Q. In other words, either v
is inner or -v is inner, and the two things never occur simultaneously. On
the other hand, given a unit vector v normal to S at p, there are a connected
neighbourhood V of p in S and a Gauss map N defined on V such that
N(p) = v. Choosing this V as w fl s with W as given in Lemma 4.1, one
has that either all the N(q) or all the -N(q), with q E V, are simultaneously
inner to the surface. Then we can apply Exercise 3.9 as mentioned, and see
that the surface has to be orientable.
Remark 4.22. It follows from the above that on any compact surface S
we have a Gauss map N : S -> S2 such that the half-line starting at each
p E S in the direction of N(p) goes into the inner domain S2 determined
by the surface-at least, a sufficiently small enough segment of it. This N
will be called the inner unit normal of the surface S. From now on, unless
otherwise specified, we will always consider the orientation supplied by the
inner normal fields of its connected components. Note that this distinction
is a differentiable inverse map of exp restricted to B(O) - {O} C
Thus, 2 - {-p} = B71.(p) is a geodesic ball of radius ir centred at p.
Then, taking p = (0,0, 1), the north pole of the sphere, the map X
(O,ir) x (O,2ir) -* S2, given by
X (p, B) = cos p (0, 0, 1) + sin p (cosO,sinO,O),
is a parametrization covering the whole sphere except a half-meridian and,
moreover, IX p A Xe = sin p. Then, if f : -* III is integrable,
fir
ff(p)dp= f(p, e) sin p d0 dp.
Exercise (1): Let q be a rigid motion of ][83 fixing the point a e R3. If
p e ST (a), with r > 0, we have
kb(p) - at = kb(p) - q(a)I = I - al =
Therefore (S(a)) C ST(a). This is also true for -1 and, thus, keeps the
sphere ST (a) invariant. The restriction of to this sphere belongs, according
to Proposition 7.1, to the group of isometrics of ST (a). But there are no other
isometrics of this sphere since each isometry f of ST (a) is the restriction of
a rigid motion i5 of ][83, as seen in Example 7.9. It suffices now to show that
q must fix the centre of the sphere. This follows, for example, from the fact
that any affine map takes the centre of a quadric to the centre of the quadric
image.
Hints for solving the exercises 243 120 4. Separation and Orientability

Therefore between inner and outer normal fields coincides with what we did for spheres;
a2F aF
axZax ax e
0- see Example 3.6.

on O, for all i, j,k = 1, 2, 3. But the partial derivatives aF/ax1, aF/ax2, and 4.5. Tubular neighbourhoods
aF/ax3 form an orthonormal basis of R3 at each point of O. Consequently
We will now establish that compact surfaces in Euclidean space-for the case
a2F of simple plane curves, see Chapter 9-have special neighbourhoods, whose
-0
existence will be very useful for proving global properties of the surface. An
on O, for each i, j = 1, 2, 3. Then the three components of F are linear example of this type of result already appeared in the Liebmann and Jellett
functions. Thus, F(x) = Ax -+- b for all x E O for some 3 x 3 matrix A and results in Corollaries 3.49 and 3.48. By the term global we mean that we
some b E R3, since O is connected. Thus, (dF) = A at each x E O and can derive some information about the entire surface from local data such as
A must be orthogonal. Hence F is the restriction to O of a rigid motion of the various curvatures and other geometric data that indicate how a surface
space R3. bends near a point.
Let S be a surface. Since ][83 is a metric space, the neighbourhoods
Exercise 7.16: Since b is self-adjoint, there exists an orthonormal basis easiest to construct for S are the so-called metric neighbourhoods: for each
of the Euclidean vector plane V that diagonalizes it. Since /i is also self- 6> 0, we denote by Bb (S) the set of the points of the space whose distance
adjoint, if A1 and A2 are its eigenvalues, then the two endomorphisms, with to S is less than S, that is,
respect to this basis, will be represented by two matrices
B5(S) _ {p e II83 I dist (p, S) <6},
=(A1 O) ?=(a bl
0 A2)' b c where dist (p, S) = infQEs Ip - 9I It is clear that Bb (S) is an open neigh-
bourhood of the surface S for each 8> 0. A first observation is that these
for some a, b, c E R. The required formula is easily proved replacing each
neighbourhoods consist of segments of normal lines with length 26 centred
endomorphism by the matrix that represents it.
at the points of the surface, at least when it is closed as a subset of I[83.
Exercise 7.17: If 0, the endomorphism ¢+b would be definite, Lemma 4.23. If S is a closed surface in II83, the set Bb(S) consisting of
say, positive definite. If {ei, e2} is a basis of the plane diagonalizing b, we the points of ][83 whose distance to S is less than S > 0 coincides with the
have union N5(S) = Us Na(p) of all the open segments of normal lines of S
('ib(e),e) > 0, i = 1,2. with radius S and with centres at the points p e S.
Consequently
Proof. If p e S is a point of the surface and b is a positive real number,
('ib(ei),ei)('ib(e2),e2) denote by N5 (p) the segment with centre p and radius b in the normal line
('ib(ei),ei)('ib(e2),e2) - ('ib(ei),e2)2 = detib, of S at p. If q e Nb(p) with p e S, then it is immediate that dist (q, S) <
which contradicts our hypothesis. Therefore det(¢ + 'i/P') < 0. Moreover, if Iq - pI < b and, so, q e Ba(S). Conversely, if q e Bb(S), consider the
equality holds, there would be at least anon-null vector in the kernel of distance function on S to the point q. Suppose that the surface S is closed
b + 'b. Let {u, u2} be a basis diagonalizing ¢ + b, that is, such that as a subset of I[83. Then, we see from Exercise 2.73 that this function attains
its minimum at a point p e S and, from Example 2.69, q is on the normal
(ui) + 'b(ui) = 0 and (u2) + 'b(u2) _ Au2i A E R.
line of S at p. Since we also have that Ip - qI = dist (q, S) <6, we conclude
From the first equality we deduce that that q E Na(p). O
(q(ui),ui) _ -((ui),ui) and -('ib(ui),ua),
This geometrical description of the metric neighbourhoods of a closed
which, together with the facts that det ¢ = det /i and ¢ and b are definite, surface suggests that, if we can arrange that the segments of normal lines
gives the equality involved do not intersect each other, then we will be able to establish that
(cb(u2),u2) = -('b(u2),u2), these neighbourhoods are topological products of the surface and an open
implying A = 0. Thus, in this case, b _ -sb. interval of R. It seems that it could be done if the radius is small enough.
4.5. Tubular neighbourhoods 121 242 7. Intrinsic Geometry of Surfaces

(20) r Consider a surface S like that studied in Exercise (7) of Chapter 3, that
is, a right cylinder over a simple curve C. Calculate all its geodesics and
show that the exponential map of this surface is defined on the whole of
TS, provided that C is compact or has infinite length. Check, however,
that there are curves C for which the resulting surface S is not closed.
(This shows that the converse of Corollary 7.41 is not true.)
(21) Show that the right cylinder constructed over a logarithmic spiral-see
Exercise 1.13-is not a complete surface.
(22) Let 2 be the unit sphere centred at the origin and d the distance defined
in Exercise (5) above. Prove that d(p, q) < it for all p, q E S2 and that
Figure 4.6. Tubular neighbourhood N5 (S) equality holds if and only if p and q are antipodal.
(23) T Given two non-antipodal points p, q E S2 of the unit sphere, prove that
there exists a unique point m E 2 such that the 180 degrees rotation
In any case, we should be able to control, at the same time, all the normal around m takes p to q and q to p. It is called the middle point of p and
lines of the surface. For an orientable surface S, not necessarily closed, this q. Show that, moreover, if d is the intrinsic distance on the sphere, then
can be done by means of the map F : S x III -p IIS3, given by 2d(p, m) = 2d(q, m) = d(p, q).
(4.1) F(p, t) = p + tN(p), b'(p, t) E S x Il8,

where N S - S2 C ][83 is a Gauss map of S. This map F is clearly Hints for solving
differentiable, in the sense of Exercises (14)-(19) at the end of Chapter 2, the exercises
and takes each pair formed by a point p of the surface and a real number t
to the point at distance t on the normal line of S at p, on the side of the
surface to which N(p) points. Hence, it is clear that Exercise 7.6: For each x E O, the map (dF) : R3 is orthogonal.
That is,
F(S x (-b, S)) = N5(S) = U Na(p), V6> 0.
PES ((dF)(u), (dF)(v)) _ (u, v), b'x E O du,v E ]E83.
Now, note that the normal segments of radius 6 not touching each other Choosing u, v as vectors of the canonical basis of ][83, one has the equality
is equivalent to the map F being injective on S x (-6, 6). In fact, we will of functions
require something more of the neighbourhoods that we are looking for. (DF aF
= bid, i, j = 1, 2, 3,
axi ax j
Definition 4.24 (Thbular neighbourhood). The union N5(S) of all the where x1, x2, x3 represent the coordinates of R3. Taking derivatives with
normal segments of radius S > 0 centred at the points of an orientable respect to xk, k = 1, 2, 3, one obtains
surface S is called a tubular neighbourhood of radius 8> 0 if it is open as
a subset of ][83 and the map F : S x (-3, 3) - N5(S) defined in (4.1) is a a2F aF \ a2F aF
= 0, i, j,k = 1, 2, 3.
-I-
diffeomorphism. 8xk8x2' axe axe axe
This equality and the equality of the mixed second derivatives imply that
We now will show that, under suitable hypotheses, these special neigh-
bourhoods exist. a2F aF
,
O xi axe Oxk
Lemma 4.25. For each point p e S of a surface S there exist an open
orientable neighbourhood V and a real number 6> 0 such that the set N5(V) is symmetric in i, j and skew-symmetric in j, k. Hence
is a tubular neighbourhood of V. Gij _ -Gi j = -G ij = G ji = G i = -Gk _ -Gig
 Exercises 241 122 4. Separation and Orientability

(15) Show that if a straight line R is contained in a surface S, then there is Proof. The differential of the map F : S x Il8 -> ][83 at a pair (p, t) E S x ][8
a geodesic of S having R as its trace. is given by
(16) Let a : I -* S be a geodesic in an oriented surface S such that (4.2) (dF)(,t)(v, 0) = v + t(dN)p(v), v e TPS,
o'«(t)(a'(t), ap(t)) = 0 (dF)(,t)(0, 1) = N(p).
for each t E I. Prove that a is a segment of a straight line. Thus, if we set t = 0, it follows that
(17) T Let BE(p) = be a geodesic ball on a surface S. Let (dF)(,o)(v,0) = v and (dF)(,o)(0, 1) = N(p)
a : I - S1 C TS be a parametrization b.a.l. of the unit circle of TS. for each v e TPS. Thus, (dF)(,o) is a linear isomorphism from TS x Il8
We define a map F : (-e, e) x I -p S by
to ][83. By the inverse function theorem-see Exercise (19) at the end of
F(t, s) = exp(ta(s)). Chapter 2-there exist a neighbourhood V of p in S and a number 3> 0
such that
Show that the Gauss curvature K of S satisfies
aF aF
FIvx(-a,a) : V x (-S, 3) -f F(V x (-b, b)) = N5(V)
as tt
+(KoF) as
=0,
is a diffeomorphism. We conclude the proof by appealing to the fact that
any surface is locally orientable; see Lemma 3.2. O
for all 0 < t <e and all s E I. (See Exercise (12) of Chapter 3.)
(18) T Minding's theorem. Let S and S' be two surfaces with the same con- Assume now that the surface S is orientable and that 1Z C S is a rel-
stant Gauss curvature and let p E S and p' E S' be any two points such atively compact open subset of S. Using the corresponding cover of 1Z by
that B5 (p) = expP(Ba(Op)) and B5 (p') = are geodesic open sets given in Lemma 4.25, we can extract a subcover and choose, as
balls. Prove that a common radius, that of the smallest of the tubular neighbourhoods of
f = expP, o cbo expp 1 the different sets of the cover. This idea leads us to the following existence
result.
is an isometry between the two geodesic balls, where ¢ : TS -p TS' is
any linear isometry. Hence, two surfaces with the same constant Gauss Theorem 4.26 (Existence of tubular neighbourhoods). Let S be an ori-
curvature are locally isometric. entable surface and let R. C S be a relatively compact open subset. Then
(19) T Bertrand-Puiseux and Diquet formulas (18.8). Let BE(p) be a geo- there exists a number > 0 such that the set NE(R) consisting of all the seg-
desic ball on a surface S. For each t E (0, e), we call the curve expP(S1(t))
ments of normal lines centred at the points of R. with radius e is a tubular
the geodesic circle of radius t and centre p. Analogously, we call the sub-
neighbourhood of the surface R; that is, it is an open subset of Il83 and the
set of S given by expp(Bt(0)) the geodesic disc of radius t and centre map F : R. x (-i, e) -* NE(R), given by
p. Let Lt and At, respectively, denote the length and the area of the F(p, t) = p +tN(p), (p,t) E R. x
geodesic circle and the geodesic disc of radius t centred at p. Show that
is a diffeomorphism. In particular, when S is compact, there exists a num-
Lt ber > 0 such that the metric neighbourhood BS(S) =NH(S) is a tubular
K (p)
p - t->o 3 neighborhood of S.
'
rt3

K(p)
l22_- t Proof. By the compactness of 1Z, it can be covered by finitely many open
-to subsets of S, each of them having a corresponding tubular neighbourhood, as
From this, compute the limit
From proved in Lemma 4.25. Let 8> 0 be the smallest radius among them. Note
that, in this situation, the map F, defined in (4.1), restricted to R. x (-8, 6)
4-rAt - Lt is a local diffeomorphism. Let us see that, moreover, we can find a number
lim
t-'O r 2t4 e (0, 8) such that the map F restricted to R. x (-i, ) is injective, i.e.,
and give an interpretation of its sign. such that the segments N6 (p) of normal lines, with p e 1Z, do not intersect
4.5. Tubular neighbourhoods
240 7. Intrinsic Geometry of Surfaces
123

each other. Suppose that this is not the case. Then, for each n e N, there where N is a Gauss map of S and prove that the function
would be points pn, qn E S with pn qn and p E S' ' ((dV)p(ei), e2) - ((dV)(e2), ei),
Nl/n(pn) n Nl,in(gn) 0. where {ei, e2} is any orthonormal basis of TPS, is differentiable and has
zero integral on S.
Since 1Z is compact, there are subsequences of {Pn}N and of {qfl}flE that
(8) T Let f, g : S - ][83 be two differentiable functions on a surface S. Show
converge in 1Z. By restricting ourselves to these subsequences, we can assume
that the subset of S defined by
that there are points p, q e 1Z such that
lim pn = p and
{p E S I= g(p) and (df)_ (dg)p}
n- too
lim qn = q.
n- too is closed.
On the other hand, let rn e N1/n (pn) f1 Nl/n (qn) . Then (9) T Use the previous exercise and the existence of geodesic balls on a
pn gn - pn - rn + rn - gn I < n11
< _2
+n n ,
surface to prove that if two local isometries f, g : S -p S' between two
surfaces, where S is connected, coincide along with their differentials at
and the two limits p and q coincide. Now we apply Lemma 4.25 to the point a point, then they coincide everywhere.
p = q e S and obtain an open neighbourhood V of p = q in S and a real (10) Darboux trihedron and Darboux equations. Suppose that S is an ori-
number p > 0 such that NP (V) is a tubular neighbourhood. But there is entable surface and N : S -* S2 is a Gauss map on S. If a : I -p S is a
also an N E N such that geodesic of S p.b.a.l. defined on an interval of ][8, we define three maps
pn,gnEV, T, N, B : I -* ][83 by
n>N - 1 /n < p. T(s) =as(s), N(s) _ (N o a)(s), and B(s) = T(s) A N(s).
Therefore, we would have the contradiction Then T, N, and B are differentiable and, moreover, {T(s), N(s), B(s)}
Nl/n(pn) n C N,,(pn) (1 NP(g1) _ 0, form an orthonormal basis of ][83 for all s E I with T(s), B(s) E T«(3)5
and N(s) 1 Ta(s)S. Prove the following:
since NP (V) is a tubular neighbourhood and thus F restricted to V x (-p, p)
must be injective. In conclusion, we can find a positive number such that ( T'(s) = (Ta(s)(T(5),T(5))N(5),
F : 1Z x (-E, E) -f N6 (R.)
N'(s) _ -a«(s)(T(S),T(S))T(S) - a«(s)(T(S), B(S)) B(S),
B'(S) _ a«(s)(T(S), B(S)) N(S),
is injective and is a local diffeomorphism. Since it is obviously surjective, where v is the second fundamental form of S corresponding to N.
the proof is complete. 0
(11) T Find all the traces of the geodesics joining two given points of a plane,
Remark 4.27. One of the consequences that can be deduced from the of a sphere, and of a right circular cylinder. What can you say about
existence of tubular neighbourhoods for compact surfaces is that the sets of their number? Deduce that, in the case of the unit sphere, a geodesic of
points at a distance less than a certain > 0 from the surface just the length less than it always minimizes the length between its ends.
number supplied by Theorem 4.26 are topological products of the surface (12) T If all the geodesics of a connected surface are plane curves, show that
and an open interval of R. In particular, the metric neighbourhoods of a the surface is contained either in a plane or in a sphere.
compact surface with small enough radius have the same homotopy type as
the surface. (13) T Show that each meridian-generating curve-of a surface of revolution
is a geodesic. On the other hand, show that a parallel is a geodesic if
Remark 4.28. If S is a compact surface, SZ is the inner domain, and NE(S), and only if it is at a critical distance from the axis of revolution.
with > 0, is a tubular neighbourhood of S, then the set (14) T Clairaut's relation. Let a : I - S be a geodesic p.b.a.l. on a surface of
VA(S) = F(S x (0, E)) C NH(S) revolution whose axis is a line R. Represent by r and p the two functions
that take each t E I, respectively, to the distance from a(t) to the axis
is connected and does not intersect S. If the orientation that we are using R and the cosine of the angle made by a at the time t with the parallel
for S is derived from the inner normal field, then it is clear that VE(S) C S2.
through a(t). Prove that r(t)p(t) is a constant function.
Exercises 239
124 4. Separation and Orientability

Exercises
(1) T Show that the group of isometries of a sphere coincides with the sub-
group of the group of rigid motions of ][83 fixing its centre and is isomor-
phic to the three-dimensional orthogonal group.
(2) T Find a local isometry between the helicoid-defined in Example 3.46-
and the catenoid, which was introduced-see Exercise (9) in Chapter
3-as the surface of revolution whose generating curve is a catenary.
(3) What can you say about a compact surface S whose Gauss map N
S - S2 is an isometry? Figure 4.7. Parallel surfaces
(4) Consider the two following differentiable maps X and X':
Remark 4.29 (Parallel surfaces). Suppose again that for some e > 0, NE(S)
X (u, v) _ (ucosv,usinv,logu), b(u, v) E ][8+ x (O,2ir), is a tubular neighbourhood of a compact surface S with a fixed Gauss map
X'(u, v) _ (ucosv,usinv,v), d(u, v) E ][82. N. Since the map F : S x (-e,e) - NE(S) defined in (4.1) is a diffeomor-
phism, for all t E (-E,e), the map Ft : S ---> Ne(S) C 1, defined by
Prove that S = X(][8+ x (0, 2ir)) and S' =X'(][82) are surfaces and that
X and X' are parametrizations for each of them. Show that the map Ft(p) F(p, t) = p + tN(p),
f : S - S' given by f = X' o X-1 satisfies K = K' o f but it is not a
local isometry. is a homeomorphism onto its image. On the other hand, if p c S and
(5) T Let S be a connected surface. For each p, q E S, we define el, e2 E TpS are principal directions of S at p, we have
(dFt)p(ei) = ei + t(dN)p(e$) = (1 - tki(p))ei, i = 1, 2.
d(p, q) = inf L(y),
Now then, from (4.2), we see that Ft has an injective differential at each
where I'p,q is the set of piecewise differentiable curves on S joining p point, for t E (-e, e). Hence, using Exercise (9) at the end of Chapter 2, the
with q. Prove that d : S x S -p ]ES is a distance (called the Riemannian image Ft (S), which we will represent by St, is a surface and Ft : S - St is
or intrinsic distance). What relation does it have with the standard a diffeomorphism for each t, ItE C E. We will call the surface St the parallel
Euclidean distance? Also prove that, if S is closed in ]E83, then the surface of S at an oriented distance t. In Exercise (16) at the end of Chapter
metric space (S, d) is complete and that any isometry between surfaces 3, the reader has already seen some useful geometrical properties of these
is an isometry for this distance. surfaces.
(6) T Let S be a compact connected surface, and let f : S x [0,1] - ][83 be
Remark 4.30. When S is a non-compact surface that is a closed subset of
a differentiable map such that, for each t E [0, 1], St = f(S x {t}) is a
Euclidean space, there also exist tubular neighbourhoods, although they do
surface and the map ft : S - St, given by ft(p) = f(p, t) for all p E S,
not necessarily have a constant radius. In fact, it can be shown that there
is an isometry with fo = I. Suppose that vt is the second fundamental
is a positive continuous function e : S -} lR and that the map
form of St relative to the inner normal of St. Prove that, if K(p) > 0 at
a point p E S and if Qp is positive definite, then v ft(P) is positive definite F:E(S,E) CSxR--->R3
for all t E [0, 1].
defined in (4.1) is a diffeomorphism onto its image, where E(S, E) denotes
(7) Let S be a compact surface and let V : S - ][83 be a differentiable vector the open subset of S x IR given by
field. Calculate the divergence of the field JV defined by
E(S, e) = {(p, t) E S x IR Itl < E(p)j.
JV=VnN,
I
Exercises 125
Figure 4.8. Tubular neighbourhood of a closed surface
Exercises
(1) T Let S be an orientable surface and let No(S) be a tubular neighbour-
hood, with 6 > 0. Let it and r, respectively, represent the first and
second components of the map
F-1 : Ns(S) S x (-6 6).
They are called projection and the oriented distance, respectively. Study
their properties and compute their differentials.
(2) T Let S be a compact connected surface, f its inner domain, and p
a point of S. Show that there is a sphere §r, r > 0, contained in n,
tangent to S at p with S f1 Sr = {p}.
(3) Suppose that S is a surface and Ne(S) is a tubular neighbourhood.
Prove that, if f : S -+ (-p, p) C Ifs is a differentiable function, then
Sf {x E Na(S) I r(x) = (f o ir)(x)} is a surface by checking that 0 is
a regular value of r - f o 7r.
(4) T Show that, if S is a compact surface, there are points of S whose
second fundamental form relative to the inner normal field is positive
definite.
(5) T Let S be a compact surface, and let P C S be the subset of S formed by
the points with non-negative Gauss curvature. Prove that the restriction
of the Gauss map of S to P is surjective. Is it also true if one takes the
set of points with positive Gauss curvature?
(6) The Jordan-Brouwer and Brouwer-Samelson theorems for closed sur-
faces. Let S be a connected surface closed as a subset of R', and let
B = B,(pe) C JR3 - S be an open ball of radius e > 0. Show the
following.
238 7. Intrinsic Geometry of Surfaces
a I [c,d] has the same trace as a radial geodesic starting from a(c). Since this
occurs for any subinterval with length less than 8, the proof is complete. D
We finally make sure that, on any surface closed as a subset of ][83, we
can join any two points by means of a curve that minimizes the length,
which will necessarily be a geodesic according to Corollary 7.22.
Theorem 7.43 (Hopf-Rinow's theorem). Let S be a connected surface closed
in ]E83 and let p, q E S be two points of S. Then there exists a geodesic
ry : [a, b] - S with -y(a) = p and y(b) = q and length less than or equal to
the length of any piecewise differentiable curve joining the two points.
Proof. The result is obvious when p = q. Suppose then that p ; q and take
any two numbers a # b. Let I'p,Q be the set of all the piecewise differentiable
curves ry defined on [a, b] taking values on S and with -y(a) = p and y(b) = q.
Let d(p, q) (see Exercise (5) of this chapter) be the number
d(p, q) = inf Ly).
This infimum exists since all these lengths are non-negative numbers and,
moreover, I'P,Q is non-empty-one can always join two points in a connected
surface by means of differentiable curves. Let {7n}nEN be a sequence in I'p,Q
such that
lim L(y) = d(p, q).
n-' 00
We will consider all the curves in this sequence to be parametrized propor-
tionally to the arc length on each regular piece, that is, h'(t)I = Cn for each
t E [a,b]. Then
L) = a
b
I(t)I dt = (b - a)C.
In particular {Cfl}flEN is a bounded sequence, since it converges. Thus, if
t1, t2 E [a, b], we have
t2
Vy(t1) - `Yn(t2)I yn (t) dt cnIti - t21,
Jtl
for each n E N. By the Ascoli-Arzela theorem, there is a subsequence, which
we denote in the same way, that converges uniformly to a Lipschitz curve
y : [a, b] -* R3. Since S is closed in R3, the curve y has its image on S and
b
I'(t)I dt < (b - a) n moo Cn n moo d(1>> 4)
a
Therefore y has length less than or equal to any other piecewise differentiable
curve joining p and q. We complete the proof by applying Theorem 7.42. 0
7.6. The exponential map 237
Proof. Let (p, v) E TS, with v 0, and suppose that the positive number
e(p, v) given by Theorem 7.30 is finite. We choose a sequence {tom}flEN of
points of (-E(p, v), E(p, v)) tending to e(p, v). This is a Cauchy sequence
and, since
tom,
VY(tn)'Y(tm)l= I f 1''(t)dtl < lVlltntmI,
t,,,
the sequence {'y(tfl)}flEN of points of S is a Cauchy sequence as well. Hence,
it converges to a point q of ][83 that, since S is closed, must be a point of S.
Consider a normal neighbourhood V of this point q E S with radius e> 0,
whose existence is given by Proposition 7.40. Since ry is continuous, we know
that the image through -y of an interval of the form (S, E(p, v)) is in V for
some S > 0. Choosing to E (S, e(p, v)) so that e(p, v) - to <e and putting
r = 'y(to) E V, we have that 'y(t) = exp,.((t - to)u) for each t E (S,e(p,v))
and where u E TrS, lul _ lvi. This shows that ry extends on the right-hand
side beyond E(p, v). Therefore e(p, v) _ +oo for all (p, v) E TS and thus
ryP,v is defined on the whole real line R. As a consequence, 1 < e(p, v) for all
(p, v) E TS and so (p, v) E El. It follows that the domain of the exponential
map is all of TS.
A second consequence of Proposition 7.40 on the existence of normal
neighbourhoods will be to prove that all curves, not necessarily differen-
tiable, minimizing the length between their endpoints have the same trace
as regular geodesics. In other words, geodesics minimize the length in a much
wider class than differentiable curves. This can be interpreted to mean that
minimizing the length forces the curve to be smooth.
Theorem 7.42. Let a : [a, b] - S be a Lipschitz curve joining two points
p = a(a) and q = a(b) of a surface S whose length is less than or equal to
that of any piecewise differentiable curve joining them. Then there exists a
geodesic and, thus, a differentiable curve whose trace coincides with that of
the curve a.
Proof. For each t E [a, b], there exists, by Proposition 7.40, a normal neigh-
bourhood in S of the point a(t), which will be denoted by V. Since a is
continuous, each reciprocal image a1 (Vi) is an open subset of the compact
interval [a, b]. Let S > 0 be the Lebesgue number of this cover. Then, if
[c, d] C [a, b] is a subinterval with length less than S, its image a ([c, d]) is
contained in at least one normal neighbourhood and hence it is in a geodesic
ball centred at a(c). Then the curve minimizes the length between
its endpoints since, if not, there would exist a piecewise differentiable curve
/3 defined on [c, al] joining them with a shorter length and, then, replacing
d] by ,(i, we would obtain another piecewise differentiable curve shorter
than a joining a(a) and a(b). Thus, by Proposition 7.37, the restriction
126 4. Separation and Orientability
a: For each pl E I183 - S one can find a point p2 E B such that the
segment [P1 ,P2] cuts S transversely.
b: If p3 is another point of B such that [pl, p3] cuts S transversely as
well, then there is an even number of points in
([P1,p21 U [P1,P3J) fl S.
c: I183 - S is not connected. As a consequence, show that the Brouwer
separation theorem extends to closed surfaces.
Now, observe that the proof of the Brouwer-Samelson Theorem 4.21 is
valid in this case without any changes.
(7) T Let S be a compact connected surface and let St be the inner domain
determined by S. Suppose that a straight line R intersects St only
at a point p. Prove that p e S. Using the restriction to R of the
oriented distance function in a tubular neighbourhood of S, show that
R is tangent to S at p.
(8) T Let S be a compact connected surface with positive mean curvature
everywhere and let 1 2 be the inner domain determined by S. Take q e St.
Show that, if p e S is the point of S closest to q, then q = p + tN(p),
where N(p) is the inner normal and 0 < t < 1/k2(p).
(9) T Let S be a compact connected surface and let St be its inner domain.
Prove that if x St, then there are at least two lines tangent to S passing
through x.
(10) T Prove that any compact surface can be expressed as the inverse image
of a regular value of a differentiable function defined on II83.
(11) T Prove that any surface has Lebesgue measure zero in I[83.
Hints for solving
the exercises
Exercise 4.11: There exist, by definition, affine parametrizations an (t) _
qn + twn, t e Ilk, of each line Rn, n e N, such that
lim an(t) = a(t), t e Ilk,
n-400
where a(t) = q + tw. Hence q e R is the limit of w 0 is the
limit of {W}EN, and it is a direction vector of R. Since pn e Rn and vn is
a direction vector of Rn, we have
pn = qn + town and vn n wn , n e N,
Hints for solving the exercises 127
for some to e Ilk. It follows from these expressions that
to = (pn - qn,2 wn) and An =
(yn, wn )
n E N,
Iwni Iwni 2
and, thus, the sequences {tn}fEN and {Afl}flEN are convergent to real num-
bers t and A. Therefore, taking limits in the first two equalities, we get
p = q + tw and v = Aw,
and thus p e R and v is a direction vector of R.
Exercise 4.12: Let K be a compact subset of that intersects all the lines
of the sequence {Rn}fEN. We choose, for each n e N, a point pn e Rn f1 K
and a direction unit vector vn of R. By the compactness of K and of
the unit sphere in there exist subsequences of the sequences {Pn}nEN
and {vfl}flEN, respectively, which we will still denote by the same symbols,
converging, respectively, to p e K and v e §2. Then
lim (pn + tvn) = p + tv, bt E R.
n-400
Consequently, {Rfl}flEN converges to the line passing through p in the direc-
tion of v.
Exercise 4.13: By definition, one can find a sequence {vn}fEN of direction
vectors of the lines Rn converging to a direction vector v of R. Moreover,
we may suppose, forgetting finitely many points if necessary, that the entire
sequence {Pn}nEN is contained in an open neighbourhood of p where we have
at our disposal a Gauss map N. Then, it is enough to take limits in the
equality (vn, N (p)) = 0, which is valid for each n E N.
Exercise (1): If N5(S) is a tubular neighbourhood of S, each point x e
Nb(S) is in a unique normal segment of radius S. Denote the point of S
on which this normal segment lies by mi(x) E S and by r(x) E (-6, 6) the
distance from the point x to its projection on S, if x is on the side of S
toward which N(ir(x)) points, and, otherwise, the opposite of this distance.
These two maps are the components of F-1. Both are differentiable since
- pl ° Fsx(-s,a) and r = P2°
where pl and p2 are the projections of the product S x (-S, 8) onto each of
the two factors. The following statement summarizes the most important
properties of ir and r.
Let S be an oriented surface, with Na(S), 5 > 0, a tubular neighbour-
hood of S, and ir N5(S) -* S, r N5(S) -* (-5, S) the corresponding
projection and the oriented distance. Then the following hold.
A: r({0}) = S.
236 7. Intrinsic Geometry of Surfaces
relatively compact open neighbourhood of qo in U and p is a positive number
less than b, such that
E:U'xBp-SxS
is a diffeomorphism onto its image, which is an open neighbourhood of
(P0, po) in S x S. Since Iand Iare bounded on U', we can
choose two positive numbers ri <p and T such that
(dX)q(Bij) C C (dX)q(Bp),
for each q E U'. Then, we can find a neighbourhood W of po in S such that
WxWCE(U'xB).
A first conclusion is that, if q E U', the map
E : {q} x BP -{X (q) } x S
is a diffeomorphism onto its image, that is,
expX(4) : (dX)q(Bp) S
is also a diffeomorphism onto its image. Hence, the open ball BT (OX(Q))
centred at the origin of TX(Q)S, which is contained in (dX)y(Bp), has as its
image a geodesic ball BT (X (q)) in S. A second conclusion is that, if pi and
p2 are points of W, one has
pi=X(q) and p2 = expX(a)(dX)e('w),
for some q E U' and w E ][82 with wI < Ii. Therefore
W C expel [(dX)4(B)] C B,-(pi),
which is true for all pi E W. We have just proved the following improvement
of Proposition 7.35.
Proposition 7.40 (Existence of normal neighbourhoods). Each point of a
surface S has an open neighbourhood that is contained in a geodesic ball cen-
tred at each of its points, with a common radius e> 0. Such a neighbourhood
will be called a normal neighbourhood of radius e.
The first consequence of this result will be an answer to the second
question that we posed after Example 7.28: when can we be assured that
the geodesics of a surface extend indefinitely on it? Or, equivalently, what
can we say about the positive number e(p, v) > 0 that we associated to each
(p, v) E TS in the existence and uniqueness Theorem 7.30?
Corollary 7.41. If S is a surface that is closed as a subset of Euclidean
space, then the geodesic ry = with initial conditions (p, v) E TS is defined
on the whole of R. Thus, the exponential map of S is defined on the whole
of TS.
7.6. The exponential map 235 128 4. Separation and Orientability

