ELEMENTS

OF
DIFFERENTIAL
GEOMETRY

RICHARD S. MILLMAN

GEORGE D. PARKER

Southern 1/linois University

Carbondale,Illinois

Prentice-Hall Inc., Englewood Cliffs, New Jersey 07632

Library of COll6rtSICatato,1ng In P1'bllcatlonData
MILLMAN, RICHARD S (date)
Elements of differential pometry,
Biblioaraphy: p. 248
Include• Index.
I, Geometry, Differential. I. Parker,
Oeorae D., (date) joint author. II. Title.
QA641,MS2 Sl6',3604 76-28497
ISBN 0·13-264143·7

© 1977 by PRENTICE-HALL, INC.

Englewood Cliffs, New Jersey 07632

All rights reserved. No part of this book may

be reproduced in any form or by any means without
permission in writing from the publisher.

Printed in the United States of America

10 9 8 7 6 S 4 3 2

PRENTICE-HALL INTERNATIONAL, INC., London

PRENTICE-HALL OP AUSTRALIA PTY. UMITED, Sydney
PRENTICE-HALL OP CANADA, LTD., Toronto
PRENTICE-HALL OP INDIA PRIVATE UMITBD, New Delhi
PRENTICE-HALL OP JAPAN, INC., Tokyo
PRENTICE-HALL OP SOUTHEAST ASIA PTB. LTD., Singapore
WHl'I'BHALL BOOKS LIMITBD, Wellington, New Zealand
 Contents

Preface xi

1 Preliminaries 1
1-1. Vector Spaces
1-2. Linear Transformations and Eigenvectors 4
1-3. Orientation and Cross Products 6
1-4. Lines. Planes, and Spheres 8
1-5. Vector Calculus 10

2 Local Curve Theory 13

2-1. Basic Definitions and Examples 14
2-2. Arc Length 20
2-3. Curvature and the Frenet-Serret Apparatus 24
2-4. The Frenet-Serret Theorem and Its Corollaries 29
2-5. The Fundamental Existence and
Uniqueness Theorem for Curves 41
2-6. Non-Unit Speed Curves 46

vii
viii

3 Global Theoryof Plane Curves 49

3-1. Line Integrals and Green·sTheorem 50
3-2. The Rotation Index of a Plane Curve 52
3-3. Convex Curves 60
3-4. The lsoperimetric Inequality 63
3-5. The Four-Vertex Theorem 66
3-6. A Preview 72

4 Local Sudace Theory 74

4-1. Basic Definitions and Examples 76
4-2. Surfaces 88
4-3. The First Fundamental Form and Arc Length 9
4-4. Normal Curvature, Geodesic Curvature.
and Gauss's Formulas 102
4-5. Geodesics 109
4-6. Parallel Vector Fields Along a Curve and
Parallelism 116
4-7. The Second Fundamental Form and
the Weingarten Map 122
4-8. Principal. Gaussian. Mean. and Normal Curvature~
4-9. Riemannian Curvature and
Gauss's Theorema Egregium 141
4-10. lsometries and the Fundamental Theorem of
Surfaces 146
4-11. Surfaces of Constant Curvature 153

5 Global Theoryof Space Curves 161

5-1. Fenchers Theorem 161
5-2. The Fary-Milnor Theorem 167
5-3. Total Torsion 170

6 Global Theoryof Surfaces 173

6-1. Simple Curvature Results 174
6-2. Geodesic Coordinate Patches 176
6-3. Orientability and Angular Variation 180
6-4. The Gauss-Bonnet Formula 185
6-5. The Gauss-Bonnet Theorem and
the Euler Characteristic 188
6-6. The Theorems of Jacobi and Hadamard 192
6-7. The Index of a Vector Field 195
Conltlllts

7 Introduction to Manifolds 198

7-1. Some Analytic Preliminaries 199
7-2. Manifolds-Definition and Examples 203
7-3. Tangent Vectors and the Tangent Space 208
7-4. Vector Fields and Lie Brackets 216
7-5. The Differential of a Map and Submanifolds 219
7-6. Linear Connections on Manifolds 223
7-7. Parallel Vector Fields and Geodesics on a
Manifold with a Linear Connection 227
7-8. Riemannian Metrics. Distance, and Curvature 232

Appendix: Historical Notes 243

Bibliography 248

Index
I. Notational 255
II. Topical 257
 Preface

The intent of this book is to provide an elementary and geometric intro-

duction to differential geometry. We adopt an approach that is elementary
enough for undergraduates but still both conveys the spirit of modem differ-
ential geometry and prepares the student for a more advanced course on the
subject. In particular, we have given significant emphasis to global considera-
tions while providing the basic classical results on curves and surfaces. In
these global results we have been guided in part by the beautiful article of
Chern [1967].This book is an outgrowth of courses taught by the first author
at Ithaca College and by both authors at Southern Illinois University over a
period of six years.
In differential geometry there is calculus and there is geometry, neither
of which should be slighted. All too often the geometry is hidden in either
machinery, abstraction, or symbolism. Furthermore, an unfortunate thing has
happened to the subject in the last ten years-it has been relegated to the
graduate curriculum. There is no question that differential geometry (or any
subject for that matter) can be done more efficiently if the student has the
background that every second year graduate student has, but to wait until
graduate school to teach the subject is doing an injustice to both the student
xi
xii Preface

and the subject. We hope that this book will aid in the return of differential
geometry to the undergraduate curriculum and that it can be taught by non-
specialists, as well as by specialists.
A traditional undergraduate course in classical differential geometry
usually hides the geometry in a myriad of symbols such as r,/ or Rlk,·We
have minimized this difficulty by making use of linear algebra throughout.
Modem vector space terminology can and should be used effectivelyto make
the material more comprehensible to the students. Furthermore, we find that
the students are very excited about using the linear algebra they have just
learned in another course. They view differential geometry as an application
of linear algebra, which it is. After the students understand the geometric
content of a result (by use of linear algebra) the result is then expressed in
classical notation. This enables the reader to consult texts in classical differ-
ential geometry without having to develop a whole new vocabulary.
In the advanced approach, which is usually restricted to the graduate
level, the geometry is hidden in the machinery and abstraction. We have
avoided topics which require building complicated machinery, such as differ-
ential forms or cohomology. We have resisted the temptation to prove the
most general results possible if these would require a lot of machinery to
state or to prove. In particular, for the first six chapters we stay in R 3 • In order
to avoid the trap of abstraction we include many examples-not only of the
definitions but also of the theorems because it is usually much easier to
follow a proof with a firm example in mind.
When first introduced to differential geometry, individuals often have
trouble reconciling the two views of the subject-classical and modem (mani-
folds). It is sometimes hard to tell what the connection (no pun intended) is
between the two approaches. We attempt to remedy this situation in the last
chapter. There we explain how the earlier material motivates the definitions
of manifold, tangent space, Riemannian metric, etc. and give a brief intro-
duction to the modem terminology. We hope that having finished this
chapter the reader will have an easier transition to books such as Hicks [1965],
Boothby [1975], Warner [1971], Matsushima [1972], or Kobayashi and
Nomizu [1963).
There is only one way to learn mathematics and that is to get your hands
dirty working problems. Differential geometry is no exception. We have
included more than 350 problems. The easier problems are usually at the
beginning of the problem sets. Those whose results are used in the text are
marked with an asterisk (*). There is one exception to this rule-in Chapters
1-6 we have not starred any problems which would be needed in Chapter 7.
It should be noted that an asterisk does not denote a hard problem: we have
reserved the dagger (t) for this.
In Chapter 1 we present the material which is the necessary background
 xiii

for the book. Most of it is review and may be emphasiz.edas needed. We have
found that our students are generally weak in describing lines and planes in
vector notation. The material on eigenvalues and linear transformations is
not used until Chapter 4 and may be delayed until then.
Chapter 2 develops the basic local theory of space curves. The Frenet-
Serret Theorem which expresses the derivative of a geometrically chosen
basis of R 3 in terms of itself is proved, and many important corollaries are
derived. We include the Fundamental Theorem of Curves, showing the depen-
dence of differential geometry on the theory of ordinary differential equations.
Several classical topics are covered in the exercises, including sphere curves,
contact, Bertrand curves, evolutes, and involutes.
In Chapter 3 we give our first sampling of global theorems. Here we
stay in the plane and cover the Rotation Index Theorem of Hopf, the Isoperi-
metric Inequality, the Four-Vertex Theorem, and curves of constant width.
We feel that this is a basic chapter because global differential geometry is a
very important and popular subject which is all too often slighted in a first
course. It is certainly easier to prove only local results. However, this really
cheats the student who thereby misses one of the most fascinating parts of
the subject and is robbed of the insight gained by examining results in the
plane.
Chapter 4 presents the basics of local surface theory and serves as the
motivation for the ideas of Chapter 7. We study surfaces in R 3 and their first
and second fundamental forms. Geodesics and their length-minimizing prop-
erties are discussed. Next comes an investigation of curvature (both Gaussian
and mean), its relationship to the curvature of curves on the surface, and
Gauss's TheoremaEgregium. The chapter ends with optional material on
isometrics, the Fundamental Theorem of Surfaces, and surfaces of constant
Cllrvature. Again, several classical topics are covered in the exercises, includ-
ing ruled surfaces, developable surfaces, and asymptotic curves.
In Chapter S we return to global notions, this time for space curves.
Using the concepts of geodesics on the sphere and integration on a surface,
we prove Fenchel's Theorem about the total curvature of closed space curves
and the Fary-Milnor Theorem about the total curvature of a knot. We also
include the nonstandard topic of total torsion of a closed space curve.
Chapter 6 gives various global results for surfaces. We start off with
Meusnier's Theorem which states that a compact surface, all of whose points
are umbilics, is a sphere. We then go on to the Gauss-Bonnet Formula, which
relates the curvature of a region with the curvature of its boundary, and the
Gauss-Bonnet Theorem, which gives the total curvature of a compact surface.
As applications of the Gauss-Bonnet Theorem we prove theorems due to
Jacobi, Hadamard, and Poincare.
Motivated by the first six chapters, we introduce in Chapter 7 the basic
xiv Preface

definitions of manifold theory and Riemannian geometry. This chapter is

written in an open-ended fashion, referring the reader to other books for
details.
A brief historical summary and bibliography are given in the appendix.
References to the bibliography are in the form of a name foilowed by a date
in brackets, such as Chern [1967).
The first four chapters require only a knowledge of calculus and finite-
dimensional vector spaces (mainly bases and linear independence). Chapter 5
has no additional prerequisites, but the material is more difficult. For Chapter
6 it would be helpful if the reader is familiar with the topology of R 3 • In par-
ticular, the reader should be familiar with the notion of the limit of a sequence
of points in R 3 • We define compact as closed and bounded. In Chapter 7 it
would be useful if the reader knew some metric space terminology, such as
open sets and continuity. A knowledge of the inverse and implicit function
theorems would also make the going a bit easier.
Several different courses can be made from this book. A one quarter
undergraduate course could be based on Chapters 1, 2, and 3 or 4; a one
semester course on 1, 2, 3, 4 (or more, depending on the pace of the course);
a two quarter course could cover the first six chapters; the entire book could
be used for a year course if the last chapter is supplemented. A graduate
course could start with Chapters 2, 4, and 7 and then go on to a book like
Hicks [1965), Boothby [1975), Matsushima [1972), or Warner [1971).
The foilowing table shows the dependence of the various chapters.

We would like to thank Sharon Champion for her magnificent job typing
the manuscript, from the first set of class notes to the final draft. Her patience
and attention to detail through the jth rewrite of the ith author (i = 1 or 2,
1 ~j ~ oo) were truly amazing. We thank Marjorie Parker for her help with
the proofreading. Finally, we thank Barbara Blum of the Prentice-Hall edi-
torial staff for her help in making this book a reality.

Carbondale,Illinois RICHARD S. MILLMAN

GEORGE D. PARKER
 ELEMENTS
OF
DIFFERENTIAL
GEOMETRY
 1

Preliminaries

Differential geometry has two primary tools-linear algebra and cal-

culus. In this text we assume that the reader is familiar with both. We shall
recall the basic facts of linear algebra and vector calculus and urge the reader
to review with care anything which is unfamiliar.

1-1. VECTOR SPACES

DEFINmoN. A real vectorspace is a set V, whose elements are called vectors,

together with two binary operations + : V x V-+ V and , : R x V-+ V,
called additionand scalarmultiplication,which satisfy the following eight
axioms for all u, v, w E V and r, s E R:
(a) u + v = v + u;
(b) u + (v + w) = (u + v) + w;
(c) there is an element O in V such that O + u = u;
(d) (rs),u = r,(s,u);
(e) (r + s),u = r,u + s,u;
(f) r,(u + v) = r,u + r,v;
(g) O,u = O; and
(h) 1,u = u.

We shall normally omit the multiplication symbol ,

1
2 Preliminaries Chap.1

EXAMPLE 1.1. Ordinary three-dimensional space, R 3 , is a real vector space.

EXAMPLE 1.2. The set of all polynomials with real coefficients, R[x], is a real
vector space.

DBFINmON. A set {v1 Ii e /} c Vis linearly independentif whenever a finite

linear combination I; a'v, is zero, then each a' must also be zero. If it is
possible to find a finite linear combination I; a'v, = 0 with some ak =t= 0,
then the set is linearly dependent.

DEFINITION. A subset Sc V spans V if for each vector v e V there are

vectors Vi, v2 , ••• , v, in S and real numbers a 1, a2, ... , a' such that
v = I; a'v,. (The number of elements used ("r") may depend on v.)

DEFINITION. A basis of a vector space Vis a linearly independent spanning

set.

THEOREM 1.3. If Vis a vector space, then V has a basis. Any two bases have
the same number of elements, or all have infinitely many elements. This
number is called the dimensionof V.

ExAMPLB 1.4. R 3 has dimension 3.

EXAMPLE 1.5. R[x] has infinite dimension. {l, x, x 2 , ••• , x", ... } is a basis.

If {v,Ii e /} is a basis of V, then every vector v e V can be uniquely

written as a finite sum v = I; a'v,. The numbers a' are called the components
of v with respect to the given basis.

DEFINITION. An inner product on a vector space Vis a function ( , ):

V x V - R such that for all u, v, w e V and r, s e R:
(a) (u, v) = (v, u);
(b) (u, rv + sw) = r(u, v) + s(u, w); and
(c) (u, u) ~ 0 with equality if and only if u = 0.

ExAMPLE 1.6. In R 3 we may use the ordinary dot product:

((a1' az' a3), (bl' b2' b3)) = a1b1+ a2b2+ a3b3.
EXAMPLE 1.7. In R[x] we may set (p(x), q(x)) = f ~ p(x)q(x) dx. See Problem
1

1.4.

DEFINmON. If V has an inner product and v e V, the length of v is

Jvl= ,Jlf,v5.
s,c.,_, Vector Spaces 3

LEMMA1.8 (Cauchy-Schwarz Inequality). If u, v E V, then I(u, v) I ~ Iu 11v I·

Furthermore, I(u, v) I = Iu 11v I if and only if u and v are linearly depen-
dent.

Proof: Problem 1.5. I

This lemma tells us that -1 ~ (u, v)/1u 11v I ~ I (unless Iu 11v I = 0, in
which case (u, v) = 0 also). Hence an angle 8 may be defined by the formula
~
cos(J = TiilTvT
.
(J, which is defined only up to sign, will be called the angle between u and v.
If the ordinary dot product is used in R 2 , then this concept of an angle coin-
cides with the usual notion of an angle in the plane. Note that
(u, v) = Iu 11v I cos e
holds if one of the vectors is zero, even though 8 is not defined then.

DBPINITION. u is orthogonal(or perpendicular)to v if (u, v) = 0.

ExAMPLB 1.9. In R 3 , (I, 2, 3) is perpendicular to (4, -5, 2).

ExAMPLB 1.10. In Example 1.7, xis perpendicular to (x 2 + 1).
1'11BoRBM1.11. If V has dimension n and an inner product, then there exists
a basis {v1, v2 , ••• , v.} such that Iv,!= I and v, is perpendicular to v,
whenever i :;t: j. Such a basis is called orthonormal.

This is proved by showing how to create an orthonormal basis from any

given basis by a process called the Gram-Schmidt orthogonalization. See a
linear algebra text such as Hoffman and Kunze [1971)for details.
Suppose V has a basis {u1, u2 , ••• , u.} and an inner product ( , ).
The inner product may be associated with a matrix in the following manner.
Let g,1 = (u,, u1). (g,1) is a positive definite symmetric matrix which repre-
sents ( , ) with respect to the given basis. Ifu = I; du, and v = I; b'u,,
then (u, v) = .l: a'b'g,1•
Note that if {u1, u2 , ••• , u.} is orthonormal, then (g,1) = (611) is the
identity matrix. The Kronecker symbol 6,1 (and its variations 6/, 611, 6'') is
defined by

~
if i =j
(1-1) 6,, ={ if i :;t:j.
If {vi,v2 , ••• , v.}is another basis of V, we have u1 = I; a•,v•. The matrix
representing ( , ) with respect to the basis {vi,v2 , ••• , v.} is given by
i., = (v., v,). Then g,1 = (u,, u1) = I; a•,a',i., or
(1-2) (g,1) = (a•,)'(i.,,)(a',).
4 Preliminaries Chap.1

This equation shows the effect of a change of basis on the matrix which
represents the inner product.

PROBLEMS

1.1. Prove that the set of polynomials of degree less than or equal to n forms
a real vector space. What is the dimension? Do the polynomials whose
degree is exactly n form a vector space?
1.2. Prove that D 5 = {/: R--+ RI/is differentiable at x = 5} is a real vector
space. What is the dimension of D 5 ?
1.3. Represent the standard inner product on R 3 (Example 1.6) with respect
to the basis {(l, 0, 1), (0, 1, 1), (1, -2, 3)}.
1.4. Prove that < , ) as defined in Example 1.7 is an inner product.
*1.5. Prove Lemma 1.8. (Hint: Let g(t) = (u + tv, u + tv) ~ 0. How many
real roots does g have? What does the quadratic formula say?)

1-2. LINEAR TRANSFORMATIONS AND

EIGENVECTORS

DEFINITION.A linear transformationis a function T : V--+ W of vector spaces

such that T(av + bw) = aT(v) + bT(w) for all a, b E R and v, w E V.
An isomorphismis a one-to-one onto linear transformation.

If T : V--> W is a linear transformation, we may associate a matrix

with it. Let {vi, v2 , ••• , vn}be a basis of Vand let {w1, w2 , ••• , wm}be a basis
of W. Then there are mn real numbers T' 1 such that T(v1) = I; T' 1 w,.We say
that (T' 1) represents T with respect to the given bases. (If V = W, it is cus-
tomary to use the same basis for Vand W.) If v = I; alv 1 E V, then
T(v) = I; (1; T' 1 al)w,.
Thus if we view v as a column vector, the i in T 11 is the row index and thej
is the column index.
Suppose T : V--+ V is a linear transformation and V has two bases
{u1 , u2 , ••• , un}and {vi, v2 , ••• , vn}related by u, = I; a\v,.. If(T' 1) represents
T with respect to the u, and (1'",) represents T with respect to the v,., WC: have
I; T 11 a",v,. = I; T' 1 u, = T(u1) = T(I; aP1 v1) = I; aP1 T(v,) = I; aft1 T" 1 v,..
Hence I; T' 1 a"1 = I; aP1 T",, or (a"1)(T' 1) = (T",)(aP1), or
(2-1) (T' 1) = (a",)-1(T",)(aP1).
This equation shows the effect of a change of basis on the matrix which
represents a linear transformation.
Sec. 1-2 Linear Transformations and Eigenvectors 6

DEFINITION. Let T: V--+ V be a linear transformation. A real number .Ais

an eigenvalueof T if there is a nonzero vector v such that T(v) = .Av.v
is called an eigenvectorof T corresponding to .A.

If (T 11) represents T, the eigenvalues of T are the real solutions of the

polynomial equation det (T 11 - x~ 1,) = 0. Then if the dimension of Vis n,
there are at most n eigenvalues (counting multiplicity). There may be fewer,
since some of the solutions might be complex but not real. Once the eigen-
values are known, the eigenvectors are found by solving appropriate linear
equations.

EXAMPLE 2.1. Let T: R 2 R 2 be represented with respect to the standard

--+

basis {(1, 0) 1, (0, 1)1} by the matrix ( 3 4).

4 -3

-3-l
4 I= (3 - l)(-3 - l) _ 16

= -9 + A. 2 - 16
= l2 - 25
= (.A- 5)(.A+ 5).

(! -~)(;)= 5(;) has

The eigenvalues are therefore 5 and -5.

( ~) as one solution. (! _;)

(;)= -5(;)has ( _~) as one solu-
tion. Hence ( ~ ) is an eigenvector corresponding to 5 and ( _ ~) is one
corresponding to -5.

PROBLEMS

2.1. In Example 2.1 we have T represented by (! -34)with. respect to the

basis {(1, 0) 1, (0, I)'} of R2.
(a) Represent T with respect to the basis {( I, I ) 1, ( I, -1) 1}.
(b) Represent T with respect to the basis {(2,I ) 1, ( 1,-2) 1 } of eigenvectors.
2.2. Find the eigenvalues and eigenvectors of T: R 2 ----->R 2 given by the matrix
(-5
-6
3).
4
6 Preliminaries Chap. 1

1-3. ORIENTATION AND CROSS PRODUCTS

Let {u1 , u2 , ••• , Un}and {v1 , v2 , ••• , vn}be two ordered bases of V and
define a matrix (a) by v1 = I: a11 u1•

DEFINITION.The ordered bases {u1 , u2 , • •• , un} and {v1 , v2 , ••• , vn}give the
same orientation to V if det (a11) > 0. They give opposite orientations if
det (a11) < 0.

EXAMPLE
3.1. Let {(l, 0, 0), (0, 1, 0), (0, 0, l)} and {(l, 1, 0), (1, 0, -1), (2, 1, 3)}

be two ordered bases of R 3 • Since (a 1) = (11 01 2)

1 has determi-
0 -1 3
nant -4, these ordered bases give opposite orientations.

EXAMPLE3.2. {(l, 0, 0), (0, 1, 0), (0, 0, l)} and {(l, ], O), (2, 1, 3), (1, 0, -1)}
give the same orientation.

The basis {(l, 0, O), (0, 1, 0), (0, 0, 1)} of R 3 will be denoted {ei, e 2 , e3 }.
Its orientation will be called right handed.

DEFINITION.If u = I: a' e, and v = I: b1e1 are vectors in R 3 , the cross (or

vector) product of u and v is
u x v = (a 2 b 3 - a 3b2 )e 1 + (a b
3 1 - a 1b 3)e2 + (a b
1 2 - a 2 b 1)e 3 •
By abuse of notation this may be written as
e1
u xv= (
det a 1
bl b2

LEMMA3.3. Let u, v, w E R 3 and r E R. Then

(a) u x v = -v x u;
(b) (ru) x v = r(u x v);
(c) u x v = 0 if and only if u and v are dependent;
+
(d) (u v) x w = (u x w) (v x w); +
(e) u x vis perpendicular to both u and v under the usual dot product
ofR 3 (Example 1.6);
(f) lu x vi= lul lvl sin 0, where Ois the angle between u and v;
(g) {u, v, u x v} gives a right handed orientation to R 3 if {u, v} is linearly
independent.

Proof: Problem 3.1. I

sec.t-3 Orientationand CrossProducts 7

LEMMA3.4. Let u, v, w E R 3 • Then ((u xv), w) = (u, (v x w)).

Proof: Problem 3.2. I

DEFINITION. The mixed (or triple) scalarproduct of u, v, w is
[u, v, w] = ((u x v), w).

It is important to remember that [u, v, w] is always a number and not a

vector.

LEMMA3.5. I[u, v, w]I is the volume of the parallelopiped spanned by u, v, w.

(See Figure 1.1.)

FIGURE 1.1

Proof: Problem 3.3. I

PROBLEMS

*3.1. Prove Lemma 3.3.

*3.2. Prove Lemma 3.4.
3.3. Prove Lemma 3.5.
3.4. Prove a x (b x c) = (a, c)b - (a, b)c.
*3.5. Let {a, b, c} be an orthonormal basis of R 3 with a x b = c. Prove
b xc= a and c x a = b.
3.6. Prove (a x b) x (c x d) = [a, b, d]c - [a, b, c]d.
3.7. Let u,, U:a, u 3, Vi, V:a, v3 be six vectors in R 3 • Let a11 = (u,, v,). Prove
det (a,1) = [u1, u2 , u 3][v1, v2 , v 3].
8 Preliminaries Chap.1

1-4. LINES, PLANES, AND SPHERES

We shall recall the vector equations of lines, planes, and spheres in R 3 •

See a standard calculus text such as Thomas [1968) for more details. Geo-
metrically, a straight line is determined by a point on the line and a vector
parallel to it. A plane is determined by a point on the plane and a vector
perpendicular to the plane. A sphere is determined by its center and its
radius.

DEFINITION. The line through x 0 e R 3 and parallel to v =;cO has equation

cz(t) = x 0+ tv. (See Figure 1.2.)

FIGURE 1.2

If we write x 0 = (x 0 , Yo, z0 ), v = (v1 , v2 , v 3) and cz(t) = (x(t), y(t), z(t)),

then the definition gives
x(t) - Xo = IV 1, y(t) - Yo = W2 , z(t) - z0 = W3 •
Assuming that v' =;cOfor i = 1, 2, 3, these equations yield the classical defini-
tion of a straight line after solving each equality for t and setting these
quantities equal:
x-x y-y z-z
~ = ,L_______LJJ.= ~.
v• ~ vJ
where we have suppressed the t from the notation as is common classically.
If x 1 and x 2 are distinct points in R 3 both of which lie on a line /, then
the vector x 2 - x 1 is parallel to I. This observation proves:

LEMMA4.1. The line through x 1 and x 2 in R 3 has equation

cz(t) = x 1 + t(x 2 - x 1).
(See Figure 1.3.)
SBC, 1-4 lines, Plenes, and Spheres 9

FIGURE 1.3

DEFINITION. The plane through x 0 perpendicular to n -=I=0 has equation

<x - x 0 ,n) = 0.
(See Figure 1.4.)

FIGURE 1.4

The following lemma is clear from Figure 1.5 since u x vis perpendicular
to the desired plane.

FIGURE 1.5 v
10 Preliminaries Chap.1

LEMMA 4.2. If {u,v} is linearly independent, the plane through x 0 parallel to

both u and v has equation (x - x 0 , u x v) = 0.

The sphere in R 3 with center m and radius r > 0 has equation

DEFINITION.
(x-m,x-m)=r 2•

If m = (a, b, c) and x = (x, y, z), then this definition is the familiar

formula
(x - a) 2 + (y - b) 2 + (z - c) 2 = r 2 •

PROBLEMS

4.1. Let a = (2, I, -3), b = (1, 0, 1), c = (0, -1, 3). What is the equation of
(a) the line through a parallel to b;
(b) the line through band c;
(c) the plane through b perpendicular to a;
(d) the plane through c parallel to a and b;
(e) the sphere with center a and radius 2?

1-5. VECTOR CALCULUS

If abstract vector spaces are unfamiliar, the reader may assume that
V = R 3 below. Letf: R---->V, where Vis a real vector space. If{v 1 , v2 , ••• , v,.}
is a basis of V, then f(t) = I: !'(t)v,. If the component functions f'(t) are
differentiable or integrable, we may differentiate or integrate f component-
wise:
df df'
-dt = I;-v,dt
and
s: f(t) dt = I: u: f'(t) dt) v,.
We should check that these definitions do not depend on the choice of
basis for V. However, this is a simple consequence of the linear properties
of d( )/dt and f! ( ) dt and is left to the reader.
Similarly if f is a vector-valued function of several variables, we may
take partial derivatives or multiple integrals.

LEMMA5.1. Let f, g : R ---->V and suppose that V has an inner product

( , ). Then
Jr(f, g) = <1~
·g) + <f, 17)
·
In particular, if If I is constant then df/dt is perpendicular to f.
Sec. 1-5 Vector Calculus 11

Proof: Problem 5.1. I

5.2. Let f, g : R --+ R 3 • Then
LEMMA
d df dg
di<' x g) = dt x g + f x at.

Proof: Problem 5.2. I

DEFINITION. f: R-+ R is of class Ck if all derivatives up through order k
exist and are continuous. f: Rn--+ R is of class Ck if all its (mixed)
partial derivatives of order k or less exist and are continuous. A vector-
valued function is of class Ck if all its components with respect to a given
basis are of class Ck.
Note that if f is of class Ck it is also of class Ck·•.
Rather than continually worrying about what class a differentiable func-
tion belongs to, we shall usually assume it is of class C3. We shall point out
those cases where higher class is needed or lower class is sufficient.
Finally, before we start our study of curves, we want to remind you what
form the chain rule for differentiation takes. Suppose that x is a function of
several variables u 1 , u2 , • • • , un and that the u1 are functions of variables
t11, '12 , ••• , " ... Then
h naxau 1
(S-1) f-1
cftji = TuiTv-' ce= 1, 2, ... , m.
Note that we are writing the coefficients on the right of the vectors
instead of the left as would be usual in linear algebra. This is done so that
Equations (5-1) look more like the chain rule from calculus.
Special cases that we shall often use arise when n = m = 2 as in Equa-
tion (5-2), or n = 2 and m = 1, as in Equation (5-3).

(S-2) :. = ~ ~ + ~ ~· ce= 1, 2
<s-3> dx h du1 axdu 2
dt = T,ii dt + 1ii2dt.

PROBLEMS

•s.1.Prove Lemma 5.1.

*5.2. Prove Lemma 5.2.

Set A-Hyperbolic Functions

For some of the examples and problems in this book it will be necessary
to use certain transcendental functions which behave in many ways like the
12 Preliminaries Chap.1

trigonometric functions. Since they are not always covered in a calculus course
we briefly cover them in this problem set. The hyperbolicsinefunction is the
function sinh: R ---+R given by sinh(t) = (e' - e-')/2. The hyperboliccosine
function is the function cosh: R---+ R given by cosh(t) = (e' + e-')/2. For aid
in reading, sinh is pronounced "cinch" while "cosh" rhymes with "gosh."

5.3. Prove that for any t e R, cosh 2 (t) - sinh 2 (t) = 1. (Since x 2 - y 2 = 1
is a hyperbola, this gives the origin of the term "hyperbolic function."
Also, you should think about the analogy between this equation and
cos 2 8 + sin 2 8 = 1.)
5.4. Prove that cosh'(t) = sinh(t), and sinh'(t) = cosh(t) for all t e R.
5.5. Using Problem 5.4 show that sinh is one-to-one. Show that cosh is one-
to-one when restricted to [O, oo). Show also that sinh is onto the reals
whereas cosh(R) = [I, oo). (This means we may define the appropriate
inverse function.)
5.6. Show that the general solution to the differential equation/"= a 2f is
given by /(t) = A cosh(at) +
B sinh(at) where A and Bare real con-
stants. (If you cannot show this is the general solution, at least show
that it satisfies the differential equation.)
5.7. Compute f dx/,./1 + x 2 by means of the substitution x = sinh(t).
 2

Local Curve Theory

We shall begin our study of differential geometry with an investigation

of curves in three-dimensional Euclidean space, R 3 • This is a good place to
start for four reasons: (1) since curves in R 3 are easy to draw and visualize,
we can develop some geometric insight into the subject by looking at exam-
ples; (2) it is a very complete subject-the Frenet-Serret apparatus of a curve
completely determines the local geometry of curves and gives a complete set
of invariants for the problem of determining whether two curves are the
same; (3) the theory of curves will introduce us to some techniques that are
the mainstay of modem differential geometry (e.g., linear algebra); (4) we
shall base our study of surfaces in Chapter 4 on the behavior of curves on the
surfaces.
The history of the theory of curves (and of all differential geometry) is a
fascinating one. Suffice it to say at this point that the many results in the
theory of curves in R 3 , which we discuss in this chapter, were initiated by
G. Monge (1746-1818) and his school (Meusnier, Lancret, and Dupin). Our
approach is due to G. Darboux (1842-1917) whose idea of moving frames
unified a great deal of the classical theory of curves. He accomplished this
in 1887-1896. It is interesting to note that the approach to the theory of sur-
faces that we will take in Chapter 4 is that ofK. Gauss in 1827. This anomaly
of dates is not due to the fact that the curve theory is more difficult than the
13
14 Local Curve Theory Chap.2

theory of surfaces (just the opposite is true) but rather to the pervasive genius
of Gauss. The history of the theory of curves is discussed more completely in
the historical notes at the end of this book.
There are two ways to think of curves. The first is as a geometric set of
points, or locus. When this is the case, we refer to a geometric curve, or the
geometricshapeof the curve. Intuitively we are thinking of a curve as the path
traced out by a particle moving in R 3 • The second way of thinking of a curve
is as a function of some parameter, say t. Intuitively it is not always enough
to know where a particle went-we also want to know when it got there.
(The parameter t is often thought of as time.) It is necessary to view curves the
second way if we are to apply the techniques of calculus to describe the
geometric behavior of a curve. This means that we must pay careful attention
to how the curve is parametrized (e.g., if you change the parameter you also
change the velocity vector field to the curve). However, we are also interested
in geometric properties of curves (e.g., arc length, tangent vector field). These
should not depend on the way a curve is parametrized as a function but only
on the image set of the function, that is, only on the geometric shape. Thus
we shall ask whether our constructions and descriptions depend upon the
parametrizations.
In Section 2-1 we define and give examples of parametrized curves. In
Section 2-2 we introduce a particularly useful parametrization-that by arc
length. Section 2-3 develops the Frenet-Serret apparatus which is the basic
tool in the study of curves. It consists of three vector fields along the given
curve (the tangent T, the normal N, and the binormal B) and two scalar-
valued functions (the curvature 'IC and the torsion -r). The Frenet-Serret
Theorem is proved in the fourth section. This theorem expresses the deriva-
tives of T, N, and B in terms of T, N, and B. We then make several applica-
tions. There is also a long collection of problems at the end of this section,
many of which deal with topics from the classical differential geometry of
curves. Section 2-5 gives the Fundamental Theorem of Curves and shows that
the Frenet-Serret apparatus does completely determine the geometry of the
curve. Finally, in Section 2-6 we develop the necessary techniques for comput-
ing the Frenet-Serret apparatus for curves which are not parametrized by arc
length.

2-1. BASIC DEFINITIONS AND EXAMPLES

Our study of curves will be restricted to a certain class of curves in R 3 •

Not only do we want a curve to be described by a differentiable function so
that we may use calculus to describe the geometry, we also want to avoid
certain pathologies and technicalities. If da,/dt = 0 on an interval, then
a,(t) is constant in that interval, which is geometrically very uninteresting. If
sec.2-1 Basic Definitions and Examples 1&

dtl,/dtis Oat some point, then the graph of II can have a sharp corner, which is
geometrically unappealing. (Consider the graph of ci(t) = (t 2 , t 3 , 0).) Be-
cause of these considerations we will only work with regular curves.

DEFINITION. A regular curve in R 3 is a function Cl: (a, b)--+ R 3 which is of

class Ck for some k > I and for which dci/dt :¢:.0 for all t e (a, b).

In this text, a regular curve will be assumed to be of class C 3 unless stated

otherwise.
Note that from this point of view the curve is the function and not the
image set (geometric curve). Two different curves may have the same image
set (see Examples 1.2 and 1.4 below). A regular curve need not be one-to-one,
but as Problem 1.9 shows, it cannot intersect itself too often.
Given a regular curve ci(t), we can define some vector fields along 11.
This means that for each t we will have a 3-vector v(t). The reader should
think of the tail of v(t) to be at the point ci(t). The mapping t--+ v(t) is a
vector-valued function and so we may use the material of Section 1-5.

DEFINITION. The velocity vector of a regular curve ci(t) at t = t 0 is the deriva-

tive dci/dt evaluated at t = t 0 • The velocity vector field is the vector-
valued function dci/dt. The speed of ci(t) at t = t 0 is the length of the
velocity vector at t = t 0 , I(dci/dt)(t0 ) I·

If we view the curve as the path of a moving particle, the velocity vector
at t = t 0 points in the direction that the particle is moving at time t 0 • (See
Figure 2.1.) The regularity condition says that the speed is always nonzero-
the particle never stops moving, even instantaneously.

DEFINITION. The tangent vector field ta a regular curve Cl(t) is the vector-
valued function T(t) = (dci/dt)/1dci/dt I·

FIGURE 2.1
18 Local Curve Theory Chap.2

Note that we are able to define T (i.e., divide by Idtt,/dt I) precisely be-
cause of the regularity condition. T is the unit vector in the direction of the
velocity vector.
We shall see later in this section that Tis a geometric quantity: it depends
only on the image set of t1, and not the particular way this set is parametrized.
For each value oft, say t 0 , there is a unique straight line through tt,(t0)
parallel to T(t 0 ). This line is a linear approximation of the curve near tt,(t0 ).
(This is an example of one of the basic techniques in differential geometry:
an object of study (a curve) is replaced by a linear approximation (a tangent
line). This is done because linear mathematics is so much better understood
than nonlinear mathematics.) More formally:
DEFINITION. The tangent line to a regular curve t1, at the point t = t O is the
straight line
I= {w E R3 lw = tt,(t0) + lT(t 0 ), l E R}.
Note that the tangent line is a subset of R 3 which contains the point
tt,(t0 ) and actually is a straight line. Intuitively it is the line that most nearly
approximates the curve near tt,(t0 ). (See Figure 2.2.)

FIGURE 2.2

Since dtt,/dt :¢: 0 and dtt,/dt = Idtt,/dt IT, the tangent line at t = t O is also
given by

EXAMPLE 1.1. Let u and v be fixed vectors in R 3 • Then the curve t1, : R --+ R 3
given by tt,(t) = u + tv is a regular curve if and only if v :¢: 0. In this
case it is a straight line and dtt,/dt = v. The tangent line at each point is
the given straight line and T = vflvl,
ExAMPLE 1.2. Let ti,: R --+ R 3 be given by tt,(t) = (t, o.0). This is a special
case of Example 1.1.
S.c. 2-1 Basic Definitions and Examples 17

:ExAMPLE 1.3. Let iz: R - R 3 be given by iz(t) = (t 3 , 0, 0). diz/dt = (3t 2 , 0, 0),
which is zero at t = 0. iz is not a regular curve, even though its image is
the same as the curve in Example 1.2 !

ExAMPLE1.4. Let IZ: R - R 3 be given by IZ(t) = (t 3 + t, 0, 0).

~~ = (3t 2 + 1, 0, 0) =t=0.
This is a regular curve whose image set is the same as the curve in
Example 1.2.

ExAMPLE1.5. Let g: R- R be a differentiable function. Let iz(t) = (t, g(t), 0).

Then diz/dt = (1, g'(t), 0) =t= 0 and iz is a regular curve. iz(t) is the
graph of g except for the extra (third) coordinate. The tangent line at
t = t 0 is {(t0 + l, g(t 0 ) + lg'(t 0 ), 0) 11E R}. In terms oft, y, z coordi-
nates this line is z = 0, y - g(t 0 ) = (t - t 0 )g'(t 0 ), which should be
familiar from Calculus I as the equation of the tangent line to the graph
of g(t) at t = t 0 • (Remember t = x in this example.)

ExAMPLE1.6. Let iz: R - R 3 be given by IZ(t) = (r cos t, r sin t, ht), where

h > 0 and r > 0. This is called a right circular helix. (If h < 0 it would be
a left circular helix.) Circular refers to the fact that the projection in the
(x, y) plane is a circle. Since diz/dt = (-r sin t, r cos t, h) =t= 0, iz is a
regular curve. At t = t 0 , T = (-r sin t 0 , r cos t 0 , h)/,./r 2 + h 2 • (See
Figure 2.3.)

FIGURE 2.3

In Examples 1.2 and 1.3 we saw a situation where the same image set
(the x-axis) was given two different parametrizations, one of which was not
regular. We wish to know what parametrizations can be used to describe a
liven image curve.

DEFINITION. A reparametrization of a curve iz: (a, b)- R 3 is a one-to-one

onto function g: (c, d)---+ (a, b) such that both g and its inverse
h: (a, b)--+ (c, d) are of class Ck for some k > 1.
18 Local Curve Theory Chap.2

What we have in mind is the new curve P= ~ o g. If r denotes the vari-

able in the interval (c, d), then dp/dr = (d~/dt)(dg/dr) by the chain rule. p is
thus regular if~ is regular and dg/dr * 0. But g(h(t)) = t so that by the chain
rule (dg/dr)(dh/dt) = 1 and dg/dr *0. Thus the composition of a regular
curve with a reparametrization yields a regular curve. Note that if ~ is of
class cmand g is of class Ck, then pis of class enwith n = min (k, m). (See
Figure 2.4.)

~
c -----
'o
t~ d a
\ b
FIGURE 2.4

The image of a curve and any of its reparametrizations are the same.
This means that any quantity which stays the same when we change para-
meters (i.e., make a reparametrization) is a quantity which depends only on
the geometric shape of the curve. Briefly, the quantity is a geometric invariant.
We will show below that the tangent line to a regular curve is such a geometric
invariant after we give some examples.

EXAMPLE 1.7. Let g: (0, 1)----+(1, 2) be given by g(r) = 1 + r 2 • g is one-to-

one and onto with inverse h(t) = ,./t"=T. g is infinitely differentiable
on (0, 1) and so is h on (1, 2). Thus g is a reparametrization of any regular
curve on (I, 2).

EXAMPLE 1.8. Let g: R--+ R be given by g(r) = r 3 • g is one-to-one, onto, and

infinitely differentiable. However, h(t) = t 113 is the inverse and h'(O) does
not exist, so that h is not C 1 • This is one reason why Example 1.3 was
not a regular curve.

Now we shall show that the tangent vector field is a geometric property
of the image set of a regular curve and does not depend on the parametriza-
tion. This means that the tangent line to a curve is a geometric property also.

PRoPOSITION 1.9. Let ~: (a, b)---+ R 3 be a regular curve and let

g: (c, d)----+(a, b)
be a reparametrization. Set p = ~ o g. If t 0 = g(r 0 ), the tangent vector
field T of~ at t O and the tangent vector field S of Pat r O satisfy S = ± T.
$11(:.2-1 Basic Definitions and Examples 19

p,oof·
~ dt1,dg
dr dt dr
s = ldJT= ldt1,lldgl
17rl dt dr
dt1, dg

= 1!11~1
dt dr
= ('1)(±1) = ±T. I

Note that S = T if dg/dr > 0 (g is increasing) and S = -T if dg/dr < 0

(g is decreasing). Geometrically the difference is whether t1,and ISindicate
particles moving along the image curve in the same or opposite directions.

PROBLEMS

1.1. (a) Show that t1,(t)= (sin 3t cos t, sin 3t sin t, 0) is a regular curve.
(b) Find the equation of the tangent line to t1,at t = n/3.
1.2. (a) Which of the following are regular curves 'l
(i) tJ,(9) = (cos 9, 1 - cos 9 - sin 9, -sin 9).
(ii) IS(9)= (2 sin 2 9, 2 sin 2 9 tan 9, 0).
(iii) y(9) = (cos 9, cos 2 9, sin 9).
(b) Find the tangent line to each of the above curves at 9 = n/4.
*1.3. (a) In Example 1.6, what is the equation of the tangent line at t = t 0 'l
(b) Show that the angle between (0, 0, 1) = u and dt1,/dtin Example 1.6
is a constant (i.e., independent of t).
1.4. Show that /: (-1, 1) - (-oo, oo) given by /(t) = tan (nt/2) is a
reparametrization.
1.S. Let g: (0, oo)--+ (0, 1) be given by g(r) = r 2 /(r 2 + 1). Is this a repara-
metrization 'l
1.6. Let t1,(9)= (e' cos 9, e' sin 9, 0). Prove that the angle between t1,and
T is constant. (A curve with this property is called a logarithmicspiral.)
1.7. Let t1,(t)be a regular curve. Suppose there is a point a e R 3 such that
tJ,(t) - a is orthogonal to T(t) for all t. Prove that t1,(t)lies on a sphere.
(Hint: What should be the center of the sphere?)
*1.8. Consider the function t1,: R --+ R 3 by t1,(t)= (t 2 , t 3 , 0).
(a) Show that t1,is C 1 but not regular.
(b) Show that the image of t1,has a sharp comer by graphing t1,,
*1.9. Let t1,: (a, b)--+ R 3 be a differentiable curve. Suppose there is a
sequence of points {t,,}in the interval (a, b) such that the t,, are all
20 Local C11rveTheory Chap.2

distinct, lim tn = t* E (a, b), and fl(tn) = X 0 for all n (thus "fl intersects
n-oo
itself infinitely often at x 0 "). Show that fl is not regular. (Hint: Show
that dt1/dt' = 0 at t* .) This shows that on any finite closed interval, a
regular curve can intersect itself only a finite number of times at the
same point.

2-2. ARC LENGTH

Sometimes it is useful for technical reasons to consider curves with end

points, that is, curves defined on closed intervals :

DEFINITION. A regular curve segment is a function fl: [a, b] - R 3 together

with an open interval (c, d), with c < a < b < d, and a regular curve
y: (c, d) - R 3 such that t1(t) = y(t) for all t e [a, b].

Thus a curve segment is a curve defined on a closed interval which can be

extended to a curve on a slightly larger open interval. In this case it is possible
to talk about dt1/dt at the end points of the curve segment because we define
(dt1/dt)(a) to be (dy/dt)(a) and (dt1/(it)(b) to be (dy/dt)(b).
Now we shall define the length of a curve segment. The intuitive justifica-
tion for what we do is as follows. If t1(t) is viewed as the path of a particle
moving in space, then Idt1/dt Iis the speed of the particle as a function of time.
The integral of speed should be the distance traveled by the particle just as
it is in one dimension.

DEFINITION. The length of a regular curve segment fl: [a, b] - R 3 is

Note that this is really the familiar formula for the length of a curve in
= (x(t), y(t), z(t)), then Idt1/dt I = ,J(x') 2 + (y') 2 + (z') 2, so that
R 3 : if fl(t)
J!
the length of the curve is given by ,J(x') 2 + (y') 2 + (z') 2 dt.
It makes sense to talk about reparametrizations of curve segments (see
Problem 2.7). One would hope that the length of a curve is a geometric prop-
erty and does not depend on the choice of parametrization. This is the con-
tent of the next proposition.

PROPOSITION 2.1. Let g: [c, d] - [a, b] be a reparametrization of a curve

segment fl: [a, b] - R 3 • Then the length of fl is equal to the length of
p= fl, 0 g.
 s.c, 2-2 Arc Length 21

Proof: The length of II is

[11,ldr = f l(~i)(~~ldr
= f 1~,11~1d,.
Case 1: If dg/dr > 0, then ldg/drl = dg/dr, g(c) = a, g(d) = b, and
J:jdtt,/dtl ldg/drl dr = J!ldtt,/dtl (dg/dr) dr = J!ldtt,/dtl dt by the substitution
rule of integral calculus.
Case2: If dg/dr < 0, the proof is similar, using ldg/drl = -dg/dr,
· g(c) = b, and g(d) = a.
In both cases we have that the length of tt, equals the length of II, I
Note that the definition of the length of a curve does not really require
Cl to be regular to make sense-it is sufficient for tt, to be of class C 1 • How-
ever, if tt, is not regular, some segments of the curve may be traversed twice
and the formula will count the doubled section twice.
Using the concept of length of a curve, we are able to define an impor-
tant way'to reparametri7.e a curve. Let tt,: (a, b)- R 3 be a regular curve and
let t 0 e (a, b). Set h(t) = J~.Idtt,/dtI dt. s = h(t) is called arc length along tt,.
It actually measures signed arc length along tt, from tt,(t0 ) with h(t) < 0 if
t < t 0 and h(t) > 0 if t > t 0 •
TIIBoREM
2.2. his a one-to-one function mapping (a, b) onto some interval
(c, d) and is a reparametrization.

Proof: By the fundamental theorem of calculus, dh/dt = Idtt,/dt I > 0 (since

Clis regular). Thus his increasing and so is one-to-one. It is easy to check that
if Cl is of class Ck so is h. Let g: (c, d) - (a, b) be the inverse of h and
denote the parameter in (c, d) bys. This means of course that g(s) = t if and
· only if h(t) = s so thats is the arc length parameter. Because g and hare
inverse functions dg/ds = l/(dh/dt), where the right-hand side is evaluated at
t = g(s). This quotient makes sense since dh/dt :;c 0. g can be differentiated
as often as h can. Thus h is a reparametrization. I
As was pointed out in the proof of the above theorem, sis the arc length.
By using g(s), any regular curve tt, can be reparametrized in terms of arc
length from a point. Once this has been done we say that the curve has been
/J(lrt1metrizedby arc length. The importance of a curve being parametrized by
arc length is carried in the observation that its velocity vector field is its
tangent vector field, as may be seen in the following computation.
If ll(s) is parametrized by arc length, thens= J~
ldlS/dO'ldO'. By the
fundamental theorem of calculus, 1 = (d/ds)(J~ ldll/dO'I
dO')= ldllfdO'I at
22 Local Curve Theory Chap.2

<1 = s; that is, 1 = IdP/ds j. Hence the velocity vector field dPfds is a unit
vector field and is thus T. When a curve pis parametrized by arc length (or
equivalently if its velocity vector field is T) we say that pis a unit speed curve.
The preceding paragraph and the proof of Theorem 2.2 contain some
important facts which we now isolate. We shall assume that O is in the
domain of cz(t)and base arc length at 0.

COROLLARY 2.3. If cz(t)is a regular curve ands= s(t) is its arc length, then
(a) s = s(t) = J~
Idcz/dtI dt;
(b> ds/dt = Idcz/dtI;
(c) dcz/dt = (ds/dt)T; and
(d) T = dcz/ds.
To take a given regular curve czand reparametrize it by arc length, while
always possible in theory, may be very difficult in practice. There are two
obstacles to such a program. In the first place, the integral

h(t>= I.I!~I dt
may not be elementary (see Example 2.6) and hence not computable.
Secondly, even if h(t) can be determined, it may not be possible to find the
inverse function g(s) (see Example 2.7).

ExAMPLE 2.4. Let cz(t)= u + tv be the straightlineofExamplel.1. dcz/dt = v,

J~
s = h(t) = Iv I dt = t Ivj. Thus t = g(s) = s/1v I, P(s) = u + sv/1v I
gives the unit speed parametrization of a straight line. Note that the
tangent vector field to xis T = v/lvl, and dT/ds = 0.

ExAMPLE 2.5. Let cz(t)= (r cos t, r sin t, 0) with r > 0.

!~= ( -r sin t, r cos t, 0)

and ldcz/dtl = r. s = h(t) = rt and t = g(s) = s/r.

·p(s) = (r cos G),r sin G),o)
is the unit speed parametrization of a circle of radius r. Note that the
tangent vector field of czis

and that
T(s) = (-sin (:),cos e),o)
dT 1
ds = ( -,cos (s)
,: , -,sin1 . (s)
,: ,0 )
has length 1/r.
StH:.2-2 Arc Length 23

BxAMPLE 2.6. If or.(t)is the ellipse (2 sin t, cost, 0), then

~= (2 cos t, -sin t, 0).

I
I~~ = ,J4 cos2 t + sin 2 t
,.= ,J4 - 3 sin 2 t = 2,Jl - (i) sin 2 t.
But ,Jl - J
sin 2 t does not have an elementary antiderivative and so
the integration h(t) = J~
Idtt,/dt I dt cannot be carried out by using the
fundamental theorem of calculus. (Definite integrals of this kind are
called elliptic integrals because they can be interpreted as the arc length
of an ellipse. Their values are tabulated in many books of mathematical
tables.)

2.7. Let tt,(t) = (t, t 2 /2, 0). Then dtt,/dt = (1, t, 0) and
ExAMPLE

l~~I =,JI+ ,2.

Hence

s = h(t) = J;,JT+7 d-r = !(t,JT"""+t2+ In (t + ,v'J+t2)).

However, it is extremely difficult to find t = g(s) from this equation.
Note that tt, is a parabola, a very simple curve geometrically!

What have we accomplished? Suppose that we want to study regular

curves and we are interested only in their geometric shape (that is, we don't
care about parametrization). Theorem 2.2 says that we may as well assume
that the curve is parametrized by arc length. This will be a very useful tech-
nical device in setting up the Frenet-Serret apparatus in the next section.

PROBLEMS

2.1. Find the arc length of the circular helix in Example 1.6 for O < t < 10.
2.2. Find the arc length of tt,(t) = (2 cosh 3t, -2 sinh 3t, 6t) for O < t < 5.
· 2.3. Reparametrize the right circular helix of Example 1.6 by arc length.
2.4. Reparametrize the curve tt,(t) = (e' cost, e' sin t, e') by arc length.
2.5. Reparametrize the curve tt,(t) = (cosh t, sinh t, t) by arc length.
2.6. Show that or.(s)is a unit speed curve where
tt,(s) = !(s ·I ,J s 2 + I, (s + Js2'+Tt 1, ,J2 In (s + ,J s 2 + I)).
*2.7. Formulate an appropriate definition of a reparametrization of a curve
segment.
24 Local Curve Theory Chap.2

2.8. Let ~(t) be a regular curve with Id~/dt I = a, where a is a fixed positive
constant. Show that ifs is arc length measured from some point, then
t = (s/a) + c for some constant c.

2-3. CURVATURE AND

THE FRENET-SERRET APPARATUS

This section is devoted almost entirely to making definitions and develop-

ing an intuitive feeling for these definitions. The reader should look again at
the last paragraph of the previous section before starting on this section.

DEFINITION. A curve~: (a, b)---+ R 3 is a unit speed curve if ld~/dtl = 1.

Note that for a unit speed curve~ = ~(t), the arc lengths= t - t 0 • We
shall assume that t 0 has been chosen to be O (so thats= t) and will write~
as a function of s. Because of this convention we can write unambiguously
~· = ~'(s) = T(s) where T is, as always, the tangent vector field (see Corol-
lary 2.3). For the rest of this section we shall assume that ~(s) is a unit speed
curve. This assumption amounts to the philosophical statement that we are
only interested in the geometric shape of a regular curve since any regular
curve can be reparametrized by arc length (Theorem 2.2) and reparametrizing
does not change the shape of a curve. In Section 2-6 we will compute the
quantities defined below if the curve is not given in a unit speed parametriza-
tion.
We now motivate the definition of "curvature" (of a curve). "Curvature"
will measure bending and will serve as the central concept of study in this
book (and, indeed, in all of differential geometry). The reader probably has
some intuitive idea of what "curvature" is. Whatever the definition of "curva-
ture" is, it should satisfy two criteria: (1) the curvature of a straight line
(Example 2.4) is zero; and (2) the curvature of a circle (Example 2.5) is the
same at each point. In terms of the curvature measuring bending, (1) says
that a straight line does not bend at all and (2) says that a circle has constant
bending. What is it about a straight line that does not change (i.e., what
might we choose to be a measure of bending?)? A glance at Example 2.4
(~(t) = u + tv) shows that the tangent vector field <?fastraight line does not
change with the arc lengths. It is v/1vI,which is independent of s. A glance at
Example 2.5 shows that the tangent vector field of a circle of radius r does
change withs but that its derivative has constant length 1/r. Because of these
considerations we are led to make the following definition.

DEFINITION. The curvature of a unit speed curve ~(s) is K(s) = IT'(s) I-

From the above discussion, it is clear that K(s) = 0 (for all s) if~ is a
Se,:.2-3 Curvature and the Frenet-Se"',t Apparatus 26

straight line and ,c(s) = 1/r (for all s) if cz is a circle of radius r. This last
equality is especially appealing because it says that the smaller the radius is
the larger the curvature is (that is, the faster or tighter the circle is bending),
which conforms with our intuition.
We may also give a heuristic description of curvature for curves lying in
the plane. If czlies in the (x, y) plane, then cz(s)takes the form
cz(s) = (x(s), y(s}, 0),
hence cz'(s) = T(s) = (x'(s), y'(s), 0). We now let 9 be the angle between the
horizontal and the tangent vector field to cz at s as in Figure 2.5. (Technical
remark: This angle 9 is not really well defined. It is defined only up to a
multiple of2n. We will remedy this problem in Section 3-2. This is the reason
we do not separate out what follows as a theorem but merely call it a heuristic
description.)
T(s)

FIGURE 2.5
~
Because of the description of the angle 9 we have
x'(s) = (T(s), (1, 0, 0)) = cos 9(s)
and T(s) = (cos 9(s), sin 9(s), 0). Thus
T'(s) = ( -sin 9(s), cos 9(s), o)(~
and ,c(s) = IT'(s) I = Id9/ds I· This shows that the curvature of a plane curve
la the rate of change of the angle the tangent vector field makes with the hori-
zontal (up to sign). This approach to the curvature of plane curves is essen-
tially due to L. Euler (1736).
Having justified the definition of curvature, we shall now develop some
machinery to study curvature. This machinery which is called the Frenet-
Serret apparatus, is the key to studying the geometry of curves in R 3 and in
fact uniquely determines the curve as we will see in Section 2-5.
It is usual in both elementary physics and mathematics to think of a
vector as an arrow with a head and a base point (or "point of application").
If we imagine at each point cz(s)on the curve the set of all vectors whose base
point is cz(s),then we obtain at each point cz(s)a 3-dimensional vector space.
From the point of view of geometry, what is a natural basis for these vector
spaces? Certainly if e 1 = (1, 0, 0), ,e2 = (0, 1, 0), and e3 = (0, 0, 1), then
26 Local Curve Theory Chap. 2

{e1, e 2 , e 3} is a basis for these vector spaces (for each s). The problem is that
{e1, e 2 , e 3} reflects the geometryofR 3 instead of the geometry of the curve and
is thus unsatisfactory from a geometric viewpoint. Our line of attack is to
take the one "geometric" vector we have (the tangent vector field), find
another one (the normal vector field), and use their cross product (the binor-
mal vector field) to obtain an "intrinsically geometric" basis. The following
*
definitions are only valid at those points where TC(s) 0. Note that T'/,c is
a unit vector.

DEFINITION. The principal normal vector field to a unit speed curve ~(s) is the
(unit) vector field N(s) = T'(s)f,c(s). The binormal vector field to ~(s) is
B(s) = T(s) x N(s). The torsion of~ is the real-valued function
-r(s) = -(B'(s), N(s)).

We will give the geometric meaning of torsion in the next section. (It
will measure "how far from lying in a plane" ~ is.)

DEFINITION.
(a) The Frenet-Serret apparatus of the unit speed curve ~(s) is
{,c(s),-r(s), T(s), N(s), B(s)}.
(b) If P(t) is a regular curve we ,may write t = t(s) or s = s(t) by
Theorem 2.2. Let ~(s) = P(t(s)) be a unit speed reparametrization
of p and let {,c(s), -r(s), T(s), N(s), B(s)} be the Frenet-Serret appa-
ratus of the unit speed curve~- The Frenet-Serret apparatus of P(t) is
{,c(s(t)), -r(s(t)), T(s(t)), N(s(t)), B(s(t))}.
Because it is so difficult (and often impossible) to find t as a function of
s explicitly, we shall need a computational tool for finding the Frenet-Serret
apparatus for a non-unit speed curve. This is done in Section 2-6.

EXAMPLE 3.1. Let ~(s) = (r cos (s/r), r sin (s/r), 0) be a circle of radius r > 0.
T(s) = (-sin (s/r), cos (s/r), 0) so that TC(s)= IT'(s) I = l/r, as was men-
tioned earlier in this section. Note that

N(s) = ~gj=(-cos(;), -sin (;),o)

and B(s) = T(s) x N(s) = (0, 0, 1). Since B' = 0, we see that -r(s) = 0.
We have completed the computation of the Frenet-Serret apparatus in
this case. We sketch {T(s), N(s), B(s)} at some points in Figure 2.6.

ExAMPLE
3.2. Consider the unit speed circular helix
~(s) = (r cos ros, r sin ros, hros),
where ro = (r 2 + h 2)-11 2 • (See also Example 1.6.)
Sec. 2-3 Curvature and the Frenet-Serret Apparatus 27

FIGURE 2.6

T(s) = a>(-r sin ros,r cos ros,h)

T'(s) = -ro 2 r(cos a>s,sin a>s,0).
Thus K = a>2r, which is a constant, yet cz is not a circle.
N = (-cos ws, -sin ws, 0)
B =T x N = w(h sin ws, -h cos a>s,r)
B' = ro2h(cos ws, sin ws, 0).
Thus T = -(B' N) = a>2 h. We sketch {T(s),N(s), B(s)}at several points
in Figure 2.7. Note that the curvature and torsion are both constant in
this example.

T
T

FIGURE 2.7

U!MMA
3.3. Let Cl(s) be a unit speed curve. Then for every s such that
K(s) * 0, the set {T(s), N(s), B(s)} is an orthonormal set.
Proof· T(s) is a unit vector, so by Lemma 5.1 of Chapter l, (T, T') = 0.
Since T' = KN and K *
0, we have (T, N) = 0 and so T and N are orthogo-
nal. Since B = T x N, Lemma 3.3g of Chapter 1 shows that Bis orthogonal
to both T and N. Since T and N are unit vectors, so is B = T x N. I

Because of this lemma, at every point on the curve where K 0 we have *

an orthonormal set of vectors {T, N, B}that move and twist as we move along
28 Local Curve Theory Chap. 2

the curve. This is the reason that {T, N, B} is classically called a "moving
frame" or "moving trihedron." Look again at Figures 2.6 and 2.7.
If a curve has isolated points at which K is z.ero, then our construction
is not valid. However, if K is z.ero on an interval, then our intuition demands
that the curve be a straight line. This is indeed the case.

PRoPOSmON3.4. Let cz(s) be a unit speed curve with K = 0 on an interval

[a, bJ.Then the curve segment iz: [a, bJ- R 3 is a straight line.