Exercise 7.39. Give an example of a geodesic on a surface having length B: If 0 < p < S, then Np(S) is another tubular neighbourhood of S.
greater than another differentiable curve with the same endpoints. C: Reversing the orientation of S, the projection ir remains unchanged
and the oriented distance r changes its sign.
Let po E S be a point of the surface, let V be an open neighbourhood of
this point in S, and let e> 0 be a number such that the exponential map D: If p E S, el, e2 E TS are principal directions of S at p, and t E (-8, S),
exp is defined on T (V, e); see Remark 7.34. Consider now a parametrization we have that 1 - tk(p) > 0 and
X : U -p S such that po = X (qo) E X (U) C V and a real number 8> 0 such
that, if w E I[82 and wI <8, then I<e, that is, (dX)q(w) E T(V,E) (d1r)P+tN() (e)
- 1 -t z(p)ez, Z = 1, 2, (d1r)+tN()(N(p)) _ 0,
for each q E U. From this, define a map E : U x Bb C II82 x ][82 -S x S by (dr)+N()(e) _ 0, 2 = 1, 2, (dr)P+tN(P)(N(p)) = 1.
E(q, w) _ (X(q),expx(q)(dX)q(w)), `d(q, w) E U x Bb. Proof. It is clear that r-1({0}) = S from definition (4.1). Furthermore,
This map is differentiable since its first component is obviously differentiable the two assertions in C about the changes that occur to the projection and
and the second one can be written as the oriented distance when we reverse the orientation of the surface are
immediate from the corresponding definitions. The same happens for B.
expX(e)(dX)e(w) = exp (X (4), (dX)q(w)) = exp (X (4), wiXu(9)+w2Xv(9')), Now, take a curve a : I -f S, where I is an interval of I1 8 containing the
for each q E U and each w = (wi, w2) E ]E82 with Iwl <8. Clearly (qo, 0) E origin and such that a(0) = p and a'(0) = e, i = 1, 2. We define another
U x Bb and curve ,6 in II83 by
E(4o,O) _ (X(qo),expx(q0)Op) _ (po,po). ,6(s) = F(a(s), t) = a(s) + tN(a(s)), s e I.
We next compute the differential of E at this point in its domain. This Then this new curve ,6 belongs to the open set N5(S) and
differential is a linear map
3(0) = p + tN(p) and ,3'(0) _ (1 - tkz(p))ei = (dF)(,t) (e, 0),
(dE)(q0,o) : ][82 x ][82 - T0S x Tao S. ir(,3(s)) = a(s) and r(,Q(s)) = t, s E I.
Take w E ]E82. Then Thus, since F is a diffeomorphism when restricted to S x (-6, S), we have
d that (dF)(,t) (e, 0) is a non-null vector, that is, 1 - tkz(p) 0. But (-6, 5)
(dE)(q0,o)(W,O) _ E(qo +tw, 0) is connected and 1 - tki(p) is positive for t = 0. Thus 1 - tki(p) > 0 on
dt t=o S x (-S, 8). Moreover,
d
(X(qo + tw), expx(go+tw)OX(go+tw)) - tlCi(p)]ei) - 'c(p(s)) _ « (o) - ei.
dt t=o ds s=0
Hence By the linearity of the differential and the fact that 1- tk(p)
(p) 0, we obtain
(dE)(q0,o)(w,O) - ((dX)qo(w), (dX)qo(w)), b'w E l[82. (d)+tN (e) - 1 - tki(p) ei, i = 1, 2.
On the other hand,
Furthermore,
(dE)(q0,o)(O,w) = dt E(9o,t'w)
= dt (po,expp0t(dX)qo(w)), (dr)+N()(e) - ds Is_o r(Q(s)) =0.
I t_o I
t=o
and, consequently, Similarly, if we consider the segment ry of the normal line in the tubular
neighbourhood Na(S), given by
(dE)(q0,o)(O,w) = (0, (dX)qo(w)), ew E R2.
ry(s) = F(p, t + s) = p + (t + s)N(p), s E I,
Since the differential (dX)q of the parametrization is injective at each q E
U, the differential (dE)(q0,o) is a linear isomorphism. Applying the inverse we clearly have that
function theorem, we have the existence of an open neighbourhood of (qo, 0) 7(0) = p + tN(p) and ry'(0) = N(p),
in U x 13, which can be assumed to be of the form U' x Bp, where U'is a ir(ry(s)) = p and r(ry(s)) = s + t, s E I.
Hints for solving the exercises 129 234 7. Intrinsic Geometry of Surfaces

Taking derivatives at s = 0 in the two last equalities, we get Proof. First, it is worth noting that a Lipschitz function is absolutely con-
(d'1r)+tN()(N(p)) - O and (dr)P+tN(P)(N(p)) - 1, tinuous and, thus, it has a bounded derivative almost everywhere. This is
why it makes sense to talk about the arc length of Lipschitz curves using
as required. D the same definition as in the differentiable case. With this established, if the
Exercise (2): If t > 0 is small, the point pt = p + tN(p), where N is the curve a passes again through the centre of the geodesic ball, for example,
inner normal of S, belongs to St and, according to Exercise (1) at the end if a(c) = p for some c E (a, b], then the radial geodesic with which we may
of Chapter 3, p is an isolated local minimum of the square of the distance compare the arc c]
is the constant curve p. That is, the proof can be
function to pt. Then there exists an open neighbourhood V of p in S such reduced to the case where a(t) p for each t E (a, b]. Then we have that
that V is outside the open ball Bt with centre pt and radius t and such that a(b) # p. Thus a(b) = expp(v) with 0 < lvi <S and
V fl St = {p}, where St is the corresponding sphere. Since St tends to the a(t) = expp(/3(t)), t E [a, b],
point p as t goes to zero and S - V is a compact set that does not contain where /3: [a, b] -p TS is a Lipschitz curve in the tangent plane of S at p
p, we have S fl Bt = Ql for small t. Since Bt is connected, either Bt C 1 2 or with /3(a) = Op, Q(t) # for each t E (a, b], and ,(i(b) = v, and, hence,
Bt C IE83 - S2. The correct choice is the first one, since pt E Bt fl 12. Thus,
/' b
St C Bt C SZ and St fl S = {p}.
La(a) = I a t)I dt
a
Exercise (4): We will show that any point p of S where the square of the
distance function to a fixed point a e R3 reaches a maximum must have
positive definite second fundamental form with respect to the inner normal
= f a
b i p ° fi)'(t)i dt = f Ib (dexp)(t)(fi'(t)) I dt.
a
field. Indeed, all the points of the half-line starting at a and passing through If we choose w(t) E TS so that together with /3(t)/i/3(t)l they form an
p that are beyond p cannot belong to S. That is, orthonormal basis, we have, for t E (a, b] where the derivative of ,6 exists,
p + t (p - a) bt > 0.
8, that
Since this set is not bounded, we have that p + t (p - a) E SZ for each + (fi'(t),w(t))w(t).
I

t > 0. On the other hand, we know, from Example 3.39, that for some Therefore, using Lemma 7.36,

l(dexpP)Q(t)(Q'(t))i (fi'(t), /3(t)) I,

a = p + AN(p) and vJ2 - AQp(v, v) < 0, v E TpS. /3(t)
Hence, we have where equality is attained for t E (a, b] if and only if ,3'(t) is proportional to
fi(t). Consequently,
p + t(p - a) = p - tAN(p) E ][83 - St, t>0.
f
b
Thus, if N is the inner normal, tA > 0 for each t> 0 small enough, then
A _ lp - al > 0 and consequently La(a) > Q(t)i
(t)t dt

f /3(t)lt)ttl = [lfi(t)l]t=a = lvi =

b
lvi 2, v E TS.
lp 1 ai

Exercise (5): Let a E 2, and let q be a point of S where the height function
h (p) _ (p, a), p E S, relative to a fixed plane perpendicular to the vector
a, attains its minimum. By Proposition 3.35 and Example 3.36, the second
fundamental form of S at q, relative to any choice of normal, is semi-definite
and, thus, K(q) > 0. That is, q e P. Now it suffices to show that N(q) = a,
where N is the inner normal of S. Since q is a critical point for h, part A of
Example 2.59 implies that N(q) _ ±a. On the other hand, since 8 is over
If equality occurs, Q'(t) is proportional to fi(t) for each t E (a, b] where this
derivative exists. Thus, the trace of ,(i(t) is contained in the line starting
from Op in TS and passing through v. Hence, the curve a = expp o /3 has
the same trace as the geodesic ry. O
Exercise 7.38. T If p E S2 is a point of the unit sphere, prove that 2-{-p}
is a geodesic ball centred at p. What is its radius? Use Proposition 5.8 to
obtain a formula for integrating functions on 2.
7.6. The exponential map 233 130 4. Separation and Orientability

for s E I and ti < 1 + S. Using the fact from calculus that mixed second the plane passing through q perpendicularly to a, St is over this plane as well.
derivatives in two variables are equal, we have Then, since q+tN(q) E St, we have h(q+tN(q)) = h(q)+t(N(q),a) > h(q)
fort > 0 small. It follows that (N(q), a) > 0 and, thus, N(q) = a.
The final answer is negative. One can see this by studying the region of
at2
the elliptic points in the torus of revolution, as proposed in Exercise (8) at
Taking into account that the end of Chapter 3.
2

8t2 (t' S) =1'"(p, a(s), t)

Exercise (7): Since St is an open subset of ][83, if p e St, Rf1St would be open
has no component tangent to the surface, since -y(p, a(s), t) is a geodesic, in R and contained in Rfl St = {p}. Hence p e S2 - St = S. Let E > 0 be such
and that (aF/as)(t, s) is tangent to S, we obtain that NE(S) is a tubular neighbourhood of S and v e ][83 is a direction vector
8 8F aF of the line R. For t close to 0, one has that p + tv e NE (S) . Since R touches
at as ' at - SZ only at p, taking Exercise (1) into account, we have r(p+tv) <0 for each
t 0 small. That is, there is a maximum of the function t H r(p + tv) at
that is t = 0. Therefore, (dr)(v) = 0 and, using Exercise (1) again, we deduce that
aF aF (N(p),v) =0.
as' at
is constant in the variable t. But, since
Exercise (8): Taking Example 3.39 into account, we know that q = p +
(t,s) = dsexpp(ta(s)) = tN(p) for some t e ll8 and that 1 - tk2(p) > 0. Since k2(p) > H(p) > 0, it
t(dexpP)t«(3)(a/(s)),

as follows that t < 1/k2(p). It only remains to show that t > 0. Since p is the
we have that point of S closest to q e St, the segment ]p, q] does not intersect S. This
t((dexpp)t«(s)(a(s)), (dexpp)ta(s)(a"(s))) same segment is connected and does cut 12. Hence, ]p, q] C St. If t < 0, then
does not depend on t. Computing this expression for t = 1 and s = 0 and, p + AN(p) E St for A E [t, 0), which is not true according to Remark 4.22.
after that, for t = 0 and s = 0, it follows that Thus t > 0.

0,
which is the second required equality.
Therefore, at any point of the surface, the exponential map preserves the
length of radial curves starting from the origin of the tangent plane and the
perpendicularity between these radii and the images of the circles centred at
this origin. This property, discovered by Gauss in his Disquisitiones, implies
the following result on the local minimization of the length, which is a local
converse of Corollary 7.22.
Proposition 7.37. Let B5(p) be a geodesic ball of radius 6> 0 with centre
at a point p of a surface S. If a : [a, b] - B5(p) C S is a Lipschitz curve
with a(a) = p, then
La(a) La(y),
where y is the only radial geodesic p. b. a.l. defined on [a, b] joining p with
a(b). If equality holds, then the traces of a and of y coincide.
Exercise (9): Consider the support function f based on the point x relative
to the inner normal of S, as was defined in Exercise (4) of Chapter 3. Since
S is compact, we can find two different points p, q e S for which the distance
to the point x reaches a maximum and a minimum, respectively. If we argue
in a way similar to Exercise (8) above, we can see, using that x 12, that
f(°) _ -Ip-x <0 and that f(q) _ Iq - x > 0. Since S is connected, there
must exist a point r E S where 0 = 1(r) _ (r - x, N(r)). Furthermore, one
can draw a continuous curve joining p and q in S avoiding the point r. In
this way, we obtain another point, different from r, where f vanishes. Now,
it is enough to prove that any point where the support function vanishes
supplies a point from which we can draw a tangent line passing through x.
Exercise (10): We take e > 0 so that NE(S) is a tubular neighbourhood
and a differentiable function h : I[8 - I[8 such that h(t) _ -e for each t < -E
and h(t) = e for each t > E, and which is increasing on the interval (-e, E).
4.6. Appendix: Proof of Sard's theorem 131 232 7. Intrinsic Geometry of Surfaces

Then we define f : I183 -f I1 8 by This image, which we will denote by Ba(p), will be called the geodesic
ball with centre p and radius b. It should be noted that, if Ba(p) is a geodesic
(hor)(x) if X e NE(S),
ball, then the exponential map expP : Bb(Op) -> S is a parametrization that
f(x) = e if X C -Ne(S),
exactly covers this open subset of the surface. Moreover, if q E Ba (p) is a
-e otherwise,
point of the geodesic ball with centre p and radius b, any geodesic joining p
where r is the corresponding oriented distance on the tubular neighbour- with q inside of Ba(p) has to be a homothetic reparametrization of the curve
hood, studied in Exercise (1) above, and St is the domain determined by S. where v = (expP)-1(q), which is called the radial geodesic joining
This definition and the choice of h force f to be differentiable. Moreover, it p with q. All these assertions follow from the fact that the differential of the
is clear that f1({O}) C NE (S) and, thus, exponential at 0 E TpS is the identity map for each p E S. We will now
11({o}) = r-1(h-1({0})) = r-1({0}) = S. study the behaviour of this differential at other points.

On the other hand, if p e S, Lemma 7.36 (Gauss's lemma). Let p E S be a point of a surface and let
e > 0 be such that the exponential map exp is defined on C TBS.
(df) = W(O)(dr) 0, Then, if v E TpS with v < E, we have
since h is increasing and (dr)(N) = 1. when

Exercise (11): Cover the surface by countably many open subsets like those 0 when and w 1 v.
provided by Lemma 4.1. In fact, each of them has Lebesgue measure zero.
This is immediate if we remember that having measure zero is a property Proof. Let a : I -> TpS be a differentiable curve defined on a neighbour-
invariant under diffeomorphisms and that a plane in I183 has measure zero. hood of the origin in II8 such that a(0) = v, a'(0) = w, and
for all s E I, that is, whose trace is the circle centred at the origin of TpS
4.6. Appendix: Proof of Sard's theorem passing through v. For 8 > 0 sufficiently small, we have that ta(s) E
if t E (-1 - b, l + b). Then, we define a map
The version of Sard's theorem that we stated as Theorem 4.6 is a direct
consequence of a stronger result concerning differentiable maps between sur-
faces. The latter requires the notion of a set of measure zero in a surface by
for its statement and proof. We will base our study on the corresponding F(t, s) = ry(p, ta(s),1) = y(p, a(s), t).
concept of subsets of I[82 of Lebesgue measure zero. Recall that, as a conse- Differentiating with respect to t, we obtain
quence of the formula for change of variables for the Lebesgue integral on
I[82, a diffeomorphism between open subsets of I[82 takes subsets of measure
8t t s) _ 'Y p a(s), t).
(d exPp)t«(s)
zero into subsets of measure zero with this same property. This is why the
following definition makes sense. Thus
Definition 4.31 (Set of measure zero). A subset A of a surface S in Eu- d exPp)t«(s)
8t
t' s
clidean space ll83 is said to have measure zero if, for each parametrization
X : U - S of the surface, the set X1 (A) has Lebesgue measure (area) zero since the velocity vector of the geodesic y(p, a(s), t) has length independent
in I[82; compare this with Remark 5.9. of t. Putting t = 1 and s = 0 in the above, we get

The following properties follow directly from the previous definition and
from the corresponding ones for Lebesgue measure on I[82. and, thus, we have shown the first equality of the lemma. Another conse-
quence of the same equality, taking derivatives with respect to s, is that
Proposition 4.32 (Properties of the measure zero sets). Let S be a surface
a2F aF
in Euclidean space and let A C S be a subset. The following assertions are
true: asat' at - °'
7.6. The exponential map 231 132 4. Separation and Orientability

A: A has measure zero if and only if there exists a family {XZ}EI of

parametrizations of S whose images cover S and such that XZ 1(A)
has Lebesgue measure zero in I[82 for each i e I.
B: If A has measure zero, then it has empty interior.
C: If A is a minion of countably many subsets of S of measure zero,
then A has measure zero.

CX]. ,,(If' We now give a stronger and more common version of Sard's theorem
than what was needed in Theorem 4.6. In fact, that result can be deduced
from the following statement, along with property B of Proposition 4.32.
Figure 7.4. Exponential map Theorem 4.33 (Sand's theorem). Let f : Si - 52 be a differentiable map,
where Si and S2 represent either surfaces or open subsets of ][82. Then the
defined by complement of the set of regular values of f has measure zero in 82.
expp(v) = exp (p, v), Vv E e(TpS).
As in other places where the proof is somewhat longer than usual, we
Clearly it is a differentiable map taking each vector tangent to the surface will divide the proof of Sard's theorem into several intermediate steps.
S at the point p in its domain into the point of the geodesic with initial
conditions p and v that is attained at the time t = 1, that is, the point of Proposition 4.34. Let O be a relatively compact open subset of I[82 and let
the geodesic p.b.a.l. starting from p in the direction of v to the point where b: O -p ][82 be a differentiable map. Let Al denote the set
we arrive at time Iv I . Al = {x e 01 (Jacq5)(x) = O}.
An immediate property of the exponential map expp : S(TpS) -> S of a
If K is a compact subset of N, then, for each e > 0, there exists an open
surface S is
subset VE of I[82 such that
expp(Op) = exp (p, Op) = 'Y(p, Op, 1) = p, VP E S.
K C VE C O and E.
Moreover, if v E TopTpS = TpS, we have Thus ¢(N) has Lebesgue measure zero in I[82.
(d expp)o, (v) = I expp(ty) Proof. We will separate the proof into steps.
dt t=o
Assertion 1. If x, y e O are two points of the open set O such that the
I
segment [x, y] is contained in O, then, for each a e I[82, there exists t e [0, 1]
= dt -Y(p, tv,1) _
t t=o 7(p, y, t) = v, such that
t=Q I

(cb(y) - q5(x),a) = ((dcb)+t(_)(y - x),a).

where we have used the homogeneity of geodesics. Therefore
(7.13) (d expp)op = IT8, Vp E S.
In fact, if [x, y] C O and a is a vector in ][82, we can define a function f on
Then, in this situation we may apply the inverse function Theorem 2.75, [0, 1] by
assuring in this way that all points near p E S are attained by a unique f(t) = (q5(x+t(y-x)),a),
geodesic starting from p. which is differentiable. Now it is enough to apply the Cauchy mean value
Proposition 7.35. For each point p E S of a surface, there is a number theorem to this function and utilize the chain rule to compute its first de-
5 > 0 such that the map rivative.
Assertion 2. For each e > 0 there exists 8 > 0 such that, if C is a
expp:Bs(Op)CTpS-,S square in ][82 included in O with C fl Al 0 and diam C < 5, we have that
is a diffeomorphism onto its image. area ¢(C) < eaxea (C).
4.6. Appendix: Proof of Sard's theorem 133 230 7. Intrinsic Geometry of Surfaces

Given an arbitrary E > 0, set E' _ e/c > 0 where c = (2-,/)2M2 and which is an open subset of TS x {1}. Hence £1 is an open subset of TS C
M = 1 + maxXEO (dq5)> 0. Since the map S x R. We define on this subset the exponential map
4 exp : £1 -p s
is continuous on the compact set O, it is then uniformly continuous. Hence, by
for the chosen e' > 0, there exists 8 > 0 such that exp (p, v) = y(p, v, l),
which is, by construction, a differentiable map on 51 C TS C S x R3.
}
I(d - (d)I <e'.
Remark 7.32. For each point p E S we have that the geodesic yp,o1,, with
Once this S is determined, let C be a square of I[82 such that C C O with initial conditions the point p and vanishing velocity at this point, is the
constant curve p and, thus, it is defined on the whole of R. That is, s(p, Op) =
diam (C) < S and such that there is at least one point a e C fl Al', that
+00 for each p E S. Hence, 1 < s(p, Op) and then
is, a point a e C with (Jac ¢)(a) = 0. Then the linear map is not
regular and, thus, its image is not all of ][82. Then we can find a unit vector (p, 0 p) E Si and exp (p, 0 p) = p, V p E S.
v e ][82 perpendicular to this image. Let {v, w} be an orthonormal basis of Remark 7.33. Suppose that (p, v) E 51 is a point of TS in the domain
I[82. Since C is convex from Assertion 1 above, if x E C, we have
of the exponential map. By definition, 1 < E (p, v). Hence, if t E (0, 1], it
(/(x) - /(a),w) = ((d)a+t(x_a)(X - follows by the homogeneity of geodesics that
where t e [0, 1]. Using the Schwarz inequality, we have that 1 < 1
< s(p,v) - t s(p,v) = s(p,ty)
(¢(x) - q5(a), w) I < I - aI < M diam C, and, from this, that (p, tv, l) E S and that (p, tv) E 51. On the other
for each x e C. On the other hand, again by Assertion 1 and by the choice hand, if t = 0, we have (p, tv) = (p, Op) E Si, as we saw in Remark 7.32.
of the vector v, Summarizing,
(/(x) - /(a),v) - ([(d)a+t(x-a) - (d)aJlx - (p,v) E E1 == (p,tv) E 51i t E [0, 1].
for some t e [0, 1]. From this, the Schwarz inequality gives Moreover,

I((x) - I(a),v)I f I- (d)aIIX - al < e'diam C, exp (p, tv) = y(p, tv, l) = y(p, v, t), t E [0, 1].

since a + t(x - a) - a _ Itlix - al < Ix - a < diam C < S. Since the length Remark 7.34. As stated in Remark 7.32, (p, OP) E El for each p E S. But
of the side of a square is exactly its diameter divided by \/, we have, for El is an open subset of TS whose topology is induced from S x ][83. Thus,
each x e C, that there exist an open subset V of S containing the point p and a number E > 0
such that
(- ¢(a), w) I < \/M side C, T(V,E) = {(p,v) ETVIIvI <E} c51.
I(cb(x) - q5(a),v)I \/'sideC. That is, given p E S, the exponential map is defined, at least for the points
That is, ¢(C) is contained in a rectangle whose sides have lengths in a certain neighbourhood of p and the tangent vectors at these points with
2V'M side C and 2/e' side C. uniformly small length.

Thus, we obtain
area ¢(C) < (2/)2M2E" (side C)2 = E area (C),
as we wanted.
We can now prove Proposition 4.32. Given e> 0, put e' = e/vol (O) > 0
and denote by S > 0 the positive number provided by Assertion 2 above.
Now take a partition of the compact set K with norm less than 5 by means
For any point p e S of a surface, we see from Remark 7.33 above that
£(TPS) ={v E TpS I(p, v) E Si}
is clearly an open subset of the tangent plane TS including the origin Op
and which is star-shaped with respect to it. We call the exponential map at
the point p E S the map
expP : 5(TS) -* S
7.6. The exponential map 229
134 4. Separation and Orientability

interval (-6(p, v), 6(p, v)). From Lemma 7.29, we know that y = p1 o g is
of squares C1,. . . , Cn C O. Then we define the open set V6 as the interior
a geodesic of S with initial conditions 'y(0) = p and 'y'(0) = v. This is
the required 'yp,v. On the other hand, the subset 5 of TS x R is, locally, of Cl U U C. Clearly K C V6 C O and
n n
nothing more than (TS x R) f1 D, and so it is open in TS x R. Finally,
the map y : 5 - S is only the restriction to D of the map F. Hence, it is area q5(V6) < area (ic)Z = area Uq5(Ci).
2=1 2=1
differentiable. D
Since the diameter of each square Ci, i = 1, ... , n, is less than b and since
Remark 7.31 (Homogeneity of geodesics). For a point p E S in the sur- each CZ cuts K and, thus, N, we have area q5(Ci) <E ' area (Ci). In this way,
face Sand avector v E TPS, set ry = ryto be the geodesic defined on we get
n
(-(p, v), (p, v)). Define
area E(VE) < E area (CZ) < E' area (O) _
g:
(--(p, v), Se(p, v)J _-* S, z=i
which was the required condition. Hence K has Lebesgue measure zero in
where s > 0, by the relation ][82. To see that the same holds for N, it suffices to choose an open subset
W of II82 with O C W and such that q5 is defined on W. Now, changing O
g(t) = ry(st).
to W and K to Jam, we apply the above and are done.
Then g is another geodesic, since it is nothing more than a homothetic
repaxametrization of the first geodesic, and Proof of Sard's theorem. We will show that the image through f of the
set
g(0)=7(0)=p and g'(0) = sry (0) = sv. N={pESlI(Jacf)(p)=0}
Consequently E(p, v)/s < E(p, sv) and g coincides with ryP,s on this common has measure zero in S2. Let X2 : U2 - S2 be any parametrization of S2
domain. Since this is valid for any vector v E TS and any s > 0, we have defined on an open subset U2 of I[82. Since f is continuous, Vi = f'(X2(U2))
is open in Si. Then, we may cover the surface Vl with countably many
SE(p,v) E(p,sv) SE(p,ssv).

parametrizations Xi : Ui - Si of Si, n e ICY, defined on relatively compact

That is, for each p E S and each v E TS, we have open subsets of Il82 that can be extended to differentiable maps on each (Jr.
We denote by qn the compositions
E(p,Sv) = SE(p,v) 7(p,Sv,t) = 7(p,v,st) bn =X' 1 o f o Xi : Ui C I[82 - I[82,
for all s > 0 and all ti < E(p, sv). which are differentiable and obey the conditions that we need for applying
Proposition 4.32. Therefore, the set
7.6. The exponential map Nn = {x E Ul I (Jac q5n)(x)_ O}, nEN
The fact that geodesics on a surface are solution curves of a second-order has the property that qn(Nn) has measure zero and, thus,
differential equation, or components of solutions of a first-order equation,
area U n(Nn) = 0.
as we have seen in the previous section, will also allow us to handle at the nEN
same time sets of geodesics corresponding to sets with close initial conditions,
Now, it is enough to observe that we have the equality
through the map -y : S -p S defined on the open subset S of TS x R, which
appeared in the existence and uniqueness Theorem 7.30. U bn(Nn) - X2 l(f (N))
Consider the following subset Si of TS: nEty
to conclude that f(N) has measure zero in S2, according to Definition 4.31.
Si = {(p, v) E TS I (p, v,1) E S}.
Notice that this Si is homeomorphic through the first projection of TS x R
onto TS to
Sfl(TSx{1}),

Chapter 5
Integration on Surfaces
5.1. Introduction
In this chapter, we will provide additional tools needed for the study of
global geometry that we develop later. At this time, they will not be of
a topological nature, as in Chapter 4, but will have mainly an analytical
flavour.
Perhaps the most famous result that we will prove in this chapter is the
divergence theorem, whose statement requires, in some sense, the JORDAN-
BROUWER separation theorem for a compact surface in R3. If S is a compact
connected surface in Euclidean space and 1 is the inner domain determined
by S, the divergence theorem will enable us to relate the integrals of certain
functions defined on 1 with integrals on the surface S. In this way, it is
a generalization of the fundamental theorem of calculus. Even the names
of the concepts involved in this theorem fields, divergence, etc. show its
physical origin. Indeed, the first formulation of this theorem appears in
a postscript of a letter dated in 1850 and addressed by LORD KELVIN to
G. G. STOKES, professor at the University of Cambridge. LORD KELVIN
himself gave two different proofs of the result. There is also another one
due to MAXWELL in this same period. But all the above presuppose the
development of a certain integral calculus on the surface.
We begin with the task of endowing surfaces with an integral calculus,
although we will do so in a non-standard, though natural, way. In fact,
we constructed the differential calculus on surfaces definition of a differ-
entiable function, differential of a function, etc. in Chapter 2, by using
differential calculus on the plane and then translating it onto surfaces by
135
228 7. Intrinsic Geometry of Surfaces
since x is a solution of (7.12). As a consequence, f(t) = r(x(t)) = 0 for any
t E (-E, E), that is, x(t) E V C S for any time t. But, moreover, in this
case,
[xll(t)]T = G(x(t),x'(t))T = 0,
and thus x is a geodesic of S with x(0) = p and x'(0) = v. It is easy to see
that the converse is also true. Indeed, if x : (-E, E) -* V C S is a geodesic
of S with x(0) = p E V and x'(0) = v E TPS, then x"(t) is a vector at each
t normal to the surface. Thus
x"(t) = a(t) (N o x)(t)
for some function A. After scalar multiplication by (N o x) (t), we get
A(t) = (x"(t), (N o x)(t)) = -(x'(t), (dJV)(t)(x'(t))),
since (x', N o x) = 0. Since ir(x(t)) = x(t) and (d7r)(t) is the identity map,
x"(t) = G(x(t),x'(t)).
Therefore, g = (x, x') is a solution of (7.11) with initial condition g(0) _
(x(0), x'(O)) and x(0) = p, x'(0) = v. This then allows us to make the
following assertion.
Lemma 7.29. The geodesics of the surface S are exactly the solutions of the
system of second-order differential equations (7.12) with initial conditions a
point of the surface and a tangent vector at this point.
From this, we have the following fundamental result.
Theorem 7.30 (Existence and uniqueness of geodesics). Let S be a surface
in Euclidean space. For each point p E S and each tangent vector v E TS
there exist a real number 0 < E(p, v) < -{-oo and a unique geodesic -y
(-E(p,v),E(p,v)) - S such that
ryP,,,(0) = p and 'YP(0) - v,
which is maximal tender these conditions. Furthermore, the set
E _ {(p,v,t) E TS x RI Iti < E(p, v)},
where TS = {(p, v) E S x I[83 I v e TPS}, is an open subset of TS x I[8 C
S x Il83 x ]E8, and the map
defined by y(p, v, t) = yP,(t) is differentiable.
Proof. Given p E S and v E TES, we have (p, v) E TV C N(V) x ]E83, for
some orientable open set V containing the point p. Then, let b > 0 be the
number corresponding to the solution of (7.11), g : (-b, 8) - N(V) xIl83 with
initial condition (p, v). This g extends to a maximal solution on a symmetric
7.5. Geodesics
where G : N(V) x Il83 -> I[83 is given by
G(x, y) _ -(y, d(N o ?r)x(y)) (N o
for each x E N(V) and y E Il83, where N is any local Gauss map for S.
Furthermore, G is clearly differentiable and, thus, is as well. We now
consider the system of first-order differential equations
(7.11) (x,y)' = c(x,y).
Solving them is equivalent to solving the second-order system of equations
(7.12) x" = G(x, x') .
The theorem of existence, uniqueness, and dependence of the initial condi-
tions implies that, if (a, b) E N(V) x I[83, there exist E(a, b) E (0, +00] and a
map ga,b : (-E(a, b), E(a, b)) -* N(V) x ]E83 such that
9a,6(0) _ (a, b), 91,b(t) _ c(9a,b(t)), t E (-(a, b), E(a, b));
that is, ga b is a solution of (7.11) with initial condition (a, b) and, moreover,
9a,b is maximal-with respect to the symmetric interval of definition-among
all those having the same initial condition. On the other hand, the set
D = {(a,b,t) E N(V) x I[83 x RI Iti < E(a, b)}
is an open subset of N(V) x ][83 x ]E8 C ][87, and the map
F : V -* N(V) x
given by
F(a, b, t) = 9a,b(t),
is differentiable.
Now let g : (-E, E) - N(V) x I[83 be a solution of (7.11), with E> 0 and
initial conditions
g(0) _ (p, v) E TV = {(p, v) E V x I[83 Iv E TS} c N(V) x Il83.
Denote by x : (-e, e) - N(V) the first component of g. Then x is a solution
of the system (7.12) with x(0) = p and x'(0) = v. To measure how much
the trajectory x goes away from the surface S, consider the composition
f = r o x : (-E, E) -II. We have 1(0) = r(x(0)) = r(p) =0 and
f'(t) _ (dr)(t)(x'(t)), t E (-E,E),
and, using the properties discussed in Exercise (1) of Chapter 4, we have
f'(t) = (x'(t), (N o
Hence, f'(0) _ (v, N(p)) = 0. Taking derivatives again, we have
f"(t) _ (x"(t), (N o irXx(t))) + (x (t), d(N o 7r)x(t)(x'(t))) = 0,
227 136 5. Integration on Surfaces
means of parametrizations, although we were always careful that the defini-
tions we gave did not depend on the choice of the precise parametrization.
A similar method could be followed to build integral calculus on surfaces:
use any theory of integration already known on the plane R2, for instance,
Lebesgue integration, and try to transpose it on the surface by means of
parametrizations. This is the usual way to introduce an integral calculus on
surfaces.
Instead, in this book, the integral of a function on a compact surface or
on a relatively compact open subset of an orientable surface will not be
defined from the integral of functions on the plane through parametrizations
of the surface and the so-called partitions of unity, but we will use the fact
that our surfaces are not abstract manifolds since they are embedded in
R3. We will define the integral of a function on a compact surface from the
integral on any of its tubular neighbourhoods which is a relatively compact
open subset of R3 of the function obtained by extending the original one as
a constant along each normal segment. That is, we will exploit the following
result that we established for surfaces: even though a relatively compact
subset of a surface does not have to be diffeomorphic to an open subset of
the plane, it has a tubular neighbourhood that is diffeomorphic to an open
subset of Euclidean space R3.
We will develop the usual properties of Lebesgue integration, give a proof
of the divergence theorem, and, later, we will finish with the Brouwer fixed
point theorem, as an example of an application of the first result. In Chapter
6, we will see that integration is a very powerful tool for obtaining theorems
of a global nature in the theory of surfaces.
5.2. Integrable functions and integration on S x III
Let S be an orientable surface. We will use the term region for any relatively
compact open subset of S. For example, if S is a surface closed as a subset
of Euclidean space, then from Exercise (6) of Chapter 4 any bounded open
subset is a region of S in the sense that we have just defined. Note that if
S is compact, then S itself is a region. We will usually reserve the symbol
R. for regions in orientable surfaces.
Let R. be a region in an orientable surface S and let O be an open subset
of ][83. Suppose that b : O -* R. x (a, b) is a diffeomorphism, where a < b
are two real numbers. We introduce the absolute value of the Jacobian,
by analogy with the usual definition for differentiable maps between open
subsets of Euclidean spaces-see Exercise (13) of Chapter 2 for the case of
surfaces-as the function IJac q : O -* Il8 given by
Jac q5(x) = d
5.2. Integrable functions and integration on S x Ilk 137
for each x e O, where the absolute value of the determinant of the linear
map (dq5) : Il83 -* x I[8, where the first component q51 : O -* S of ,
must be understood in the sense of the following exercises.
Exercise 5.1. Let Vl and V2 be two Euclidean vector spaces with the same
dimension, and let f : Vi - V2 be a linear map between them. If M is the
matrix of the map f relative to two orthonormal bases of Vi and V2, prove
that I det MI does not depend on the bases chosen. This non-negative real
number is called the absolute value of the determinant of f and we denote
it by I det f Show that I det f > 0 if and only if f is an isomorphism.
I.
Exercise 5.2. Let f : Vl - V2 and g : V2 -* V3 be two linear maps between
Euclidean vector spaces of the same dimension. Prove that
I det (9 ° f) I= I det g I I det f ,
that is, the absolute value of the determinant is multiplicative with respect
to the composition of maps.
In general, we see that this absolute value of the Jacobian can be defined,
in a completely analogous way, for differentiable maps between objects such
as open subsets of Euclidean space and products of surfaces by intervals
of Ilk and that these are still multiplicative with respect to composition.
Explicitly, if b : O1 -4 02 and b : 02 -- 03 are two differentiable maps
between either open subsets of R3 or open subsets of products of surfaces
and Ilk, we then have, by the chain rule and Exercise 5.2 above, that
oq5)I = [IJac'bIoc1IJaccI.
Moreover, by Exercise 5.1, such a map q5: Ol - 02 is a diffeomorphism if
and only if Jacq5I > 0.
We will now take advantage of the fact that, since O is an open, and thus
a measurable, subset of ][83, we know how to integrate functions on O. This
will allow us to define the integral of functions on the product R x (a, b) with
a < b. The idea is to work as if O and R x (a, b) were both open subsets
of ][83 and to transfer integrals from each to the other by imagining that
the formula for the change of variables is still valid for any diffeomorphism
q5:O -* R. x (a,b).
Definition 5.3 (Integral on S x II8). We fix a diffeomorphism b O -*
R. x (a, b) defined on an open subset O of Il83, and we say that a function
h : R. x (a, b) -* II8 is integrable on R. x (a, b) if (h o q5) IJac c5I is integrable
on O in the sense of Lebesgue. In this case,
(5.1) f(ho)IJacI
226 7. Intrinsic Geometry of Surfaces
Figure 7.3. Geodesics of the cylinder
That is, the geodesics of the cylinder are circular helices including, as a
limiting case, its straight lines and circles.
In every example so far, two facts stand out. First, for each point of the
surface being considered and for each tangent vector at the point, there is
a geodesic passing through the point with this given velocity. Second, these
surfaces mentioned above share the property that any of their geodesics are
defined on the whole of R. Which of these two facts can be generalized to
an arbitrary surface? Of course, the answer is definitely not for the second
one. To see this, consider the surface S = P - {p}, where P is a plane of R3
and p E P is any of its points. On this surface, the geodesics are straight
lines parametrized with constant velocity. If q E S and if we take the vector
p - q E TqS, the geodesic passing through q with velocity p - q cannot be
defined on the entire real line Ilk. However, we will see that the first property
is not exclusive to the three examples studied above.
In fact, let S be an arbitrary surface. From Lemma 4.25, we know that,
for each p E S, there is an orientable open neighbourhood Vp of p in S and
a number sp > 0 such that N (Vp) is a tubular neighbourhood. Let N(V )
be any tubular neighbourhood of this type on an orientable open subset V
of the surface. Such a subset of R3 is an open set containing the surface V
and where the projection it : N(V) - V C S and the oriented distance r
are defined, with the properties discussed in Exercise (1) of Chapter 4. We
define a map really, a vector field on an open subset of R6
g: IV (V )X R3 -*II83 X II83
by
g(x,y) = (y,G(x,y)),
7.5. Geodesics 225 138 5. Integration on Surfaces

Hence a is geodesic if and only if a" = 0, that is, a(t) = tc + d where will be called the integral of h on R. x (a, b), and it will be denoted by either
c, d E I[83. Therefore, we can say that the geodesics of a plane are just its of the symbols
straight lines parametrized proportionally to the arc length.
Example 7.27 (The geodesics of a sphere). Let a be a differentiable curve
h or f Zx (a,b)
h(p, t) dp dt.

p.b.a.l. on a sphere S2(r) centred at a point a E I[83 with radius r > 0. We first check that this is a good definition in the sense that it does
We have f - a12 = r2 for all t. Differentiating this expression twice, we not depend on either the open set O or the diffeomorphism q: O -* R x
obtain (a(t) - a, a"(t)) _ -1 at each instance. Since the plane Ta(t)S2(r) is (a, b). We will decide later whether we have at our disposal open sets and
the orthogonal complement of the vector a(t) - a, we have that diffeomorphisms of this type. In fact, if Ol and 02 are two open subsets of
Il83 and qZ : OZ -* R. x (a, b), i = 1, 2, are two difFeomorphisms, then the
[a"(t)]T = a"(t) + (a(t) - a). composite map =X21 0 q51 : Ol - 02 is a difFeomorphism between open
Thus, a is a geodesic if and only if it satisfies the equations subsets of II83 satisfying
r2a"(t) + a(t) - a = 0, Ia(t) - a (t) 12 = 1. I_ Ion
It is easy to see that the solutions are O. Thus, if h : R. x (a, b) -* II8 is a function such that (h o q52) IJac c52 1 is
integrable on 02i the formula for the change of variables for the Lebesgue
a(t) = a + p cos t + rv sin t , 1p12 = r2, 1vI2 = 1, (p,v) = 0; integral tells us that
r r
that is, the great circle determined by the plane spanned by p and v ran with [(hOq52)
constant speed.
is integrable on Ol and that
Example 7.28 (The geodesics of a cylinder). Let C be the right cylinder
of unit radius whose axis is the z-axis of Il83, and let a [a, b] -p C be [(ho
102 101
a differentiable curve on C given by a(t) _ (x(t),y(t),z(t)). Then a is a
That is, in this case, we have
geodesic if and only if
(ho2)IJac2I= f (hoi)Jac,.
J02 0l
for each t E [a, b]. Using Exercise 2.55, this is equivalent to Second, to make sense out of the previous definition, we need to have at
(x"(t), y"(t), z"(t)) (x(t), y(t), 0). our disposal, for each region R on the surface, a difFeomorphism b : O -
Since x(t)2 + y(t)2 = 1, differentiating twice and assuming that a is p.b.a.l., R x (a, b), where O is an open subset of Il83. In fact, note that, indeed, we
we have that x(t)x"(t) + y(t)y"(t) _ -1 + (z'(t))2. Hence the curve is a can take as O any tubular neighbourhood NE(R) supplied by Theorem 4.26.
geodesic of the cylinder if and only if In this way, q5: (IJ?. x h) o F-1 : O -* R. x (a, b), where h : (-e, e) -* (a, b)
is any diffeomorphism, is the diffeomorphism that we are looking for.
x"(t) + (1 - z'(t)2)x(t) = 0,
y"(t) + (1 - z'(t)2)y(t) = 0, From this definition, it is more or less immediate that the integral of
functions on subsets of S x ][8 of the form R. x (a, b) retains all the elemen-
z"(t) = 0.
tary properties-linearity, monotony, etc.-of Lebesgue integration. In the
Suppose now that a(0) _ (1,0,0) and that a'(0) _ (0, a, b) with a2 +b2 = 1. following proposition, we gather the most significant of them.
Then
z(t) = t b and x"(t) + a2 x(t) = y"(t) -I- b2 y(t) = 0, Proposition 5.4 (Properties of the integration on S x Il8). Let R. be a region
of an orientable surface and let (a, b) be a non-empty open interval of R. The
and so x and y are of the form
following standard properties of the Lebesgue integral remain valid for the
Asin at + µ cos at, A, µ ER, integration of functions R. x (a, b):
and, therefore, A: Any continuous function on R. x [a, b] is integrable on R. x (a, b).
a(t) _ (cos at, sin at, bt) with a2 + b2 = 1. B: The integral of functions is linear and monotone.
5.3. Integrable functions and integration on surfaces 139
C: There are convergence theorems and theorems concerning the con-
tinuous and differentiable dependence parameters, under the same
conditions as for the Lebesgue integral on R3; to recall them, see
Remark 5.22.
D: Let R' C R. be a subregion of R. and suppose that the function h
is integrable on R. x (a, b) and vanishes on (R. - R') x (a, b). Then
f Zx (a,b)
h=
RJ x (a,b)
h.
E: If RI is a region of another orientable surface S', (a', b') is an-
other open interval of Ilk, and b R. x (a, b) --* RI x (a', b') is a
:
Rfor
diffeomorphism, then
fW x (a',b')
h= f of a geodesic; that is, there are ways of moving along it inertial or with
each integrable function h on RI x (a', b').
Proof. The only assertion that perhaps is not a direct consequence of the
definition is the last one. Property E follows from two things: first, that the
definition of the integral on R. x (a, b) depends neither on the open set O nor
on the diffeomorphism q5: O -* R x (a, b) and second, the multiplicativity
of the absolute value of the Jacobian. O geodesic of S if and only if f o a is a geodesic of S'.
5.3. Integrable functions and integration on surfaces
Consider now a function f : R. -* I[8 defined on a region R. of an orientable
surface S. We can think of f as a function on the product R x (a, b), for
any real numbers a < b, as
f(p,t)=f(p), b'p E R C S b't E (a, b).
For this class of functions, we have that
1 for each s E (-b, b). Hence L'c(0) = L'1(0) = 0 and f o a is a geodesic
f(p)dpdt
b - a 1x(a,b)
is independent of (a, 6), provided that this integral exists. The reason is
that, if we take another number c < d, we can define a diffeomorphism
q5: S x (a,b) -* S x (c,d) by b = ids x h, where h : (a,b) -* (c,d) is an
affine diffeomorphism. We now apply property E above. This enables us to
make the following definition.
Definition 5.5 (Integral on S). Let R. be a region of an orientable surface
and let f : R -* Il8 be a function. We will say that f is integrable on R
224 7. Intrinsic Geometry of Surfaces
Remark 7.24. If a : [a, b] - S is a geodesic of S, then
= 2(a"(t), ap(t)) = 0
dt l
and, thus, any geodesic of a surface has constant length velocity vector
at each time t. That is, any geodesic is either a constant curve or it is
parametrized proportionally to the arc length. Notice that being geodesic
is not a geometrical property because it is a property of curves that does
depend on the parametrization, and so it is a physical property. In fact, if
we reparametrize a geodesic curve, the new curve that we obtain is also ge-
odesic if and only if the reparametrization is homothetic. Instead, it is true
that any regular curve having the geometrical property of minimizing in
general, making critical the length between its endpoints must be the trace
constant speed that make it a geodesic.
The first property arising from the relation between the definition of a
geodesic and the length of curves is its invariance under isometries, which
converts them to an object of intrinsic geometry.
Theorem 7.25. Let f : S - S' be a local isometry between two surfaces S
and S", and let a : [a, b] - S be a differentiable curve on S. Then a is a
Proof. Since the fact that a curve is a geodesic is of a local nature, we can
restrict ourselves to pieces of the curve and surface that are small enough for
f to be a diffeomorphism. Thus, we may assume that f is a global isometry.
Then, suppose that a is a geodesic of S and let G : [a, b] x (-b, b) -> S' be
a variation of f o a on S' with 6> 0. Then F = f1 o G is a variation of a.
Moreover, since f1 preserves the length of curves, we have
jiF(S) - I'Q(F's) - I'a(.f-1 ° Cas) - I'a(C's) - I'G(S)
because it has velocity vector of constant length and thus we can apply
Corollary 7.22. The converse can be shown in the same way by changing
the roles played by S and S' and using f_i instead of f. O
Example 7.26 (The geodesics in a plane). Let P = {x E ][83 I (x, a) = b}
be a plane perpendicular to the unit vector a E Il83. If a : [a, b] -> P is an
arbitrary differentiable curve on P, we have (a(t), a) = b for each t E [a, b].
Thus (a"(t),a) = 0, that is,
a" (t) E T«(t)P, `dt E {a,b].
7.5. Geodesics 223 140 5. Integration on Surfaces

Corollary 7.22. A regular curve a : [a, b] -> S on a surface S has criti- when the function (p, t) E R. x (0, 1) H f(p) is integrable on R. x (0, 1). The
cal length; that is, LF(0) = 0, for every proper variation F if and only if real number
[aFF(t)}T n ap(t) = 0 for each t e [a, b], in other words, if and only if its
(5.2) f(p)dpdt
tangential acceleration [a(t)}T at each point of the surface is proportional
to the velocity ap(t). If we assume that a is parametrized proportionally to will be called the integral of f on R. and it will be represented by any of the
the arc length, then this is equivalent to the fact that [aFF(t)]T = 0. symbols
Proof. First, it is an elementary exercise to show that the condition
7z f ff(P)dPi or ffdA.
[aFF(t)]T n a'(t) = 0 is invariant under reparametrizations of a, and the
same applies to the fact of having critical length for all proper variations. The following properties are inherited, through definition (5.2), from the
Consequently, we suppose from now on that the curve a is p.b.a.l. In this analogous ones of either the integral that we defined on S x ][8 in (5.1) or
case, since Ia'(t)12 = 1, the acceleration a"(t) is perpendicular to the veloc- the Lebesgue integral on II83.
ity ap(t) at each time t. This is why the two conditions [aFF(t)]T n ap(t) = 0
and [aFF(t)}T = 0 are equivalent. By the formula of the first variation of the Proposition 5.6 (Properties of the integration on S). Let R. be a region of
length in Theorem 7.21 above, if L'F(0) = 0 for every proper variation F of an orientable surface and let f, g : R. -* II8 be two functions defined on R.
a, one has The following assertions are true:
b
A: If f is continuous on R, then it is integrable on R.
f (V(t), a"(t)) dt = 0,
where V is the corresponding variational field. But, by using Proposi-
B: If f and g are integrable and A E Il8, then f + g and Af are
integrable and, moreover,
tion 7.20, we see that there exists a proper variation whose variational field
is V(t) = h(t) [a" (t)]T, where h is a function differentiable on [a, b], positive ff+g=ff+fg and fAf=Aff.
on (a, b), and with h(a) = h(b) = 0. Hence
C: If f and g are integrable and f < g, then
f
b

h(t)I[a"(t)]TI2dt = 0.

Thus the acceleration a"(t) does not have a component tangent to the sur-
face. Conversely, if the curve p.b.a.l. a satisfies [aFF(t)]T = 0 for each
t e [a, b], the formula of the first variation of the length in Theorem 7.21
implies that L(0) = 0 for each F proper.
In particular, the condition supplied by Corollary 7.22 is a necessary
condition for a regular curve to minimize the length between its two end-
points: it must be the path of a particle subjected exclusively to the force
needed to remain at each time on the surface. That is, from the point of
view of a two-dimensional physicist living on the surface, it is the path of
a particle that is not subjected to any exterior perturbation an inertial
trajectory, if our physicist accepts Newton's laws.
Definition 7.23 (Geodesic). In general, any differentiable curve a :
S on a surface S will be called a geodesic when
a" (t) 1 Ta(t) S, Vt e [a, b],
independently of having or not having the property of minimizing the length
with respect to close curves or distant curves with the same endpoints.
If, moreover, f and g are continuous, the equality holds if and only
iff=g.
D: Let R' C R. be a subregion of R. and suppose that f vanishes on
'R.-W. Then
E: If R' C R. is a subregion of R. such that R. - RI is a finite union
of curves and points of S, then
f f
f integrable.
[a, b]
Proof. It is only worthwhile to say something about the proof of E. We
have that R' x (0,1) C R. x (0, 1) and that the difference of the two sets
is a finite union of products of the form C x (0, 1) and {p} x (0, 1), where
C is a curve of S and p is a point. But, since b O -* R. x (0, 1) is a
diffeomorphism, it is enough to realize that q5' (C x (0, 1)) is a surface and
5.3. Integrable functions and integration on surfaces 141 222 7. Intrinsic Geometry of Surfaces

that '-1({p} x (0, 1)) is a curve, both in the open subset O of Il83. Note C: [a,b] x (-e,e) - R by
now that both curves and surfaces have Lebesgue measure zero in II83, as we
have seen in Exercise (11) of Chapter 4. G(t, S)_
Remark 5.7 (Area of a region). Since the constant function 1 is continuous, we have that G is continuous and that
properties A and C above assure us that the integral of the constant function
1 exists and that it is positive on each region R. of an orientable S. This G(t, 0) = Ia'(t)I = 1>0, Vt E [a,b].
number will be called the area of the region R. It will be denoted by A(R) Hence, there exists a number S such that 0 < S <e and such that
and so we can write
A(7)=f 1. IsI <6 = G(t, s) >0, Vt e [a, b].
Thus G [ab] X is positive and, so, differentiable. As a consequence, the
At the beginning of this chapter, we alluded to the standard way of length function of the variation F, suitably restricted, LF : [a, b] x (-S, S)
presenting a theory of integration on surfaces. One bases it on a known Ilk, is differentiable and, moreover,
theory for I[82 and transfers it by means of parametrizations. The following
statement relates this procedure to our definition of integral and, by the way,
provides a useful manner for computing integrals when we have an explicit
LF(S) =
J
fb
S (t, S) dt
parametrization which covers the surface almost entirely. for all s e (-S, S). Our aim now is to compute LF(0). For this, we have
Proposition 5.8. Let R. be a region of an orientable surface and let X 02F\ J OF
(t,°)
U -* S be a parametrization defined on an oven subset U C ][82. Suppose
that X (U) C R. and that the difference R. - X (U) consists of finitely many
(7.10)
as -
asa (t, o),
bt e [a, b}.

points and curves. Then With these preliminaries, one can already learn something about curves
minimizing the length between its endpoints, even among only close curves.
Theorem 7.21 (First variation of the length). Let a : [a, b] -* S be a curve
for each integrable function f : R -* R. p. b. a.l. on a surface S and let F : [a, b] x (-S, S) - S be a variation of a.
Then, if LF is the associated length function,
Proof. Note that X (U) is a region of S included in R. and that, by part E 6

of Proposition 5.6, we can put L(0) = (V(b), a'(b)) - (V(a), &(a)) f (V(t), a"(t)) dt,
-
ff=f X(U)
f where V is the variational field of F.

for each function f integrable on R. Definition (5.2) of the integral on the Proof. From (7.10) it follows that
region X(U) gives pb 2
o dt.
x(u) f x(u)x(o,i)
f(p)dpdt. LF(°) Ia - aF' asa fit,

Integrating by parts, we have

But the map q5: U x (0, 1) -* X (U) x (0, 1), given by b = X x id(o 1), i.e.,
by /OF OF (b,°) b 102F aF
(t,0)dt.
(u,v,t) _ (X(u,v),t), (u,v) E U, t e (0, 1), at ' as c>o> - at2 as
is clearly a diffeomorphism defined on the open subset U x (0, 1) of Il83. We complete the proof by utilizing the fact that
Furthermore, a
u = (X, 0), ctv = (Xv, 0), and ct = (0,1) 8t t' -
0) a '(t),
at2
t, 0) = a"(0), and S (t, 0) = V (t).
7.5. Geodesics 221 142 5. Integration on Surfaces