Proof: According to Section 1-4 we must produce a vector v and a point x 0

of R 3 such that iz(s) = sv + x 0 • Since IT' I = K =
0, Tis constant. Now for
any curve, cz(s) = f T(u) du + .s(a}. Thus, since T(u) = T is constant,
cz(s)= (s - a)T + cz(a). Hence we may let v = T and x 0 = -aT + iz(a). I
We note that any unit speed curve can be broken into segments with
K =0 on some and K = 0 only at the end points of others: By the above
proposition, we completely understand the geometry of these segments where
K =0. We shall usually consider only the second type of curve and in fact
will quite often restrict ourselves to 11: (a, b) - R 3 with K :;z!:0. This is because
at this stage we are interested in the local behavior of curves, which means the
behavior of the curve near a particular point. (If K(c) :;z!:0, then K(s) :;z!:0 for
all s near c since K is continuous.)
At an isolated point where K = 0 strange things can happen. Problem
4.14 gives an example of a C"° curve iz: (-oo, oo)- R 3 with the image of
(-oo, OJlying in one plane and the image of [O, oo) lying in another. Note
that K is z.ero only at one point (cz(O}).

PROBLEMS

3.1. Show that iz(s) = (fJcos s, fJ - sins, -tt cos s) is unit speed and
compute its Frenet-Serret apparatus.
3.2. Show that

cz(s) -
- (CI+3s)''2 ' (I - 3s)''2 ' -:::TT
s )

is a unit speed curve and compute its Frenet-Serret apparatus.

3.3. Show that iz(s) = f(cos- 1 (s) - s~, I - s 2 , 0) is a unit speed
curve and compute its Frenet-Serret apparatus.
3.4, Show that cz(s) = (~, 2s, In (s +~))/~is a unit speed
curve and compute its Frenet-Serret apparatus.
3.5, Compute the Frenet-Serret apparatus of the curve in Problem 2.6.
ssc.2-4 The Frenet-Serret Theorem and Its Corollaries 29

•3.6. Let «z(s) be a Ck curve in the (x, y) plane. Prove that if K =I= 0 then
T =0. (Hint: There are differentiable functions x(s) and y(s) such that
cz(s) = (x(s}, y(s), 0). Show that B = ±(0, 0, I).}
3.7. Let «z(s) = (x(s),y(s}, 0) be a unit speed curve. Prove that
K = Ix'y" - x"y' I,

2-4. THE FRENET-SERRET THEOREM AND

ITS COROLLARIES

Because the Frenet-Serret apparatus has been defined so geometrically

we would expect to get a great deal of information from it. This is indeed the
case. After a preliminary lemma, we prove the Frenet-Serret Theorem from
which we can derive many geometric corollaries. Even though the lemma
below belongs-to the realm of linear algebra, we shall prove it in detail here
because it is so crucial in the proof of the Frenet-Serret Theorem.

LEMMA4.1. If E = {e1, ... , en} is an orthonormal set of n elements of an

n-dimensional inner product space V, then
(a) Eis a basis for V; and
(b) if v E V, then v = :E(e" v)e
l=l
1•

Proof:
(a) Because the number of elements in Eis the dimension of V we need
only prove that E is linearly independent. Let c 1 , c 2 , ••• , en be real numbers
with :Ec'-e, = 0. Then O = (:E c'-e,, e1) = :Ec'-(e" e1) = :Ec1011 = c1 for
eachj. (Recall that 011 is the Kronecker delta as defined in Equation (1-1))
of Chapter 1.) Therefore, {e1, e 2 , ••• , en}is linearly independent and a basis
for V.
(b) Since Eis a basis, we know that for each v E V there are real num-
bers vJ such that v = :Evier Therefore,
(ei, v) = (e,, :Evie1) = :Ev1011 = v',
which proves (b). I
The important thing about the above lemma is not that it tells us that
v E V can be expressed as a linear combination of the elements of E, but
rather that it tells us how to express v as a linear combination of the elements
of E. To appreciate this, consider what you must do if you are given an
arbitrary basis {u1tU2, ••• , Dn}of V and are asked to write a given vector
v E Vas v = :Ea'u,. You must solve n linear equations in n unknowns,
which is, in practice, very difficult. If, however, the basis is an orthonormal
30 Local Curve Theory Chap. 2

one, (Vis an inner product space), then it is easy-you need only compute
(u,, v) for eachj.
THEOREM 4.2 (Frenet-Serret). Let ~s) be a unit speed curve with K * 0 and
Frenet-Serret apparatus {K,T, T, N, BJ. Then
(a) T'(s) = K(s)N(s)
(b) N'(s} = -K(s}T(s) + T(s)B(s)
(c) B'(s) = - T(s)N(s)
for each s.

Proof: Lemma 3.3 asserts that {T(s},N(s), B(s)} is an orthonormal set. There-
fore, according to Lemma 4.1, we may write.any vector v as
(4-1) v = (T, v)T + (N, v)N + (B, v)B.
(a) T' = KN is the definition of K and N.
(b) Since N' is a vector, we may apply Equation (4-1) with v = N'. We
first compute the coefficient (T, N'). Differentiating O = (T, N), we have
0 = (T', N) + (T, N') = (KN, N) + (T, N') so that (T, N') =-JC.Since
N is a unit vector, (N, N') = 0 and the second coefficient of (4-1) is zero.
To compute the third coefficient, (B, N'), notice that (B, N) = 0 so that
by using the definition of T, 0 = (B', N) + (B, N') = -T + (B, N') and
(B, N') = T. Putting these three coefficients in (4-1) with v = N' yields
Formula (b).
(c} Formula (c) is obtained by using Equation (4-1) with v = B'. The
first coefficient (T, B') is zero because O = (T, B) implies
0 = (T', B) + (T, B') = (KN, B) + (T, B') = (T, B'),
since (N, B) = 0. The definition oh gives (N, B') = (B', N) = -T. Finally,
since Bis a unit vector, the last coefficient (B, B') is 0. I

The reason that we have left spaces in the statement of Theorem 4.2 is
that it is easy to remember the form of the equations in matrix format:

(N'T')= ( 0
-K
K
O
~)(:).
B' 0 -T

The fact that the matrix is skew symmetric is very important in more abstract
differential geometry.
The equations in the above theorem are naturally called the Frenet-
Serret equations. They were independently found by Frenet (1847, published
in 1852) and Serret (1851). We show how powerful these equations are by
drawing some corollaries and leave other applications for the exercises. The
first corollary shows that the vanishing of torsion characteriz.esplane curves
with K *o.
.StlC.2-4 The Frenet-Serret Theorem and Its Corollaries 31

COROLLARY 4.3. Let cz(s)be a unit speed curve with K :;z!:0. The following are
equivalent:
(a} The image of czlies in a plane (more simply, czis a plane curve).
(b) B is a constant vector.
(c) T(s) = 0 for all s.

Proof: The equivalence of (b) and (c) is given by the Frenet-Serret equation
B' = -iN. If czis a plane curve, we may assume (by appropriate choice of
coordinates in R3 } that czlies in the (x, y) plane. For this case we have already
computed the torsion (Problem 3.6) and it is zero, so (a) implies (b).
We now show (b) implies (a). Let x 0 be any point on cz,say x 0 = «(O).
According to Section 1-4, we must find a vector v (which does not, of course,
depend on s) such that (cz(s} - x 0 , v) is identically zero. A glance at the
Frenet-Serret equations or a close look at Example 3.1 (when v = (0, 0, l}}
suggests that we should let v be the binormal vector B.
Since B is a constant vector field, B' = 0. Thus
(~(s) - x 0 , B)' = (~'(s), B) + («(s) - x 0 , B') = (T, B) =0
and (cz(s) - x 0 , B) is therefore constant. But («(s) - x 0 , B) = 0 at s = 0,
so that this constant is zero and «(s) lies in a plane. I

Note that the proof actually gives the plane in which czlies. It is the plane
through x 0 perpendicular to B. This theorem is actually false if the assump-
tion K :;z!:0 is omitted. See Problem 4.14.

DEFINfflON. The osculatingplane to a unit speed curve czat the point cz(s}is
the plane through cz(s) perpendicular to B (and hence spanned by T
andN).
The normalplane of «(s) is the plane perpendicular to T.
The rectifyingplane of cz(s) is the plane perpendicular to N.

One of the standard topics in classical curve theory is an investigation

into the projections of a given curve onto the above planes. The interested
reader may consult, for example, Hicks (1965), Stoker [1969), Laugwitz
(1965), Goetz [1970),or Struik [1961).
What is the significance of the osculating plane? lf a curve actually lies
=
in a plane (T 0), that plane is the osculating plane. More generally, the
osculating plane at cz(s)is that plane which czis the closest to being in, just as
the tangent line at «(s) is the line that czis the closest to being in. See Figure
2.8. (Note that the tangent line lies in the osculating plane.) The curve twists
out of the osculating plane, and -r measures this twisting or torsion. As s
increases, the curve twists toward the side B points to if -r > 0 and toward the
opposite side if -r < 0. This gives geometric meaning to the sign of -r. Also,
since B' = -TN and B is the normal to the osculating plane, -r measures how
32 Local Curve Theory Chap.2

FIGURE 2.8

the osculating plane is turning as s increases. This provides the geometric

interpretation of T which we promised in the last section.

DEFINITION. A (general) helix is a regular curve er.such that for some fixed
unit vector u, (T, u) is constant. u is called the axis of the helix.
Intuitively, the helix grows linearly in the direction of the axis.

EXAMPLE4.4. The right circular helix of Example 1.6 is a helix with axis
(0, 0, 1) according to Problem 1.3.

EXAMPLE 4.5. Any regular plane cutve is a helix since B is constant and may
serve as u.

COROLLARY 4.6 (Lancret, 1802). A unit speed curve a.(s) with 0

K =I=- is a
helix if and only if there is a constant c such that T = cK.

Proof· Assume er.is a helix. Since (T, u) is a constant, we may write

(T, u) = cos8
where 8 is some fixed angle (called the pitch of er.).If 8 is an integer multiple
of n, then u = T or u = -T. In either case this implies that K = 0, which
is a contradiction. We may therefore assume that 8 is not an integral multiple
of n. The following computation shows that N is perpendicular to u:
0 = (T, u)'= (T', u) = K(N, u).
Hence Lemma 4.1 shows that u = aT + bB, where a = cos 8 and b = (B, u).
Since lul = 1, we may choose the sign of 8 so that b = sin 8. Then
0 = u' = (cos 8)KN - (sin 8)'t'N
and K co~ 8 = T sin 8. Since 8 is not an integral multiple of n, we have that
T = cK where c = cot 8.
We now assume that T = cK and show that er.is a helix. Motivated by
the first half of the proof, we define 8 by cot 8 = c with O < 8 < n and let
Sec. 2-4 The Frenet-Serret Theorem and Its Corollaries 33

+
u = cos OT sin OB. An application of the Frenet-Serret equations shows
that u' = 0, so that u is constant. Note also that (T, u) = cos 0, which is
constant, so that «i is a helix. I

Although Lancret stated this theorem in 1802, the first proof was given
by De Saint Venant in 1845.
One of the beautiful things about the above corollary is its constructive
nature. If you are given the constant c such that 't = cK, then you can actually
compute the axis of the helix by first finding its pitch and using the values of
T and B at a single point.
In Problem 5.4 you will show that if both K and -r are constant, then «i
is a circular helix.
We end this section with several propositions which illustrate how the
Frenet-Serret Theorem can be applied to derive simple geometric results.
Many more such results are contained in the exercises. The reader should note
that the general method of proving these results is as follows: (1) express the
geometric hypotheses as an algebraic equation using linear algebra; (2) dif-
ferentiate an appropriate expression (possibly several times), using the
Frenet-Serret Theorem and the hypotheses; (3) interpret the result geometri-
cally.

PROPOSITION 4.7. «i(s) is a straight line if and only if there is a point x 0 E R3

such that every tangent line to «i goes through x 0 •

Proof· If «i(s) is a straight line, any point on «i(s) may be chosen as x 0 since
the image of «i(s) is the tangent line at each point.
Now suppose that every tangent line to «i goes through x 0 • Then
«i(s} - X0 = l(s)T(s)
for some function l(s). Either ,c = 0 or N(s) is defined. In the second case
T(s) = «i'(s) = l'(s)T(s) + l(s}T'(s) = l'(s)T(s) + l(s)K(s)N(s).
Hence (A'(s) - l)T +
l(s)K(s)N = 0. Since T and N are linearly independent,
l'(s) =
I and l(s},c(s} =
0. Thus l(s) = s +
c, which is not constant, and
hence K(s) = 0. Then «i is a straight line by Proposition 3.4. I

PROPOSITION 4.8. Let ll(s) be a unit speed curve such that every normal plane
to «i(s) goes through a given fixed point x 0 E R 3 • Then the image of «i
lies on a sphere.

Proof· The normal plane is orthogonal to T, so that («i(s) - x 0 , T) = 0.

Then («i - X 0 , ll - x 0 )' = 2(«i - x 0 , T) = 0 and («i - X 0 , ll - x 0 ) is a
constant a?:_ 0. If a = 0, then ll(s) = x 0 and «i(s) is not regular. Hence a> 0
and cz(s) lies on a sphere with center x 0 and radius .,./a. I
34 LocalCurveTheory Chap.2

PROPOSITION4.9. Let «z(s) be a unit speed curve with K 0. Then «z(s) lies in*
a planeif and only if all osculating planes are parallel.

Proof- If IX(s)
lies in a plane, that plane is the osculating plane at each point.
Hence all osculating planes are parallel.
Supposeall osculating planes are parallel. Then the values of B(s) at
any two pointsare parallel. Hence B(s) is constant and «zlies in a plane by
Proposition 4.3. I

PROPOSITION4.10. Let cz(s) be a unit speed curve whose image lies on a sphere
of radiusr and center m. Then K 0. If T *
0, then *
«z- m = -pN- p'aB,
wherep = 1/K and a= 1/-r. Hence r 2 = p2 + (p'a) 2•

<
Proof· Wehave cz(s) - m, «z(s}- m) = r 2 , so that
0 = (cz(s) - m, «z(s} - m)' = 2<iz(s) - m, T).
Then
0 = (cz(s) - m, T)' = (T, T) + <iz(s) - m, T')
= I + («z(s) - m, KN),
or K(«z(s)- m, N) = - I 0. Thus * K * 0.
*
Assume-r O. cz(s) - m = df + bN + cB, where the coefficients a, b,
c may be found by Lemma 4.1.
a= («z(s) - m, T) = 0.
b = (iz(s) - m, N) = _ __!_= -p.
K
= c («z(s) - m, B).
Since («x(s)- m, N) = -p,
-p' = (cz(s) - m, N)' = (T, N) + («z(s) - m, -KT+ -rB)
= 0 - K(«z(s) - m, T) + -r(«z(s) - m, B)
= 0 + -r(«z(s)) - m, B).
Hence c = (cr.(s)- m, B) = -p'/-r = -p'a. Thus
«z(s) - m = -pN - p'aB.
Since N and B are orthonormal,
r2 = (cz(s) - m, «z(s) - m)
= 1-pN - p'aBl 2 = P2 + (p'a)2. I
P is called the radius of curvature and a the radius of torsion.
It is possible to generalize much of the previous material to curves in
Sec. 2-4 The Frenet-Serret Theorem and Its Corollaries 36

higher dimensional Euclidean spaces-CZ: (a, b) ·- • R•. The interested reader

might consult H. Gluck [1966, 1967). The Frenet-Serret frame {T, N, B} is
just one geometrically nice basis for the vector spaces along a curve. There
are other possibilities. See R. L. Bishop [1975).

PROBLEMS

General
4.1. Prove that = -<T', B').
1C'C

4.2. If cz(s)is a unit speed curve, prove that [cz',cz", cz"'] = K 2 T.

4.3. Let cz(s)be a C4 unit speed curve with K > 0.
(a) Prove that [cz", cz"', cziv]= K5 ('r/K)'.
(b) Prove that cz is a helix if and only if [cz", cz'", cz1v) = 0.
4.4. Let cz(s)be a unit speed curve with K =t=0. Prove
[cz' cz" cz'"J
1: = ' ' .
<
cz", cz")
4.5. Describe (if possible) any special geometric shape or form of the
curves in Problems 3.1, 3.2, 3.3, and 3.4. If any are helices, find the
axis and pitch.
4.6. Find the equation of the normal plane to cz(t) = (e', cost, 3t2 ) at
t = I. (Note: t is not arc length!)
4.7. Ifcz(s) is a unit speed curve with K =t=0, show that the equation of the
osculating plane through cz(O)is [x - cz(O),cz'(O),cz"(O)J 0. =
4.8. Find the equations of the osculating plane for the curves in Problems
3.1, 3.2, 3.3, and 3.4 at s = 0.
4.9. What is the angle between the axis of a helix and the normal to its
osculating plane in terms of the pitch?
*4.10. Prove:
(a) T = N x B =-Bx N;
(b) N = B x T = -T x B;
(c) B = T x N = -N x T.
4.11. Let cz(s) be a unit speed curve with K =I=0. Find a vector w(s) such
that T' = w x T, N' = w x N, B' = w x B. (w is called the Dar-
boux vector after G. Darboux who investigated curves from a kine-
matical viewpoint. If a rigid body is thought to move along the curve
with unit speed, then the instantaneous motion of this body is
described by a translation vector and a rotation vector. The transla-
tion is given by the velocity vector T, while the rotation is described
by the Darboux or angular ,,efocity ,,ector w.)
38 Local Curve Theory Chap.2

4.12. Prove that cz(s)is a straight line if and only if all its tangent lines are
parallel.
*
4.13. Let cz(s) be a curve with 'IC 0. Prove that cz(s) lies in a plane if and
only if there is a point x 0 in R 3 such that every osculating plane
passes through x 0 •
*t4.14. Let

f(t) = {
e-11,•
O
ift *o
if t = 0.
You may assume/is C with J<•>(O)= 0 for all n. Let cz(t) be given
00

l
by
(t,f(t), 0) if t <0
cz(t) = (0, 0, 0) if t = 0
(t, O,f(t)) if t > 0.
(a) Prove that cz is regular and COO.
(b) Show that 'IC = 0 at t = 0. (Note: tis not arc length.)
cz consists of curves in two different planes joined together at a point
where 'IC = 0.

Set A-Spherical Images

The next seven problems deal with the notion of tangent, normal, and
binormal spherical images. These notions, especially that of the tangent
spherical image, are very important in Chapters 3 and 5.
If cz(s) is a unit speed curve, then T: (a, b) - R3 gives a curve defined
bys - T(s). This curve may not be regular. Since IT(s) I = 1, the image of T
lies on the sphere of radius 1 about 0. This curve is called the tangentspherical
image of cz. We may also consider the normal sphericalimage N (defined by
s - N(s)) and the binormalsphericalimage (defined bys--+ B(s)) of cz.
4.15. Find the tangent, normal, and binormal spherical images of the
helix in Example 1.6.
4.16. Let cz(s)be a unit speed curve.
(a) Prove that the tangent spherical image of czis a constant curve if
and only if cz is a straight line.
(b) Prove that the binormal spherical image of cz is a constant curve
if and only if cz is a plane curve.
(c) Prove that the normal spherical image of cz is never constant.
*4.17. Lets be arc length along the tangent spherical image of cz so that
=s J;IT'(a)I da.
(a) Prove that ds/ds= "'·
Sec. 2-4 The Frenet-Serret Theorem and Its Corollaries 37

(b) Find a necessary and sufficient condition for the tangent spherical
image of ex to be a regular curve.
(c) Lets* be the arc length along the normal (resp. binormal) spher-
ical image of ex. Prove that ds*/ds = .j,c 2 + T2 (resp. !T !),.
4.18. Let ex(s) be a unit speed curve with ,c > 0. Lets* be arc length on the
normal spherical image. Prove ,c = Ids*/ds I if and only if exis a plane
curve.
*
4.19. Let ex(s) be a unit speed curve with ICT 0. Prove that the tangent to
the tangent spherical image is parallel to the tangent to the binormal
spherical image at corresponding points.
*4.20. If the tangent spherical image of a unit speed curve ex(s) lies in a
plane through (0, 0, 0), prove that ex(s) is a plane curve.
4.21. Prove that a unit speed curve ex(s) is a helix if and only if its tangent
spherical image is an arc of a circle.
Problem 6.10 will also deal with the tangent spherical image of a curve.

Set 8-Sphere Curves

The next six problems deal with sphere curves, i.e., curves ex(s) such that
there are constants m and r > 0 with (ex(s) - m, ex(s) - m) = r 2 • The pur-
pose of this set is to give some recent characterizations of sphere curves
(Problems 4.26 and 4.27). Bishop [1975] also gives a recent characterization.
4.22. Let ex be a regular curve and let a be a point that belongs to each
normal plane of ex. Prove that ex is a sphere curve. (See also Problem
1.7.)
4.23. Show that ex(9) = (-cos 29, -2 cos (J, sin 29) is a sphere curve by
showing ( -1, 0, 0) belongs to each normal plane.
4.24. Letex(s) bea unit speed curve with,c * *
0, T 0 andp = 1/,c, <T = 1/T.
Assume p 2 + (p'a) 2 = constant = a 2 where a > 0. Prove that the
image ofex lies on a sphere of radius a. (Hint: Show ex+ pN + p'aB
is constant. Call this m. It should be the center of the sphere. This is
motivated by Proposition 4.10.) Assume p'¢ 0 on any interrals.
4.25. Combine the previous result with Proposition 4.10 to prove that if
ex(s) is a unit speed curve with ,c *0, T *
0, then ex(s) lies on a
sphere if and only if T/IC= (,c'/T1C2 )' (or TP = - (p'/T)'}.
4.26. Prove that a unit speed curve ex(s) lies on a sphere if and only if
,c > 0 and there exists a differentiable function /(s) with /T = p',
f' + Tp = 0. (This result is due to Y-C Wong [1963].)
38 Local Curve Theory Chap.2

t4.27. Use the previous result to show that ct(s) lies on a sphere if and only
if there are constants A and B with

"( A cos u:
T ds) + B sin ( f:
T ds)) =-I.
(This result is an improvement by Y-C Wong [1972) of a result of S.
Breuer and D. Gottlieb [1971).)

Set C-Bertrand Curves

The next five problems deal with a classical topic in the theory of curves,
namely Bertrand curves. These curves were first investigated by J. Bertrand
[1850). Similar types of curves are treated in E. Salkowski [1909) and A. Voss
(1909). These problems will not be used in the body of the text.
ct(s) and P(s) are called Bertrand curl'es if for each s 0 , the normal line
to ct at s = s 0 is the same as the normal line to Pats= s 0 • (s need not be arc
length on both ct and p.)We say that pis a Bertrand mate for ct if ct and pare
Bertrand curves. (Note that this means N .. = ± N 11.)
4.28. (a) Show that any two circles in the plane with the same center are
Bertrand curves.
(b) Let
ct(s) = !(cos-• s - s~, 1 - s 2 , 0)
and let
P(s) = !(cos-• s - s,Jl=7 - s, 1 - s 2 + ~. 0).
By Problem 3.3 ct(s) is unit speed. pis not unit speed. Show that
ct and p are Bertrand curves.
4.29. Prove that the distance between corresponding points of a pair of
Bertrand curves is constant. (Hint: If ct(s) is one of the curves with a
unit speed parametrization, then there is a function ,l(s) such that
ct(s) + l(s)N(s) = P(s) gives the other curve. Prove ,l is constant.
P(s) may not be a unit speed curve.)
4.30. Prove that the angle between the tangents to two Bertrand curves
at corresponding points is constant. (Hint: Show that (T .., T 11) is
constant.)
t4.31. Let ct(s) be a unit speed curve with TCT=t=0. Prove there is a curve
P(s) (s not arc length on P)so that ct and Pare Bertrand curves if and
onlS-if there are constants l =t=0 andµ with 1/,l = ,c + µ1:.
4.32. Let ct(s) be a unit speed plane curve. Prove there exists a curve P(s)
so that ct and pare Bertrand curves.
Problems 4.41, 5.6, and 6.12 will also deal with Bertrand curve::.
Sec. 2-4 The Frenet-Serret Theorem and Its Corollaries 39

Set D-Spherica/ Contact

In Corollary 4.3 and the discussion that followed we saw that the osculat-
ing plane at at(s0 ) is the plane that at is the closest to being on and 't'(s0 ) mea-
sured how close at is to actually lying on it. Motivated by the problems on
sphere curves, we can ask what is the sphere (osculating sphere) at ~s 0 ) that
at is closest to lying on and how close is at to actually lying on it (spherical
contact). Note that if at actually lies on a sphere with center m, then
I at(s) - m 12 is constant (hence all of its derivatives with respect to s are zero)
and so we are led to the following definition of contact. Let m and r > 0 be
given and let c(s) = I at(s) - m 12 • We say that at has jth order spherical
contact with the sphere of radius r and center m at s = s 0 if c(s 0) = r2,
c'(s 0 ) = c"(s 0 ) = ... = c<J'(s0 ) = 0. Note that the larger j is, the closer c(s)
is to being a constant function and so the closer at(s) is to lying on a sphere
of radius r and center m.
4.33. If ,c -:f::.0 compute the first three derivatives of c(s) in terms of T, N,
B, ,c and 't'.
4.34. Prove that at(s) has second order spherical contact at s = s 0 if and
only if m = at(s 0 ) + (l/,c(s 0 ))N(s 0 ) + ..tB(s0 ), where ..t is arbitrary.
Note that as ..t varies with s 0 fixed we get a straight line of possible
centers called the polar axis of at at s 0 • The point
I
me = at(s0 ) + -( )N(so)
1C So
is called the center of curvature of at at s 0 • p(s 0 ) = l/,c(s 0 ) is the radius
of curvature at s 0 • The circle of radius p(s 0 ) with center me and lying
in the osculating plane is called the osculating circle of at at s 0 • See
Figure 2.9.
4.35. If 't'(s0 ) -:f::.0, prove that at(s) has third order spherical contact if and
only if ..t = -,c' /'t'K2 , where the right-hand side is evaluated at s = s 0 •
The sphere with center
_ ,., ) I N( ) ,c'(s 0 ) B( )
m, - <N\So + -( )
1C S0
So -
't'( So) 1C2 ( So) So
with third order contact is called the osculating sphere of at. See
Figure 2.9.

FIGURE 2.9
40 Local Curve Theory Chap.2

4.36. Prove that at a point where -r(s0 ) = 0 there can be third order spher-
ical contact if and only if ,c'(s0 ) = 0. Hence plane curves may not
have third order spherical contact anywhere.
4.37. Let Cl(s) have constant curvature. Prove that the osculating sphere
and the osculating circle have the same radius.
4.38. Under what conditions does the osculating sphere have a constant
radius, i.e., independent of s 0 ?

Set E-lnvolutes and Evolutes

The last six problems deal with another classical topic. Let ct and p be
two regular curves defined on an interval (a, b). p is an involute of ct if P(t 0 )
lies on the tangent line to ct at ct(t0 ) and the tangents to ct and pat ct(t0 ) and
P(t0 ) are perpendicular. p is an evolute of ct if ct is an involute of p.
4.39. Suppose that Cl(s)is a unit speed curve.
(a) If P(s) is an involute of ct (not necessarily unit speed), prove that
P(s) = ct(s) + (c - s)T(s), where c is a constant and T = ct'.
(b) Under what conditions is ct(s) + (c - s)T(s) a regular curve and
hence an involute of ct?
Note that because of part (a) 11' - ct I is a measure of the arc length of
ct. For this reason I' can be formed by unwinding a string from the curve
ct(s). See Figure 2.10 for the case of a plane curve. Because of this, the term
string involute is sometimes used, especially for plane curves. We shall use this
concept in Section 3-5 to give an example of a curve of constant width. The
idea of a string involute is due to C. Huygens (1658), who is also known for
his work in optics. He discovered involutes while trying to build a more
accurate clock. (See Boyer (1968).)

FIGURE 2.10
,ec.
2-5 The Fundamental Existence and Uniqueness Theorem for Curves 41

4.40. (a) Prove that an involute of a plane curve lies in that same plane.
(b) Prove that an involute of a helix is a plane curve. (Hint: What
should be the normal to this plane?)
4.41. Let ct(s) be a unit speed plane curve and let ,S(s) and ,y(s) be two dif-
ferent involutes of ct. Prove that pand 'Yare Bertrand mates.
4.42. Let ct(s) be a unit speed curve with Frenet-Serret apparatus
{K,T, T, N, B} and let P(s) be an evolute of cz.
(a) Prove that P(s) = Cl(s)+ lN + µB.
(b) Prove that l = 1/'IC and (l' - -rµ)/l = (µ' + d)/ µ.
(c) Prove that T = (µp' - pµ')/(µ 2 + p 2 ), where p = I/ICas usual.
(d) Prove thatµ= p cot ([ i- ds + constant).
4.43. Prove that an evolute of a plane curve is a helix.
4.44. Let ct{s) be a unit speed plane curve.
(a) Prove that the locus of the centers of curvature of cz (Problem
4.34) is the unique evolute of cz lying in the plane of ct.
(b) Show that this curve is regular if K' *
O. (A point where 'IC'= 0
is called a vertex:)

2-5. THE FUNDAMENTAL EXISTENCE AND

UNIQUENESS THEOREM FOR CURVES

After the Frenet-Serret Theorem we drew several corollaries which char-

acterized several types of curves by properties of their curvature and torsion:
IC =0 straight line (Proposition 3.4)
IC * 0, T =0 plane curve (Corollary 4.3)

!!. = constant helix (Corollary 4.6).

T

As problems we will have:

i- =0, IC = constant > 0 circle (Problem 5.3)
i- = constant *
0, 'IC = constant > 0 circular helix (Problem 5.4).
There were also a large number of exercises where curvature and torsion
played a major role in describing what was occurring geometrically.
This is not at all accidental. Theorem 5.2 will tell us that as long as
*
'IC 0, the functions 'IC and T completely describe the curve geometricaUy,
except for its position in space. Philosophically, this is the same as saying that
two angles and one side completely describe a triangle in plane geometry, or
that the radius completely describes a circle. Classically, mathematicians have
called 'IC = K(s), -r= i'(s) the intrinsic equations of a curve.
42 Local Curve Theory Chap.2

The actual proof of Theorem 5.2 will not be used in the sequel and may
be omitted. However, the content of the theorem is important and will be
used. The proof will depend on the following basic result from the theory of
ordinary differential equations which is due, in various formulations, to
Picard, Lindelof, Peano, and Cauchy. See Birkhoff and Rota [1969, Chapter
6) or Coddington and Levinson (1955, Chapter 1). The dependence of the
Fundamental Theorem of Curves on a major theorem of ordinary differential
equations shows that in the differential approach to geometry, the heart of
geometry is ultimately in differential equations. We will see this again when
we study geodesics in Section 4-5 and the Fundamental Theorem of Surfaces
in Section 4-10.

THEOREM 5.1 (Picard). Suppose that the R•-valued function A(x, t) is defined
and continuous in the closed region Ix - c I < K, It - a I < T, and
satisfies a Lipschitz condition there. Let M = sup IA(x, t) I over this
region. Then the differential equation dt1./dt= A(,x,,t) has a unique solu-
tion on the interval It - a I ::;:min (T, K/ M) satisfying Cl(a) = c. (The
technical requirement about the Lipschitz condition is satisfied if A has
bounded partial derivatives with respect to the coordinates of R•.)

THEOREM 5.2. (Fundamental Theorem of Curves). Any regular curve with

,c > 0 is completely determined, up to position, by its curvature and tor-
sion. More precisely, let (a, b) be an interval about z.ero, ic(s) > 0 a C 1
function on (a, b), f(s) a continuous function on (a, b), x 0 a fixed point
ofR 3 , and {D, E, F} a fixed right handed orthonormal basis ofR 3 • Then
there exists a unique C 3 regular curve ,x,: (a, b) .- R 3 such that:
(a) the parameter is arc length from ,x,(O);
(b) Cl(O)= x 0 , T(O) = D, N(O) = E, B(O) = F; and
(c) ,c(s) = ic(s), -r(s) = f(s).

Proof: Consider the system of ordinary differential equations

3
u/ = I:
l=l
a'is)u" j = 1, 2, 3,

where (a'1) is the matrix

0
-f
Picard's Theorem implies that this system has a unique solution uh) with
u 1(0) = D, u2 (0) = E, and uJ(O) = F. We shall show that the u, give the
moving trihedron of a regular C 3 space curve ,x,with the required curvature
and torsion.
Sec. 2-5 The Fundamental Existence and Uniqueness Theorem for Curves 43

Step 1. The vectors 01 are orthonormal:

Proof- Let p 11= (ui, u1) so that

pl/' = (u,', u1) + (u" u/) = I: a"1PkJ + I: a\P1k·
Thus p 11satisfies the initial value problem
Pu'= I: (a\pkJ + a\p,k)
0 if i -=l=j
P,/0) = ~If = { l
if i =j.
By Picard's Theorem, such a system has a unique solution. But
L (a\~ki + a\~1k) = a 11 + a 1 =0 =~
1
1/.

Hence ~ii =
p,1 gives a solution, and thus the only solution: (0 1, u1) = ~11
and the 0 1 are orthonormal.

Step 2. Let a.(s) = x 0 + Jiu (u) du1 for s E (a, b). Then ~s) is C 3 ,
regular, and unit speed:

Proof- da./ds = u.(s) by the fundamental theorem of calculus.

dza. , -( )
dsZ = U1 = IC S Oz.

Both ic and Dz are differentiable.

dla. -, Oz
dsl = IC - ,=
+ ICUz -,
IC Oz
+ IC-(-ICU1
- + "rU3
- )•
Since ic is C 1 , f is continuous, and the 0 1 are differentiable (hence continu-
ous), d 3 a./ds 3 is continuous, and a. is C3. lda./dsl = ju 1 j = 1 by Step (1).
Hence a. is regular and unit speed.

Step 3. ic = IC, f = -r, 0 1 = T, Oz = N, u3 = B:

Proof- a.'= u 1 so 0 1 = T. 1CN= T' = 0 1 ' = Ruz. Since both N and Dz are
unit vectors and ic > 0, ic K. Hence Dz = N also. Now [u1 , Oz, u3 ] = = ±1
since the vectors are orthonormal. At s = 0 we have
[0 1 , Oz, U 3] = [D, E, F] = + 1.
Since [ui, Dz,u3 ] is continuous on (a, b), it is always + 1 and {ui, Dz,u3 ] is
right handed. Thus B = T x N 0 1 x Dz= =
ul" Finally,

--rN = B' = o/ = -fuz,

so· f = -r.
This completes the proof of the existence of a curve a.(s) with the required
curvature, torsion, starting point and initial moving trihedron. On the other
44 Local Curve Theory Chap. 2

hand, the definition of ~s) was forced if cz was to solve the problem. Hence
there is a unique curve with the required property. I

In general, given ICand -r it is very difficult to solve the Frenet-Serret

equations and find the curve cz.However, it can (almost) be done in the case
of a helix.

EXAMPLE 5.3. Let ~s) be a helix with IC> 0, -r = clCfor some constant c.
It will be useful to reparametrize cz by a parameter t given by

t(s) = J:IC(a)da.
Note that this is an allowable change of coordinates since t' = IC> 0
implies t(s) is one-to-one and both t(s) and s(t) are differentiable. Since
-r = clC,the Frenet-Serret equations are
T' = ICN, N' = -ICT + clCB, B' = -clCN.
In terms of the parameter t they are
dT dN dB ·
dt =N, dt = -T + cB, dt = -cN.
Thus d 2 N/dt 2 = -N - c 2 N = -co 2N, where co= ../1 + c2 • Since N
solves this differential equation, we have N = cos cot a + sin cot b for
some fixed vectors a and b. dT/dt = N may be integrated to give
T = (sin cot a - cos cot b + c)/w. Hence

cz(s) = ! (S: cot(a) da S:

sin a- cos cot(a) dab+ sc + d).
However, the integration constants a, b, c, dare not arbitrary.
dN = -co sm
dt . cot a + co cos cot b
so that

0= (N,~r;)
= (-co Ia 12 + coIb 12 ) sin cot cos cot + co(a, b)(cos 2 cot - sin 2 cot).
<
At t = 0 this equation is a, b) = 0. Then
!(-cola 12 + colbl 2) sin 2rot= 0
and 191
2 = jbj 2 •
1 = IN 12 = Ia 12 cos 2 cot + Ib 12 sin 2 cot
yields Ia I = Ib I = 1. Thus a and b are orthonormal.
Similarly O = <T,N) yields (a, c) = 0 and then <h,c) = 0. 1 =IT 12
Sec. 2-5 The Fundamental Existence and Uniqueness Theorem for Curves 46

gives lei= lei and soc= ±c(a x b). dN/dt = -T + cB implies that
cB = dN/dt + T. Hence

£.(sin wt a - cos wt b
w
+ c) = cT = N x cB

= [cos wt a + sin wt b] x [ (! - w)<sin wt a - cos wt b) + ! c]

= ..!.[cos wt (a x c) + sin wt (b x c) + c (a x b)].
2
w
Thus c = +c(a x b). In terms of the orthonormal basis {a, b, a x b} we
have

(5-1) cx(s)= ! (f sin rot(a) da, - f cos wt(a) da, cs)+ d 1,

where t(a) = J;,c(s) ds and d 1 = cx(O).

Note that the above solution requires that you be able to compute some
nontrivial integrals, which may not be all that easy to do. See Problems 5.1
and 5.2.

ExAMPLE 5.4. Using Theorem 5.2 we offer an analytic proof of Corollary 4.3.
=
We must show that if exis a unit speed curve with -r 0, then cxlies in
a plane. In Example 5.3 we may put c = O. Equation (5-1) yields the
curve P(s) = (J~sin t(a) da, - J~
cos t(a} da, 0), where t(a) = J;
,c(s) ds.
This is clearly a plane curve, lying in the plane spanned by a and b and
has curvature ,c(s) and torsion O. Hence, by Theorem 5.2, it is the same
as cx(s),up to position. Thus cx(s)is also a plane curve.

The preceding proof, although lacking the geometric appeal of the first
proof, does contain a valuable insight: a unit speed plane curve may be ob-
tained from its curvature by three integrations. See Problems 5.1 and 5.2.

PROBLEMS

5.1. A plane curve (-r = 0) is a helix. Perform the integration as outlined in

Example 5.4 for
(a) ,c = constant > 0;
(b) ,c = l/(ms + n), where m, n are constant.
5.2.• Find a unit speed curve cx(s)with ,c(s) = 1/(1 + s 2 ) and -r = 0.
*5.3, Prove that the only plane or spherical unit speed curves of constant
curvature are circles.
46 Local Curve Theory Chap.2

5.4. Let cz(s)be a unit speed curve with 'IC and -r constant so that a. is a
helix. Prove that a. is a circular helix by showing that its projection in
a plane perpendicular to u is a circle. (Hint: Equation (5-1).)
5.5. Let a. be a helix with axis u. Let 11 be a plane perpendicular to u and let
p(s) be the projection of a. into this plane. (P is not unit speed.) Let ic
be the curvature ofp and prove that 'IC = ic sin 2 (), where <u,
T) =cos(},
(Hint: u = a x b.)
*
5.6. Let a.be a C 3 regular curve with -r 0. Prove that a.is a circular helix
if and only if a. has at least two Bertrand mates. (Hint: Problem 4.31.)
5.7. Let cr.(s)be a unit speed curve with 'IC> 0 and 't > 0. Let

P(s) = f:
B(u) du.

(a) Prove that ti is unit speed.

(b) Show that the Frenet-Serret apparatus {ic,f, T, N, B}of p satisfies
ic = -r, f = 'IC, T = B, N = -N, and B = T.
5.8. Let cz(s)be a helix with 'IC= -r > 0. If P(s) is defined as in Problem 5.7,
show that a.and pare congruent; that is, they are the same curve up to
position in space.

2-6. NON-UNIT SPEED CURVES

In this final section we shall determine the Frenet-Serret apparatus for a

curve that is not given a unit speed parametrization. As was pointed out in
Section 2-2, in practice it may not be possible to reparametrize by arc length.
Thus one must find alternative computational techniques.
Let P(t) be a regular curve and let s(t) denote the arc length function.
Then P(t) = cr.(s(t)), where cr.(s) is P(t) reparametrized by arc length. Note
that ds/dt= ldP/dtl> 0. We wish to determine the Frenet-Serret apparatus
in terms of the variable t. We denote derivatives with respect to t by dots:
fl= dp/dt,Ii = d2 P/dt2 , and = d 3 p/dt 3 •
p
PROPOSITION6.1. If P(t) is a regular curve in R 3 , then
(a) T = fl/lPI;
(b) B = fl x Ii/IPx Ji;
(c) N =Bx T;
(d) K = IP x Ii1/1P13 ; and
iJ x Ii12
(e) i = [P, Ii,P1/I

Proof· (a) Since P(t) = cr.(s(t)), we have fl= a.'s= sT. Because s > 0,
s = IPIand T =Pis= P/IPI·
Sec. 2-6 Non-Unit Speed Curves 47

(d) P = sT +st= sT + s 2T' = sT + ,cs2N. (Since pis the velocity,

this gives the acceleration in terms of its tangential and normal compo-
nents.) Hence p x P = sT x (sT+ ,cs2N) = ,cs3 B and ,cs3 = Ip x pI or
"= IP x Pl/s 3 = IP x Pl/lPl 3 •
*
(b) If 1' 0, then B = P X P/,cs3 = P X P/1P x P I,
(c) Since {T, N, B} is a right handed orthonormal basis, N = B x T.
(e)
p= sT+st+ (,cs2)'N + ,cs2N
= sT + ssT'+ (1cs2)'N + ,cs3N'
= sT+ ,cssN+ (,cs2)'N - ,c2s 3T + 1C'r.f3B
= (s- ,c2s 3)T + (,css+ (,cs2)')N + ,cd 3B.
Hence [P, P, P1
= (P X P, P>= (,cs 3B, P>= -r(7'.f3) 2 = TIP x P 12. Thus
T = [P,p,P1/I
p x p1 I 2•

ExAMPLE 6.2. Let P(t) = (1 + t 2, t, t 3). Then p = (2t, 1, 3t 2), P = (2, 0, 6t),
and = P (0, 0, 6). Hence P x P = (6t, -6t 2, -2) and [P, P, = -12 P1
so that
- (36t 2 + 36t 4 +
4) 112
" - (4t 2 1 + + 9t 4)3/2
-12
T = 36t 2 + 36t 4 +4
T_ (2t, 1, 3t 2 )
- (4t2 + 1 + 9t4)1t2
- (6t, -6t 2 , -2)
B - (36t 2 + 36t 4 + 4) 112
and
(-18t 4 + 2, -4t - 18t 3, 6t 12t 3) +
N = (4t 2 + 1 + 9t 4) 112(36t 2 + 36t 4 + 4) 172 .

EXAMPLE 6.3. We show that

P(t) = (t, 1 + t) r 1 ,r 1 -

fort> 0 is a plane curve. p = (1, -r 2, -r 2 - 1), = (0, 2r 3 , 2r 3), p

and p
= (0, -6r , -6r ). Clearly P x = 0 so that [P,
4 4 P= 0. P,P1
=
By Proposition 6.1, -r 0 and p is a plane curve, provided ,c 0. *
p
p x = (2t- 3, -2r 3, 2r 3) *
0. Hence, ,c 0. *
For a non-unit speed curve P(t), the Frenet-Serret apparatus is a func-
tion oft, nots, and does not satisfy the Frenet-Serret equations. However, we
do have the following variation.
48 Local Curva Theory Chap.2

PROPOSITION 6.4. Let P(t) be a regular curve in R 3 , and let 11(t)= IPI, Then
(a) t = ,c11N
(b) N= -,cvT + f11B
(c) B= - -r11N.

Proof· Problem 6.5. I

PROBLEMS

6.1. Compute the Frenet-Serret apparatus for P(t) = (t, t 2 , t 3) with t > 0.
6.2. Compute the Frenet-Serret apparatus for P(t) = (cosh t, sinh t, t).
6.3. Compute the Frenet-Serret apparatus for P(t) = (t - cost, sin t, t).
6.4. Find the curvature and torsion of P(t) = (e' cos t, e' sin t, e').
*6.5. Prove Proposition 6.4.
6.6. What form do the expressions in Proposition 6.1 take for a curve of
constant speed 11?
6.7. Let P(t) be a regular curve with speed 11= IdPfdt I· Prove that
"= 11- -v<P.P>- v
2 2•

6.8. Prove that P(t) is a straight line if and only if Pand Pare linearly
dependent.
t6.9. Let P(t) = (at, bt 2 , t 3).
(a) Prove that pis a helix if and only if 4b4 = 9a 2 •
(b) What is the axis in this case?
6.10. Let ci(s) be a unit speed curve with ,c :;c 0. Let it and i be the curvature
and torsion of the tangent spherical image. Prove that it= ../1+ (1:/,c)"
and i = (1:/,c)'/,c(l + (-r/,c)2), where ,c and -r: are the curvature and
torsion of 11. (See Problem Set 4A.)
6.11. Let P(t) be a curve with ,c :;c0. Prove that Pis a plane curve if and only
if [P, P,ii1 o. =
t6.12. Prove that the product of the torsions of two Bertrand curves at cor-
responding points is constant. (See Problem Set 4C.)
 3

Global Theory
of Plane Curves

The geometry of a curve at or near a point in no way restricts or reflects

the behavior of the curve at other points "far away" from the given point. In
the previous chapter we adopted the microscopic approach (i.e., we looked in
very small neighborhoods of points and made assumptions about ,c, -r,etc.).
In this chapter we shall adopt the macroscopic or global approach and look
at the entire curve. The global theorems we shall present are very much in the
spirit of much of the modem research in differential geometry. Because we
are dealing with global notions, the theorems become harder to prove, but
they still do not lose their geometric flavor. To understand the difference be-
tween local and global discussions, the reader is urged to look at the curve of
Problem 4.14 in Chapter 2 again. About each point of this curve, except one,
there is a segment of the curve that lies in a plane. Yet the global behavior is
quite strange.
Our first global theorems will deal with plane curves (so that -r = 0). We
shall assume that a suitable choice of coordinates has been made in R 3 so that
the curve lies in the (x, y) plane. We shall then disregard the z-coordinate and
write the curve as if it lies in R 2 •
Section 3-1 contains a briefreview of line integrals and Green's Theorem,
which should be familiar from Calculus III. In Section 3-2 we define the
plane curvature of a plane curve. This new notion of curvature has the
49
50 Global Theory of Plane Curves Chap.3

advantage of being a signed quantity and enables us to globally define a nor-

mal vector field to a plane curve. This new notion of curvature coincides with
the old except for sign. We also introduce the rotation index of a curve. This
measures how many times the tangent vector goes around (the unit circle) as
a closed plane curve is traversed once. In Section 3-3 we discuss convex curves
and give an alternate description of them in terms of curvature. The Isoperi-
metric Inequality (which says that the circle is the curve of largest area for a
given perimeter) is proved in Section 3-4. In Section 3-5 we study ovals,
proving the Four-Vertex Theorem and certain results about curves of con-
stant width. We end the chapter with a brief statement in Section 3-6 about
some global theorems for space curves. These results will be proved in
Chapter 5.

3-1. LINE INTEGRALS AND GREEN'S THEOREM

We review here the very basic concepts of line integrals for use in this
chapter and in Chapters 5 and 6. We shall not use the vector notation since
we have no need of it.
Suppose ci(t) = (x(t), y(t)) is a C 1 parameterization of a geometric curve
e in R 2 with a::;::t::;::b. If/ and g are real-valued functions of two variables,
then by the line integral fef dx +
g dy we mean the ordinary integral

f [!(x(t), y(t)) !:+ g(x(t), y(t)) 'Z]dt.

(Note that there is an obvious generalization to curves in R 3 .)

EXAMPLE
1.1 Let e be the unit circle in R 2 parameterized as
11(8)= (cos 8, sin 8), 0::;:: 8 < 2n.
Then

fey dx + xdy = f" [-sin 2 8 + cos 2 8Jd8

= 12,,:
0 cos 28 d8
sin 2812,,:
= -2- 0 = 0.

EXAMPLE
1.2. Let e be as in Example 1.1. Then

f.ex dy - y dx =
fb
Jo [cos2 8 + sin 2 8] d8
fb
= Jo d8 = 2n.

Somotimes one is given a function to integrate along a curve, such as

curvature. fe K ds may also be thought of as a line integral and is computed
f;
in the standard fashion: feK ds = K(s) ds. If one is told to integrate a
function along a curve, it is implied that this is with respect to arc length. If
Sec. 3-1 Line Integralsand Green's Theorem 61

the function is given in terms of a different parametrization, a suitable change

of coordinates must be made. For example, if it is not practical to repara-
metrize by arc length (Examples 2.6 and 2.7 of Chapter 2), one would
compute fe ,c ds =fe dt, where e is parametrized by t. Al-
1C(&(t))(ds/dt)
though s(t) is not computable, ds/dt = Idrx./dtIis and so is ,c(s(t)) by the results
of Section 2-6. Example 1.5 of Chapter 5 carries out this program for an
ellipse.
fe
If h is a function defined along a curve, dh makes sense as the differ-
ence in the values of h at the end points:

f dh = f dh ds = s.•dh ds = h(b)- h(a).

Je Jeds • ds
If h(s) = p(x(s), y(s)) for some function p(x, y), there is a second way of
computing

Since
fe
dh:

f dh = Je
Je f X dx +
f dp = Je
~dx+~~=dh
** Y dy.

aids ayds ds
by the chain rule, this latter integral does give the same value for dh. fe
Note that Example 1.1 was of this latter type with p(x, y) = xy: dp =
ydx + xdy.
One important tool for computing line integrals is Green's Theorem,
which allows us to replace certain line integrals with double integrals. We
shall omit the proof of Green's Theorem. The interested reader may refer to
an advanced calculus text such as Fulks (1969]. See the next section for the
formal concepts of a closed curve and a piecewise C 2 curve.

THEOREM 1.3 (Green). If e is a closed plane curve made up of C2 curve

segments, which bounds a region <R,and which is traversed counter-
clockwise, then

Lfdx + gdy =IL(~

-i) dxdy,

for all differentiable functions f and g defined on <R.

ExAMPLE fe
1.4. In Example 1.2 we found x dy - y dx = 2n. On the other
hand, setting/= -y, g = x in Green's Theorem yields

fL(~- t) dx dy = fL11- c-•>1 dx dy

= 2 ffCR
dxdy
= 2(area of <R)= 2n.
52 Global Theory of Plane Curves Chap. 3

PROBLEMS

1.1. Compute fe(xy + I) dx + (x 2 /y) dy along each of the following curves:

(a) e is parametrized as IX(t) = (t, t), 0 < t < 1;
(b) e is parametrized as P(t) = (t2, t 3 ), 0 < t < 1.
1.2. Compute fe(x dy - y dx)/(x 2 + y 2) where e is
(a) the straight line from (0, I) to (1, 0);
(b) the straight line from (I, 0) to (I, I) ; and
(c) the straight line from (I, I) to (0, 1).
1.3. Verify Green's Theorem for fe(x dy - y dx)/(x 2 + y 2) where e is the
triangle with verticles at (0, 1), (I, 0), and (1, 1). (Both sides of the equa-
tion should be 0.)

3-2. THE ROTATION INDEX OF

A PLANE CURVE

Suppose that IX: (a, b)-+ R3 is a plane curve. We can choose coordinates
in R3 so that this plane is given by z = 0. We might as well assume that
IX: (a, b)-+ R2, which we do.
Because R2 has dimension 2 and the tangent vector field of a regular
curve is a nonzero vector at each point, we may define the normal to a plane
curve by using the concept of orientation (Section 1-3). This definition has
the advantage of giving a globally defined normal vector field along the curve,
as opposed to the normal of Chapter 2, which was only defined when K 0.*
Note that a similar approach for general curves in R 3 would not work because
giving one vector in a 3-dimensional vector space does not uniquely deter-
mine two others to make an oriented basis.
We shall now define the tangent and normal vector fields t(s), n(s) and
the plane curvature k(s). Lemma 2.2 shows how these concepts compare with
the Frenet-Serret apparatus of IX viewed as a curve in R3 ( which happens to
lie in the (x, y) plane R2).

DEFINITION. Let IX(s) be a unit speed C 2 plane curve. The tangent vector field,
t(s), to IX is t(s) = IX'(s). The normal vector field, n(s), to IX is the unique
(unit) vector field n(s) such that {t(s), n(s)} gives a right handed orthonor-
mal basis of R 2 for each s. The plane curvature, k(s), of IX is given by
k(s) = (t'(s), n(s)).

The following lemma gives both an analogue to the first Frenet-Serret

equation and a "concrete" realization of n(s).
Sec. 3-2 The Rotation Index of a Plane Curve 63

LEMMA 2.1. If II is a unit speed plane curve, then

(a) t'(s) = k(s)n(s); and
(b) if ci(s) is written in the form ci(s) = (x(s), y(s)) for some 0, real-
valued functions x and y, then
t(s) = (x'(s), y'(s)) and n(s) = (-y'(s), x'(s)).

Proof· (a) This follows from Lemma 4.1 of Chapter 2 and the fact that t is a
unit vector field.
(b) The formula for t is obvious. Note that m(s) = (-y'(s), x'(s)) is a
unit vector orthogonal to t. Thus we must only show that {t(s), m(s)} is a right
handed basis. Note that (using e 1 = (1, 0), e2 = (0, 1) as usual)
t(s) = x'(s)e 1 + y'(s)e 2
m(s) = -y'(s)e 1 + x'(s)e 2 •
Hence the matrix A = (a'1) for the change of basis is
A = (x'(s) -y'(s)).
y'(s) x'(s)
This has positive determinant and {t(s), m(s)} is a right handed basis. Hence
m(s) = n(s). I

LEMMA
2.2. For a unit speed plane curve II we have that
t(s) = T(s), n(s) = ±N(s)
at all points for which N(s) is defined, ,c(s) = Ik(s) I,n(s) is differenti-
able, and n'(s) = -k(s)t(s) for all s.

Proof· Problem 2.1. I

EXAMPLE 2.3. ci(s) = (r cos (s/r), r sin (s/r)) is a circle of radius r traversed
counterclockwise. t = (-sin (s/r), cos (s/r)), n = (-cos (s/r), -sin (s/r))
and k = 1/r.

EXAMPLE 2.4. ci(s) = (r cos (s/r), -r sin (s/r)) is the same circle traversed
clockwise. t = (-sin (s/r), -cos (s/r)), n = (cos (s/r), -sin (s/r)), and
k = -1/r.

These examples indicate that the sign of k is changed when the direction
of traverse is reversed. The sign of k indicates whether Cl curves in the direc-
tion of n (k > 0) or away from n (k < 0).

DEFINITION. A regular curve P(t) is closed if pis periodic, i.e., there is a con-
stant a > 0 with ll(t) = P(t + a) for all t. The period of p is the least
such number a. Note that t(O) = t(a).
64 Global Theory of Plane Curves Chap.3

LEMMA 2.5. If P(t) is closed with period a and ci(s) is p reparametrized by arc
length, then Cl is closed with period L = I IJ: dP/dt
dt.
Proof:

s(t +a)= f+.ltldt

= f ltldt+ J.'+•11,dt
= l'ltldt
L -t-
= L -f- s(t).
Thus s(t +a)= s(t) +Land
ci(s + L) = ci(s(t) + L) = ci(s(t + a)) = P(t + a) = P(t) = ci(s(t)) = ci(s).
Hence ci(s) is closed. Since a is the least positive number such that
P(t + a) = P(t) for all t,
L must be the least positive number such that ci(s + L) = ci(s) for alls. I
Note that the period of a closed unit speed curve is its perimeter (or
length).

DEFINITION. A regular curve P(t) is simple if either p is a one-to-one function

or if pis a closed curve of period a with P(t 1) = P(t2) if and only if
t 1 - t 2 = na for some integer n.

EXAMPLE
2.6. The unit circle is a simple closed curve of length 2n:
ci(s) = (cos s, sins).

EXAMPLE
2.7. The figure-eight curve is not simple. See Figure 3.1.

~ FIGURE3.1

For a closed curve ci(s) it makes sense to ask how much the unit vector t
rotates as the curve is traversed once in the direction of increasing s. This is
described by 'an angle which, because t(O) = t(L), must be an integral mul-
tiple of 2n. In order to carefully define and find this integer we employ
certain normalizations.
Let II be a closed unit speed curve in the plane. We may choose a right
handed coordinate system in the plane and the starting point of II so that the
Sec. 3-2 The Rotation Index of a Plane Curve &&

image of II lies in the upper half plane and 11(0)= (0, 0). In this case t(O) is
horizontal.
At each point P of the image of II define 8(P) to be the angle between the
horizontal and t, measured counterclockwise with O ~ 9 < 2n. 9 may not
depend continuously on the position of P. However, we can find a continu-
ous function fJ(s)defined for O < s < L which also describes the' angle. (fJ(s)
may be greater than 2n or less than 0.) This is done by breaking [O,L] into
small intervals over which t does not change much and adjusting 9 on each
interval.
More precisely, points s, may be chosen with
0 = So < s 1 < , . , < S8 = L
such that on each segment [s,, s1+ 1J t does one of four things: (1) points
into the upper half plane (0 < D< n); (2) points into the left half plane
(n/2 < D< 3n/2); (3) points into the lower half plane (n < D< 2n); or
(4) points into the right half plane (0 ~ 8 < n/2 or 3n/2 < 8 < 2n).
8 is continuous on [O,s 1] so we set fJ(s)= 8(s) for O < s ~ s1 • We as-
sume by way of induction that fJ(s)has been defined continuously on [O,sk].
Then on [sk, sk+iJthe angle between the horizontal and t is well defined up
to a multiple of 2n. If the angle is forced to lie in a certain interval of
length n ((0, n) for Case (1), (n/2, 3n/2) for Case (2), (n, 2n) for Case (3) and
(3n/2, Sn/2) for Case (4)) it is a continuous function of s. By adding an ap-
propriate multiple of2n it can be made to agree with the known value of fJ(s)
at s = sk. In this manner fJ(s) can be continuously defined on [O,sk+i1 and
eventually on all of [O,L]. Note that fJ(L) need not be 0. However, fJ(L) is an
integral multiple of 2n.
DEFINITION. The rotation index of a closed unit speed plane curve ci(s) is the
integer i. = (fJ(L) - 9(0))/2n = 9(L)/2n.
EXAMPLE 2.8. The curves in Figures 3.2. 3.3, 3.4, and 3.5 have rotation index
1, -1, 0, and 2, respectively. The reader may also want to look ahead to
Example 2.10.
Note that
t(s) = (t(s), e 1)e 1 + (t(s), e2)e2
= (cos fJ(s),sin fJ(s)).

FIGURE 3.2 FIGURE 3.3

66 Global Theory of Plana Curves Chap.3

00
FIGURE 3.4 FIGURE 3.5

Since t is differentiable (x is C 2), both cos (O(s)) and sin (O(s)) are differen-
tiable. Since 8(s) is continuous and both sine and cosine are differentiable,
O(s)is differentiable. Then O'(s) = k(s). (This equality combined with Lemma
2.2 shows that IO'(s) I = ,c(s) for a unit speed plane curve. This was one inter-
pretation of curvature which we promised to make precise in Section 2.3.)

THEOREM 2.9 (Rotation Index Theorem). The rotation index of a simple

closed plane curve ci(s) is ± l.

Comment: This theorem seems almost obvious and without need of a formal
proof. However, a simple curve can be quite complicated, spiraling a lot, and
our intuition based on oval-shaped curves may not hold true. Also, the proof
will give us a glimpse of certain techniques.

Proof· Let L be the period of ci(s). If O ~ u < v < L, let a(u, v) be the unit
vector from ci(u) toward ci(v). Let
a(u, u) = t(u) (and a(O, L) = -t(O) = -t(L)).
Then a is a C2 function in the region A of Figure 3.6. As in the case of (J on
(0, L], it is possible to define a C2 function II on A, where ti(u, v) measures
the angle between the horizontal axis and a(u, v). Note that ti(u, u) = O(u).

B = (0, L) C = (L, L)

A = (0, 0) FIGURE 3.6

Now 2ni.
1
= (J(L) = O(L) - 0(0) = j; (d(J/ds)ds = f. dO. Since
ti(u, u) = O(u),
this last line integral equals J;re
dti.
Sec. 3-2 The Rotation Index of a Plane Curve 67

and
(since ti is C 2).

Thus by Green's theorem (1.3), J:w+ci+JJ dti = 0. Hence

t/11= .bdti + Siledti.

J:udti is the angle through which OProtates as P traverses the image of II
once in the direction of increasings. (See Figure 3.7) Since II lies in the upper
half plane, OP never points downward and the angle of rotation must be 1ee,
with E = ± 1 according to whether II is traversed counterclockwise or not.

FIGURE 3.7

Likewise, Jilc
dti is the angle through which POrotates as P traverses II

once. POnever points upward and the integral is again 1ee.Thus

= 1CE+ 1ee= 21eeand i. = ±1.
211:i., I
The above proof is due to H. Hopf (1935).

EXAMPLE 2.10. Consider the simple closed curve in Figure 3.8. The following
table gives the values of 9 at the labeled points. Hence i,. = 1 for this
curve as 9( V) = 211:.

point A B c D E F G H I J K

8 0 n n 3n 2n Sn 3n 1n 3n Sn 2,r
2 2 2 2 2
L M N 0 p Q R s T u v
3n n n 0 -n -n O n n 3n 2n
T 2 TT 2 T
58 Global Theory of Plane Curves Chap.3

FIGURE 3.8

In the following corollary, and elsewhere, S 1 denotes the unit circle in

the plane: S 1 = {(a, b) E R 2 la 2 + b 2 = 1}.