Proof. Since the trace K = a([a, b]) of the curve is a compact subset of S, and, hence, the corresponding absolute value of the Jacobian is
we can, by Lemma 4.25, find a neighbourhood W of K and a number 6> 0
such that there is defined a differentiable projection r : N(W) - W C S of
I_ IXu n X I(u, v), b'(u, v) E U bt E (0, 1).
the tubular neighborhood N(W), with the properties discussed in Exercise Thus, we can use this diffeomorphism b in definition (5.1) and obtain
(1) of Chapter 4. Since N(W) is an open subset of Il83 containing the
compact set K, there exists a number e' > 0 such that, if dist (p, K) < s', f (v)x(o,i)
f(p)dpdt
= f x(o,i)
(fo)IJacI =(f
v
then p e Na(W). Put e _ e'/(1 -I- M) with M = maxtE[d,b] IV(t)I. Thus, if where we have applied theorem for the Lebesgue integral to compute
s e (-e, E), we have
the integral on the product U x (0, 1) C ][83.
dist (a(t) + sV (t), K) < I< e M <e', Remark 5.9 (Sets of measure zero in an orientable surface). Let S be an
for each t e [a, b]. Hence orientable surface and let R be a region of S. A set A C R. will be said to
have measure zero when
a(t) +sV(t) E Na(W), b(t, s) E [a,b] x (-s, E).
Then, we define the required variation F : [a, b] x (-e, e) -> S by the ex- L XA =0,
where XA is the characteristic function of A in R. For an arbitrary sub-
pression
set A of S, the notion of measure zero can be defined as follows: A has
F(t, s) _ ir(a(t) + sV (t)), b'(t, s) E [a,b] x (-s, s). measure zero when A l R. has measure zero for all regions R of S. From
It is clear that F is differentiable. When s = 0, one has Proposition 5.8 above, we deduce, among other things, the consequence that
this concept of measure zero coincides with what we introduced in Defini-
F(t, 0) _ pr(a(t)) = a(t), tion 4.31, based on parametrizations and on the Lebesgue theory on R2.
since a(t) E S and, moreover, its va,riational field is Now, it is clear, from the definition we have just proposed, that, if a func-
tion is defined on a region R. of S except maybe on a subset of measure zero,
(t, 0) = ds I (a(t) + sV (t)) we may question if it is integrable on R or not.
8s s=0
Example 5.10. Denote by S the torus of revolution of Example 2.17 whose
_ (dlT)a(t)(V(t)) = V (t) minimum and maximum distances to the revolution axis are, respectively,
a - r and a -I- r, with a > r > 0. Then the map X : (0, 27r) x (0, 27r) --*
since a(t) E S and V(t) E T«(t)S and, according to the exercise cited earlier, given by
the differential of r at the points of the surface, restricted to the correspond-
ing tangent plane, is the identity map. Finally, if V (a) = V (b) = 0, X (u, v) _ ((a + r cos u) cos v, (a + r cos u) sin v, r sin u),

F(a, s) _ 7r(a(a) + sV (a)) _ 7r(a(a)) = a(a),

is a parametrization-see Exercise (3) of Chapter 2-of S covering it entirely
except for two circles with a common point. Then we may apply Proposi-
for each s e (-e, e). Analogously, F(b, s) = a(b). tion 5.8 to this specific parametrization to integrate any integrable function
on the whole of the torus S. For instance, take the constant function 1. We
We can associate to each variation F : [a, b] x (-e, e) -k S of the curve have
a = Fo a function f
A(S) = 1= IXu A v) dude.
LF: (-e,e) -* J J J
by the expression But
Xu(u, v) _ (-r sin u cos v, -r sin u sin v, r cos u),
LF(S) = L(F8) = f a
6 (t) dt = f a
6
OF
at
(t'
S)
dt. Xv(u, v) _ (-(a -I- r cos u) sin v, (a -I- r cos u) cos v, 0),

This is the length function of the variation and clearly LF(0) = La(a) for and consequently
each variation F. Now suppose that the curve a is p.b.a.l. If we define Xu A Xv12 = IXuI2IXvI2 - (x,x)2 = r2(a -I- r cosu)2.
5.3. Integrable functions and integration on surfaces 143 220 7. Intrinsic Geometry of Surfaces

Finally, the area of the torus is

f2ir f2ir
A(S) = r(a + r cos u) du dv =
0 0
J J
Exercise 5.11. Calculate the integral of the Gauss curvature function on
the torus of revolution of the previous example by using Proposition 5.8.
Exercise 5.12. Show that the map X : (0, 'ii) x (0, 2ir) -* Il83, defined by
X (u, v) _ (al, a2i a3) -{- r(sin u cos v, sin u sin v, cos u),
is a parametrization of the sphere S2(r) with centre a = (al, a2i a3) E II83
and radius r > 0. Use this, together with Proposition 5.8, to prove that the Figure 7.2. Variation of a curve
area of this sphere is 4irr2.
Definition 7.19. Let a : [a, b] -+ S be a differentiable curve on a surface
Now let V1,. . V be a finite cover of the region R. by open subsets of
. ,
S C R3. A variation of the curve a is a differentiable map F : [a, b] x
the orientable surface S. We can define on R an integer-valued function (-E, E) --4S such that F(t, 0) = a(t) for each t E [a, b], where E is a positive
n : 1Z -* N by taking each point p e R. into the number of open sets of the number. For each s E we have a curve Fs [a, b] -> S, given
:

cover that p belongs to. Then 1 < n(p) < k for each p e R. Using this by F8(t) = F(t,s). We will call it a longitudinal curve of the variation.
function, we define functions X1, ... , Xk, on R., indexed by the open sets of When all these curves have the same endpoints as F0 = a, that is, when
the cover, by F(a, s) = a(a) and F(b, s) = a(b) for each s E (-E, E), we will say that the
(0 ifpV, variation is proper. Associated to the variation F, we define V : [a, b] - 1R3
Xi(P)= L by

for all i = 1, ... , k and all p e R.. It is clear from this definition that the V (t) =
as
(t, 0) = d F(t, s).
8-o
sum X1 -I- + x is the constant function 1 on R. This family xl,... , Xk V is clearly differentiable and, moreover,
of functions which are not of course differentiable is usually called a par-
tition of the unity on R. associated to the cover Vi,. . . , Vk. We will use it V(t) E TT(t)S, Vt E [a,b].
to see that, as mentioned before, there is an alternative definition for the
The map V is called the variational field of the variation F. Of course, if F
integral of the functions on surfaces, taking as a base Lebesgue theory on
R2 and not on R3. is proper, we have V (a) = V (b) = 0.

Proposition 5.13. Let R. be a region of an orientable surface S and let We now show that variations of curves on a surface and variational fields
f : R -* I[8 be an integrable function on S. Then associated to them have a deeper relation.
k Proposition 7.20. Let a : [a, b] -> S be a differentiable curve on the surface
JR
I f=J ((xf)oX)IXAXI,
i=1
S and let V : [a, b] -> R3 be a differentiable map such that
V (t) E Ta(t)S, Vt E [a, b].
where each Xi : U2 --* S, i = 1, ... ,k, is a parametrization of S defined on
an open subset U2 of R2 such that R. C X 1(Ul) U U X k (Uk) and X1, ... , Xk Then there exist e > 0 and a variation
is the partition of the unity on R. associated to this cover. F : [a, b] x (-E, E) -* S
Proof. Consider the partition of the unity x1,. , Xk associated to the fi- of a whose variational field is V . Furthermore, if V (a) = V (b) = 0, F can
nite cover Xl(U1),.. . , Xk (Uk) of the region R. The properties listed in be chosen to be proper.

7.5. Geodesics 219
point p e S into the determinant of the endomorphism A(f) + (dN)P is
non-positive by Exercise 7.17 above. Consequently, we have
(7.9) (H'of-H)(p)dp> 0,
Is
and, if equality is attained, then A(f) P = -(dN)P at each point p e S, that
is, (dN')f() o (df)P = (df)o (dN)P for all p e S. This is equivalent to saying
that f preserves the second fundamental form. Therefore, if equality holds
in the integral inequality above, then f is the restriction of a rigid motion,
by the Fundamental Theorem 7.7. We now show that, indeed, equality does
occur in this inequality. The reason is that if we interchange the roles of
Sand S' and substitute the isometry f by its inverse f1, we get another
integral inequality on S', namely
f(Hof_1 _ lye) (p) dp > 0.
Applying the change of variables theorem for the diffeomorphism f : S
S' to this integral and taking into account that the absolute value of its
Jacobian is identically 1, we have
(H-H'of)(p)dp 0,
Is
which, together with (7.9), gives the required equality.
7.5. Geodesics
We have already mentioned throughout this chapter that the subject of
the intrinsic geometry of surfaces is the study of those properties and ob-
jects that isometries keep invariant, in other words, according to Proposi-
tion 7.2, all the geometrical objects preserved by diffeomorphisms that keep
the length of curves invariant. Until now, we know only one invariant of
this class: the Gauss curvature, and we have profited from its invariance in
helping us to show the rigidity of positively curved compact surfaces, that is,
the fact that their intrinsic geometry determines the way they are embedded
in three-space.
Other objects which should be, in a first approach, invariant under maps
preserving lengths of curves are least length curves joining two given points,
if they exist. We now begin the study of certain curves of a special nature on
the surfaces that are fundamental in intrinsic geometry: geodesics. We pose
a question: in the case when they exist, what are the main properties of the
curves that minimize the length between two points of a surface? First, we
will only consider curves joining two points that have a shorter length than
the curves close to them.
ho!!
144 5. Integration on Surfaces
Proposition 5.6 give us
i-1
2-1 Z_i c Z)
It suffices to now apply Proposition 5.8 to each of the summands. U
5.4. Formula for the change of variables
One of the fundamental features of Lebesgue integration on Euclidean spaces
is the formula for the change of variables. It enables us to compute integrals
by passing from one domain of integration to another more convenient one.
In the context of surfaces we have an analogous result.
Theorem 5.14 (Change of variables formula). Let q5 : Rl - R2 be a diffeo-
morphism between two regions of two orientable surfaces, and let f : R2 - lI8
be an integrable function. Then the function (f o )Jac b is integrable on
Rl and
fzf= f (f I
Proof. We consider the diffeomorphism b between the products Rl x (0, 1)
and R2 x (0,1) defined by b _ q5 x id(o 1). This difFeomorphism obviously
satisfies
IJacbI(p,t) _ IJaccbl(p), d(p,t) E Rl x (0, 1).
By using this equality, property E of the integration on S x I[8 listed in
Proposition 5.4, and definition (5.1), we are done. O
Example 5.15 (Invariance under rigid motions). Let S be an orientable
surface, R. a region of S, and ¢ II83 -* ][83 the rigid motion given by
q5(x) = Ax -I- b with A E 0(3) and b E I[83. We know that q5(S) is another
orientable surface and that ¢(R) is a region of it. Then, one can view
-* (R) as a diffeomorphism between the two regions. If f is an
integrable function on ¢(R), we have, by the change of variables formula,
1=
since, if p e R. and el, e2 E TS form an orthonormal basis,
Jacq5I(p) = I IAe1AAe2I _ IdetAl = 1.
In particular, if we choose f - 1, we deduce that A(S(R)) = A(R); that is,
the area of a compact surface is invariant under rigid motions.
5.5. Fubini's theorem and other properties 145 218 7. Intrinsic Geometry of Surfaces

Example 5.16 (Odd functions on the sphere). Any rigid motion q5 :1183 Exercise 7.16. T Let b and b be two self-adjoint endomorphisms of a Eu-
][83 fixing the origin fixes the unit sphere S2 as well. Applying the invariance clidean vector plane. Show that det ¢+det b+trace btrace b-
of the previous example to this particular situation, we have trace b o b.
Exercise 7.17. T If b and b are two definite self-adjoint endomorphisms of
a Euclidean vector plane and det q = det b, prove that +b) 0 and
For example, if q5 _ is the antipodal map, we obtain that equality occurs if and only if b _ -s/i.
Theorem 7.18 (Cohn-Vossen's rigidity theorem). Any isometry between
ff(p)dp= ff(_P)dP two ovaloids is necessarily the restriction of a rigid motion of Euclidean
space.
for each function f integrable on the sphere. Then, if f is an odd function,
that is, if f(-p) _ -f(p) for each p e S2, one has
Proof. Let f : S -k S' be an isometry between two ovaloids S and S'. Since
the map f is a diffeomorphism, we can construct from it the symmetric
ff(p)dp=O. endomorphism A(f )p, given by
Important examples of odd functions on the sphere are the height functions _ -(df )p 1 ° (dN")f() ° (df), b'p e S,
f(p) _ (p, a) with a e 1183, or, in general, any restriction to the sphere of an
odd degree homogeneous polynomial. which, as was seen at the beginning of this section, has trace 2(H' o f) and
determinant KF o f. Now then, using Gauss's Theorema Egregium 7.10, we
Example 5.17 (Area of a parallel surface). We pointed out in Remark 4.29 have
that, if t 0 is small enough, then the map Ft : S -* 1[83, given by Ft(p) _ det A(f) = K' of - K,
p -I- tN(p), where S is a compact surface and N is one of its Gauss maps, is
a diffeomorphism onto its image St = Ft(S), the parallel surface at distance since f is an isometry. Then, the second Minkowski formula of Theorem 6.11
t. Moreover, can in this case be written as
(JacFt)(p) = I(dFt)p(ei) A (dFt)p(e2)I = 1 - 2tH(p) +t2K(p) 2 H = -2 (p, N(p))K(p) dp
J J
Applying the change of variables formula to the constant function 1 on St,
we obtain the following expression for the area of the parallel surface at _ - J (p, N(p)) det(dN)P dp - J (p, N(p)) det A(f) p dp.
distance t:
On the other hand, Herglotz's integral formula gives
A(St) = A(S) - 2t H(p) dp + t2 K(p) dp,
Js Js
2 H' o f - N(p)) [trace (dN)Ptrace A(f) P -trace (dN)A(f)] dp.
where H and K are the mean and Gauss curvatures of S, respectively. Js s
Subtracting these two equalities and applying Exercise 7.16-notice that
5.5. Fubini's theorem and other properties both (dN)P and A(f) P are self-adjoint endomorphisms of the plane TpS-
we obtain
We continue the development of some other properties of the integral of
functions on orientable surfaces as defined in Section 5.3. They will be used 2 f(H' o f - H)(p) dp = (p, N(p)) det[A(f )P + (dN)] dp.
in the proofs of the global theorems that we will deal with in later chapters. J
In this section, we will prove the corresponding version of the very useful and The integrand on the right-hand side in this equality is non-negative ev-
classical Fubini theorem on 1152. Later, we will recall some properties con- erywhere. The reason is that, on one hand, the support function based on
cerning continuous and differentiable dependence on parameters and show the origin-which we may consider in the inner domain of S-is negative,
how to use them with some important examples. according to Remark 6.2, and, on the other hand, the function taking each
7.4. Rigidity of ovaloids 217 146 5. Integration on Surfaces

We now consider the particular case where the vector-field V : S -* 113 Theorem 5.18 (Fubini's theorem). Let R. be a region of an orientable sur-
is given by face, a < b two real numbers, and h : R x (a, b) -* ll8 an integrable function
-
V (p) = pT = p (p, N(p))N(p), by e S, on the product R. x (a, b). Then, for almost all t e (a, b), the function
p E R H h(p, t) is integrable on R and, for almost all p e R, the function
where N is the Gauss map of S. Then
t e (a, b) H h(p, t) is integrable on (a, b). Moreover, the functions
(dV)(v)T = v - (p,N(p))(dN)(v), Vv E TPS. b

Hence pERH h(p, t) dt and t E (a, b) H h(p, t) dp

2

kZ ° f)(p)((dV)(e) , ei) =2 (H' ° f)(p) - (p, N(p)) trace (dN)pA(f)p are integrable see Remark 5.9 on R and on (a, b), respectively. Finally,
2=1 (fb
From this, Proposition 7.14 above implies, in this particular case, the integral Lx(a,b) h(p, t) dp dt = J h(p, t) dt I dp = 6 (f h(p, t) dp I dt.
formula
J
2J (0(H'o f )p, p) dp = -2J (H'of)(p) dp+J (p, N(p)) trace (dN)A(f)P dp. Proof. Since R. is relatively compact, we can cover it by a finite number
X 1(U1), ... , X ' of images of parametrizations defined on open subsets
But, we also have UZ, with i = 1, ... , k, of JR2. Let {Xi,. , xk} be the corresponding partition
of the unity. Then each product hx2 is integrable on R x (a, b) and
f(V(H'of)p, p) dp=-2J (H'of)(p) dp-2 f (p, N(p))H(p)(H'°f)(p) dp,
h(p, t) dp dt = h(p, t)x2 (p) dp dt.
as seen in Exercise (20) of Chapter 6. By eliminating the gradient term in fZx (a,b) i=1 7Zx (a,b)
the last two equalities, we obtain
By property D of Proposition 5.4 on integration on S x R, we obtain
2Jf(H'of)(p)dp
_- f (p N(p) )trace (dN)A(f)P dp -4 Js(p, N(p))H(p)(H' ° f)(p) dp f Zx (a,b)
h(p, t) dp dt =
i=1 XZ(UZ) x (a,b)
h(p, t)x2(p) dp dt.

Taking the definition of mean curvature into account, we deduce the follow- Reasoning as in the proof of Proposition 5.8, we see that each product
ing result. X i x id(Q,b): UZ x (a, b) - X i (UZ) x (a, b) is a diffeomorphism with absolute
value of the Jacobian equal to Xu A Xv , for i = 1, ... , k. Hence, we may
Proposition 7.15 (Herglotz's integral formula). If f S -> S' is a lo- use it in definition (5.1), and we have
cal isometry between two compact surfaces in Euclidean space, the integral
formula

2
Js
H' o f -
s
N(p)) [trace (dN)ptrace A(f) P -trace (dN)A(f)] dp
(5.3)
7Zx (a,b)
h=
i=1 f Z x (a,b)
(Xi° xz) (h o (xz x id(Q,b))) I Xu A Xv l .

is valid, where H' is the mean curvature of S' and A(f) P is the endomor-
phism
-(df)p 1 ° (dN')f() ° (df), b'p e S,
where N' is the Gauss map of S'.
As a consequence of Herglotz's integral formula and the following easy,
but non-trivial, exercises concerning endomorphisms of a Euclidean vector
plane, we will be able to deduce the rigidity result for positively curved
We now apply the classical Fubini theorem to each integral on the right-
hand side. This theorem assures us that the integrands of the right-hand
side are integrable with respect to each pair of variables for almost every
fixed value of the third variable. Furthermore,
(Xi o XZ)(h o (XZ x id(Q,b)))IXu n Xv
fUx(a,b)
= o X XA
X(fb

h(X z, t) dt du dv.
surfaces that we mentioned after the rigidity of spheres Theorem 7.12. Ui
5.5. Fubini's theorem and other properties 147 216 7. Intrinsic Geometry of Surfaces

This expression, together with (5.3) and Proposition 5.8, gives us the first a tangent field on a surface, as given in Section 6.3, we have
equality. On the other hand, if we apply Fubini's theorem to each summand (divW)(p) _ ((dW)X(q), X(9)) + ((dW)pXv(9), Xv(q)),
of (5.3) in the second possible way, we have
Z) (W,X)(q) + (
(Xi o X (h o (xi x id(Q,b))) Xu A Xv I
fZx(a,b) But, from (7.6), we have
b Wu = muXu+nuXv+mXuu+nXvu, WX = mvXu+nvXv+mXuv+nXvv.
= la (x2 o XZ)h(XZ, t) IXu A Xv du dv dt.
Taking Lemma 7.13 and (7.5) into account, the two equalities above imply
(fui
Using Proposition 5.8 again for the expression inside the parentheses and (divW)(p) _ (mu+rtv)(q).
substituting in (5.3), we obtain the second equality. On the other hand, recalling the relation (7.7), this expression becomes
Example 5.19 (Volume enclosed by a parallel surface). Let S be a compact (div W)(p) _ [(eu + f)b + (fu + gv)c + e'bu + f'(cu + bv) + g'cv] (q)
surface and let e > 0 be such that NE(S) is a tubular neighbourhood. In
From the particular way we have chosen the parametrization X', we have,
Remark 4.29, we introduced parallel surfaces St at distance t e (0, e). If
according to Lemma 7.13, that
Stt is the inner domain determined by St, we have, taking Remark 4.28 into
account, that f'(q) = 0, f;(q) = gu (q), and f(q) = e( q).
vole - vol SZt = vol Vt (S) = 1. Substituting, it follows that
Jvt(s) (div W)(p) _ [(eu + gu)b + (e, + gv)c + e'bu + g'cv] (q).
But the map F : S x (0, e) -* VE(S), given by F(p, t) = p + tN(p), is a
diffeomorphism. Therefore, if we use it in definition (5.1) of the integral on Finally, the properties mentioned in the lemma concerning the coefficients
S x I[8, we obtain e', f', g' at the point q also imply that
e'(q) _ (ki ° f)(p), g'(q) _ (k2 ° f)(p),
JSx(O,t) IJacFI.
vol SZ - vol SZt =
2 (H' o X')(9) _ (e' + 2 (H' o X')(9) _ (e', + 9v)(9)
But we have already computed the differential of the map F in (4.2). It
follows from this that, if (p, t) E S x (0, e), then Thus, we conclude that

It) = det((dF)(P,t)(el, 0), (dF)(t)(e2, 0), (dF)(t)(0, 1)) (divW)(p) = 2(0(H'o f)(p),V(p))
(7.8) + (k'1 °.f)(p)((dV)p(ei), ei)+(ka°.f)(p)((dV)p(ea), ea),
= 1 - 2tH(p) + t2K(p), for each p e S and where {ei, e2} is a basis taken by (df)p into the principal
using, for example, an orthonormal basis of the principal directions at the directions of S' at the point p' = f(p). This allows us to state the follow-
point p of S. Hence, the Fubini Theorem 5.18 above implies ing proposition, which is an immediate consequence of the first part of the
ft(f(i dp\
divergence theorem for surfaces.

/s
vo1SZ - vo1SZt = - 2tH(p) + t2K(p)) I dt Proposition 7.14. Let f : S -> S' be a local isometry between two compact
surfaces and let V be a differentiable tangent field on S. Then

f
2
=tA(S)-t2j H+-j K. f(V(H' ° .f), V) _ - (k° .f)(p)((dV)(e), e) dpi
3 s Z
S Z-1
From now on, we will refer to 5.18 simply as Fubini's theorem.
where {ei, e2} is an orthonormal basis of TS at the point p of S that is
Example 5.20 (Integration in polar coordinates). Assume now that the mapped by (df)to the principal directions of S' at f(p) and kl and k2 are
compact surface being considered is the unit sphere S2. We know that the the corresponding principal curvatures of S'.
7.4. Rigidity of ovaloi ds 215
for some differentiable functions b, c defined on U, we have that
W X = mXu + rtXv,
where the new coefficients m, n are functions defined on U satisfying
1b 1E' F'1 (e' f' (b
()=a(f)C)=_F,
(7.7)
c') f' g')c
This proves that W X = W o X is differentiable on U C R2 and, hence, that
the tangent field W is differentiable.
Once we have checked that this new tangent field W, which we have
defined on the first surface S in terms of the tangent field V and the dif-
feomorphism f : S -k S', is differentiable, we will compute its divergence in
the particular case when f is a local isometry. Next, when S is compact,
we will apply the divergence theorem (Theorem 6.10). To do all of this, it
is worth recalling the following facts.
Lemma 7.13. Let p e S be an arbitrary point of a surface. Then there
exists a parametrization X : U -> S of S, defined on an open subset U of
1182, such that X (q) = p for some q e U and such that the following hold.
a: The vectors {Xu(q), Xv(q)} form an orthonormal basis of TS.
b: The vector X(q) vanishes and the vectors and X(q) are
perpendicular to TS.
c: The coefficients E, F, G of the first fundamental form associated to
X have vanishing first partial derivatives at q.
d: The first partial derivatives of the coefficients e, f,g of the second
fundamental form associated to X satisfy
f,.(9) and 9u(9) =
Proof. In fact, it is enough to take a suitable neighbourhood of the point
p in the surface S as a graph on the tangent plane at this point and to
take the corresponding parametrization, as we did in the proof of Hilbert's
Theorem 3.47.
We now proceed to calculate the divergence of the tangent field W. Pick
p e S. Then, we apply Lemma 7.13 to the second surface S' at the point
p' = f(p) and obtain a parametrization X' of S' with the required properties
a, b, c, and d. We choose as a parametrization X of the first surface S the
map X = f -1 o X'. Hence, we have X' = f o X. From property a of X' and
the fact that (df) p is a linear isometry, the vectors Xu (q) and Xv (q) are an
orthonormal basis of TS. Consequently, by the definition of divergence of
148 5. Integration on Surfaces
punctured unit ball B3 - {0} is diffeomorphic to the product S2 x (0, 1),
since the map F : S2 x (0, 1) -* B3 - {0}, given by
F(p, t) = p -I- tN(p) _ (1 - t)p,
is a diffeomorphism. The absolute value of its Jacobian was computed in
Example 5.19 above and, in this particular case, is
I= 1 - 2tH(p) + t2K(p) _ (1 - t)2,
since the (inner) mean curvature and the Gauss curvature of the unit sphere
are both the constant function 1. Then, by definition (5.1), we have
f f=f _f=f
B3 3 {O} 2 X ( O,1)
f ((1 - t)p) (1 - t)2 dp dt.
Now using Fubini's theorem, we obtain
dp) dp.
1B3 f - Jol p2 \f2 f(Pp
This is the so-called polar coordinates integration formula for functions de-
fined on the unit disc.
We will complete this section by gathering some results relative to con-
vergence and dependence of parameters. Our integration inherits these
properties directly from Lebesgue theory. Finally, we will give an exam-
ple to show how it can be useful in the study of the interaction between the
geometry and the analysis of surfaces.
Remark 5.21 (Dominated and monotone convergence theorems). The main
properties of Lebesgue integration carry over without difficulty to the inte-
gration of functions on surfaces. For instance, the following is true.
Let R. be a region of an orientable surface and let fn : R -
II8, n e ICY, be a sequence of integrable functions on R such
that {ffl}flEN converges pointwise to a function f and such
that IfmI < g for some function g integrable on R. Then
f is integrable and
lim
n-*°°JR
fnf f .
In fact, this is the dominated convergence theorem, which can be obtained by
using definitions (5.1) and (5.2) from the analogue in the Lebesgue theory
on Il83. Similarly, we have the monotone convergence theorem, which says:
Let R be a region of an orientable surface and let f: R
][8, n E ICY, be a monotone sequence of integrable functions
5.5. Fubini's theorem and other properties 149 214 7. Intrinsic Geometry of Surfaces

on R that converges pointwise to a function f such that
lim
exists and is finite. Then the function f is integrable on
R and
J fnf
lim f.
n-'O° R
Remark 5.22 (Continuous and differentiable dependence on parameters).
We will utilize the dominated convergence theorem, although it could be
done directly from the definitions as in the preceding remark, to prove some
easy versions, which will be adequate for our needs, of the theorems on
the continuous and differentiable dependence of integrals whose integrands
depend on a real parameter.
Let R. be a region of an orientable surface, I an interval
of ][8, and f : R x I -* ll8 a continuous function. Taking
into account part E of Proposition 5.4, we define a new
function h : I -* ll8 by
isometrics. In fact, this is impossible, as shown in Exercise (6) at the end of
this chapter. In the following section, we will see that rigidity of spheres is
a property shared by all positively curved compact surfaces, but the proof
of this is considerably more involved than in the case of spheres.
7.4. Rigidity of ovaloids
Suppose that S and S' are two orientable surfaces and that f : S -* S' is a
diffeomorphism. If V : S -* I[83 is a differentiable tangent field on S, we can
define another tangent field W : S -f Il83 on the same surface by
W(p) = A(f)pV(p), b'p E S,
where A(f) P : TS -* TS is the linear endomorphism given by
= -(df);1 o (dN')f() o (df), p e 5,
and N' is a Gauss map for S'. This is awell-defined tangent field since
(dN')q can be thought of as an endomorphism of TyS' for each q e S'.
Furthermore, it is also clear that
detA(f)p=K'(f(p)) and traceA(f)P=2H'(f(p)),

h(A) = f .f (p,A) dpi VA e l.

Then h is continuous.
We now prove this assertion. If A E I and {Afl}flEN is a sequence in I
converging to A, we can assume that both the sequence and its limits belong
to a compact interval J C I. Since f is continuous, the sequence f(-, fin)
of functions defined on R converges pointwise to f(-, A). Moreover,
since similar matrices have the same trace and determinant, and that
W (p) E TS, `dp E S.
That is, W is another tangent field to S depending on the original field V
and on the diffeomorphism f. We now show that this vector field W is
differentiable. For this, let X : U -k S be any parametrization defined on
an open subset U of ][82. We know that X' = f o X is a parametrization
of S' with the same domain as X and that the partial derivatives {X,

IC max (9,u) ER x J
f(q,p).
and {X', provide, at each point of U, bases of the tangent planes TS
and TX'S', respectively. Using the usual notation, we set
WX = W o X, VX = V o X, and A(.f)X = -(df)xl ° (dN')' o (df)x.
Using Remark 5.21, we have
The matrix of the linear map (df)x : TS -* TX'S', relative to these bases,
lim h(en) = n->oo
n->oo
lim f(p,A)dp= f(p,A)dp=h(A). is the identity matrix since
j7 J7

X'u = (.f o X)u = (df)X(Xu) and x'v = (.f o X)v = (df)X(Xv),

Thus, h is continuous at the point A. This same continuous dependence
on parameters can be used to obtain results concerning the derivative of and, thus, the matrix a(f) of A(f )X is the same as the matrix of the endo-
an integral when the integrand depends differentiably on a parameter. For morphism (dN')x', which, according to (3.11), is
example, we have the following. fe''
Let f : R x I -* I[8 be a continuous function, where R is a
a(f) _ - (F' G') -1 ( g'
J
region of an orientable surface and I is an open interval The coefficients are clearly differentiable functions defined on the open sub-
of R. Suppose that, for each p e R, the function set U of ][82. In conclusion, if
aEIH.f(p,A) VX = bXu + cXv,
73. Gauss's Theorema Egregium 213 150 5. Integration on Surfaces

is differentiable; denote its derivative by (f/A) (p, A). If

the function
-
of : 1 x I -- R
as
is continuous, then the function h : I -p 1R, defined by

h(A) = f f(p, A) dpi

is of class Cl and its derivative is

h'(A) = f(P,A)dP.
Figure 7.1. Non-congruent isometric surfaces We now check that this statement is true. For a E I we choose a compact
interval J C I and a sequence {fin}nER contained in J converging to A.
Theorem 7.12 (Rigidity of spheres). Let S2 (r) be a sphere of radius r > 0 Then
and let S be a connected surface. Any local isometry f . §2(r) -> S is the
restriction of a rigid motion of 1R3. In particular, S is another sphere of
h(An) - h(A) -
[ f(p, An) - f(p, A) d

radius r. Since f is differentiable with respect to the a-variable, the integrand on the
right-hand side converges pointwise to (af/aA)(p, a). Furthermore,
Proof. Since f is a local isometry it is, in particular, a local diffeomorphism
f(p,AAn)--fA(p,A)l
and consequently it is also an open map. Thus, its image f (S2 (r)) is an open =l
subset of S. It is a closed subset as well since §2(r) is compact. Since S
is connected, f is surjective. On the other hand, we know that the sphere where we have utilized the Cauchy mean value theorem with 9n representing
S2(r) has constant Gauss curvature equal to 1/r2. The Gauss Theorema a point between A and An, and the continuity of 8 f /aA. By the dominated
Egregium that we have just proved implies that S also has constant Gauss convergence theorem established in Remark 5.21, it follows that
curvature 1/r2. Hence, applying the Hilbert-Liebmann theorem stated ei- h(An)h(A)
lim - h(A) of (p,A) dpi
ther in Corollary 3.49 or in Corollary 6.12, S is also a sphere of radius r.
Thus f : 52(r) -> S is a local isometry between two spheres of the same
n-*oo An - A =j
radius. From the fundamental theorem of the local theory of surfaces- and the integral on the right-hand side is continuous in a by the continuous
Example 7.9-we deduce that f is the restriction of a rigid motion. dependence of parameters that we have just proved.
Remark 5.23 (An application to the sphere). Let O C ][83 be an open subset
This rigidity of the sphere means that its geometry is completely deter- of Euclidean space containing the origin and f O - ][8 a differentiable
mined by only its first fundamental form. In this sense, it is a property that function. For r > 0 small enough, the closed ball centred at 0 with radius
is not exclusive to spheres, although there are compact surfaces that do not r is included in O and, hence, S2(r) C O as well. Thus, we can define a
possess it. For example, the surfaces in Figure 7.1-which appear in almost function q by
all books dealing with this topic-axe two compact surfaces of revolution
that are isometric but not congruent. The isometry suggested by Figure 7.1 r H q(r) -
4 r2
f(p)dp,
fixes all the points under the tangent plane of the surface along the circle
for r > 0 near zero. Let r and ro be two positive small numbers. The
of vanishing Gauss curvature and coincides, when restricted to the points
dilatation with centre the origin and ratio r/ro provides a diffeomorphism
over this plane, with the reflection about that tangent plane. Obviously,
between S2(ro) and S2(r). Then, applying the change of variables formula
this isometry does not come from a rigid motion of R3 since it has too many
(Theorem 5.14), we have
fixed points. Another question is whether we can pass continuously from one
to the other of these surfaces through isometric surfaces or whether we can
pass from the identity map to that isometry through a continuous family of f§2(r)
f(p) dp -
r2
2 Jf(p)
0 2(ro) TO
dp.
5.5. Fu bini's theorem and other properties 151 212 7. Intrinsic Geometry of Surfaces

Therefore If we apply Lemma 7.11 to the parametrization X, we obtain

(r)=f2(ro) .f( (7.2) (X v)u - (X), Xu) = (K o X)(EG - F2)
Then, the composite map X' = f o X is a parametrization of the second
which shows, by the dependence theorems, that q5 is a continuous and dif- surface S'. Thus, applying Lemma 7.11 to the second surface S' and to the
ferentiable function and that it extends to zero. By again applying the con- parametrization X', we can write
tinuous dependence result and taking into account that the area of S2(ro) is
4irro, we have (7.3) ((X') u - (X'), X'u) = (K' o X')(E'G' - F'2).
On the other hand, taking partial derivatives of X' and using the chain rule,
lim 1 2 f(p) dp = (O) = f(O).
r-'0 47r1' JS2(r) we have
From the same expression we deduce X'u = (df)x(Xu) and X'= (df)x(X),
and thus, computing scalar products and taking into account that f is a
qV (r) 4 r2
I z
(ro)
((Vf)(rop),o) dp. local isometry, we have
(7.4) E' = X'uI2 = IXul2 = E, G' = X'I2 = = G,
From this, the differentiable dependence theorem assures us that ' is also F' = (X', (X, X) = F.
differentiable. Moreover,
Differentiating again the three equalities in (7.4) with respect to the two
variables, we obtain that the six scalar products between the second partial
=
f ((Vf)(O), ) dp =0,
derivatives and the first partial derivatives of X' and of X coincide. Hence,
since (df)x is a linear isometry at each point,
where O is the gradient of functions on 1183-see Exercise (5.36) at the end
of this chapter-and where the last equality comes from, for instance, Ex- (7.5) (X)T (df)x(X), (Xv)T = (d1)x(X v),
ample 5.16. In addition, we can define a vector field X on an open neigh- (X' T_(df)X(v-T\
bourhood of 0 by the expression X (p) _ (Vf)(p), which is clearly differ- Consequently,
entiable. Moreover, div X (p) = To (Lf)(p). Using the divergence Theo-
/
rem 5.31 on the ball Bi(ro), which will be proved later, we obtain (((' ')u, X'u) = CX' v' X'u}u - X' V, X' u/

. 4i p (Of)( ro p)dp _ ((df)x(X), (df)x(X)),. - ((df)x(X), (df)x(Xfj)

30
B2 (ro)
-
ivT x) - (xf,x) =
Taking the limit as r goes to zero, we have by the continuous dependence of
parameters that and, similarly,

= 4r3(Of)(0)vo1B2(ro) _
(f)(o) These two equalities, together with (7.3), (7.4), and (7.5), give us

Hence, applying l'Hopital's rule, we have

KoX = K'oX' = K'ofoX
on U. Therefore, K' o f = K on the open subset X (U) of S. Since the
1
lim-
r 1S2(r) f(p) dp - f(0)) - ( f)(O)
6
surface may be covered by open sets of this type, the theorem is proved. E
( 4 r2 As an immediate consequence of Gauss's Theorema Egregium, we can
With the two limits that we have just computed for integration on surfaces, see, for instance, that an open subset of a sphere cannot be isometric to an
one will be able to solve Exercise (12) at the end of this chapter. This is a open piece of a plane. Furthermore, we deduce-again--a new property of
good example of the usefulness of the dependence theorems. spheres that, in some texts but not in all of them, is called rigidity.
7.3. Gauss's Theorema Egregium 211
Proof. Let X U -* S be a parametrization of our surface S. We will
work on the open set X (U) covered by it and will use the same notation as
in the proof of Proposition 3.43. We know that on the open set X (U) we
have a Gauss map N = NX o X-1, where NX = Xu A A and
that the partial derivatives {X, Xv} provide a basis of the corresponding
tangent planes TS at each of its points -we will always work with functions
defined on the open subset U of I[82, although without explicitly indicating
the point. The two fundamental forms of the surface with respect to this
basis are represented by symmetric matrices whose coefficients were denoted
by E, F, G and e, f, g, respectively. We first prove the following preliminary
equality.
Lemma 7.11. Let U be an open subset of II82 and let F : U -* II 83 be a dif-
ferentiable map taking values on a surface S, for example, a parametrization
of S. If K is the Gauss curvature of S, we have the following expression,
which is valid at each point of U:
((F v)u - (Fu )v, Fu) _ (K o F)(IFuI2IFvI2 - (Fu, Fv)2),
where the superscript T means the part of the corresponding vector belonging
to the tangent plane of the surface at the given point.
Proof. In fact, we have, if N is a local unit normal field and NF = N o F,
F v = Fvv - NF) NF = Fvv + (F, NFv) NF,
F v = Fuv - Fuv NF}NF = Fuv + (Fu, NFv)NF,
since (Fu, NF) = (F, NF) - 0. Taking partial derivatives of these two
expressions with respect to u and to v, we obtain
CFvv)u = (F, N ' )uNF + (F, NFV)NFU,
CF v)v = Fuv + (Fu, NFV)VNF + (Fu, NFV)NFV.
After scalar multiplication by Fu and subtracting, we have, by the equality
of mixed derivatives, that
CF v)u - (F), F) = (NFU A NFv Fu A F).
Thus,
NFu A NFv = (dN)F(F) ^ (dN)F(F) = det(dN)F(Fu n Fv),
and, using (3.4), we have
NF'uANF'v = (KoF)(FAF),
and the proof is complete.
152 5. Integration on Surfaces
5.6. Area formula
We will prove, in the following section, the divergence theorem. Many dif-
ferent proofs of this result can be found in the literature. We will base ours
on a particular case of the so-called area formula, an integral formula that
generalizes the change of variables formula for the Lebesgue integral, when
the transformation in question is no longer a diffeomorphism, but only a
differentiable map. This area formula is of interest itself and will enable us
to obtain more interesting consequences in what follows.
Consider a bounded open subset O of R3, a differentiable map q5: O -p
113, and a function f : O -p 1R. If x e JR3 is a regular value of the map
-with a definition similar to Definition 4.5 then the set of the inverse
images of the point x with respect to b is discrete. This a direct consequence
of the inverse function theorem. Then, since O is compact, this is a finite
subset of O. This allows us to define the following function
n(q5,f) :113-q5(N) ---*J
by the expression
with the convention that the sum is zero when x q5(O), where N C O is
the set of the points of O that are critical for , that is,
N= {xe(Jacq5)(x) =O}.
The three-dimensional version of Sard's theorem tells us that ¢(N) is a
subset of measure zero in Il83, and thus the function f) is defined almost
everywhere on Il83, that is, except on a set of measure zero. Note that, if ¢
were a diffeomorphism onto its image, then n(cb, f) would be defined on the
whole of I[83 and would coincide with the composite map f o -1 on ¢(O).
For this reason, the following result can be thought of as a generalization of
the change of variables formula for the Lebesgue integral on 1183.
Theorem 5.24 (Area formula). Let ¢ : O - 1[83 be a differentiable map,
where O is a bounded open subset of 1183 and f O -* ][8 a function on
0. Suppose that the product fiJac q is integrable on O. Then the function
f) is integrable on 1[83 and
n(, f) = ff(x)lJacI(x)dx.
1R3
Proof. Let V be a relatively compact open subset of ][83 including O such
that cb extends to a differentiable map cb: V -p ][83. Choose another open
subset W of 1[83 such that O C W C W C V. Suppose first that all the
points of W are not critical for cb. Then, for all p e W there exists, by the
5.6. Area formula 153 210 7. Intrinsic Geometry of Surfaces

inverse function theorem, an open neighbourhood Vp of p in V such that
is a diffeomorphism of Vp onto its image q(VP). Thus, the family {V}
with p e W is a cover by open sets of the compact O. Let b > 0 be the
Lebesgue number of this cover and let P = {C1,. . . , C1 } be a partition by
cubes of the compact 0 -that is, O C U=1 CZ and the interiors of the CZ
are disjoint with norm smaller than b and all of them included in W. Each
of these cubes is contained in some open set of the cover. Consequently,
c2 is a diffeomorphism onto its image for each i = 1, ... , n. On the other
hand, since fiJac cbl is integrable on O, it is also integrable on each CZ n O.
Therefore, the formula for the change of variables for the Lebesgue integral
implies that f o q5-1 is integrable on (CZ n O) and gives
f ino
for all i = 1, ... , n. Adding these n equalities, we obtain the integral equality
of the theorem since f) vanishes outside of q(O) and, in this particular
case foq1.
The case where there are critical points of ¢ can be deduced from the
previous case as follows. Consider the set
M = {p e V I 0}.
Then K = ,M fl O is a compact subset of M. For each k e ICY, we take an
open set Ok such that
K C Ok, vol Ok = vol K + , and Ok is decreasing.
By Proposition 4.34, there exists another open set Vk such that
and
This, together with the fact that ITPS + t(dN) is an isomorphism of the
plane TS for all p + tN(p) in the tubular neighbourhood, imply
(dq5)+N()(N(p)) = N'(f(p)) and (dq5)P+tN(P) IT S - (df),
valid for each p E V and tI <e. Consequently, we have that b is a diffeomor-
phism between two open subsets of II83 whose differential is a linear isometry
at each point of its domain N6(V). Thus, from Exercise 7.6, b must be the
restriction to NE (V) of a rigid motion b of II83. Thus, since the initial point
po E S was arbitrarily chosen, we have shown that, for each p E S, there
exist an open neighbourhood V of p in S and a rigid motion 'b(x) = Ax + b,
x E I[83, such that = liv and N' o liv = A o Niv. Keeping in mind that
S is connected, we deduce that all these rigid motions corresponding to the
various neighbourhoods V coincide. O
Example 7.8. Let P and P' be two planes in 1[83 and let f U -* P'
be a local isometry defined on a connected open subset U of P. Since
the second fundamental forms of P and of P' are identically zero-for any
orientation-f always preserves the second fundamental forms. Therefore,
f is the restriction of a rigid motion.
Example 7.9. If S and S' are two spheres with the same radius r > 0, U C
S is a connected open subset of S, and f : U -> S' is a local isometry, then f
necessarily preserves the second fundamental forms v and v', associated, for
instance, to the two inner normal fields, since, according to Example 3.20,
af(p)«df)p(u), (df)(v)) = ((df)(u), (df)(v)) = (u,v) _
for each p E S and u, v E TS. Then f will be the restriction of a rigid
motion.

Then the open set O - Ok obeys the conditions of the above special case,
since W = O - V is an open set containing its closure and consisting of
7.3. Gauss's Theorema Egregium
regular points of Hence
q5.
As we already pointed out in the introduction to this chapter, intrinsic geom-
etry of surfaces is the study of those properties remaining invariant through
I3 (cio-' I-o f Jacb. isometries, or, more generally, through maps preserving the first fundamen-
tal form. It is quite remarkable that the Gauss curvature, as defined in
The integral on the right-hand side converges to
(3.4) and (3.5) from the second fundamental form, turns out to be the main
flJaccI invariant of this intrinsic geometry. Gauss wrote in the Disquisitiones that
o-x the following theorem, due to himself, is certainly excellent.
as k -* oo since f is integrable on O, IJac q is bounded on O, and the limit Theorem 7.10 (Theorema Egregium). Let f : S -* S' be a local isometry
of vol (Oj - K) is zero as k -* oo. As a consequence we have that

k f n(10_, f1_) -
3 o-x f J between two surfaces S and S', and let K and K' be their respective Gauss
curvatures. Then K = K' o f. Equivalently, local isometries preserve the
Gauss curvature.
7.2. Rigid motions and isometries 209 154 5. Integration on Surfaces

linear isometry of IIS3. Prove that F is the restriction to O of a rigid motion On the other hand, from the definition of n(q, f) and the fact that q5 is
of 3. differentiable on the compact O, each regular value of q5 has finitely many
pre-images in O, and we deduce that, for each x e ]E83 - (M), we can find
It can be deduced from the second of these exercises that there are open a k E ICY such that
subsets of the plane and of the cylinder that are isometric. Such an isometry
cannot come from a rigid motion of that is, it is not a congruence, n(1o-o ,.fro-)(x) = n(q5,.f) (x), k>
since congruences take planes into planes. However, we have the following Therefore
converse of Proposition 7.1.
.fro-ok) -i n(q, f)
Theorem 7.7 (Fundamental theorem of the local theory of surfaces). Let pointwise ask --> oo and moreover, when f > 0, this convergence is mono-
S and S' be two orientable surfaces, with S connected, and let N and N' be tone. Thus, in the case where f is non-negative, we infer that f) is
Gauss maps for each of them with associated second fundamental forms a integrable on I[83 by the monotone convergence theorem, and furthermore
and a', respectively. If f : S -* S' is a local isometry preserving the second that
fundamental forms, then there exists a rigid motion ' : i3 -* IR3 such that
'is = f, that is, f is a congruence. 3 n(, f)
o-x f - IJac .

Proof. Let po E S be an arbitrary point on the first surface and let f(o) E
fnot know anything
In the case where we do about the sign of f, the domi-
nated convergence theorem gives us the same equality since
S' be its image. Lemma 4.25 along with the fact that f is a local difFeomor-
phism imply that there exist two connected open neighbourhoods V of po (o-o' In(q, INow
in S and V' of f(o) in S' such that f : V -* V' is an isometry and there is it is enough to establish that
a number e> 0 such that N6(V) and N6(V') are tubular neighbourhoods.
We define a map q5: N6(V) -* N6 (V') by I IJacI = fiiJaci
q(p + tN(p)) = f(p) + t(1V' ° f)(p), b'p E V `dItI <e.
since IJac q = 0 on K.
This b is clearly differentiable and bijective since it is nothing more than
the composite map F' o (f x Id(_E,f)) o F1, where F : V x (-e, e) -* N6(V) Exercise 5.25. T Let R. be a region of an orientable surface S, let a < b be
and F' : V' x (-e, e) -* NE(V') were defined in (4.1). Using the definition two real numbers, and let q5: R x (a, b) -p I[83 be a differentiable map. If we
itself or the commutativity relation ¢ o F = F' o (f x Id(_6,6)), we compute define
the differential of b and see that A/= {(p,t) E R x (a,b) I I= 0},
(7.1) = N'(f (p))
(N(p))
show that N) has Lebesgue measure zero in I[83.

(dq5)+tN() ° [Irs + t(dN)p] _ [ITf()s' + (df), Exercise 5.26. T Let S be a compact surface and let P be a plane in I[83
perpendicular to a unit vector a e 2. Prove that almost all the straight
for each p E V and ti But we are assuming that f preserves the second lines perpendicular to P intersect S transversely at an even number of points
fundamental form, and thus in such a way that the sign of (N, a) changes alternatively when we move
((dN')f()(df)p(u), (df)p(v)) _ (df)(v)) along the line in a given direction.
= -a(u,v) = ((dN)u,v),
The first exercise above allows us to easily obtain the corresponding
for each p e S and u, v E TS. Therefore, since (df)P is a linear isometry, versions of the area formula for differentiable maps of R. x (a, b) to R3,
(dN')f() o (df)p = (df) o (dN), b'p E S.
where R is a region of an orientable surface, and for differentiable maps
between regions of orientable surfaces.
Applying this commutativity rule to the second equality in (7.1), we get
Theorem 5.27 (Area formula for products). Let q5 : R x [a, b] -> I[83 be a
(dq5)+N() 0 [ITs + t(dN)p] _ (df)p 0 [ITs + t(dN)p]. differentiable map, where 7 is a region of an orientable surface and a < b,
5.6. Area formula 155 208 7. Intrinsic Geometry of Surfaces

and let f be a function on R x (a, b) such that f Jac is integrable on First assume that the differential of f at each point of S is a linear isometry.
R. x (a, b). Then the function n(q, f), given by Then, since (f o a)'(t) _ (df)a(t)(a'(t)) for each t E [a, b], we have

n(q, f)(x) _ f(p, t) I(foa)'(t)I = I(df)a(t)(a'(t))I = Ia'(t)I,

and thus
6
is well-defined for each x e I[83, except on a set of measure zero, and is La(f ° a) = f Ia'(t)I dt =
a
integrable on ]E83 with
as required. Conversely, suppose that the map f between S and S' leaves
f(p, t) IJac (p, t) dp dt. the length of curves invariant. Let p E S be an arbitrary point, and let
n(, .f) - fx
3 (d,b) v E TPS. There exists a differentiable curve a : (-e, e) --* S for some e> 0
such that a(0) = p and a'(0) = v. By the hypothesis on f, we have, for
Proof. As a consequence of Exercise 5.25 we see that the function 1) each t E (0,e),
is well-defined except on a set of measure zero in I[83 above. To prove the
area formula in this case, it suffices to pick a diffeomorphism b O -->
R. x (a, b), where O is a relatively compact open subset of I[83, and to apply
Theorem 5.24 to the map q5 o b and the function f o b. Then
f 1a)'(u) du
t = Lo(f ° a) = L(a)
J
=It
Taking derivatives with respect to t at t = 0-the functions are differentiable
by the fundamental theorem of calculus-we obtain

0f (fo)IJac(o)I=f R3 I(df)p(v)I = I(foa)'(O)I = Ia'(O)I = lvi.

Therefore (df) preserves the scalar product. O
Now it remains to note that n(q o b, f o - n(q, f) and that, by definition
(5.1), we have
It is useful to realize that, since a linear map between Euclidean vector
spaces with the same dimension that preserves the scalar product must be
fo (1
= I IJacI O an isomorphism, any local isometry between surfaces is necessarily a local
diffeomorphism, by the inverse function theorem. A local isometry that is,
There is also an area formula for maps between surfaces. We give the in addition, a diffeomorphism will be called an isometry. If there exists an
following version. isometry between two surfaces, we say that they are isometric.
Theorem 5.28 (Area formula for surfaces). Let q5 : Rl -> R2 be a differen- Exercise 7.3. Prove that, if f : S -* S' is an isometry between two sur-
tiable map between regions of two orientable surfaces and let f be a function faces, then the absolute value of the Jacobian of f is identically 1. As a
on Rl such that the product f IJac q is integrable on Rl. Then the function consequence, if S and S' are compact, their areas are the same.
n(q, f ), defined by
Exercise 7.4. Let P be the plane with equation z = 0 in IIg3 and C the
n(q5,f)(q)= 1(p), right unit cylinder given by the equation x2 + y2 = 1. Show that the map
PEA-1 (9) f : P -* C given by f(x,y,0) _ (cosx,sinx,y), for each (x,y,0) E P, is a
is well-defined for each q E R2 except on a set of measure zero and is inte- local isometry.

1Rz
grable on R2. Furthermore,
Exercise 7.5. Prove that the composition of local isometries between sur-

f 1ja1 dp = f n(q, f)
faces is also a local isometry and that the inverse map of an isometry is an
isometry as well. Conclude that the set of all the isometries from a surface
onto itself is a group relative to the composition law. This is the group of
Proof. To prove this version, recall that, by definition (5.2), we have isometries of the surface.
Exercise 7.6. T Let F : O -* O' be a diffeomorphism between two con-
ft fIJac5I=fzix(o,i) f dpdt.
nected open subsets of IIS3 such that its differential at each point of 0 is a
7.2. Rigid motions and isometries 207 156 5. Integration on Surfaces

for each p E S. One arrives then at the following relation between the two But, if pl : 1Z1 x (a, b) -> Rl denotes the projection onto the first factor, the
second fundamental forms o and a' of S and S': integral on the right-hand side can be rewritten and thus
Qf() ((df )p(u), (df)(v)) = Qp(u, v), b'p E S du, v E TS,
f I IJacI - fix(O,1)(f x
as was established in Exercise 3.25. This is what we mean by the invariance
of the second fundamental form under rigid motions. As a consequence, the We now choose a diffeomorphism b : O -p f2 x [0,1], where O is a bounded
mean and Gauss curvatures of S and of S' are transformed into each other open subset of R3. If we denote by 7Z1 x (0, 1) -p O the composition
:

under f, that is, H' o f =Hand K' o f = K. We summarize the effect of -1 o ( x id(o,l)) and use the multiplicativity of the absolute value of the
the rigid motions of Euclidean space on surfaces in the following result. Jacobian, the integral on the right-hand side becomes
Proposition 7.1. Let S be a surface and let b : ][83 - II83 be a rigid motion. (fop1) (IJacbI IJacI.
Let S' _ ¢(S) and f = is : S -* Il83. Then the following hold. LX(O,l)
A: S' is a surface and f : S -* S' is a difj`'eomorphism. Applying the area formula (Theorem 5.27) to the map and the function
B: For each p E S and u, v E TS, one has (f ° Pi) IJac bI o , we see that

((df)(u), (df)(v)) (u,v).

C: For each point p E S and for tangent vectors u, v E TS, the
equality
af()((df)P(u), (df)p(v)) = a(u,v)
holds, where a and v' are the second fundamental forms of S and
of S', respectively.
In general, a diffeomorphism f : S -* S' between two surfaces is said
to preserve the first fundamental form when it satisfies condition B of the
proposition and is said to preserve the second fundamental form if it obeys
C. Thus, we have confirmed that congruences between surfaces preserve the
first and second fundamental forms or, equivalently, that congruent surfaces
have the same first and second fundamental forms.
If f S -* S' is a differentiable map between two surfaces S and S'
satisfying requirement B of Proposition 7.1, we say that f is a local isometry.
The reason for this name is contained in the following proposition.
(f opl) IJac/)I off) - n(, fopi) IJac/I.
But now it is easy to check that
Keeping this in mind together with definition (5.1), we obtain
f1 fIJacI=f2x(o,1)
The proof can be completed by noting that
[n(, I 0 P1)0 '](q, t) = n(, f)(q)
for each q E R2 and again using definition (5.2) of the integral on a surface.
D
An immediate application of these area formulas is the following three-
dimensional version of the so-called Chern-Lashoff theorem.
Theorem 5.29 (Chern-Lashoff theorem). Let S be a compact surface. Then
the following hold.

Proposition 7.2. Let f S -* S' be a differentiable map between two

surfaces. Then (df)p : TS -* Tf(p)S' is a linear isometry, for each p E S;
Af K+ > where K+ = max{0, K}.

that is, f is a local isometry if and only if f preserves the length of curves.
Proof. Let a : [a, b] -* S be a differentiable curve. Then the length of the
image curve on S' satisfies
L(foa) =f I(foa)'(t)Idt.
Bf IKI > 4r and, if equality is attained, then S has non-negative
Gauss curvature everywhere.
Proof. The proof consists simply in considering the compact set A C S
formed by the points of S with non-negative Gauss curvature and applying
the area formula (Theorem 5.28) for maps between surfaces to the Gauss
5.7. The divergence theorem 157
Figure 5.1. Vector field
map N : S §2 and the characteristic function xA of the set A. Then we
have
XA IJacNI = n(N, XA)
Is fS2
We must now check that xA IJac NI = K+ = max{0, K} and that, since N
is surjective even when restricted to A, as Exercise (5) of Chapter 4 showed,
the function n(N, XA) is always greater than or equal to 1 on 2. This proves
the first inequality. The second inequality and the case of equality follow
from the fact that IKI > K. O
5.7. The divergence theorem
The divergence theorem is also known as Gauss's theorem, Green's theorem,
Ostrogradski's theorem, and even Stokes' theorem. In fact, all these math-
ematicians either used this result in a fundamental way or contributed to
stating and to proving it. It is one of the most powerful analytical tools in
geometry and physics. From the physical point of view, it might appear, in
some sense, to be a trivial assertion, since it establishes that the variation of
the volume of the inner domain determined by a compact connected surface,
when it is deformed by the flow of a vector field, coincides exactly with the
total flow of the field crossing the boundary surface.
Definition 5.30 (Vector fields and the divergence of a vector field). If A is
a subset of R3, a differentiable vector field on A is a differentiable vector-
valued function X : A -p R3. We define the divergence of the field X as the
function div X : A -p R given by div X (p) = trace (dX) for each p e A.
Therefore, if {ei, e2, e3} is any orthonormal basis of the space ]E83, we
have div X(p) _ ((dX)(e), ei). If we make the particular choice
206 7. Intrinsic Geometry of Surfaces
Finally, in this chapter we will deal with the other intrinsic invariants par
excellence: geodesics. It seems to be logical, and indeed it is, that the curves
of least length joining two given points of a surface, if they exist, should be
preserved under local isometries. It was probably JOHANN BERNOULLI in
1697 who was the first mathematician to study their existence and to see
that they are paths of vanishing tangential acceleration in the particular
case of ovaloids. L. EULER was however the first geometer to write their
equations, in 1732, in the case of surfaces given in implicit form. Its current
name and definition, apart from their minimizing properties, are due to
J. LIOUVILLE (1844), while the beginning of the study of these minimizing
features is due to C. G. J. JACOBI (1804-1851) in 1843. We will end this
chapter by proving an easy version of the existence theorem for geodesics
that minimize the length between any two points of a closed surface in 3.
In a more general form, this is nothing more than the famous result proved
in 1931 by H. HOPF and W. RINOw.
7.2. Rigid motions and isometries
Let S be a surface and let b : IIS3 -- IIS3 be a rigid motion of Euclidean space
see Definition 1.7 given by q5(x) = Ax + b, for each x E 1R3, where A is a
3 x 3 orthogonal matrix and b E i3 is a fixed vector. From Example 2.12,
we know that the image S' _ q5(S) of S under b is another surface and that
f = is : S - 5'
is a diffeomorphism; see Exercise 2.45. Those diffeomorphisms between sur-
faces obtained by restricting a rigid motion of Euclidean space, which we
will also call congruences, keep not only the topological and differentiable
structures invariant, but also their geometry. We now clarify this assertion.
First, we have
(df)(v) = (dc51s)p(v) = (dq)(v) = Av,
for each p e S and each vector v E TS. Since A is an orthogonal matrix, it
preserves the scalar product in II83 and thus
((df)(u),(df)(v)) _ (u, v), `dp e S Vu, v E TS.
That is, the differential (df)p at each point of S is a linear isometry between
the corresponding tangent planes. Continuing, if N : S -* 2 is a Gauss
map defined on S, the new map N': S' S2, defined by
N'o f =AoN,
is a Gauss map for S'. Thus, the relation between the differentials is given
by
(dN')f() o A = A o (dN)
7.1. Introduction 205 158 5. Integration on Surfaces