COROLLARY 2.11. If cx(s) is a simple closed regular plane curve, the tangent
circular image t: [O,LJ --+ S 1 is onto.

Proof: Problem 2.2. I

Quite often it is useful to consider curves that are made up of regular
curve segments joined together. At these junction points t may not be defined.

DEFINITION. A piecewise Ck regular curve is a continuous function «i:

[a, bJ--+ R3 together with a finite set of points {s, IO< i < n} with
a= s 0 < s 1 < ... < sn = b such that «zl 1,.,,.., 1 is a regular Ck curve
segment.

At a junction point cx(s1) of a piecewise regular plane curve «i let

r(s 1)= lim t(s) and t+(si) = lim t(s). The angle from r(s 1) to t+(s,) will be
,-,,- ,-,,·
denoted 40 1• (If ll is closed, M 0 = Mn = angle from t-(sn) to t+(s 0).) Let
Ml, be the ,angle through which t rotates on the segment «il1,,,,,., 1•
DEFINITION. The rotation index of a piecewise regular plane closed curve «i is
n-1 n-1
• I~ Ml,+~ M,
l
.. = 2,r; .
Sec. 3-2 The Rotation Index of a Plane Curve 59

EXAMPLE 2.12. Let cx(t)be the triangle with vertices (0, 0), (2, 0), (I, 1).
(t, 0) O<t<2
cx(t) = 1 (4 - t, t - 2) 2::::;:t < 3
(4 - t, 4 - t) 3 < t<4.
Here s 0 = 0, s, =-,2, s 2 = 3, s 3 = 4. Note that t is not arc length. The
vectors t-(si) and t+(s,) are shown in Figure 3.9. Since t is constant on
2
each segment, t50,= 0. I; M, is the sum of the exterior angles of a
1=0
triangle, and is thus 2n. Hence the rotation index is I.

FIGURE 3.9

Note that for any piecewise C 2 curve cx(s),

t50,= f.,..., k ds and i,. =

f• kds + ·-·
I; M;
1= 0 •
2n

The following proposition is again almost obvious. However, a careful

proof is even more involved than that of Theorem 2.9. One way to prove it
would be to approximate the piecewise smooth curve by a regular curve.
However, development of the necessary ideas would take us too far afield and
so we omit the proof. See H. Hopf [1935).

PROPOSITION 2.13. If cx(s) is a piecewise regular simple closed plane curve,

then i,. = ± I.

There is another "obvious" theorem about simple closed plane curves,

called the Jordan Curve Theorem. It states that any simple closed plane curve
has an "inside" and an "outside." The interested reader can find a careful
proof in Stoker [1969].
60 Global Theory of Plane Curves Chap.3

PROBLEMS

*2.1. Prove Lemma 2.2.

*2.2. Prove Corollary 2.11.
2.3. Let e be the curve made up of a straight line segment and two circular
arcs of radius 2 as in Figure 3.10. Compute the 68, and the 118 1 and

verify that the rotation index is 1.

(0, v'3)

(-1, 0) (1, 0)
FIGURE 3.10

3-3. CONVEX CURVES

A straight line I divides R 2 into two half planes, H 1 and H 2 , such that
H1 UH 2 = R 2 and H 1 n H 2 = I. We say a curve lies on one side of I if the
image of czis completely contained in one of the half planes H 1 or H 2 •

DEFINITION. A regular curve II is convex if it lies on one side of each tangent

line.

ExAMPLE 3.1. The curves in Figure 3.11 and 3.12 are convex, while those in
Figures 3.13 and 3.14 are not convex.

The next theorem will use the concept of monotonicity. A function

s
/(s) is called monotone increasingif s t implies /(s) sf(t). f(s) is mono-

FIGURE 3.11 FIGURE 3.12

Sec. 3-3 Convex Curves 61

FIGURE 3.13 FIGURE 3.14

tone decreasingifs< t implies/(s) > /(t)./is monotoneif it is either mono-

tone increasing or monotone decreasing. We note that if f is differentiable,
/is monotone if and only if f' has constant sign.(/'~ 0 implies monotone
increasing,/' s;:0 implies monotone decreasing. See Problem 3.2.)

'fHEoREM 3.2. A simple closed regular plane curve Cl(s)is convex if and only
if k(s) has constant sign.

Proof: Since k(s) = dO/ds,we must show that (J is monotone if and only if
czis convex. Suppose (J is monotone. If czis not convex there is a point A such
that cz does not lie on one side of the tangent line I at A. Since cz is closed,
there are points B and C of czon opposite sides of I which are farthest from I
(see Figure 3.15). Note that the tangent lines at A, B, and C must be distinct.

FIGURE 3.15

The tangent lines / 1 and /2 at Band C must be parallel to I. If this were

not the case, one could construct a line through B (or C) parallel to I. Since
this line is not tangent to the curve and goes through B (or C), there must be
points of the curve on both sides of the line. There would then be points on
the curve farther from I than B is.
Two of the three points A, B, C must have tangents t pointing in the same
direction. Thus there exists s 1 < s 2 with t(s 1) = t(s 2) and0(s 2) = O(s1) + 2nn.
62 Global Theory of Plane Curves Chap.3

Since (J is monotone, Theorem 2.9 implies n = -1, 0, or I. If n = 0, then

O(s,) = 0(s2 ) and by monotonicity (J is constant on the interval [s,, s 2 ]. If
n = ±1, then (J is constant on [O,s,] and [s2 , L]. In either case one of the
segments of IX between «x{s1) and 1X(s 2 ) is a straight line.
Hence the tangent lines at 1X(s1) and «x{s 2 ) coincide. But /, / 1 , / 2 are
distinct. This contradiction implies IX is convex.
Now suppose IX is convex. If (J is not monotone, there are points
s 1 < s 0 < s 2 with O(s1) = O(s2 ) *
O(s0). We shall prove (J is monotone by
showing that if O(s1)= O(s2 ), then (J(s)= O(s1) for alls between s 1 and s 2 •
If O(s1) = O(s2 ) for some O < s 1 < s 2 :::;;: L, then t(s 1) = t(s 2 ). By
Corollary 2.11, the map t: [O,L]---+ S 1 is onto so there is a point s 3 with
t(s 3) = -t(s 1). If the tangent lines at s 1, s 2 , s3 are distinct, they are parallel
and one is between the others. This can't happen since IX is convex. Thus two
of the lines coincide and there are points A and B of IX lying on the same
tangent line I (see Figure 3.16). We next show that the curve IX is a straight
line between A and B.

FIGURE 3.16

Suppose some point C of the line segment AB is not on IX. Let/' be the
line perpendicular to AB at C. I' is not a tangent line since IX is convex. Thus
I' intersects IX in at least two points, D and E, which lie on the same side of/.
If D denotes the point closer to C, the tangent line at D has at least one of the
points A, B, Eon each side, contradicting IX being convex.
Hence C is on IX and all of AB is also. Thus the tangents at A and B have
the same direction. Then A, Bare 1X(s 1), 1X(s
2) and IX 11,.,., 1 is straight. There-
fore (J is constant on [s1 , s 2], and (J is monotone on [0, L]. I
Includ~d in the proof above is the following result.
CoROLLARY 3.3. Let IX be a simple closed convex curve with horizontal angle
0. If O(s1) = O(s2 ) with s 1 < s 2 , then IX is a straight line segment on
[S1,S2]·
Sec. 3-4 The lsoperimetricInequality 63

The assumption that cz(s) is simple cannot be relaxed in Theorem 3.2.

The curve of Figure 3.5 has 6 always increasing (hence k(s) > 0), but it is
not convex.

COROLLARY 3.4. Let cz(s) be a closed convex curve. If a straight line I inter-
sects the image of czin three points, then the entire line segment joining
these points lies in the image of cz. In particular, if k :;c 0, a straight line
can intersect cz in at most two points.

Proof: Let the three points of intersection be A, B, and F with F between A

and B. I must be tangent to czat F or else A and B lie on opposite sides of the
tangent line at F, a contradiction. I must also be tangent at A and B or else
there would be points of czon both sides of I. But then the proof of Theorem
2.2 shows that the entire straight line segment from A to B is in the image of
cz. Finally, if k :;c 0, cz has no straight segments (Corollary 3.3) and I can
intersect the image of cz in at most two points. I

This corollary is true for nonclosed convex curves.

PROBLEMS

3.1. A piecewise C 1 curve may have -two tangents at each junction point.
Define a piecewise C 1 curve to be convex if it lies on one side of each
tangent line (and on one side of each tangent line at a junction point).
Give an example to show that Theorem 3.2 is false for piecewise C2
curves.
*3.2. Prove that a differentiable function/ is monotone increasing if/' :;;:::0
and monotone decreasing if/' s;;0. (Hint: Mean Value Theorem.)

3-4. THE ISOPERIMETRIC INEQUALITY

One of the standard problems of early calculus is to find the rectangle (or
possibly a triangle) offixed perimeter bounding the greatest area. The answer,
of course, is a square (or equilateral triangle). Quite often the student is told
that of all geometric figures with a fixed perimeter, the circle bounds the
greatest area. We now give a proof of this fact for regions bounded by regular
curves. The proof requires the following result.
LEMMA4.1. If czis a simple closed plane curve whose image bounds a region
CR,and which is traversed counterclockwise, then the area of CRis
J. J.
x dy = - y dx, where x and y are the coordinates of the plane.
Proof: Problem 4.1. I
64 Global Theory of Plane Curves Chap.3

THEOREM(Isoperimetric Inequality). Let exbe a simple closed regular plane

curve of length (perimeter) L. Let A be the area of the region bounded
by 11. Then L2 > 4nA with equality if and only if II is a circle. Thus, of
all curves of fixed length L, the circle bounds the greatest area.

Proof: Let / 1 , / 2 be two parallel lines tangent to II with II bounded between

them. Let p be a circle tangent to / 1 and / 2 which does not intersect II and let
r be its radius. Choose coordinates x and y for the plane with the origin at
the center of the circle and with the y-axis parallel to / 1. Let / 1 be tangent to
II at A= cz(O)and / 2 at C = cx(s2 ). See Figure 3.17. The key idea of the proof

FIGURE 3.17

is to compare the area of the region bounded by II with the area of the circle
of radius r. Since both the length and the area enclosed by II are indepen-
dent of parametrization we may assume that II is a unit speed curve. 11
may then be written in the (x, y) coordinates as 11(s)= (x(s), y(s)) where
(x'(s)) 2 + (y'(s)) 2 = 1. The curve p may be parametrized by P(s) = (z(s), w(s))
where
z(s) = x(s)
(4-1) -,.Jr2 - x2 O< s < s2
w(s) ={
,.Jr2- x2 s2 ~s<L.
Note that sis arc length on II but not on p. In fact, with this parametrization
p may not be regular, but it is C 2 •
Sec. 3-4 The /soperimetricInequality &&

By Lemma 4.1 the area bounded by czis A= J.x dy = ft xy' ds. The
area bounded by P is nr 2 = -f 11y dx = - ft wz' ds = - ft wx' ds. Thus
A + 1er2 = f:(xy' - wx') ds s foLIxy' - wx' I ds

= foLl<(x',y'),(-w,x))lds.

The Cauchy-Schwarz Inequality (Lemma 1.2 of Chapter I) together with the

fact that czis unit speed tells us that
l((x',y'), (-w, x))I s l(x',y')ll(-w, x)I = ,v'w 2 + x 2 = r
(see Equation (4-1)). Thus

A + 1er :$; foLI<(x',y'), (-w,

2 x)) IdsS r f ds = rL.

We have thus shown that

(4-2) A+ nr 2 s rL.
If we let a = A and b = nr 2 and apply Problem 4.2, we obtain

(4-3) ,./Aiir2s A i n,2 s 'f •

where the second inequality comes from (4-2). Hence Anr 2 r 2L2/4 or s
L2 ~ 4nA, which is the Isoperimetric Inequality.
Now assume that L2 = 4nA. We shall show that cz must be a circle. We
first show that x = ry'. Since L2 = 4nA, inequality (4-3) becomes an equality.
An application of Problem 4.2 yields A = nr 2 • Inequality (4-2) must also be
an equality. Considering the derivation of (4-2) we must have equality where
the Cauchy-Schwarz Inequality was used. By Lemma 1.8 of Chapter 1 there
is a real number c such that
(4-4) (-w, x) = c(x', y').
Taking the length of each side of (4-4) yields
,v'w 2 + x 2 = lcl,v'(x') + (y') = lcl
2 2

and so (4-1) implies that c = ±r. On the other hand, (4-4) also says that
c = <(x', y'), (-w, x)) which (since the first inequality in the derivation of
(4-2) must also be an equality) is nonnegative. Hence c =rand (4-4) shows
that x = ry'.
A = 1er2 implies that r depends on A and not the choice of / 1. Thus if
lines / 3 and /4 orthogonal to 11 and / 2 are used, a circle of radius r is tangent
to them. If coordinates x and y with origin at the center of this circle are used
(see Figure 3.17), we would derive x = ry',Since the y-axis is parallel to and
in the direction of the x-axis and the x-axis is parallel to and in the direction
66 Global Theory of Plane Curves Chap.3

of -y there are constants d and e such that i = y - d, y = e - x. Thus

y - d = i = ry' = -rx'
and
x2 + (y _ d)2 = (,y')2 + (-rx')2 = ,2((x')2 + (y')2) = ,2.
Therefore ot is a circle of radius r centered at (0, d) in the (x, y) coordinate
system. I

This proof, which is due to E. Schmidt [l939], and another, which in-
volves Fourier series and is due to Hurwitz (l902), may be found in Chern
[1967].

PROBLEMS

*4.1. Prove Lemma 4.1. (Hint: Green's Theorem.)

*4.2. Let a and b be positive numbers. Prove that ../ab<f(a + b), with
equality if and only if a = b.

3-5. THE FOUR-VERTEX THEOREM

We shall now consider a certain class of convex curves-those with no

straight segments or isolated points where k = 0.

DEFINITION.An oval is a regular simple closed plane curve with k > 0.

Note that an oval is convex since k does not change sign.

DEFINITION.A vertex of a regular plane curve is a point where k has a relative

maximum or minimum.

The concept of a vertex does not depend on the parametrization of a

curve, as you will show in Problem 5.3. Vertices may be found when k is a
function of some arbitrary parameter, much as in Section 2-6.

EXAMPLE5.1. The (non-unit speed) curve ot(t) = (2 cost, sin t) is an ellipse.

It has exactly four vertices given by t = 0, n/2, n, 3n/2. See Problem 5.2.

LEMMA5.2. If I is a line in R 2 , then there are a, c E R 2 with c =;eOsuch that

z E I if and only if (z - a, c) = 0.

Proof· Let I be the line given by ot(t) = z 0 +

tv where O =;ev = (v 1 , v 2 ) E R2 •
Then a= z 0 and c = (-v 2 , v 1) give the desired result. I

THEOREM5.3 (Four-Vertex Theorem). An oval ot(s) has at least four vertices.

Sec. 3-5 The Four-Vertex Theorem 87

Proof: Since cz is C 3 , k' = 0 at each vertex. If k is constant on any segment,

then every point on this segment is a vertex and we are finished. We may
therefore assume that czhas no circular arcs (and since k > 0, no straight line
segments either) and that there are distinct vertices A and B where k takes on
its global maximum and minimum. Assume also that A = Cl(O).We now show
that the assumption that these are the only vertices leads to a contradiction.
Because vertices come in pairs this will prove the theorem.
Let I be the straight line joining A and B. Since we have assumed that
there are exactly two vertices, k' is positive on one segment of czand negative
on the other. See Figure 3.18.

<•- a,c) > 0

(«-a,c) < 0

FIGURE 3.18

Lemma 5.2 says that there are constants a and c =I= 0 such that z e I if
and only if (z - a, c) = 0. Because k > 0, czmust be convex (Theorem 3.2)
and so (Corollary 3.4) I intersects czat exactly the two points A and B. Hence
(Cl(s) - a, c) is positive on one segment of cz and negative on the other.
A case by case study shows that k'(s)(cz(s) - a, c) does not change sign
on cz. (It is nonpositive in Figure 3.18.) At some point k'(cz - a, c) =I= 0
and so

s:
0 =I= k'(cz - a, c) ds = k(cz - a, c)ra - s: k(cz', c) ds

=0+ J:(-kt,c)ds
= J:(n', c) ds

= (n, c)ra = o.
This contradiction implies there are more than two vertices. Since k' changes
sign at a vertex, the number of vertices must be even. Thus there are at least
four vertices. I
66 Global Theory of Plane Curves Chap.3

This theorem was first proved by Mukhopadhyaya [1909] and Kneser

[1912). It is, in fact, valid not only for ovals, but also for any simple closed
plane curve. (See Kneser [1912)or S. Jackson [1944].)The proof we gave is due
to G. Herglotz in a letter to W. Blaschke [1930).
The theorem is false if we omit the hypothesis "closed" (Problem 5.4)
or if we omit the hypothesis "simple" as the following example shows.

EXAMPLE 5.4. Consider the nonsimple plane curve given in polar coordinates,
= l - 2 sin ef,and whose graph is sketched in Figure 3.19.
(r, ef,),by r

B (0, -3)
FIGURE 3.19

In rectangular coordinates this is the curve

P(ef,)= (cos ef,- 2 sin ef,cos ef,,sin ef,- 2 sin 2 ef,,0).
We shall show that p has only two vertices. Since the angle (J between the
horizontal and t increases as ef,increases, k > 0 and hence k = K, We
compute K via Proposition 6.1 of Chapter 2.
p= (-sin ef,- 2 cos ef,+ 2 sin ef,,cos ef,- 4 sin ef,cos ef,,0)
2 2

p= (-cos ef,+ 8 sin ef,cos ef,,-sin ef,- 4 cos ef,+ 4 sin ef,,0)
2 2

js x p= (0, 0, 9 - 6 sin ef,)

Ijs X PI= 9 - 6 sin ef,
IjsI = (5 - 4 sin ef,)112
k = K = (9 - 6 sin ef,)(5- 4 sin ef,)-31 2
k' = k~ = (24 cos ef,- 12 cos ef,sin ef,)(5- 4 sin ef,ts12 (~) •

We see that k' = 0 only when

(24 cos ef,- 12 cos ef,sin ef,)= 12(2 - sin ef,)cos ef,
Sec. 3-5 The Four-Vertex Theorem 89

is zero. That is, when cos 4,= 0, or 4,= n/2, 3n/2. Thus IShas only two
vertices, which are marked A and Bin Figure 3.19.
Let czbe an oval and Pa point on cz.By Corollary 2.11, there is a point
P where the tangent tis opposite to that at P (i.e., t(P) = -t(P)). The tangent
lines at P and P are parallel. By the reasoning in the proof of Theorem 3.2,
at no other point of czis the tangent line parallel to these two lines. Thus, given
a point Pon an oval, there is a unique point P (called the oppositepoint to
P) such that the tangents at P and P are parallel and distinct.
DEFINITION. The width w(s) of an oval czat cz(s) is the distance between the
tangent lines at P = cz(s)and P. (See Figure 3.20.)

j
FIGURE 3.20 P

DEFINITION. An oval has constant width if the width at Pis independent of

the choice of P.
EXAMPLE 5.5. A circle has constant width.
EXAMPLE 5.6. The "piston" for the Wankel engine gives a piecewise dif-
ferentiable curve of constant width (Figure 3.21). To obtain a differenti-
able curve of constant width one can take the set of points a fixed
distance outside the Wankel piston (Figure 3.22). Note this curve is C 1
but not C 2 • (Why?)

FIGURE 3.21 FIGURE 3.22

70 Global Theory of Plane Curves Chap.3

Problem 5.5. gives another example of a curve of constant width.

The next theorem is surprising because it states that the perimeter of an
oval of constant width depends only on the width.

THEOREM 5.7 (Barbier, 1860). If cz is an oval of constant width w, then the

length of czis nw.

Proof: Let IS(s)denote the opposite point to cx(s).The curve IS(s)is not a unit
speed curve (unless czis a circle). Using Lemma 4.1 of Chapter 2, we have
(S-1) IS(s)- cz(s) = vt(s) + wn(s),
where v = (IS - cz,t) and w = (IS - cz,n). Note that the unit tangent to
IS(s)is -t(s) and the unit normal is -n(s).
Let s be arc length along IS,so that dlSfds is the unit tangent, and dif-
ferentiate Equation (5-1) with respect to s.
-t(s) =~= i:~
ds d(cz + vt + wn)
= ds ds

=::ct++ + v't vt' w'n + wn')

= :;ct
+ + + v't vkn w'n - wkt).

Comparing coefficients oft and n, we have -1 = (ds/ds)(l + v' - wk) and

0 = (ds/ds)(vk + w') or (provided ds/ds =t:-0, see Problem 5.7)

(S-2) 1+v'-wk+d 9 =0
ds
and
(S-3) vk + w' = 0.
Since cz has constant width and k > 0, Equation (S-3) implies v = 0.
Then Equation (5-2) yields l + (ds/ds) = wk. Let P = cz(O)and P = cz(s1).
Then
r··(I + ds)
Jo r··
ds ds = Jo wk ds
or

l.. + s.·=··= l"

O
ds
r=O
ds w
O
k ds = w l"O
ds ds = wn.
d9

s~·ds is the arc length of CZfrom p to P. s::;_:·

ds is the arc length of ISfrom p
to P, which is the arc length of cz from P to P. Hence the length of cz is
Jt ds + J:::·tis = wn. I
Sec. 3-5 The Four-Vertex Theorem 71

CoROLLARY 5.8. If tz is an oval of constant width, the straight line joining P

to P is orthogonal to the tangents at P and P.

Proof: In the course of the above proof we showed that " = 0. This fact
coupled with ·Equation (5-1) implies p = tz +
wn and so the vector from P
to Pis wn. I

PROBLEMS

S.1. If tz(s) is an oval, prove that t• is parallel tot at at least four points.
S.2. In Example 5.1 show that k' = 0 only at the given points. Note that tz
is not unit speed.
*S.3. Prove that the concept of a vertex does not depend upon the parametri-
zation.
5.4. Prove that the Four-Vertex Theorem is false if the hypothesis "closed"
is omitted by considering the parabola tz(t) = (t, t 2).
tS.S. (a) Let tz: [O,r]--> R 2 be a unit speed curve segment and let P(s) be
the string involute (Problem 4.39 of Chapter 2)
P(s) = tz(s) + (r O - s)t(s)
where r 0 > r is some constant. Show that the unit tangent to P
at P(s) is orthogonal to t(s).
(b) Let A, B, C be three points in the plane and let tz be a closed piece-
wise C 2 curve with junction points at A, B, C and t+ = -t- at
each point. (A, B, Care called cusps.)Assume that ABis the longest
of the segments of tz. Let D be a point on the tangent line at A as
indicated in Figure 3.23. Define a curve p as a string involute of tz

FIGURE 3.23
72 Global Theory of Plane Curves Chap.3

starting at D: DEis an involute of BA,EFis an involute of fie,

etc. By the first part of the problem, DEand EFmeet the line EB
at right angles so that p is at least C 1 • Note that EB = DA + arc
length AB.Prove JA = DA so that J = D and p is a closed C1
curve. Show that p has constant width.
5.6. Let ct be a plane oval of constant width. Show that the sum of the radii
of curvature (1/k) at opposite points is a constant independent of the
choice of points.
*5.7. (a) Let ct(s) be a unit speed oval of length L. Let (J be the angle be-
tween the horizontal and the tangent t(s) as in Section 3-2. Prove
[O,2n] is a reparametrization. (Hint: (J = f k ds.)
that 0: [O,L]----->
(b) Let y((J) be the same oval parametrized by (Jso that ct(s) = y(O(s)).
Prove that the opposite point to ct(s) is P(s) = y(O(s) + n).
(c) Prove that P(s) is regular.
(d) Prove that ds/ds is nonzero and finite as needed in the proof of
Theorem 5.7.
5.8. Assume that y((J) is an oval parametrized by (J as in Problem 5.7. Let
w(O) denote the width of y at y(O). Prove that Ji"w d(J = 2L, where
L = length of y. (Hint: Equation (5-2).)
5.9. Let y be an oval parametrized by 0. Prove that
d2w I I
dU2+ w = k(O) + k(8 + n)
(Note that k((J + x) is the curvature at the point opposite to y(O).)

3-6. A PREVIEW

In this section we shall state two theorems about space curves that have
the same flavor as those of the preceding sections. However, we have to
delay their proofs until Chapter 5 so that we may use certain ideas from the
theory of surfaces, namely the concept of curves of shortest length on spheres
and integrals over surfaces.
J;
For a closed plane curve k ds measured the rotation index:

f k ds = 2ni,..

For a space curve the analogous object to study is J;

K ds. For a closed
space curve J;K ds > 0 as K > 0, and somewhere K > 0. The analogous
statement to Theorem 2.3 is
THEOREM 6. I (Fenchel). If ct is a closed unit speed curve of length L, then
f; K ds ~- 211:with equality if and only if ct is a convex plane curve.
Sec. 3-6 A Preview 73

A remarkable extension of this theorem was proved independently by I.

Fary (1949) and J. Milnor (1950). It depends on the concept of a knot, which
will be defined in Cha.pter 5. For now, think of the special case pictured in
Figure 3.24.

FIGURE 3.24

THEOREM 6.2 (Fary-Milnor). If a closed unit speed curve is knotted, then

Jt,cds
~ 4n.

This seems plausible when you realize how a knotted curve has to twist
about twice as much as an unknotted curve.
 4

Local Surface Theory

In the study of curves it was not hard to discern what the geometry was:
it was the way in which curves twisted in Euclidean space. This idea leads to
the notions of curvature and torsion and a fruitful study of the geometry of
curves. We will now discuss the geometry of surfaces. This is a much deeper
subject on both a philosophical and mathematical level. There have been
many attempts to study geometry including such axiomatic approaches as
Hilbert's [1921, 1956]. We shall not attempt to explain this approach, but
rather just call the reader's attention to the fact that there are approaches to
geometry other than ours. Our viewpoint is that differential geometry pro-
vides a unifying thread to geometry (classical plane, spherical, hyperbolic,
and projective geometry can all be placed within the framework of dif-
ferential geometry) and that the approach of Riemann to geometry is the
appropriate one. For a more abstract interpretation of Riemann's program in
terms of manifolds (as in Chapter 7), see R. Millman and A. Stehney [1973].
Modern geometry was born when Riemann first separated the concept
of geometry from the concept of space. His inaugural lecture at Gottingen,
On the Hypotheses which Lie at the Foundation of Geometry (1854), began:
"As is well known, geometry presupposes the concept of space, as well as
assuming the basic principles for construction in space.... The relationship
between these presuppositions is left in the dark." In this chapter we shall
74
Chap.4 Local Surface Theory 76

first define space (the notion of surface). This will take the first two sections.
We are motivated by the idea that a surface should look "locally" like a piece
of the Euclidean plane. Then, since our development of surfaces is based upon
the study of curves in a surface, we establish the notion of arc length and
initiate a study of the curvature of a curve on a surface in Sections 4-3 and
4-4. Having done this, we will discuss in Sections 4-5 and 4-6 what the appro-
priate notion of "straight line" is in the setting of surfaces. We do this by
listing in Section 4-5 the various properties of straight lines in Euclidean space
and taking as a definition the idea that the curve does not bend except as the
surface bends. We then explore which of the other properties of straight lines
in Euclidean space carry over to this setting. Agreeing on a definition of a
straight line is, in Riemann's words, agreeing on "constructions in space."
The actual terminology for these generalized "straight lines" is geodesics.
Next we give a development of the concept of curvature of the surface in
Sections 4-7 through 4-9. We define the curvature of a surface in two ways.
The first way is to break the surface up into (infinitely many) curves and
measure the curvature of these curves. The second way is to measure how the
normal to the surface changes. (Think how we measured the rate of change of
the osculating plane in Chapter 2. We computed B', that is, the rate of change
of its normal.) Section 4-8 has a very extensive set of problems on certain
classical topics which tie together several ideas that have been developed so
far. The chapter ends with optional material on isometrics (length preserving
mappings), the Fundamental Theorem of Surfaces, and an investigation of
surfaces of constant curvature.
A word is due on the meth~d of differential geometry for attacking these
problems. In looking over the first three chapters, the reader should be struck
by the usefulness of linear algebra in attacking geometric problems. What,
after all, are the Frenet-Serret equations but statements about a (nonobvious)
basis for R 3 ? If we are to follow the approach of earlier chapters here, then
we must set up a linear algebraic tool to let us obtain geometric information.
This tool will be the tangent space to a surface at a point. The idea is that the
linear is significantly easier to work with than is the nonlinear. (For students
who have studied differential equations: think how much easier it is to solve
linear ordinary differential equations than nonlinear.)
Another point that should be made while we are speaking in an informal
way is the difference between local and global. In the study of curves we first
introduced only local concepts (in Chapter 2) such as curvature and torsion
and obtained information about the local behavior of the curve. In Chapter
3 we then studied the global behavior of a curve (such as the rotation index
and convexity). We shall adopt the same strategy in our study of surfaces.
We shall first study the local behavior of surfaces, that is, what goes on in a
neighborhood of a given point. This is the content of Chapter 4. We then go
on to the global behavior of surfaces in Chapter 6. The difference between
78 Local Surface Theory Chap.4

local and global can be explained as follows: If you are a creature living on the
2-sphere (i.e., the earth) you can only see a very little bit of it at a time, i.e.,
you can only see "local" things. If, on the other hand, you were an astronaut,
then you would be able to see very global things (such as the shortest distance
between distant points is the distance along great circles).
This last analogy also brings out another point-the difference between
intrinsic and extrinsic. The intrinsic things are the things that an observer who
is on the surface sees. The observer doesn't know (for example) howhis or her
world lies in a bigger universe, or in fact that it doeslie in a bigger universe.
He or she only sees concepts that are independent of the embedding of the
sphere in space. The astronaut, on the other hand, is out in this space and is
cognizant of things such as the normal to the surface while the creature on
earth remains totally ignorant of such things. These are the extrinsic concepts.
It is sometimes quite difficult to tell exactly which concepts are intrinsic and
which are extrinsic. The ninth section of this chapter hasas its main theorem
the highly nonobvious fact (due to Gauss (1827)) that Gaussian curvature is
an intrinsic concept. This last property (being intrinsic) is a very important
one and one that we will turn to in the last chapter of this book when we take
a very intrinsic (and more abstract!) approach to these ideas.
In the differential geometry of surfaces there is an enormous number of
formulas to be derived. We shall emphasize the most important of these
formulas by placing a gray background behind each of them.

4-1. BASIC DEFINITIONS AND EXAMPLES

In calculus we are introduced to the concept of a surface through several
examples: graphs of functions of two variables, surfaces of revolution, and
quadratic surfaces. Here we shall make the concept of a surface more precise
and study various geometric properties of surfaces.
There will be an immediate difference between curve theory and surface
theory. For a given curve there is a natural geometric parametrization (by
arc length) and this parametrization can be used to describe every point on
the curve. For a surface there is no natural geometric parametrization. In
fact, often it will not be possible to find a parametrization that describes the
whole surface, especially in a unique way. As an example, consider the unit
sphere S 2 c R 3 • No matter how we choose a pair of parameters, there will be
at least one point that cannot be described by them. Ordinary latitude and
longitude fails at the poles ((90° N, 30° E) and (90° N, 60° W) are the same
point) and to a lesser extent along the 180° meridian (east or west?). (See
Figure 4.1).
DEFINITION. A subset 'l1 of R 2 is open if for every point (a, b) e 'l1 there is
a number E > 0 such that (x, y) e 'l1 whenever
(x - a) 2 + (y - b) 2 < E2 ,
Sec. 4-1 Basic Definitions and Examples 77

90° N

FIGURE 4.1 90° s

Thus the statement "'ll is open" means that about each point of 'l1 there
is a little disk that is contained in 'll.

ExAMPLB 1.1. 'l1 = {(x,y) E R 2 lx 2 + y2 < 1}is open.

ExAMPLE 1.2. 'U.= {(x, y) e R2 Ix 2 + y 2 s;;:l} is not open.

EXAMPLE 1.3. "l) = {(x,y) E R 2 Ix 2 + y2 > l} is open.

ExAMPLE 1.4. 'W = {(x,y) E R2 I l < x < 2, 3 < y < 5} is open.

EXAMPLE 1.5. oc= {(x, y) E R2 Ix 2 O}is not open.

DBFINmON. A Ck coordinatepatch (or simple surface) is a one-to-one Ck

function x: 'l1 - R3 for some k 2 1, where 'l1 is an open subset of R2
with coordinates u1 and u2 and (oxfou1) x (ox/ou 2) 0 on 'll. *
t.. f .,..,";,~; ... , .. , ,~J
r •:'

Notice how closely this definition resembles that of a (simple) regular

C" curve with the open interval (a, b) replaced by an open set 'l1 and the
regularity condition replaced by (ox/ou 1) x (ox/ou2) 0. We refer to *
1 2 *
(ox/ou) x (ox/ou) 0 as the regularity condition for surfaces. This makes
sense geometrically because a curve czis "one-dimensional" and so the set of
vectors tangent to it at one point should be one-dimensional. Since dcz/dt is
*
tangent to the curve, insisting that dcz/dt 0 (i.e., {dcz/dt} is a linearly
independent set) defines the tangent line to czat a point. A surface is "two-
dimensional" so that the regularity condition for surfaces should be that
we have two linearly independent vectors at each point. According to
78 Local Surface Theory Chap.4

Lemma 3.3 of Chapter 1, (ax/au 1) x (ax/au2 ) *

0 is precisely the condition
that {ax;au1, ax;au2 } is a linearly independent set.
As was the case for curves, we shall assume that k > 3 unless stated
otherwise.

EXAMPLE1.6. Let f (u 1 , u2) be a Ck function of two variables defined in an

open set 'll of R2 • Set x(u 1, u2) = (u1 , u2,f (u1 , u2)). x is Ck and one-to-
one. Let/,= aJ/au'.

;; x ::2 (1, 0,/1) x (0, l,/2)

=

(-/1, -!2, 1) 0.
= *
Thus x is a Ck simple surface. It is the graph of a function and is fre-
quently called a Monge patch.

EXAMPLE1.7. Let 'll = {(u1, u2) E R 2j(u 1)2 + (u2)2 < l} and set
x(u1, u2) = (u1, u2, ,v'l _ (u1)2 _ (u2)2).
This is a special case of Example 1.6. The image is the upper half of the
unit sphere S 2 c R3 •

1.8. Let "() = {(v1,v 2 )

EXAMPLE R2 I(v1) 2 + (v2) 2 < I} and set
E

y(v1, v2) = (vi, -,JI _ (v1)2 _ (v2)2,v2).

This is essentially another case of Example 1.6. The image is the left half
of the unit sphere.

EXAMPLE1.9. Let w = {(w1 , w2 ) E R2 j -n/2 < w 1 < n/2, -1C < w2 < 7C}
and set z(w1, w2) = (cos w1 cos w2, cos w1 sin w2, sin w1). The image of
z is all of S 2 except the 180° meridian.

:: 1 x :: 2 = (-sin w1 cos w2, -sin w1 sin w2, cos w1)

x (-cos w1 sin w2, cos w1 cos w2, 0)
= (-cos 2 w1 cos w2, -cos 2 w1 sin w2, -sin w1 cos w1).
Since cos w 1 is never zero on (-n/2, n/2) and cos w 2 is never zero when
w (az;aw
sin 2 is, 1) x (az;aw 2) *
0 on wand z is a simple surface since
it is easily shown to be one-to-one.

EXAMPLE1.10. Let x: R2 ---> R 3 by x(u 1, u2) = ((u1)2, (u2)2, u1u2 ). Then

ax;au1 = (2u1, 0, u2), ax;au2 = (0, 2u2 , u 1), and
~x ~ = (-2(u 2 )2, -2(u 1)2, 4u 1 u2 ).

At the point (0, 0), (ax/au1) x (ax/au2 ) = (0, 0, 0). Thus x is not a
Sec. 4-1 Buie Definition, and Example, 79

simple surface because the regularity condition is not satisfied. :x:is not
one-to-one either since :x:(l,2) = :x:(-1, -2). If
'U. = {(u 1 , u2) e R 2 lu 1 > 0, u2 > O},
then :x::'U.- R 3 is a simple surface.

We urge the reader to review the chain rule (Section 1-5) before con-
tinuing.

DEFINITION. AC" coordinatetransformationis a Ck one-to-one onto function

/: 'O - 'U.of open sets in R 2 whose inverse g: 'U.- 'O is also of class Ck.

Let v 1, v2 be coordinates in 'O, u 1 , u 2 in 'U.. Set

/(v 1 , v 2) = (/ 1 (v 1 , v 2 ), f2(v 1 , v2)) = (!•, f2).

Similarly, g(u1 , u2 ) = (g 1(u1 , u2), g 2 (u1 , u2)) = (g1.g 2). Then since f and g
are inverses, we have/o g(u1 , u2) = (u 1, u2). Thus
/ (g 1(u 1, u2 ), g 2 (u 1, u2)) = u 1 and / 2 (g 1(u 1, u2), g 2 (u 1 , u2))
1 = u2 •
If we differentiate using the chain rule, we obtain the four equations:
ap ag• ap ag2_ 1 a11ag• ap ag2 _ 0
'Tv'iTut+ Tv2Tut- • av•Tu2+ Tv2Tu2-

Since /'(v 1 , v2 ) = u' and g"(u1 , u2) = v•, we may write the above equa-
tions in either of the shortened forms

(1-1)
or
(1-2) ~ au' av• _ 6'
-~ "Jiji Tti1- /•
(Recall the Kronecker symbol 6 1 = 0 if i *i
while 6i = 1.) Each of the
above forms has its advantages. Equation (1-1) explicitly mentions the trans-
formations f = (/ 1 , /2) and g = (g 1 , g 2), while Equation (1-2) looks more
aesthetic in terms of the chain rule. We will use either (au'/av•), (ap/a'V"),
or J(f) to denote the matrix of partial derivatives of/. It is called the Jacobian
of/and is named after Carl G. J. Jacobi (1804-1851). Note thatJ(/) changes
from point to point of 'O, i.e., it is a matrix-valued function defined on 'O.
Equations (1-1) and (1-2) indicate that J(f) is nonsingular and thus has
a nonzero determinant. An important theorem of advanced calculus (the
Inverse Function Theorem, see Fulks [1969)) states that if/: 'O - 'U. is one-
to-one, onto, differentiable, and has nonsingular Jacobian at each point, then
80 Local Surface Theory Chap.4

the inverse of /is also differentiable and/ is a coordinate transformation. We

shall not, however, need this result.
These coordinate changes play the role in surface theory that repara-
metrizations played in curve theory. The idea is that the upper hemisphere is
the upper hemisphere whether it is parametrized as in Example I. 7 or Example
1.8 (see Example 1.12 below).

LEMMA 1.11. If x: 'l1 - R 3 is a simple surf ace and/: "O- 'l1 is a coordinate
transformation, then y = x o /: "O-+ R 3 is a simple surface with the
same image as x.
Proof: y is one-to-one since both x and fare. Its class is the minimum of
that of x and/. Certainly x('ll) = y("O). Furthermore the chain rule shows
that
(1-3)

By a straightforward computation we obtain

(1-4) oy x ~
avi ov2= det (ap)ax ax
O'ijmTui x au2= det(J(
f »au1
ax x ou2·
ox

Therefore (oy/ov 1) x (oy/ov 2) * 0 and y is a simple surface. I

EXAMPLE
1.12. Let 'l1 and "O be as in Examples 1.7 and 1.8. Let
'U.= {(u 1 , u2) E 'll Iu2 < O} and -G= {(v1 , v2) E "OIv2 > O}.
'U.and 'Oare open. Let/: 'O-+ 'U.be given by
f(v', v2) = (v', -,JI - (v')2 _ (v2)2).
We show that f is a coordinate transformation. f is one-to-one, onto,
and differentiable. Its inverse is g: 'U.--+ 'Ogiven by
g(u 1 , u2) = (u 1 , ,JI - (u 1) 2 - (u2) 2),
which is differentiable. Hence f is a coordinate transformation.
The Jacobian off is
I
( a1·
_. ) = ( v' 0
v2 )
avm .j1 - (v')2 - (v2)2 ,Jl - (v')2 - (v2)2
which has nonzero determinant since v2 ,¢:: 0. Note that in Examples 1.7
and 1.8, using 'U.and 'i5we have
x o f(v 1 , v2) = x(v 1 , -,JI - (v' )2 - (v2)2)
= (v', -,Jl _ (v')2 _ (v2)2, ..J(v2}2)
= (v', -,,/1 _ (v')2 _ (v2}2,v2)
= y(v1, v2).
Sec. 4-1 Ba1ic Definition, and Example, 81

The discussion for the remainder of this section is absolutely crucial for
understanding the rest of this book . We urge the reader to pay especially
close attention to these definitions and notations and understand them in the
special cases of Examples l.14 and I.15.
We shall use the following notation throughout the rest of this book.

NOTATION.lfx: CU- R 3 is a simple surface, then

(1-5)

In this notation Equation (l-3) becomes

au'
Y" = :Ex, oV"
.
DEFINITION.The tangent plane to a simple surface x: CU---+R 3 at the point
P = x(a, b) is the plane through P perpendicular to x 1(a, b) x xi{a, b).
The unit normal to the surface at P is n(a, b) = x 1 x x 2 /Ix 1 x x 2 I,
where the right-hand side is evaluated at (a, b). Note that n(a, b) exists
because x 1 x x 2 =I=0. It is perpendicular to the tangent plane at P.
The set {xi, x 2 , n} is linearly independent and hence gives a basis of R 3 •
In a way it serves the same purpose for surfaces that the Frenet-Serret frame
{T, N, B} does for a curve. However, as we shall see, it is not an orthonormal
basis ( in general). Eventually we shall compute the derivatives of x,, x:!,and
n with respect to the coordinates u'. The resulting expressions, due to Gauss
and Weingarten, will take the place of the Frenet-Serret equations.
The following proposition is immediate from Equation (I-4).
PROPOSITION 1.13. If x: CU---+R3 is a simple surface,/: 'O--+ CUis a coor-
dinate transformation, and y = x of, then
(a) the tangent plane to the simple surface x at P = x(f(a, b)) is equal
to the tangent plane to the simple surface y at P = y(a, b);
(b) the normal to the surface x at Pis the same as the normal to the
surface y at P, except possibly it may have the opposite sign.
This proposition says that the tangent plane at P is an intrinsic invari-
ant . This is especially important for the general theory of surfaces as we shall
see in the next section. See Problem 1.6.
EXAMPLE
1.14. In Example 1.7 we have
x i -- (1O,JI
• ,
1
-u - (u2)2)
- (u1)2 and X2 = (
2
0, I, ,JI - (~)2u - (u2)2) •
82 Local Surface Theory Chap . 4

Therefore n = (u 1 , u 2 , ,Jt - (u 1) 2 - (u2) 2). At the point x(f, !) the

tangent plane has the equation

~ (x- ~)+ 1(y- 1)+ ¥(z - ¥) = 0.

See Figure 4.2.

FIGURE 4.2

EXAMPLE 1.15. Consider the Monge patch x(u 1 , u2) = (u 1 , u2 , u 1u2 ).

x 1(a, b) = (I, 0, b) and x 2(a, b) = (0, I, a)
so that
x1 x x2
_
- (-b, -a, 1) and
_ (-b,
n(a, b) - ,vii+ -a,+I)b
02 2

Note how the normal, and hence the tangent plane , changes as a and
b change. At the point x(l, 2) = (1, 2, 2) the tangent plane has the equa-
tion
-2(x - 1) - (y- 2) + (z - 2) = 0.
See Figure 4.3. In this example the tangent plane actually intersects the

FIGURE 4.3
Sec. 4-1 Basic Definitions and Examples 83

surface in two straight lines:

~(t) = x(t, 2) = (t, 2, 2t) and P(t) = x(l, t) = (I, t, t).
See Problem 1.8.
DEFINITION.A vector X is a tangent vector to a simple surface x: 'll - R3
at P = x(a, b) if X is the velocity vector at P of some curve in x('ll),
i.e., ir"for some f > 0 there is a curve~= (-f, f)- x('ll) c R3 with
~(0) = P, (d~/dt)(O) = X, and 11(1) = x (~'(t), ~ 2 (1)), where
the ~; are C 1 •
LEMMA J.16. The set of all tangent vectors to a simple surface x: 'll - R 3
at P is a vector space.

Proof: If X and Y are tangent vectors at P, then X = ci(O) and Y = P(O)

for some curves~ and Pin x('ll) with ~(O) = lS(O)= P = x(a, b). There are
functions 111 (t), 112 (t), p1 (t), ft 2 (t) such that
~(t) = x(111 (t), 112 (t)) and P(t) = x(ft 1 (t), ft 2 (t)).
(In fact, x- 1 o ~ 1 (t), 11
defines ((% 2 (t)), where x- 1: x('ll)----->'ll is the inverse
ofx.) Then
x= ax d111 (0) + ax d112 (0)
oui dt aii2 dt
(l-6)
y = ax tff£.(0) + ax dft 2 (0).
"iJUfdt aii2 dt
Consider y(t) = x((%1(t) + P'(t) - a, 112 (t) + p (t) -
2 b). Since
111(0) = ft 1(0) = a and 112 (0) =P 2 (0) = b,
y(t) make sense and is in x('ll) if It I is small enough. We claim y(O) = X + Y.
dy _ ax (d111
dt - oui dt +
tff£.)ax (d11 dft
dt + au dt + dt2
2 2)

_ ( ax dll I ax dll 2) ( ax tjJ£_ ax tJ..IE.)

.
+ au dt + oui dt + du dt
- au 1 dt 2 2

By Equation (1-6) y(O) = X + Y. Hence the sum of two tangent vectors at

Pis a tangent vector at P (y(O) = x(a + a - a, b + b - b) = x(a, b) = P).
Assume that X is a tangent vector at P as above and r E R, and
let 1)(t) = ~(rt). For small values of t this makes sense and 1)(t) is a curve
in x('ll). 1)(0) = ~(O) = P. d1t/dt = (d~/dt)r, so that it(O) = rX. Thus a
multiple of a tangent vector is a tangent vector.
All other properties of a vector space are satisfied by the set of tangent
vectors to x at P since the set is a subset of the vector space R3 • Hence the set
of all tangent vectors to x at P forms a vector space. I

The next definition is important because it will give examples of tangent

vectors to a surface x.
84 Local Surface Theory Chap.4

DEFINITION.Let x: 'll--+ R 3 be a simple surface. The u 1-curve, cz, through

P = x(a, b) is given by cz(u1) = x(u 1 , b). The u 2 -curve, p, through Pis
_givenby P(u2 ) = x(a, u 2 ). A parametric curve on xis one of these curves.

Note that the standard rectangular grid on 'll (i.e., the (u 1 , u 2) plane) has
as its image under x a grid on x('ll). This grid is sometimes called a curvilinear
coordinate system on x('ll). The grid lines are the parametric curves. See
Figure 4.4.

:x('lL)

FIGURE 4.4

EXAMPLE1.17. Consider the upper hemisphere in Example 1.7. The

u 1-curve through P = x(a, b) is cz(u1) = (u 1 , b, ~l - (u 1) 2 - b 2). The
u 2 -curve through Pis P(u2 ) = (a, u 2 , ,v'l - a 2 - (u 2 )2). See Figure 4.5.

FIGURE 4.5

Let x: 'll--+ R 3 be a simple surface and (a, b) E 'll. What is the velocity
vector of the u 1-curve cz(u1) = x(u 1, b) at u 1 = a? It is
d«z ax
dul (a) = au1(a, b) = X1(a, ).
b

Similarly the velocity vector of the u 2 -curve P(u 2 ) = x(a, u 2 ) at u 2 = b is

xia, b). This means that both x 1 (a, b) and xia, b) are tangent vectors at
x(a, b).

PROPOSITION 1.18. The set of all tangent vectors to x at P = x(a, b) is a

vector space of dimension two with basis {x1(a, b), xia, b)}. Further-
more, viewed as a plane in R 3 through the origin, it is parallel to the
tangent plane at P.
Sec. 4-1 Basic Definitions and Examples 86

Proof: x 1(a, b) and xz(a, b) are tangent vectors at P and are linearly inde-
pendent. Thus the set of all tangent vectors at P is a vector space (Lemma
1.16) of dimension at least two. It will have dimension two if we can show
that x 1(a, b) and xz(a, b) span it.
Let X be tangent to x at P and let y be a curve in x('U.)with y(O) = P
and y(O) = X. There are functions 7 1(t), 72 (t) such that y(t) = x(7 1(t), 72 (t)).
((71 (t), 72 (t)) is defined by x- 1(y(t)).) By the chain rule we have
dy - ax ~
2dtax ~ - t!1:..
(1-7)
dt - ou1dt + au - L x, dt
Thus
(l-8)
where
X' = 71(0).
Equation (1-8) says that x 1(a, b) and x 2 (a, b) span the set of all tangent
vectors at P. Thus the dimension of this vector space is two.
The tangent plane at Pis perpendicular to n(a, b) by definition. n(a, b),
which is a multiple of x 1(a, b) x xz(a, b), is certainly perpendicular to
x 1(a, b) and x 2 (a, b). Hence n(a, b) is perpendicular to the plane spanned
by x 1(a, b) and x 2 (a, b) and these two planes are parallel. I
We quite often identify the vector space of tangent vectors to x at P
with the tangent plane at P by viewing the tail of a tangent vector to be at P.
In the above proof we showed that any vector X tangent to a surface x
can be written in the form X = :E X'x,. Suppose/: 'O .- 'U.is a coordinate
transformation and y = x of. Then X = :E X"y,,.. How are the coefficients
{X1 , X 2 } and {X1 , X2} related? Since y,,.= :Ex, (au
1/av•),

:E Xix,= X = :E X•y,,.
= :E x·(~x,) = :E ( X"'~)x,.

The vectors x 1 and x 2 are independent and so we must have

(t-9) x1 = ~
~
x· a'V"
au'.
This formula tells us that if we know the components of the vector X in the
'0-coordinate system, we can find the coordinates in the 'U.-coordinatesystem
by multiplying by the Jacobian of/: 'l) .- 'U.:
au• au•)
(1-10) (x•)
X
2
= ("Jvi Tv2(~·)·
au au2x2
2

av• Tv2
Note that we also have
(1-11)
88 Local Surface Theory Chap.4

PROBLEMS

1.1. Let 'll = {(u1, u2) E R 2 I -1t < u• < n, -n < u2 < n} and define
x(u 1, u2 ) = ((2 + cos u 1) cos u2 , (2 + cos u 1) sin u2 , sin u 1).
(a) Prove that x is a simple surface. (x('ll) looks like the surface of a
donut or innertube.)
(b) Compute x 1, x 2 , and n as functions of u• and u2 •
This problem is a special case of the next one, which we shall investigate
in detail throughout the chapter.
*1.2. Consider a curve in the (r, z) plane given by r = r(t) > 0, z = z(t). If
this curve is rotated about the z-axis, we obtain a surface of revolution.
(See Figure 4.6.) We may parametrize this surface in the following

FIGURE 4.6

manner. It is useful to use coordinates t and 8 instead of u• and u2 ,

where t measures position on the curve and 8 measures how far the
curve has been rotated. The surface is given by
x(t, 8) = (r(t) cos 8, r(t) sin 8, z(t)).
(a) Prove that x is a simple surface if the original curve
at(t) = (r(t), z(t))
was regular and one-to-one and if -n < 8 < n by computing
X 1 = ax/at,X 2 = ax/afJ, and D,
(b) Show that Problem 1.1 is a surface of revolution.
(c) Show that Example 1.9 is a surface of revolution.
The t-curves are called meridians and the 8-curves are called circles of
latitude. The z-axis is called the axis of revolution.
(d) What are the meridians and circles of latitude of Problem 1.1 and
Example 1.9?
Sec. 4-1 Basic Definitions and Examples 87

1.3. Let 'U. = {(u1 , u 2 )IO < u 1, 0 < u2 < 2n} and
x(u 1, u 2) = (u cos u2, u 1 sin u2, u 1
1 +u 2 ).

Show that x is a simple surface.

1.4. Let 6(u 1) be a Ck function. Prove
x(u 1 , u 2 ) = (u 2 cos 6(u 1), u 2 sin 6(u 1), u 1)
is a simple surface.
1.S. Let x(u 1, u 2 ) = (u 1 +
u2, u 1 - u2, u 1u 2 ). Show that x is a simple
surface. Find the normal n and the equation of the tangent plane at
u1 = I, u 2 = 2.
*1.6. In Example 1.7 compute the equation of the tangent plane at u 1 = !,
u 2 = -f. Show that this is the same as the tangent plane in Example
1.8 at v 1 = f, v2 = 1/,./2.
*1.7. Prove Proposition 1.13.
*1.8. In Example 1.15 verify that the curves exand , do actually lie in the
tangent plane.
1.9. Let x((J, v) = (cos 6, sin 0, O) v(sin +
cos (J, sin to
sin 6, cos !O)with to
-n < (J < n, -! < v < i· Compute n(O,0) and show that
lim n(O,0) = -lim n({J,0)
e--.. -
while
lim x((J, O) = lim x((J, O).
e--.. 11-.....

See Figure 4.7. This is called the Mobius band.

0 = ±n

FIGURE 4.7 (-) = 0

1.10. Let S2 = {(u, v, w) E R 3 Iu 2 +v +w2 2 = I} and

R 2 = {(u, v, w) E R 3 Iw = 0}.
If (u, v, 0) belongs to R 2 , the line determined by (u, v, 0) and (0, 0, I)
intersects S 2 in a point other than (0, I). Denote this point byo:
x(u, v). Compute the actual form of x(u, v) and show that x: R2 - R 3
is a simple surface. SeeFigure 4.8. (The inverse mapping to xis called
the stereographic projection .)
88 Local Surface Theory Chap.4

(0, 0, l)

FIGURE 4.8

1.11. Let x(r, s) = (r - s, r + s, 2(r 2 + s 2)), for all (r, s) e R 2 • Show x is

a simple surface. Can you describe the surface geometrically?
1.12. Let x(8, if,)= (sin if,cos 8, 2 sin if,sin 8, 3 cos if,) with -n < 8 < n,
0 < if, < n. Show x is a simple surface. What is it?
1.13. Let x(u, v) = (~ cos v, ~ sin v, u) with -1 < u < 1
and -n < v < n. Show that xis a simple surface. What is it?
1.14. Let ex.
: (a, b) - R 3 be a unit speed curve with ,c ::;cO. Let
'U. = {(s, t) e R 2 la< s < b, t-:¢: O}.
Define x: 'U.- R 3 by x(s, t) = cx.(s)+ ta.'(s). Prove that x is a simple
surface, provided x is one-to-one. x is called the tangent developable
surface of ex..

4-2 . SURFACES

The various parametrizations we have given for S 2 (Examples 1.7, 1.8,

1.9, and Problem 1.10) all had one defect in common: they did not describe
the entire sphere . However, we shall want to consider geometric sets such as
S2 as surfaces . To do this will require more sophisticated concepts . Basically,
a surface will be a collection of simple surfaces that overlap. On the overlap
we shall require that the two parametrizations be related by a coordinate
transformation . Because we want to emphasiz.e that a surface is made up of
patches, we shall now refer to a simple surface exclusively as a coordinate
patch . (Both terms mean exactly the same thing .)

DEFINITION.Let M be a subset of R 3 and E > 0. The E-neighborhoodof

Pe Mis the set of all points Q e M such that d(P, Q) < E, where d
denotes the usual Euclidean distance in R 3 •

Note that the e-neighborhood of P is the intersection of M with the ball

in R 3 of radius E and center P. (The student who has studied topology should
note that what we are doing is defining the relative topology of Min R 3 . )
Sec. 4-2 Surfaces 89

DEFINITION. If Mc R 3 , a functiong: M- R 2 is continuousatP e Miffor

every open set 'U.in R 2 with g(P) e 'U.there is an E-neighborhood m.of
P with g(&) c 'U..

Intuitively, a continuous function is one that sends nearby points to

nearby points.

DBFINIDON. A coordinate patch x: 'U.- R 3 is proper if the inverse function

x- 1: x('U.)- 'U.is continuous at each point of x('U.).

EXAMPLE 2.1. The function that sends the open rectangle (0, 2) x (0, 1)
onto the surface in Figure 4.9 is not proper. Any e-neighborhood of the
point P is mapped in a discontinuous fashion by x- 1, if E is sufficiently
small.

FIGURE 4.9
•
DBFINIDON. A Ck surface in R 3 is a subset Mc R 3 such that for every point
P e M there is a proper C" coordinate patch whose image is in Mand
which contains an E-neighborhood of P for some E > 0. Furthermore,
if both x: 'U.- R 3 and y: 'U - R 3 are such coordinate patches with
'U.' = x('U.),'l)' = y('U),theny- 1 0 x: cx- 1('U.' (') '{)')) - (y- 1('U.' (') '{)'))
is a Ck coordinate transformation. (See Figure 4.10.)

'l1
~----y-••x

FIGURE 4.10

In practice one usually gives a collection of proper patches such that

each P e Mis in at least one of the patches. We shall say the patches cover M.
90 Local Surface Theory Chap.4

EXAMPLE 2.2. The unit sphere S 2 = {(x, y, z) E R 3 Ix 2 + y 2 + z 2 = I} is a

surface. We will cover S2 with six patches. (See also Problem 2.3.) Let
'lJ.+= 'U- = 'lJ+ = 'l)- = w+ = w- = {(a, b) E R 2 la 2 + b 2 < I}. The
table gives the various patches.

Patch Definition Inverse

x+: 'U+-+ R3 x+(ul, u2) = (ul, uz, ,v'l - (u1)2- (u2)2) (x+)- 1(x, y, z) = (x, y)
x-: 'U- -+ R3 x-(ul, u2) = (ul, uz, --v'l - (ul)2 - (u2)2) (x-)-l(x, y, z) = (x, y)
y+:'l)+ -+ R3 y+(vl,v2) = (vi, -v'l - (vi )2 - (v2)2,v2) (y+)-l(x, y, z) = (x, z)
y-: 'l)- -+ R3 y-(vl, v2) = (vi, --v'l - (v1)2- (v2)2,v2) (y-)-l(x, y, z) = (x, z)
z+: w+-+ R3 z+(wl,w2) = (,v'l - (w1)2- (w2)2,wl, w2) (z+)-l(x, y, z) = (y, z)
z-: w--+ R3 z-(wl, w2)= (--v'l - (wl )2 - (w2)2,wl, w2) (z-)-l(x, y, z) = (y, z)

The images of the patches are (respectively) the upper, lower, right, left,
front, and back hemispheres. Note every point of S2 belongs to at least
one of these hemispheres. We should show that on each overlap the
appropriate composite function is a Ck coordinate transformation. We
shall only do this for x+ and y+. Example 1.12 did it for x+ and y-. We
leave the other cases to the reader. We omit all+ signs on x+, 'U +, etc.
'U' n 'O' = {(x, y, z) E R 3 IY> 0, z > 0} so that
x-l('U.' n 'O') = {(ul' uz) E 'U Iuz > O}.
Now
y-1 0 x(ur, uz) = y-•(u1, uz, ,JI _ (u1)2 _ (u2)2)
= (u1, ,JI _ (u1)2 _ (u2)2),
which is certainly a Ck function for any k :2:::I since (u 1) 2 + (u 2) 2 < I.
We need also show that y- 1 o xis a one-to-one function on x- 1('U.' n 'O').
If (y- 1 o x)(a, b) = (y- 1 o x)(c, d), then
(a, ,JI - a 2 - b2) = (c, ,JI - c 2 - d 2)
so that a = c and b = ±d. However, since both (a, b) and (c, d) belong
to x- 1('U.' n 'O'), both b and dare positive and b = d. Thus y- 1 o x is
one-to-one. The inverse function is x- 1 o y given by
x-• o y(v1, vz) = (v•, ,JI - (v1)2 - (v2)2)
which is differentiable. Hence y- 1 o x is a coordinate transformation.

EXAMPLE 2.3. We show that the circular cylinder S 1 x (0, I) is a surface.

Let 'U = t-3n/4, 3n/4) x (0, I) and x: 'U -+ R 3 by
x(O, t) = (cos 0, sin 0, t).
xis proper. Let 'O = (n/4, 7n/4) x (0, I) and y: 'O-+ R 3 by
y(,f,, s) = (cos ,f,,sin ,f,,s).
y is proper. Every point of S 1 x (0, I) is in the image of x or y. The
Sec. 4-2 Surfaces 91

overlap consists of two separate pieces, a left piece and a right piece.
Note that on the right piece both (J and if, are between n/4 and 3n/4.
Hence on this piece y- 1 o x(fJ, t) = (0, t), which is certainly a coordinate
transformation. (We may also write this as if, = (J and s = t.) On the
left piece (J lies between -3n/4 and -n/4 whereas if, lies between Sn/4
and 7n/4, so that it is impossible for (J = if,.In fact, since cos (J = cos if,
+
and sin (J = sin if,, we have y- 1 o x(fJ, t) = (8 2n, t) (or if,= (J 2n +
ands= t). This is also a coordinate transformation. See Figure 4.11.

FIGURE 4.11
( - 31! 3
4 ' 4
2:) x (0 l)
'
SI X (0, l)

EXAMPLE 2.4. A surface of revolution (Problem 1.2) can be made into a

surface in a similar manner. See Problem 2.1.

In advanced calculus there is a difficult theorem, which we shall not

prove, called the Implicit Function Theorem. (See Fulks [1969).) It states
that if f: R 3 --,. R is a differentiable function such that (f",h.fz) 0 at all *
points of M = {(x, y, z) If(x, y, z) = O}, then M is a surface. In fact, if
fz* 0 at a point P of M, there is a Monge patch in M that contains P.