any other, the measure of the curvature at each point re- el = (1,0,0), e2 = (0,1,0), and e3 = (0,0,1), then
mains unchanged.
(dX)(e,) _ (ax aa 1 ()as 2 (p), as 3
This is the Theorema Egregium of Gauss that asserts that the Gauss cur-
vature is invariant, not only under the group of rigid motions of J3, but
under a much wider class of transformations, i.e., the local isometries be-
(dX )(e2) _
ay
1, aX2
(() ay (p),
ax 3
ay /
tween surfaces. We will prove this famous result in Section 7.3 by utilizing (dX)(e3) - 1 as 1(p), a 2 (p), as 3 (p))
an expression for the Gauss curvature in terms of an arbitrary parametriza-
tion that does not make use of the normal field of the surface. This will provided that X = (X1, X2, X3), that is,
be an expression that can be calculated inside of the surface, that is, it will
be an intrinsic expression. One might wonder if such an expression see X=
axl axe ax3
Lemma 7.11 leading to an invariance greater than Euclidean invariance
div
ax + ay - az
should be taken as a more appropriate definition than the original one as and, thus, div x is a differentiable function.
the product of the two principal curvatures. On one hand, the original def-
inition is easy to make as well as clearer and esthetically more agreeable Theorem 5.31 (Divergence theorem). Let S be a compact connected surface
than the alternative offered by Lemma 7.11. On the other hand, it is quite and St the inner domain determined by S. If x : S2 -> I[83 is a differentiable
an awkward exercise to verify that the expression for the Gauss curvature vector field, then
contained in the proof of the Theorema Egregium does not depend on the
choice of parametrization. The problem is not that the original definition fdivx = _f(x,N),
was bad, but that, as Spivak points out in [17], the definition of the notion §2
where N : S is the inner unit normal field.
of the Gauss curvature was too good.
It turns out in this way that the Theorema Egregium leads to an intrinsic Proof. Denote by P the plane in R3 with equation x = 0 and consider the
geometry for surfaces in Euclidean space. From this perspective we are orthogonal projection 7r : S -p P given by
interested in all properties such as Gauss curvature that are preserved under
local isometries, or, in other words, under differentiable maps leaving lengths 71(p) =p- (p,a)a, p e S,
of curves unchanged. This situation does not have an analogue in the case of
curves and that is why we did not discuss their intrinsic geometry. Indeed, where a is the vector (1,0,0) which is normal to P. This projection is clearly
all regular curves are locally isometric to each other: the arc length function differentiable and its differential at each point p of S is the projection itself
restricted to TS; see Example 2.58. Hence
allows us to establish the corresponding isometries. For this reason, there
are no obstructions to inelastically deforming plane or space curves.
As a consequence of the Theorema Egregium and of global results in pre-
I= I
vious chapters, we will be able to show another property of spheres relative where N is a Gauss map of S. Consider the following function f : S --> I[8
to the deformation problem that we have considered: its rigidity. It is impos- defined by
sible to deform a sphere preserving lengths of curves. This was suggested by 1(p) _ (sign (N(p), a)) (X, a) _ (signN')x',
F. MINDING (1806-1885) as early as 1838 and was proved by H. LIEBMANN
where the superscript 1 means the first component of these maps and the
as a corollary to his well-known result obtained in 1899. This rigidity of function sign N' takes the values 1, 0, or -1 depending on whether Nl is
spheres can be viewed in quite a wider context, since it is really a property positive, zero, or negative. This function f is integrable on S since x is
shared by all positively curved compact surfaces, that is, all ovaloids are continuous and S is compact, and therefore it is bounded. We apply the
rigid. This statement is the so-called rigidity theorem by S. CoHN-VossEN, area formula for surfaces (Theorem 5.28) to obtain
proved in 1927, which we will deduce from an integral equality due to HER-
GLOTZ and which had probably been announced, according to CAUCHY, by X 1 Nl = f f) (0, y, z) dy dz.
LAGRANGE in 1812. 2
5.7. The divergence theorem
Taking Exercise 5.26 and the definition of n(7r, f) into account, we easily
deduce that, for each regular value (0, y, z) of 7r, we have
n(7r, f)(O,y, z) = - [X'(xi,y, z) - X 1(xi, y, z)]
i=2jE{1,...,k}
where (xi, y, z), i = 1, ... ,k, are the points of intersection of S with the
vertical line passing through (0, y, z), arranged according to their heights
relative to the plane x = 0. Now, if we take the functions X' and aX 1 /ax
extended by zero outside SZ, the fundamental theorem of calculus, together
with the previous equality, give us
+°° ax 1
f ax
(x, y, z) dx = -n(7r, f)(0, y, z).
Now integrating on R2 and using Fubini's theorem for the Lebesgue integral
and the equality (5.4), we get
-Jf X1N1 =f Ia dxdydz. 1
The same formula can be obtained for the second and third components.
Adding the corresponding three equalities, we have the required integral
formula.
Remark 5.32 (Volume enclosed by a compact surface). If S is a compact
surface and St is its inner domain, consider the identity vector field X : St -->
I[83 given by X (x) = x for all x e SZ. Its divergence is the constant function
3. Hence, by the divergence theorem
vol SZ = - 3 f(P,N(P))dP,
where N is the inner unit normal of S. For example, if we take as the
surface S a sphere S2(r) of radius r > 0 and centre a point po E 1[83, we have
= Br (po) and
vol Br(p0) _ - 3 f2 T (p, r (P0 - p)) dp
Now using Exercise 5.12 and Example 5.16, we obtain
vo1Br(po) = 37rr3.
Remark 5.33 (Divergence theorem on regular domains). We will use the
term regular domain of I[83 for any bounded connected open subset S2 such
that Bdry St is a compact-not necessarily connected-surface that we will
denote by aS2. For example, if Sl and S2 are two compact connected disjoint
surfaces such that their corresponding inner domains satisfy S22 C Stl, then
SZ = S21 - St2 is a regular domain such that 8St = Si U 52. A particularly
159 204 7. Intrinsic Geometry of Surfaces
are two necessary and sufficient conditions: to preserve the arc length of
curves and the second fundamental forms. This is a version of the so-called
fundamental theorem of the theory of surfaces, first proved by O. BONNET
(1819-1892) in 1867. In other words, it is clear that, if a surface were made
from a flexible but inelastic material, any deformation that we produced of
it would preserve the length of curves. Let us assume that, in fact, surfaces
are made from such a material, and let us try to perform a deformation
of the surface that preserves the curvatures of all its normal sections. The
only way to accomplish this is to move the surface rigidly in IIS3. That is,
the deformation would be impossible. But, what happens if we remove the
condition imposed on the second fundamental form? It seems intuitively
clear that, if we restrict ourselves to a small enough piece of a surface, we
could deform it, preserving the length of curves. But, it would be difficult
to imagine that, for instance, an entire compact surface might be deformed
in this way without translating or rotating in IIS3.
If a surface can be mapped to another one preserving lengths of curves,
then we say, using current terminology, that the two surfaces are locally
isometric. In GAuss's time, it was said that the first surface could be de-
veloped onto the second one. Topographers, and GAUSS was one of them,
while trying to draw maps of the Earth's surface, thought of surfaces, un-
like our usual view, not as a two-dimensional film in three-space but as the
boundary of a three-dimensional rigid body. This idea leads in a natural way
to considering deformations of surfaces preserving lengths of curves. They
knew, for example, that a piece of a sphere the Earth's surface cannot
be developed onto a plane. On the other hand, many people have had the
experience of making a cylinder by bending a piece of paper -a flexible but
inelastic material and also of having made conical hats from paper. A
good question relative to these situations is: is there some quantity or geo-
metrical object remaining invariant under local isometries? If so, we would
find some obstructions to the existence of this class of maps between two
given surfaces. On page twenty, Art. 12 of the Disquisitiones of 1827, GAUSS
states:
Formula [(4.6)] itaque art. praec. sponte perducit ad egre-
gium theorema. Si superficies curva in quamcunque aliam
superficiem explicator, mensura curvaturae in singulis
punctis invariata manet.
This statement, in spite of the rigidity of the Latin-that GAUSS often com-
plained about-to express some mathematical nuances, can be translated as
follows:
Then the formula in the previous article leads by itself to
an excellent theorem. If a curved surface is developed onto
 160 5. Integration on Surfaces

Chapter 7

Intrinsic Geometry
of Surfaces
Figure 5.2. Regular domain

interesting example of this situation is the following: if S is a compact

connected surface and St is a parallel surface with t > 0 sufficiently small,
then the open set Vt (S) is a regular domain with 8Vt (S) = S U St and is such
7.1. Introduction that St - STt = V(S), where ll and Pt are the inner domains determined by
We have been studying differential geometry of curves and surfaces in Eu- S and St, respectively.
clidean space R3 and, until now, the only distinction that we have made For this kind of domain Il it also makes sense to talk about its inner unit
about the results obtained is on their local or global nature. It is true that normal field, tubular neighbourhoods of 811, and the fact that the divergence
we titled Chapter 6 with the adjective extrinsic-referring to the geometry theorem remains valid, i.e.,
of surfaces and that we start this chapter using the term intrinsic. This
label of intrinsic, applied to the geometry of surfaces, might not appear to fdivX f(x,N),
a
be appropriate, since the very essence of surfaces, as we have defined them, where the integral on 811 means the sum of the corresponding integrals on
is to be embedded in R3. That is, the geometry of surfaces consists of study- each of its connected components.
ing how they bend and extend in Euclidean space and, in this sense, it is
extrinsic. This assertion is clearly confirmed if we consider that both the Remark 5.34 (Gauss's theorem for electrostatics). If S is a compact surface
separation properties and the existence of tubular neighbourhoods, estab- and St is its inner domain, the vector field
lished in Chapter 4, which could be extrinsic properties par excellence, have p-a
been the key in the proof of several of the theorems that we have considered. lp -
Then, why use the expression intrinsic geometry, and what does it refer to?
is differentiable on n if a S1. One easily sees that
The answer, as with so many matters in this field, goes back to GAUSS and
requires further explanation. a,
(dY)p(v) lp I all v - 3 - l5) (p - a)
Since we are studying the differential geometry of curves and surfaces in
Euclidean space R3, it seems to be logical, and indeed we have done so until for each v E R3. Thus, the divergence of Y vanishes everywhere. Conse-
now, to check that the notions that we introduce and the results that we state quently
are invariant under rigid motions of R3 since these are the isomorphisms f (P - a, N(p)) dp = 0, Va S2.
of Euclidean geometry. We will begin in Section 7.2 by discovering that, is Ip-a13
for a map between surfaces to be extended to a rigid motion of R3, there Suppose now that a E Q. Since Il is open, there exists e > 0 such that
the open ball BE (a) centred at a with radius - is included in D. The set
S2' = Ii - B, (a) is a bounded connected open subset of the plane whose
203 topological boundary is S U S,2 (a). Hence, S2' obeys the conditions required
5.8. The Brouwer fixed point theorem 161 202 6. Global Extrinsic Geometry

in Remark 5.33 above to be a regular domain. Since the vector field Y is Exercise (12): Substituting K = 2 - H into inequality (3.6), we have
differentiable on SZ' and has null divergence, we have that H2 + H - 2 > 0 and, thus, that H > 1 (with respect to the inner
0_ (p-a,N(p))
dp + f (q-a,N(q)) normal), with equality only at the umbilical points. Going back to the
sIp -ads E(a) Iq-aI3 dq original equality, we see that K < 1, and then that K < 1 < H. Recalling
that the support function is strictly negative, the two Minkowski formulas
But in the second integral we have give
N(q) - :1: =(q-a). A(S) _ - f H(p)(p, N(p)) dp ? - f K(p)(p, N(p)) dp = J H > A(S).
Then we can explicitly compute this second integral and finally obtain Therefore, 5 is totally umbilical.
/' (a_PN(P))dP_4l.r, `da E St.
is Ia-pI3 Exercise (14): Since the support function is negative, keeping in mind the
Note that this equality is usually known as Gauss's theorem for electrostat- hypotheses, we have
ics. <H and -K(p)(p, N(p)) (p, N(p)),

5.8. The Brouwer fixed point theorem for each p E S. Integrating and using the Minkowski formulas, we have

We will finish this chapter by using the divergence theorem to prove one rA(s) f H = -f K(p)p,N(p)) dp -r f H(pXp,N(p))dp = rA(s)
of the most famous results of algebraic topology: the Brouwer fixed point
theorem, which we will consider only in the differentiable case. Our proof Thus S has constant mean curvature 1/r and constant Gauss curvature 1/r2.
also works in the two-dimensional case, after suitable modifications that we We can conclude the proof by referring either to Corollary 3.49 of Hilbert
will point out in Chapter 9. and Liebmann or Alexandrov's Theorem 6.17.

Exercise 5.35 (Archimedes' principle). Let S be a compact surface and SZ Exercise (19): The definition of the Laplacian and Exercise (16) above
its inner domain. Apply the divergence theorem to the vector fields X and imply directly that the integral on S of L f vanishes. Therefore, since L f
Y defined on I[83 by X (x) _ (x, a)b and Y(x) _ (x, a) (x n b), for all x e I[83, is continuous, we have L f = 0. From Exercise (18), we have
and where a, b e I[83. We obtain in this way the two equalities f2 = 2fLf +21 Of 2 = 2IOf 12.
(p, a)N(p) dp = -(vo1St)a, a E I[83, Integrating this equality and taking into account the fact that the integral
Is of j f 2 is zero, we arrive at
(p, a) [p n N(p)] dp = -(vol St) (G n a), a e 1[83,
Is
Is
where
11 x dx Thus V f = 0 and f is constant on each connected component of S.
G vol SZ Js
is the centre of gravity of SZ. The first equality can be interpreted as follows:
if a force acts at each point of the boundary S of a solid body St, perpendic-
ularly to S and with a strength equal to the height from this point to a given
plane placed over the body, then the body experiences a total upthrust whose
strength is equal to its volume. The second equality refers to the moments
of the forces involved in the previous statement and can be paraphrased as
follows: the resultant effect of all these forces acting on the boundary of the
solid body is the same as we would obtain by applying the total upthrust at
the centre of gravity of the body.
Hints for solving the exercises 201 162 5. Integration on Surfaces

Exercise (9): The same calculation as in Exercise (4) gives us Exercise 5.36. We can associate to each differentiable function f : A -p IfS,
Ft
defined on a subset A of Euclidean space, a vector field V f , also defined on
Jac (p) = 1 3 I1 - tk1(p)II1 - tk2(p)II (p> N(p)) + tJ, A, by
Ftl
which does not vanish, since kl, k2 < 0, and (p, N(p)) > 0, p E S, since N ((Vf)(p),v) _ (df)(v), Vv E 1[83.

is the outer normal. The monodromy theorem implies that cz5 = Ft/IFtI is a This field is called the gradient of f. Prove that the three components of
difFeomorphism for each t > 0. As a consequence, z5 and, thus, each Ft are V f are
of = (DI of of
injective. From Exercise (9) of Chapter 2, St = Ft(S) is a compact surface
and Ft : S -j St is a diffeomorphism. Finally, Exercise (16) of Chapter 3 8x' 8y' az) .
asserts that its Gauss curvature is always positive.
Show that, if the gradient of a differentiable function vanishes everywhere,
then f is constant on each connected component of A.
Exercise (10): We know, after solving Exercise (9), that it makes sense to
consider outer parallel surfaces St at arbitrary distance t > 0. Since each St Exercise 5.37. If A C R3 and f, g : A -p R are two differentiable functions,
is compact and connected, we can apply the isoperimetric inequality (6.6) then
to each of them and obtain aiv(v f n vg) =0.
A(St)3 > 36ir(volStt)2, b't > 0.
Exercise 5.38. If A C 1[83 and f, g, h : A -p I[8 are differentiable, then
Taking into account Example 5.17 and the analogue of Example 5.19 for
outer parallel surfaces, we obtain the polynomial inequality det(Vf, fig, Oh) = div [h(Vf n fig)].

CA(S) + 2t H+ 13 > 36 (vol St + tA(S) + t2 H+ 3 1312 ,

The following result, which prepares us for the proof of the fixed point
Js Js theorem, asserts that, in some sense, a differentiable transformation of the
inner domain determined by a compact surface that fixes the surface behaves
which is valid for each t> 0. Dividing by t6 and denoting by s the inverse
almost like a orientation-preserving diffeomorphism.
1/t, we have that the function
Proposition 5.39. Let S2 be the inner domain determined by a compact
f(s) = (4 2s H + 52A(S) 13-36 14 + sJ H + s2A(S) + s3vo1 S2 12 connected surface S, and let q5 S2 -p I[83 be a differentiable map such that
Js / s \3 / s = Is. Then
:

is non-negative for s > 0. We easily see that f(0) = 0 and f'(0) = 0.

Therefore, we have (Jac)(x) dx = vol (a).
/p \2
0< f"(0) = 24ir I H) - 96ir2A(S). Proof. We have _ (1, 2, 3), where i, i = 1, 2, 3, is the component
\s /
J ui) relative to an orthonormal basis {Ui, u2i u3} of I[83. Using Exer-
Exercise (11): By integrating the inequality K > 1, we obtain cises 5.37 and 5.38 above, we have

A(S) 4'ir = A(S2), (Jac)(x) = det(Vq51,

0q52, 0q53) = div [3(V' ^ 0q52)].
since S is an ovaloid. On the other hand, if an open unit ball B is in the Hence, the divergence Theorem 5.31 gives us
inner domain St of S,
vol 1 > vol B. (Jac)(x) dx = - f (V2)(p), N(P))dP.
s
Consequently, But our hypothesis about asserts that restricted to the surface S is the
A(S)3 < A(S2)3 = 36ir(vo1B)2 < 36ir(vo1SZ)2, identity map. Thus 3(p) _ (p, u3) if p e S and, moreover,
which, along with the isoperimetric inequality, completes the exercise. ((V)(p),v) = (d)(v) = (v,u)
Exercises 163 200 6. Global Extrinsic Geometry

for i = 1, 2, 3 and v e TBS. By substituting this into the previous equality, Hence el, e2 E TS = are eigenvectors for (dN)() and we are done.
we obtain

f Exercise (4): The map q5: S §2, given by

(Jac )(x) dx = - (p, u3) det(ul, u2i N(p)) dp.
J - p - cN(p dp E S,
We complete the proof by using Exercise 5.35 above, where we studied the ,

Archimedes principle. is well-defined and differentiable, since p - cN(p) cannot vanish anywhere
because the support function does not take the value c. From the definition
Theorem 5.40. Let S be a compact connected surface and let S2 be the
we obtain
corresponding inner domain. There are no differentiable maps q5: St -> I[83
such that = Is and q5(SZ) C S. v - c(dN)p(v) _ (p - cN(p), v - c(dN)(v))
(dq) (v) - (p - cN(p)),
Ip - cN(p)I Ip - cN(p)I3
Proof. If such a map q5 existed, then, for each x e SZ and each v e R3, we for each p E S and v E TPS. Hence, if el, e2 E TS are principal directions
would have atpES,
(x + tv) E Tc7()S. IIBy
t=o
Therefore, the image of (d) would be two-dimensional at each x E SZ our hypotheses, this Jacobian never vanishes. Thus z5 is a local diffeo-
and, thus, (Jac )(x) = det(dq5) = 0. This would be a contradiction to morphism by the inverse function theorem. We conclude the exercise by
Proposition 5.39 above. using the monodromy theorem cited above.
Theorem 5.41 (Fixed point theorem). Let q5 B,. -> B,. C I[83 be a dif-
:

ferentiable map, where BT is the closed disc of radius r > 0 centred at the Exercise (5): If S is contained in an open ball of radius r > 0, we can and
origin. Then has at least one fixed point, that is, there exists x E Br such do assume the origin is the centre of this ball. Thus, if p E S,
that q(x) = x. II <r.

Proof. Suppose that this is not the case. Then, for each x E B,., we denote
by 'b(x) the only intersection point of the half-line starting at (x) and
passing through x with S2(r). The resulting map b takes Br into S2(r)
and satisfies = I. The proof will be finished if we establish that Eb is
differentiable. But this can be easily seen by expressing it in terms of
using the differentiability of the latter.
Exercises
(1) T Let S be a surface of revolution and let a : I C 1[8 -> S be a curve
p.b.a.l. generating the surface. Prove that the area of S is equal to
A(S) = 271
J1
dist (a(s), R) ds,
Then, the support function based on the origin never takes the value r. Now
apply Exercise (4) above.
Exercise (6): By Remark 6.3, if R n 1Z 74 0, then R n S consists of exactly
and two points. Suppose, then, that R n 1Z = 0 and that R n S 74 0. Hence
R n S = R n 1Z must, by Theorem 6.1, be a non-empty closed convex subset
of R, that is, a non-empty closed interval. If it had more than one point,
by Exercise 3.33, S would have non-positive Gauss curvature at the interior
points of this interval. Since S is an ovaloid, R n S has only a point. Finally,
apply Exercise (7) of Chapter 4 to see that R is tangent to S at this point.
Exercise (8): If the centre of the ball where S is included is the origin, we
use the first Minkowski formula and obtain
A(S) _- N(p))H(p) dp f IpII H(p)I dp < A(S).

where R is the revolution axis. This result is known as the Pappus Then we complete the proof, for example, by using Alexandrov's Theo-
theorem. rem 6.17 or Exercise (4) of Chapter 2.
Hints for solving the exercises 199 164 5. Integration on Surfaces

Exercise (2): Let S be an ovaloid that attains equality and let St be its (2) T Let S be a compact surface with non-vanishing Gauss curvature ev-
parallel surface at distance t, with t small. The function erywhere and an injective Gauss map. Show that
\2
tH ( Ht(p) dp ) - cA(St) K=4i-.
\ St / Is
has a minimum at t = 0. By using Exercise (16) of Chapter 3, Theorem 5.14 (3) T Prove that, if a, b E Il83,
on change of variables, and Example 5.17, we can explicitly compute this 4
function, namely I (p,a)(p,b)dp=
JS2
'Tf(a,b),

t H (J H- t J K)2- c (A(S) - 2t J H+ t2 K ). and, consequently,

J 4
This function is differentiable and its derivative at t = 0 vanishes. Hence (Ap, p) dp = 3 trace A,

IJ HIIc-J KI =0. for each square matrix A of order three.

(4) T Let A be a regular 3 x 3 matrix. Show that the map : S2 -k §2
Since S is an ovaloid, H2 > K> 0. Therefore defined by q5(p) = is a diffeomorphism. Use the change of
variables formula to check that
c=J[K=4,
s
where the last equality follows from the Hadamard-Stoker Theorem 6.5. 2 IApI3 dp det AI .
(5) T Let :
§2 ->
§2 be the map defined by q5(p) _ (p + a)/Ip + aI, for
Exercise (3): It is clear that the required involution q5: S -j S exists since all p E §2, where a E ll83 is a vector with al < 1. Show that q is a
the Gauss map N : S §2 is a diffeomorphism and, thus, it is enough to diffeomorphism and that
put z5 = N-1 o A o N, where A is the antipodal map of the sphere. Applying
the chain rule to the equality N o z5 _ -N, we have f 1+(p,a)
Ip+a13
(*) (dN)() o (dq5) _ -(dN)P, b'p E S.
(6) Let be a dilatation of Euclidean space I[83 with centre po E Il83 and
On the other hand, if q5(p) -p = c for each p E S and some c E IEB, we have ratio 0. Prove that the areas of a compact surface S and of the
that c> 0 and image surface S' _ q5(S) are related by A(S') = µ2A(S).
(**) (q(p) - p, (dq)P(v) - v) = 0, b'p ES Vv E TPS. (7) T Prove that the integral of the Gauss curvature on a compact surface
We now show that (dq)P - ITS is trivial. In fact, if (dq)(v) = v for coincides with the integral of the Gauss curvature on any of its parallel
some v E TES, from (*) we have -(dN)P(v). After scalar surfaces.
multiplication by v, we see that (8) T Show that, if S is a compact surface,
a()(v,v) = -a(v,v). (N, a) = 0,
Since S is an ovaloid, v is definite and, thus, v = 0. Then (dq)P - ITS Is
is injective and, as a consequence, its image is the whole of the plane TBS. where N is a Gauss map and a is an arbitrary vector.
Equality (**) can then be rewritten in another way: (9) T Assume that S is a compact surface. Use the divergence theorem and
gy(p) - p 1 TS, b'p E S. the formula for the change of variables to show that
f 3 f
Therefore q5(p) - p = cN(p), where N is the inner normal (for the choice of vo1SZt = vo1SZ + 3
J
(2H(p,N) -1) + 3 (2H - K(p, N)) - 3 K,
sign, consult Remark 6.2). Therefore, if el, e2 E TS are principal directions, J J
we obtain where SZt is the inner domain determined by the parallel surface at dis-
(d)(e) - ez = -ckZ(p)ez, 2 = 1, 2. tance t. (See Example 5.19.)
 Exercises 165 198 6. Global Extrinsic Geometry

(10) Let A C 1R3, let X, Y be two vector fields defined on A, let f : A - III for each p E S and show that
be a differentiable map, and a, b E Ilk. Prove that

div (aX + bY) = a div X + b div Y and div (fX) = f div X + (Vf, X).
f((vf)p,) dp + 2 ff (p) dp = -2 J f(p)H(pXp,N(p)) dp
(11) Let A C J83, and let f, g : A -> Il8 be two differentiable functions. We
define the Laplacian of f as the function Hints for solving
the exercises
- 52/ 52/ 52/

Exercise 6.14: If H = K, by subtracting the two Minkowski formulas, we

Prove that L f = div V f and O(f g) = g0 f + fLg +2 (Vf, Vg). have
(12) Let O be an open subset of J83, and let f : O -k Il8 be a differentiable 1
map. Show that the following assertions are equivalent: Is f H.
The function f is harmonic, that is, z f = 0. On the other hand, inequality (3.6) gives us H = K <H2 and, thus, H > 1,
For each x E O and r > 0 small enough, since H = K> 0. We complete the exercise by using Proposition 5.6, which
concerns properties of integration.
f(x) = 1 2 JS(r) x1(P) dp

Exercise 6.15: Subtract the two Minkowski formulas again. If HK = 1,

For each x E O and r> 0 small enough, we see that

((Vf)(p),p) dp = 0,
where S(r) is the sphere of radius r> 0 and centre x E O.
(13) T Let S be a surface and let P be a plane such that the orthogonal
projection onto P restricted to S is a diffeomorphism on an open set
U C P. Prove that A(U) < A(S). That is, orthogonal projections on
planes decrease area.
(14) T Let S be a compact connected surface that is star-shaped with respect
to a point, say, for example, the origin. Show that the central projection
¢ : S -> §2 given by q5(p) = p a diffeomorphism.
Thus S = S(f) for some positive function f defined on §2. (See Exercises
(10) and (11) at the end of Chapter 2.)
(15) T Spheres minimize the integral of the square of the mean curvature.
Prove that, if S is a compact surface, then
H(p)2 dp > 4ir,
Is
and equality occurs if and only if S is a sphere.
Is
Cl - H(p) + (N(p), p) (H(p) - H(p))) dp =0.
Again inequality (3.6) and the hypothesis HK = 1 imply that H > 1. To
complete the exercise, use the fact that the support function based on the
origin-assumed to be in 1-is strictly negative.
Exercise 6.18: Suppose that A C IEBn is open. By definition,
A+B= U(A+b).
bEB
But each A + b, with b E B, is open because translations of Rn are homeo-
morphisms. Then A + B is open.
Exercise 6.20: The required equality follows directly from the definition
of the sum of sets. Suppose that A n B = 0. Then, at least one of the
intersections I1 n J1, I2 n J2, 13 n J3 has to be empty. Suppose that it is the
first one. Thus, there exists a E Ilk such that the intervals I1 and J1 are on
different sides of a. Hence, the plane of the equation x = a separates the
two sets A and B.
 Exercises 197 166 5. Integration on Surfaces

(12) T An ovaloid whose (inner) mean and Gauss curvatures satisfy H+K = 2
must be a unit sphere. Hints for solving
(13) An ovaloid whose mean and Gauss curvatures satisfy HZ+KZ = 2 must the exercises
also be a unit sphere.
(14) T Let S be an ovaloid with H > 1/r and K G H/r, where r > 0. Show
that S is a sphere of radius r. Exercise 5.25: As we argued after Definition 5.3, one can find a diffeomor-
phism zb : O -> R x (a, b), where O is an open subset in J83. Let JVt be the
(15) Let S be a compact surface. Show that S has constant mean curvature set of the points of O where IJac (q5 o b) I - (IJac cbl o b) Jack vanishes. It
if and only if, for each differentiable function f defined on S, such that is enough to see that this equality implies that N = b(M) and to apply a
vol SZt(f) = vol St, ti <6, one has that three-dimensional version of Sard's theorem to the map b o b.
d
A(St(f)) = 0.
dt t=o Exercise 5.26: Let S - P be the restriction to S of the orthogonal
(16) Let S be a compact surface and V S -p II83 a tangent vector field. projection ir of ll83 to the plane P. Then, since
Prove that ir(x) = x - (x, a)a + ca, bx E I[83,
(divV)(p) dp = 0, for some c E Ilk, we have

(dcb)(v)=v-(v,a)a, `dv E TpS `dp E S,

{ki(p)((dV)(ei), ei) + e2)} dp = 0,
Is and, thus, an easy computation leads to
where {ei, e2} is a basis of principal directions at TS for each p E S. IJacc5I = Iwhere

(17) Let f : S -p II8 be a differentiable function defined on a surface S. We

use the term gradient of f for the vector field of tangent vectors denoted
by V f : S -j ][83 and given by
((Vf)(p),v)=(df)(v), v E TPS,
J (N(p), a) 0, that is, a TS and, so, S and R meet at p transversely.
1 ((Vf)(p),N(p)) =
where N(p) is a unit normal to S at p. Prove that V f is a differentiable
vector field and that, if it is identically zero, f is constant on each
connected component of S.
(18) If f : S -p R is a differentiable function on a surface S, we define the
Laplacian of f by L f = div D f . Prove that
O(f9) = fLg+gLf+2(Vf,Vg),
for any two functions f and g defined on S.
(19) T Let S be a compact surface and let f S -p II8 be a differentiable
function. Prove that, if L f > 0, then f is constant on each connected
component of S.
(20) If f : S -p IE8 is a differentiable function on a compact surface, apply the
divergence theorem to the vector field V defined on S by V (p) = f(p)p
N is a Gauss map of S. Hence, a point x E P is a regular value
of if all the points p E S which project onto x satisfy (N(p), a) 0.
Taking Sard's theorem into account, we deduce that almost every line R
perpendicular to the plane P satisfies the property that, if p E R n S, then
Suppose now that R is a line in this class. From Remark 4.3, we know
that R n S consists of finitely many points, since S is compact. The open
intervals resulting after removing these points from R belong either to S2 or
to JR3 - ft Remark 4.3 implies that there are points of S2 and of JR3 - SZ on
each side of the points of R n S. Since S2 is bounded, this, in turn, implies
that these intervals lie alternatively in JR3 - SZ and in S2, if we move along R
in any of the possible directions, for example, corresponding to the vector
a. Thus R n S has an even number of points. Let p E R n S be a point that
is a lower extreme of one of these intervals. Then, for t near zero,
p+taES2 ift>0 and p+taER3-1Z ift<0.
If we work with t sufficiently small so that p + to is in a tubular neighbour-
hood of S, then the expressions above can be rewritten as
r(p + ta) >0 if t >0 and r(p + ta) <0 if t <0,
Hints for solving the exercises 167 196 6. Global Extrinsic Geometry

where r is the oriented distance to S relative to the inner normal (see Exer- (3) T Let S be an ovaloid and N : S §2 its Gauss map. Show that there is
cise (1) of Chapter 4). Therefore a unique diffeomorphism q5: S -p S such that N o z5 _ -N. If Iq5(p) - pI
d
is the constant function c E R on S, we say that the ovaloid has constant
0< r(p +ta) _ (dr)(a) _ (N(p),a). width c. Prove that, in this case, k2 0 z5 - k2/(1 - ckz), i = 1, 2.
dt t=o
(4) T Let S be a compact connected surface and let N be a Gauss map
Hence, since (N(p), a) 0, we have (N(p), a) > 0. Analogously one can see for S. Let s and kl, k2 be the corresponding support function based
that (N, a) < 0 at the upper extremes of such intervals. on the origin and the principal curvatures. If there is a real number
c E II8 - {0} such that neither s takes the value c nor kl, k2 the value
Exercise (1): Use a parametrization X : I x (0, 2ir) -> S of the form -1/c, prove-applying the monodromy theorem, which asserts that a
local difFeomorphism from a compact surface to a sphere has to be a
X (u, v) = x(u) a + y(u) cos v b -y(am) sin v c, Vu E I Vv E (0, 2ir), diffeomorphism, to the map p E S H (p - cN(p))/Ip - cN(p)I E 2-
like the one described in Exercise (3) at the end of Chapter 2, where a(u) _ that S is diffeomorphic to the sphere.