EXAMPLE 2.5. Let f(x, y, z) = x 2 +y + z 2 2 - I. Then the set of points

where f = 0 is the surface S 2 •

EXAMPLE 2.6.
x2 y2 z2
f(x,y, z) = 2 + b2 + 2 - I
a c
gives a surface called an ellipsoid.

EXAMPLE 2.7. f(x, y, z) = x2 - y2 - z gives a surface called a hyperbolic

paraboloid.

EXAMPLE 2.8.

gives a surface called a hyperboloid of one sheet.

92 Local Surface Theory Chap.4

ExAMPLE 2.9.

f(x,y,z)=-,-
x2 l.:
b -
z2
- I
2 2
a c
gives a surface called a hyperboloid of two sheets.

PROBLEMS

2.1. The coordinate patch of Problem 1.2 does not cover the entire surface
of revolution-it omits points that would correspond to 8 = ±n.
Define a second coordinate patch with O < if, < 2,r, check the overlap
condition, and thus show that a surface of revolution is a surface as in
Example 2.3.
2.2. Problem 1.1 gave one coordinate patch for a torus (or inner tube; see
Figure 4.12). Find three more patches.making the entire torus a surface
(including the points where u 1 = n and those where u2 = n).

FIGURE 4.12

2.3. Note that every point of S 2 is either in the image of the patch in Example
1.7 or the patch in Problem 1.10. Show that these two patches make S2
into a surface by computing the coordinate transformation on the
overlap.
2.4. Describe some possible parametrizations of the ellipsoid
x2 l.: z2
2a + b2 +2 c = I.
2.S. Give another coordinate patch for the surface in Problem 1.9 so that the
curve 8 = ±n is included, thus making that example into a surface.
Note that Problem 1.9 says that it is impossible to make a choice of unit
normal at each point of this surface in a continuous fashion.
Sec. 4-3 The First Fundamental Form and Arc Length 93

4-3. THE FIRST FUNDAMENTAL FORM

AND ARC LENGTH

In this chapter we are concerned with local properties of surfaces. We

must understand them before we can consider global properties. For that
reason we shall not use the terminology of the last section very much now.
Normally we shall assume we are in a coordinate patch, which is equivalent
to looking at just one small portion of a surface at a time.
One problem will arise by staying in one coordinate patch. Any defini-
tion that seems to depend upon the coordinates will have to be checked as to
its form in another coordinate patch in order to determine if the concept is
really geometric (i.e., independent of coordinate patch). We have already
done this in the case of the tangent plane at a point. (Proposition 1.13.)
Let M be a surface in R 3 and P e M. If X and Y are two vectors tangent
to Mat P, it makes sense to compute (X, Y) . In terms of a proper coordinate
patch x: 'll--. R 3 about P we can write X = I: X'x, and Y = I: Y'xr
Then, since the inner product is a bilinear function,
<X, Y) = I: X'Y'<Xt, x1) = I: X' Y 1g,1,
where we make the definition

(3-1)

The functions g,1 define a symmetric matrix at each point in the image of
x. It is important to note that g,1 is a function defined on 'll and depends
critically on which coordinate patch we have picked. SeeExample 3.1. The
functions g11 are sometimes referred to as the metric coefficients,the coeffi-
cients of the metric tensor, or the coefficientsof the Riemannianmetric. Their
definition is due to K. Gauss (1827) and, in a more abstract formulation, to
G. F . B. Riemann (1854).

DEFINITION . The tangent space of a surface M at P e M is the set TpM of

all vectors tangent to Mat P.

Propositions 1.13 and 1.18 show that this is the same as the tangent plane
at P to any of the coordinate patches whose image contains P. The standard
inner product< , ) of R 3 can be restricted to this vector space, where it is
still an inner product . The matrix (g,1) is the representation of the restricted
inner product with respect to the basis {x,, x 2 }. Thus (g,1) is a nonsingular
positive definite matrix. (Recall that a two by two matrix ( a,1) is positive
definite if and only if a 12 = a21t a 11 > 0, and det (a,,) > 0.)
The rule which assigns to any two tangent vectors X, Y e TPM their
94 Local Surface Theory Chap.4

inner product is called the first fundamental form of the surface. The termi-
nology "form" refers to the fact that an inner product is a bilinear form in the
sense of linear algebra.

EXAMPLE
3.1. Consider the coordinate patch of Example 1.7:
x(ut, u2) = (u1, u2, ,v"l _ (u1)2 _ (u2)2)

X1 = ( 1, 0, ,y"l - (-;;)~!- (u2)2)

-u2 )
X2 = ( 0, 1, ,v"l - (u1)2 - (u2)2 .
Then

g11 = 1 _ (u1)2 _ (u2)2

u1u2
g12 = 1 _ (u1)2 _ (u2)2 = g21
- 1 - (u1)2
g22 - 1 _ (u1)2 _ (u2)2
Thus

Note that this matrix is positive definite since (u 1) 2 + (u 2 ) 2 < I. We

also note that det (g11) = 1/(1 - (u 1) 2 - (u2) 2 ) > O.
In the coordinate patch of Example 1.8 you can compute that the
corresponding matrix is
- 1
(g ..p) = 1 _ (v1)2 _ (v2)2
(1-v1v2(v2)2
Note that this matrix is not found by replacing u 1 and u2 by v 1 and
-,v"l - (v 1) 2 - (v2) 2 as given by f: 'O--+CU..This is because the two
matrices represent the inner product with respect to different bases.
Later in this section we shall determine the actual relation between (g11)
and (K..,).

EXAMPLE 3.2. Consider the Monge patch x(u1, u 2) = (u1, u2, f(u 1, u 2)) of
Example 1.6. x 1 = (1, 0,/ 1) and x 2 = (0, l,/ 2). g11 = 1 + U1) 2,
g 12 = g21 = fi/2, and g22 = 1 + (/2) 2. Also,
det (g,1) = 1 + U1>2 + U2) 2 > o.
EXAMPLE3.3. Consider the surface x(u, v) = (u2, uv, v 2), where u > 0 and
v > 0. Then x 1 = (2u, v, 0) and x 2 = (0, u, 2v). g 11 = 4u 2 + v2,
Sec. 4-3 The First Fundamental Form and Arc length 95

K12= K21= uv, and K22= u 2 + 4v 2. Note that

det (g,1) = 4u4 + 16u2v 2 + 4v4 > 0.

The term "metric" used with the functions g,1 refers to the fact that they
play an important role in measuring. They obviously help to measure lengths
and angles in TPM. We shall see below that they appear in formulas for the
lengths of curves and (later) for areas of regions.
Let cz(t) be a regular curve whose image is contained in that of a coor-
dinate patch x: 'U-----+R3 so that cz(t) = x(ix'(t), ix2(t)) for some real-valued
functions ix'(t) and ix2(t) defined by (ix'(t), ix2(t)) = x-• o cz(t). If we wish to
know the length of the segmentcz: [a, b]-----+ R 3 , we must compute f0bldcz/dtl dt.
But dcz/dt = :Ex, dix'/dt so that

Idcz
dt
I= J\dt/dcz, dcz)
dt

= J(L X,~, L X1 ~/)

I dix' dix'
='VLK,; dt dt.
Hence the length of the segment is J:
,./!:, g,i(dix'/dt)(dix 1/dt) dt. Thusifa
curve segment cz is defined in terms of local coordinates by ix• and ix2, it is
easy (at least theoretically) to compute its length with the use of the metric
coefficients Ki;·

NOTATION.

(3-2)
(3-3)
LEMMA 3.4. For a coordinate patch x: 'U ___.R 3
(a) g = Ix, x x 2 12 ;
(b) g 11 = K22/g,g 12 = g 21 = -g12/g, g22 = K11/g;and
2
(c) for all i and j, ~ g,kgki = ~/.
tt:"1
Proof·
(a) Let 8 be the angle between x 1 and x 2 so that
Ix I x x 2 l2 = Ix 1'2 1x 2 12 sin 2 8 = Ix, 121
X2l2 ( I - cos 2 8)

= Ix, 121x 2 12(1 - lx1l

(X1,X2)2)
2 lx2l2
= lx1 l2 lx2 l2 - (xi, X2)2
= K11K22- K12K21= g.
98 Local Surface Theory Chap.4

(b) A straightforward calculation shows that

g22 _g12)
( g g and Ku K12)
_g;, g~, ~" K22

are inverse matrices.

(c) :E g1kgkl is the (i,j) entry of the product (g11)(gk 1), which is the iden-
tity matrix(~,'). I

ExAMPLE3.5. In Examples 1.7 and 3.1 we have g = 1/(1 - (u 1) 2 - (u2)2)

so that g 11 = 1 - (u 1)2, g12 = g21 = -u 1u2, and g22 = 1 - (u 2) 2•

Suppose now that there is a coordinate transformation /: 'O --+ 'll.

How are the metric coefficients g,1, their determinant g, and the inverse
coefficients gk' in the 'll-coordinate system related to those in the 'O system
for a surface x: 'l1 --+ R 3 ?
Let

Ym-
- av-'
ay
g = det Cir1.,),and (irl) = Cir1.ir)-
1•

Just as Ym= :Ex, (au1/av•) (Equation (1-3)), we have X; = :E Y. (av•/au')

so that

Thus
(3-4)

As a matrix equation this is

gll
(3-5) (
g2l

If J is the Jacobian of/: 'O --+ 'll, Equation (3-5) is

(3-6) (go)= (J-•)•(i.,)J-•
since
Sec. 4-3 The First Fundamental Form and Arc length 97

is the Jacobian of g: 'll - 'U and the inverse of the Jacobian of/: 'U -+ 'll.
(See Equation (1-2) of Chapter 1 and the preceding discussion.)
g = det (g,1) = det ((J- 1)'(i ..pXJ- 1))
= det(J- 1)det(i ..p)det(J- 1) = g(det(J- 1)) 2 •
Thus
(3-7) g=g av")2
(deta,7 ·

Finally
(gkl) = (g,,tt = [(J-l)'(g,.p)(J-1))-1 = J(g,.p)-IJt = J(gr6)J'.

Thus
(3-8)

For future reference we gather together the various formulas associated

with a coordinate transformation /: 'U-+ 'll for a coordinate patch
x: 'll - R 3 •
(3-9)

where X = L X'x, = L X"y,. is a tangent vector to the surface,

- av" avP
(3-10) g,, = L g ..p ail p

(3-11)

and
(3-12)

Interchanging the roles of 'll and 'U, we obtain

(3-9') X" = L X' ~

1 1
(3-10') g- ..p = Lg,,~au aiii
au

(3-11 ')

and
(3 - 12') -ra ~ kl av1ava
g = ~g auk au'·

In Equations (3-9), (3-10), (3-11), and (3-12) there is one fundamental

difference. While all four tell us how to write an expression in the 'll-coor-
98 Local Surface Theory Chap. 4

dinates if we know how to write the analogous expression in the 'l) system,
some involve the Jacobian of/: 'U - 'll ((3-9) and (3-12)) and some involve
the Jacobian of g: 'll - 'U ((3-10) and (3-11)). Classically, transformation
laws (3-9) and (3-12) are called contravariant transformations, while (3-10)
is called a covariant transformation. Transformation (3-11) is neither covari-
ant nor contravariant because it does not involve matrix multiplication. We
shall not belabor the difference between covariant and contravariant. How-
ever, from time to time we may mention whether a given quantity is
covariant, contravariant, or neither. In the abstract approach of Chapter 7
it is easy to distinguish between covariant and contravariant.
Our notation (due to Ricci, Levi-Civita, and Einstein, circa 1900) is
designed to aid in remembering such formulas. Whenever we are summing
an expression, such as :EX'x,, it is with an upper index versus a lower index.
In any transform~tion equation (like (3-9) or (3-10)) the dummy summation
indices are upper versus lower. All other indices appear on both sides of the
equation in the same position (upper or lower). We shall use Latin indices
(i,j, k) in the 'll-coordinate system and Greek indices (iz, p, 7) in the 'l)
system. With these rules in mind it is easy to recall Equations (3-9), (3-10),
(3-11), (3-12). In some texts the summation sign is omitted completely and
one must remember to sum whenever there is a repeated index.

EXAMPLE 3.6. Consider the sphere S 2 and, in particular, the overlap of the
coordinate patches in Examples 1.7 and 1.8. To make the following
computation a little easier to read, we set r = u 1, 9 = u'-,a = v 1, and
b= v2 ,
f(a, b) = (a, -,v'l - a 2 - 62) = (r, 9)
and
g(r, 9) = (r, ,v'l - r2 - 9 2) = (a, b).

The matrix (J(f))- 1 = J(g) is

~r 0
-9 ) .
(
,v'l - r2 - 92 ,v'l - r2 - 92

1 - b2
_ ( 1 - a 2 - b2 1 - aab2 - b2)
(i.,) = ab
1 - a2
1 - a2 - b2 1 - a 2 - b2
r2 + 92
( 92
 The Fim Fundamental Form and Arc Length 99

Thus

,vii - -rr 2 - s2 )
v-•Y<i.,iv-•i -(: -s
,vii - r 2 - s2
r2 + 92 r,v11- , 2 - s 2)
92 92
(
I - r
_s_2_
2

-s0 )
,vll - r 2 - s2
I - s2
= ( I - r2 - s2 I - rrs2 - s 2)
= (g,1)·
rs I - r2
I - r 2 - s2 1 + r2 - 92

DIGRESSION (This material may be omitted.)

If cz(t) is a curve in a simple surface x: 'U.- R 3 , and ifs is arc length

oncz, then
t1s I Ja1dr,,J (ds) 2 dal d(I)
dt = 'Vl; l,1 di di or dt = l; 111di tit'
A classical geometer would drop all reference to the parameter t (and in fact
the curve itself) and write this equation as
(3-13) ds2 = l; 1, 1 du' dul
and call ds 2the first fundamental form. To a classical geometer the symbols
ds, du', dul had meaning as infinitesimals. To a modem geometer they have
meaning as linear functionals, as we shall describe below.
Equation (3-10) can also be derived from (3-13). We must have
ds 2 = l; i .., dv"'dv'.
t1v· = :t cav-1au')du', so that
~ -
~ lrt.p UV
.I •.• .1../1
UV"
~ - av•av1.,_.,
= ~ 1 • 1 aui au} uu·
dul
0

If we accept the fact that the symbols du1du' are independent, then
ds 2 = l; 1 11 du1 du'
must yield
 Local Surface Theory Chap.4

Classically the expressions du1 and du2 expressed the difference in the
coordinates of two infinitesimally close points. The modern viewpoint is to
view them as linear functions on TpM. Before we indicate how this is done, let
us review some more linear algebra.

DEFINITION.A linear functional on a real vector space V is a linear function

p: V--+ R.

3.7. Let V ·-=R 3 and define p by p(x, Y, z)

EXAMPLE = x -·· 2y + z.
EXAMPLE3.8. Let V be any real vector space with an inner product. Let
p E V and define p: V--+ R by p(v) = (p, v).

The set of all linear functionals defined on V forms a vector space under
the usual concepts of function addition and multiplication by real numbers:
(r,f, + IJl)(v}= r(,f,(v}) + IJl(v). This vector space is called the dual space of
V, and is denoted V*.

3.9. If Vis a real vector space of dimension n, then so is V*.

THEOREM

Proof: By the previous observation we need only show that V* has dimension
n. Let {ui, u 2 , ••• , Un}be a basis of V. Define a linear functional ,f,1 : V--+ R
by ,f,1(u1) = 611. Then ,/,'(~ a'u,) = ~ a1,f,1(ui) = ~ al6 11 = a'. We claim
{,f,1, ••• , ,f,n}is a basis of V*. First we show that the ,f,1 are independent. Sup-
pose there are real numbers a, with l; a1,f,t being the zero function. Then
0 = (l; a 1,f,1)(u1) = l; a 1(,f,1(u1)) = ~ a,6 11 = a1. Hence the ,f,1 are indepen-
dent. Now we need to show that they span V*. Let p E V* and let c1 = p(u,).
Then a straightfoward calculation shows that l; c,,/,'(v) = p(v) for any v E V
so that p = ~ c,,f,t.Hence {,f,1, ••• , ,f,n} is a basis and V* has dimension n. I

The basis {,f,1, ,f,2 , ••• , ,f,n}

found in the previous proof is called the dual
basis of V* with respect to the basis {u1, u 2 , ••• , Un}·Suppose one has another
basis {vi, v2 , ••• , vn} of V with u 1 = l; a•,v •. Let {IJl1, IJl2,... , IJlnJbe the
basis of V* dual to {vi, v2 , ••• , vnl· Then IJI• = l; b•,,f,tfor some matrix (b•,).
What is b• 1 '!
b• 1 = l; b•,6 11 = l; b• 1,f,1(u1) = (l; b• 1,f,1)(u1) = IJl•(u1)
= IJI·(~ atJ1vp) = l; atJ11J1•(vp)= ~ atJ16•p = a• 1.
Thus (b• 1) = (a• 1). If u 1 = l; a•,v,., then
(3-14)
If (c'p) denotes the inverse of (a•,), then
(3-15) ,t,1= ~ clplJI/J.
Now we shall indicate how the differentials du1, du2 may be viewed as
linear functionals on TpM. If X = ~ X'x, e TPM, then du·1(X) is defined to
be xi. This is certainly a linear functional, and {du•, du2 } is the dual basis
Sec. 4-3 The first Fundamental Form and Arc Length 101

(of TpM*) with respect to {xi, x 2 }. This is also consistent with the manner in
which dual bases transform:

X1 =L ov"y,.
T'""' while dui = L oui
~ dv".
uu' uv°'
Note that (ou 1/ov") is the inverse of (oi,"/oui) as required by Equation (3-15).
It thus makes sense to call ds 2 = L g 11 dui dul the first fundamental form
because it is a bilinear form:
ds 2 (X, Y) = L g; 1 du1 dul(X, Y) = L g 11 du1(X) dul(Y)
= L gijX y1 = (X, Y).
1

The matrix (glJ) can be viewed as representing an inner product on TpM*,

namely (du 1, du') = gil. This is natural in the following sense. xw = ~ g 11x1
has the property that (XCi>,Y) = du1(Y). g 11 = (XW, X<i>). That is, the inner
product of dui and dul is defined to equal that of the vectors that represent
du1 and dul via the inner product in TpM.

PROBLEMS

*3.1. Show that the metric coefficients for a surface of revolution (Problem
1.2) are given by the matrix
·2
(r
+z ·2 0)
0 ,2 .
3.2. Compute the metric coefficients for the parametrization of S 2 in Prob-
lem I.IO.
3.3. Compute the inverse of metric coefficients in Examples 3.2 and 3.3.
3.4. A coordinate patch x: 'l1----+R 3 is called an orthogonal net if the
u 1-curves meet the u2 -curves at right angles, that is, x 1 is perpendicular
to x 2 everywhere. Prove that the meridians and circles of latitude of
a surface of revolution form an orthogonal net.
*3.5. For a coordinate patch x: 'll-> R 3 show that u 1 is arc length on the
u 1-curves if and only if g 11 = I.
3.6. Let x and y be cartesian coordinates of the plane while rand (J are polar
coordinates. Show that x = r cos(}, y = r sin (J is a C 1 coordinate
transformation for r > 0. Show that the metric coefficient matrix for

the plane with respect to the x, y coordinates is (~ ~). Determine

the metric coefficients for polar coordinates.

3.7. Consider the coordinate patch x(r, 0) = (r cos(}, r sin(}, r) where r > 0.
Find the coefficients of the metric tensor (g11). Now consider the curve
r(t) = e1< 2 , O(t) = t/vT
00 t P>l in this surface, where O < t < n and P
is a constant. Find the length of the curve. Show that p is the angle
between this curve and the line (J = constant.
102 Local Surface Theory Chap.4

4-4. NORMAL CURVATURE, GEODESIC

CURVATURE, AND GAUSS'S FORMULAS

Let y(s) be a unit speed curve whose image lies on a surface M c R 3 • y

has Frenet-Serret apparatus {K,-r,T, N, B}. The reader is warned not to con-
fuse N, the normal vector field to y as defined in Chapter 2, with n, the normal
vector to a coordinate patch in M. They can very well point in different direc-
tions (and usually do!).

4.1. Let M = S2 and let y(s) be the curve

EXAMPLE

y(s) = (¥ cos J'r s,-y-sin J'r s, ¥) ·

Then
y' = T = (-sin J'r s,cos J'r s,0)
T' = (-J'r cos J'r s, -J'r sin J'r s,0)
N= ( -cos J1r s, -sin J1r s, 0).
Ats= n/2J'r, N = (0, -1, 0). Using Example 1.14 we find

n=(o.y,¥) at 1(~)=(0.~,v7)=x(o.-y-)·
See Figure 4.13.

FIGURE 4.13

NOTATION.If x: 'U.- R 3 is a simple surface and y(s) is a unit speed curve in

the image of x, then

(4-1)

(4-2)

(4-3)
Sec. 4-4 Normal Curvature,Geodesic Curvature,and Gauss's Formulas 103

We shall call S the intrinsic normal ofy. It is well defined on a surface M

up to sign, just as n is. The two choices of sign correspond to different orien-
tations. Do not let the terminology mislead you : S is normal to the curve but
tangent to the surface. Note that x,1 = x,1 under the assumption that x is
c2.
For the curve y(s) whose image is contained in the coordinate patch
x : 'll-> R 3 , we write, as usual, y(s) = x(,,•(s), ,,2 (s)). By Equation (1-7) we
have T = y' = I; (ax/au')(.,,')'= I; x,(,,')' or
(4-4) y'(s) = I: xi(,,.(s), ,,2 (s))(,, 1(s))'.
We wish to differentiate this equation with respect to s. What is the derivative
of x,(y1(s), ,,2 (s))? An application of the chain rule yields
dx,(y1(s), y2 (s)) _ ax, ~ + ax, ~
ds - Diii ds Tu'-ds
= X11ds
~
+ x,2 dv
2 <!t..
~ = I: x,, ds .
This equation and (4-4) imply
(4-5) y"(s) = f1
.,
x,,C,,')'(,,')' + ~ x,("')".
"r
If P E M, we may let NPM = {rn Ir E R}. NPM is the set of all vectors
perpendicular to M at P and is called the normal space of M at P. Clearly
R 3 = TPM + NPM, and thus any vector can be decomposed as a sum of a
vector tangent to M at P and a vector normal to M at P . In particular, this
may be done for the vector y"(s):
y"(s) = X(s) + V(s),
where X(s) is tangent to Mand V(s) is normal to M .
Since T(s) is tangent to M, <V, T) = 0. <1", T) = 0 and thus<X(s), T)
= 0 also. But <X(s), n) = 0. Hence X(s) is perpendicular to both n and T
and is thus a multiple of S = n x T. We may then define two functions
Kn(s) and K.(s) by K,,(s) = <1"(s), n(,, 1(s), ,,2 (s))) and K.{s) = <1"(s), S(s))
so that

(4-6)

DEFINmON.The normal curvature of a unit speed curve y is the normal com-

ponent of y" (i.e., is KJ.The geodesic curvature of y is the component of
y" in the direction of S = n x T (i.e., is K.).

In Section 4-7 we will see that the normal curvature helps measure how
M is curving in R 3 • In Section 4-5 we will see that the geodesic curvature
measures how y is curving in M.
104 Local Surface Theory Chap. 4

Note that although K > 0, Kn and K• can be negative. In fact, a change of

the sign of n changes the signs of K,. and K •. Note also that K, K,., and "• are
related by
(4-7)
(See Problem 4.1.)

DEFINITION.The coefficients of the second fundamental form of a simple

surface x: 'll-> R 3 are the functions L, 1 defined on 'll by L, 1 = <x,1, n).
The Christoffel symbols are the functions r,/ (1 < i,j, k < 2) defined on
2
'll by r,/ = ~ <x,j, x,)g'k.
r-1
Since x,1 = x1,, we have L, 1 = L 1, and r,/ = r,/. Proposition 4.2 will
show that the L, 1 measure the normal component of x, 1 while the r,/
measure the tangential components. The reason that the Lil are called the
coefficients of the second fundamental form is that the assignment
II(X, Y) = ~ L, 1X 1 y1 is a symmetric bilinear form (but not necessarily
t:1
positive definite) on TpM just as the first fundamental form
l(X, Y) = <X, Y) = I; g11X ' YI
is. We will study the second fundamental form and its relation to Kn in Section
4-7 .

PRoPOsmoN 4.2. Let x : CU.-> R 3 be a simple surface . Then

(a) (Gauss's formulas)

(4-8)

(b) for any unit speed curve, y(s) = x(y 1(s), y 2 (s))

(4-9)

and

(4-10)

Proof·
(a) Since n, x 1 , x 2 are linearly independent, and hence a basis ofR3, we
may express any arbitrary vector as a linear combination of these three vec-
tors . In particular, there are functions, a,1 and h,/", defined on 'll such that
+
x,1 = a,1n I; b,,"'x,,..Thus L,1 = <xii, n) = a,1 since <x,,.,n) = 0. Also
Sec. 4-4 Normal Curvature. Geodesic Curvature. and Gauss's Formulas 105

<x,i, x 1) = L b,/"g,n1so
that <x,1, x 1)g'k = L b,/"g,,,,g'k. If we sum this
"'
equation over I and apply Lemma 3.4, we obtain
"'

r,/ = '°"'
-6r
<x,,, x,)g'k = '°"'bl/"'g,n1g'k= L b,,"'li,,,k= b,/.
~ ,n

Thus a,1 = Lil and b,/ = r,/, which proves Gauss's formulas.
(b) Substitute Equation (4-8) into Equation (4-5) to get
1" = L xk<t)" + L xkr,/(1')'(11>' + L nL,h'>'<11>'
or
'Y"= L [(yk)" + L r,/(r)'(y1)']xk + (L L,,(y')'(y1)')n.
Since x 1 and x 2 are tangent to the surface, a comparison of this equation with
Equation (4-6) completes the proof. I

From Equation (4-8) we obtain <x,1, x 1) = ~ r,/gkl. This quantity

is classically written ri 11, and is called a Christoffel symbol of the first kind,
while r,/ is called a Christoffel symbol of the second kind. We have no use for
r,Jll and will use the word Christoffel symbol to refer tor,/. These symbols
were first used by G. B. Christoffel (1829-1900) who used { i} for r,/.
The remainder of this section is devoted to the calculations of the
Christoffel symbols in terms of the metric coefficients and several applications
of these calculations.

PROPOSITION 4.3. For a coordinate patch :x.:'l1 -+ R 3 with metric coefficients

gil•

(4-11)

Proof: This proof (which is due to Gauss) makes use of a classical technique
called "cyclic permutation of indices."

(4-12) ~= {;;<x;, x 1) = <x,k, x 1) + (x,, x,k)

(4-13) ~~; = <x,;, xk) + <x,, xk,)

(4-14) °fz1l
= <x1,, xk) + <x1, xk 1).

Since we assume that xis of class C 3 , x,1 = x 1,. Hence, we may combine
(4-12), (4-13), and (4-14) to obtain

(4-15)
108 Local Surface Theory Chap. 4

If we multiply this result by gk' and sum over k, we obtain

2 ~
_!_~ oiil"_ ~g~
gk'(OK1k U
+ alk')
UI
=~
~
(x IJ• ·xk )gkl = r.•J'. I

Note that this formula shows that r,/ is completely determined by the
metric coefficients. We say that the r,l are intrinsic. This means that they can
be determined by measurements within the surface. More generally, an
intrinsic concept is one which depends only on (g,1). If we were two-dimen-
sional beings living on a surf ace, we would be able to compute lengths and
angles and thus the metric coefficients and anything that depends on them.
The intrinsic concepts would be the only geometric concepts we would have.
We would know nothing about a normal vector. (An amusing interpretation
of this is in E. A. Abbott's Flatland, a story about a two-dimensional creature
in a two-dimensional world with insight into the third dimension.)

PROPOSITION 4.4. The geodesic curvature of a surface curve is intrinsic.

Proof: Let E 11 = [n, x,, x1] (see Section 1-3 for triple bracket notation) so
that E11 = E22 = 0 and, by Lemma 3.4, E12 = -E 21 = ,./g.
Since E11 is
either zero or depends only on g, E,1 is intrinsic. Now
"'• = (K 6 S, S) = [K6 S, n, T].
Equation (4-10) shows that
"'• = L (rk" + L r,/'Jl''r'')[xk,n, L :x,r''l
= L <rk"+ L ri,t'Jl''r''>r''[n,x,, xkl
or
(4-16) "'• = L (rk" + L r,/'Jl'''Jl'')'Jl''E,k,
which is intrinsic. I
Although (4-16) gives a formula for"'•• sometimes it is useful to have an
extrinsic formula.
(4-17) "'• = (T', S) = [T', n, T] = {n,T, T'] = (n, T x T')
= (n, T x KN) = K(n, B) = 'ICcos CZ,
where cz is the angle between the unit normal n of the surface, and the
binormal B of the curve. Note that "'• = 'ICcos czmakes sense even when
K = 0 and neither B nor czis defined because "'• = 0 when K = 0.

EXAMPLE 4.5. Let x(u 1, u 2 ) = (u 1, u 2 ,f(u 1, u 2)) be a Monge patch. We

compute some of the coefficients of the second fundamental form and
then some of the Christoffel symbols both extrinsically and intrinsically.
x 1 = (1, 0,/ 1)andx2 = (0, 1,/ 2),n= (-/i, -/ 2, l)/,./I + (/1) 2 + (/2)2,
Sec. 4-4 Normal Curvature, Geodesic Curvature, and Gauss's Formulas 107

x 11 = (0, O,f 11), x 12 = x 21 = (0, 0,! 12 ) and x 22 = (0, 0,! 22). By Gauss's
formulas (4-8), X11= L11D+ r11 1X1 + r11 2X2· Thus
L - (x n)- !11
II·- 11, - ,v'I + U1)2 + U2>2
Considering the first coordinate of x 11, we note

0 = L11<-!1> r
+ U1 )2 + U2) 2 +
I
I II·
,V J
Thus we obtain r 111 = fd 11/(l + (! 1) 2 + (! 2) 2). A similar calculation
yields r1i2 = !zf11/(I + U1>2 + U2) 2).
Alternatively we could computer 112 intrinsically by using Equation
(4-11).
12 -!dz d 22 _ 1 + U1) 2
g = I + U1) 2 + UzY an g - I + U1) 2 + U2) 2
agl 1 _
du1 -
ao+auiU1F> -_ •1
2:/1 11,

and

_ 1-[<
-fdz)(2fd11 - 2fd1 + 2fd11) I
- 2 + U1>2 + U2) 2)
(I
+ 0 + U1) 2)U1d2 + !d21 - 2fd12 + !1d2 + fd21)]
(I + U1) 2 + U2) 2)
_ [-2U1) 2fzf11 + 2f1d2 + 2U1>2!1d21
- 2(1 + U1) 2 + U2) 2)
_ !11!2
- 0 + U1>2 + U2) 2)
Notice how much easier it was to computer 112 extrinsically than intrin-
sically.

EXAMPLE 4.6. Consider the upper hemisphere x(r, s) = (r, s, ,v'l - r 2 - s 2).
Let y(t) = (sin t, 0, cost). This is a unit speed curve in x for -n/2 < t <
n/2.
T = (cost, 0, -sin t}, T' = (-sin t, 0, -cost),
and
n = (r, s, ,,.jl - r 2 - 82).
108 Local Surface Theory Chap. 4

At the point (sin t, 0, CC\St), n is (sin t, 0, cost). Then

'ICn = (T', n) = -sin 2 t - cos2 t = -1.
=
Since 'IC= I, and 'IC2 = 'IC/ + 'IC/, we have 'IC11 0 for this curve. Note
that the image of this curve is part of a great circle.

PROBLEMS
2 = 'ICn
*4.1. Prove 'IC 2 + '/C112.
*4.2. Show that the matrix (L; 1) for a surface of revolution (Problem 1.2) is

~I (
,.;;2+ z2
;z-i; O)·
O ri
4.3. Prove that for a surface of revolution det (L 0 ) = 0 if and only if each
meridian is a straight line.
4.4. Let L,.,,be the corresponding expression to L 0 in a coordinate system
'U. Let/: 'U - 'l1 be a coordinate transformation. Show that
- iJv" iJvfl
Li} = ± L L,.,,T,ii au1,

where the sign is that of det (iJv"/iJu').This shows that up to sign, the
functions L;1 transform covariantly, (i.e., just like g!J).
4.5. In the Monge patch x(u, v) = (u, v, u 2 + v2 ) find the normal curvature
of the curve y(t) = x(t 2 , t) at t = I.
*4.6. The plane is a simple surface: x(r, s) = (r, s, 0). Show that the geodesic
curvature of a curve in the plane is its plane curvature: 'IC• = k.
4.7. The sphere is a surface of revolution. Find the geodesic curvature of a
circle of latitude w1 = constant in Example 1.9.
*4.8. Prove that if x is a coordinate patch of the sphere and n is the unit
normal, then at any point x(r, s), n(r, s) = ±x(r, s) as vectors.
4.9. Let y be a curve on the sphere. Prove 'ICnis constant.
4.10. Let y be a curve on the sphere with 'IC11 constant. Prove T is a circle.
(Hint: Problem 5.3 of Chapter 2 and Problem 4.9.)
4.11. Let f'.,,/ denote the Christoffel symbols for a coordinate system 'U, anc
'l1 a coordinate transformation. Use Gauss's formulas to provt
/: 'U-----+
- ( iJu'iJul iJ2uk) iJvr
r ..,,r= I: I:r,/~TvP + av"'av,, auk·
This formula is the first example of a transformation law that is no·
well behaved (because of the term (iJ2uk/iJv"'iJvfl)).
Sec. 4-5 Geodesics 109

4.12. Prove that

J a(lng) _ r I + r 2 and J (}(Ing) _ r 21 I + r 22· 2
2~- 11 12 2~-

4.13. Does the sign of the geodesic curvature have any meaning in a coor-
dinate patch? In a surface?

4-5. GEODESICS

In plane geometry, the straight lines play a very important role as the
basis for most constructions and for the formation of most figures studied. We
would like to find curves on an arbitrary surface that play an analogous role.
However, there are several properties of straight lines and it is not clear which
is the most important. That is, it is not clear which of these properties should
be taken as the definition of a "straight line" on an arbitrary surface. For
example:
SLI Straight lines have (plane) curvature zero.
SL2 Straight lines give the shortest path between two points.
SL3 Given two points there is a unique straight line joining them.
SU The tangent vectors to a straight line are all parallel.
We shall generalize each of these properties to curves on surfaces and
explore their interrelationships.

DEFINITION.A geodesic on a surface M is a unit speed curve on M with

geodesic curvature equal to zero everywhere.

This is an immediate translation of property SLI above. Since the geo-

desic curvature of a plane curve is its plane curvature (Problem 4.6), this means
that straight lines in the plane are geodesics.

PROPOSITION 5.1. A unit speed curve y(s) in Mis a geodesic if and only if
[n, T, T'] = 0.

Proof: By Equation (4-17), "• = [n, T, T'] is the geodesic curvature. I

PROPOSITION 5.2. Let y(s) be a unit speed curve, x a coordinate patch, and
write y(s) = x(y 1(s),y 2(s)). 'Yis a geodesic if only if

(5-1)

Proof: By Equation (4-10) ,c6 S = I; ('yk" + I; r,/y''yl')xk. Since x 1 and X2

are independent, the result follows. I
110 Local Surface Theory Chap.4

PROPOSITION 5.3. A unit speed curve T(s) on a surface M is a geodesic if and

only if y" is everywhere normal to the surface (i.e., is a multiple of the
normal to M).

Proof: y" = ,cN = "•S + 7'.n. y" 1s normal to the surface if and only if
"• = o. I
EXAMPLE 5.4. Consider the great circle y(s) =(sins, 0, cos s) on S 2 •
y' = (cos s, 0, -sin s), y" = (-sin s, 0, -cos s). Then y"(s) is the
inward pointing normal to S 2 at y(s). By Proposition 5.3, y is a geodesic
since a normal to S 2 at y(s) is ±y(s). Since there is nothing geometrically
special about this particular great circle, every great circle of S 2 is a
geodesic. The converse is true: if y is a geodesic of S 2 then y is an arc
of a great circle (see Problem 5.3).

PROPOSITION 5.5. Let M be a surface of revolution generated by the unit

speed curve (r(t), z(t)). Then
(a) every meridian is a geodesic; and
(b) a circle of latitude is a geodesic if and only if the tangent x 1 to the
meridians is parallel to the axis of revolution at all points on the
circle of latitude.

Proof: M may be parametrized by x(t, 0) = (r(t) cos 0, r(t) sin 0, z(t)) as in

Problem 1.2. Since t is arc length along the generating curve, the metric
matrix is, according to Problem 3.1,
;2 + z2
(g,1) = ( 0

Let u 1 = t, u 2 = 0 so that the Christoffel symbols can be computed by Propo-

sition 4.3: r 12 2 = r 21 2 = ;Jr, r 22 1 = -r;, and all other r 1/ are zero. By
Proposition 5.2, a unit speed curve y(s) = x(t(s), O(s)) is a geodesic if and
only if it is a solution of the differential equations
(5-2) t" - ,;(J'O' = 0

(5-3) 0" + 2(: )t'O' = o.

(a) A meridian is given by O = constant. Then O' and O'' are zero and
(5-3) is satisfied. Along a meridian t = s, so that t' = I and t" 0. Thus =
(5-2) is satisfied. Hence each meridian is a geodesic.
(b) A circle of latitude is given by t = constant. Then t" =-=t' = 0.
Since y(s) = x(t(s), O(s)) has unit speed,

I= IT'(s)l 2 = 1i,,+ ~0'1

= 2
g22(8')2.
We therefore have
Sec. 4-5 Geodesics 111

This last equation implies that O * ()'

= ± l/r. (Note r is constant if t is.)
Thus ()" = 0 and a circle of latitude satisfies (5-3). Since ()' 0 and r > 0, a *
circle of latitude satisfies (5-2) if and only if; = 0. This occurs if and only if
x 1 = (;cos(),; sin(), i) is parallel to the axis of rotation (0, 0, I). See Figure
4.14. I

FIGURE 4.14

EXAMPLE 5.6. Let M be the Monge patch x(u, v) = (u, v, u 2 - v 2 ). Let y(t)
be the (non-unit speed) curve y(t) = (t, 0, t 2) in M. Now

"I' = t dtds and "f" = 1(dt)2

ds
+ t d2t
ds 2
so that [n, y', y") = [n, t, Tl(dt /ds) 3 •
0 = (-2u, 2v, l)
~l + 4u 2 + 4v2
n(u(t) v(t)) =
, ~l + 4t 2
<-
2t, O, l)

t= (I, 0, 2t) and f = (0, 0, 2)

so that [n, t, il = 0. Hence [n, y', y") = 0 and "f is a geodesic by Propo-
sition 5.1.

In the plane, a straight line is determined once a point on the line and the
direction of that line (at that point) are given. The next theorem says that this
is true for geodesics in general.

THEOREM 5.7. Let P be a point on a surface Mand let X be a unit tangent

vector at P. Then, if s0 E R is given, there exists a unique geodesic "f
with "f(s0 ) = P, "f'(s0 ) = X.

Proof- Let x be a patch about P with P = x(O, 0), and X = ~ X'x,. If there is
a geodesic y(s) = X(}' 1 (s), }'2 (s)), then we must have }'k" = - ~ r,/}"'}'J'
with initial conditions }"(s 0 ) = 0, y1'(s0 ) = X 1• Conversely, any solution to
112 Local Surface Theory Chap. 4

this initial value problem is a geodesic with the required properties, if the
solution is a unit speed curve. Picard's theorem (Theorem 5.1 of Chapter 2)
implies there is a unique solution to this initial value problem, for values of s
near s 0 • We must show that the solution y(s) = x(r1(s), 'Jl2 (s)) to the differen-

tial equation is unit speed.

Let/(s) = IT' 12 = l: gur''r''·Then

f'(s) 1' + :Egu'Jl''r

= :E~'Jlk'r''r'' + :Eg,,'Jl'"r 1"
:Eg,,r1,.'rk'r
= 1'r1
' + :Eg,,r,k'rk'ri''1''
+ :Eg,,r'"r''+ :Eg,,r"r''
= :Egu(r'"+ r1k'r1'rk')r1' + :Eg,,Cr'" + rir''rk'>r''·
1, 'Jl2) solves the differential equation. Thus/(s) is con-
which is zero since ('Jl
stant. Since f (s0 ) = IX12 = 1,f must be identically 1. Thus y' is a unit vector
and y is a unit speed curve, hence in this case a geodesic.
Repeated application of this gives y(s) defined for sin some open interval
(a, b), even though the image of y may not be all in one coordinate patch. The
proof shows that y is unique at all points where it is defined. I

EXAMPLE 5.8. Let M be the surface consisting of all the points in R2 except
(2, 0). Let P = (0, 0) and let X = (1, 0) be the unit vector pointing in the
direction of the positive x-axis. The associated geodesic is y(s) = (s, 0)
and is defined for all s < 2. Since y is continuous, it is impossible to
define y(2) as anything but (2, 0), which is not in ¥· It is therefore pos-
sible for the geodesic, which is the solution of the differential equation,
to be defined only for Is I < f < oo.

(The question as to whether a geodesic extends indefinitely, i.e., is de-

fined for all s E (-oo, oo), is metrical. One calls a surface complete if
every geodesic extends indefinitely. A famous theorem due to Hopf and
Rinow says that Mis complete if and only if it is complete as a metric space.
See Hicks [1965].)
The next theorem tells us that geodesics also possess property SL2 of
straight lines. The proof we give is a variant of Gauss's proof of Theorem 5.3.
The technique of the proof is to start with a length-minimizing curve y and
assume its geodesic curvature is not zero. We "wiggle" the curve to form a
family cz,of curves with the same end points as y and with cz0 = y. The function
L(t) which gives the length of cz, must have a minimum at t = 0 so that
L'(O) = 0. This fact together with integration by parts yields a contradiction.
The idea of differentiating with respect to the "wiggle" parameter t is due to
Gauss. A proof like this is called "variational" and is a cornerstone in the
calculus of variations.
Sec. 4-5 Geodesics 113

THEOREM 5.9. Let 1 be a unit speed curve in a surface M between points

P = T(a) and Q = y(b ). If T is the shortest curve between P and Q, then
y is a geodesic.

Proof: Let a< s 0 <band"'• be the geodesic curvature of y. We shall show

that "'•(s0 ) = 0. Suppose "'•(s0 ) =t=0. Then there exist numbers c and dwith
a < c < s 0 < d < b, "'• =t=0 on [c, d], and the image of [c, d] under y con-
tained in a coordinate patch x. Note that the segment of y from y(c) to y(d)
must be the shortest curve joining y(c) and y(d) or else there is a piecewise
regular curve from y(a) to y(c) to y(d) to y(b) that is shorter than y. But y is
assumed to give the shortest curve from y(a) to 1(b).
Let l(s) be a C 2 function defined for c ::;;;s < d such that
A(c)= l(d) = 0, l(s 0 ) =t=0, and l(s)K.(s) > 0 for c < s::;;; d.
(If T is C4, then l(s) = (s - c)(d - s)K.(s) will work .) If S = n x y', then
in the patch x we have l(s)S = I; v'(s)Xi for some v': [c, d]-+ R. Let y(s) be
given by y(s) = x(p 1(s), p 2(s)) and define a family of curves by
cz,(s)= x(p 1(s) + tv 1(s), p2(s) + tv 2 (s)),
where It I is small enough. cz, is a curve from y(c) to 1(d) for each choice oft
with cz0 = y. We may also write cz,(s)= cz(s;t). (See Figure 4.15.)

FIGURE 4.15

The length of cz(s; t) is L(_t)= J: ) 112ds. L(t) has a mini-

(ucz/us, ucz/us
mum fort = 0 since cz(s; 0) = y(s) gives the shortest path .
d
L'(t) = dt c
J.d(acz acz)112
us' Ts
J.da(acz acz)112ds
ds = cut Ts' Ts

= J.d2I 2(:ifs·~) J.d(li~i'~)

(acz acz)112ds = (acz acz)l/2 ds.
c Ts' Ts c Ts., Ts
At t = 0,

(~· ~) = (9s·
~) = l.
114 Local Surface Theory Chap.4

Thus

L'(O) = J"1\osa,
c ot'Ts l ds
oa,)
02
=O

0 a,)lJds
ds\ ot oa,)I
_ f "[.!!_Joa,, _Joa,, 2
- c OS Tt os t=O \ 2 =O

=/oa,,oa,)l
\Tt OS
l"-f"Joa,,0 a.)ld
\Tt ai2 -o s.
=O c c
2

i-l,=o
= :Ev'(s)'.Xt = A(s)S.

But l was constructed so that l(c) = A(d) = 0. Thus

0 = L'(O) = 0- J."<l(s)S, ,c.(s)S + ,cn(s)n) ds

=- r
l(s)7C.(s) ds < o.
This contradiction implies ,c.(s 0 ) = 0. I
The above theorem is true if y is only assumed to be piecewise regular
unit speed (see Hicks [1965]). The converse is false. A geodesic need not mini-
mize distances. Let P and Q be two points on S 2 with P =I=± Q. There are
two geodesics of different lengths joining P to Q, corresponding to the two
arcs of the great circle through P and Q. The longer geodesic does not mini-
mize length.
The next two examples show that, in general, property SL3 of straight
lines is false for geodesics.

EXAMPLE 5.1O. Let M be the surface of Example 5.8. Then there is no geode-
sic joining the points (0, 0) and (4, 0).

EXAMPLE 5.11. Let M be the unit sphere with P and Q the two poles. There
are an infinite number of great circles through the poles. Hence there is
not a unique geodesic from P to Q.

The closest we can come to having property SL3 is the following theorem
which is local in nature (i.e., it says nothing about geodesics between "distant"
points). To get stronger, global theorems requires further topological assump-
tions (e.g., simple connectivity) and further geometric assumptions (e.g., on
curvature). The reader is referred to M. Berger [1965]. The proof of White-
head's theorem below is beyond the scope of this book. It may be found in
Hicks [1965].
Sec. 4-5 Geodesics 116

THEOREM 5.12 (J. H. C. Whitehead, 1932). Let P be a point on a surface M.

Then there is an €-neighborhood 'll of P such that any two points of 'll
can be joined by a unique geodesic of shortest length, and this geodesic is
contained in 'll.

PROBLEMS

5.1. Show that a meridian of a surface of revolution is a geodesic without

solving the differential equations as was done in Proposition 5.5. Also,
determine which circles of latitude are geodesics. (Hint: Proposition
5.3.)
5.2. Let M be a surface and II a plane that intersects Min a curve y. Show
that y is a geodesic if II is a plane of symmetry of M, i.e., the two sides
are mirror images.
*5.3. Combine Problems 4.9 and 4.10 to show that any geodesic on S 2 is a
great circle.
5.4. Let y be a straight line in a surface M. Prove y is a geodesic.
*5.5. Suppose xis a coordinate patch such that g 11 = I and g 12 == 0. Prove
that the u 1 -curves are geodesics. (Such a patch is called a geodesic
coordinate patch.)
t5.6. Prove that if M is a surface of revolution and y is a geodesic, then
r cos ft(s) = constant, where ft(s) is the angle between y'(s) and the
circle oflatitude (of radius r) through y(s).
5.7. Let y(t) be a geodesic not parametrized by arc length. Prove
d 2•yi ~ r I d11i d11k__ =t-
d11 1 d 2t (ds) 2

dt 2 + ~ Jk dt dt -
::.L.. =-- for i = I and 2.
dt ds 2 dt
t5.8. Prove that a regular curve y(t) is a geodesic if and only if
d2-y1~ _ d2-y2r!J.:.. (~ I _ <!t 2) efi._ef:i!_
=0
dt 2 dt dt 2 dt + I; dt r,, dt r,i dt dt - ·
5.9. Let Mbe the Monge patch x(u, v) = (u, v, uv). Show that the non-unit
speed curve y(t) = (t, -t, -t 2 ) is a geodesic when parametrized by
arc length. Can you find other geodesics in M?
5.10. Let «x(s)= (j(s), g(s)) be a simple unit speed plane curve. Let x(s, t)
be the surface x(s, t) = (f(s), g(s), t). (This is called a cylinder over
«x(s).)Let P be a fixed constant and let y(O) = (f(O), g(O), (} tan ft).
Prove y is a geodesic (0 is not arc length). Prove y is a helix.
5.11. Let M be the surface given by x 2 + y 2 - z 2 = I (Example 2.8). Find
as many geodesics as you can.
116 Local Surface Theory Chap. 4

5.12. Find as many geodesics as you can on the Monge patch (u, v, u2 - v 2 ).
(Hint: Problems 5.2, 5.4)
5.13. Find as many geodesics as you can on the surface
2
xz +~ - z2 = I.

t5.14. Let x and y be two different coordinate patches for part of a surface M.
Let X = I: X 1x 1 = I: Xflyfl and Y = I: Y'x, = I: f•y,. be two
vector fields. Define symbols Zk and 2,1by
zk ~ ~~xaul
~
~r ,, kY x 1
1 -1 ~ 1

and
z,--
-
~ af,
avfl
~
x/J+ ~ r •fl,f•xfl•
~

Prove that 2 7 = I: V(av 1/auk). (Hint: Problem 4.11.) This proves

that I: Vxk = I: Z7 y7 defines a vector field Z = VxY, called the
covariant derivative of Y with respect to X. This is one of the most
fundamental concepts of modern differential geometry. It is due to
Levi-Civita (1917) and will be used in Chapter 7.
t5.15. Let x be a patch where g 11 = g 22 = a(u 1) + b(u 2 ) and g 12 = 0. Let
y(s) be a geodesic and let 9 be the angle between y and the u 1-curves
(i.e., between y' and x 1). Prove a sin 2 9 - b cos 2 9 = constant. (Hint:
Equation (4-16).)

4-6. PARALLEL VECTOR FIELDS ALONG

A CURVE AND PARALLELISM

Finally we would like to generalize property SL4 which involves the

concept of "parallel." In plane geometry there are concepts of both parallel
lines and parallel vectors. We shall treat the concept of parallel vectors as
the more primitive of the two. Indeed, there are geometries (e.g., spherical
geometry) in which parallel lines do not exist.

DEFINITION.A vectorfield along a curve y: [a, b]-+ Mis a function X which

assigns to each t E [a, b] a tangent vector X(t) to Mat y(t).

EXAMPLE 6.1. Let y(s) be a unit speed curve in M. Then T(s) is a vector field
along y.

EXAMPLE 6.2. Let y(s) be a unit speed curve in a coordinate patch x. Then
S(s) = n x T is a vector field along y. Since T and S are linearly inde-
pendent, any vector field X along y has the form
X(s) - a(s)T(s) + b(s)S(s).
Sec. 4-6 Parallel Vector Fields Along a Curve and Parallelism 117

DEFINITION. A vector field X(t) along y(t) is differentiable if as a function

X: [a, b]-> R3 it is differentiable.

That dX/dt is normal to the surface means there is no tangential com-

ponent in the change of X as t varies. Thus an "intrinsic" being living on the
surface could detect no change in X along y. The being would view the vectors
X(t 1) and X(t 2 ) (for t 1 near t 2 ) as parallel if dX/dt is normal. Hence the fol-
lowing:

DEFINITION. A differentiable vector field X(t) along y(t) is parallel along y(t)
if dX/dt is perpendicular to M.

EXAMPLE 6.3. Let y(t) = (a(t), b(t), 0) be a plane curve. Let

X(t) = (A(t), B(t), 0)
be a vector field along y. dX/dt = (dA/dt, dB/dt, 0). The normal to the
surface is (0, 0, 1). Hence dX/dt is perpendicular to the surface (plane)
if and only if dA/dt = 0 = dB/dt. Therefore X is parallel along y if and
only if A and Bare constants. This is the usual notion of parallel vectors
in the plane. Note that in this case, the notion of parallel along y is inde-
pendent of y. This is atypical and in fact characterizes the plane locally.

EXAMPLE 6.4. Let M be the unit sphere, y(t) the equator, and X(t) the
unit vector pointing north at each point on T. Then X is parallel since
X(t) = (0, 0, I) and dX/dt = (0, 0, 0) is certainly perpendicular to the
surface.

EXAMPLE 6.5. Let M be the unit sphere, and y(t) the circle of latitude

T(t) = ('{;cost,~ sin t, ~) ·

Let X(t) be the unit vector field that points toward the north pole at each
point of y. Analytically

X(t) = (-ycost,=#sin t,~)

so that
~~ = (J[ sin t, =#cost, o),
which is not normal to the surface. X is not parallel along T.

PROPOSITION 6.6. If y(t) = x(·y1(t),y 2 (t)) is a regular curve in a coordinate

patch x and X(t) is a differentiable vector field along y with X = :E X'x,,
then X(t) is parallel along y if and only if
O = dXk
dt
+ I; r ..kX
I)
1 dy',
dt
k = I and 2.
118 Local Surface Theory Chap.4

Proof: X is parallel if and only if for all /,

0=(~~·
x,)= (Ldi,'x"x,)+ (LX'x,,r,,x,):
that is, if and only if

(6-1) 0= L gll + L (x,,, x,)X' t·

d;,' I= I, 2.

If these equations are multiplied by g 1k and summed over /, the result is

O= L (dX'/dt) 6/ + Lr,/ X' (d71/dt) or
(6-2)

Conversely, if Equations (6-2) are multiplied by gk 1 and summed over k,

the result is (6-1). Thus X(t) is parallel if and only if (6-2) holds. I

Notice that we did not assume that tis arc length along y. Further note
that the differential equation that X must satisfy depends only on the given
curve 1 and the intrinsic quantities r,/. Hence the concept of parallel along
a curve is intrinsic.

THEOREM 6.7. Let y(t) be a regular curve on a C 2 surface M. Let :Xbe a

vector tangent to Mat y(t 0 ) . Then there exists a unique vector field X(t)
that is parallel along y(t) with X(t0 ) = X.

Proof: Let x be a patch about y(t 0 ) . y(t) = x(y 1(t), y 2 (t)). Consider the
initial value problem

d:,k
= - L C/(y 1(t), y 2 (t))X'(t) dJ:' k = I, 2,

Xk(t 0 ) = Xk.
By Picard's Theorem, this has a unique solution for values oft near t 0 • By
Proposition 6.6 this solution is parallel along 1 (where defined) and any
parallel field along 1 whose value at t 0 is X must solve this problem .
Repeated application of this gives a unique X defined along all of y. I

DEFINITION . The unique vector field X(t) parallel along y(t) such that X(t0 )
= X is called the parallel
translate of X along y.

EXAMPLE 6.8. Let M = S 2 and 1 be a circle of latitude q,= q,0 so that

y(t) = (sin q,0 cost, sin q,0 sin t, cos q,0 ). We shall work in the coordinate
patch x(q,, 0) = (sin q,cos 0, sin q,sin 0, cos if,). Let X = x 1(q,0 , 0) so
that X is a unit vector pointing south . We now compute (using
Theorem 6.7) the parallel translate of X.
Sec. 4-6 Pere/le/ Vector Fields Along II Curve and Parallelism 119

Recall that by Problem 3.1

(g,,) = (~
sin~;)
so that r 1l = r 212 = cot;, r 2/ = -sin (JScos;, and all other r,/
are zero. Now y(t) = x(y 1(t), y2 (t)) where 7 1(1) = (JS
0 and 7 2 (t) = t so
that the equations for parallel translation (Equation (6-2)) become

d:• - sin (JS 0 X2

0 cos (JS = 0 and d! 2
+ cot (JSX = o·
0 1

with X 1(0) = I and X 2 (0) = 0 (i.e., X(O)= X). It is easy to check that
X 1(t) = cos ((cos (JS
0 )t) and X 2 (t) = (-sin ((cos (JS
0 )t))/sin (JS
0 are solu-
tions to the system of differential equations which satisfy the initial
conditions . Thus the parallel translate of i is
X(t) = cos ((cos (JS
0 )t)x 1 - sin ((~tpo)t)x 2•
SID o
Note that X(O)=!=X(2n) . See Figure 4.16.

FIGURE 4 .16

PROPOSITION 6.9. Let X(t) and Y(t) both be parallel along a regular curve 1
in M. Then IX(t) I is constant and so is the angle between X(t) and Y(t}.

Proof : Let /(t) = (X(t), Y(t)) . df/dt = <dXfdt, Y) +

<X, dY/dt) = 0 0, +
so/is constant. IfX = Y, this implies IX I is constant . The cosine of the angle
between X and Y is f (t)A XII YI, which is then constant and so is the
angle. I

Comment: This theorem says that when we parallel translate vectors, angles
and lengths are preserved, just as in plane geometry. However, we shall see
120 Local Surface Theory Chap.4

below in Example 6.11 that if two different curves join P to Q and X is a

tangent vector at P, the value of the vector field at Q depends on which curve
is used. This is one of the critical differences between geometry in the plane
and the geometry on an arbitrary surface. It was T. Levi-Civita (1917) who
first realized the importance of parallel translation.
DEFINITION. A regular curve 1(t) on a surface M is maximally straight if
d1/dt is parallel along T·
This is the generalization of property SU of straight lines in terms of the
given notion of parallel.
PROPOSITION 6.10. A regular curve 1(t) on a surface Mis maximally straight
if and only if dt/ds is constant and 1(t(s)) is a geodesic, wheres is the arc
length.
Proof: Assume 1(t(s)) is a geodesic and dt/ds =c. Then
dy/dt = T ds/dt = T/c.
1(t) is maximally straight if d1/dt is parallel along T· d 2 1/dt 2 = T'/c 2 • Since
1(t(s)) is a geodesic T' = 1" is normal to the surface. Thus d1/dt is parallel,
and 1(t) is maximally straight.
Now assume that 1(t) is maximally straight. d 21/dt 2 is normal to
M. We need to show that d_t/ds= constant and dT/ds is normal to M.
d1/dt = (d1/ds)(ds/dt) = T(ds/dt). Since d1/dt is parallel, its length lds/dtl is
constant by Proposition 6.9. Since ds/dt is continuous, ds/dt is a constant :;=0.
Thus dt/ds, which is the reciprocal of ds/dt, is a constant.

!: 9 = = (~)(;!f + (!I)(~;!)·
Since dt/ds is constant, d 2 t/ds 2 = 0 and dT/ds is normal because d 2 1/dt 2
is. I
We might note that if 1(t) is maximally straight, then to an intrinsic
being 1(t) has no acceleration (the acceleration vector d 21/dt 2 is normal to
the surface) and hence is traveling in uniform or "linear" motion. When
differential geometry is applied to physics, and particularly to general rela-
tivity, the concept of uniform motion mentioned in Newton's first law of
motion (the law of inertia) is translated to motion along a geodesic because
the Euclidean notion of straight line has been lost. (See Misner, Thorne,
and Wheeler [1973].)
We shall now give two examples to show that the notion of parallel
translation depends upon the path. In the first example we take two curves
with the same end points and parallel translate a given vector along each curve
and get different answers. In the second example we translate a vector around
a "geodesic triangle" but do not return to the original vector.
Sec. 4-6 Parallel Vector Fields Along a Curve and Parallelism 121

EXAMPLE 6.11. On S2 consider two different meridians T, cz from the north

to the south pole. Let :f be the unit tangent to T at the north pole. If
i is parallel translated along T to the south pole, we get the unit tangent
to T at the south pole since T is a geodesic. If i is parallel translated along
cz,it must keep constant angle with cz', which is also parallel. Let 9 be the
angle between 1' and cz' at the north pole. The angle between the two
translates of X at the south pole is 29. Note that 29 is the area of the
region bounded by T and cz. See Figure 4.17.
ExAMPLE6.12. On S 2 consider a great circle segment T 1 from the north pole
N to point A on the equator, followed by part of the equator 7 2 to B,
then a great circle segment 7 3 back to N. Let i be the unit vector point-
ing south along 7 1 at N. This may be parallel translated along 7 1 to A,
where it points south, then along 7 2 to B, where it points south, and then
along 7 3 to N, where it points south along 7 3 • The final result is a vector
different from :f. In fact, the angle between the two vectors at N is
9 = angle between 7 1 and 7 3 • 9 also equals the area of the region sur-
rounded by 7 1, Tz and 7 3 • See Figure 4.18.

FIGURE 4.17 FIGURE 4.18

PROBLEMS

Equation (4-6) will be useful in many of these problems.

6.1. Let T(t) be a maximally straight curve. Prove there are constants a, b E
R such that s = at + b, where s is arc length.
6.2. Let 7(s) be a unit speed curve on a surface Mand let S be its intrinsic
normal. Prove that S is parallel along T if and only if T is a geodesic.
122 Local Surface Theory Chap.4

6.3. Let XN be the tangential component of the normal vector N of a unit

speed curve 1 on a surface M.
(a) Prove that XN = N - (N, n)n and that XN is a vector field along 1·
(b) Prove that the following are equivalent:
(i) XN = O;
(ii) 1 is a geodesic;
(iii) XN is parallel along 1.
6.4. Let Xa be the tangential component of the binormal vector B of a unit
speed curve 1 on a surface M.
(a) Prove Xa = -(K./K)S.
(b) Prove that the following are equivalent:
(i) Xa = B;
(ii) 1 is a geodesic;
(iii) Xn is not zero and is parallel along 1 ·
6.5. Let 1 be a curve with 1(0) = P, 1(1) = Q. If X e TPM, let X be the
parallel translate of X along 1. Define 1•: TPM - T0 M by 1•(X) =
X(l).
(a) Prove that 1• is a linear transformation.
(b) Prove that 1• is an isometry, that is, that
<1•(X),1•~» = (X, 'i').
(c) Prove that 1• is an isomorphism. 1• is called the parallelism defined
by 1·
(This problem requires an understanding of the proof of Theorem 6.7.)