Ib'u E I
x(u) a + y(u) b for each u E I. Then

and we complete the exercise by using Proposition 5.13.

`dv E (0, 2ir),
(5) T Let S be a compact connected surface contained in an open ball of
radius r > 0. If its principal curvatures satisfy kl, k2 > -1/r, then S
is diffeomorphic to the sphere. In particular, prove that a compact con-
nected surface whose principal curvatures are non-negative everywhere
must be difFeomorphic to a sphere.
Exercise (2): The Gauss map N : S -k 2 is a local diffeomorphism since, (6) T Let S be an ovaloid and let R be a line in IE83. Prove that S fl R = 0,
according to (3.4), or S fl R = {p} and R is tangent to S at p, or S fl R has exactly two
points.
0 K(p) = det(dN)p, by E S,
(7) Show that the following integral formulas are always valid:
and, thus, we can use the inverse function Theorem 2.75. Thus, it is an
open map and, thus, N(S) is an open subset of S2, which is also closed (N,a)H = 0 and f(N,a)K = 0,
since S is compact. Consequently, N is a diffeomorphism and we may apply Is
the change of variables formula (Theorem 5.14) taking into account that where S is a compact surface, N its Gauss map, H and K its mean and
IJac NI _ IKI = K (see, for this last equation, Exercise 3.42) and that Gauss curvatures, and a E ][83 an arbitrary vector.
A(S2) = 4ir (see Exercise 5.12).
(8) T Let S be a compact surface contained in a closed ball of radius r > 0
and such that its mean curvature satisfies IHI < 1/r. Prove that S is a
Exercise (3): Apply the invariance of the integral on 2 under rigid motions sphere of radius r.
preserving the origin. If (a, b) = 0, then there is a symmetry E 0(3) such
(9) T Let S be an ovaloid and let Ft : S -p ][83 be the map given by Ft(p) _
that (a) = a and q5(b) _ -b. Then
p + tN(p), where N is the outer Gauss map of S. Show that, for each
t > 0, the image St = Ft(5) is an ovaloid and, moreover, Ft : S -j St is
b) dp = f2 ((p) a)((p), b) dp a diffeomorphism. (Use the monodromy theorem-see Exercise (4)-for
the map cb =
= f 2 (p, (a)) (p, (b)) dp =- (p, a) (p, b) dp (10) T Use the isoperimetric inequality to prove that on each ovaloid S we
have
On the other hand, if a, b E ll83 are unit vectors, there exists ¢ E 0(3) such If 2

H(p) dp
that q5(b) = a. Hence > 4irA(S).

2 p a)2 dp = f 2
f(p,b)2dp. (11) T Let S be a compact connected surface with K > 1. If one can find an
open unit ball inside the inner domain of S, then S is a unit sphere.
Exercises 195 168 5. Integration on Surfaces

Theorem 6.25 (Space isoperimetric inequality). Let S be a compact con- Furthermore, given a unit vector a E R3, we can choose an orthonormal
nected surface. Then A(S)3 > 36 ir (vol )2, where 1 2 is its inner domain. basis {u1 = a, u2, u3} of JR3 and we get
Moreover, equality occurs if and only if S is a sphere. 3

3 (p, a)2 dp = (p, ui)2 dp = L dp =

Proof. The inequality was already proved in (6.6). Furthermore, we know i=1 2
that equality is satisfied for spheres of arbitrary radius. If S is a surface that
This suffices for solving the exercise since, if {e1, e2, e3} is the standard basis
attains equality and f : S -j ][8 is any differentiable function, we consider
of JR3, we have
the variation St(f) of S relative to f, which is defined for tI <6 with 6 > 0. 3
Then, the function h : (-6, S) -j R, given by (Ap, p) _ ez) (p, ei)
h(t) = A(St (f ))3 - 36 ir (vol Sgt (f ))2, i,a=1

is differentiable from the above examples and has a minimum at t = 0 by Exercise (4): The map b is differentiable since it is the restriction to §2 of
our hypothesis. Consequently, a differentiable map defined on ll83 - {0}. It is a diffeomorphism since one
can check immediately that b : §2 -> §2, given by
0 = h'(0) = 3A(S)2 dt A(St(f)) - 72irvo11 2 vol SZt (f )' i
dt t=o
t=o (p)= I ' b'p E S2,
That is, if S attains equality in the isoperimetric inequality, one has, using
the first two variation formulas computed earlier, that is its inverse map. To complete the exercise, compute the absolute value of
the Jacobian of g5. Since
f(p)[12irvo1 S2 - A(S)2H(p)] dp =0
J (dq5)(v) = Ilgpl
(APAv)Ap, by E TS `dp E §2,
for any differentiable function f defined on S. For instance, choosing f as
we have, if p E §2 and el, e2 E TpS2 is an orthonormal basis,
f(p) = 12 ir vol SZ - A(S)2H(p), dp E S,
IAdet
we deduce that f itself must be identically zero. In conclusion, the surface IJac I I pAI
S has constant mean curvature and we complete the proof by applying since {ei, e2, p} is an orthonormal basis of I[83.
Alexandrov's Theorem 6.17.
Exercise (5): The map ¢ is well-defined on §2 and is differentiable since
it is the restriction of a differentiable map defined on Il83 - {-a}. It is a
Exercises difFeomorphism since zb : §2 -> §2, given by

(1) Let S be an ovaloid in ][83. Show that (p) = ((a,p) + (a,p)2 + 1 - 1aI2) p - a,
is an explicit inverse map for it, which is clearly differentiable (recall that
fH2>41r I al <1). As in Exercise (4) above, it only remains to compute the absolute
value of its Jacobian.
and that equality occurs if and only if S is a sphere.
(2) T Suppose that there exists some c> 0 such that Exercise (7): Let S be a compact surface, St the parallel surface at dis-
\2 tance t, and Ft : S -* St the corresponding diffeomorphism studied in Re-
cA(S) < (j
\s
H(p) dp I
/
mark 4.29. It suffices to combine the information about Kt o Ft supplied by
Exercise (16) of Chapter 3 with the calculation of IJac FtI that we made in
for all ovaloids S and that there exists an ovaloid that attains equality. Example 5.17. In fact, we can show in this way that (Kt o Ft) IJac Ft I = K.
Using the parallel surfaces of S, prove that c = 4ir. Now, use the change of variables Theorem 5.14.
Hints for solving the exercises 169 194 6. Global Extrinsic Geometry

Exercise (8): Apply the divergence theorem (Theorem 5.31) on the domain Example 6.24 (First variation of volume). Let Stt(f) denote the inner
S2 determined by S to the constant vector field equal to the vector a. domain determined by the compact surface St(f ). We will prove that the
function
Exercise (9): According to Remark 5.32, we have t E (-S, 6) H V (t) = vol Stt (f )
1 is differentiable and that
vol Sgt = - 3 ISt (p, Nt (p)) dpi

where Nt is the inner normal of the surface St parallel to S at distance t. Now dt t=o
I
vol Stt f _- fs f (p) dp
use the change of variables formula (Theorem 5.14) for the diffeomorphism
For this, recall that
Ft : S -k St, taking into account that Nt o Ft = N and using the value of
IJac Ft I computed in Example 5.17. V (t) _ -1 f Nt p p dpi
3 st
Exercise (13): From the change of variables formula, we have that by Remark 5.32, where Nt is the inner normal of St(f ). Again using the
change of variables formula for the diffeomorphism fit, we obtain
A(U) = JU 1=
J V(t) =
where b S -+ U is the restriction of the orthogonal projection in this s
exercise. But in the hint for solving Exercise 5.26, we saw that Since the difFeomorphism qt between S and St (f) is the identity map when
IJaccbl _ I(N,a)I 1, t = 0, we have
where N is a Gauss map of S and a is a unit vector perpendicular to P. (dq5t)(ei) A (dqt)(e2) _ (dq5t)(ei) n (dq5t)(e2)
Nt o
IA (dt)p(e2)I Iand
Exercise (14): In Example 2.79, we computed the differential of and, thus, from (6.7), we have
from this, it can be deduced that
IpI3(N(p),p),
V (t) _ f
-3 ((dt)(ei) A (dt)(e2), t(p)) dp
where the last equality, if N is the inner normal, is true since we know that
the support function based on the origin does not change its sign, according
=-1 ftf() 3
dp
to Exercise (4) of Chapter 3. By the area formula for surfaces found in
Theorem 5.28, we have
1
-3
t2
f
- Js
IS2 n(,1), D is a differentiable function defined on S x (-6, S) and V f again
where n(q5,1)(x) is the number of inverse images through b of each regular denotes the gradient of f; see Exercise (17) at the end of this chapter. Hence
value x E 2 of b. Since S is star-shaped with respect to the origin, all V is differentiable at t = 0 and
the points of 2 are regular values of b, that is, any half-line starting from
the origin cuts S transversely. Moreover, Proposition 4.14 assures us that
V'(O) = 3 f[-f() + Zf (p)H(p)(N(p),p) + ((Vf),p)} dp
the function is locally constant and, thus, continuous. Since S is Now, assuming we have solved Exercise (20) at the end of this chapter, we
compact, 1. Therefore have that
4 < J n(,1) - _f
2
IpI3
(N(p),p)dp= 2 ff (pXp,N(p))H(p) dp = -f [2f(p) + ((Vf),p)] dp
where the latter is due to Gauss's theorem given in Remark 5.34. Hence Substituting this into the previous equality, we get the integral formula that
n(, 1) is the constant function 1. That is, is injective. On the other we called the first variation of volume formula.
6.5. The isoperimetric inequality 193 170 5. Integration on Surfaces

called the variation of S corresponding to the function f. Note that each
St (f) is diffeomorphic to S under the map q5t : S - St (f) , given by
c5t (p) = p + t f (p) N (p) = F (p, t.f (p)) , dp e S,
where F S x IIS -* IIis, as usual, the map defined in (4.1), which is
inj ective and which has as an inverse the projection it of the tubular neigh-
bourhood N6(S) restricted to St(f ); see Exercise (1) at the end of Chapter
4. The following two examples, which are studied by using the change of
variables formula, the divergence theorem, and the differentiable dependence
of parameters, will help us see what happens in the case of equality in the
isoperimetric inequality.
Example 6.23 (First variation of area). Let St (f) , t I <6, for some b > 0,
be the variation of the surface S corresponding to the function f : S -* III.
We will show that the function
t e (-8,8) H-* A(t) = A(St(f))
is differentiable and that
dt tot =
s a contradiction. Therefore, all the points of S are elliptic and, consequently,
In fact, if ei, e2 are the principal directions of S at the point p e S, we have
hand, b is surjective since its image is closed in §2, since S is compact, and
open in §2 as a consequence of the inverse function theorem.
Exercise (15): Equation (3.6), together with the fact H2 is non-negative,
gives us the inequality H2 > K. Integrating on S and using the Chern-
Lashoff Theorem 5.29, we obtain
fsH2 > fK+ > 4ir.
If equality holds, Proposition 5.6 implies that H2 = K+ and, thus, all the
elliptic points of S are umbilical. There is at least one such point, say po E S,
according to Exercise 3.42. Hence, the connected component C containing
po of the open subset of S formed by its elliptic points is included in a sphere
by Theorem 3.30. If there exists some point p E S with K(p) < 0, since
S is connected, we could take a continuous curve a : [0,1] -> S such that
a(0) = po and a(1) = p. Then, there would exist a first time x E (0, 1]
where K(a(x)) < 0. Thus a([0, x)) C C and, hence, a([0, x)) would be
included in a sphere and the Gauss curvature along it would be a positive
constant. By continuity, the same would occur at a(x), which would lead to
umbilical. We finish the exercise by using Corollary 3.31.

(dq5t)p(ei) _ (1 - tf (p)k1(p))e2 -I- t(df)p(e2)N(p), i = 1, 2.

Then, we see that
(6.7) (dq5t)(ei) n (dt)(e2) _ [1 - 2tf (p)H(p)]N(p) - t(Vf)p + t2G(p, t),
where p e S, tI <5, Of is the gradient field of the function f-see Exercise
(17) at the end of this chapter-and G is a certain differentiable function
defined on S x (-S, 8) taking values in I[83. Thus, since from the change of
variables formula

A(t) =14(St(f)) - ft(f)i=f IJactI,

equation (6.7) says, keeping in mind the differentiable dependence of param-

eters of Remark 5.22, that A(t) is differentiable at t = 0 and
d
A'(0) d IJacc5tI (dq5t)p(ei) n (dq5t)p(e2)I dp.
- dt t=o IS dt t=o

Therefore, using (6.7), we have

dt t=o
l IA (dt)p(e2)I = -2f(p)H(p)
From this we can easily obtain the required formula.
 192 6. Global Extrinsic Geometry

Chapter 6
from the points of S to q has to be attained at some point po E S such that
q is on the normal line of S at po and, since q E It, we have Io - q I > t.
Thus, the sum set SZt + B, which is connected by Exercise 6.21 above, is
included in one of the two connected components of ii - S. Since SZt C 1,
we finally have that

Global Extrinsic 1t + B C 1, dt E (0, s).

Now applying the Brunn-Minkowski inequality that we have just proved, it
Geometry can be seen that
vol SZ > vol (It + B) > [(vol 1)113 + (vol B)1"3]3,
for each t e (0, s). Developing the cube on the right-hand side and removing
the terms with exponent greater than 1/3 for the volume of the ball B, we
conclude that
vol St > vol It + 3(vol Bt)1/3(vol SZt)2/3,

bt E (0, e).
6.1. Introduction and historical notes Taking into account that vol Bt = (4ir/3)t3, as shown in Remark 5.32, we
get
We have already seen in Chapter 3 convincing reasons to believe that the vol 1 - vol 1 t 1/3
second fundamental form and the corresponding curvatures associated with >3 (vol1t) 2/3 , `dt E (o,).
t 3
the Gauss map are efficient tools for controling the shape of a surface. In
addition, in Chapter 5 we have developed a theory of integration on compact In Example 5.19 we computed the volume vol 1 t enclosed by the parallel
surfaces with the usual properties of Lebesgue integration. These are the surface S. Then, taking the limit in the previous inequality as t goes to
two main bases for carrying out a deeper study of the geometry of surfaces zero, we have
in JR3. 1/3
A(S) > 3 (vol
3
From the point of view of modern differential geometry, which has been
greatly influenced by topology, deeper means being able, even in a small This inequality involving the area of a compact connected surface and the
measure, to skip from local to global properties. For example, the Brouwer- volume of its inner domain is the isoperimetric inequality, which can also be
Samelson theorem, the Jordan-Brouwer separation theorem, and the Brouwer written as
fixed point theorem, proved in Chapters 4 and 5, are of course results of a A(S')3
global nature, but they are theorems that perhaps belong more to differen- (vo11)2 (volB3)2
tial topology. In this chapter, we deal with a class of results and proofs that That is, the function S H A(S)3/(vol1)2, defined for each compact con-
we think of as the core of differential geometry. nected surface S, attains its minimum at any sphere. But the proof that we
Whenever one thinks about global problems in geometry, it is usual have just given does not permit us to show that these are indeed the only
to impose some global condition to avoid the case when pieces of larger minima. However, we can do this in a small roundabout way.
surfaces may appear as possible solutions, that is, surfaces that are proper For each differentiable function f : S -* R we can find a number b > 0
open subsets of another larger surface. One of these conditions is clearly such that, if t I < S, then t f (S) C (-s, s) , where s > 0 is chosen so that
compactness. Another global condition that is weaker than compactness, N6 (S) is a tubular neighbourhood. In this way, it makes sense to consider
which appears in many texts, is geodesic completeness. Since our point of the set
view is mainly concerned with extrinsic geometry, we will restrict ourselves
St(f) _ {x e NE(S) I x= p+ t f (p)N(p), p E S, tI <8},
which, according to Exercise (3) of Chapter 4, is a compact surface such
171 that So (f) = S, for any f. The family of surfaces St (f) with ItI <8 will be
6.5. The isoperimetric inequality
A+, B+ and A-, B- since the total numbers of blocks satisfy n+ + m+
n + m and n- + m- <n + m, respectively. As a consequence,
vol (A+ + B+)> [(volA!3 + (volB+)V3]3,
(6.5) vol (A- + B-) > [(volA)'13 + (volB) 1/3]3
On the other hand, since A+ C P+ and B+ C Q+, then A+ + B+ C
P+ + Q+ _ (P -I- Q)+ and, analogously, A- + B- C (P + Q)-. Keeping
in mind that P + Q is another plane of ]I83, we conclude that A+ + B+ and
A- + B- are disjoint. Thus, taking (6.5) into account,
vol (A + B) > vol (A+ + B+) + vol (A- + B-)
-> [(volA!3 + (vol8+1/3113 +. [(volA-)1!3 + (volB-1/3113
But, from (6.3) and (6.4),
1!3 3
Vl B VOlB
vol (A +B) > vol A+ [i+ (:)1!3]3+vo1A_ 1-}-
lA vol A
3
1,'3
> volA 1-+ -
VOl B
_ [(vol A)1!3 + (vol B)1/3JJ13,
and, thus, the induction is complete and the result is also true for this second
particular case.
Now consider the general case. For this, denote by A and B any two
bounded open sets in 3. Using the elementary theory of Lebesgue inte-
gration, there are two sequences An and Bn, n e N, of open sets in IIS3 of
the type that we have considered in the second particular case, such that
AnCAand BnCBand
lim vol An = vol A and lim vol Bn = vol B.
n-400 n-400
Hence An + Bn C A + B, for every n E N, and so
vol (A -I- B)1"3 > vol (An + Bn)1"3 > (vol An)1/3 + (vol Bn)1/3
for every n e N. Taking limits as n tends to infinity, we obtain the Brunn-
Minkowski inequality, which is what we wanted to prove. 0
Now, given a compact connected surface, we look for a natural situation
where we can apply the Brunn-Minkowski inequality. In fact, takes > 0
sufficiently small so that N6 (S) is a tubular neighbourhood. For every t e
(0, s), the parallel surface St at oriented distance t is in the inner domain 1
determined by S. Hence, we also have It C 1, where 1Zt is the inner domain
determined by St; see Remark 5.33. Denote by B the open ball of radius t
centred at the origin of 3 . Then, if p e (1+B)nS, we have p = q + v with
q e SZt and v e B. Thus I p - q _ lvi <t, and hence dist (q, S) <t. But
this is impossible since the minimum of the square of the distance function
191 172 6. Global Extrinsic Geometry
to surfaces that are closed subsets of JR3. Thus, we will restrict ourselves
in this chapter to studying closed (as subsets of Euclidean space) surfaces
and mainly to compact surfaces. In this way, we will be able to use, as a
fundamental analytical tool, the theory of integration of Chapter 5.
The first problem that we will consider, which will repeat some questions
that we have already posed for curves, is to determine compact surfaces with
one or more of their curvatures having some specific behavior. For exam-
ple, can one characterize compact surfaces with constant Gauss or mean
curvature? We first consider the Gauss curvature. The local geometry of
surfaces with constant Gauss curvature was studied by F. MINDING in 1838,
with respect to their relation to non-Euclidean geometries. In the compact
case, we know from Chapter 3 that the only example is the sphere, as the
HILBERT-LIEBMANN theorem asserted. This result was first proved by H.
LIEBMANN in 1899 and was rediscovered by D. HILBERT in 1901 our proof
is a simplification of Hilbert's. It is then natural to generalize the ques-
tion and consider compact surfaces whose Gauss curvature does not change
sign. From what we have seen until now, we deduce that this sign must be
positive. We take up the study of compact surfaces with positive Gauss cur-
vature, the so-called ovaloids, and, more generally, positively curved closed
surfaces. Study of this class of surfaces was initiated at the beginning of the
20th century by H. MINKOWSKI and W. BLASCHKE in relation to the the-
ory of convexity. We will prove HADAMARD's theorem (1857), which states
that ovaloids are topologically spheres, and STOKER's theorem (1936), which
states that closed non-compact surfaces with positive Gauss curvature are
diffeomorphic to JR2. Furthermore, both are characterized as boundaries of
convex domains in Euclidean space. Just this information about ovaloids
will enable us to again prove the HILBERT-LIEBMANN theorem using as a
tool some integral expressions known as the MINK0w5KI formulas. We will
also obtain a new version of J. H. JELLETT's result (1853) which, in this
context, states that a compact surface with constant mean curvature, which
is star-shaped with respect to a given point, must be a sphere centred at
this point.
The last results about ovaloids lead to the second question that we posed
above: which compact surfaces have constant mean curvature? We know,
as mentioned in the introduction to Chapter 3, that the Gauss and mean
curvatures were introduced in the 18th century and, since then, they have
competed with each other. In spite of the fact that the initial competition
was won, in some sense, by the Gauss curvature, once GAUSS proved its
intrinsic character, the global problems on the mean curvature have proved
to be more complicated and, in some sense, more interesting. In fact, it
was only in the years between 1956 and 1962 that A. D. ALEXANDROV
succeeded in proving his celebrated theorem that, in our setting, asserts that
6.2. Positively curved surfaces 173 190 6. Global Extrinsic Geometry

the only compact surfaces with constant mean curvature are, again, spheres. P
This theorem was quite a surprising result, not because of its contents, but
because of the method that ALEXANDROV invented to prove it, one that was B
completely different from those of LIEBMANN and JELLETT. It was based
A2
on an ingenious use of properties of certain partial differential equations.
Al B
Including this proof is not feasible for a book of our level, and that is also
why this theorem, though it is basic to the global theory of surfaces, does
not even appear in the most prominent books on the subject. In 1978, R. A3
REILLY found a new proof that, although less geometrical, was easier to
understand and, especially, to explain. Unfortunately, it is also difficult to
include in a first course in differential geometry. During the last ten years,
the authors have studied some problems related to the mean curvature which Figure 6.6. Choice of P and Q
have allowed them to outline the proof that we will present here, based on
a paper by H. HEINTZE and H. KARCHER. Then A+ and A- are also finite unions of parallelepipeds whose faces are
Finally, the determination of compact surfaces with constant mean cur- parallel to the coordinate planes, namely
vature will lead us to answer one of the oldest questions in geometry, posed n+ n
by the Greeks: the isoperimetric problem, which can be stated as follows. A+ = U At and A- = A,
Among all the compact surfaces in JR3 whose inner domains have a given 2=1 2=1
volume, which has the least area? The ancient Greeks themselves knew that
the answer was spheres, as was first shown by H. A. SCHWARZ according where n+ <n and n- <n, since P separates at least Al and A2. On the
to earlier ideas of STEINER and MINKOWSKI, who had looked for the solu-
other hand, we can find another plane Q parallel to P such that
tion among convex domains. Our proof will consist, following [5] and [6], of vol A+ vol B+
obtaining the BRUNN-MINKOWSKI inequality, which is nothing more than (63)
vol A vol B
another property of Lebesgue measure, and showing that surfaces, for which where we have used the same notation as before. The reason is that, as soon
equality is attained, must have constant mean curvature. as P is chosen, the fraction on the left-hand side is a number between 0 and
Although we think that this chapter should furnish a more than ade- 1 and, if one takes Q on the left of all the blocks of B and displaces it to
quate introduction to global geometry of surfaces in Euclidean space, the the right until it surpasses all of them, then the fraction on the right side
compactness hypothesis required by the theory of integration we use and our is a continuous function of the position of Q taking values from 0 to 1; see
desire to not complicate this text excessively will keep us from studying neg- Figure 6.6.
atively curved surfaces or surfaces with vanishing Gauss curvature. Thus, In this way, since vol A+ + vol A- = vol A and vol B+ + vol B- = vol B,
a suitable follow-up for the reader would be the search for readable proofs we obtain
of the HILBERT and EFIMOV theorems on the non-existence of closed nega-
tively curved surfaces in JR3 or for the HARTMAN-NIRENBERG and MASSEY (6.4)
vol A
= ____
vol B
theorems, asserting that the only surfaces with null curvature are cylinders
having any plane curve as a directrix. (See Appendix 7.7. ) Moreover, B+ and B- are also disjoint finite unions of open parallelepipeds
with faces parallel to the coordinate planes
6.2. Positively curved surfaces m+ m-