4-7. THE SECOND FUNDAMENTAL FORM

AND THE WEINGARTEN MAP

In the previous sections we considered concepts related to the geodesic

curvature of a curve 1(s) on a surface M. We now turn our attention to the
normal curvature of the curve. It will tell us how Mis curving in the direction
of T (see Proposition 7.2). In order to study how Mis curving at a point,
without reference to a direction, we define the Weingarten map L. This
map will be essentially the directional derivative of the normal vector. (If
11:M- S2 c R 3 is the normal spherical image given locally by 11(P)= n(P),
then L can be viewed as the derivative of 11.)It will be the eigenvalues of L
at a point which will tell us how M curves at that point, as we shall see in
Section 4-8.
As before, let 1 be a unit speed curve on M whose image is contained
in a coordinate patch x('ll). Let 1(s) = x(7 1(s), 72 (s)). We have (from (4-6)
and (4-8))
(7-1) "(" = 'KnD + K,S
Sec. 4-7 The Second Fundamental Form and the Weingarten Map 123

(7-2) x,,= L,,n + I: r,lxk.

Recall that the L 11 were called the coefficients of the second fundamental
form.
DEFINITION. The secondfundamental form II on M is the bilinear form on
TPM (for each P E M) given by Il(X, Y) = I: L 11X 1 YJ, where
X= I: X'x,,
PROPOSITION 7.1. Let M be a surface. Then
(a) II is a symmetric bilinear form on TPM for each P E M;
(b) if Tis a unit speed curve with tangent T, then"'•= II(T, T);
(c) if cz and p are regular curves with cz(O)= P(O) and whose velocity
vectors are dependent at t = 0, then cz and phave the same normal
curvature at t = 0.

Proof:
(a) It is easy to show that
Il(X 1+ X2 , Y) = II(Xi, Y) + Il(X 2 , Y),
Il(X, Y 1 + Y2) = Il(X, Y 1) + Il(X, Y 2)
and
Il(X, rY) = Il(rX, Y) = rll(X, Y)
(for all r E R and X, Xi, X 2 , Y, Y 1, Y2 E TPM), which is what is meant by
bilinearity. Since L,1 = L 1,, II(X, Y) = Il(Y, X) and II is symmetric.
(b) This is precisely Equation (4-9).
(c) Let T and T denote the (unit) tangent vectors to cz and P at
cz(O)= P(O).Since 4(0) and ~(O)are dependent, T = ± T. Hence, the normal
curvature of cz at t = 0 is Il(T, T) = Il(±T, ±1') = 11(1',T), which is the
normal curvature of pat t = 0. I

PROPOSITION 7.2. Let 1(s) be a unit speed curve in a surface M with normal
curvature "'· at P. Let j be the curve formed by the intersection of M
with the plane IT through P spanned by n and T'· (See Figure 4.19.)
Then I'IC.I is the curvature ic of the plane curve j.
Proof· Part (c) of Proposition 7.1 shows that T and j have the same normal
curvatures. However, j is a plane curve with normal ±n at P. Its plane cur-
vature k satisfies Ik I = ic. On the other hand, k is ±(normal curvature of
j) = ±'IC•. Thus !'IC.I=lkl = ic. I
We now discuss directional derivatives in preparation for the Weingarten
map.
A subset <Rof a surf ace Mis open if for each P E <Rthere is an £-neigh-
borhood of P in M contained in <R.(The image of a proper coordinate patch
124 Local Surface Theory Chap . 4

FIGURE 4.19

is open.) A function /: CR--+ R is differentiable if for every C 1 curve ex(t)

with ex(O)E CRthe derivative (d(f o ex)/dt)(O)exists.

DEFINITION. Let/be a differentiable function defined on an £-neighborhood

of PE M. Let X E TPM and let ex(t) be a curve on M su:h that
ex(O)= P and X = (dex/dt)(O). The directional derivative off in the
direction X is X/ = (d(f o ex)/dt)(O).

Note that X/ is a number for each X e TpM and each differentiable

function/. We will show that X/is well defined (i.e., independent of choice of
ex) and give an easier way to compute ,Vafter an example.
EXAMPLE 7.3. Let
M = S2, P = (), 0, 0), X = (0, I, 0), and f(x,y, z) = x 2 + y.
H we choose ex(s) = (cos s, sin s, 0), then f o ex = cos 2 s + sin s so that
(f o ex)'(s) = -2 cos s sins+ cos s. Hence 'if= (! o ex)'(O)= I.

PROPOSITION 7.4. Let x: 'U --+ M be a coordinate patch for M about

2
P = x(O, 0). H X = I; X'x,, then X/ = ~ X'(a(J o x)/au 1)(0, 0). In
particular, X/ does not depend on the choice of ex such that X = ci(O)
and ex(O)= P.

Proof : We may write ex(t) = x(ii 1(t), ii 2 (t)). Therefore

(! o ex)(t) = (! o x)(ii 1(t), ii 2 (t)).
The chain rule implies
*'(/ 0 ex)(t) = ~ ac~:,
x) ~ •
Sec. 4-7 The Second Fundamental Form and the Weingarten Map 125

Since X' = (da//dt)(O) (Equation (1-8)), the above expression when evaluated
at t = 0 completes the proof. I

Using Proposition 7.4 it is easy to show:

LEMMA7.5. Let f be a differentiable function in an £-neighborhood of

P EM. H X, YE TpM and r ER, then (X + Y)f= X/ Yf and +
(rX)f = r(Xf).

Hf= (f 1, / 2 , / 3 ) is a vector-valued function defined on an open set in

M, we can let Xf = (Xf 1 , X/2, X/3) be the definition of the directional
derivative of a vector-valued function. In particular, this may be done for the
function f = n. (Remember that n is really only defined up to sign: in a
coordinate patch n is determined and if a change of coordinates is made, the
worst that can happen is that the sign of n is changed.)

DEFINmON.The Weingarten map L is, for each P E M, the function

L: TpM--. R 3 given by L(X) = -Xn. (The minus sign is for future
convenience.)
Because n is only determined up to sign, the same is true of L.

PRorosmoN 7.6. Let M be a surface. Then

(a) Lis a linear transformation from TpM to TPM;
(b) If L(xk) = L Dkxl, then nk = L L,kg' 1•
Proof·
(a) It is immediate from Lemma 7.5 that L(X + Y) = L(X) + L(Y)
and L(rX) = rl(X). Thus Lis a linear transformation. We must show that
L(X) E TPM for-each X E TpM. Since Lis linear;we need only cheek this
on the basis {xi, x 2 }. Let cz,be the u'-curve through P. Then
d an
L(x,) = -x,(n) = - du,(n o cz,)= -"Jui·

Since 1 = (n, n), 0 = (antaun)1, and antau'

is tangent to Mat P. Hence

(7-3)

and L(x,) is tangent to M. Note that (7-3) defines n,.

(b) Since x, is tangent to M, (n, x,) = 0. Hence

O = a<j;,x,) = (~. x,) + (n, x,k> = -(L(xk), x,) + L,k

= L,k - (I: Llkx 1, x,) = L,k - I; L'k(x,, x,)
= L,k - L L' kK11·
126 Local Surface Theory Chap.4

Thus
(7-4)

(7-5) I
A classical geometer would say that IJk is obtained from L,k by raising
an index. L (or I}k) is a "tensor oftype(l, })"because it has one upper and one
lower index and behaves in a certain fashion under change of coordinates
(see Problem 7.6). Note that (L'k) is the matrix representing L with respect
to the basis {x1, X2},

PRoPOSITION 7.7. On a surface M we have Weingarten's equations

(7-6)

Proof· By Equation (7-3) L(x,) = -an/au,

so that

n, = t = -L(x 1) = -L Lk 1xk. I

For a surface M, the basis {x1, x 2 , n} ofR 3 plays a role analogous to the
Frenet-Serret frame {T, N, B} of a curve. Gauss's formulas (7-2) and the
Weingarten equations (7-6) give the analogues of the Frenet-Serret equations:
(7-2) Kt, = L,,n + L r,/xk
(7-6) n, = -:E u,xk
EXAMPLE 7.8. Let M = R 2 • We show that L is the zero transformation (at
each point) in two ways. First by the definition of L: n (0, 0, 1) is =
independent of P so that L(X) = 0 for all X E TpM. The second way is
to compute the L,k for M: x(u 1, u 2 ) = (u 1 , u2, 0) and
L,k = (x,k> n) = (0, n) =0
for all i and k. Hence raising an index by (7-5) gives L'k = 0 so that
L=O.

EXAMPLE 7.9. Let M = S 2 and choose n to be the outward pointing normal.

Let X e TpS 2 and choose cz to be a curve on M such that ci(O) = X,
cz(O)= P, and cz(t) = (x(t), y(t), z(t)). At cz(t), n = cz(t) so that
L(X) = -Xn = -(Xx, Xy, Xz)
= -(i(O), j(O), i(O)) = -ci(O) = -X.
That is, L is the negative of the identity. H we had chosen the inward
pointing normal, we would have had L equals the identity.
Sec. 4-8 Princip11/,G11ussi11n,
Me11n,11ndNormal Curv11tures 127

The following result says that L is a self-adjoint (or symmetric) linear

transformation.

LEMMA
7.10. If PE Mand x, YE TpM, then
ll(X, Y) = (L(X), Y) = (X, L(Y)).

Proof: Problem 7.4. I

PROBLEMS

7.1. Show that for M = S 2 , Lis a plus or minus the identity by computing
the L,k in a coordinate patch and raising an index.
7.2. Show that for M = S 1 x (0, 1) (Example 2.3) L can be represented by

the matrix (~ ~) ·

7.3. Find L for the torus in Problem 1.1.

*7.4. Prove Lemma 7.10.
7.5. Find the matrix (L' 1) for a surface of revolution. (Compare your answer
with 7.3)
7.6. Let/: 'U --+ 'l1 be a coordinate transformation. How are the l•, related
to the L' 1 ? Compare your result with Equation (2-1) of Chapter 1.
7.7. Let 7(s) = X('Jl1(s), r 2 (s)) be a unit speed curve in a surface M. Note
that {T, S, n} give a right-handed orthonormal basis of R 3 at each point
of 7. View n as n(s) = n(r 1(s), 'Jl2 (s)) and prove the following analogues
of the Frenet-Serret equations:
(a) T' = Il(T, T)n + ,c6 S;
(b) S' = -,c 6 T + II(T, S)n;
(c) n' = -II(T, T)T - Il(T, S)S.

4-8. PRINCIPAL, GAUSSIAN, MEAN, AND

NORMAL CURVATURES

We now have two ways to measure how a surface is curving. The first is
through the normal curvature of curves. This is aesthetically unsatisfactory
because it forces us to break up a surface into infinitely many curves. The
second measurement involves the change of the normal, which is given by L.
Since L is a linear transformation, there are two associated numerical
invariants: the determinant of (L',) (the Gaussian curvature) and one-half
the trace of (L'1) (the mean curvature).This section is devoted to understand-
ing the relationship between these different kinds of curvatures.
128 Local Surface Theory Chap.4

We know that the normal curvature of y at P depends only on the unit

tangent of y at P. If we knew all the possible values that ,c. takes on at P we
would know. how M curves. One step in this direction would be to find the
maximum and minimum values that 1e. takes on. That is, determine the
maximum and minimum of Il(X, X) as X runs over all unit vectors in TpM.
This means we are maximizing (and minimizing) II(X, X) subject to the con-
straint (X, X) = 1. The method of Lagrange multipliers tells us to find the
critical values of
/(X, .l) = ll(X, X) - .l((X, X) - 1) = (L(X), X) - .l(X, X) + .l
= (L(X) - .lX, X) + .l
atP.
In terms of a coordinate patch x we consider the function
f(X 1, X", .l) = .t + I: (L', - .l6',)XIX1t:g,1c,
Following the method of Lagrange multipliers we must have of/oX 1 = O,
of/oX" = 0, and of/o.l = 0. Now of/o.l = (-X, X) + 1, so of/o.l = 0 is
the equation (X, X) = 1.

/{, = 2 I: (L', - .l6',)g,,XI.

Hence of/oX' = 0 implies I: (L'i - .l6'1)g,,X,Y' = 0 for all choices of Y'.
That is, (L(X) - AX,Y) = 0 for all Y, or L(X) =AX.Thus .l is an eigenvalue
of L, and Xis a corresponding unit eigenvector. L has real eigenvalues since
Lis self-adjoint (Lemma 7.10) and so the Lagrange problem has a solution.
(Alternatively, the problem has a solution since II(X, X) does have a maxi-
mum and minimum: the set of unit vectors in TpM is closed and bounded,
i.e., compact.) The eigenvalues are the roots of
0 = det (L - .ti)= .l" - (trace L).l + det L.
Denote these roots by ,c 1 and ,c 2 with ,c 1 > ,c 2 •
LEMMA 8.1. Let P e M and .l, X be an eigenvalue-eigenvector pair for L
at P. Let Y be a unit vector in TpM such that (X, Y) = 0. Then Y is
also an eigenvector.

Proof· Note that {X, Y} is a basis for TpM,

0 = (.lX, Y) = (L(X), Y) = (X, L(Y)).
Hence L(Y) is orthogonal to X and L(Y) = µY for someµ since TpM has
dimension 2. Thus Y is also an eigenvector. I

PROPOSITION 8.2. At each point of a surface M there are two orthogonal

directions such that the normal curvature takes its maximum value in
one direction and its minimum along the other.
Sec. 4-8 Principal, Gaussian, Mean, and Normal Curvatures 129

Proof: By the preceding discussion we know that the maximum and mini-
mum values of Kn are taken in the direction of eigenvectors of L. Note that if
X(i) is a unit eigenvector of L with eigenvalue ,c 1, then the value of Kn in the
direction of Xcn is
Il(X(I),Xw) = (L(Xcl)),Xw) = ,ci(Xcn,X(i)) = ,c,.
*
Thus the maximum value of Kn is 1C1 and the minimum is ,c2 • lf,c 1 K2 , then
1C1(Xco,Xm) = (L(Xm), Xc2,) = <Xw,L(Xc2,)>= ,c2(Xm, Xm) implies that
(X 0 ,, XC2l)= 0. If ,c 1 = K 2, Lemma 8.1 implies that a unit vector orthogonal
to Xco is also an eigenvector. Hence it may be chosen as Xc2,. In either case
(X 0 ,, XC2l)= 0. I

DEFINITION.The principal curvatures of a surface M at a point P are the

eigenvalues of L there (K 1 and K 2 ). Corresponding unit eigenvectors are
called principal directions at P.

DEFINITION.An umbi/ic is a point where 7' 1 = 7' 2 •

EXAMPLE 8.3. Every point of either S 2 or R 2 is an umbilic. We shall see that

the converse is true in Section 6-1.

DEFINITION.A line of curvature on a surface Mis a curve whose tangent

vector at each point is a principal direction at that point.

In Problem 8.5 you will show that the meridians and circles of latitude
of a surface of revolution are lines of curvature.

8.4 (Euler). Let Y be a unit vector tangent to M at P. Then

THEOREM
II(Y, Y) = K 1 cos 2 0 +
K 2 sin 2 0,

where O is the angle- between Y and the principal direction Xco corre-
sponding to ,c 1 •

Proof· Let Xm be the principal direction corresponding to K 2•

L(Xw) = ,c,Xw.
Thus in terms of the orthonormal basis Xm, Xm, L is represented by

(~I ~J·
Y = cos OXw + sin OXm.
Il(Y, Y) = (L(Y), Y)
= (,c:1 cos OXrn + K 2 sin OXm, cos OXm + sin OXm)
= K 1 cos 2 0 + K 2 sin 2 0. I
Note that this theorem can be used to compute normal curvatures.
130 Local Surface Theory Chap. 4

EXAMPLE 8.5. Consider the Monge patch (u, v, v 2 - u 2) defined for

u2 + v 2 < 1. The curves u = 0 and v = 0 are lines of curvature. See
Figure 4.20.

FIGURE 4.20

DEFINmoN. The Gaussian curvature of Mat Pis K = ,ci7c2 = det L. The

mean curvature of Mat Pis H = !(,c 1 + ,c2 ) = ! trace (L).

The terminology "mean curvature" reflects the fact that His the average
normal curvature (see Problems 8.3 and 8.11). Surfaces for which H= 0 are
called minimal surfaces. A minimal surf ace has the local property that if a
"small" region of the surf ace is deformed slightly without changing the
boundary curve, the area of the region is increased. That is, the given region
has minimal area among all nearby regions . (The concept of area is developed
below.) Such a surface may be made experimentally by dipping a simple
closed curve made out of wire into some liquid soap. The resulting soap film
bounded by the wire frame is a minimal surface. See do Carmo (1976) for a
more extensive study of the differential geometry of minimal surfaces. See
Almgren (1966) for a detailed study of soap films and their geometric prop-
erties.
We shall now give a geometric interpretation of the Gaussian curvature.
This will require the concept of integration on a surf ace.

DEFINITION.If x: 'll--. R3 is a parametrized surface, the area of a subset

<Rc x('ll) is ffx-•«R>
[x 1, x 2, n] du 1 du 2 or

(8-1)

,Jg du 1 du 2 is called the area element and is denoted dA.

If it is necessary to integrate a function, such as K, over a surf ace, it is done
with respect to dA: ff<llK dA = ffx-•<CR> K(u 1, u 2)Ji' du 1 du 2• See the digres-
Sec. 4-8 Principal,Gaussian, Mean, and Normal Curvatures 131

sion below for a justification of this definition. In this chapter, the concept
of area is used only in Proposition 8.6, which can be omitted.

DEFINITION.The normal spherical image (or Gauss map) of a surface is the

function 11: M - si which sends each point of M to the normal at M.
(Actually 11 is only defined on a coordinate patch, where there is a well
defined normal.)

PROPOSITION 8.6. The Gaussian curvature K at a point P is the limit of the

as the region CRshrinks to the point P and A denotes
ratio A(ll(CR))/A(CR)
(signed) area.

[x 1, x,., D]dui du,.= JI.,./g dui dui. The function

Proof: A(CR)= JJ-''<CR>
D:x- 1(CR)- S 2 has image ll(CR)and is a parametrized surface that might not
be regular (Di x D,. = 0 ?). This lack of regularity does not affect the formula
for the absolute area of 11(CR): IA(11(CR))I [D1 , Di, m) dui du,., where
= JJ-¥-•<CR>
m = D 1 x Di/I D 1 X Di I = ±D. The plus sign holds if 11preserves orientation,
the minus if it reverses it. Thus signed area is given by

A(11(CR))
= ff[D 1, Di, D]du 1 du,.,
Then
C(P) = lim A(11(CR))
= [D1, n,.,D].
CR-PA(CR) [x1, x,., DJ
By Problem 8.9 Di x Di = K.,./g n. Thus [D1 , Di, D] = K,,./g and C(P) = K.
I
This theorem gives geometric significance to the sign of K: K negative
means 11reverses orientation at P.
Note that neither the matrix L 11nor the mean curvature is intrinsic. I~ is
a remarkable fact that the Gaussian curvature is intrinsic, as we shall see in
Theorem 9.2.
We have given some geometric meaning to the Weingarten map Lin
terms of its trace 2H and determinant K. We shall now give a further geome-
tric interpretation to the second fundamental form. We know that Il(X, Y)
is a symmetric bilinear form on TpM for each P e M. If we think of II as a
generalization of the length of a vector, then we can ask which vectors have
"generalized length" + 1 and -1 with respect to II. This information (which
is classically called the Dupin indicatrix) provides information about the
geometry of M near P.

DEFINITION.The Dupin indicatrix D of Mat PE Mis the subset of TPM

given by
D = {X E TpMI Il(X, X) = I} u {X e TpMI Il(X, X) = -1}.
132 Local Surface Theory Chap. 4

We shall write D = D+ u D-. In terms of local coordinates,

D+ = {:EX'x,l:EL, 1X'X 1 = l} and D- = {L X'x,l:EL, 1X'X 1 = -1}.
(Since the sign of n is not well defined, neither is that of II. A change in the
sign of n will interchange D+ and D- but does not, of course, change their
union which is D.) The local coordinate description shows that D+ and D-
are (possibly degenerate) conic sections when viewed as graphs in the tangent
plane. Note that the axes (xi, x 2 ) that we use for TpM may not be orthogonal
since g 12 may not be zero.

EXAMPLE 8.7. Let M = S 2 • By Problem 7.1 Lis the identity transformation.

Hence Il(X, X) = (L(X), X) = (X, X). Then D = D+ = the unit circle
in TpM. See also Problem 8.20.

Depending on the signs of the principal curvatures and whether they

vanish or not, the Dupin indicatrix may be an ellipse, two conjugate pairs of
hyperbolas (i.e., having the same asymptotes and semi-axes), two parallel
lines, or empty. We leave the proof of the precise statement (below) as
Problem 8.21.

PROPOSITION 8.3. Let P be a point on M with Gaussian curvature Kat P.

Then
(a)" if K > 0, the Dupin indicatrix is an ellipse;
(b) if K < 0, the Dupin indicatrix is two conjugate pairs of hyperbolas;
and
(c) if K = 0, the Dupin indicatrix is two parallel lines if one principal
curvature is not zero and is empty if both principal curvatures are
zero.

Because of this proposition, the point P is called elliptic if K > 0, hy-

perbolic if K < 0, parabolicif only one principal curvature vanishes, and flat
(or planar) if both principal curvatures are zero.
The geometric significance of the Dupin indicatrix comes from the
classical interpretation we now describe. The basic idea is to shift the tang_ent
plane parallel to itself in the direction of the normal to the surface. The curve
of intersection of this new plane and the surface is "approximately the Dupin
indicatrix." See Figures 4.21, 4.22, 4.23 for the case of elliptic, hyperbolic,
and parabolic points, respectively.
We now make the classical notion that the intersection is "approximately
the Dupin indicatrix" more precise. Let x: 'U.- R 3 be a coordinate patch
about P with x(O, 0) = x 0 = P. The tangent plane at P is given by
TpM = {y e R 3 I(y - x 0 , n) = O},
where n is the normal to M at P. Let E be any (possibly negative) number
and n. = {y e R 3 I(y - Xo,n) = E}. n. is the plane parallel to TpM at a
Sec. 4-8 Principal, Gaussian, Mean, and Normal Curvatures 133

FIGURE 4 .21 FIGURE 4.22

FIGURE 4.23

signed distance E from it. Since we wish to know what M n IL is for small
E, we shall examine

x('ll) n IT, = {x(u', u 2 ) I(u', u 2) E 'l1 and (x(u', u2 ) - Xo,n) = E}.

To this end we expand x in a Taylor series about (0, 0):
x(u', u2 ) = x 0 + :Ex,(O,O)u' + ! :Ex,i(O,O)u'u'+ r(u', u 2 ).
Then
E = (x(u 1, u2) - x 0 , n)
= :E(x,(O. 0), n)u' + ! :E(x,iO, 0), n)u'u' + (r, n)
or
(8-2)
Providing x is sufficiently differentiable we may assume that the remainder
r(u1, u 2 ) satisfies
1. r(u1, u2 ) 0 ( 1 2) (O O)
im (u•)2 + (u2)2 = as u'u - ' .
Thus for u' and u 2 small, r is quite small and the intersection is approximated
by the conic section E = ! :EL,1 u'u1• We shall make a change of coordinates
that will have the effect of normalizing this conic section, and will stretch it
as E ----. 0. Currently we are using u' and u2 as coordinates in the plane, with
axes given by x,(O, 0) and x 2 (0, 0), which may not be perpendicular. Define
new coordinates here by ...1.'= u'/.../2JET . Notice that as E----. 0 this has the
effect of stretching the graph of M n IL relative to a fixed (...1. 2 ) coordinate
1 , .i1.

system.
134 Local Surface Theory Chap.4

If we substitute ui = l\/21 € I into Equation (8-2) and divide by I€ I we

obtain
± 1= ~ L- l' l' + (r(,J2f"€T i 1,,J2jif i2), n).
~ •J 1€1
We compute the limiting position of this locus as € --> 0. By assumption,
. r(~ A.i, ,J2f"€TA.2)
~~ 21€ I ((l 1)2 + (l2)2) = o.
Hence
1, ,J2fEfl
1. r(~A. 2) _
0
im
,-o I€ I - .
Thus the limiting position as €--> 0 of these "stretched conic sections" is
given by
{I:A.X, I ±1 = I: L,1A.A.1} =
1 1 D.
This is the classical interpretation of the Du pin indicatrix.

DEFINITION.Two non-zero vectors X and Y are called conjugate directions

(or conjugates) if Il(X, Y) = 0.

DEFINITION.A tangent vector X at P is an asymptotic direction (or is self-

conjugate) if II(X, X) = 0. An asymptotic curve is one whose tangent
vector is an asymptotic direction at each point.

The terminology asymptotic direction is due to the fact that the asymp-
totic directions at a hyperbolic point are the asymptotes of the Dupin in-
dicatrix, which is an hyperbola. See Example 8.9 below. Various properties
of asymptotic and conjugate directions are developed in Problems 8.22
through 8.35.

EXAMPLE 8.9. Let x be the Monge patch x(u, v) = (u, v, u2 - 2v2). Then
x 1 = (1, 0, 2u), x 2 = (0, I, -4v), n = (-2u, 4v, 1)/,v'l + 4u2 + 16v2 ,
X1 1 = (0, 0, 2), X12 = (0, 0, 0) = X21, X22 = (0, 0, -4),

L - 2
i1 - ,JI+ 4u2 + 16v2'
L12 = L21 = 0,
and
-4
L22 = ,v'l + 4u2 + 16v2
At x(O,O) the coordinate axes x 1 and x 2 are orthogonal. The Dupin indicatrix
there is given by
 Principal.Gaussian, Mean, and Normal Curvatures 136

and consists of two conjugate pairs of hyperbolas, as pictured in Figure

4.24. The asymptotic directions are given by
0 = II(X, X) = 2(X 1) 2 - 4(X2f 2 = 0
or X 1 = ±~ x2.. Note that these give the asymptotes of the hyper-
are conjugates. If X = x 1 x 2
bolas in Figure 4.24. Note that x 1 and X:z. +
then Y = Y1x 1 + y2.X2.is conjugate to X if and only if
o= (L(X), Y) = 2x 1 y 1 - 4x2.y2. = 2y 1 - 4y2.
or Y 1 = 2y2.. Hence 2x 1 + X2.is conjugate to x 1 + X:z.,

FIGURE 4.24

DIGRESSION

We indicate here in a very brief fashion, the origin of the area formula
in a coordinate patch. For a more rigorous foundation seeFulks [1969].
Consider a region ct contained in the image of a coordinate patch
x: 'U.- R 3 bounded by four parameiric curves: u1 = a, u1 =a+ 4u 1,
u2.= b, u2 = b + &2.. In the tangent plane at x(a, b) there is a parallelogram
<Bspanned by 4ul x 1 and 4u2.x2,.If 4u 1 and 4u 2 are quite small, the area
of <Band ct should be almost the same. See Figure 4.25.
Theareaof<Bis 1&1 x 1 x 4u2.x2.I= 4u 1 4u 2 lx 1 x x2.I= 4u14u2 ../ g.
The area of a large region CRin x('U.)should be the sum of lots of small areas of
regions like ct and thus should almost equal the sum of terms 4u 1 4u2.../ g •
In the limit this says
area of CR= ff ,./g du 1 du2.
.,.-•(CR)

FIGURE 4.25
138 Local Surface Theory Chap.4

If <Ris also contained in the 'U coordinate patch we must have area of
<R= J.fr•clll> 2,
,.fg' dv1 dv where i = det (i.,). Since (g 11) = (J- 1)'(i.,)r1,
g= det (g,1) = det (J- 1) det (i.,) det (J- 1) = i(det (J- 1))2 •
Thus ,lg'= ,./g'ldet (J- 1)1.Thus

f·rJ.,-•(dl),,/g dul du'-= s·r (,./g' a/-l)ldet(J-l)ldul

J.,-,(dl)
du'-,

which, by Jacobi's rule for transformation of integrals (the substitution or

change of variables formula for multiple integrals), equals ffr'<lll>
,,/i dv1 dv2 •
(/: 'U - 'U.is the transformation of coordinates.)

EXAMPLE 8.10. For a Monge patch x(r, s) = (r, s, f(r, s)) we have
= (1 + /,'-)(1 + !.'")- (/,/.)'- = 1 + 1,2 + !.'-.
g
The area is given by ff ,,/1 + /,'-+ !.'"dr ds. This should be a familiar
formula from Calculus III.

EXAMPLE 8.11. For the upper hemisphere (Example 1.7 of Chapter 4) we have
f(r, a) = ,,/1 - r 2 - s 2• I, = -r/,,/1 - r 2 - s 2, f. = -s/,,/1 - r 2 - s 2,
and ,,/1 + f,2 + /. 2 = ,,/1/(1 - r 2 - s 2). Hence the area of a hemis-
phere must be

= fI

-I
(
arc sin ,,/l '_ 82) 1,/T=',I
ds
-~T=',1

Recall from Calculus II that if a function/ defined in R 2 is given in terms

of polar coordinates, then the integral of f over a region is given by
ff/(r, 8) r dr d8, not ffI (r, 8) dr d8. Furthermore, the expression r dr d8 is
precisely what is integrated to find area: area of <R= ffdl r dr d8. Similarly, if
wehave a function/ defined on a surface, its integral over a region <Rcontained
ff
in a coordinate patch will be defined as /(u 1, u2),/g' du1 du2 •

PROBLEMS

General
8.1. Find the Gaussian and mean curvatures of R 2 , S 2 , T 2 (Problems 1.1
and 7.3), and S 1 x (0, I) (Problem 7.2).
8.2. Prove H 2 > K. When does equality hold?
Sec.4-8 Principal. Gaussian. Mean. and Normal Curvatures 137

8.3. Let X and Y be orthonormal vectors at P. Prove

H = }(II(X, X) + II(Y, Y)).
8.4. Prove that the u'-curves are lines of curvature if and only if
=Lu=
K12 0.
8.5. Prove that the circles of latitude and the meridians of a surface of
revolution are lines of curvature.
t8.6. Suppose g 12 =0. Prove
1 ( iJ
iJg,I
au2 iJ T,7
Ogzz)
K = - 'f:7g iJuz-::Jg + iJu' -::Jg .
*8.7. Suppose g 11 = I, g =0. Prove
12

iJz-v'fu+ K r::- - 0
(c7u')2 "1 Ku - ·
8.8. Determine which points of the Monge patch x(u, v) = (u, v, u 3 + v 3)
have K > 0 and which have K < 0.
*8.9. Prove n 1 x n 2 = K,/g n.
8.10. What are the principal curvatures for a surface of revolution?
8.11. Prove H = (1/2n) J:"Kn d8, where 8 is as in Euler's Theorem (8.4).
This shows that the mean curvature is the average of the normal cur-
vature over all directions.
t8,12. Suppose two surfaces M 1 and M 2 intersect in a curve e. Let K be the
curvature of e, l, the normal curvature of e in Mi, and 8 the angle
between the normals of M 1 and M 2 • Prove
K 2 sin 2 8 = 1 1 2 + 1 2 2 - 21 11 2 cos 8.

8.13. Let y(s) be a unit speed curve on a surface M. The geodetic torsion of
T is T6 = -(S, dn/ds) (where S = n x T).
(a) Prove T6 =0 if and only if 'Yis a line of curvature.
(b) Prove if T is a geodesic then T = T•.
(c) Does T = 1:6 imply 'Y is a geodesic? (Give proof or counter-
example.)
t8.14. Suppose that in a coordinate patch x: 'l1 - M the Weingarten map
satisfies L = f I where f: 'l1 - R and I is the identity linear transfor-
mation. Prove f is a constant in a neighborhood of each point.

Set A-Minimal Surf aces

8.15. Let M be the surface of revolution generated by the non-unit speed
curve cz(t) = ((I/a) cosh (at + b), t). Show that Mis minimal (H ==
0).
M is called a catenofd. .
138 Local Surface Theory Chap.4

8.16. Let M be the surface x(u 1 , u 2 ) = (u 2 cos u 1 , u 2 sin u 1 , pu 1). Show that
M is minimal. M is called a helicoid.
8.17. Prove that a surface is minimal if and only if
g11Lu - 2guLu + gzzLu O. =
t8,18. Prove that Problem 8.15, together with the plane, gives all surfaces of
revolution that are minimal.

Set 8-Dupin lndicatrix, Asymptotic and Conjugate Directions

8.19. Classify all the points of the Monge patch x(u, v) = (u, v, u 3 +v 2)

as to being elliptic, hyperbolic, parabolic, or flat.

8.20. Let P = (1/2, 1/2, 1/,./2) E S 2 • Graph the Dupin indicatrix of S 2
at P, using axes parallel to x 1 and x 2 , where the coordinate patch is
the upper hemisphere (Example 1.7). Be careful, x 1 and x 2 are hot
orthogonal. Your result should still be a circle as in Example 8.7.
8.21. Prove Proposition 8.3. (Hint: Write elements of TPM in terms of a
basis of principal (orthogonal) directions.)
8.22. Show that there are 0, 1, or 2 linearly independent asymptotic direc-
tions at P, depending on whether P is elliptic, parabolic, or hyper-
bolic.
8.23. Prove that the principal directions bisect the asymptotic directions at
each hyperbolic point.
8.24. Prove that the parametric curves are asymptotic curves if and only if
Lu= Lzz = 0.
8.25. Let "I be a curve on M. Suppose that the curvature ,c of "I is nonzero.
Prove that "I is an asymptotic curve if and only if B is normal to the
surface, i.e., the osculating plane of"f coincides with the tangent plane
of Mat each point of "I·
8.26. Let "I be a straight line lying on a surface M. Prove that "I is asymptotic.
t8.27. Let x be the Monge patch x(u, v) = (u, v, (u 2 /4) - v 2). Prove that all
asymptotic curves are straight. (Hint: Problem 8.26 may be useful.)
·8.28. Prove that a geodesic is an asymptotic curve if and only if it is a
straight line.
8.29. Two surfaces x(u 1 , u 2 ) and y(u 1 , u 2 ) are said to have contact of order r
at (a, b) if x - y and all its derivatives up through order rare zero at
(a, b). If x: 'll - R 3 is a surface with P = x(O, 0), determine under
what conditions y = a + :Eb1u1 + :Ec 11u 1u1 has second order
contact with x at (0, 0). The resulting surface y is called the osculating
paraboloid. Show that x and y have the same normal vector at P.
Sec. 4-8 Principal. Gaussian. Mean. and Normal Curvatures 138

Consider a plane parallel to the tangent plane at P. How is its inter-

section with the osculating paraboloid related to the Dupin indicatrix?
(This problem is an analog for surfaces of Problem Set D of Section
2-4 which deals with spherical contact for curves.)
8.30. Prove that if K * *
0 and X 0 then there are exactly two unit vectors
Y1 and Y 2 such that X and Y, are conjugates for i = 1 and 2.
8.31. Prove that the tangents to the parametric curves are conjugates at
each point if and only if L 12 =
O. In this case we say that the para-
metric curves form a conjugate family of curves.
8.32. Let M be a surface with no umbilics. Prove that the parametric curves
form an orthogonal conjugate family of curves if and only if they are
lines of curvature.
8.33. Can a line of curvature ever be an asymptotic curve?
8.34. Prove that (L(X), L(Y)) - 2H(L(X), Y) + K(X, Y) =
0. (Hint:
Write X and Y as linear combinations of the principal directions or use
the Cayley-Hamilton Theorem from linear algebra.)
t8.35. Prove the Beltrami-Enneper Theorem: If T has non-zero curvature,
torsion T, and is an asymptotic curve, then T 2 = - K, where K is the
Gaussian curvature. (The proof will make use of Problem 8.34.)

Set C-Ruled Surfaces

A ruled surface is a surface swept out by a moving straight line, much as
a curve is swept out by a moving point. Such a surface can be parametrized
in the form x(s, t) = cz(s) + tP(s), where cz(s) is a unit speed curve and
IP(s)I= 1. However, not every function of this form gives a surface. There
are minor problems with self intersections if x is not one-to-one. These will
be ignored. However, there may be singular points, that is, points where
x 1 x x 2 = 0. We allow this below but will keep track of any singular points.
8.36. Prove that the surfaces in Problems 1.4, 1.9, and 1.14 of Section 4-1
are ruled.
t8,37. Prove that the surface z = x 2 - y 2 is doubly ruled; that is, through
each point of the surface there are two straight lines that lie on the
surface.
t8.38. Prove that the surface
xi y2 z2
:;J
a + b2 - c = 1
-:z
is doubly ruled.
8.39. Prove that the curvature of a ruled surface is never positive.
140 Local Surfaca Thaory Chap. 4

=
A ruled surface for which P' 0 is called a cylinder. (See Problem 5.10
for an example.) A ruled surface for which P'(s) *
0 for alls is called non-
cylindrical.
+
8.40. Let x(s, t) = cz(s) tP(s) be a noncylindrical ruled surface. Prove
there exists a unique curve T(s) = cz(s) + r(s)P(s), which in general
=
is not unit speed, such that (T', P') 0. This is called the line of
striction.
8.41. Find the line of striction for the surfaces in Problem 8.36.
8.42. If7(s) is the line ofstriction of the ruled surface x(s, t) = cz(s)+ tP(s),
prove that y(s, u) = 7(s) + uP(s) is a reparametrization, where
u = t- r(s).
8.43. Show that any point on the line of striction where 7' = 0 is a singular
point.
8.44. Show that every singular point of a ruled surface lies on the line of
striction. (Hint: Parametrize as y(s, u) = 1(s) + uP(s).)

Set D-Deve/opable Surfaces

A ruled surface is called developable if along each line of the ruling the
tangent planes are all parallel.
8.45. Let x(s, t) = cz(s) + tP(s) be a developable surface. Show that the
normal vector n does not depend upon the parameter t.
8.46. Let x(s, t) = cz(s) + tcz'(s) be the tangent developable surface of a unit
speed curve cz(s) (Problem 1.14). Show that xis developable.
8.47. Compute the metric matrix for the surface in Problem 8.46.
8.48. Show that a surface of revolution generated by a straight line is
developable.
8.49. Prove that x(s, t) = cz(s) + tP(s) is developable if and only if
=
[cz',p,P'J o.
8.50. If cz(s)is a unit speed curve, under what conditions is
x(s, t) = cz(s) + tN(s)
developable?
8.51. When is y(s, t) = cz(s) + tB(s) developable?
8.52. Show that a developable surface has Gaussian curvature equal to
zero.
8.53. Show that each line in the ruling of a developable surface is both an
asymptotic curve and a line of curvature.
. Sec. 4-9 Riemannian Curvature and Gauss's Theorema Egregium 141

8.54. Let M be a developable surface without umbilics. Let cz(s) be a (unit

speed) line of curvature corresponding to the nonzero principal cur-
vature. Show that cz(s)is orthogonal to each line in the ruling.
8.SS. Let cz(s) be chosen as in Problem 8.54 and parametrize the develop-
able surface M by x(s, t) = cz(s) + tP(s) with IPI = 1. Show that
P'(s) = l(s)ct'(s) for some function ..l(s).
8.56. If ..l(s) is as in Problem 8.55, show that
(a) ..l= 0 implies Mis a cylinder;
=
(b) ..l constant =I=0 implies all lines of the ruling have a point in
common and in this case that point is the line of striction (Mis a
cone, but not necessarily circular);
(c) ..l and ..l' both nonzero implies Mis the tangent developable sur-
face of its line of striction.

4-9. RIEMANNIAN CURVATURE AND

GAUSS'S THEOREMA EGREGIUM

In this section we introduce the Riemannian curvature tensor (R/ kr),

which is an intrinsic invariant. We then obtain Gauss's equations and the
Codazzi-Mainardi equations and use them to prove the surprising fact that
the Gaussian curvature is L R/ 21 g 12/g and hence intrinsic .
Let x: 'll -+ R 3 be a coordinate patch on M with Christoffel symbols
r,/ and second fundamental form coefficients Lw
DEFINITION . The Riemannian curvature tensor with index (i, l,j, k) is given
by

(9-1)

for all 1 ~ i, l,j, k < 2.

Most readers complain about the definition of R/1k because it is so
replete with symbols that they feel (correctly) that the geometric meaning of
the Riemannian curvature tensor is lost. In order to define R/Jk more geo-
metrically we would have to take a much more abstract approach using the
formalism of covariant derivatives. We will do this in Chapter 7.

PRorosmoN 9.1. For all I< i, /,j, k ~ 2

(a) (Gauss's equations)

(9-2)
142 Local Surface Theory Chap.4

(b) (Codazzi-Mainardi equations)

(9-3)

Proof· We prove both equations simultaneously by computing ox,1'ari'

= x,1k
two ways and comparing the tangential and normal components of the
result. Using Gauss's formulas (4-8), we have x,1k= a(L,
1n + L r,/x,)/ari'.
Therefore

x,,k= ~~n + L,,nk

+ L a::ix,+ L r,/x,k
= ~Jn - L L,,L'kx,+ L a{,;lx,+ L r,/r,kmxffl
1 + L r,/L,kn
= ($;+ L r,/L,k)n+ L (a!,;l
- L,,L'k+ L r,/rpk')x,.
Similarly

x,kl= (~;k+ L r,k'L,,)n+ L (T,;'- L,kIJJ

+ L r,kPrp/)x,.
Since we assume that x is at least C3 ,
ax 3 ax 3
x,,k= oil'au
1a,1= auaukau'= x,k}'
1

{x1, x 2 , n} is a basis of R 3 at each point so that the various components of

x,1k with respect to this basis must equal those of x,k,. Hence

(9-4) ($7+ L r,,'L,k)= (~~~k + L r,k'LIJ)

(9-5) eJ,;l
- L,,L'k+ L r,/r Pk')= (af;' - L,kIJJ+ L r,kprp/)
Equation (9-4) can be rewritten as

$Ll aL,k ~ <rlk'LIJ- r IJ'L,k>

U -w=~
which is the Codazzi-Mainardi equation (9-3). Equation (9-5) may be re-
written as
ar,k' al~,+~
a',iT -U erlkprPi, - r IJprpk') = LlkL'
~ J -
L IJ L'k"

Since the left-hand side of this equation is R,'1k (by the definition), we have
Gauss's equation (9-2). I
Sec. 4-9 Riemannian Curvature and Gauss's Theorema Egregium 143

The important point here is that the Riemann curvature tensor is defined
intrinsically by Equation (9-1). Gauss's equation (9-2) gives an extrinsic
description in terms of the second fundamental form and the Weingarten
map.
We use the above to prove:

THEOREM 9.2 (Gauss's Theorema Egregium.) The Gaussian curvature Kofa

surface is intrinsic.

Proof· By Gauss's equation (9-2) R,' 1k = L,kL' 1 - L, 1L 1k. Hence

I; R/,kglm = L,kLm, - L,,Lmk•
Let i = k = I and j = m = 2.
I; R/21g,2 = L11L22 - L12L21 = det (L,1) = det ((L\)(g,k))
= det (L\) det (g,k) = Kg.
Hence K = I; Ri'21g,2/g. I
Gauss called this theorem egregium because it is truly remarkable.
(Look up the meaning of egregious if it is unfamiliar.) K is defined very
extrinsically-in terms of n, or in terms of L, or in terms of v, none of which
are intrinsic. Yet K is intrinsic.
It can be shown that if a different coordinate patch is used, the coef-
ficients R.//17defined there satisfy
- OV"ou'ovfl
OV7'
R/ Jk = L 7 ail
R,.'1,, ov6 aiiiaiik.
From this it follows that there is a trilinear function mapping
TpM x TpM x TPM--+ TpM,
called the Riemann curvature tensor, by R(X, Y)Z = I; R/ 1kXJYkZ'x 1• We
then note that (R(x 2, x 1)x 1, x 2)/I x 2 x x 112 = K. In modern differential
geometry (of higher dimensions) the expression (R(X, Y)Y, X), where X
and Y are orthonormal, is called a sectional curvature. It plays a significant
role in describing the geometry of M. This will be covered in more depth in
Chapter 7.
One important aspect of the Codazzi-Mainardi equations (9-3) is that
they give integrability conditions. The Fundamental Theorem of Surfaces
states that if g,1 and Lil are symmetric functions such that g 11> 0, det (g,1) > 0,
and both Equations (9-2) and (9-3) hold, then there is a surface in R 3 ,
unique up to position in space, having the g 11 and L,,as coefficients of the
first and second fundamental form. See Section 4-10. Thus the situation for
surfaces is a bit different than that for curves, where only the (natural) restric-
tion 'IC > 0 is placed upon 'ICand T in order to determine a curve.
144 Local Surface Theory Chap. 4

PROBLEMS

9.1. Show that if M = R.2 , then R,' 1k = 0 for all 1 ~ i, l,j, k ~ 2 both
intrinsically and extrinsically.
9.2. Prove that the Riemann curvature tensor has the following symmetry
properties :
(a) R,'1" = -Rl1c,, hence Rlu = O;
(b) R,'1k + R/1c,+ R/, 1 O;=
(c) if R,m11c= L Rf ,1cK1m
then R,lflJk= -Rm,Jk; and
(d) R1mJk = R11c1m•
9.3. Compute the Riemann curvature tensor for M = S 2
(a) extrinsically;
(b) intrinsically. (Hint: Use Problem 9.2 to cut down on the calcula-
tions.)
9.4. Compute the Rl,1cfor
(a) T 2 (Problem 1.1);
(b) S 1 x (0, I) (Example 2.3).
The following problems are on the whole more difficult than most of our
exercises. They are intended to introduce some of the modern terminology to
help bridge the gap between this text and. more advanced texts. They may be
omitted if Chapter 7 is not covered.
9.5. Let X and Y be vector fields defined on a surface M. In a coordinate
patch x : 'U.- M define
z, = L (x,oY'_ y,ox'),
J Ttl T,ii
where X = L X 1x 1 and Y = L Y1xr Similarly, in the patchy: '0- M
define
z· = ~ (x,~ _ t,~:':)·
Prove that Z1 = L z•ou1/ov•.This means that L Z'x, = L z•y,,. = z
is a vector field. Z is denoted [X, Y] and is called the Lie bracket of X
and Y. Prove that [x,, x 1] = O. Prove that [X, Y] = -[Y, X]. Show
that [u1x 2 , u 1u 2 x 1] = (u 1) 2 x 1 - u 1u 2 x 2 •
9.6. Recall that in Problem 5.14 we defined the vector field

VxY = L X'(?u'~ 1)xk

+ L r,,"Y
Sec. 4-9 Rlemannian Curvature and Gauss's Theorem, Egregium 146

in a coordinate patch x: 'U---+ M. Prove that

Vx Y - VyX - [X, Y] = 0.
(Hint: r,/ = ri/ (Why?).)
9. 7. In this problem we interpret a vector field as a "directional derivative."
Let f: M---+R be a function defined on aJI of M. We say that f is
differentiable if for every coordinate patch x: 'U --+ M, fa x: 'U.--+ R
is differentiable. If X is a vector field on M, then in the patch x set
X(f) = I: X' of/ou'.Prove that in terms of another patch y: 'U --+ M
we have _I;X• of/ihJ•= I; X' of/ou'.Prove that if a ER, and/and g
are differentiable functions on M then: (I) X{f + g) = X(f) + X(g);
(2) X(/·g) = g-(X(f)) + /·(X(g)), where · denotes multiplication;
=
and (3) X(a) 0, where a denotes the constant function h(P) =
a. It
can be shown that any operator X that satisfies (I), (2), and (3) actually
is a vector field. This will be the definition in Chapter 7.
9.8. Prove [X, Y](f) = X(Y(f)) - Y(X(f)). Let/·X denote the vector field
which atPis/(P) times XatP. Prove [/·X, Y] = /·[X, Y]- (Y(f))·X.
Prove [X + Y, Z] = [X, Z] + [Y, Z].
9.9. Prove Vx(Y + Z) = Vx Y + VxZ, Prove Vx+Y(Z) = Vx.Z + VyZ.
Prove V1 .x.(Y)= f· VxY. Prove Vx{/• Y) = X{f) •Y + f· VxY.
9.10. If X, Y, Z are vector fields on M, set
R(X, Y)Z = Vx(VyZ) - Vy(VxZ) - Vcx,Y1Z,
Prove that in a patch x: 'U.--+ M we have R{X, Y)Z = I: W'x, with
W1 = I: R,',kZ 1XJYk and Rl,k is defined by (9-1). This is the same
concept as that at the end of this section.
9.11. Prove R(X, Y)Z = -R(Y, X)Z.
Prove X((Y, Z)) = (Vx Y, Z) + (Y, VxZ).
Prove (R(X, Y)Z, Z) = 0.
Prove (R(X, Y)Z, W) = -(R(X, Y)W, Z).
Prove R(X, Y)Z + R(Y, Z)X + R(Z, X)Y 0. =
Prove (R(X, Y)Z, W) = (R(Z, W)X, Y).
9.12. Prove (R(X, Y)Y, X) = IX x Yl 2 K, where K is the Gaussian curva-
ture.
9.13. Let X and Y be two fixed vector fields. If Z is a vector field define
S(Z) = R(Z, X)Y. Prove S(/· Z) = f· S(Z) and
S(Z + W) = S(Z) + S(W).
In particular, this means that S is a linear transformation from T;M
to TpM. Prove that the trace of S is IX, Y 12 K.
146 Local Surface Theory Chap. 4

4-10. ISOMETRIES AND THE FUNDAMENTAL

THEOREM OF SURFACES

In this section we shall discuss the question as to when two surfaces are
geometrically the same. There will be two concepts: isometry, which refers
to intrinsic geometry, and rigidity, which refers to both the intrinsic and
extrinsic geometry. In the latter case, two surfaces will be considered the same
if they differ only in their position in space, as for example the two spheres
{x E R 3 I I xi = 1}and {x E R 3 I Ix - (1, 0, 0) I= l}. It is the former concept
that shall be our primary interest.

DEFINmON. Letf: M--+ N be a function between surfaces.fis differentiable

if for each P E M there are coordinate patches x: 'l1 --+ M and
y: 'U --+ N about P and f (P) respectively such that y- 1 o f o x : 'l1 --+ 'U
is differentiable as a function of two variables.

In practice, f is often given in local coordinates by giving y- 1 of o x,

as in the following example.

ExAMPLE 10.1. Let M be the surface of revolution

x(u, 0) = (cosh (u) cos 0, cosh (u) sin 0, u)
for -sinh- (1) < u < sinh- 1 (1) and O < 0 < 2n. Let N be the heli-
1

coid y(v, if,)= (v cos if,,v sin if,,if,) for -1 < v < 1 and O < if, < 2n.
The function y- 1 of o x(u, 0) = (sinh (u), 0) is clearly differentiable and
gives a differentiable function/: M--+ N by /(x(u, 0)) = y(sinh (u), 0).
Note that this particular function is also one-to-one and onto. See
Figure 4.26. The meridians of M are sent to straight lines on N and
the circles of latitude are sent to circular helices.

FIGURE 4.26
Sec. 4-10 lsometries and the Fundamental Theorem of Surfaces 147

DEFINITION.An isometry from M to N is a one-to-one, onto, differentiable

function/: M - N such that for any curve T: [c, d] - M, the length of
T equals the length off o T. M and N are isometric if such an isometry
exists.

In Problem 10.1 you will verify that a rotation of S 2 is an isometry. A

less obvious example is the following.

EXAMPLE10.2. If!, M, N are as in Example 10.1, then/is an isometry. This

is seen as follows. With respect to the coordinates (u, 9), the metric
matrix of M is
g,,-
( ) _ (cosh (u)
2

0
With respect to the cordinates (v, ;), the metric matrix of N is

(h,,) = (~ 1: v2)·
If T: [c, d] - M is given by T(t) = x(u(t), 9(t)), then d'f/dt has length
cosh (u(t)r/ Cl2 + 82 • Since (f o T)(t) = y(sinh (u(t)), 9(t)), d(f o 1)/dt
has length ,J cosh 2 (u(t))r.i2 + (I + sinh 2 (u(t)))8 2 = cosh (u(t)~
also. Hence T and/ o T must have the same length and/is an isometry.

In the above example we saw that the vectors d'f/dt and d(f o 1)/dt had
the same length for a particular isometry. This is true in general.

PROPOSITION 10.3. Let/: M - N be an isometry and let 1: [c, d] - M be a

regular curve. Then d'f/dt and d(f o 1)/dt are vectors of the same length
for each value of t E (c, d).

Proof· For each t* E (c, d), the curve T: [c, t*] - M has the same length
as IO.,: [c, t*] - N. Hence r ld1/dtl dt = s:·
ld(f O1)/dtl dt. If we dif-
ferentiate with respect to t* and evaluate at t* = t we obtain

DEFINITION.Two surfaces M and N are locally isometric if for each P E M

there are open sets 'U.' c: M, 'l)' c: N with P E 'U.', and an isometry
f: '\1' - 'l)' (and vice versa for Q E N). Such an/is called a local isometry.
ExAMPLE10.4. In Example 10.1 we now allow O ~ 9 ~ 2n, so that Mis the
entire surface of revolution, and - oo < ; < oo so that N is an infinite
staircase./gives one local isometry. Choosing a second coordinate patch
148 Local Surface Theory Chap. 4

to cover M we can easily define another local isometry from M to N

so that M and N are locally isometric. However, they are not isometric
be1tause M has finite area while N has infinite area and we shall see in
Proposition 10.5 that isometrics preserve the intrinsic geometry, in
particular, the area. This proposition says that isometrics preserve the
first fundamental form.

PROPOSITION 10.5. Two surfaces Mand N are locally isometric if and only if
for each P E M there is an open set 'U. c: R 2 and coordinate patches
x: 'U.--+ M, y: 'U.--+ N with PE x('U.)such that the metric coefficients
of x and y are the same. Hence locally isometric surfaces have the same
local intrinsic geometry at corresponding points.

Proof· First assume that /: 'U.'--+ 'l)' is a local isometry with P E 'U.'.
Let x: 'U.--+ M be a coordinate patch about P with x('U.) c: 'U.'. Let
y = f o x : 'U.--+ N, which is clearly a coordinate patch about /(P).
(K,1) will denote the metric tensor of x and (h,1) will denote the metric
tensor of y. We shall first show that KtJ(a,b) = h,i(a, b) for all (a, b) E 'U.
when i = j. Let y(t) = x(a + t, b) (i.e., a u 1-curve). By Proposition 10.3,
IT(t)I = 1(/ o y)'(t)I for all t. Since y(O)= x 1(a, b) and (Jo y)'(O) = y 1(a,b),
Ix 1(a, b)I = IY1(a, b)I and Ku(o, b) = hu(a, b). Similarly Ku(a, b) = h22 (a, b)
by using a u2 -curve.
Now we show that Ku(a, b) = h 12 (a, b). Let cz(t) = x(a +
t, b t). +
+ +
ct(O)= x 1(a, b) x 2 (a, b) and (/ o cz)'(O) = y 1(a, b) y 2 (a, b). Proposi-
tion 10.3 implies Ict(O)I = I(/ o cz)"(0) I so that
(x, + X2, X1 + X2) = (y, + Y2, Y1 + Y2)
or

or
Ku + 2Ku + Ku = hu + 2h 12 + hu-
By the previous paragraph Ku(a, b) = hu(a, b) and Ku(a, b) = h22 (a, b) so
that K12(a, b) = h 12 (a, b) also. Hence (K,1) = (h,1).
On the other hand, if there exists an open set 'U.and coordinate patches
x: 'U.--+ Mand y: 'U.--+ N with the same metric coefficients, let 'U.' = x('U.),
'l)' = y('U.) and define /: 'U.'--+ 'l)' by /(x(u 1, u 2)) = y(u 1 , u 2). Then
y- 1 of o x(u 1 , u 2 ) = (u 1, u2 ) is clearly one-to-one, onto, and differentiable.
If T: [c, d]--+ 'U.', y(t) = x('Jl1(t}, 'Jl2 (t)) and (Jo y)(t) = Y('Jl 1 (t}, 'Jl
2 (t)}, then

'Y and f o T have the same length J: J:

,Jt't'K,1 dt = ,Jt't 1h,1 dt since
K,1 = h,r Hence f is a local isometry. I

EXAMPLE 10.6. Let f, M, N be as in Examples 10.1 and 10.2 so that

y- 1 of o x(u,8) = (sinh(u),8). Then the Jacobian matrix of y- 1 of ox is
Sec. 4-10 lsometries end the Fundementel Theorem of Surf,ces 148

eOS:z
= U
1 J
+ s~nh2

= (cos:z u
cos~z J,
which is the metric matrix for M in the (u, fJ)coordinates.
By using Proposition 10.5 we shall show in the next section that any two
surfaces of revolution having the same positive constant curvature are locally
isometric. In fact, in Section 6-2, after we have introduced the concept of a
geodesic coordinate patch, we shall prove the more powerful theorem stated
below.
THEOREM 10.7. Let Mand Nbe two surfaces with the same constant curva-
ture K = c. Then for each P e M, X E TpM, Q E N, Y E T0 N with
X and Y unit vectors, there is a local isometry f such that f (P) = Q
and f o 11 = y, where II is the unique geodesic through P in the direction
X and y is the unique geodesic through Q in the direction Y.
Note that the above theorem must hold in particular for M = N and
P = Q. Thus for each P on a surface of constant curvature there is a local
isometry that acts like a rotation about P. Also, if M = N there must be a
local isometry that sends any given point P to any other point Q. This is a
situation that is clear on S 2 and in the plane, where each local isometry is
actually global, i.e., defined on all of S 2 or R 2 • (The isometrics of sz consist
of compositions of rotations about various axes and reflections. The isome-
trics of R 2 are compositions of translations, rotations, and reflections. See
Problem 10.7.)
Gauss's TheoremaEgregiummay be stated in the language of isometrics
(and in fact originally was, using the word "developed" which means mapped
by an isometry):
THEOREM 10.8. If two surfaces are locally isometric, their Gaussian curva-
tures at corresponding points are equal.
160 Local Surface Theory Chap. 4

The analogue of Theorem 10.7 for mean curvature is false (see Problem
10.6). Further, the condition of constant curvature cannot be dropped from
Theorem 10.7. Problem IO.IOgives an example of two surfaces and a func-
tion from one to the other which is not a local isometry yet corresponding
points have the same nonconstant Gaussian curvature.
We now turn to the stronger notion of equivalence of surfaces called
rigidity.

DEFINITION. Ann x n matrix A is a special orthogonal matrix if AT= A- 1

and det A = 1. When thought of as a linear transformation from R• to
R•, A is called a rotation.

The terminology rotation comes from the situation when n = 2.

If A= (: :). then A-1 = (_: -:) /det (A). If det A= 1 and
AT=A- 1 , then a=d and b=-c. l=detA=od-bc=a 2 +b 2

implies we may write a= cos; and b = -sin; with O:::::;;;:::::;;

2n. Hence

A= (cos; -sin;)·
sin; cos;
lfv = (' CC:s0) E R 2 , then
r sm (J
A,= (cos; -sin;)(' cos (J)
sin; cos; r sin (J
= ('cos; cos (J - r sin·; sin (J)
+
r sin ; cos (J r cos ; sin (J
r cos ((J + ;))
(
= r sin ((J + ;) .
Hence the special orthogonal matrix A rotates ·R2 through an angle of;
radians.

DEFINITION. I: R• ---+ R• is called a rigid motion of R• if there is a rotation A

of R• and a vector b e R• such that f (Y) = (A,) + b.

DEFINITION. Two surfaces Mand N are rigidly equivalent if there is a rigid

motion/: R 3 ---+ R' such that/(M) = N.

Note that a rigid motion is the composition of a rotation (by A) and a

translation (by b). Since a rotation preserves distances, as does a translation,
a rigid motion must preserve distances and the length of curves. Thus the
restriction of a rigid motion/ to a surface Mis an isometry from M to /(M).
However, the converse is false and an example is given by Example 10.2.
Sec. 4-10 lsometries and the Fundamental Theorem of Surfaces 161

The two surfaces are isometric but not rigidly equivalent. It can be shown that
rigidly equivalent surfaces have the same first and second fundamental forms
at corresponding points so that they must have the same mean curvatures as
well as the same Gaussian curvature. In Problem 10.6 you will give another
example of isometric surfaces which are not rigidly equivalent by exploiting
this idea.
Recall that the Fundamental Theorem of Curves (Theorem 5.2 of Chap-
ter 2) said that the curvature and torsion determine a space curve up to posi-
tion, that is, up to a rigid motion. Further, any differentiable function ic > 0
and continuous function f determine a curve with curvature ic and torsion
f. We now state the analogous theorem for surfaces and give a brief discus-
sion of its proof.

1'HEoREM 10.9 (Fundamental Theorem of Surfaces). Let 'll be an open set

in R2 such that any two points of 'll may be joined by a curve in 'll and
let L,1 : 'll - R and g11: 'll - R be differentiable functions for i = 1, 2
and j = 1, 2 such that
(a) Lu = L21, g12 = g21, gu > 0, g22 > 0, and gug22 - (g12)2 > 0,
and
(b) L,1 and g11satisfy Gauss's Equations (9-2) and the Codazzi-Mainardi
Equations (9-3) where the r,/' are determined by Equation (4-11).
Then, if P E 'll, there is an open set "O c: 'll with P E 'U and a simple
surface x: "O- R 3 such that (g11) and (L0 ) are the matrices of the first
and second fundamental forms. Further, if y: "O- R 3 is another simple
surface with first and second fundamental forms (g,1) and (LiJ), then
y("O)is rigidly equivalent to x("O).

It is clear that conditions (a) and (b) are necessary (this is Proposition
9.1). What is surprising is that these conditions are sufficient! A detailed
proof may be found in the Appendix to Chapter 4 of Do Carmo (1976].
We discussed the analogy between the Frenet-Serret frame {T, N, B} and
the "frame" {:x1 , x 2 , n}just before Proposition 1.13. Because of this, the ideas
of the proof of Theorem 10.9 are very similar to those of the Fundamental
Theorem of Curves, but the technical details are more formidable. The idea is
to set up a system of differential equations for the derivatives of Xi, X2, and n
in terms of g11 and L" and then solve. To actually carry this out for surfaces
requires an existence and uniqueness theorem from partial differential
equations (instead of ordinary differential equations). In general, one cannot
always solve a system of partial differential equations. In the case at hand,
the equations can be solved if and only if condition (b) of the theorem holds l
This once again points out how the theory of differential equations lies at the
heart of geometry. _
The first proof of Theorem 10.9 was given by 0. Bonnet in 1867.
162 Local Surface Theory Chap. 4

PROBLEMS

10.1. Let!,: S 2 - S 2 be given by

cos (J -sin9
( cos(J
f ,(_x,y, z) = s~ (J
0
Show that f ,(0, 0, 1) = (0, 0, 1) and that f, is a rotation around the
z-axis through an angle 0. Prove that/, is an isometry.
10.2. Let S 2(r) = {x E R' 11x I = r} and let f: S 2 - S 2 (r) by /(x} = rx.
Prove that f is one-to-one and onto but not an isometry.
10.3. Show that R 2 and the cylinder S 1 x (-oo, oo) are locally isometric.
10.4. Prove that R 2 and the torus T 2 are not locally isometric. (Hint: Prob-
lem 8.1.)
10.5. If f: M - N is an isometry and 'f is a geodesic in M, show that f o T
is a geodesic in N.
10.6. Find an example to show that the following statement is false: If M
is locally isometric to N, then the mean curvature of Mis equal to the
mean curvature of Nat corresponding points.
10.7. Prove that any rigid motion of R 2 is an isometry of R 2 onto itself
(where R 2 = [(x, y, z) E R 3 Iz = O}).
tl0.8. An isometry f is called orientationpreservingif the Jacobian off has
positive determinant. Prove that/: R 2 - R 2 is an orientation pre-
serving isometry if and only if f is a rigid motion. Do this by setting
2}
f(u 1, u = (f 1(u1, u 2},/2 1, 2))
(u u and proceeding as follows.
(a) Assume that /(0, 0) = (0, 0), (oj1/ou 1}(0, 0) = (of2/ou
2}(0, 0) = 1

and (of1/ou}(0, 0) = (o/ /ou )(0, 0) = 0. Show that

2 2 1

a11)2+ (af2)2
(T,l Trl
= 1

for i = 1 and 2 at all points of R 2 • Differentiate this last expression

to show that o 2fk /ou' ouJ = 0 at all points in an open set about
(0, 0) so that ojk/ouJ is constant there. Use this to show that
f(u 1, u2) = (u1, u2}.
(b) Assume that g: R 2 - R 2 is an isometry. Show that there is a
rigid motion T: R 2 - R 2 such that f =go r- 1 satisfies the
assumptions of part (a}. (Hint: b = g(O,0).)
(c) Use parts (a), (b) and Problem 10.7 to finish the proof.
Note that if f: R 2 - R 2 is an isometry that does not preserve orien-
tation and if S = ( 1 0) is the reflection in the u 1 axis, then So f
0 -1
Sec. 4-11 Surfaces of Constant Curvature 163

is an orientation preserving isometry. Hence every isometry of R 2 is

either a rigid motion or a rigid motion followed by a reflection.
tl0,9. If M is a surface show that the set of (global) isometries from M to
M forms a group under the operation of composition.
10.10. Let M be the surface of revolution
x(t, 0) = (t cos(}, t sin(}, log t) for O < t, 0 < (} < 2n
and let N be the helicoid
y(t, 0) = (t cos(}, t sin(}, 0) for O < t, 0 < (} < 2n.
(a) Show that the metric tensors for Mand N are respectively
t2 + l ) )
(
: ; and (~ 1: 12

so that the obvious function from M to N is not a local isometry.

(b) Show that the Gaussian curvature of both surfaces at (t, 0) is
K = -1/[(l + t 2) 2]. (Hint: Problem 8.6.)

4-11. SURFACES OF CONSTANT CURVATURE

In this last section we shall discuss certain surfaces of constant Gaussian

curvature. Our goal is to determine and describe all such surfaces that are
actually surfaces of revolution. This topic is a good one for the final section
of this chapter because this description will require most of the ideas of this
chapter. Unfortunately, it will also entail the use of integrals that are nonele-
mentary, and hence not computable by elementary functions. We shall find
that such surfaces of revolution fall into three general classes depending upon
the sign of their curvature. Within each class all surfaces with the same curva-
ture will have the same local intrinsic geometry although their global and
extrinsic properties will differ; that is, each surface is locally isometric to
every other surface in the class with the same curvature, but is not globally
isometric to any of the others.
We recall briefly some facts about surfaces of revolution. Let
cz(s) = (r(s), z(s))
be a unit speed curve defined for s in some interval (s0 , s 1) with r(s) > 0
and
M = ((r(s) cos(}, r(s) sin(}, z(s)) IO~ (}~ 2n, s E (s0 , s 1)}.
M is a surface and can be covered with two coordinate patches. Since cz is
unit speed we know (Problem 3.1) that the metric matrix with respect to s
and(} is
.(g,,) = (01 o2)
r
.
154 Local Surface Theory Chap.4

We also know (Problem 4.2) that the second fundamental form has matrix
r' z" - z'r" 0)
(L,1) = ( .
O rz'
From these equations you can derive (Problem 1l.l) the formula for the
Gaussian curvature K of M:
r"
(11-1) K=--·
r
We shall integrate Equation (ll-1) to find all surfaces of revolution of
constant curvature, handling first the case of positive curvature, then zero,
and finally negative curvature.

Case 1: K= a2, a> 0.

In this situation Equation (11-1) becomes r" + a 2 r = 0, which can be

solved to yield r(s) = a 1 cos (as)+ a 2 sin (as) or by rearranging the constants
of integration r(s) = A cos (as+ b). By an appropriate choice of the point
from which we measure arc length we may force b = 0. Since r(s) > 0 we
must have A > 0 and Is I < n/2a. Since «i is unit speed, z' = ±~l - (r') 2
and we have

PROPOSITION 11.l. If M is the surface of revolution generated by the unit

speed curve «i(s) and M has constant positive Gaussian curvature
K = a 2 , then «i(s) = (r(s), z(s)) is given by
r(s) = A cos (as), Is I < n/2a
(11-2)
z(s) = ± f:~l - a 2 A2 sin 2 (at) dt +C
where A > 0 and C are constants.

Note that if A > 1/a, z(s) will be defined only for

Is I < _!_sin- 1 (- 1) < .!!..

a aA 2a
If A ;c 1/a, the integral giving z(s) is nonelementary (it is an elliptic integral)
and cannot be computed by elementary functions.
If A = 1/a, then z(s) = ±f; cos (at) dt + C = ±(1/a) sin (as) + C.
Thus if A = 1/a, M is part of a sphere of radius 1/a. The three cases
A < 1/a, A = 1/a, A > 1/a are pictured in Figures 4.27, 4.28, and 4.29
respectively. Note that for A < 1/a the surface tends to a sharp point on
the axis of revolution and that for A > 1/a the limiting tangent plane for
s = ±(1/a) sin- 1 (1/aA) is perpendicular to the axis.
We thus have a whole family of surfaces of revolution with constant
curvature a 2 > 0. This family is parametrized by the integration constant A.
Sec. 4-11 Surfaces of Constant Curvature 166

FIGURE 4.27

FIGURE 4.28

)
FIGURE 4.29

The extrinsic geometry varies with the parameter A because different values
of A give geometrically distinct surf aces. The intrinsic geometry is independ-
ent of the parameter A (but not a).

PRorosmoN 11.2. Let M 1 and M 2 be two surfaces of revolution of unit

speed curves a.1 and a.2 • H both M 1 and M 2 have constant curvature
a 2 > 0, then M 1 is locally isometric to M 2 •
166 Local Surface Theory Chap. 4

Proof· We shall show that there is a change of coordinates so that the metric
matrix for any surface of revolution in the family ( 11-2) takes the form

I 0 ).
a 2 cos2 (as)
The result will then follow from Proposition 10.5.
With respect to the coordinates (s, (}), the metric matrix is

(~ AZ co~2 (as))·
Define new coordinates (s, tf,) by tf, = aAO. Then the metric matrix with
respect to (s, tf,)is

(: .~)(: I 0
a2 cos2 (as)
).

Thus locally, in terms of the coordinates (s, tf,), all members of this family
have the same metric matrix and so are locally isometric by Proposition
10.5. I
Thus all surfaces of revolution with constant curvature a 2 are locally
isometric to the sphere of radius 1/a. However, their global properties are
different. For example, a surface' of revolution given by Equation (II-2)
with A> 1/a cannot be extended past the boundary Isl= (1/a) sin- 1 (1/aA)
and remain a surface of revolution of curvature a 2 • A path around the "equa-
tor" cannot be stretched and deformed to a single point without leaving the
surface. For a sphere the equator can be so shrunk by passing it up over the
north pole. This is a topological difference between the surfaces. See Figure
4.30.

FIGURE 4.30

Case 2: K= 0.
PROPOSITION 11.3. If Mis a surface of revolution of a unit speed curve«
and M has constant curvature zero then M is either
(a) part of a circular cylinder;
(b) part of a plane; or
(c) part of a circular cone.
Furthermore, these surfaces are locally isometric. (See Figures 4.31,
4.32 and 4.33.)
Sec. 4-11 Surfaces of Constant Curvature 167

FIGURE 4.31 FIGURE 4.32 FIGURE 4.33

Proof · Equation (11-1) becomes r" = 0 so that there are constants a, b e R

such that r =as+ b. Since z' = ±..Jl - (r')2, z(s) = ±,vii- a2 s + d.
Thus «i(s) is a straight line and is given by
(11-3) «x(s)=(as+ b, cs+ d), where a2 +c 2 = 1.
If a= 0 then Mis a part of a circular cylinder, which is possibility (a).
If c = 0, Mis part of a plane (possibility (b)). If ac =I=0 then« is parallel to
Reither the r nor z axes and Mis part of a circular cone (possibility (c)).
Again these surfaces are quite different extrinsically but not intrinsically.
As before we could carefully describe a change of coordinates that would
indicate the local intrinsic geometry is like that of the plane . However, the
formulas become quite messy in the case of the cones, and we shall be content
with the following heuristic description. If M is a piece of a cylinder or a
cone, make a slit along one of the meridians . The surface may now be un-
rolled flat and laid on the plane. The placement gives rise to a coordinate
patch for M with respect to which the metric matrix is the identity matrix.
This is true because the process of unrolling the surface did not cause any
distortions and hence did not change the length of any curve. I

Case 3: K = -a 2, a > 0.
Equation (11-1) becomes r" - a 2 r = 0. This has the general solution
(see Problem 11.5)
(11-4) r(s) = c, cosh (as)+ c2 sinh (as).
This may be rewritten for some real numbers B, b or C, c as
Bcosh (as+ b) if C1 > C2
(11-5) r(s) = { . A e'" if c, = C2 = A
C sinh (as + c) if C1 < C2,
168 Local Surface Theory Chap.4

By choosing the point from which arc length is measured we may force b = 0
and c = 0 in the above . Thus we get three basic kinds of surfaces.
r(s) = Ae"'
(11-6) {
z(s) = ± f:
,vii - a 2 A 2eiot dt +D
r(s) = B cosh (as)
(11-7) {
z(s) = ± f:
,vf~l---a.,...
2 B= dt + D
2=-s-..in--.h,....
2,....,(,....at"""")

r(s) = C sinh (as)

(11-8) {
z(s) = ± f: 2 c"=""
,vf~l---a""" 2 -co-s-h- dt
2 ~(a-,-) +D
In order that z be defined we must haves< (I/a) ln (1/aA) in Equation
(11-6). By use of the substitution aAe"' = sin;, the integral describing z(s)
may be computed . See Problem 11.4. Geometrically ; measures the angle
between the tangent to «i and the z-axis. This curve is called a tractrix or
drag curve. The terminology drag curve comes from the geometric description
of the curve : walk along the z-axis dragging an object by a rope of length
1/a, starting at the origin with the object at a distance 1/a out the r-axis. The
resulting path of the object is the tractrix . The surface of revolution generated
by the tractrix is called the pseudo-sphereof radius 1/a. The limiting tangent
plane as s--+ (I/a) In (1/aA) is perpendicular to the axis of revolution. See
Figure 4.34. The pseudo-sphere was one of the first models of non-Euclidean
geometry and was discovered by E. Beltrami (1835-1900) in 1868.

FIGURE 4.34

In Equation (11-7) we must have Isl< (I/a) sinh- 1 (1/aB). The result-
ing surface is pictured in Figure 4.35. The limiting tangent planes as
s--+ ±(1/a) sinh- 1 (1/aB) are perpendicular to the axis.
In Equation (11-8) we must have O < aC < I and then

O< s < ! cosh- 1 (

0 ~)·
Sec. 4-11 Surfaces of Constant Curvature 169

The resulting surface is pictured in Figure 4.36. Again the limiting tangent
planes are perpendicular to the axis.
In analogy with the case of positive curvature, the surfaces of constant
curvature -a 2 < 0 are locally isometric to the pseudo-sphere of radius I/a .
This result will follow from a more general result to be proved in Section 6-2.

FIGURE 4.35
c

FIGURE 4.36

PROBLEMS

11.t. Let M be the surface of revolution generated by the unit speed curve
ci(s) = (r(s), z(s)) and show that K = -r"/r. (Hint: First show
K = z'(z''r' - r"z')/z and then use the relationship obtained by dif-
ferentiating <ci', ci') = l.)
11.2. In Case 1, if A > l/a, verify that as s--+ ±(I/a) sin- 1 (1/aA) the
tangent to ci becomes perpendicular to the z-axis. Use this to conclude
that the limiting tangent planes are perpendicular to the axis as
claimed.
11.3. In Case I, if A < 1/a, determine the angle between ci' and the z-axis
as s --+ ±n/2a.
11.4. Use the trigonometric substitution aAe"' = sin ¢, to compute
f~,,/1 - a 2 A 2e2.., dt .and then parametrize the tractrix as a function
of,f,.
180 Local Surface Theory Chap.4

11.5. Verify that if K = -a 2 then Equation (11-4) does give the general
solution to K = -r"/r. (You must be able to solve linear differential
equations with constant coefficients.)
11.6. Derive the basic forms of Equation (11-5) from Equation (11-4).
11.7. Verify that the limiting tangent planes for Case 3 behave as described.
For the surface given by Equation (11-8) determine the angle between
the meridian and the axis of revolution at the sharp point.
11.8. Prove that the surface :x:(s,t) = (cos s, 2 sins, t) has constant Gaussian
curvature but is not a surface of revolution. Hence the results of this
section do not give all surfaces of constant curvature.
11.9. Compute the area of each of the surfaces in Cases 1 and 3 in terms of
the arbitrary constants.
 5

Global Theory
of Space Curves

In this chapter we will prove analogues of the Rotation Index Theorem

(Theorem 2.9 of Chapter 3) for space curves. For a quick overview you should
reread Section 3-6 at this time. The notion of the tangent spherical image of a
curve (Problems 4.15-4.21 of Chapter 2) is critical for this discussion because
the total curvature of a curve is the length of the tangent spherical image of
that curve. Anyone wishing to delve further into the material covered in this
chapter should see Fenchel [1951], who gives a good survey of related
questions.

5-1. FENCHEL'S THEOREM

DEFINITION. The total curvature of a regular curve «i: [O,LJ - R 3 is f; K ds.
Since K = IT' I, the total curvature of «i is the length of the tangent
spherical image
T: [0, LJ- S2 = {(x, y, z) E R 3 lx 2 +y + z
2 2 = I}= {a E R 3 llal = I}.
Thus it is not surprjsing that we need some results about sphere curves in
order to prove Fenchel's Theorem on the total curvature of a space curve.
161
162 Global Theory of Space Curves Chap.5

DEFINITION. If A, BE S2, then ABdenotes the distance from A to B along

any shortest geodesic (i.e., along a great circle through A and B).

DEFINITION. The open hemisphere with pole N is {XE S 2 INX< x/2}. If S 2

is viewed as the set of unit vectors, the open hemisphere with pole n is

{x E S 2 I(n, x) > O}= {x E S 2 IO< ~ (n, x) < ; }

where ~(a, b) denotes the (positive) angle between a and b.
The closed hemisphere with pole N is {X E S 2 j NX< x/2}. In vector
notation, the closed hemisphere with pole n is

{x E S 2 j(n, x) > O}= {x E S2 IO< ~(n, x) < ~} ·

DEFINITION. If A, BE S2,then A and Bare antipodal points if AB=x. In

vector notation, a and bare antipodal if a= -b. (See Figure 5.1.)

FIGURE 5.1

LEMMA 1.1. If «i is a regular closed curve, then its tangent spherical image
does not lie in any open hemisphere.

Proof· We use vector notation and give a proof by contradiction. Suppose

that the tangent spherical image lies in the open hemisphere with pole a,
so that ( a, T) > 0. Then

0< f (a, T) ds = f (a, of) ds = f (a, ll)' ds = (a, ll) 1:= 0

(since°' is closed), which is impossible. I
COROLLARY 1.2. The tangent spherical image of a regular closed curve does
not lie in any closed hemisphere unless it lies in the great circle that
bounds the hemisphere.

Proof· Problem I. I. I
Note that the next lemma does not require that the curve 'Y be regular.
Sec. 5-1 Fenchel's Theorem 183

LEMMA1.3. Let "f(I) be a closed C 1 curve on the unit sphere S2. The image
e of "f is contained in an open hemisphere if either of the following two
conditions hold:
(a) the length I of "f is less than 2n; or
(b) I= 2n but the image of "f is not the union of two great semicircles.

Proof:
(a) Assume I< 2n. Let A and B be two points of e which divide "f into
two segments of equal length //2. We say A and B bisect "f. AB:::;;:
1/2 < n since
geodesics (great circles) give the shortest distance between points. Thus A
and B are not antipodal and there is a unique great circle through A and B
(formed by the intersection of the sphere with the plane determined by A,
B, and the center of the sphere). The shorter of the two segments of this circle
has length AB.Let M be the midpoint of this segment. We will show that e
lies in the open hemisphere with pole M. This means we must show that if
X is a point on e then XM< n/2.
Since the spherical distance from X to M is a continuous function and
since AM< n/2, if there is a point Yon e whose distance from Mis not less
than n/2, the Intermediate Value Theorem of calculus (see Fulks [1969D

Y, e e with -
implies that for each real number r with AM< r < 'YMthere is a point
Y,M = r. In the next paragraph we show that if C e e with

-
CM< n/2 then, in fact, CM,::;;:
Y,M = r if //4
//4. This says that there is no Y, e e with
< r < n/2. This contradiction of the Intermediate Value
Theorem implies there is no Y e e with YM ~ n/2.
Suppose C e e with CM< n/2. Let D e S2 be chosen so that C, M,
Dall lie on the same great circle, D =/:.C, and CM=MD.(See Figure 5.2.)
D is called the reflection of C through M. Note that CD= CM+ MDsince
CM< n/2 and that AC= BDsince the spherical triangles AMC and BM D
are congruent. Hence 2CM= CM+ MD= CD:::;;: CB+BD= CB+ AC

FIGURE 6.2 A
184 Global Theory of Space Curves Chap.6

< distance along "f from A to Cto B = //2. Thus CM< //4. Hence there is
no Yon the image of "f with YM~ n/2 and (a) is proved.
(b) Now assume that I= 2n. This time we must choose A and B more
carefully. We first show that there is no pair of antipodal points one. If P
and Q are on e and are antipodal, then P and Q divide "f into two segments
whose lengths are at least 'IT.since PQ= n. Since I = 2n, each of these seg-
ments must have length 'IT.and must give a curve of shortest length joining
P to Q, and so must be a geodesic by Theorem 5.9 of Chapter 4, i.e., a great
semicircle. But this means that e is a union of two great semicircles, which
contradicts the hypothesis. Thus there are no antipodal points on e.
Next we show that there is a pair of points A, B that bisect "f such that
AF+iii< 'IT.for all Fe e. Suppose P, Q bisect "f· If P, Q have the desired
property (i.e., if we can take A = P and B = Q) we are done. Otherwise there
is a point X on e with PX+ XQ= n. Since the distance along "f from P
to X to Q is also n, each of the two segments PX and XQ are great circle
segments (geodesics). They are not part of the same great circle since P and
Q are not antipodal. (See Figure 5.3.)

c FIGURE 6.3

Consider the great circle segment XP. We know that it does not continue
along e past X. However, it may continue along e past P. Let C be the point
of the e beyond which this great circle segment does not continue in e. Like-
wise the great circle segment XQ extends to a point D. a< 'IT.and XD < 'IT.,
Let Y be chosen so that X and Ybisect "f. If X and Y serve as A and B, we are
done. Otherwise there is a point R on e with XR+ RY= n. Then the seg-
ment of "f from X to R to Yis the union of two great circle segments as before.
Thus R = C or R = D. Without the loss of generality we may assume that
R = D. The great circle segment D to Yin e extends to a point E.
If E = C, then "f is a spherical triangle of perimeter 2n. Then the solid
angle at the center O of the sphere has 2n as the sum of its face angles (see
Figure 5.4). Hence it is flat and "f is a great circle, a contradiction. Thus
E=t:.C.
Sec. 5-1 Fenche/'s Theorem 186

FIGURE 5.4

Let A be a point on e between C and E and let B be chosen so that A,

B bisect T· Note that B must be between D and X. Then no matter what point
Fon e is chosen, the segment of T from A to F to B has either D and Y or C
and X on it. Thus this segment is not the union of just two great circle seg-
ments . Hence AF+ Fil< 1Cand we have the desired A and B.
A and Bare not antipodal . Let M be the·midpoint of the unique great
circle segment of length AB< 1Cjoining Cis on e with CM~ n/2,
A to B. If
let D be its reflection through M as before . 2CM = CM+ MD = CD
< CB+ BD= CB+ AC < 1C by choice of A and B. Thus CM<n/2 and
there are no points on e whose distance from Mis n/2. Since AC< n/2, all
points Xe e satisfy XM< n/2, or else the Intermediate Value Theorem
implies there are points one at a distance n/2 from M. Hence e lies in the
open hemisphere with M as pole. I

THEOREM 1.4 (Fenchel, 1929). The total curvature of a closed space curve II
is at least 2n. It equals 2n if and only if II is a plane convex curve.

Proof · Let T be the tangent spherical image of 11. Its length is ft K ds. If this
is less than 2n, Lemma 1.3 shows that the image of T lies in an open hemi-
sphere, which contradicts Lemma I.I. Thus ft K ds :2::. 2n.
Now assume that ft K ds = 2n. Then T is the union of two great semi-
circles by Lemma 1.3. If y is not a single great circle, the image of T lies in a
closed hemisphere but not entirely in the boundary, which contradicts Corol-
lary 1.2. Thus T is a single great circle. By Problem 4.20 of Chapter 2, 11 is a
plane curve . ft K ds measures the length of the parametrized curve T and
counts those segments traversed more than once with appropriate (positive)
multiplicity . Since the image set of Tis a great circle whose length is 2n, no
segment of this circle can be traversed twice, or else ft K ds > 2n. If the plane
curvature k = dfJ/dsof II takes on both positive and negative values, then
some segment of the tangent spherical image Tis traversed at least twice, a
contradiction . Hence k does not change sign and , by Theorem 3.2 of Chapter
3, 11 is convex. I
166 Global Theory of Space Curves Chap. 5

Fenchel's Theorem is also true for piecewise smooth curves provided that
the total curvature is defined as f 1e ds + L fj.(}" where the !,.(} 1 are the (posi-
tive) jump angles of T at the junction points. The tangent spherical image 'Y
of such a curve consists of a collection of disconnected segments. These may
be joined by great circle segments of lengths !,.(} 1 to get 11,
whose length is the
total curvature. Lemma 1.1 still holds with virtually the same proof. Lemma
1.3 holds for piecewise smooth curves, in particular p. (A piecewise smooth
curve may be parametrized to be C 1 but need not be regular or unit speed.) It
is possible to have piecewise smooth convex curves, although we have not
made a formal definition. The proof of Theorem 1.4 can be modified to
handle this case. We have not done this because there is a tendency to get
lost in the technicalities.
The proof of Fenchel's Theorem that we have given is due to B. Segre
[1934]and to H. Rutishauser and H. Samelson [1948].A proof due to K. Voss
[1955] may be found in Laugwitz [1965].This proof involves surrounding the
curve with a tubular surface and relating f 1e ds to ffx~o K dA, where K is the
Gaussian curvature of the surface. Both Fary [1949] and Milnor [1950] gave
proofs of Fenchel's Theorem while studying the total curvature of knots.
In Problem 2.3 you will prove the easy part (f 1e ds > 2n) by using Crofton's
Formula.
Fenchel's Theorem is also true for curves in Rn, as has been proved by
Borsuk [1948]. In fact, it was Borsuk who first conjectured the Fary-Milnor
Theorem.

EXAMPLE 1.5. We verify Fenchel's Theorem in the special case where e is the
ellipse ct(t) = (2 cost, sin t, 0) for O < t < 2n. Ifs is arc length one,

i e 1Cds =
r 2"
Jo 1C(s(t)) it dt
d
= 4
rn11e(s(t))dit dt
Jo
2

because of the symmetry of the ellipse. Now

ds = JezI= ,v"4sin 2 t
dt
+ cos2 t.
By the results of Section 2-6, 1e(s(t))= 2/,v"4 sin2 t + cos2 t 3 • Hence

i
e
1eds=4 1" O
12 2dt
4 sin2 t + cos2 t
=4 f"' +
O
2 2 sec2 t dt ·
1 4 tan 2 t
Let u = 2 tan t so that du = 2 sec2 t dt.

i
e
1Cds = 4 r~l +du
Jo U
2 = 4 tan-I U,~
o
= 4~ = 2,i,

Note that e is plane and convex.

Sec. 5-2 The Fary-M/lnor Theorem 187

PROBLEMS

*1.1. Prove Corollary 1.2.

1.2. Let ci(s) be a closed space curve. Suppose O ~ ,c ~ 1/R for some real
number R > 0. Prove that the length of Cl is at least 211:R.
1.3. Compute the tangent spherical image for the ellipse in Example 1.5.
What does the proof of Fenchel's Theorem say about this image?
1.4. Consider the piecewise regular plane nonconvex curve II in Figure 5.5
made up of three circular segments of radius 2. Compute the total
curvature of 11.

(0, v'3)

FIGURE 5.5
c-,.O)w
.. O)

5-2. THE FARY-MILNOR THEOREM

Suppose that Ol is an orientedgreat circle on S2. Then there is a unique

point W e S2 associated with m, namely the pole of the hemisphere to the
left as m is traversed. (See Figure 5.6.) Conversely every point of S 2 is as-
sociated with some oriented great circle. Thus the set of oriented great circles
is in one-to-one correspondence with the points of S2. The orientation is what
allows this to be one-to-one-it gives a well defined choice of a pole (from two
possibilities) for each oriented great circle. If the north pole corresponds to
the equator with one orientation, the south pole corresponds to the equator
with the opposite orientation.

FIGURE 5.6

DBFINmON. The measure of a set of oriented great circles (counted with

multiplicity) is the ar~a of the corresponding subset of S2 (counted with
multiplicity).
188 Global Theory of Space Curves Chap.5

If WE S 2 , let W.L be the associated great circle. For a regular curve 1

with image set e, let n,(W) be the number of points in en W.L(which might
be infinite). Note that n,(W) does not depend upon the parametrization of y.
We shall usually omit the subscript 1 in n,(W) if no confusion will result.

THEOREM 2.1 (Crofton's Formula). Let e be the image of a regular curve y(t)
on S2 of length /. The measure of the set of oriented great circles that
intersect e, counted with multiplicity, is 4/.

Proof· Since neither n( W) nor I depends on the parametrization, we may

assume that 1 is unit speed. Let S ={WE SZI W.L n e *
0}. The measure
of the set of oriented great circles that intersect e is the area of S; that is,
Ifs dA. We wish to count the circles with multiplicity-if en W.L has three
points, we want to count W.L three times. This measuring is done by the
integral Ifs n(W) dA. We are to prove Ifs n(W) dA = 4/.
Let a(t) = y(t), b(t) = dy/dt and c(t) = a x b. Then analogously to the
Frenet-Serret equations, for some function l(t) we have the equations
a'= b
b' = -a + l(t)c
c' = - l(t)b.
(See Problem 2.1.) a, b, and c give an orthonormal basis of R 3 for each
t E [O,/]. If W.L meets eat y(t 0 ), then the unit vector w which represents the
pole of W.L is orthogonal to y(t 0 ) = a(t 0 ). Hence
w = -sin~ b(t 0 ) + cos~ c(t0 ),
where ~ is the angle between the tangent vector b of 1 and the tangent vector
at
of w.L y(t o>·
Define a map w: [O,/) x [O,2n) - S2 by
w(t, ~) = -sin~ b(t) + cos~ c(t).
The image of w is S. This makes S a parametrized surface. Despite the fact
that w might not be regular (i.e., w 1 x w 2 might be zero sometimes), the area
of the parametrized surface S is given by

fz..f' Iow x I
OWdt d~.
Jo Jo Tt ~
just as for a regular surface (Equation (8-1) of Chapter 4). However, this is
not the area of the set S, but rather the area counted with multiplicity (w need
not be one-to-one either). Now, if YE S, precisely n(Y) points in [O,/)
x [0, 2n) are mapped to Y. Thus

JLes
n(Y)dA = iz.S:li X ~,dt~.
Sec.5-2 The Fary-Milnor Theorem 169

This last equality can also be derived from Jacobi's rule for change of variables
(Fulks [1969]). By Problem 2.2 we have
awx att,
lai .
awl= lsmt/JI.
Thus
ffsn(Y) dA = fo2,rf:Isin ef,Idt def,= If"lsinef,I def,
=
f"'2
41 Jo sin if,def,= 4/. I
COROLLARY
2.2. Crofton's Formula holds for piecewise smooth curves.

Proof· Use Theorem 2.1 on each segment of the curve and add. I
DEFINITION.D 2 = {(x, y) E R 2 Ix 2 + y 2 < I} is the closed unit disk. S 1 =
{(x, y) E R 2 Ix 2 + y 2 = I} is the unit circle.

DEFINITION.A simple closed curve~ is unknotted if there is a one-to-one con-

tinuous function g: D 2 ----+ R3 that sends S 1 onto the image of~. Other-
wise ~ is knotted.

THEOREM 2.3 (Fary-Milnor). If~ is a simple knotted regular curve, then the
total curvature of~ is at least 4n.

f
Proof: We shall prove the contrapositive: IC ds < 4n implies~ is unknotted.
f
Assume IC ds < 4n. Let "f be the tangent spherical image of ~. Then
!ffs• ny(W) dA =I= f IC ds < 4n. Then for some point YE S2, n(Y) < 4,
ff
otherwise ! n( W) dA > !4 • 4n = 4n. Let y be the unit vector representing
Y and set f(s) = (y, ~(s)). Then f'(s) = (y, T(s)) = (y, r(s)) and f'(s) = 0
if and only if Y.l intersects eat r(s). Thusf(s) has exactly n(Y) < 4 critical
points. There is an absolute maximum, an absolute minimum, and possibly
an inflection point (if n( Y) = 3). There are no relative extrema other than
the absolute extrema.
We may assume that coordinates for R 3 have been chosen so that
y = (0, 0, 1). If M and m are the maximum and minimum values off and
m < r < M, the plane z = r intersects ~ in exactly two points, or else the
Mean Value Theorem of calculus would imply some more relative extrema.
Join these two points with a straight line segment. The totality of these seg-
ments gives the image of a one-to-one continuous function g: D 2 ----+ R3
making~ unknotted. See Figure 5.7. I

The above proof is essentially that outlined by Fenchel [1951]. The

theorem holds for piecewi~e regular curves, but this proof does not work.
(Why?) Milnor (1950] proved the theorem by approximating the curve with
170 Global Theory of Space Curves Chap.5

FIGURE 5.7

piecewise regular curves made up of straight segments. He used very geometric

methods to compute the total curvature and then took the necessary limits.
Fary [1949] also approximated at by piecewise linear curves. One important
step in his proof was the result that the total curvature of at is the average of
the total curvature of the projections of at into the planes through a fixed
point. Voss [1955] also outlines a proof.

PROBLEMS

*2.1. Prove that a' = b, b' = -a + le, c' = -lb, as claimed in the proof
of Theorem 2.1.
*2.2. Prove that l(ow/ot)x (ow/oif,)I
= lsinif,I as claimed in the proof of
Theorem 2.1.
2.3. Use Crofton's Formula to prove that f IC ds > 2n for all closed regular
curves.
2.4. Letat(s) be the curve on S 2 given by at(s) = (cos s, sins, 0), 0::::;;s::::;;2n.
Verify that Crofton's Formula is true for this curve.

5-3. TOTAL TORSION

After considering the total curvature of a space curve it would seem

natural to investigate the integral f T ds. Results here are not as strong as for
f IC ds. In fact, if r E R, there is a closed curve at with f -rds = r. To see this,
one need consider curves made up of a circular helix with the ends joined, as
in Figure 5.8. By varying the pitch, the null\ber of coils, and the right- or
left-handedness of the helix, any desired total torsion may be found .
Despite this, there are some results on total torsion worth studying.
Recall that p = I/ICis the radius of curvature and that a = 1/'ris the radius
of torsion of the curve at.

THEOREM 3.1. The total torsion, f -rds, of a closed unit speed curve at(s) on
S 2 is zero.
Sec. 5-3 Total Torsion 171

FIGURE 5.8

Proof· ci(s) = lT + µN + vB. Since Cl is a unit vector,

l = <11,T) = !(11, 11)' = 0.
Thus 11= µN + vB. T = 11' = µ'N + µ(-,cT + -rB) + v'B + v(--rN).
Thus T = -µ,cT + (µ' - -rv)N + (µ-r + v')B. Hence I = -µ,c or
µ = -1/,c = -p, and µ' - -rv= 0. Since <11,11) = I, µ 2 + v2 = I and
v = ±,JI - µ 2 = ±,Jl - p 2 •
Hence -p' = ±-r,Jl p 2 and -r = ±p'/,Jl - p 2 (when p =t=1). If
p < I always so that p'/,Jl - p 2 always makes sense, and if a constant
choice of sign in -r = ±p'/,Jl p 2 could be made, then -r = ±p'/,Jl p2
everywhere. In this case

iL-rds = ± iL,Ji'~ p 2ds = ±sin- 1 p(s{ = 0

since p(L) = p(O). However, this may not be true. In fact, we shall see below
that p = I whenever the geodesic curvature of Cl is zero, and the choice of
sign depends on the sign of the geodesic curvature.
+ TC.S where S = n x T. If we choose the inward
We have 11" = TC,1.1
pointing normal to S2 as n, then n = -Cl and "• = +I.
"• = <S,11") = <n x T, 11") = -<11 x T, T')
= -<11, T x ,cN) = -,c<ci, B).
On the other hand, <11,B) = -p'u so that "• = ,cp'u or 1' = ,cp'/,c6 if
1' =t=0. Since ,c2 = ,c.2 + ,c6 2 = I + ,c6 2 , we have "• = ±,J,c 2 - I and
"•'" = ±,Jl p 2 • Thus -r = ,cp'/,c6 = p'/(,c.f,c) = ±p'/,Jl p 2 and the
sign ± is that of"•· Note also that p = I if and only if ,c = I, which is true
if and only if,c 6 = 0. Thus -r = ±p'/,.;r=:,;i whenever"• =t= 0 and the sign
is that of"•·
To simplify the proof we shall assume th~re are a finite number of points
0 = So < S1 < ... < s. = L where"• is zero and such that on (s,-i, s,) "• is
either strictly positive, ·strictly negative, or identically zero. (This finiteness
172 Global Theory of Space Curves Chap. 5

need not be true. In order to handle the general situation one needs some
topology and a more sophisticated concept of integration. However, the
basic idea is the same.)
J: T ds = I: J.:\
T ds. At each s11 1C6 is zero, ,c = I and thus p(s1) = I.
If 1C6 > 0 on (s1_ 1, s,), then T = p'/,Jl - p 2 and J.::,-r ds is the improper
integral

J~,JI_p·
1,-1 p
2 ds = sin- 1 (p(s))
I~
,,...
1
= (~ - ~) = O.
Likewise if ,c• < 0 on (s1_ 1, s,), then J.:~.
T ds = 0. If ,c• 0 on (s,_1, s,)
then ,c = 1 on (s,_1, s,). Problem 5.3 of Chapter 2 implies that this segment of
=
II is part of a circle and hence is a plane curve. Thus T =
0 on [s1_ 1, s,] and
J.::,T ds = 0. Therefore J;
T ds = 0. I

W. Scherrer [1940] has proved the converse of this theorem: if M is a

J
surface in R 3 such that T ds = 0 for all closed curves in M, then Mis part of
a plane or part of a sphere. B. Segre [1947b] has studied the total absolute
torsion J
ITI ds and has obtained some results on lower bounds on the
integral.

PROBLEM

J
3.1. Prove that (-r/,c) ds = 0 for any closed unit speed sphere curve. (This
result and its converse is due to B. Segre [1947a].)
 6

Global Theory
of Surfaces

After the brief hiatus of Chapter 5 we return once again to surfaces. We

now study results about surfaces which are of a global nature-that is, they
do not just depend on the behavior of a surface in a neighborhood of a point
but rather on the surface as a whole. These results are very much in.the spirit
of modern research in differential geometry. We get quite unexpected rela-
tionships between quantities that seem totally different (e.g., Euler char-
acteristic and total curvature), especially mixing results of a topological nature
with those of a geometric one. One word of caution: the concepts in this
chapter are more difficult than those of Chapter 4 because we are dealing
with global concepts. Note how much more difficult Chapters 3 and 5 were
than was Chapter 2.
In the first section we define basic topological concepts and prove some
simple results to give the reader the flavor of the kinds of theorems that we
are aiming at. Sections 6-2 and 6-3 are rather technical, giving some results
(geodesic coordinate patches) and introducing concepts (orientability and
angular variation) which will be needed to prove the Gauss-Bonnet Formula
in Section 6-4. This formula is truly beautiful because it relates the total
geodesic curvature of a curve to the sum 01'the jump angles of the curve and
the total Gaussian curvature of the area it encloses-seemingly unrelated
things! The last three ·sectionsprove corollaries of the Gauss-Bonnet Formula,
173
174 Global Theory of Surfaces Chap. 6

all of which are important theorems in global differential geometry (the

Gauss-Bonnet Theorem, Hadamard's Theorem, Jacobi's Theorem, and the
Poincare-Brouwer Theorem). It is startling relationships of this sort between
unexpectedly related quantities like these which give global differential
geometry its fascination.

6-1. SIMPLE CURVATURE RESULTS

In this section we will be concerned with some simple results relating the
curvature of the surface to its "topology." The first theorem (Theorem 1.2)
shows how a topological assumption (compactness) yields a geometric con-
clusion (that the curvature must be positive at some point). The second
theorem (Theorem 1.3) shows how a geometric assumption (all points are
umbilics) can yield an extremely strong conclusion (Mis a sphere).

DEFINmoN. If P E R3, then !Pl denotes the distance from P to the origin
O of R 3 • (This agrees with the vector notation: if pis the vector from O
to P, IPI = IPI-)

DEFINITION. Let B, ={PE R 3 I IPI::;;;:r}. Mc: R3 is bounded if there is an

r > 0 such that M c: B,. M is closed if for each sequence {Pnl such that
Pn E Mand Jim Pn = P exists (in R3 ), Pis actually in M.
n~oo

Toe ball B = {P E R 3 I IPI < l} is bounded but not closed. Toe (x, y)
plane in R 3 is closed but not bounded. Observe that if Mis bounded, then (1)
there is a minimum r such that Mc: .B,,and (2) if {Pn}is a sequence in M,
then {Pn}must contain a subsequence which converges in R 3 •

DEFINmON. If Mc: R 3 is closed and bounded, then Mis compact.

LEMMA I.I. Let M be a compact surface in R 3 and r the minimum number

such that Mc: .B,.Then there is a point P E M with IPl= r, i.e.,
Mn S, =!=.0, where S, is the sphere of radius r.

Proof: Let rn = r - l/n for n > 0. By the choice of r we have M - B,.=!=.0.

Let Pn E M - B,•. Since Mis bounded and Pn EM, there is a subsequence
(which we also name Pn) which converges to P E R 3 • Since M is closed,
PE M. Clearly, Iim IPnl = r so that IPI = r. I

THEOREM 1.2. Let M be a compact surface in R 3 • Then there is a point P E M

such that the Gaussian curvature at P is positive.

Proof: Let r be the minimum number such that Mc B,. By Lemma 1.1 we
Sec. 6-1 Simple Curvature Results 175

know that the sphere S, of radius r and center O satisfies S, n M =fa0. We

will show that M has positive curvature at all points of S, n M. Let
p E S, n M. M and S, have the same unit normal n at P, or else there is a
point of M outside S,. (This argument is similar to that about parallel lines
in the proof of Proposition 3.2 of Chapter 3.) Hence Mand S, have the same
tangent plane at Palso. Let X be tangent to both Mand S, at P. Let II be the
plane spanned by n and X. The normal curvature Kn of Min the direction X
is the curvature (with respect ton) of the curve II n M. This has the same
sign as the normal curvature of S, at P, which is ± 1/r. Clearly IKn I :2: 1/r. If
Ki, K2 are the principal curvatures of Mat P we have K = K1K2 > l/r 2 > 0.
See Figure 6.1. I

IlnS,
FIGURE 6.1

This theorem is false without the assumption that M is compact. If

M = R2 c R3 , then the Gaussian curvature is zero everywhere. If M is the
hyperboloid of revolution x 2 + y 2 - z 2 = 1 (Example 2.8 of Chapter 4),
then the Gaussian curvature is negative everywhere so that we cannot even
hope for K = 0 at some point.

THEOREM1.3 (Meusnier, 1785). Let M be a compact connected surface all of

whose points are umbilics. Then M is a sphere.

Proof· We first prove that the curvature is constant in any coordinate patch.
R3 be a coordinate patch of M. Then since every direction is
Let x: CU.----+
principal at an umbilic, (L',) = (K'
O
O)· Weingarten's equations ((7-6) of
K1
Chapter 4) are thus n, = -K 1x 1• Hence n;1 = -(iJKifiJu')x, - K 1X11 and
0 = 0 12 - n21 = (iJK1/iJu1)x 2 - (iJKi/iJu2)x 1 • Since x 1 andx 2 are independent,
iJKi/iJu' = 0 and K 1 is constant. Hence K = K 1K 1 is constant.
In overlapping patches K must have the same (constant) value. Thus K
is constant on M. By Theorem 1.2, K > 0 and K 1 =fa0. In a patch x we have
n = f n 1 du1 = -K, J x 1 du 1 = -K 1(x - c). Take the length of each side
of this equation to obtain I = K(x - c, x - c) so that the patch is con-
178 Global Theory of Surfaces Chap. 6

tained in a sphere of radius 1/,./ K centered at c. An overlapping patch must

be contained in a sphere of radius 1/,./K and centered at some point d. The
overlap of the two patches is a simple surface and lies in both spheres. Hence
c = d. Thus Mis contained in a sphere. Using compactness, we see that M
is the entire sphere. I

PROBLEMS

1.1. If Mis a surface in R 3 each of whose points is an umbilic, prove that

each point of M has a coordinate patch whose image is either an open
subset of a plane or an open subset of a sphere. (Hint: Examine the proof
of Theorem 1.3 carefully.) What invariant differentiates between these
two cases?
1.2. Parametrize the level surface x 2 + y 2 - z 2 = 1 as a surface of revolu-
tion. Compute its Gaussian curvature and show that it is negative at
each point.

6-2. GEODESIC COORDINATE PATCHES

In this section we will show that any point of a surface M is contained

in a special kind of coordinate patch, called a geodesic coordinate patch.
This kind of patch is very useful for the theoretical computations involved in
proving the Gauss-Bonnet Theorem. (See Sections 6-4 and 6-5.)

DEFINmoN.If x: '11- R 3 is a coordinate patch such that g 11 = 1 and

=
g 12 0 (hence g 21 = 0) in '11, then x is called a geodesic coordinate
patch. If, in addition, there is a curve 1 on M defined on [a, b] such that
y([a, b]) c: x('ll) and such that the u 2 -curve through a point of the image
of 1 is 1 itself, then x is called a geodesic coordinate patch along 1 ·

In Problem 5.5 of Chapter 4 we saw that the u 1-curves in a geodesic

coordinate patch are geodesics. We will obtain a geodesic coordinate patch
by starting with any nonclosed curve II and constructing geodesics perpendi-
cular to II which fill out a neighborhood. (See Figure 6.2.) This will mean that
the u 1-curves are geodesics and the u 2 -curve through any point of Cl is 11.

PRoPosmoN2.1. Let M be a surface and let 11: [a, b] - M be a simple

regular curve which is not closed. Then there is a geodesic coordinate
patch x: '11- R 3 of M along 11.

Proof: We may assume· that Clis defined on (c, d) with c <a< b < d.
Choose e,f e R so that c < e < a <b <! < d. For each t, let X(t) = -S,
Sec. 6-2 GeodesicCoordinatePatches 177

FIGURE 6.2

where S is the intrinsic normal n x 4/Icz I and n is a continuous choice of a

unit normal to M along iz. (Such a choice of n can be made because iz is not
closed.) For each t e [e, /] let iz,(r) be the unique geodesic satisfying
iz,(O)= iz(t) and (diz,/drXO)= X(t). Let x(r, t) = iz,(r). xis of class C3 because
of the smooth dependence of the solution of differential equations upon a
smoothly varying parameter t. (Recall that a geodesic satisfies a differential
equation: Equation (5-1) of Chapter 4.) Because [e, /] is a closed interval,
there is a t5> 0 such that x is one-to-one on [-'5, t5] x [e, /].
x 1(0, t) = X(t}, xi{O, t) = diz/dt.
x 1 x xi{O, t) * 0 since X and diz/dt are orthogonal and nonzero. Thus
*
there is an f > 0 with f ~ t5 such that x 1 x x 2 0 on [-f, e] x [e, /].
Then x: (-f, e) x (e, /)-+ R 3 is a C 3 coordinate patch containingiz([a, b]).
g 11 =1 since x 1 = 8iz,/8r and iz,(r) is a geodesic with r arc length.

5;
8 2 = f,(x1, X2) = (Xu, X2) + (X1, X21)

or
(2-1)

Now Xu equals 8 2iz,/8r 2, which is normal to M since iz, is a geodesic. Thus

(xw x 2) = 0. Since (xi, x 1) = 1, we have 2(x 12 , x 1) = 0 and hence, by
Equation (2-1), 8g12 /8r = 0. This means that g 12 is constant with respect to
r so that g 12 (r, t) = gu(O, t) = (cz,
X) = 0. Therefore g 12 =0 and x is a
geodesic coordinate patch (with u• = r and u 2 = t). The u 2-curve through
iz(a) = iza(O)= x(O, a) is x(O, a + t) = iz.+,CO)= iz(a + t}, i.e., is iz. I

We note that the first fundamental form in the coordinate patch has as
. . (1 h0)
its matrDC O 2 , where O < h = Ix 2 I.

In the special case where iz in Proposition 2.1 is a (unit speed) geodes-

ic, the function h = Ix 2 I has two important properties. In the first place,
h(O,u2) = IX2(0,u 2) I ==Iiz'(u 2) I = 1. Secondly, since iz is a geodesic,
178 Global Theory of Surfaces Chap.6

x 22(0, u 2) = a."(u2) is normal to the surface. Thus

ach 2) a
aui = Tzii.(Xz, Xz) = 2(Xm Xz) = 2(x 12 , x 2)

= 2(b<x1, X2) - (x1, X22)) = -2(x 1, X22).

Hence at u 1 = 0 we have (a(h2)/au 1)(0, u2 ) = 0 and hence (ah/au 1)(0, u2 ) = 0.

These results, that h(O,u 2) = I and (ah/au1)(0, u2) = 0 if a. is a geodesic,
are very useful in proving the next theorem, which gives a nice application of
geodesic coordinate patches.

THEOREM 2.2. Any two surfaces of the same constant Gaussian curvature
K =c are locally isometric.
Proof: Suppose Mand N both satisfy K = c. Let P E M, X E T pM, Q E N,
and Y E TaM, where X and Y are unit vectors. Let er.:(-71, 71)-+ Mand
'Y: (-71, 11)-+ N be the unique geodesics such that cr.(O)= P, a.'(O) = X,
y(O) = Q, and y'(O) = Y. Construct geodesic coordinate patches
x: (-f, f) x (-11, 71)-+ Mandy: (-f, f) X (-71, 71)-+ N along a. and 'Y
respectively, denoting the parameter in (-f, f) by t and that in (-71, 11)bys.
The metric matrices for x and y are

(g,,) = (~ O) and (h,,) = 1

(0
eZ
where e and/ satisfy
e(O,s) = f(O, s) = I and ae (0, s) = Tt(O,
at aJ s) = 0.

According to Problem 2.3, the Gaussian curvature of x and y are given

by K = (-a 2 e/at 2 )/e and K = (-a 2J/at 2 )/f, respectively. Hence, for each
fixed value of s, e and f both solve the initial value problem
Q= -cu,
(2-2)
u(O) = I, u(O) = 0.
By Picard's Theorem, Equation (2-2) has a unique solution. In fact, it is
given by
cos (../ct) if c > 0
(2-3) u(t) = { 1 if c = 0
cosh (,./=ct) if c < 0.
In particular, e = f and there exist coordinate patches x: 'l1 -+ M, y: 'll -+ N
such that the metric matrices are the same. Hence, M and N are locally iso-
metric by. Proposition 10.5 of Chapter 4. I
Sec. 6-2 Geodesic Coordinate Patches 179

It should be noted that the above proof gives more than is required. In
fact, it shows that for surfaces of the same constant curvature if there are
given any two points P and Q in M and N and geodesics tz and T through
P and Q, then there is a local isometry f of a neighborhood 'll' of P with a
neighborhood 'U' of Q such that f (P) = Q and T = f o ,z, In particular, if
M = N there is a local isometry mapping P to Q (a generalized "translation")
and if P = Q there is a local isometry sending a given geodesic through P to
a second given geodesic through P (a generalized "rotation").

ExAMPLE 2.3. The standard parametrization of a surface of revolution is a

geodesic coordinate patch along a circle of latitude.

PROBLEMS

*2.1. Prove the assertion in Example 2.3.

2.2. Suppose the curve tz used in Proposition 2.1 is a geodesic. Prove that
along this curve (g11) = (6u), Are the other u2-curves geodesics, or even
unit speed, in general?
*2.3. Let x be a geodesic coordinate patch with (g,1) = ( ~ Z and h > 0.
2)

Show that r 12 2 = r 21 2 = h1/h, r 22 1 = -hh 1, r 22 2 = h2/h and all

other r,/ are zero, where h, = oh/ou'.Show that K = -h 11 /h.
2.4. (Hyperbolic Half Plane.) Suppose there is a simple surface x: 'll- R3
with
1
(
y2
'l1 = {(x,y) E R 2 ly > O} and (g,1) = 0

Show that r 11 2 = -r 12 1 = -r 21 1 = -r 22 2 = l/y and all other r,/

are zero. Compute K. (See the comment below.)
2.5. (Poincare Disk.) Suppose there is a simple surface x: 'll - R 3 with
'l1 = {(x,y) E R 2 jx 2 + y2 < 1}
and

(1 - x~ - y2)2
(g,,) = (
0
Show that
r 22 2-r1-r1-
- 12 - 21 - -
r2-
11 - 1 -x 2y
2
-y 2
180 Global Theory of Surfaces Chap.6

and
r 11 1- - r 12 2- - r 21 2_ - r 22 1- - 2x
- 1 -x 2
-y 2

Compute K.
Comment: These last two exercises are a bit misleading because there is no
function x giving the desired properties. They are examples of abstract sur-
faces. Both are models of hyperbolic geometry, just as the sphere is a model
of elliptic geometry. We shall return to them below. Seealso Chapter 7.

6-3. ORIENTABILITY AND ANGULAR

VARIATION

Let y(t) be a piecewise regular simple closed curve in a surface M with

periodL. Let Z(t) be a continuous vector field along y which is differentiable
*
along the regular segments of y. In general Z(O) Z(L). We shall be inter-
ested in the angle between these two vectors. In the plane we measured angles
counterclockwise from the x-axis. In our more general setting we do not have
an "x-axis" to use as a reference. We must therefore find some reference
''line" from which to compute angles as well as to define angles themselves.
Given any two vectors, the cosine of the angle between them is well defined,
but the angle itself is not. (Recall the problems we had in Chapter 3 in defining
the rotation index.) We need a concept analogous to "counterclockwise" in
the plane. We will do that first by defining orientation.

DEFINITION. A surface M is orientable if there is a continuous function

v: M --+ S 2 with v(P) normal to M at P for all P E M.

Thus for an orientable surface the Gauss map is globally defined.

ExAMPLE 3.1. S 2 is orientable; let v(P) be the outward pointing normal at

P. v(P) = p, where p is the vector from the origin to P.

EXAMPLE 3.2. The torus T 2 (see Problem I. I of Chapter 4) is orientable; let

v(P) be the outward pointing normal at P. See Figure 6.3.

FIGURE 6.3
Sec. 6-3 Orientability and Angular Variation 181

EXAMPLE 3.3. The Mobius band is not orientable. See Figure 6.4 and Prob-
lem 2.5 of Chapter 4.

FIGURE 6.4

EXAMPLE 3.4. Recall the example of a surface mentioned between Examples

2.4 and 2.5 in Chapter 4. Let/: R3 - R be a differentiable function and
(V/)(P) = Ux.fy,fz) be the gradient off (where the derivatives are all
evaluated at P). We mentioned that M = {P E R3 I/(P) = O}is a sur-
face if (V/)(P) * 0 for all P E M. Mis called a level surface off. A level
surface off is always orientable. (Just take v(P) = Vf (P)/ IVf (P) 1-)This
example subsumes most examples we will run into in practice. Level sur-
faces are treated in depth in Chapter 7.

The following is a very deep topological result and is beyond the scope
of this book. A recent proof may be found in Samelson [1969]:

3.5. Every compact surface in R3 is orientable.

THEOREM

An orientation (choice of n globally) aids in measuring angles. If X and

Y are tangent to M at P, the angle from X to Y is defined up to sign by
cos() = <X, Y)/1X 11Y 1-The sign of() is that of [X, Y, n]. This corresponds to
counterclockwise, or the right-hand rule. We write ~(X, Y) for the angle()
from X to Y.

DEFINITION.A subset CRof a surface M is a region in M if CRis open and if

any two points of CRmay be joined by a curve in CR.If CRis a region in
M, then the boundaryof CR,bd CR,is the set {P E MI P ¢ CRand there is
a sequence {Pn}in CRwith Iim Pn = P}. A curve T bounds a region CRif the
image of y is the boundary of CR.and S points into CRat all points of y
while -S points out, where Sis ·the intrinsic normal of y.

EXAMPLE 3.6. Let M = R2 . CR= {(x,y, 0) E R 2 lx2 +y 2 < I} is a region.

The curve t1,({})= (cos(), sin 0, 0) bounds CR.

EXAMPLE 3.7. Let M be the torus T 2 and let CR~ all the points of T 2 except
those on the image ofy, as in Figure 6.5. CRis a region and the boundary
of CRis the image of T but y does not bound CRsince both S and -S
point into <R. ·
182 Global Theory of Surfaces Chap.6

I
I

'\.
\

FIGURE 6.5 FIGURE 6.6

ExAMPLE 3.8. Let M be S2 and let <Rbe the set pictured in Figure 6.6. <Ris
not a region because there is no curve in <Rjoining A to B.

Now in order to measure the total angular change of Z along y we need

to replace the x-axis. Suppose y bounds a region <R.It can be shown that
there is a region &in M containing <Rand the image of y and a field of unit
vectors in&, i.e., an assignment of a unit tangent vector V(P) to each point
P E &,which is differentiable. (See also Problem 3.1.)
Let «(t) = ~(V(y(t)), Z(t)) with« continuous, and hence differentiable.
Once V is fixed, « is unique up to an integral multiple of 2n and so d«/dt
depends only on Z.

DEFINITION. The total angular variation of Z along y with respect to V is

6v« = J:(d«/dt) dt.
In general the total angular variation depends on V. However, if the
curve T can be continuously shrunk in <Rto a point, we will show that 6v«
is independent of V. In this case we shall write 6« for 6~(V(y(t)), Z(t)) for
any choice of the unit vector field V. We shall now make the idea of"shrink-
ing" precise.

DEFINITION. Let "( be a closed curve which bounds a region <R.Let a be any
closed curve of period L which is either y or lies in <R.Let a(O) = x 0 •
a is null-homotopic in <Rif for each s E [O, 1) there is a closed curve a,
in M such that
(a) a,(O) = x 0 ;
(b) a 0 (t) = a(t) and a 1(t) = x 0 (i.e., a 1 is the constant curve);
(c) a,(t) E <Rfor all O < s < 1 and t E (0, L); and
(d) the function r: [O,L] x [O, 1) -- M given by r(t, s) = a,(t) is con-
tinuous.
Sec. 6-3 Orientabilityand Angular Variation 183

The a, represent the shrinking and the fourth condition is a precise state-
ment that the shrinking is continuous. Note that the third condition says
that each of the curves a, must lie in <Rifs > 0 (except possibly for the end
point :x:
0 ). The notion of homotopy is discussed more fully in Singer and
Thorpe [1967).

EXAMPLE 3.9. Let M = R2 and y.(t) = (cos t, sin t), 0 ~ t:::;;;2n. Then <R=
{(x,y)lx 2 + y 2 < l}. That Tis null-homotopic is clear from Figure 6.7.
More formally, if a,(t) = (I - s)y(t) + (s, 0), then a, satisfies the defini-
tion above.

Xo = (\, 0)

FIGURE 6.7

ExAMPLE 3.10. Let y(t) be as above with <R= {(x, y) IO< x 2 + y 2 < l} and
M = R 2 - {(O,O)}.Then y(t) is not null-homotopic. This is intuitively
clear because y(t) goes around the hole (i.e., the origin) and so cannot
be shrunk through it. Notice that an attempt to mimic the procedure of
Example 3.9 fails because a 112 (n) = (0, 0) which is not in <Rand so vio-
lates condition (c).

Because the proof of the next lemma is very technical we shall only
sketch it here.

LEMMA 3.ll. If y bounds the region <Rand is null-homotopic in <R,then

6vr,, does not depend on the choice of V.

Proof: Let W be another field of unit vectors.

r,,= ,t(V, Z) = ,t(V, W) + ,t(W, Z) + 2nn = 8 + p + 2nn.
Hence dr,,/dt= dfJ/dt + dp/dt. We need only show that
ri:.d8
Jodt dt = 68 = 0.

Since V(O) = V(L) and W(O) = W(L), 68 = 2nr for some integer r.
To say that T is null-homotopic means we have a family of curves a,(t)
for O < s < I with a 0 = T, a 1 _. = a very small circle, a,(t) continuous in s
and t, and a,(t) piece:wise differentiable in t. Then 68, = f...(dfJ/dt)dt is
184 Global Theory of Surfaces Chap.6

defined and continuous in s. Since it is always an integral multiple of 2n it

must be constant. But along a very small circle V and W can't change much
relative to each other and so Ml,_,= 0. Then ~O= MJ0 = ~01 _, = 0 and
~va, = ~wP· I
ExAMPLE 3.12. Let M be S 2 and take "I to be the circle of latitude</,= t/,0 so
that T(O) = (sin t/,0 cos 0, sin t/,0 sin 0, cos t/,0 ). We work in the coordi-
nate patch x(t/,,0) = (sin</,cos 0, sin</,sin 0, cos</,). Let

Z(O) = cos ((cos t/,0 )(J)x1 - sin (~~;~to)O)x2

as in Example 6.8 of Chapter 4. Recall that Z(O) is parallel along "I with
Z(O) = x,. Let V be a field of unit vectors on S 2 defined everywhere
except at the south pole S (see Problem 3.2). We will compute
~a,= .q:(V, Z) and show that ~a,= JfmK dA where CRis the cap of the
sphere bounded by "I and K is the Gaussian curvature. The fact that
~a, = JfmK dA is a special case of the Gauss-Bonnet Formula, which we
prove in Section 6-4.
Note that
(3-1) ~rx= y) + ~.q:(y,Z).
~.q:(V,
We first compute ~.q:(y,Z). Now y = xi{t/,0 , O) and g 22 = sin 2 </,so that
cos (.q:(y,Z)) = <!•Z) = -sin_((cos t/,0)(J)g22
ITIIZ I sm tf,0 ,Jg;";

= -sin ((cos t/,0 )0) = -cos ( ~ - (cos t/,0 )0) ·

Thus
.q:(y,Z) = 3n - (cos t/,0 )0
2
and

~.q:(y,Z) = d (3
f dO
2
Jo "; - )
(cos t/,0 )0 dO = -2n cos tf,0 •

Clearly ~1:(V, y) = 2nn for some integer n. y is null-homotopic in CR,

and for a very small circle pit is clear that ~1:(V, P)
= 2n. Hence n = I
and ~1:(V, y) = 2n. By Equation (3-1)
f
~a, = 2n - 2n cos t/,0 = 2n(I - cos t/,0 ) = 2n :·sin</, def,

= f"f :·sin¢, def,dO = fImdA fImK dA

=

since K = 1 on S 2 •
Sec. 6-4 The Gauss-Bonnet Formula 186

PROBLEMS

*3.1. Prove that if y bounds a region CRthat is entirely contained in a single

coordinate patch, then a field of unit vectors exists in CR.
*3.2. Prove there is a field of unit vectors defined on all of the sphere except
the south pole. (Hint: Problem 1.10 of Chapter 4.)
3.3. In Example 6.12 of Chapter 4 we considered parallel translating a vector
along a certain piecewise regular curve. Show that the angular variation
equals the area of the region surrounded in that case .