Before determining the compact surfaces with constant Gauss or mean cur-
B+= B and B-=
j=1 j=1
vature, we will consider surfaces whose Gauss curvature does not change
sign. Recall Exercise 3.42 that, on any compact surface, there is always where m+ < m and m- < m, because we cannot suppose that Q separates
an elliptic point. Therefore, if the sign of the Gauss curvature does not two blocks of B. Hence we can apply the induction hypothesis to the pairs
6.5. The isoperimetric inequality 189 174 6. Global Extrinsic Geometry

and that, if A and B are disjoint, there exists a plane parallel to one of the
coordinate planes separating A and B.
Exercise 6.21. If A and B are two arc-wise connected subsets of prove
that the sum A + B is also an arc-wise connected subset.
Theorem 6.22 (The Brunn-Minkowski inequality). Let A and B be two
bounded open subsets of Euclidean space Then
(vol A)1/3 + (vol B)1/3 < (vol (A + B))1"3.
Proof. Suppose first that A and B are the parallelepipeds I1 X 12 X 13 and
J1 x J2 x J3, respectively, where the Ii and the Ji, i = 1, 2, 3, are bounded
open intervals of IIS with lengths a2 and bi. Then Figure 6.1. Ovaloid

rTi=1
3
1/3 3
1/3
(volA)1"3 + (volB)1!3 ai + \111 bi change, then it must be positive everywhere. We will use the term ovaloid
(vol (A + B))1/3 13 ' for any compact connected surface with positive Gauss curvature every-
(fl.1(ai + b))' where. This is equivalent, according to definition (3.5), to the fact that the
and thus second fundamental form is definite at each point and, taking Exercise (4) of
3 1/3 3 1/3 Chapter 4 into account, to the statement that the second fundamental form
(vol A)1!3 + (vol B)1!3 az bi relative to the inner normal is positive definite at each point. For example,
(vol(A + B))1!3 Z=1
a2 + b2 + Z=1
ai -I- b2 according to Exercise 3.24, all ellipsoids are ovaloids.
But the geometrical mean of three numbers is always less than or equal to More generally, in this section, we will give a complete description of the
their arithmetical mean. Hence shape that positively curved closed surfaces can have. In the following, S will
(vol A)1/3 + (vol B)1/3 1
3
ai
3
bi
denote a connected surface that is closed as a subset of the Euclidean space
1
1R3 and has strictly positive Gauss curvature everywhere. By Exercise (6)
(vol (A + B))1!3 - 3 i=1 az + bz 3 i_1 az + bz
of Chapter 4, which generalizes the Jordan-Brouwer and Brouwer-Samelson
So, in this particular case, the inequality claimed in the theorem is verified. theorems to closed surfaces, we know that the surface S separates 1R3 into
As a second case, suppose now that the bounded open sets A and B two connected components and that it is orientable. Since the second fun-
are disjoint finite unions of bounded open parallelepipeds whose edges are damental form of S, relative to any choice of normal field, is definite, we
parallel to the coordinate axes. That is, can choose as the orientation of S the one corresponding to the Gauss map
n m N : S -+ Sz for which the second fundamental form is positive definite at
A= Ai and B = B3 , each point. Then, all the half-lines starting at a point p E S with direction
i=1 j=1 N(p) lie locally on the same side of the surface, since any two points of S
where Ai and Bj are as in the first case above. We are going to prove the belong to a relatively compact open subset of S, which possesses a tubular
Brunn-Minkowski inequality for this kind of set using induction on the total neighbourhood. This privileged connected component 0 of R3 - S toward
number n + m of parallelepipeds involved. It is clear that the result is true all the half-lines in the direction of the so-chosen Gauss map point will be
if n + m = 2, since this is the particular case above. Assume that n + m > 3 called the inner domain of the surface S. It is clear that, if S is compact,
and that the desired inequality is true for disjoint finite unions of bounded this notion of inner domain coincides with the one that was already defined
open parallelepipeds with edges parallel to the coordinate axes and total in Remark 4.20. Examples of closed connected surfaces with positive Gauss
number less than n + m. Since n + m> 2, we have either n> 1 or m> 1. curvature are, according to Example 3.45, the elliptic paraboloids.
Suppose the first case. From Exercise 6.20, there exists a plane P parallel to A concept related, in a natural way, to the Gauss curvature is that of
the coordinate axes separating Al and A2. Let P+ and P- be the two open the convexity of surfaces. Usually, a surface is said to be convex when it lies
half-spaces in which P divides IIS3 and let A+ = A f1 P+ and A- = A f1 P. on one of the closed half-spaces determined by the affine tangent plane at
6.2. Positively curved surfaces 175
Figure 6.2. Inner domain
each of its points. We will say that it is strictly convex when it is convex and
touches each tangent plane only at one point. It is worth pointing out that,
at first, this notion of convexity does not have much to do with the notion
of affine convexity. However, we have the following result relating them.
The one that refers to compact surfaces was first shown by Hadamard in
1857 and the analogous statements for closed surfaces are consequences of
a result due to Stoker in 1936.
Theorem 6.1 (First part of the Hadamard-Stoker theorem). Let S be a
connected surface closed as a subset of 113 with strictly positive Gauss cur-
vature, and let SZ be its inner domain. Then the following are true.
A: For each x, y E SZ, ]x, y[C R. In particular, Si is a convex subset
ofiRe.
B: The intersection of S with the affine tangent plane of S at any of
its points is only the point itself. Moreover, a is contained in all
the closed half-spaces of 13 determined by the acne tangent planes
and the inner normals of the surface.
Proof. Part A. We first show that St is convex. Consider the subset A =
{ (x, y) E c x SZ I [x, y] C 1l} of the product S2 x St. Since SZ is connected, if
it were not convex, then the boundary of A in S2 x St would be non-empty.
Take a pair (x, y) in Bdry A. By definition of the topological boundary,
there are sequences {xn}fEN, {4}1EN, {yn}nEN, and {4}1EN of points of
St such that limnyw xn = x' = x, yn = y; = y,
and [xn, yn] C St, [x', y',,.] ci, for each n E N. This implies that one
could find a point z E]x, y[f1S of first contact between the segment and the
surface. From Lemma 4.15, we deduce that Ix, Y [c S2 would lie in the affine
tangent plane of S at z and that the height function relative to this plane,
when restricted to the normal section at z in the direction of the tangent
segment ]x, y[, would attain a maximum at z. On the other hand, we said
188 6. Global Extrinsic Geometry
1o=
Figure 6.5. Sum of sets
6.5. The isoperimetric inequality
Given a compact connected surface S in Euclidean space, it makes sense,
after Chapter 4 where we proved the Jordan-Brouwer theorem, to talk about
the volume enclosed by S and, after Chapter 5, about the area of the surface
S. Thus, it is natural in this context to pose the isoperimetric question,
which can also be considered in the case of plane curves (see Chapter 9) :
Among all the compact surfaces enclosing a given volume, which has the
least area? To give an answer, we will first establish a useful geometrical
property of the Lebesgue integral on Ii: the Brunn-Minkowski inequality.
If A and B are two arbitrary subsets of Euclidean space W, we define
its sum A + B as the set
A+B = {a+bla E A, be B}.
For example, if A is the open interval (al, a2) of IIS and B is the open interval
(bi, b2) of TI, the sum is
A + B = (ai, a2) + (bi, b2) = (ai + a2, bi + b2)
The following exercises, which point out some of the properties of this sum
of sets, will be used in what follows.
Exercise 6.18. 1 Let A, B C W. If at least one of these subsets is open,
prove that the sum A + B is open as well.
Exercise 6.19. Suppose that A, B C W are two bounded subsets of Eu-
clidean space. Show that A + B is bounded.
Exercise 6.20. 1 Let Ii, 12,13 and Ji, J2, J3 be six bounded open intervals
of the real line. The two subsets of TI, given by
A=I1x12x13 and B=J1xJ2xJ3,
are called parallelepipeds with edges parallel to the coordinate axes. Prove
that
A + B = (I1 + J1) x (12+J2) x (13+J3)
6.4. The Alexandrov theorem 187 176 6. Global Extrinsic Geometry

Clearly n(F, XA) > 1 on F(A) j 1 and, then, also using F ubini's theorem,
we have
vol S2 <
Js Jo
a Xa(p't) IJac FI (p, t) dt I dp.

We now substitute the absolute value of the Jacobian of F and take into
account the definition of the set A. Thus we have
p / fl/k2(p)
vol St < I (/ I(1 - tkl(p))(1 - tk2(p))Idt dp.
is \Jo
But when 0 < t < 1/k2(p), the previous integrand is non-negative. More-
over, using inequality (3.6), we see that
(1 - tkl(p))(1 - tk2(p)) = 1 - 2tH(p) + t2K(p) < (1
and equality is attained only at the umbilical points. Hence, since 1/k2(p) <
-
1/H(p) for each p e S and since the function (1 tH(p))2 is non-negative,
1/x(P)
vol St < / (1-tH(p))2dt) dp,
( /
s o
with equality only if the surface S is totally umbilical. The proof is com-
pleted by computing the integral on the right-hand side and recalling Corol-
lary 3.31, which classified closed totally umbilical surfaces.
The Heintze-Karcher inequality that we have just proved, applied to the
case of surfaces of constant mean curvature, and the first Minkowski formula
supply complementary information. They can be combined to yield one of
the most beautiful theorems of classical differential geometry.
Theorem 6.17 (Alexandrov's theorem). If a compact connected surface has
constant mean curvature, then it is a sphere.
Proof. If S is a surface satisfying the conditions of the theorem, then The-
orem 6.16 implies that vol 1Z < A(S)/3H, where 1 is the inner domain of S
and H > 0 is the value taken by the mean curvature relative to the inner
normal. Moreover, if equality occurs, then S is a sphere. But, indeed, equal-
ity does occur, since in our case the first Minkowski formula of Theorem 6.11
gives
A(S) + H (p, N(p)) dp = 0.
J the set S - {p} is connected and does not intersect H. But, since there
It is enough to note that
(p, N(p)) dp = -3vo1 SZ,
Is
according to Remark 5.32.
Figure 8.3. Hadarnard-Stoker's theorem
- tH(p))2
that the positivity of the Gauss curvature forces the second fundamental
form relative to the orientation chosen on S to be positive definite at each
point. Then, Remark 3.28-see also Proposition 3.37-says that this same
height function attains a local strict maximum at z. This is a contradiction
and, thus, S2 is a convex subset of Euclidean space. Now, an elementary fact
about convex sets is that the closure of a convex set is convex (the reader
should try to prove this as an exercise). Therefore, if part A were not true,
then there would exist two points x, y E S such that ]x, y[f1S 0 and this is
impossible from the above. In fact, arguing in the same way, we would have
a first contact point at each z E]x, y[f1S and this is again a contradiction.
Part B. Let p E S be a point of the surface. Let IIp denote the affine
tangent plane of S at p, and let rHp denote the closed half-space of R3
determined by this plane ILp and the inner normal vector N(p). In other
words,
IIp+ = hp1[D, +oo) where hp(q) = (q - p, N(p)),
for each q E 113, is the height function relative to H.
Suppose now that q E Hp f1 S for some p E S. Then p and q belong
to 1 and, from part A, we would have ]p, q[C ft n Hp. But, since p is an
elliptic point, Proposition 3.37 and Remark 3.28 tell us that all the points in
a neighbourhood of p in Hp, except p itself, lie in the outer domain of S, and
this leads us to a contradiction, unless p = q and so ]p, q[= 0. Therefore,
IIp l S = {p}, for each p E S, which is the first assertion of part B.
To see that the second assertion is also true, observe that, for p E S,
are points of S near p that lie in IIi,+, since p is an elliptic point, we have
S - {p} C IIp+ and, since p E Hp, S C Hp. Hence R3 - Hp C R3 - S.
Thus, the connected open set 13 -- IZp belongs to one of the two connected
D components of the complement of S and has points in the outer domain
6.2. Positively curved surfaces
of S, again since p is elliptic. Consequently, R3 - ll c IL Taking
complements, we conclude that 11 C ll , and this holds for each p E S. D
Remark 6.2. If p e S, assertion B of Theorem 6.1, 11 C ll , is equivalent
to
h(q) = (q-p,N(p)) 0, Vq E1,
and the equality implies that q = p. Hence
(p-q,N(p))<O, `dp E S b'q E SZ;
that is, the support function of S corresponding to the inner normal based
on any point of 1 2 is negative everywhere. In particular-see Exercise (4) of
Chapter 3-we have that S is star-shaped with respect to any point of its
inner domain.
Remark 6.3 (Intersection of a straight line and an ovaloid). Assume that
the surface S is an ovaloid and that SZ is the inner domain determined by
it. Let R be a straight line of cutting the inner domain, that is, with
R f11 0. As a consequence of the Hadamard-Stoker Theorem 6.1, R f1 SZ
is a non-empty bounded convex set, which is open as a subset of the line
R. Thus, it has to be an open interval. This, along with the fact that R
cannot be tangent at any point of S, by part B of Theorem 6.1, implies
that R f1 S consists of exactly two points; that is, a straight line touching
the inner domain of an ovaloid cuts it at exactly two points. See Exercise
(6) at the end of this chapter.
Theorem 6.1 above asserts that the inner domain of a closed connected
surface with positive Gauss curvature is convex and that, furthermore, the
surface itself is strictly convex. To proceed with this type of result, we will
need a lemma, in which we will use the following notation. If x e Il83 is a
point of Euclidean space and v e ][83 - {0} is a non-null vector, we denote
the straight line passing through x in the direction of v by R(x, v) and by
R+(x, v) the set
R(x,v) ={x+tvlt O},
that is, the half-line with origin at x and direction v.
Lemma 6.4. Let S be a closed connected surface with positive Gauss cur-
vature and let SZ be its inner domain. Then the following are true.
a: If R+(x, v) C S2 for some x e SZ and v e Il83 - {0}, then R+(y, v) C
St and, moreover, R+(y, v) - {y} C SZ, for each y e SZ.
b: There are no straight lines contained in ft
Proof. Part a. If we suppose that, for some x e SZ and a certain non-null
vector v, R+ (x, v) C SZ, then x -I- tv E 11 for each t > 0. If y E 11, by part
177 186 6. Global Extrinsic Geometry
Exercise 6.14. T Use the Minkowski formulas to show that, if the Gauss
and mean curvatures coincide everywhere on an ovaloid, then it is necessarily
a unit sphere.
Exercise 6.15. 1 Manipulate the Minkowski formulas to prove that an
ovaloid such that HK = 1 has to be a unit sphere.
6.4. The Alexandrov theorem
The study and characterization of compact surfaces with constant mean
curvature are typically more involved than the case of constant Gauss cur-
vature. The Alexandrov theorem solves this problem completely. As we did
in the case of the Gauss curvature, we will examine the case in which this
curvature does not change its sign before considering the problem of con-
stant mean curvature. It is worth noting that the fact that this sign is either
positive or negative is not significant, but it depends on the orientation that
we have chosen on the surface. In any case, if the mean curvature does not
vanish anywhere when we take the inner normal field on a compact surface,
then it must be positive, as Exercise (4) of Chapter 4 shows.
Theorem 6.16 (The Heintze-Karcher inequality). Let S be a compact sur-
face whose (inner) mean curvature H is positive everywhere. Then
vol SZ
1
_
1
dp,
3 s H(p)
where 1 is the inner domain determined by S. Moreover, equality holds if
and only if S is a sphere.
Proof. Define a subset A of S x IIS by
< 1
t
A={(p,t)SxIO<
k2 (p)
where k2 is the largest principal curvature of S corresponding to the inner
normal which, since k1 < H < k2, is positive everywhere. If we denote
by a > 0 a real number larger than the maximum on S of the continuous
function 1/12, then A is a compact set contained in S x [0, a). Exercise (8)
of Chapter 4 says that the inner domain 1 determined by S satisfies
1 c F(A),
where F : S x IIS -* IIS3 is the map defined in (4.1) by F (p, t) = p + tN (p)
for all p e S and t e R. Now, apply the area formula of Theorem 5.27 on
S x (0, a) to the map F and the characteristic function xA of the set A to
obtain the integral equality
xA IJac FI.
n(F, XA) =
113 S x (O,a)
6.3. Minkowski formulas and ovaloids 185
Proof. Let S be a surface satisfying our hypotheses. We know, by Exer-
cise 3.42, that K is a positive constant. Then S is an ovaloid to which we can
apply the Hadamard-Stoker Theorem 6.5. After doing this, we are assured
that the support function based on any inner point is negative. Now, we
may also suppose that the origin of IIis in SZ. If not, a suitable translation
would take it to this situation without changing either the hypotheses or
the conclusions of the theorem. The Minkowski formulas the second one
divided by tt?-give us
IH(p)
(1 + (p, N(p))H(p)) dp = 0 and +(p,N(p))) dp=0.
Is I Using again part A of the Hadamard-Stoker Theorem, we would get
Subtracting these two equalities, we have
((1 - H(P)) + (p,N(p))(H(p) - dp= 0.
I which is impossible.
But, since H2 > K, we obtain that either H(p) > /it? or H(p) < -s/k, for
all p e S. The correct alternative is the first one, as Exercise (4) of Chapter
4 shows. Therefore
1- H(p) < 0 and (H(p) - v'k) (p, N(p)) 0
at each point p of S. Then, the integrand above is non-positive and has
vanishing integral. Using Proposition 5.6, we infer that the integrand is
zero. This implies that H2 = K and then S is totally umbilical. Thus, to
complete the proof, apply Corollary 3.31. D
Corollary 6.13 (Jellett's theorem). The only compact connected surfaces
that are star-shaped with respect to a given point and have constant mean
curvature are the spheres centred at this point.
Proof. If S is a compact connected surface and it is star-shaped with respect
to, say, the origin of TI, then 0 E SZ, where SZ is the inner domain of S.
Otherwise, from Exercise (9) of Chapter 6, there would exist some line
tangent to S passing through the origin. Since the mean curvature H of S
is constant, we write the Minkowski formulas, the first one multiplied by the
constant H, and we have
Is
(H+H2(p,N(p)))dp= 0 and
Subtracting the two equalities, we get
f (H+K(p)(p,N(p)))dp = 0.
f (H2 - K(p))(p, N(p)) dp = 0.
But the integrand is non-positive, from inequality (3.6) and Exercise (4) of
Chapter 3. We conclude the proof as we did in the previous result. D is a diffeomorphism between S and the sphere S2.
178 6. Global Extrinsic Geometry
A of the Hadamard-Stoker Theorem 6.1 above, we have ]y, x -I- tv [ C 11, for
t > 0. Taking the limit as t - oo, we obtain that R+ (y, v) C 1. If this
half-line had some point z of S different from its origin, we would have a
contradiction to part A of Theorem 6.1.
Part b. Suppose that there is a straight line R in 11. We pick a point
x E R and let v e S2 be a direction vector of R. Then the two half-lines
R+(x, v) and R+(x, -v) are in 11. Thus, if p e S C 11, by part a, the
half-lines R+ (p, v) and R+ (p, -v) are contained in SZ. Therefore,
p -I- v e R+(p, v) C 12 and p- v e R+(p, -v) C SZ.
pE]p-v,p+v[CSt,
Theorem 6.5 (Second part of the Hadamard-Stoker Theorem). Let S be a
connected surface, closed as a subset of Il83 and with strictly positive Gauss
curvature.
C: If S is an ovaloid, that is, if S is compact, then the Gauss map
N : S - S2 is a diff eomorphism. Consequently, S is diff eomorphic
to a sphere.
D: If S is not compact, then the Gauss map N : S - S2 is injective
and S is a graph on an open convex subset of a plane.
Proof. We are assuming that the surface S has positive Gauss curvature
everywhere. Then, equation (3.4) implies that the Gauss map N : S -
is a local diffeomorphism, by the inverse function theorem (Theorem 2.75).
We now show that, under our hypotheses, N must be injective. Indeed,
if p, q E S are two points of the surface with N(p) = N(q), then the two
corresponding affine tangent planes L[ and IIq are parallel. By part B of
Theorem 6.1, we have S C SZ C IIp f1 IIq . This intersection coincides with
one of the two half-spaces, since the boundary planes are parallel, say, with
IIp . Hence q e IIq C Hp and thus q E IIp f1 IIp = IIp and IIp = IIq .
Assertion B also gives us that
{p}=Sflllp=Sflllq={q}.
Then p = q and N is injective, whether or not S is compact.
Part C. If we suppose now that S is compact, then N(S) C 2 is com-
pact, and thus, it is closed as a subset of the unit sphere. Since N is a local
diffeomorphism, this image is also an open subset of 2 and, by connected-
ness, N(S) = S2; that is, N is surjective. In conclusion, the Gauss map N
6.2. Positively curved surfaces 179 184 6. Global Extrinsic Geometry

Part D. Now consider the case where S is not compact and, hence, it is of a vector field defined on Euclidean space. Putting all this into (6.2), it
not bounded. Then, we may select a sequence {n}neN of points qE S such follows that
that qn I = +oo Thus, if we fix every q e S, since q, qn e S C SZ,
by part A of Theorem 6.1, we have [q, qn] C St for each n e N. Extracting aJ divV - la2J [k2(p(dV)(ei), ei)+ki(p)((dV)p(e2), e2)J dp
s 2 s
a subsequence, if necessary, we can find a e S2 such that R+ (q, a) C SZ. We
apply part a of Lemma 6.4 above to deduce that
_ -2a f(V N)H + a2 f(V, N)K,
R+(p, a) - {p} C SZ, `dp E S.
From this, if p e S and if there were another point p' E S in R(p, a), either which is valid for each a E (0, s). Comparing coefficients in this polynomial
p or p' would be on the open half-line starting from the other point in the equality, we have established the following theorem.
direction of a and, so, it would belong to SZ. Therefore Theorem 6.10 (Divergence theorem for surfaces). Let S be a compact sur-
R(p, a) fl S = {p}, `dp E S. face and let V : S -* 13 be a differentiable vector field defined on S. Then
the following hold.
Furthermore, none of these lines can be tangent to S. The reason for this
is that, if R(p, a) were tangent to S, of course at p, then there would be a A: div V = -2 (V, N)H.
neighbourhood of p in the line that would be contained, except for p itself, Is J
in the outer domain of S-as we have already argued several times above- B:
which would contradict the fact that R+(p, a) - {p} C SZ. Consequently, the Js [k2(p)((dV)(ei), ei)+kip)((dV)p(ea), e2)] dp=-2J(V,
s
N)K,
orthogonal projection P of ][83 to the plane where {ei, e2} is an orthonormal basis of principal directions at the
point p E S.
ll={xeIt3I(x,a)=O}
As usual, H and kl, k2 stand for the mean and the principal curvature func-
passing through the origin and perpendicular to a is injective when restricted tions of the surface.
to S. Moreover, if p e S and v e TS is in the kernel of (dP), we have
0 = (dP)p(v) = P(v) = v- (v,a)a. As a direct consequence we can obtain the Minkowski integral formulas;
the reader should try to deduce them also from Exercise (9) of Chapter 5
Since a TS, we infer that v = 0. That is, the projection P is also a local and Example 5.19.
diffeonnorphism, by the inverse function theorem. The final conclusion is
that P is a diffeomorphism between S and its image in II, which is an open Theorem 6.11 (Minkowski formulas). Let S be any compact surface, N its
convex subset of this plane. In other words, S is a graph over this open set inner Gauss map, and H and K its mean and Gauss curvatures. Then, we
of a function which is necessarily convex, since SZ is convex. have the following two integral equalities:

Remark 6.6. If S is an ovaloid, assertion C of the second part of the A (1 -I- (p, N(p))H(p)) dp = 0,
Hadamard-Stoker Theorem 6.5 says that N : S -* 2 is a diffeomorphism. Is
We can apply the change of variables formula (Theorem 5.14) to the constant B (H(p) + (p, N(p))K(p)) dp = 0.
function 1 on the unit sphere and obtain Is
4 = A(S2) = 1= lJac NI. Proof. It suffices to apply Theorem 6.10 to the vector field V S -* I[83
J2 Js given by V (p) = p for all p e S. D

(= I
But-see Exercise (2) at the end of Chapter 5-if p e S,
n (dN)p(e2)I,
where {ei, e2} is any orthonormal basis of TS. For example, taking ei and
From these two Minkowski formulas we can deduce, in an alternative
way, the two global theorems that we established in Section 3.6: the Jellett-
Liebmann and Hilbert-Liebmann results in Corollaries 3.48 and 3.49.
e2 as principal directions of S at the point p, we have Corollary 6.12 (The Hilbert-Liebmann theorem). A compact connected
(JacN)(p)I = Iki(p)k2(p)Ilei n e21 = K(p), surface with constant Gauss curvature must be a sphere.
6.3. Minkowski formulas and ovaloids 183 180 6. Global Extrinsic Geometry

where N and Na are, respectively, the inner and the outer normal fields
of the surface S and of the parallel Sa at distance a e (0, ) We use the .

change of variables formula in the second integral on the right-hand side for
the diffeomorphism Fa : S * Sa that we already studied in Remark 4.29.
Moreover, if we take into account that Na o Fa = -N and that X o Fa = V,
we have see also Example 5.17

_- f(V, N) (1 - 2aH -I- a2K).

Therefore
(6.2)
f a(s)
divX = -2a
Js
(V, N)H -I- a2
Js
(V, N)K, Figure 6.4. Lemma 6.7
for all a with 0 < a <. On the other hand, we will now compute the diver-
gence of the vector field X that we have built on the tubular neighbourhood since the Gauss curvature is positive. It follows that
NE(S) from the vector-field V given on the surface S. For each p e S and
t e (-s, ), we have, from Definition (6.1), is
(dV)p(ez(p)) Z_1Z
0 and (dX)P+tN&(ei) _ 1 - tk
These two Hadamard-Stoker theorems are the strongest results that one
where {ei, e2, N(p)} is the basis of Il83 formed by the unit normal of S at p can expect for positively curved closed surfaces. However, we can obtain
and the principal directions of the surface at this point. Thus more information about this type of surface if we assume that at least one
((dV) (ei), ei) e2) of the curvatures of the surface, Gauss or mean curvature, is bounded away
( di v X )(p+ tN (p)) = trace ( dX )+tN _ 1 - tki(p) 1 - tk2(p) from zero. To arrive at this kind of result, we will need to show that some
pieces of the surfaces that we are dealing with can be expressed as graphs.
for all p e S and t e (-s, ). The left-hand side of (6.2), after using (5.1)
and Theorem, becomes Lemma 6.7. Let P be a plane cutting the inner domain S2 of a closed
fa f connected surface S with positive Gauss curvature. Then there exists an
f a (s
divX =
J Js [(divX) o F](p, t)IJacFI(p,t) dpdt.
o
open subset V of S that is a graph over P fl Q.
If we substitute the expression I (1 - tk1 (p))(1 - tk2(p)) and Proof. Denote the non-empty open subset P fl Il of the plane P by U and
the value of the divergence of X that we have just computed, we obtain suppose that, given v E S2, normal to P, there is a point y c U such that
[(divX) o F](p, t)IJacFI(p, t) _ ((dV)(ei),ei) + ((dV)(e2),e2) the half-line R+(y, v) does not intersect the surface S. Then R+ (y, v), which
is connected, must lie in one of the two connected components of 1R3 --- S
-t[k2(p)((dV)(ei), ei) + ki(p)((dV)(e2), e2)]. and, since y E R+(y, v) fl 0, we will have R+(y, v) C Q. Thus, by part a of
Lemma 6.4, we will obtain that R+ (x, v) C 0 for each x E U. But part b
Incidently, this calculation shows that the functions of the same lemma asserts that ci contains no straight lines. Consequently,
p E S((dV)(ei), ei) + ((dV)(e2), e2), any half-line R+(x, -v) cannot be entirely included in i, whatever x E U
is. However, each of them has the point x E R+ (x, -v) in Q. Hence, we
p E S'-' k2(p)((dV)p(ei), ei) + ki(p)((dV)p(e2), e2), have as a consequence that R+(x, -v) fl S # 0 for each x E U.
defined on the surface S from the vector field V are differentiable, where We may assume from the above that there exists a vector v E S2, normal
{ei, e2} is an orthonormal basis of the principal directions of TS. The first to P, such that all the half-lines R+ (x, v) starting at the points x of U in the
function will be denoted by div V and will be called the divergence of the direction of v meet the surface S. Thus, we have, by Theorem 6.1, that each
field V. The reason is obvious if we recall Definition 5.30 of the divergence intersection R+(x, v) f1 S2 is a non-void bounded convex open subset of the
6.2. Positively curved surfaces 181
half-line R+(x, v), which, moreover, contains its origin x, for every x E U.
Therefore, each R+ (x, v) fl S has a unique point for each x E U and also
the straight line R(x, v) is not tangent to S at this point, since otherwise,
by part B of Theorem 6.1, R(x, v) could not contain any points of the inner
domain S1. This means that, if f : S - P is the restriction to S of the
orthogonal projection onto the plane P, we have that f is a diffeomorphism
from the open subset V = f1(U) of S to U, that is, V is a graph over the
open set U. D Hadamard-Stoker Theorem 6.5 imply that S is a graph over a convex open
We will now show that, on a closed connected positively curved surface
that is not compact, the Gauss curvature must approach zero at infinity and
the fact that the mean curvature is bounded away from zero on this class of
surfaces supplies us with some bounding properties.
Theorem 6.8 (Bonnet's theorem). Let S be a connected surface that is
closed as a subset of IE83. The following assertions are true:
A: If infpES K(p) > 0, then S is compact, that is, S is an ovaloid.
B: If S is non-compact, has positive Gauss curvature, and its mean
curvature satisfies infS H(p) = h > 0, then S is a graph over a
convey open subset of a plane whose diameter is less than or equal
to 2/h.
Proof. Part A. Assume that S is not compact and denote by k > 0 the
infimum of the Gauss curvature on S. By part D of the Hadamard-Stoker
Theorem 6.5, we know that S is a graph over a convex open subset of a plane
P and that the Gauss map is a diffeomorphism of S over an open subset of
S2. Thus, if R is any region of the surface S, one has
k A(R) = k < f K(p) dp = I (Jac N) (p) I dp
J J
Using the change of variables formula, we see that
1=A(N(1))<47r,
and this bound serves for each region R of the surface S. On the other
hand, if ]a, b[ is an open segment of the inner domain SZ of S, we will have
R+(x, v) C St for some v e 2 normal to the plane P and each x e ]a, b[.
But, if n e I`N, the set
Un = {x+tvIx E ]a, b[, 0< t <n}
is an open subset of the plane Q determined by ]a, b[ and the vector v.
Applying Lemma 6.7 to this plane Q, we deduce that, for each n E I`N, there
182 6. Global Extrinsic Geometry
exists an open subset Rof the surface that is a graph over the open subset
Un of the plane Q. Therefore, by Exercise (13) of Chapter 5, we obtain that
4k > A(Rn) > A(Un) = nia - bI
for all natural numbers n. This is absurd and, hence, S must be a compact
surface.
Part B. As in the previous case A, our hypotheses and part D of the
subset U of some plane P. Let ql and q2 be any two points of U, and let
pl and p2 be the two points of S projecting onto them. One of the two
unit vectors normal to the plane P, which we will represent by v, satisfies
R+(pl, v) C 12 and, from Lemma 6.4, R+(x, v) C SZ for each x E ]pl, p2[
Then, the set
V = {x +tv lx E ]pi, p2 [, t > 0}
is an open subset of the plane Q determined by the segment ]pi , P2[ and the
vector v. Applying Lemma 6.7 to this plane Q, we infer the existence of an
open subset S' of the surface S that is a graph over this open subset V of
Q. Note now that this V belongs to a plane strip whose width is Ii - q21.
We complete the proof using Exercise (18) at the end of Chapter 3. O
Remark 6.9. The above result implies that for positively curved closed sur-
faces, there are two or three independent directions, depending on whether
we do or do not allow the mean curvature or the Gauss curvature to come
near zero, with respect to which the surface is bounded. Assertion A con-
cerning the Gauss curvature is a weak version of a result, due to Bonnet,
which has a very well-known generalization to R.iemannian geometry: the
Bonnet-Myers theorem.
6.3. Minkowski formulas and ovaloids
Let S be a compact surface. We will work with differentiable vector fields
V on this surface, as introduced in Definition 3.1. That is, V : S -> IE83 is a
differentiable map. If NE(S) C Il83, with e > 0, is a tubular neighbourhood
of the surface, we define a vector field X on it-see Definition 5.30-by
extending V as a constant along the normal segments of N6 (5). That is,
(6.1) (X o F) (p, t) = X (p -I- tN(p)) = V (p), b'(p, t) E S x (-e, ),
where the map F was introduced in (4.1). As a consequence of this defini-
tion, X is differentiable as well. We now apply to this vector field X the di-
vergence theorem (Theorem 5.31) on the regular domain Va(S), a E (0,E)-
see Remark 5.33-and obtain
f a(S)
div X =-
JS
(X, N) - f (X, Na),
a