6-4. THE GAUSS-BONNET FORMULA

DEFINITION.A region of a surface is simply connected if every closed curve

in that region is null-homotopic.

ExAMPLE4.1. Any coordinate patch which has 'l1 equal to a disk or a

rectangle is simply connected.

EXAMPLE 4.2. The entire sphere S2 is simply connected. This is easy to

visualize but difficult to prove. SeeFigure 6.8.

FIGURE 6.8 FIGURE 6.9

EXAMPLE 4.3. The entire torus, 1'2, is not simply connected-a circle going
around either "hole" cannot be shrunk to a point while staying in 1'2.
Neither of the curves cz and ISin Figure 6.9 is null-homotopic .

THEOREM 4.4 (Gauss-Bonnet Formula) . Let y be a piecewise regular curve

contained within a simply connected geodesic coordinate patch and
bounding a region CRin the patch. Let the jump angles at the junctions
be ix" ••• , ix•• Then JJ
111K dA + +
J,.'IC6 ds I; ix, = 2n.
186 Global Theory of Surfaces Chap.6

Proof· We first point out that the geodesic coordinate patch referred to in
the hypothesis is not a geodesic coordinate patch along 'Y.It can't be because
'Yis closed.
Recall that for a geodesic coordinate patch, the metric tensor takes the
form (~ Z2 ), where h > 0 is a function of u 1 and u 2 • By Problem 2.3,
2_ 2 _ h1
r12 -r21 -7i'
r 22 I _
-
-hh 2 _ h2
1, r22 -7i,

r11 1 = r12 1 = r2i1 = ru 2

= 0 and K =
-h
T'
where h, = oh/ou'.
Since all the quantities in the conclusion are independent of parametriza-
tion, we may assume that 'Yis parametrized by arc length. We compute the
angular variation of a vector field parallel along 'Y.
Let T be the unit tangent to 'Y(where it exists) and let P be a unit vector
field parallel along 'Y,starting at a junction point. (Such a P exists by Theorem
6.7 of Chapter 4.) Note that x 1 is a field of unit vectors in CR.Let ot = ~(x 1 , T},
if,= ~(x 1 , P) and (J = ~(P, T). dot/ds= dif,/ds + dO/ds.
First we write y(s} = x(y1(s}, y2(s)). Note that cos if,= <x1, P) so that
= <xi', P) + <xi, P') = <xi', P) since P is parallel along 'Y·
-(sin ef,)if,'
However, xi'= y1' x 11+ y2' x 12so that Gauss's formulas (Equations (4-8)
of Chapter 4) yield
-(sin if,)ef,'= <y1'x 11+ y2'x 12, P)
= <(y1T111+ y2T121)x1+ (y1T112+ y2T1l)x2, P).
The specific form of the r,/ allows us to conclude
(4-1) -(sin if,)if,'= ,,2,h1<x2,P).
h
Because g 12 = 0, {xi, x2/lx 21} is an orthonormal basis of TaM for each
Q E x('ll); hence by Proposition 4.1 of Chapter 2,
p _ <P x )x
- , i i
+ <P, X2/I
X2l)x2
lx2 I .
Since P is a unit vector and <P, x 1) = cos if,, we have <P, x 2 )/h = sin if,.
Hence Equation (4-1) becomes if,' = -h 1y 2 ' so that

(4-2) ~if,= - { h 1y 2 ' ds = -{ h 1dy2 = - l h1 du2.

We now show that
(4-3) f T
K 11 ds = f
T
(J' ds.

Note cos (J = <T, P) so that -(sin 0)0' = <T', P) + <T, P') = <T',P).
Sec. 6-4 The Gauss-Bonnet Formula 187

K = (n x T T') = (n T x T') = (P x T, T x T')

11 ' ' sin 6
_ (P, T x (T x T')) _ (P, -T') _ (J'
- sin 8 - sin 8 - ·
J, J,
Hence K 11 ds = (J' ds, which is Equation (4-3).
Because a: ds = 4>'ds + (J' ds, Equations (4-2) and (4-3) imply
J, J, J,
J,a: ds + :Ea,, = - h 1 du2 + K 11 ds + :Ea,,. a,' ds + :Ea,, is precisely
J, J, J,
the angular variation of T along y taking into account the jump angles at the
junctions. Since T bounds CR,~a, = a,' ds + :Er,,1 = 2n. (This is essentially
J,
the Rotation Index Theorem.)
Finally, Green's (Theorem 1.3 of Chapter 3) states that

-i h1 du 2 =- IL h 11du 1 du 2 =- IL hi,1hdu 1 du 2 =+ILK dA.

Thus 2n = JJCR K dA + J,K 11 ds + :Ea,,. I

EXAMPLE 4.5. On S 2 let y be a triangle whose sides are geodesics and whose
interior angles are p1 , p2 , ft3 • Then a,,= n - p,.

2n = ff K dA + f
CR K 11 ds + :Er,, = 1 area(CR)+ O + 3n - :Ep,.
Thus (:EP,)- n = area(CR).The number (:EP,)- n is called the angular
excess of T. The fact that the angular excess is equal numerically to
area is a standard result of spherical geometry due to Legendre.

ExAMPLJ:4.6. Let y be a circle of latitude 4>= 4> 0 on S 2 • The geodesic curva-

ture of T is K cos 1/1,where K = 1/sin 4> 0 is the curvature of T and 1/f is
the angle between the tangent plane of S2 and the osculating plane of T
(this is the extrinsic formula for K 11). This angle is also 4> 0 • Hence
K 11 = cot 4>
0• J
K 11ds = K 11(length ofy) = K 11(2n sin 4> 0 ) = 2n cos 4>0 • Thus

2n = ff K ds + {
CR K 11 ds + :Er,, = 1 area(CR)+ 2n cos 4>
0

or
2n(l - cos 4>0 ) = area(CR)
as we found in Example 3.12.

PROBLEMS

4.1. Use the Gauss-Bonnet Formula to prove that the sum of the exterior
angles of a Euclidean polygon is 2n.
188 Global Theory of Surfaces Chap.6

4.2. In Problems 2.4 and 2.5 we had "surfaces" with K =

-1. Suppose 1 is
a geodesic triangle in either of these cases. What can you say about the
sum of the interior angles? (Hint: Example 4.5.)
4.3. (Hyperbolic Half Plane, see Problem 2.4.) Prove x = constant gives a
geodesic. Prove x = a + r cos (},y = r sin (} gives a geodesic with a and
r constant and O < (}< n. (Use Problem 5.8 of Chapter 4.) Note that in
'll these curves are vertical straight lines and circles which meet the
x-axis at right angles. They give all the geodesics.
4.4. (Poincare Disk, Problem 2.5.) Prove that y = 0 gives a geodesic.
Prove that x = a+ r cos 0, y = r sin 0, where a and r are constants
related by a 2 = 1 + r 2 , gives a geodesic (for a suitable range of 0). Note
that in 'll this curve is a circle that intersects x 2 + y 2 = 1 at right angles.
Due to the rotational symmetry of the matrix (glJ) (it depends on
x 2 + y 2 , not really on x or y) any such circle (or straight line through the
origin) gives a geodesic. These give all the geodesics.
4.S. In Problems 4.3 and 4.4 let 1 be a geodesic and let P be a point not on
the geodesic. Show that there are at least two geodesics through P that
never meet y. (Proof by reasonable picture is sufficient.) These "surfaces"
give examples of geometries where Euclid's parallel axiom fails, if by
straight line you mean geodesic and by parallel you mean never inter-
secting. These models of hyperbolic geometry are discussed more fully
in Chapter 7.

6-5. THE GAUSS-BONNET THEOREM AND

THE EULER CHARACTERISTIC

DEFINITION.
A polygon on a surface Mis a piecewise regular curve 1 whose
segments are geodesics and which bounds a simply connected region CR.

Let M be a compact surface in R 3 • Suppose M can be broken into

regions bounded by polygons, each region contained in a simply connected
geodesic coordinate patch. Let V equal the number of vertices, Ethe number
of edges, and F the number of faces (i.e., the number of polygonally bounded
regions). Let the Euler characteristic of M (with respect to this decomposi-
tion) be the integer x = F - E + V.
The next theorem is a truly amazing result. It relates two seemingly un-
related quantities, curvature (which is a geometric quantity) with the Euler
characteristic x (which is a topological or combinatorial quantity).

5.1 (Gauss-Bonnet). If M is compact then

THEOREM JIMK dA = 2nx.
Sec. 6-5 The Gauss-Bonnet Theorem and the Euler Characteristic 189

Proof· Let the polygonally bounded regions be {CR, F}, the polygons
i::::;;;:
11::::;;;:
i < F}, and {P,
themselves {T,11::::;;;: 1} the interior angles of y,. An application
of the Gauss-Bonnet Formula shows that

fLK dA = t Jf K dA = t (2n -
dlc ~ (n - p,1))
F F
= 2nF- I: I:n
1•1 J
+ f:1
~ I:P,1·
J

Now
I: n = n (number of vertices of 1,) = n (number of edges of 11).
J F F
Since each edge belongs to two polygons, ~ I: n = 2nE. I: I: P,1 is the sum
t=1 J I• 1 J
F
of all interior angles. At each vertex the sum is 2n. Hence~ I: P,1 = 2nV.
t:'1 J
Thus ffMK dA = 2nF- 2nE + 2nV = 2nx. I

The quantity JIM K dA is called the total curvatureof Mor the curvatura
integra. Since it is a well defined number, X cannot depend on how the
poly!lons are chosen, so long as each is small enough to be contained in a
simplj connected geodesic coordinate patch! This fact alone is surprising but
we may also look at the Gauss-Bonnet Theorem the other way. From its
definition x must be an integer, hence (1/2n) JIM K dA is an integer! There is
no a priori reason for this. (It is similar to the situation for plane curves,
where (1/2n) f k dsis an integer.) The Gauss-Bonnet Theorem shows us one
of the beautiful things about global differential geometry (and its present
popularity): the mixing of topological and geometric results in a very unusual
and striking manner. To us, this theorem is even more egregious than the
TheoremaEgregium.
To define X we do not really need geodesic polygons, just piecewise
regular curves. They were chosen to be geodesic so that the terms J,"•ds in
the Gauss-Bonnet Formula would not appear. They would cancel anyway
since each edge is traversed twice and in opposite directions because it bounds
part of two regions.
On the other hand, if M is broken into polygons that are not small
enough to be in simply connected geodesic coordinate patches, we could still
define i = F - E + V, where F, E, V are the number of faces, edges, and
vertices. If a particular polygon is not small enough, let P be a point inside it.
Join P to each of the vertices of that polygon by a regular curve segment.
(Since Mis compact, these curves can even be geodesics.) If the polygon
has r vertices, we have added r edges, 1 vertex, and r - 1 faces (where there
was t face there are now r). F = F + r - 1, E = E + r, V = V + 1.
190 Global Theory of Surfaces Chap.&

Then F - E + V = F - E + V + r - 1 - r + 1 = i· Hence we have not

changed the value of i, but we have eliminated one large polygon. This can
be continued until all polygons are small enough.

DEFINITION.
xis called the Euler characteristic of M.

EXAMPLE
5.2. x(s 2) = 2~ JL K dA = ;: = 2.

x has been defined combinatorially. Our theorem shows that it is an

intrinsic geometric invariant. x can also be defined topologically.

The proof of the following theorem (which gives the classification of

surfaces) and the formal definition of "looks like" can be found in Massey
[1967].

THEOREM 5.3. Every compact surface in R 3 "looks like" a sphere with

handles. The number of handles is called the geometric genus of M and
is denoted g. (See Figures 6.10 and 6.11.)

FIGURE 6.10

FIGURE 6.11

5.4. If M is a compact surface in R3 , then

THEOREM x= 2(1 - g).

Proof (by induction on g): If g = 0, then M looks like S 2 and x(S2) = 2

= 2(1 - 0). Assume the theorem is true for g = n and let M have genus
+
n I. Break M into polygons with F, E, V faces, edges, and vertices and so
that one handle looks like that in Figure 6.12.
If three faces are removed as in Figure 6.13 and the resulting holes filled
in, there is a net loss of one face and three edges. The new surface has Euler
characteristic i which is two more than the original. It has genus n. By induc-
tion i = 2(1 - n). Hence the Euler characteristic of the original surface is
X = i - 2 = 2(1 - n) - 2 = 2(1 - (n + 1)) = 2(1 - g). I
Sec. 6-5 The Gauss-Bonnet Theorem and the Euler Characteristic 191

FIGURE 6.12 FIGURE 6.13

Note that the Euler characteristic of a compact surface in R 3 is even and

less than or equal to 2.
There is a more general version of the Gauss-Bonnet Theorem in our
context.

THEOREM5.5. Let r be a piecewise regular curve in an oriented surface M.

Suppose r bounds a region CR.Then

ff K dA + l "·ds + I; (1t -
(R IX,)= 2nx(CR),

where the IX1 are the interior angles ofy and x(CR)is the Euler character-
istic of CRfound by breaking CRinto polygons and counting those
edges and vertices lying on y in addition to those in CR.

Proof· Problem 5.1. I

There are many generalizations of the Gauss-Bonnet Theorem. Note that
in its simplest form it states that positive curvature everywhere implies posi-
tive Euler characteristic. The conjecture that this is true in higher even
dimensions (in the context of Chapter 7) has long been one of the outstanding
problems in differential geometry. It is called the Chem-Hopf Conjecture
(Conjecture 8.16 of Chapter 7).

PROBLEMS

*5.1. Prove Theorem 5.5.

5.2. Call a polygon an n-gon if it has n vertices (so a 0-gon is a closed
geodesic). Let M be a surface with K < 0. Prove that there are no
n-gons for n = 0, 1, 2.
192 Global Theory of Surfaces Chap. 6

S.3. Find an example of a surface with K < 0 that has a closed geodesic 1·
(1 cannot bound a simply connected region by Problem 5.2.)
S.4. Consider a "triangle" in the Poincare disk whose vertices are on the
boundary of the disk ("at infinity"). What do you expect the area to
be? (See Figure 6.14.) What would be the area of an n-gon if all the
vertices are at infinity? (Figure 6.15).

FIGURE 6.14 FIGURE 6.16

5.5. Prove that a compact surface in R 3 which has K";?:; 0 "looks like" a
sphere. (Hint: You may use Theorem 5.3.)

6-6. THE THEOREMS OF JACOBI AND

HADAMARD
In this section we give two more examples of theorems from global
differential geometry. The first is due to C. G. J. Jacobi (1804-1851)and the
second to J. Hadamard (1865-1963).Both are proven as applications of the
Gauss-Bonnet Theorem.
Recall that if 1 is a regular space curve with ,c > 0, then its normal
spherical image o is the curve on S2 given by o(s) = N(s), where N is the
normal to 1· Since N' = -JCT+ TB =t= 0, o(s) is regular.
THEOREM 6.1 (Jacobi, 1842). Let 1 be a regular closed unit speed space curve
with ,c > 0. Assume that o, the normal spherical image, is a simple
curve. Then it divides the unit sphere into two regions of equal area.
Proof: Let t be arc length on o and R. the geodesic curvature of o viewed as
a curve on S2. Let CRbe the region bounded by o. We will show that the area
of CRis 2n by applying the Gauss-Bonnet Formula. (Since the area of S 2 is
4n, this will complete the proof.) To this end, we must compute R•. We shall
show that
(6-1) ,c- =- d( arctan- T)ds-r·
• ds ,c at
Since t is arc length on o, the Frenet-Serret equations of 1 show that
(6-2) do= dN =(-JCT+ TB)~•
dt dt ut
Sec. 6-6 The Theorems of Jacobi and Hadamard 193

which is a unit vector so that

ds 1
(6-3) dt = ,.f ,c2 + Ti"
Also
d2N = d2s(-,cT +TB)+ (ds)2(-,c'T + T'B - (,c2 + T2)N)
~2 dt2 ~ •
or, after applying Equation (6-2),
(6_4) d 2N = d 2 s dt dN +
dt 2 dt 2 ds dt
(ds)
dt
2(-,c'T + T'B _ (,c2 + i- 2)N)
·
Now the surface normal to S 2 at p is p. Hence the surface normal to S 2
at o(t) is N(t) so that the intrinsic normal to o is N x dN/dt. By using (6-2),
(6-4), and the fact that i. = (S, d 2N/dt 2), we obtain

it•= (Nx ~~· ~~) = (N,~~ x ~~)

= (N, ~~ x (:) 2(-,c'T + T'B - (,c2 + T2)N))

= (:r (N, (-,cT + TB) X (-,c'T + T'B - (,c2 + T2)N))

= (:) 3 (-T,c' + ,er').

Because of Equation (6-3), this last equation is

it = ds (T',c - ,c'T) = ds !!...
(arc tan.!...)
• dt T2 + ,c2 dt ds ,c
as desired.
We integrate Equation (6-1) to obtain

f.
•
it• dt = J..ds!!...(arc tan .!...)
,c dt
ds dt = J.
,ds
(arc tan .!...)
!!... ,c
ds = 0

since 1 is closed. Thus the Gauss-Bonnet Formula gives us

Thus the area of CRis 2n and o breaks S 2 into two regions of equal area. I
The next result generalizes Theorem 3.2 of Chapter 3 relating convexity
to the sign of the curvature of a plane curve.

THEOREM 6.2 (Hadamard, 1897). Let M be a compact surface in R 3 • If K > 0

everywhere, then ~e surface is convex (lies on one side of each tangent
plane).
194 Global Theory of Surfaces Chap.6

Proof· K > 0 implies O < JIM K dA = 2nx = 2n2(1 - g) and g < I. Hence
g = 0 and JIMK dA = 4n.
Let v: M ----+ S 2 be the Gauss normal map defined by the outward point-
ing normal. We show that vis onto. Let n E S2. Let IT, be the plane in R 3
given by <x, n)=r. Since Mis bounded, there isanr>Owith.-rs(p, n)sr
for all p E M. Let R be the smallest real number such that (p, n) s R
for all p E M. R may be negative. Then M n ITR=I=0. If P E Mn ITR,then
ITRis the tangent plane at P, v(P) = n, and v(M) = S2. Note ffmK dA is the
area of v(CR)counted with multiplicity.
Next we show that v is one-to-one. Suppose v(P) = v(Q) with P =I=Q.
Then there is a small open set 'll about Q such that v(M - 'll) = S 2 also
since v is continuous. Hence JIM-'ll K dA ~ 4n. JI'ILK dA > 0. Hence
JIMK dA = JIM-'IL K dA + JI'IL K dA > 4n, a contradiction. Hence vis one-
to-one.
Now suppose there is a point P on M such that there are points of M
on both sides of the tangent plane at P. Let Q, R be points of Mon opposite
sides of this plane and farthest from the plane. The normals at P, Q, R must
be parallel. Hence two are equal and v is not one-to-one, a contradiction.
Thus M is convex. I

PROBLEM

The following problem generalizes the notion of curves of constant

width to surfaces. (See Section 3-5.)
t6.1. Let M be compact with K > 0 (hence M is convex).
(a) If P E M, prove there is a unique point P =I=P with the tangent
planes at P and P parallel.
(b) From now on assume that the distance between these planes is a
constant independent of the choice of P. Prove that the line joining
P to P is normal to M at both P and P.
(c) Prove that the principal directions at P and Pare parallel (in the
Euclidean sense).
(d) Let X be a principal direction at P with principal curvature K 1 and
X the corresponding principal direction at P with principal curva-
ture R 1 • Prove that l/1e1 + 1/R1 is a constant independent of P.
Does H(P) = H(P), where His the mean curvature? (See Struik
[1961, p. 202].)
(e) Prove JIM H dA = 2nc, where c = width (constant distance from P
to P). (You will need the Gauss-Bonnet Theorem for this last
result.)
Sec. 6-7 The Index of e Vectof'Field 196

6-7. THE INDEX OF A VECTOR FIELD

Let M be a compact surface with a vector field V on M. We say P e M

is an isolatedzero of V if there is an open set 'U.about P with P the only point
in 'U.where Vis zero. Suppose Pis an isolated zero of V, and let 'f be a simple
closed piecewise regular curve which bounds a simply connected region <R
with P the only zero of V in <R.

DEFINITION. The index of Vat Pis ip(V) = (1/2n) «S<X(U,V), where U is any
field of unit vectors in <R.

We know that ip(V) does not depend on U. It" is an integer which does
not really depend on T either-a continuous deformation of T cannot change
the integer.

ExAMPLE 7.1. The vector fields in Figure 6.16 have indices as indicated.

ip =I ip =I ip = -I

,-, -- '\
\/'f /'
----
'l~
~~
, t
{ (........__'t ~'--7L.
7~~
t,/ i '-.../
ip = 2 ip
-
'........

= I
/
ip = -2
FIGURE 6.16
196 Global Theory of Surfaces Chap.6

Let M be a compact surface in R 3 with V a vector field on M with only

a finite number of zeros. The total index ofV is /(V) = I: i,(V).

PROPOSITION 7.2. If V and Ware two vector fields on a compact surface M

with only a finite number of zeros, then /(V) = I(W).

Proof· Break M up into triangles y, so that each triangle is in a simply con-

nected coordinate patch, each triangle has at most one zero of V and at most
one zero of W in its interior, and all zeros are inside triangles. Let U, be a
field of unit vectors in the patch about y,.
I
/(V) = 21t l: 64:(U,, V).

/(V) - I(W) = 2I1t l: (64:(U,, V) - 64:(U,, W)) = 2I1t l: 64:(W, V)

= _!_ I: l: 64:(W,
21ttriaqles ed1e1
V) = 0,

since each edge is traversed twice, in opposite directions. Thus /(V) = I(W).
I
THEOREM 7.3 (Poincare-Brouwer). If M is a compact surface and V is a
vector field on M with only a finite number of zeros, then /(V) = x(M),
the Euler characteristic of M.

Proof· We break Minto triangles. We can define a vector field W on M

which has a zero at each vertex, at the midpoint of each edge, and at the center
of each triangle as in Figure 6.17. The zeros at the vertices and at the centers
have index + I while those on the edges have index -1. Thus
/(W) = l: i, = (+I)V + (-I)E + (+l)F= V- E + F= X·
But /(V) = I(W) by Proposition 7.2. I

FIGURE 6.17
Sec. 6-7 The Index of a Vector Field 197

This theorem implies that any vector field on S 2 must have a z.ero. This
fact is sometimes referred to by saying "you can't grow hair on a coconut,"
meaning somewhere there will be a swirl or a cowlick.

PROBLEMS

7.1. Let M be a compact surface. Prove that M has a vector field with no
zeros if and only if M "looks like" a torus.
7.2. Prove that on M = R 2 there are vector fields with infinitely many z.eros
Pn such that the sequence {Pn}does not have a convergent subsequence.
7.3. Prove that on M = R 2 there are vector fields with no z.eros.
7.4. Find an example to show that Proposition 7.2 is false if the hypothesis
"compact" is omitted.
 7

Introduction
to Manifolds

In this chapter we return to the theme of the introduction to Chapter 4-

that of geometry being built with the concept§ of "space" and "line." In
Chapter 4 we used surfaces for the concept of space and geodesics for the
concept of line. In this chapter we will take a much more abstract view of
space by defining "space" to be anything that looks locally like Euclidean
n-space (a manifold). Note that this is similar to what we did for surfaces;
they were locally like R 2 • One important difference here is that we are not
assuming that our manifolds lie in a Euclidean space of one dimension higher
(as we did for surfaces) or even that they lie in Euclidean space at all!
Having settled on a definition for space (which occupies the first two
sections), we must find a suitable geometric structure to add to space to give
the concept of "lines" (which we again call geodesics). A careful reading of
Chapter 4 shows that there are two geometric structures which led inexorably
to the notion of lines. One was the notion of differentiating a vector field with
respect to a vector field and the other was the notion of measuring lengths
and angles (the metric tensor). We shall take the view here that the first idea
is primitive and call such a structure a "covariant differentiation" in Section
7-6. We will also introduce the notion of a Riemannian metric on a manifold
and show in Section 7-8 that every Riemannian metric gives rise to a covariant
differentiation. We conclude this last section with a discussion of the various
198
Sec. 7-1 Some Analytic Preliminaries 199

notions of curvature which can be defined on a manifold with Riemannian

metric and their relationship with curvature as defined in Chapter 4.
There are a couple of things to notice. The first is that we must be very
careful to define everything intrinsically because there is no such thing as
extrinsic! This is one reason why it was so important in Chapter 4 to get an
intrinsic description of something we had already defined extrinsically. In this
chapter we will use the intrinsic description for the very definition of the con-
cept. (This set of ideas is best exemplified by the notion of curvature. See
Gauss's Theorema Egregium of Section 4-9. Compare that section with
Section 7-8.) A more philosophical question is: Since this material has become
so very abstract, has it lost all possible "relevance" to anybody but a mathe-
matician? The answer is an emphatic no! There are many applications of this
material to high energy physics (the theory of Lie groups is useful in quantum
mechanics) and relativity theory. In fact Albert Einstein gave much impetus
to differential geometry at the turn of the twentieth century by introducing
relativity theory.
On the question of what is the proper level of abstraction, we would
certainly want our approach to be abstract enough to include all of the clas-
sical geometries. In fact, we do have as special cases Euclidean, spherical,
hyperbolic, and projective geometries. There are, of course, other ways of
obtaining these classical geometries. For example, a group theoretic ap-
proach is taken in Millman [1977].
Another point to make about this chapter is that we have written it in a
different manner than the previous six. We have written it in an open-ended
style; that is, we have not tried to be encyclopedic but rather have given
references for the results which we do not prove here. What we are doing here
is motivating the abstract definitions of manifold theory so that if the reader
wishes to study some of the more advanced works in the field he or she may
do so without being attacked by a formalism which seems to be totally un-
motivated. We have also adopted the notation which is most common in
manifold theory. In this we follow Kobayashi and Nomizu [1963].
We make one further comment on this chapter and that is about the
level of sophistication required to read it: We outline in Section 7-1 some of
the analytic preliminaries necessary. The reader should have some familiarity
with metric spaces and with advanced calculus (at least to the point of the
Inverse and Implicit Function Theorems).

7-1. SOME ANALYTIC PRELIMINARIES

In this section we review some basic results from analysis, especially

metric spaces and the Inverse and Implicit Function Theorems.
200 Introduction to Manifolds Chap. 7

DEFINITION.Let X be a set. A metric don Xis an assignment of a nonnega-

tive number d(x, y) to each pair of points x, y E X such that:
(a) d(x, y) = 0 if and only if x = y;
(b) d(x, y) = d(y, x); and
(c) if z E X then d(x, z) < d(x, y) + d(y, z).
A metric space is a set X together with a metric d.

1.1. Let X = RN. If x = (x 1 ,

EXAMPLE ••• , xN) and y = (y 1 , ••• , yH), we
define d(x,y) = .J'ti (x' - y')z =Ix - y I· This is the usual Euclidean
distance on RN.

The next example is inserted to show that "strange looking" spaces can
be higher dimensional surfaces. This example may be skipped on a first
reading.

EXAMPLE 1.2. Let pn be the set of all straight lines through the origin in
Rn+1 • pn is called real projective n-space. P 2 is called the projective plane.
If I E r, then I may be represented by any nonz.ero point in Rn+r (say
a= (a 1 , ••• , an+1)) through which I passes. Note that a and b E Rn+1
represent the same line if and only if there is a O =I=l E R such that
a= lb. Motivated by this fact, we define an equivalence relation on
Rn+i - {O} by a,.., b if there is a O =I= l E R such that a= lb. We write
[a] for the equivalence class of a E Rn+1 • Note thatr = (Rn+i - {O})/,..,.
The coordinates of any a E Rn+i such that I= [a] are called homogene-
ous coordinates for /. (They are only determined up to a real multiple, of
course.) Let / 1 , / 2 E pn and write /1 = [a,] for a, E Rn+i where i = 1, 2.
Define the distance from / 1 to / 2 to be

d(l 1, 12 ) = minimum of I i::

I- I::11 and i::
I I + I::,I·
The motivation for this is as follows: For each i = 1 or 2, aifla,I is one
of the two points on snwhich represent /1• (They are antipodal points.)
We then measure distance by computing the Euclidean distance between
the nearest pair of representatives of/ 1 and / 2 • See Figure 7.1.

DEFINITION.If x E X and E > 0 then the (open) ball of radius E about xis
Bix) = {y E XI d(x, y) < E}. A subset S of Xis open in X if for each
x E S there is an E > 0 such that B,(x) c S. If S is open and x E S,
then S is a neighborhood of x.

The following is immediate.

Sec . 7-1 Some Analytic Preliminaries 201

FIGURE 7.1

PROPOSITION 1.3. Let X be a set with a metric d.

(a) If U and V are open in X, then U (') Vis open.
(b) If U,. is open in X for each ix e /, then U U,. is open.
* &El
(c) If p, q e X and p q, then there are open sets U and Vin X such
that p e U, q e V, and U (') Vis empty.

For students who have seen the definition, the following proposition is
immediate from the definition of continuity in a metric space. For students
who have not, the following may be taken as the definition of continuity.

PROPOSITION 1.4. Let X and Y be metric spaces and /: X-> Y. f is contin-

uous if and only if for every open set Vin Y, 1- 1 (V) is open in X.

We now recall some facts about calculus. If/: R•-> Rm, then /(x) may
be written as (f1(x), ... , /m(x)) for some real-valued functions / 1 , ••• , /"'
of n variables. f is Ck if all partial derivatives of each f' exist and are
continuous up to order k . We will usually omit reference to the k by saying
that f is differentiable. As in the case m = n = 2, we may form the
Jacobian matrix J(f) at each point p e R•:

a11
rxr<P> ~(p)

a12 a12
rxr<P> rx-(p)
J(!Xp) =

af"'
-axr(p)
202 Introduction to Manifolds Chap. 7

EXAMPLE 1.5. Let/: R 2 - R 3 be given by f(x 1, x 2) = (sin x 1 , x 2x 1 , x 2 cos x 1).

At p = (x 1 , x 2) we have
cos x 1
J(f)(p) = ( xz xi0 ).
-x 2 sin x 1 cosx 1

DEFINITION. Suppose U is open in R•, V is open in R•, and/: U - V. f is a

Ck diffeomorphismif f is Ck and there is a Ck function g: v- U such
that f o g (resp. go/) is the identity map on V (resp. on U). g is called
the inverse of f (These were called coordinate transformations in
Chapter 4.)

EXAMPLE 1.6. Let U = V = the open interval (-1, 1) in Rand /(x) = x 3 •

The inverse off is g(y) = ,J'y so that f is a Ck diffeomorphism if and
only if k = 0.

For a proof of the following theorems see Fulks [1969].

THEOREM 1.7 (Inverse Function Theorem). Let Ube an open set in R•, p e U
and/: U - R•. If det J(f)(p) =/:.0, then there are neighborhoods N, of
p and N1c,,of ,f,(p) such that/IN,: N, - N1c,,is a diffeomorphism.

ExAMPLE 1.8. Let U = {(x1, x 2) I -!< x 1 < !}and define

/: U - R 2 by f(x 1, x 2) = (e--" cos x 2 , e-"' sin x 2 ).
1

An easy computation shows that det (J(f)(x 1, x 2)) = -e-2ar•=t:.0 so the

Inverse Function Theorem applies for any point p = (x 1 , x 2). Note,
however, that although/is one-to-one when restricted to small enough
neighborhoods of any point,/: u- R 2 is not one-to-one! However,/
is one-to-one on the open set
Nco,01= {(x 1, x 2) I -! < x 1 < !, -n < x 2 < n}.
Let i be a fixed integer between 1 and n + 1. If
w= (w 1, ••• , w•+1) e R•+ 1 ,
then we shall write lii = (w 1, ••• , w1- 1 , w'+1, ••• , w•+1) e R•. If W c R•+ 1 ,
then W= {Jiilw e W} c R•. Note that if Wis open in R•+ 1 , then Wis
open in R" (Problem 1.3). We shall now state the Implicit Function Theorem,
which roughly says that if a11au1 =/:.0, then we can solve for the ith variable
in terms of the others.

THEOREM 1.9 (Implicit Function Theorem). Let/: R•+ 1 - R be a Ck func-

tion (k ~ 1), a e R•+ 1 and assume that (aJ/au')
(a)=/:. 0 for some fixed
Sec. 7-2 Manifolds-Definition and Examples 203

i. Then there is a neighborhood W of a in Rn+1 and a Ck function

g: W-+R such that, forw= (w1, ••• , w•+1) E R•+ 1, /(w 1, ••• , w•+1) = 0
if and only if w' = g(Jio).

PROBLEMS

1.1. Prove that P' is a metric space.

*1.2. Let U, = {/E P' 11= [(a 1, ••• , an+1)] and a':¢: O}.Prove that U, is well
defined and open in P'.
*1.3. If W c R•+ 1 is open, show that W is open in R•. Show that the con-
vtrse is false.
1.4. Prove Theorem 1.9 for f (x 1, ••• , xn+1) = -1 + I: (x') 2 •
7-2. MANIFOLDS-DEFINITION AND
EXAMPLES
In this section we define the main object of study of this chapter, a dif-
ferentiable n-manifold, and give a large collection of examples. It will be
obvious from the definition of manifolds that surfaces (as defined in Chapter
4) are exactly those 2-manifolds which are subsets of R 3 . (See Section 7-5 for
a more precise description.) We shall prove the important fact that if
/: R•+ 1 -+ R is a differentiable function, then 1- 1(0) (the hypersurfacedefined
by/) is an n-manifold if a certain condition on/holds.
Just as a surface looked locally like R 2 , an n-manifold will look locally
like R•.
DEFINmoN. Let M be a metric space and p E M. A coordinatechart about
p of dimensionn is a neighborhood U of p and a one-to-one continuous
function 4,: U-+ R• such that O = 4,(U) is open in R•. ( U, 4,)is a proper
coordinatechart if 4,-1 : ef,(U)-+ Uc Mis continuous.

EXAMPLE 2.1. Let S" = {(x 1, ••• , xn+1) e R•+ 1 I~ (x') 2 = 1}be the n-
sphere and let p = (0, ... , 1). A proper coordinate chart about p of
dimension n is given by U= {(x1, ••• , xn+1) E s•1xn+ 1 > O} and
4':U - R• by ef,(x1, ••• , x-+ 1) = (x 1 , ••• , X-). Note that
U= 4,(U) =B 1(0),
which is certainly open. Clearly
4,-•cu•,... , un)= (u•,... ,un,,/1 - t.(u') 2)

which is continuous so that (U, 4,)is a proper coordinate chart.

204 Introduction to Manifolds Chap.7

DEFINITION. Let M be a metric space. M is an n-dimensional C manifold if

00

there is a collection Ci of coordinate charts (U, tf,),called the atlas of M,

such that
(a) for each p E M there is a proper coordinate chart (U, tf,) E Ci of
dimension n with p E U;
(b) if (U, tf,), (V, v,) E Ci with U (') V =I=0 then
v, o tf,-1: tf,(U(') V)-+ v,(U (') V)
is a C diffeomorphism (of open sets of Rn); and
00

(c) Ci is maximal with respect to conditions (a) and (b), i.e., Ci contains
all possible charts with these properties. (See Figure 7.2.)

1/1 0 ;- I

FIGURE 7.2

The content of condition (b) is that v, o ef,-1 is C Having proven this

00
•

for all v,, ef,shows that v, o ef,-1 is a diffeomorphism since

(v, otf,-1)-1= "'ov,-1.
We will usually abbreviate "n-dimensional C manifold" by "differentiable
00

n-manifold" or "n-manifold," or write Mn. It is possible to define a Ck

manifold for any l ~ k < oo. However, in this case it is necessary to con-
tinually check the degree of differentiability. We prefer to avoid this.
In practice we usually give a collection of proper coordinate charts
(Ui, ef,,)on M which cover M (i.e., M = U U,). There is then a unique atlas
I
determined which includes the given subcollection of (U 1, ef,i),much as a
subbasis determines a topology. It is possible to have two different atlases on
a metric space M, making M into a manifold in different ways. We shall not
get involved with such problems here.
Note that this definition is essentially the same as that of a surface as
given in Section 4-2, with two exceptions: the direction of the maps has been
reversed and U is a subset of M instead of being a subset of R 2 • A coordinate
patch was a map from R 2 to the surface; a coordinate chart is a map from the
Sec. 7-2 Manifolds-Definition and Examples 206

manifold (surface) to Rn (R 2). This change emphasizes the fact that the
manifold is the primary object of study, not Euclidean space.

ExAMPLE 2.2. R• itself is an n-manifold, for example, with the single coordi-
nate chart (Rn, identity). Similarly any open set in Rnis an n-manifold.

EXAMPLE 2.3. snis an n-manif old. This can be seen by using Example 2.1 and
imitating the proof that S2 is a surface. See also the proof of Theorem 2.6
below.

EXAMPLE 2.4. Let G = GL<.n,R) be the group of all nonsingular n x n

matrices. We show that G is an n2-dimensional manifold. G is a metric
space with distance functiond(A,B) = ,vIi;
,.,_,(a,1 -b, 1) 2 , where A= (a,1)
and B = (b,1). If A = (a,1) e G, let
if,(A) = (a 11 , a 12 , ••• , a,.,a 2 ,, ••• , a.J E R"'.
To show that (G, if,) is a proper coordinate chart we need only show
that if,(G)is open in R•'. We define a function A: Rn'--+ R by
A(x, 1, ••• ~ (-1)"
'x .. ) = .,n. X1,,(1) Xz,,(2) ••• Xn,,(n)•

wheres. is the group of permutations of n letters. Note that A is con-

tinuous and Ao if,(A) = det A. Therefore if,(G)= A- 1(R - {O})is open
by Proposition 1.4.

If we write M(n) for the set of all n x n matrices, then we have identified
M(n) and Rn'via the mapping if,.

ExAMPLE 2.5. We show that projective n-space I"' is an n-manifold. We

define (n + 1) sets U, = {/ e P"ll = [(a 1, ••• , a.+ 1)] and a, :;e O}. By
Problem 1.2 these are open subsets of I"'. Note that {U1, U 2 , •• • , U.+1}
covers P•. We define if>,:
U,--+ R• by

J..t/)
Y't\
=(a
a,
1
' ••. '
a,_1' a1+ 1' ••.
a, a, '
a.) E
a,
R" fort ~i~n+ 1.

is called the ith nonhomogeneouscoordinatesof/.) Note that if,,is

(if,1..1)
well defined and that if,,(U,)= R•, which is certainly open. (U,, if,,)is
proper because
(2-1) ef,,-1(u 1, ••• , un) = [(u', u 2 , ••• , u1- 1, l, u', ... , u")].
To save on notation we will only show that if,2 o if,3 - 1 is C 00
• By (2-1)
,I. -I (U I , •••
,,I,. o y,3
'PZ , u8) --
.
,I.
Y,2(U I , U2 , 1, U3 , • , • , U" ) -- 2u , 2u1 , 2u
("'
u
u•)
u3 , • , • , ::Z "
206 Introduction to Manifolds Chap. 7

Since u 2 ::;,!:0 on cf,3( U2 n U3 ), which is the domain of cf,2 o cf,3- i, cf,2 o cf,3- 1
is c=.
Although P 2 is a 2-manifold, it is not a surface in R 3 (see Section
7-5).

The next result is really another example, but because it includes so many
familiar examples, we separate it out as a theorem. This theorem is the
rigorous basis for the assertion of Section 4-2 about level surfaces of a
function. Because most students have trouble with the constructions involved
in this theorem, we will carry along (inside bold-face brackets [ ]) the special
case which gives S 2 •

DEFINITION.If/: Rn+i-> R is c= (with k > l), then the gradient off is the
function grad /: Rn+1 -> R"+1 given by

(grad f)(p) _
- (af a1 (p)) •
azii(p), • • • • aun+I

2.6. Let/: Rn+i---->R be a

THEOREM c= function and
M1 = {x E R"+1 i/(x) = 0}.
If (grad /)(p) *
0 for all p E M 1 , then M 1 is a c= n-manifold (called
the hypersurface defined by!).

Proof: We write M for M 1 . [/ (x, y, z) = x 2 y2 + +

z 2 - I for S2]. Let
m 0 E M be fixed. Since (grad f)(m 0 ) ::;,!:0, there is an I < i ~ n 1 such +
that aJ/au' (m0 ) ::;,!:0. For notational convenience we assume that i = n l. +
[Take m 0 = (0, 0, l) for S 2 .] This can be accomplished by renumbering the
variables. As in Section 7-1 ifw = (w1 , ••• , wn+1) E Rn+I and W c Rn+l, we
will write li>= (w1 , ••• , wn) E Rn and W= {~lw E W}. By Theorem 1.9
there is a neighborhood w of mo in Rn+I and a c= function g: w-R such
that w = (w1 , •••• wn+1) E w ('\M if and only if wn+I= g(w 1 , ••• 'w").
[For S2 : W = {(x,y, z) Ix> O} and g: W-> R is given by
g(ui, u2) = ,v'l _ (u1)2 _ (u2)2.)
Since W is open in Rn+1 , U = W n M is open in M [ U is the upper
hemisphere for S2]. Let cf,:U-> Rn be defined by
cf,(w1 , •••• wn+1) = (w1, ... 'wn).
The point is that (U, cf,)is a chart about m 0 for M. First note that cf,(U)= U,
which is open by Problem 1.3. To show that cf,is one-to-one and proper we
need only write down ct,-1 : ct,- 1(u 1 , ••• , U") = (u 1 , ••• , un, g(u 1 , ••• , u")).
[For s2: ct,-1(u1,u2) = (ui, uz, ,v'l - (u1)2 - (u2)2).]
We now check the overlap condition: Suppose that (V, 1p) is another
patch about m 0 • We may assume that this corresponds to a1;ax1 (m0 ) ::;,!:0 and
Sec. 7-2 Manifolds-Definition and Examples 207

so by the construction above 'II: V--+ R" is given by

'1/(x•, . .• 'X"+•) =(xi, ... ' X"+•).
Therefore.
'If O ;-•cu•,
.. ,, u") = 'lf(U 1, ., • , u"g(u•, ,. • , u"))
= (u 2, ••• , u", g(u 1, • •• , u")),
which is C on ;- 1(V n U). (For S2 :
00

'II o ;-•(u•, u2) = (ui, .../1- (u•)2 - (u2)2).] I

The next example shows that Theorem 2.6 can also be used to produce
fairly abstract examples.

ExAMPLE 2.7. Let SL(n) be the set of all matrices of determinant one. We
may view SL(n) as a subset of R111 by stringing out coordinates as in
Example 2.4. Showing that SL(n) is a manifold of dimension n 2 - I
amounts to investigating the function/ : R111 -+ R given by f (A) = det A.
Since 1- 1(1) = SL(n) we can use Theorem 2.6 once we know that
(grad /)(A)* 0 for all A e SL(n). This computation is done in Guil-
lemin and Pollack [1974).

ExAMPLE 2.8. Let C = {(x, y, z) e R3 Ix 2 + y 2 - z 2 = O}be a double cone.

Then C is not a 2-manifold. (See Figure 7.3.) Suppose C were a 2-mani-
fold. Intuitively, we can disconnect C by removing the origin which
could not happen to a 2-manifold. More formally (for those who have
had some topology) let ( U, q,) be a coordinate chart about p = (0, 0, 0)
with U connected. Then U - {p} must be connectedsinceq,(U)- {q,(p)}
is connected. On the other hand, U - {p} is obviously not connected
(from the picture). Note that if f(x,y, z) = x 2 + y 2 - z 2 then (grad/)
(0,0,0) = 0.

FIGURE 7.3
208 Introduction to Manifolds Chap. 7

PROBLEMS

snis an n-manifold directly from the definition (i.e., without

2.1. Prove that
Theorem 2.6).
2.2. If M 1 (resp. M 2 ) is a C n 1-manifold (resp. n 2 -manifold), prove that
00

+
M 1 X M 2 is a C n-manifold for n = n 1 n 2 • (Hint: If (U,, ~,) is a
00

chart about m, e M,, then make (U 1 X U2 , ~ 1 X ~ 2) a chart about

(m1 , m2 ).)
2.3. Let T• =
{(x 1, ••• , x 2 •) E R 2 • I(x 1) 2 + (x =
2) 2 1, ... , (x 2 •- 1) 2 + (x 2 •) 2 = 1}.
Prove that Tn is an n-manif old. T• is called the n-torus. (For those who
have had topology: T 2 is homeomorphic to the inner tube of Chapter 4.)
2.4. Let M = {A e M(n) Itrace (A) = O}.Prove that M is a manifold (of
what dimension?). Can you picture M?
t2.5. Let O(n) = {A e M(n)IAA' = I} be the set of orthogonal matrices.
Since O(n) c M(n) = R•1 , we may view O(n) c R•1 •
(a) Show that O(n) is compact (i.e., closed and bounded).
(b) Show that O(n) is an (n(n - 1)/2)-manifold.

7-3. TANGENT VECTORS AND

THE TANGENT SPACE

Because the concept of a tangent vector was so useful in surface theory,

we would like to define an analogous concept on an arbitrary manifold. An
immediate generalization of the definition is not possible. On a surface we
defined a tangent vector to be the velocity vector of some curve on the surface.
This meant we viewed the curve as lying in R3, where we knew how to dif-
ferentiate. In our current setting a curve on Mis only a curve on M (there is
no ambient Euclidean space in which it is included) and it does not make
sense to differentiate in a metric space. The reader may argue that locally M
is an open set in Rn, where we do know how to differentiate. This is true, but
there is no canonical way to make M look locally like Euclidean space. Dif-
ferent people could choose different coordinate charts and obtain different
notions of differentiation.
In elementary calculus, a vector v at a point p e R• may be viewed as a
directional derivative. Ih = (a 1, a 2 , ••• , an) and/: R• - R is differentiable,
then the directional derivative off at p in the direction v is

v(/) = :Ea' 3-G,(p).

Sec. 7-3 Tangent Vectors and the Tangent Space 209

We will imitate this idea by defining a tangent vector as a real-valued oper-

ator on the set of differentiable functions on M which obeys the properties
of a derivative. First we will need the concept of a differentiable function
defined on a manifold.
Let M be a c~ n-manifold. We shall use the word "chart" for "proper
coordinate chart."

DEFINITION.Let p E Mand let/: M - R. f is differentiable of class C1

(l < oo) at p if there is a chart (U, ef,)about p such that
f O ef,-1: ef,(U)--+ R
is of class C' at ef,(p) E Rn.

There are two things to note about this definition. First of all, it is not
circular because cf,(U) is an open set in Rn,where we know already what it
means to be differentiable. Also, the definition does not really depend upon
the choice of chart because if (V, v,) is another chart, then
f o v,-1 = (! o ef,-1)o (ef,o v,-1)
is also of class C'.

EXAMPLE 3.1. Let/: S2-> R be given by f(x, y, z) = x 2 +

z. We show that
/is differentiable at p = (0, 0, 1) by using the chart (U, cf,)in which U is
the upper hemisphere and cf,(x,y, z) = (x, y). Since
ef,-1(u1,u2) = (u1, u2, ../1 _ (u1)2 _ (u2)2),
we have f o ef,-1(u 1, u 2) = (u 1) 2 + ../1- (u 1) 2 - (u2) 2, which is of class
c~ at ef,(O,0, 1) = (0, 0).
DEFINITION.If/: Mm-> Nn is a function between c~ manifolds, f is dif-
ferentiable of class C1 (l < oo) at p E M if there is a chart (U, ef,)about
p on Mand a chart (V, v,) about f(p) on N such that f(U) c V and
v, of o ef,-1: ef,(U)-> y,(V) is of class C1at ef,(p)./ is of class C1 if it is of
class C' at each point. If the precise value of l is unimportant, then we
simply say that/ is differentiable.

ff(M)
DEFINITION. = {!: M-> RI/ is of class c~}.
3.2. The function of Example 3.1 is an element of ff(S2).
EXAMPLE

DEFINITION.If G is a manifold which is also a group such that both

µ: G x G ---> G by µ(x, y) = xy and 1: G - G by i(x) = x- 1 are of
class c~, then G is called a Lie group.

Under addition, Example 2.2 is a Lie group. Under matrix multiplica-

tion, Examples 2.4 and i.1 and Problem 2.5 are Lie groups. We have defined
210 Introduction to Manifolds Chap.7

Lie groups because of their importance in both mathematics and physics.

We shall not pursue the study of Lie groups here. The reader is urged to con-
sult Warner [1971) for more information.
Before we continue we need certain technical results on the existence of
differentiable functions with certain properties on a manifold M. We shall
only sketch the proofs and leave the details to the reader.

LEMMA3.3. Let W be an open set in M with p E W. Then there is an open

set V with p E V c: W and a C function/: M ---+ R such that
00

(a) /(x) = I if x E V;
(b) /(x) = 0 if x ¢ W; and
/(x) < 1 for all x E M.
(c) 0::::;;;;

Proof: Let (U, 4,)be a chart about p with Uc: Wand ef,(p)= 0 E R•. With-
out loss of generality we may assume that B 2 (0) c: 4,(U).Leth be the func-
tion h: R ---+ R of Problem 3.6. Define g: R• ---+ R by g(y) = h(ly 12) and
define/: M ---+ R by

I (x) = { (g o t)(x)
otherwise.
Then this/satisfies the conditions of the lemma if V = 4,-1(B 112 (0)). I
LEMMA3.4. Let W be a neighborhood of p E Mand let F: W---+ R be dif-
ferentiable. Then there exists a differentiable function F: M---+ Rand an
open set V with p E V c: W such that F(x) = F(x) for all x E V.

Proof: Let f and V be as in Lemma 3.3 and define F by

F(x) = { f (x) F(x) if x E W
0
ifx¢w.1
LEMMA3.5. Let p E M and let W be a neighborhood of p. Then
(a) there is a differentiable function/: M---+ R with /(p) = 1,/(x) = 0
if x ¢ W; and
(b) there is a differentiable function g: M---+ R with g(p) = 0, g(x) = I
ifx¢W.

Proof: Problem 3.7. I

We are now able to define the concept of a tangent vector and prove that
the set of tangent vectors to M• at a point forms a vector space of dimension n.

DEFINITION.A tangent vector to Mat pis a function X,: ff(M)--+R, whose

value at/ is denoted X,f or X,(f), such that for all I, g E ff(M) and
r ER,
Sec. 7-3 Tangent Vectors and the Tangent Space 211

(a) X"(f + g) = X"(f) + X,(g);

(b) X"(rf) = rX"(f); and
(c) Xifg) = f(p)X,(g) + g(p)X"(f), where/g is the ordinary product
of functions/ and g and f(p)X,(g) is the product of real numbers
f (p) and Xig).

We urge the reader to read "X"f" as "the directional derivative of/in

the direction X" at p." Condition (c) is the product rule for derivatives.

LEMMA 3.6. Let X" be a tangent vector at p e M.

(a) If/, g E ff(M) with f(p) = g(p) = 0, then Xifg) = 0.
(b) Ifr is the constant function r(m) = r, then Xir) = 0.
(c) If/, g E ff(M) and f(x) = g(x) for all x in some neighborhood W
of p, then X"(f) = X,(g).

Proof: (a) This is immediate from the definition of a tangent vector.

(b) Let 1 be the constant function l(x) = 1. Then
X,(1) = X,(1·1) = X,(1) + X,(1) = 2X,(1).
Hence X,(l) = 0 and X,(r) = X,(rl) = rX,(1) = 0.
(c) It suffices to show that if h = 0 on W, then X,(h) = 0. (Leth=
f - g.) By Lemma 3.5 there is a differentiable functionµ: M--+ R such that
µ(p) = 0 and µ(x) = 1 if x ¢ W.
We first show that (hµ)(x) = h(x) for all x E M. If x E W, then
(hµ)(x) = h(x)µ(x) = O· µ(x) = 0 = h(x).
If x ¢. W, then (hµ)(x) = h(x)µ(x) = h(x). Thus hµ = h. Hence
X,(h) = X,(hµ) = h(p)X,(µ) + µ(p)X"(f) = 0. I
Because of part (c) of this lemma, it makes sense to differentiate a func-
tion defined only in a neighborhood U of p: if/ e ff(U), then X,(f) is defined
to be X,(]) where J is any function in ff(M) that agrees with f on some
neighborhood ofp. Lemma 3.6(c) says this does not depend upon the choice
of J and_Lemma 3.4 says that such an extension J off exists.

EXAMPLE 3.7. Let ix: (-f, f)-+ M be a differentiable curve in M with

«x(O)= p. Let X" .. be defined by X" ""(/) = (d(f o «x)/dt)(O),where the
derivative on the right-hand side is the usual derivative of the real-valued
function of a real variable,/ o ix. That X/ is a tangent vector at p is the
content of Problem 3.5.

This example shows that our new concept of tangent vector agrees with
that on surfaces. In Proposition 7.4 of Chapter 4 we showed that a tangent
vector X" to a surface may be viewed as a directional derivative by choosing
212 Introduction to Manifolds Chap. 7

some curve whose velocity vector is X, and then differentiating along the
curve. This is what Example 3.7 is doing.
We now define some important tangent vectors (8/8x'), to M atp. They
will play the same role for manifolds as the x, did in surface theory. In par-
ticular, they will serve as a basis of the tangent space of Mat p.

DEFINITION. Let (U, f,) be a chart about p e M and let u 1, u2 , , , • , un be

Cartesian coordinates in Rn. Then (8/8x'), is the tangent vector given by

(3-1)

(That (8/8x'), really is a tangent vector is Problem 3.4.)

EXAMPLE 3.8. Let /: S2 --+ R be given by f (x, y, z) = x 2 + z and let

p = (1/,./2, 0, 1/,./2). If (U, f,) is the chart given by the upper hemi-
sphere U and the function f,(x, y, z) = (x, y), then
(Jo ;-1)(u1, u2) = (u1)2 + ,v'l _ (u1)2 _ (u2)2
and f,(p) = (1/,./2, 0). Hence

(_!_)
8x ,
1
(/)= 8(/ o
8u1
f( 1)~
'
o)
= ( 2u 1 ui )( I )
-
j'I _ (u1)2 _ (u2)2 ;:rt ,0
=,./2- I.
Similarly

We cannot stress enough that the value of (8/8x'),f depends critically on

the chart used. SeeProblem 3.1.

DEFINITION. The ith (local) coordinate function on M with respect to a chart

(U, f,) is the function x': U--+ R given by x'(m) = u'(f,(m)).

Note that since x': U--+ R, it makes sense to compute X,x' if p E U.

In particular, we have as a result of Equation (3-1)

(3-2) (fx;),
(x') = ~,' i, j ::;;;:n.
for all 1 ::;;;:

As on a surface, we have parametric curves given by all but one param-

eter u' held constant. It should be true (by the general "poetry" of mathe-
matics) that if a., is a parametric curve then X,"' (which on a surface was
Sec. 7-3 Tangent Vectors and the Tangent Space 213

xi(p)) corresponds to (iJ/iJx1)r The following computation shows that this is

indeed the case.
Let p = ef,-1(a) and write a= (a 1 , ••• , an) E Rn. Let
cx;(t)= ¢,-1(a1, ... , a' + t, ... , an).
This is exactly the analogue of a parametric curve. (Compare with the
definition of parametric curve in Section 4-1, remembering that for a surface
¢,-1 is x.) We show that X/• = (iJ/iJx1)P by showing that Xp''f = (iJf/iJx1)(p)
for all/ E lJ{M).

X/'f = d(fd~ cx;)(O)= o(f ;l-1)(a1, ... ' an)= (a:1)

(f)p

where the next to the last equality uses the chain rule and the last equality is
Equation (3-1).

DEFINITION. The tangent space to Mat p, TpM, is the set of all tangent vectors
to M atp.

PROPOSITION 3.9. TPM is a vector space.

Proof: If Xp, YP E TpM and r E R, we define XP + YP and rXP by

+ Yp)(f) = Xpf + Ypf and
(Xp (rXp)f = r·Xp(f).
By Problem 3.3, XP + YP and rXP belong to TPM. It is then trivial to
check the rest of the axioms of a vector space. Note that the O of this vector
space is the operator which is zero on all functions. I

We next wish to show that the set {(iJ/iJx')p11< i < n} is a basis of

TpM. This requires the following technical result which is essentially Taylor's
Theorem with remainder.

LEMMA3.10. If F: Rn --+ R is a function of class Ck and (a 1, ••• , an) E Rn,

then there exists functions h11 of class Ck- 2 such that

F(u 1, ••• , un) = F(a 1, ••• , an)+ I; :~(a1, ... an)(u' - a')
+ I;h,i(u 1 , ••• , un)(u1 - a')(u' - a1)

Proof: Let u = (u 1 , ••• , un) and a = (a 1 , ••• , an). Then

F(u) - F(a) = it I
(F(t(u - a) + a)) dt
= i (I: 1
:~(t(u - a)+ a)(u' - a')) dt

= I; (u' .- a') i 1
°F.(t(u -
a 8 u•
a) + a) dt = I; g,(uXu' - a').
214 Introduction to Manifolds Chap. 7

Note that g,(a) = It (oF/ou') (t(a - a) + a) dt = (oF/ou') (a). Similarly

g;(u) - g,(a) = I: h,iu)(u' - al). Hence
F(u) = F(a) + I: g,(_u)(u'- a')
= F(a) + I: (g,(_a)+ I: h,iu)(u 1 - al))(u' - a')
= F(a) + I: g1(a)(u' - a') + I: h,1(u)(u' - a')(ul - al)

= F(a) + I: aaF(a)(u' -
u,
a)+ I: h,,(u)(u' - a')(u' - al). I

PROPOSITION 3.11. Let (U, 4,) be a chart aboutp e M. Then

(B = {(o/ox'), II < i < n}
is a basis of T,M. In particular, T,M has dimension n.

Proof· First we show <Bis linearly independent. Suppose I: c1(o/ox'), = 0.

Then we must have O = I: c'(o/ox'),(xl) = I: c'6/ = cl. Hence <Bislinearly
independent.
Now we must show that <Bspans T,M. Let/ E ff(M), a= ef,(p),and
set F =Jo 4,- 1 • By Lemma 3.10 we have form E U

/(m) = F o 4,(m) = F o ef,(p) + I: :(ef,(p))(x'(m) - a')

+ I: h,iefJ(m))(x'(m) - a')(xl(m) - a')

or
I= /(p) + I: [ (-k,),
(/) (x' J - a') + I: (h,1 o 4,)(x' - a')(xl - al).

Hence, since /(p) is a constant, X,(f(p)) = 0 by Lemma 3.6(b) and

X,(f) = X,(f(p)) + I: [(b),(/)]X,(x' - a')

+ I: X,[(h 11 o 4,)(x' - a')(xl - al)]

= 0 + I: X,(x,) (--b),
(/)+ 0,
where the last term is zero by Lemma 3.6(a) because it is X, on the product
of the two functions (h,1 o 4,)(x' - a') and (xi - al) both of which vanish at
p = 4,- 1(a 1, ••• 'a•). Thus
(3-3) X, = I: X,(x') (b),
and <Bspans T,M. Hence <Bis a basis of T,M. I
ExAMPLE 3.12. Let M = R• and p E R•. Since (R•, id) is a chart for all of
Rn, we have {(o/ox'), 11::;;;:
i:::;;:n} is a basis for T,M for each p e R•. If
X, E T ,R•, then X, = I: a'(o/ox'), for some numbers a 1, a 2 , ••• , a•.
Sec. 7-3 Tengent Vectors end the Tengent Space 216

Hence we may associate with X, the 2n-tuple (p; .a', ... , a•). Under this
identification, the effect of (p; a 1, ••• , a•) on a function/is I: a'(of/ou'),,
which is exactly the Euclidean idea of directional derivative. We fre-
quently write X = (p; a 1, ••• , a•) (if XE T,R•) and T~· = R•.

If M 1 is the hypersurface defined by!, we will show in Section 7-5 a way

to visualize T,M 1 as a subspace of T~·+ 1 for eachp E M 1 •

PROBLEMS

3.1. Let/: S 2 -+ Rand p e S 2 as in Example 3.8. Let (V, 1/1)be the chart
defined by V = right hemisphere and 1/f(X,·Y,z) = (x, z). Show that
with respect to this chart (off ox'), = ~ and (of/ox 2 ), = I. Note
that this is a different result than was obtained by using the chart of
Example 3.8.
3.2. Let ct,: S"-+ p• be defined by cl,(a) = the line through the origin and
a. Show that ct, is c1rfor all k and that ct, is two-to-one. (If you have
had some algebraic topology, ct, is actually a (double) covering map.)
*3.3. If X,, Y, E T,M and r E R, prove that both X, + Y, and rX, are in
T,M.
*3.4. If (U, 4,)is a coordinate chart about p, prove that (o/ox'), as defined by
Equation (3-1) is a tangent vector at p for each i.
*3.5. If 11: (-f, f)-+ Mis a curve with 11(0)= p, show that X,• E T,M.
*3.6. The point of this problem is to show that there is a C function h: 00

h(x) :::;;;:
R -+ R such that O :::;;;: t and
I for all x E R, h(x) = I if Ix I :::;;;:
h(x) = 0 if lxl:::::I. The graph of h will look like that of Figure 7.4.
h is called a bump function.
(a) Let
if x> 0
if XS o.
Show that h 1(x) is c1rfor all k.

I
FIGURE 7.4 -1 4
218 Introduction to Manifolds Chap. 7

(b) Let h 2 (x)=h,(x+ l)h 1(--i-x). Show that h 2 (x)>O if

-1 < x < --i but hi(x) = 0 otherwise.
(c) Show that h 3 (x) = 0 if x < -1, h 3 (x) = I if x > -!, and
s
O h3 (x) s1 for all x E R, where

h3(X) =
r hi(t) dt
J::b,hi(t) dt ·
-1

(d) Show the existence of the function h: R - R described above.

(Hint: Leth(x) = h 3(x)h 3 (-x).)
*3.7. Prove Lemma 3.5.

7-4. VECTOR FIELDS AND LIE BRACKETS

In this section we introduce two concepts (vector fields and Lie brackets)
which we will need later. Let M be a fixed C n-manifold.
00

DEFINITION. A field of vectors X is an assignment of a tangent vector

X, e T,Mto eachp e M. If Xis a field of vectors and/ e o:(M), then
we may define a real-valued function Xf on M by (Xf)(p) = X,f.
If Xf e ij(M) for each/ e 0:(M), then Xis called a vectorfield.
Note that if Xis a vector field and (U, ~) is a patch, then (since the x'
are differentiable functions) Equation (3-3) shows that X can be written as
X,.,= I: X 1(p)(8/8x'), for p e U.
This is usually written X = I: X 1(8/8x'). This definition makes sense
if we only want X to be defined on an open set U of M, in which case we have
a vector field on· U. We will write l:(M) for the set of all vector fields on M.
The zero vector field, X, is defined by X, = 0 e T,M. There are certainly
nonzero vector fields in l:(M), as Problem 4.2 shows. The question of
whether, given M, there is a vector field which is never zero on Mis much
more difficult. (Remember: Xis nonzero if X, =t= 0 for some p. We are look-
ing for a vector field X with X, =t= 0 for all p.) In Theorem 7.3 of Chapter 6
we proved that for a compact surface there was a vector field on M which is
never zero only if the Euler characteristic of M is zero. This theorem gen-
eralizes to compact, orientable, even-dimensional manifolds for a suitable
definition of Euler characteristic. See Milnor [1965].
Since T,M is a real vector space, it comes as no surprise that l(M) is
also a vector space (of infinite dimension). We can also define JX for
f e o:{M) and X e l:(M). More precisely, if X, Y e l:(M), r e R, and
f e o:{M), we may define X + Y, rX, and/Xby
(X + Y), = X, + Y,
(rX), = r·X,
Sec. 7-4 Vector Fields and Lie Brackets 217

and
(fX), = f(p)X,.
The reader is warned: fX is a vector field while Xf is a function.
Note that Xfis a c- function so that Y,(Xf) makes sense.

DEFINmoN.If X, Y E x(M), then the Lie bracket of X and Y, [X, Y], is the
field of vectors defined by
(4-1) [X, Y],f= X,(Yf)- Y,(Xf) for/ E g:(M) andp EM.

LEMMA 4.1. [X, Y] is a vector field on M.

Proof: It is immediate from (4-1) that [X, Y],(f + g) = [X, Y],f + [X, Y],g
and [X, Y],(rf) = r[X, Y],f for r E R and f, g E g:(M). To show that
[X, Y], E T,M we compute [X, Y],(fg). Note that since
Y ,(Jg) = f(p)Y ,g + g(p)Y,f,
we have Y(fg) = fYg + g Yf, where we are multiplying the two differentiable
functions in each term of the right-hand side together (i.e., f and Yg are
multiplied as functions). Now
[X, Y],(fg) = X,(Y(fg))- Y,(X(fg))
= X,(!Yg + gYf)- Y ,(!Xg + gXf)
= f(p)X,Yg + (Yg)(p)X,f + g(p)X,Yf + (Yf)(p)X,g
- f(p)Y,Xg - (Xg)(p)Y ,f - g(p)Y,Xf - (Xf)(p)Y ,g
= f (p)[X, Y],g + g(p)[X, Y],f.
Therefore [X, Y], E T ,M.
To show that [X, Y] E x(M) we note that Yfand Xfare in ff(M); then
so are X( Yf), Y(Xf), and their difference [X, Y]f. I

ExAMPLE 4.2. Let M = R 2 , X = x 1x2(a/ax•) and Y = x 2(a/ax 2). If

g(x', x 2) = (x 1) 2x 2, then gX = (x 1) 3(x 2) 2(a/ax 1) but
Xg = x 1x 2(2x 1x 2) = 2(x 1) 2(x 2) 2.
If/ E ff(R2 ), then

(X, Y]f = x•x 2 -b(x2 Jfi)- x 2 fxz(x'x 2 Jf.)

= x'(x2)2 a 2J - x2xl a1 - xl(x2)2 a 2J
ax• ax 2 ax• ai2P
= -x2xl af,
Txi
Thus [X, Y] = -x 1x 2(a/ax 1).
218 Introduction to Manifolds Chap. 7

The next to last line in the above computation is quite revealing

because in it we see that second order terms (the mixed partials) cancel
out. This is the reason that, although/-+ X;(Yf) is not a tangent vector,
J-+ X,(Yf) - Y;(Xf) is a tangent vector.
The following lemma is left as an exercise.

LEMMA 4.3. If X, Y, Z E ~(M) and r E R, then

(a) [X, Y] = -[Y, X] and [rX, Y] = r[X, Y];
(b) [X + Y, Z] = [X, Z] + [Y, Z], [Z, X + Y] = [Z, X] + [Z, Y];
(c) (Jacobi's Identity.) [[X, Y], Z] + [[Y, Z], X] + [[Z, X], Y] = 0.

Note in particular that the Lie bracket "multiplication" of vector fields

is not associative but instead satisfies the rather strange substitute of Lemma
4.3(c). Conditions (a), (b), and (c) should remind you of the crossproduct of
vectors in R 3 •

PROBLEMS

*4.1. Let X, E T,M for some point p E M. Show that there is a neighbor-
hood U of p and a vector field X defined on U such that X, = X ,.
(Hint: Use local coordinates.)
*4.2. Let X, E T ,M for some point p E M. Show that there is a vector field,
X, on M such that X, = X,. (Hint: Use Problem 4.1 and Lemma 3.5.)
In particular, there is a nonzero vector field on M.
4.3. Let (U, if,)be a chart with {(a/ax1),, ••• , (a/axn),}
as basis for T,M for
p EU.

(a) Show that [fx;,a!,]=

0.
(b) Show that if X = :EX 1(a/ax1) and Y= :E Y1(a/ax1), then
ex.
'
YJ= :E(x'ay,_ y,ax'\.J.__
ax, ax ax 1} 1

Compare this with Problems 9.5 and 9.8 of Chapter 4.

4.4. If/, g E ff(M) and X, YE ~(M), show that
[JX,gY] =fg[X, Y] + f(Xg)Y-g(Yf)X.
*4.S. Prove Lemma 4.3.
4.6. Mn is called parallelizable ifthere are n vector fields, X1t ... , Xn, such
that {(X1),, ••• , (Xn),} is a basis for T,M for each p E M. Prove that
Rn and T" are parallelizable. It is often quite difficult to decide if a
Sec. 7-5 The Differential of a Map and Subman,1olds 219

manifold is parallelizable. s• is parallelizable if and only if n = 1, 3,

or 7. s, X S' is parallelizable if p and q are both odd. (Do not try to
prove these last assertions I)

7-5. THE DIFFERENTIAL OF A MAP

AND SUBMANIFOLDS

In this section we introduce the concept of the differential of a mapping

between manifolds and then apply this notion to the idea of submanif olds of
a given manifold. It is this last idea that will provide many examples of the
geometric concepts which have been treated already and those which follow.
In particular, we will have a concrete way of visualizing the tangent space of
the hypersurface defined by a function f We shall assume throughout that
M and N are manifolds and that ct,: M - N is differentiable. We want to
define for each X, e T,M an element (ct,.),(X,) of T'llc,1Nin such a way that
the assignment X, - (ct,.),(X,) is a linear transformation for each p e M.
One way of doing this would be to define (ct,.), on a basis of T,M, i.e., view
(ct,.), as a matrix instead of a linear transformation. There is a natural candi-
date for the matrix definition of (ct,.),-just take the Jacobian of'I' o ct, o ~- 1
(evaluated at ~(p)) where (U, ~) is a chart aboutp and (V, 'I') is a chart about
ct,(p). The trouble with this definition is that it is not an invariant definition;
that is, the definition itself depends on a coordinate chart. This is contrary
to our desire for invariance. It also causes many technical problems (in the
proof of other results) involving picking the "right" coordinate chart to do
the computations. The definition which we will use avoids these technical and
aesthetic problems but is, unfortunately, quite abstract.

DEFINITION. If It,: M - N is differentiable, the differential of ct, at p is the

function (ct,.),: T,M - T'llc,1N defined by
(5-1) (ct,.),(X,)(f) = X,(f o ct,)
where X, e T,M and/ e o:(N).

LEMMA 5.1. (ct,.), is well defined; that is, if ct,: M - N and X, E T,M, then
(ct,,)*(X,) E T'llc,,N.

Proof: For notational convenience we drop the subscript p in this proof.

We must show that, for all/, g e ij(N) and r e R,
(a)ct,.(X)(/ + g) = .i,.(X)(f) + .i,.(X)(g),
(b)ct,.(X)(rf) = rct,.(X)(f), and
(c) .i,.(X)(fg) = f(e~p)')'b.(X)(g) + g(ct,(p))ct,.(X)(f).
220 Introduction to Manifolds Chap. 7

First we prove (a):

«»*(X)(f +g)= X((f+g)o«.)= X(fo«.)+ X(go«.)
= «,*(X)(f) + «»*(X)(g),
where we have used Equation (5-1) repeatedly.
Next we prove (b):
«.*(X)(rf) = X(rf o «,) = rX(fo «») = r«»*(X)(f).
For (c) we note that ((Jg) o «»)(q) = (/g)(«.(q)) = f(«.(q))g(«.(q))
for any q E M so that
(Jg) 0 «» = (! 0 «»)(g O «»).
Hence,
«»*(X)(fg) = X((fg) o «») = X((f o «.)(go «»))
= (! o «.)(p)X(g o «») + (go «»)(p)X(f o «»)
= f(«.(p))«»*(X)g + g(«,(p))«»*(X)f. I
PROPOSITION 5.2. Let«»: M------>
N and p E M. Then («»*)P: T 11M-+ Tq;c11,N
is a linear transformation.

Proof· Let r E R, X 11, Y 11 E T11M. We shall prove that

(«»*)p(rXP +Y 11) = r(«.*)PXP + («»*)PYr
For any f E ~(N):
((«»*)ArXP + YP))(f) = (rXP + YP)(f o «»)
= rXif o «») + Yp(f o «»)
= r(«»*)AXP)f + («»*)AY )f.
11 I
PROPOSITION5.3. If«»: M---> N and 'I': N---> Pare differentiable maps of the
manifolds M, N, and P, if p E M and q = «»(p), then
(('I' O «»)*)P = ('I' *)q O («»*)P"
(More succinctly: ('I' o «»)* = 'I'* o «,*.)

Proof: Problem 5.1. I

PROPOSITION 5.4. Let«,: M-+ N, p E M. Let ( U, cf,)be a chart about p and
(V, v,) be a chart about q = «.(p) such that «,(U) c V. If we write
{(a/ax1)p Ii= I, ... , m} for the basis of TPM derived from (U, cf,)and
{(a/ay 1) 9 li =I, ... , n} for the basis of T9 N derived from (V, v,), then
the matrix of («»*)P with respect to these bases is the Jacobian of
VI o «» o ct,-1:
( 5_2) <«» ) (( a) ) = ~ a(yi o «» o ct,-\s) ( a ) ,
* p 'uxi p '7" 1 au lyi q
wheres= cf,(p).
Sec. 7-5 The DiHerential of a Map and Submanifold, 221

Proof: By Equation (3-3) we know that

= ~ [<"'•>,(fx;},
((b.),(fx;), (y') J (btc,1
a ) , (y' "'>]
= E [( 1xi j
(ay1
a) 0
~(p)

An application of (3-1) gives (5-2). I

It is also convenient to write (I)' = y' o (I) so that

"' 0 (l)(p) = ((l)'(p), ... ' (l)•(p)).

With this notation (5-2) becomes

(5-3) ((I).),
a), =
(1xi J~
n a(/)/
N(p)
(a)
oi1'
EXAMPLE 5.5. Let M = M 1 be the hypersurface induced by/: R•+ 1 - R.
(See Theorem 2.6.) Let (I): M - R•+ 1 be inclusion. Let p e M such that
a11au•+1 (p) *
0. In the natural coordinates for R•+ 1 and the coordi-
nates about p given by Theorem 2.6 we see that

(5-4) ((I),)* (fx;),

=e,(p) + Pu,(p)en+ (p)1

where e1(p) = (p; 0, ... , 1, 0, ... , 0) as discussed in Example 3.12.

Note that in the notation of the proposition, V = Rn+1, 1/1is the
identity, and tt,·1(u 1 , ••• , u•) = (u 1, ••• , u•,g(u 1 , ••• , u•)). Thus
1/1o (I) o tf,"1(u 1, ••• , u•) = (u 1, ••• , u•,g(u 1, ••• , u•)) and Equation
(5-4) follows by computing the Jacobian of 1/1o (I) o tt,·1.

DEFINmoN. Let M,. and N" be manifolds. Mis a submanifold of N if there is

a differentiable function (I): M - N such that (I) is one-to-one and ((I).),
is one-to-one for each p e M. Such a map (I) is called an embedding of
MinN.

We remark that: (1) if Mis a submanifold of N, then dim M ~ dim N;

(2) it is possible for ((I).), to be one-to-one for all p without (I) being an
embedding (Example 1.8); and (3) it is possible for (I) to be one-to-one with-
out (I) being an embedding (take (I): R 1 - R 1 with (l)(t) = t 3).
Because ((b.),isone-to-one for an embedding we may identify T,M and
((l).),T,M c:: T~1, 1N. Using this identification we are simply viewing the
tangent space to M as a subspace of the tangent space to N. This is very
important intuitively in the case when N = am+k because it is exactly how we
define the tangent plane to a surface (i.e., when m = 2 and k = 1). (Recall
that the tangent plane to a surface at p was defined as a certain subspace of
T ,R 3 = R 3 .) One word of caution: we must be careful not to forget the em-
bedding itself. (There are many ways to embed R 1 in R 2 1)
222 Introduction to Manifolds Chap, 7

The next proposition gives us even more faith in this abstract approach
because it reaffirms our intuition.

PROPOSITION 5.6. If M = M1 is the hypersurface defined by/: R•+l - R,

then T,M is isomorphic (as a vector space) to
('1>.),T,M = {X, e T,R•+ 1j((gradf),, X,) = O},
where '1>is the inclusion of M in R•+1.

Proof· We already know that ('1>.),is an isomorphism, so we need only show

the equality. Actually since both subspaces have the same dimension we need
only show that ('1>.),(T,M) c {X, E T,Rn+1I((grad/).,, X,) = O}= &. It
is sufficient that we show that ('1>.),(i/ix), e & for each i = l, 2, ... , n,
where we are now in some chart about p. We shall do this assuming that
(at p) if/iun+ 1 #:-0. (The computation for the other charts is similar.) Using
Example 5.5 we see that
a) (grad!), ) = (e,(p) + art(p)e.+1CP),
(('1>.),(rx;; ig ft1o,il(p)eip)
if n+ l )

or
(5-5) (<'1>.),(-k,); (grad/).,)= :G,(p) + 3:,(p) a:!+
(p).1

Recall, however, that g = g(u 1, ••• , u•) was defined in such a way that
f (u 1, ... , u•, g(u 1, ••• , u•)) = 0. Differentiate this equation with respect to
u1 to obtain
§G+8;{.~=o.
so that the right-hand side of Equation (5-5) is zero. ('1>.),(i/ix), is therefore
in & for each l :::;;;i < n. I

We have defined embeddings in this book so that we could present

Proposition 5.6 which enables us to concretely visualize T,M if Mis a hyper-
surface defined by a function. The subject of embeddings is a very deep and
subtle one. Even some very simple sounding questions (e.g., "For what k
does P• embed in Rn+k ?") are very difficult. Because the theory of embedding
is actually a part of differential topology (not differential geometry), we will
only give the following theorem (whose proof is in Guillemin and Pollack
[1974)) and make some comments. (One word of caution: the definition of
embedding in differential topology is slightly different than that in differ-
ential geometry-in the former one also assumes that M has the subspace
topology.)

THEOREM 5.7 (Whitney Embedding Theorem, 1937). Every n-manifold em-

beds in R 2 •+1.
Sec. 7-6 Linear Connections on Manifolds 223

A surface is a 2-manifold which is embedded in R 3 • It can be shown that

P 2 cannot be embedded in R 3 but can be embedded in R 4 so there are 2-
manifolds that are not surfaces. (In Problem 5.6 we give an embedding of
P 2 into R 5 .) An immersion (): M--+ N is a differentiable map with (()*)P
one-to-one at each point (so that an embedding is a one-to-one immersion).
Although P 2 cannot be embedded in Rl, it can be immersed in Rl (as Boy's
surface, see Hilbert and Cohn-Vossen [19521).

PROBLEMS

*S.1. Prove Proposition 5.3.

5.2. If pis the north pole (0, 0, l) and(): S2--+ P 2 as in Problem 3.2, find
(()*)" as a matrix. Show it is nonsingular.
5.3. Prove if J: M--+ Mis given by I(x) = x, then(/*)"= identity for all
p EM.
5.4. A differentiable map (): M--+ N is a diffeomorphism if there is a dif-
ferentiable map 'I': N--+ M such that () o 'I' = I and 'I' o () = I.
(a) If() is a diffeomorphism, prove that (()*)" is an isomorphism for
allp EM.
(b) If there is a diffeomorphism between M and N then dim M =
dimN.
S.S. Let G be a Lie group and g E G. Define L 11: G--+ G by L 11(x) = gx
(multiplication in the group).
(a) If x. E T.G, prove that the field of vectors X given by X 11 =
(L 11)*.(X.) is actually a vector field. Xis called a left-invariant vector
field on G. (e denotes the identity of the group G.)
(b) Show that any Lie group is parallelizable. (See Problem 4.6.)
5.6. Let(): P 2 --+ Rs be defined as follows: Let I E P 2; then I= [(a 1, a 2, al)]
for some a= (a 1 , a 2, al) E Rl with (a 1) 2 + (a 2) 2 + (al) 2 = l. (There
are two such points a.) Let()(/)= ((a 1)2, (a 2)2, a 1a2, a 1a3, a 2al). Prove
that() is an embedding of the projective plane P 2 into Rs. (Does this
help you visualize P 2 ?)

7-6. LINEAR CONNECTIONS ON MANIFOLDS

In Chapter 4 we saw how important it was to differentiate a vector field

along a curve 11. This really meant that given a vector field Y we could dif-
ferentiate Yin the direction of the tangent to 11. This led naturally to the
definition of a parallel vector field and gave a necessary and sufficient condi-
tion for a curve to be a geodesic (that the tangent vector field be parallel
224 Introduction to Manifolds Chap. 7

along the curve). In this section we formalize the notion of differentiating a

vector field with respect to a vector field by defining a linear connection and
we present some examples and technical results in preparation for the dis-
cussion of parallel vector fields and geodesics in the next section.

DEFINITION.A linear connection on M is a function V: x(M) x x(M)--+

x(M) (which we write Vx Y) such that for X, Y, Z E I(M), r E R and
/E mM)
+ +
(a) Vx( Y Z) = Vx Y VxZ and Vxr Y = rVx Y;
+
(b) Vx+rZ = VxZ VrZ and V 1xY = fVxY; and
(c) VxfY = (Xf)Y fVxY. +
Note that Condition (c) makes sense both technically and philosophical-
ly. Technically, Xf E mM) so that (Xf)Y makes sense and the right-hand
side of (c) is a vector field. Philosophically, (c) is just a "derivative of the
product" rule. V is called a linear connection, but Vx Y should be read as the
covariant derivative of Y with respect to (or in the direction of) X. It can be
shown that any manifold admits many linear connections (Hicks [19651).
If we are in a chart ( U, if,) so that {(a/ax'),Ii = I, ... , n} is a basis for
T,M for all p E U, then any vector field X can be expressed locally in the
form I; X' ( a/ax'),where X' E m
U). Because of the various linearity prop-
erties of V and the product rule, the behavior of V is completely determined
by the values of V,wx•(a/axi). These values must be expressible as linear
combinations of the a;axkwith the coefficients in mu).

DEFINITION.Let V be a connection on Mand let (U,ef,)be a proper coordi-

nate chart. The Christoffel symbols of V with respect to (U, ef,)are the
functions r,/ E mu) defined by

(6-1) V11111x{k)= ~r,/ a!k·

Note that nothing in this definition allows us to conclude that
r,/ and rJ are equal. In fact this is often not true.
We will show in the next section that (in the presence of a Riemannian
metric) these Christoffel symbols play the same role that those of Chapter 4
did. Condition (c) of the definition ofV x Y shows why r,/ had such a strange
transformation law (see Problem 4.11 of Chapter 4). Note that if M can be
covered with one chart, then giving the r,/ with respect to this chart defines
a unique linear connection. We will use this approach in Example 6.2.

EXAMPLE6.1 (Flat Euclidean Space). Let M = Rn. If YE x(Rn), then

Y = :Ef'e, for some / 1 E ij(Rn). We define the flat connection on Rn
by VxY = l: (Xf')e
I= I
1• That this defines a linear connection is the content
Sec. 7-6 Linear Connections on Manifolds 226

of Problem 6.1. Note that Vmx•(a/ax') = 0 for all i and j so that

r 1/ = 0 for all i,j, and k. V x Y is the usual directional derivative of a
vector-valued function.

ExAMPLE6.2 (Hyperbolic (or Poincare) Upper Half Plane). Let

M = H = {(x,y) E R 2 !y > OJ.
r
Define a linear connection on H by setting all 1/ to zero except that
r 12 1 = r 21 1 = -r 11 2 = r 22 2 = -1/y. This gives a connection onH
because H can be covered by one coordinate chart.

Before giving another example we shall need some technical results. We

shall fix a manifold M with linear connection V.

LEMMA6.3. If X, YE I(M) and X, = Y, for some p E M, then {VxZ), =

(VrZ), for allZ E I(M).

Proof: If W = X - Y, then W, = 0. Thus, in some coordinate chart about

p, W = L p ca;axi) and fi(p) = 0 for all i. This means that for any Z E
I(M), {VwZ), = L /1(p){V,1,x.Z), = 0. I

What Lemma 6.3 means is that (VxY), depends only on the value of X
at the point p. The Y dependence is not so simple. It depends on the value of
Y along any curve which "fits" X as we see in Lemma 6.6.

DEFINITION.Z is a vector field along the curve ii: I-+ M if Z assigns to each
t E I an element z..
<,>E T,.,,1M such that t-+ z..
<,i(/) is a differentiable
real-valued function oft for each/ E ff(M).

DEFINITION.Let ii: I-+ M be a curve. The tangent vector field to ii, T,., is
given by (T ..) ..<,>= (ii*),(d/dt), where t is the natural coordinate for I.

If there is no ambiguity, we write T for T,.. Note that T.. is an example of

a vector field along ii. In local coordinates about ii(t 0 ) we have

(6-2) dii1 (
T ..(,) = ~ dt (t) 8x'
a) IX(t)

(see Equation (5-3)) just as in the Euclidean case. This is the reason one often
sees dii/dt used instead of T ... We have no concept of length at this point so
we cannot talk about T.. being a unit vector. Note T,. is a function of t, not
ii(t). This was called the velocity vector field in Chapter 2.

DEFINITION.Let ii: I-+ Mand YE I(M) be given. We shall now define the
covariatJt derivative of Y with respect to ti, Vr Y. Let t O be given and
let X be any vector field on M such that X,.<,,>= T,.<,,>·(Such an X
228 Introduction to Manifolds Chap. 7

exists by Problem 4.2.) VTY is the vector field along iz defined by

(VTY).<,,>= CV x Y).<ro>·
(This is well defined by Lemma 6.3.)
EXAMPLE 6.4. Let M be flat Euclidean 2-space, iz(t) =(cost, sin t) for
0 < t < 2n and Y = ye, - xe 2 • Since T = -sin t e 1 cost e2 , if we +
+
let X = -ye 1 xe 2 , then X.<,>= T.w Therefore,
VTY= VxY= Xye 1 - Xxe 2 = xe 1 + ye 2
so that (VTY) = cos t e 1 + sin t e2 •
LEMMA 6.5. If f e ff(M) and iz: 1- Mare such that f o c,r;is constant, then
T,.f= 0.
Proof· T,J = iz.(d/dt)f = (d/dt) (f o iz) = 0. I
LEMMA 6.6. Let iz : I - M be a curve and Y, Z e !(M) such that
=Z,.<,>for all t e /. Then VT,.Y = VT,.Zalong c,r;.
Y,.<,>
Proof· Let W = Y - Z so that W,.<,> = 0. We now show that VTW = 0.
In local coordinates about some point on the curve we may write
W = ~ f' (i/ix'), where /'(iz(t)) = 0. Therefore using Property (c) of the
definition of V, we have
VTW= :E T,.f',b + :Ef'VT(-b)·
Since f' o iz is constant (it is actually zero), the first term of the right-hand side
is zero by Lemma 6.5. Since f' is zero at iz(t), the second term is zero when
evaluated along c,r;(t). I
Three remarks about Lemma 6.6 are in order: (1) the conclusion
VTY = VTZ is only valid at points along the curve 11,(in fact, it does not
make sense at any other point); (2) an examination of the proof shows
that if t O e I and Y,.<,>= Z,.<,>for t in some interval about t 0, then
CVT Y),.<,,> (3) if Y is a vector field along iz, then we may define
= (YTZ),.<,,>;
VT. Y to be a vector field along iz by picking any Y e !(M) such that
Y•<r>= Y,.<,>in some neighborhood of t O in I and setting
Y).(,,) = (VTf).(,,)·
<YT
This is well defined by Lemma 6.5.
We shall now assume that Mis a submanifold of N and that M has been
identified as a subset of N and T,M as a subset of T,N. If N has a linear con-
nection ~ and X, Y e !(M), then we may extend X and Y to vector fields
on N which agree (at least locally) with X, Y; hence ~xY makes sense for
X, Y e !(M). (It is well defined by Lemmas 6.3 and 6.6.) Since ~ obeys the
rules for a linear connection, the overly optimistic reader is tempted to think
that ~ is a linear connection on M. Alas, this is not true because ~x Y need
not be in l(M) (although it must be in '£(N), of course). Specifically look at
Sec. 7-7 Pare/le/ Vector Fields end Geodesics on e Menffold 227

the case when M = S1, N = R 2 , X = -ye,+ xe 2 , and Y = ye, - xe 2 •

Since S1 is the hypersurface of R 2 defined by f(x, y) = x 2 + y 2 - 1 and
grad/= 2xe 1 + 2ye 2 , Proposition 5.6 shows that both X and Y are in I(S 1).
However, in Example 6.4, we computed Vx Y to be xe 1 + ye 2 which is not
in I(S 1). However, guided by Gauss's formula (Equation (4-8) of Chapter 4)
we can prove the following proposition:

PROPOSITION 6.7. Let M = M 1 be the hypersurface defined by!, let V be

the linear connection of flat Euclidean (n + I )-space, and write
N = grad f IIgrad f I· If X, Y E I(M), then
(6-3) VxY=VxY-(VxY,N)N
defines a linear connection on M.

Proof· We only show that Vx Y E I(M) and leave the rest as Problem 6.2.
Vx Y E I(M) if and only if (V x Y, N) = 0 (by Proposition 5.6), but this is
clear. I

It will be shown in Section 7-8 (Problem 8.4) that

(VxY, N) =-(Y, VxN).
If we write down explicitly what VxN is, we find that VxN = -L(X) is our
old friend the Weingarten map (if n = 2)1 Now write out Equation (6-3)
in local coordinates. It should be Gauss's formulas (4-8) of Chapter 4.

PROBLEMS

6.1. Prove that flat R• (Example 6.1) is a manifold with linear connection.
*6.2. Prove Proposition 6.7.
6.3. Let H be the hyperbolic upper half plane (Example 6.2) and let
IZ(t)= (t, 1). If
(-b) (h) •
Y(t) = sin t ux· •<t>+ cos t ux- ,.,,,
prove that VT,.Y = 0.
6.4. Prove that Equation (6-3) yields Gauss's formulas when n = 2.

7-7. PARALLEL VECTOR FIELDS AND

GEODESICS ON A MANIFOLD WITH
A LINEAR CONNECTION

In this section we define the notions of parallel vector fields, parallel

translation, and geodesics on a manifold with linear connection. We show
that parallel translation is the "global" version of covariant differentiation
228 Introduction to Mil/nifolds Chap. 7

and also compute the geodesics of the classical geometries, R 2 , H, and S2. We
fix a manifold M and linear connection V.
What should we take for the definition of a vector field Y being parallel
along a curve«? For a surface it was that the derivative of Y was normal to
the surface. Equation (6-3) tells us that this happens precisely when VT"Y = 0.

Let Ybe a vector field along«. Yisparallel along« if VT.. Y

DEFINITION. = 0.

EXAMPLE 7.1. For flat Euclidean space a vector field Y is parallel along« if
and only if Y..1,, = («(t); a 1, ••• , an) for some constants a1 E R. (See
Problem 7.1.)

A review of Section 4-6 will enable you to guess all of the results (and
proofs) of this section (except Proposition 7.6).

THEOREM 7.2 (T. Levi-Civita, 1917). Let «: [c, d]--+ M, p = «(c) and
YJI E TJIM. Then there is a unique vector field Y parallel along« such
= y I>'
that YGC(c)

Proof: The proof proceeds exactly like that of Theorem 6.7 of Chapter 4.
Assume that we are in a chart (U, if,)aboutp and that Y = l: Y'(t)(a/ax') ..1,,
is a vector field along«; where the Y' are differentiable. Y is parallel along«
if and only if
0 = VTY = ~ [ e:;)(a!,)
+ Y'VT a!,J
or
(7-1) o= l: [(dJ,)+ Y r,/] b 1~ 1

where «(t) = ef,-1(« 1(t), ... , «n(t)). Thus Y is parallel along« if and only if
the Y 1 solve the initial value problem
dYk + l: Y' d«' r k= o k = 1, ... , n
(7-2) { di dt JI
Yk(c) = fk.
An application of Picard's Theorem finishes the theorem in a neighborhood
of ix(c). Now repeat this process until we get to ix(d). I

EXAMPLE 7.3. Let H be the hyperbolic upper half plane of Example 6.2 and
let «(t) = (t, 1) with c = 0 (and d > 0) so that p = «(O) = (0, 1). Let
YJI= (a/ax 2 )JI so that in the notation of Equation (7-2), Y1 = 0 and
Y2 = 1. Equation (7-2) becomes
yi dix2 _l __ Y2dlx1 _I_= O
dY 1 _

{ dt dt «2 (t) dt «2(t)
dY + yi dix _l_ _ y2 dix2 _1_ = o
2 1
dt dt «2 (t) dt «2 (t)
Sec. 7-7 P11111/le/
Vector Field, and Geod111ic1
on II M11n,Yold 229

or
(7-3) dY' - Y2 = 0
dt
and dY"'
dt
+ Y1 = 0.
This coupled system (7-3) can be solved by obtaining
(d 2 Y 1'fdt+ Y 2 = 0 so that
2) Y 2 (t) = c 1 sin t c 2 cost. Since +
Y (0) = Y = 1 we see that c2 = 1. Going back to Equation (7-3) yields
2 2

Y 1(t) = -c 1 cost+ sin t. Since Y1 (0) = 0, c1 = 0, and so

Y(t) = Y 1(t)(a/ax 1),.,,, + Y (t)(a/ax
2 2 ).,,,

or
Y(t) = sin t(a/ax 1),.,, 1 + cos t(a/1Jx 2),. 111
•

(See Figure 7.5.)

FIGURE 7.6 - - - - - - - - -

DEFINITION.Let iz be a given curve. We define parallel translation along iz,

P•: T,.,,,,M__. T,.,,.,Mby P"'(Y,.,,,,)= Y,.,,.1 where Y is the unique paral-
lel vector field along IZ such that Y,.,,,,= Y,.,,,,.
7.4. For each curve iz,P•: T ,.,,,,M __.T,.,,,,Mis an isomorphism.
PROPOSITION
Proof: Problem 7.2. I
EXAMPLE7.5. Let M = H be the hyperbolic half plane and let iz(t) = (t, 1).
Example 7.3 shows that if t 1 = 0 then

p• (-b)
ux-
= sin t (-b) 2
ux· c,,,t>
+ (cos t.,.)(h)
ux- c,,,o
·
It should be pointed out that Example 7.3 was picked with malice afore-
thought; that is, we picked it carefully so that we could actually solve the
initial value problem (7-2). In practice it is often quite difficult to actually
solve this initial value problem.
The next proposition shows either that parallel translation is a global
version of covariant differentiation or covariant differentiation is the infini-
tesimal version of parallel translation, depending on your viewpoint. We feel
that it gives a better intuitive feeling for covariant differentiation.

PROPOSITION 7.6. Let P•<t>:T,.1o,M__. T,.<t,Mbe parallel translation along IZ

(with respect to the linear connection V). Then
(V Y) = lim (P•<t>)-'(Y,.,,,)- Y.co>
Ts s(O) ,-...o t
230 Introduction to Manifolds Chap.7

Proof: Let Z, be the unique parallel vector field along II guaranteed by

Theorem 7.2 such that (Z,) ..ca>= (P•<•>t1(Y..c,,). In a chart we write
(Z,) ..ct>= :EZ,'(t)(i/ix') ..c,,,
T,. = :E(da.'/dt)(i/ix') ..c,>
and
Y,.c,,= :EY'(t)(i/ix') ..w
Since Z, is parallel along 11,we apply Equation (7-2) to obtain

(7-4) ~/ + t1!' Z/(t)r /(11(t)) = 0,

1 k = 1, ... , n.
Z,k(s) = Yk(s).
Applying the Mean Value Theorem to the function Z/(t) on the interval
0 ~ t ::;;:s we see that Z/(s) = Z/(0) +
sZ,k'(,k) for some O < ,k < s.
Thus the kth component of
(P"<•>t1(Y..1,,) - Y..101
s
is
Z/(0) - Yk(O) _ Z/(s) - sZ,k'"k) - Yk(O)
s s
= -Z/'(,k) + Z/(s) -; Yk(O).
Using Equation (7-4) this becomes

(7-5) ti":r'C,k)Z/(,4-)r,/(11(,k)) + Yk(s) -; Yk(O)·

Lettings--+ 0 in Formula (7-5) and remembering that ,k--+ 0 ass--+ 0, we

get the kth component of
fun (pm<•>t1(Ym(1))- Ys(D)
1-+D S
is
:E !' (O)Za'(O)r /(11(0)) + 4J:
1 (0)

= (O)Y'(O)r,/(11(0))+ d::
t1":r' (0)

= kth component of VT,.Y by Equation (7-1). I

A direct translation of the definition of geodesic applies in this setting.
Unlike Chapter 4, we do not assume T,. has length one because we can'tf
There is no notion of length of vectors on a manifold with linear connection.

DEFINITION. A curve II on Mis a geodesic (with respect to V) ifVT,.T.. = 0.

Sec. 7-7 Pere/le/ Vector Field1 end Geode1ic1 on e Menifold 231

Note that applying (7-2) in the case Y = T,. yields

(7-6) fork= 1, ... , n

as the equation of a geodesic. From this we conclude

THEOREM 7.7. Let M be a manifold with linear connection V. Let p e Mand

X, e T,M. Then there is an E > 0 and a unique geodesic iz such that
iz: (-f, f)-+ M, iz(O)= P and (diz/dt) (0) = X,.

We shall finish this section by actually computing the geodesics of the

hyperbolic upper half plane. We leave the geodesics of flat Euclidean space
as a problem (see Problem 7.1).

ExAMPLE7.8. Let H be the hyperbolic upper half plane (Example 6.2). With
the given Christoffel symbols, a geodesic iz(t) = (iz1(t), cz2 (t)) must
satisfy
d 2iz1 2 diz1 diz2
F - iz2 (t) cit cit = o
(7-7)
d2iz2
dt 2 + 1Z1(t) (d°'1)2
2 dt -
I (d°'2)2
2 (t) dt
1Z = o,
which results from Equation (7-6). Recall that
e' - e-, 2
tanh t = e' + e-' and sech t = e' + e-,
are the hyperbolic tangent and hyperbolic secant. Solutions of (7-7) are
of the following form:
(7-8) iz1(t) = a+ b tanh (rt) and iz2 (t) = b sech (rt)
or
(7-9) iz1(t) = c and iz2 (t) = de'',
where a, b, c, d,p, r E R.
Those geodesics which are in the form (7-9) are straight lines per-
pendicular to the x-axis. Those geodesics which are in the form (7-8)
are upper semicircles of radius Ib I with center at (a, 0) since
(iz1(t) - a) 2 + (1Z
2(t) - 0) 2 = b 2•
See Figure 7.6.

It is important to examine the hyperbolic upper half plane in regard to

Euclid's Fifth Postulate. His a space on which classical hyperbolic geometry
lies. Note that ifwe use line to mean geodesic and parallel lines to mean they
never meet (when infinitely extended), then given a line in H and a point not
232 Introduction to Manifolds Chap. 7

FIGURE 7.6

on the line there are infinitely many lines through the point which are parallel
to the given line.

PROBLEMS

*7.1. Prove Example 7.1 and find the geodesics.

*7.2. Prove Proposition 7.4.
7.3. Let Mn be a parallelizable manifold with parallelization
X,, ... , Xn E I(M).
(See Problem 4.6.) Show there is a linear connection V on M such
that each X 1 is parallel along any curve. (Hint: Analyze the example
of flat Euclidean space.)
7.4. Let M = R 2 and define V on R 2 by r 1/ = 0 except r 21 1 = r 1 / = 1.
Find all geodesics throughp = (a, b) E R 2 • (Note that the geodesics of
this example are not the same as the geodesics of flat R 2 .)
7.5. Use the connection of Proposition 6.7 to find all the geodesics on
S2 c R3.
*7.6. Let {Wi, W2 , ••• , Wk} be parallel vector fields along ot. Show that the
W, are linearly independent at each point of at if and only if they are
linearly independent at one point of ot.
7.7. Verify that Equations (7-8) and (7-9) do indeed give the solution of
Equation (7-7).

7-8. RIEMANNIAN METRICS, DISTANCE,

AND CURVATURE

In this final section we define a Riemannian metric and a Riemannian

manifold. We show that every Riemannian manifold has a natural linear con-
nection associated to it (called the Riemannian connection) so that the notion
of geodesic makes sense on a Riemannian manifold. We then put this natural
Sec. 7-8 Riemannian Metrics. Distance. and Curvature 233

connection on a Riemannian manifold and discuss the distance-minimizing

properties of geodesics. (Compare with Theorem 5.9 of Chapter 4.) We also
define several different kinds of curvature on a Riemannian manifold.
A Riemannian metric is defined as a differentiable assignment of an
inner product on T 11M for each p E M. More formally

DEFINITION. Afield of metrics g on a manifold Mis an assignment of a map

g11:T 11Mx T11M->Rtoeachp E MsuchthatforallX 11, Y11,Z 11 E T11M
and r ER:
(a) g11(X 11 + Y11, Z 11) = gp(X11, Z 11) + g (Y
11 11, Z 11) and
gp(rX 11, Y11) =rgp(X 11, Y11);
(b) gp(X11, Y11) = gp(Y11, X 11);and
(c) gp(X11, X 11) > 0 with gp(X 11, X 11) = 0 if and only if X 11 = 0.

Note that Conditions (a) and (b) imply that

gp(X11, Y11 +Z 11) = gp(X11, Y11) + gp(X 11, Z 11)
and
g11(X 11, rY 11) = rgp(X 11, Y11).
Thus a metric is an inner product on T11M because it is symmetric (b), bi-
linear (a), and positive definite (c). If g is a field of metrics on M and X,
Y E I(M), then g(X, Y) is a real-valued function on M, whose value at p is
gp(X11, Y11).

DEFINITION. A Riemannian metric on a manifold M is a field of metrics g

such that g(X, Y) E mM) for all X, YE I(M). A Riemannian manifold
is a manifold M together with a fixed Riemannian metric.

It can be shown that every manifold possesses a Riemannian metric (see

Hicks [19651).Note that in geometry the term metric does not refer to a dis-
tance function but rather to an inner product. We shall see below that there
is a natural distance function associated with a Riemannian metric.
As in Chapter 4, we may define n 2 real-valued functions g,1 on a coordi-
nate chart (U, ef,)by

EXAMPLE 8.1. Let N = Rn and let g =( , >(the usual inner product)

so that
gp(X 11, Y11) = :Ea1b1,
where X 11 = (p; a 1 , ••• , an) and Y11 = (p; b 1 , ••• , bn). Clearly g,iP) = ~11
(the Kronecker delta).
234 Introduction to Manifolds Chap. 7

ExAMPLE8.2. Assume that Mis a submanifold of a manifold N and that M

(resp. T,M) has been identified as a subset (resp. subspace) of N (resp.
T,N). If N has a Riemannian metric gN and X,, Y, e T,M c:: T,N, then
g, N(X,, Y,) makes sense and we may define g, M(X,, Y,) = g, N(X,, Y,).
This is exactly the way in which we made a surface in R 3 into a Rieman·
nian manifold.

DEFINITION. The Riemannian metric gM defined in Example 8.2 above is

called the Riemannian metric induced by gN (or the induced metric).

EXAMPLE 8.3. Let M = H be the hyperbolic upper half plane of Example 6.2.
If X,, Y, E T,H, then (X,, Y,) will denote the dot product of Example
8.1. We define, at the point p = (x, y) e H,

( 8-1) g, (X ,. y), = (X,,y2Y,).

g is a Riemannian metric on H. This example is what we were alluding

to in Problem 2.4 of Chapter 6. g is not an induced metric.

We are now in the pleasant situation of having an entire chapter for

motivation. We will examine some of the results of Chapter 4 and see which
carry over to the more general setting of Riemannian manifolds. Assume M
is a manifold with a fixed Riemannian metric.

DEFINITION.Suppose that A: T,M--+ T,M is a linear transformation. A is

an isometry if for all X,, Y, E T,M, g,(X,, Y,) = g,(AX,, A Y,).

DEFINITION. Let V be a linear connection on the Riemannian manifold M.

V is called metrical if for all X, Y, Z E I(M)
(8-2) Xg( Y, Z) = g(V x Y, Z) + g( Y, V xZ).
A word is in order about what Equation (8-2) means. g( Y, Z) is a function
so Xg( Y, Z) makes sense (and is a function). Since V x Y and V xZ E I(M),
the right-hand side is also a function. Intuitively, there are two ways to
view (8-2). The first comes by reading g( Y, Z) as " Y dot Z." From this
viewpoint Equation (8-2) reads exactly like the formula for the derivative of
a dot product where the derivative on the left is a directional derivative and
those on the right are covariant derivatives. The second way to view (8-2) is
geometrically via the following proposition.

PROPOSITION 8.4. Let V be a linear connection on a Riemannian manifold

M. V is a metrical connection if and only if for every curve IZ on M,
parallel translation along IZ is an isometry.
Sec. 7-8 Riemannian Metrics, Distance, and Curvature 235

Proof: First assume that V is metrical and let « be a curve in M. If

Y, Z E T,,,co>M,let Y and Z be their parallel translates along « so that
= P"'<t>(Y)
Y,,,<•> =P"'< 1>(Z). We must showthatg,,,<o>(Y,Z)= g,,,<r>(
andZ,,,<•> Y,Z)
for each t in the domain of«. We shall do this by showing the function
f (t) = g,,,c,>(Y, Z) is constant. By Equation (8-2) we have
'!t= T,,,(f) = T,,,g,,,ui(Y,
Z) = Z) + g,,,c,i(Y,VTZ) = 0
g,,,ct>CVTY, +0 = 0,

since both Y and Z are parallel along «. Hence f is constant.

Now assume P"' is an isometry for each curve«. We shall verify Equa-
tion (8-2). Let XP E TPM and let« be the unique geodesic such that «(O) = p
and T,,,coi= X;, as guaranteed by Theorem 7.7. Let {Jf't,U:\,... , Wn} be an
orthonormal basis of T;M and let W, be the parallel translate of W, along
«. The uniqueness of a parallel translate (Proposition 7.2) guarantees that the
W, are linearly independent at each point of « (see Problem 7.6). Hence
{W1 , ••• , Wn} is a basis at each point of«. By hypothesis,
(8-3)
and the W, give an orthonormal basis of the tangent space at each point of«.
Let Y, Z E l:(M). There are differentiable functions Y'(t) and Z'(t) such
= :EY1(t)(W 1),,,c,iand Z,,,c,i= :EZ 1(t)(W 1),,,c,i·By Equation (8-3),
that Y,,,c,>
g,,,<o(Y, Z) = :E Y1Z 1g,,,ct>CW,,
W1) = :E y 1z1t511= :E y1z1.
Then
(8-4) T,,,g(Y, Z) = :ET,,,(Y'Z1).
Since the W 1 are parallel along «,
VT y = :ET(Y')W, + Y'VTWI = :ET(Y 1)W,.
Thus
g(VTY, Z) + g(Y, vTz)= :ET(Y )Z 1g(w1, w,) + :EY'T(Z')g(W,,
1 w,)
= :ET(Y'Z 1)g(W 1, W1) = :ET(Y'Z').
If we combine this result with Equation (8-4) and evaluate at t = 0 (remem-
bering that T,,,101 = X;), we obtain
(8-5) X~( Y, Z) = g(V x. Y, Z) + g( Y, Vx.Z).
Hence V is metrical. I
DEFINITION. A linear connection V is torsion-free (or symmetric) if
VxY- VrX = [X, Y] for all X, YE I(M).

This condition is not as intuitive as the property of being metrical. In

particular, it has nothing to do with the torsion of a curve. The reason for the
236 Introduction to Manifolds Chap. 7

terminology "symmetric" is the following lemma, whose proof we leave as

Problem 8.2.

LEMMA
8.5. V is torsion-free if and only if in every coordinate chart,
r,/ = r,/ for all I ~ i,j, k ~ n.

THEOREM 8.6 (Fundamental Lemma of Riemannian Geometry). Let M be a

Riemannian manifold with a Riemannian metric g. Then there is a u-
nique torsion-free metrical linear connection V on M.

Proof: A straightforward calculation using the metrical and torsion-free con-

ditions shows that if V exists, then
(8-6) g(V xY, Z) = f(Xg(Y, Z) - Zg(X, Y) + Yg(Z, X) - g(X, [Y, Z])
+ g(Z, [X, Y]) - g(Y, [Z, X]))
for all X, Y, Z E I(M). Since the right-hand side is totally determined by
g, the value of Vx Y is determined. Hence V is unique if it exists. Equation
(8-6) is equivalent to the equation

(8_7> r IJk = _!_"'

2 ~
g'k(agu
ail _~go+
,I
ag")·
F,ii
where (g 1k) is the inverse of (g,1) (Lemma 8.5 has been used). This latter for-
mula should be familiar as the intrinsic formula for the r,/ (see Equation
(4-11) of Chapter 4) and is the actual motivation behind Equation (8-6).
On the other hand, Equation (8-6) may be used to define a connection
V: (Vx Y), is the unique element of T,M whose inner product with an arbi-
trary Z, is given by the right-hand side of (8-6) evaluated at p. A tedious
calculation shows that this does give a linear connection and that it is metrical
and torsion-free. I

In Problem 8.3 you will verify that (8-7) follows from (8-6). Note that
(8-7) can be used to find the Christoffel symbols for this connection.

DEFINITION.The linear connection V guaranteed by Theorem 8.6 is called

the Riemannian connection of the Riemannian manifold.

EXAMPLE 8.7. The Riemannian connection of Example 8.1 gives flat Eucli-
dean space (Example 6.1). The Riemannian connection of Example 8.2
is given in Problem 8.7. The Riemannian connection of the hyperbolic
upper half plane is given in Example 6.2 as may be seen by computing
the r,/ via Equation (8-7).

We now put a notion of distance on a Riemannian manifold and discuss

the length-minimizing properties of geodesics (with respect to the Rieman-
7-8 RiemannianMetrics,Distance,and Curvature 237

nian connection). Again we fix a manifold M and a Riemannian metric g

and let V be the Riemannian connection. We shall assume that Mis path-
connected (i.e., if p, q e M, then there is a curve 11: [a, bJ ----+M such that
11(a)= p and 11(b)= q). We shall not prove any of the results in the rest of
this section but merely give references.

DEFINITION. If 11:[a, b]----+M then the length 0/11 is

l«I = f ,,Jg.c,,(T.c,,,T..c,>)
dt.
If p, q e M, the distance from p to q is d(p, q) = inf III I, where the in-
fimum is taken over all broken C curves joining p to q.
00

PROPOSITION 8.8. The Riemannian manifold M with the distance defined

above is a metric space.

The symmetry and the triangle inequality for dare very easy. The hard
part of Proposition 8.8 is showing that d(p, q) = 0 implies p = q. This
requires Theorem 8.11 below. Note that we have assumed that Mis a metric
space to start with. Is the distance defined by g the same as the original
metric? The answer is an emphatic no. The upper half plane is a metric space
(as a subset of R 2). The metric space structure of H coming from a Rieman-
nian metric is entirely different (see Problem 8.8). However, we can show the
topologies are the same. (If you don't know what a topology is, then skip
the next result.)

PROPOSITION 8.9. The topology of a Riemannian manifold M as a metric

space is the same as the topology induced by the metric space structure
coming from the Riemannian metric g (Hicks [1965, p. 70]).

Because we are assuming that Mis a Riemannian manifold it makes

sense to talk about a parametrization by arc length.

DEFINITION. The curve II is parametrized by arc length if g(T., T.) = I.

We see that the analogue of Theorem 5.9 of Chapter 4 is true even in

this very general setting. A reference for the next four results is Milnor
[1963, pp. 55-67] or Hicks [1965, Chapter 9].

THEOREM 8.10. Suppose 11:[a, bJ----+Mis a curve parametrized by arc length.

IPI for all broken C curves P such that P(a) = 11(a) and
If 1111::;;; 00

P(b) = 11(b),then II is a geodesic.

This says that any curve which minimizes distance is a geodesic. The
converse is false as S 2 with the induced metric shows. (Great circles are
238 Introduction to Manifolds Chap, 7

geodesics, as usual.) However, locally things are better (compare with

Theorem 5.12 of Chapter 4).

THEOREM 8.11 (J. H. C. Whitehead, 1932). For every point pin a Rieman-
nian manifold M there is a neighborhood U such that
(a) any two points q and r in U may be joined by a geodesic czwhose
image lies in U;
(b) the geodesic czis the unique geodesic joining q to r which has length
d(q, r); and
(c) there is a local coordinate chart (U, 4,) such that the geodesics
through p take the form cz(t)= 4,-1(a 1t, ... , a"t) for some constants
a 1, ••• ,a• ER.
As mentioned in Chapter 4 it need not be possible to join any two points
by a geodesic but:

THEOREM 8.12 (Hilbert, 1902). If a connected Riemannian manifold M is

complete as a metric space with the metric induced by g, then any two
points can be joined by a geodesic which minimizes distance between the
points.

Hilbert's Theorem has a partial converse:

THEOREM 8.13 (Hopf-Rinow, 1931). If every geodesic on M may be extended

indefinitely (i.e., is defined on all of R), then any two points may be
joined by a geodesic of minimal length and M is complete as a metric
space under the distance determined by g.

We now turn to the notion of curvature. It is not even clear what the
definition of curvature should be on a Riemannian manifold. We cannot use
the Weingarten map L because there is not one on an arbitrary Riemannian
manifold (there is no concept of a normal vector). It is Gauss's Theorema
Egregium that tells us what to do. Indeed, that is the whole point of the
Theorema Egregium.

DEFINITION. The Riemann-Christoffel curvature tensor of type (l, 3) is the

map R : l(M) x l:(M) x l:(M) --+ l:(M) by
R(X, Y)Z = VxVyZ- VyVxZ - Vcx,l'1Z.

In local coordinates we have

R (-k•-b)b = '£ R,',k k·

where R,11,. is defined by Equation (9-1) of Chapter 4 (see Problem 8.9). Note
Sec. 7-8 Riemannian Metrics, Distance, and Curvature 239

that if X = 8/8x' and Y = 8/8x1, then R(X, Y)Z measures how much the
"equality of mixed partials" fails to hold for covariant derivatives.
A simple calculation shows that R = 0 in flat Euclidean space, which
intuitively does not curve. On the other hand, R is nonzero in S" (as we have
actually computed in the case of S 2 in Chapter 4). As a hypersurface in R•+ 1,
S" does have an intuitive notion of curvature. Thus in some sense R does mea-
sure curvature, or lack of flatness. (It can be shown that a manifold with
R = 0 cannot be distinguished locally from R• geometrically-about each
point there can be found a coordinate chart in which g,1 is identically equal
to t51J")However, we would prefer to have a numerical invariant for curva-
ture. There are several ways to obtain a numerical invariant from the Rie-
mann-Christoffel tensor. We shall give the definition of three: sectional,
Ricci, and scalar curvatures. Each is defined in terms of the Riemann-
Christoffel curvature tensor of type (l, 3) or type (0, 4).

DEFINmoN. The Riemann-Christoffel tensor of type (0, 4) is the map R:

l(M) x ~(M) x ~(M) x ~(M) -- ij(M) given by
R(X, Y, Z, W) = g(R(X, Y)Z, W).

A straightforward calculation shows that the values of R(X, Y)Z and

R(X, Y, Z, W) at p e M depend only on the values of X,, Y,, Z,, and W,
and not on the vector fields themselves.
In surface theory we studied curvature by breaking the surface up into
curves. Analogously, we can try to break a manifold up into two-dimen-
sional submanifolds, or we may look at two-dimensional subspaces of T,M.

DEFINmON. Let n be a two-dimensional subspace of T,M. The sectional

curvature of n is K,(Il) = R(X, Y, Y, X)(p), where {X,, Y,} is an ortho-
normal basis of n.

For surfaces in R 3 , K,(Il) is exactly the same as the Gaussian curvature

atp.
The other two kinds of curvature, the Ricci and scalar curvatures, are
quite frequently used in physics (especially relativity) as well as mathematics.
If X,, Y, e T,M, we may define a map S:,(X,, Y,): T,M -- T,M by
S,(X,, Y,)V, = R(V,,X,)Y,. Problem 8.10 shows that S:,(X,, Y,) is a linear
transformation for eachp and X,, Y, e T,M and so we may take the trace
of S,(X,, Y,).

DEFINmON. The Ricci curvature tensor S is an assignment, to each p e M,

of a function S,: T,M x T,M -- R defined by
S,(X,, Y,) = trace(S:,(X,, Y,)).
240 Introduction to Manifolds Chap, 7

DEFINITION. If Mn is a Riemannian manifold with Ricci tensor S, then the

scalar curvatureof Mat p is ~
• S,((X ),, (X,),), where {(X ),, ••• , (X.),}
1 1
t:1
is any orthonormal basis of T,M.

It should be pointed out that the applications of the above to relativity

theory is not Riemannian geometry. In relativity theory one does not have a
Riemannian metric but rather a nondegeneratemetric; Condition (c) of the
definition is replaced by
(c') g(X,, Y,) = 0 for all Y, e T,M if and only if X, = 0.
This lack of positive definiteness does create certain problems in translating
Riemannian geometry into relativity theory.
One area of research in differential geometry revolves around the effect
of assumptions about curvature on the topology of the manifold. The results
in this area are fascinating (and quite difficult). Most of them require more
topology than we can assume, so we will only mention a few of them and
refer the interested reader to Kobyashi and Nomizu [1963, especially p. 294).
Readers who have gone through Chapter 6 have seen some results of this
type.
There are many very simple sounding questions which are extremely
difficult to answer. F<;>rexample, there are very few examples known of com-
pact manifolds with every where positive sectional curvature. In fact, it is not
even known what the spaces with constant positive sectional curvature l/a 2
are. However, we do have the following.

TIIEoREM 8.14. Every connected complete Riemannian manifold M of even

dimension which has constant positive sectional curvature l/a 2 is either
a sphere of radius a or a suitable projective space.

If we are willing to assume that Mis simply connected, then there is the
following classification.

TuEoREM 8.15. If Mis a complete, connected, simply connected Riemannian

manifold of constant sectional curvature c, then M is either
(a) a sphere of radius ,./1/c (c > O);
(b) flat Euclidean space (c = O); or
(c) hyperbolic space form (see Problem 8.13) (c < 0).

x
In Section 6-5 we saw that the Euler characteristic was (1/2n) times the
integral of the Gaussian curvature so that K > 0 implies X > 0. There is an
analogous formula in higher even dimensions but the following conjecture,
with the appropriate meaning for Euler characteristic, has remained un-
solved for over fifty years:
Sec. 7-B Riemann/an Metrics, Distance, and Curvature 241

CONJECTURE 8.16. If Mis a compact, orientable, even-dimensional manifold

with everywhere positive sectional curvature, then M has positive Euler
characteristic.

PROBLEMS

8.1. If g 0 and g 1 are Riemannian metrics on M, prove that tg 0 + sg 1 is

also a Riemannian metric on M if t and s are both positive.
*8.2. Prove Lemma 8.5.
*8.3. Prove that Equation (8-7) follows from Equation (8-6).
*8.4. Prove that the Riemannian connection of Example 8.1 is flat Eucli-
dean space. Verify that (VxY, N) = -(Y, VxN) as claimed at the
end of Section 7-6.
*8.5. Prove that the Riemannian connection of the hyperbolic upper half
plane (Example 8.3) is the linear connection given in Example 6.2.
*8.6. Prove Proposition 8.8 assuming Theorem 8.11.
t8.7. (Riemannian Connection for Submanifolds). Let Mbe a submanifold
of N, Mc N and write g for the Riemannian metric of N (and~
for the Riemannian connection of N). If p E M, then the normal
space to M at p is
v, = {X, E T,Njg,(X,, Y,) = 0 for all Y, E T,M}.
Since T,N = T,M(fJ v, for each p E M, we may talk about the
tangential and normal components of an element of T ,N. If
X, Y E l:(M), then V(X, Y) is (by definition) the normal component
of Vx Y. (Vis called the second fundamental tensor of Min N.) Define
VxYby:
(8-8) VxY= VxY- V(X, Y).
(This is known as Gauss's formula.)
(a) Show that V(X, Y) = V(Y, X) and V(JX, Y) =fV(X, Y) for
any IE ff(M).
(b) Show that V is a linear connection on M.
(c) Show that V is the Riemannian connection for the induced
metric on M.
(d) If M = M 1 is the hypersurface of N = R•+ 1 defined by f, show
that Equation (8-8) reduces to Equation (6-3).
*8.8. Let O < E < 1. Show that the distance between p = (0, I) and
qe = (0, E) is (1/E) - 1 in the hyperbolic upper half plane. (Hint:
Consider the curve in H, <1(t)= (O, l - t) for O < t < 1 - E.) Note
that the Euclidean distance between p and qe is 1 - E (hence stays
242 Introduction to Manifolds Chap. 7

finite as E----,. 0) whereas the "origin" is infinitely far away from

(0, 1) in the hyperbolic distance.
8.9. Show that, in local coordinates,

R (a:" a:k)
a:,=~ {a;;:' b
- 8{;l+ ~ cr,k'r,/- r11,r,k')}
for all I :::,;;:
i,j, k < n. Compare with Equation (9-1) of Chapter 4.
*8.10. Show that E,(X,, Y,): T,M----,. T,M is a linear transformation for
eachp EM.
8.11. Compute the sectional, Ricci, and scalar curvatures of Rn and sn.
8.12. Let M2 be a surface in R with Gaussian curvature K(p) and mean
3

curvature H(p) at p. Express the sectional, Ricci, and scalar curva-

ture of Mat p in terms of Kand H. Are you surprised at the answer?
8.13. (Hyperbolic Space Form.) Let Hn= {x E Rnl lxl < l}. Define the
Riemannian metric on Hn at x E Hn by
2c511
g,,(x ) = (1 4a
- Ix 12)2
We are taking the chart on Hn given by the identity map.
(a) Find the Riemannian connection of Hn. (Answer: r,/ is z.ero
unless at least two of the indices i, j, k are equal, and then
r/ = r,/ = -ri/ = ru1 = 2x 1/(l - lxl 2).
(b) Show that the sectional curvature of any plane at any point is
-1/a 2 • (Hint: Show R/k, = 0 unless k = i and r =j, or k =j
and r = i.)
Comment: H 2 is not the hyperbolic upper half plane. It is the
Poincare disk. We choose H as the model for hyperbolic geometry
rather than H 2 because the computation and visualization of
geodesics on His easier. It can be shown that H 2 and Hare the same
in the sense that they are isometric.
tt8.14. Prove Conjecture 8.16 and win a priz.e.
 Appendix

Historical Notes

The history of mathematics, in addition to being a fascinating field in its

own right, is a very valuable aid in learning a new field of mathematics. For
this reason we have included these historical notes. They are not meant to
be a history of differential geometry or even a history of the results contained
in this book. Such a history itself would be at least as long as this volume. For
a more complete history see Stroik [1933), [1967), or Coolidge [1940). There is
also an interesting chapter in Boyer [1968).
Problems in the differential geometry of plane curves have been studied
since the invention of calculus, but we shall not concern ourselves with this
aspect of the history. (One amusing fact: some historians date the beginning
of differential geometry before the invention of the calculus!) The first major
contributor to the subject was Leonhard Euler (1707-1783). In 1736, he intro-
duced the intrinsic coordinates of a plane curve (arc lengths and the radius
of curvature p instead of the curvature k), and so began the study of intrinsic
geometry. You may recall that the description of curvature as the rate of
change of a particular angle was due to Euler. The original theory of curves of
constant width is due to him. He also contributed to the theory of surfaces,
especially (with John and Daniel Bernoulli (1667-1748) and (1700-1782))
some work on geodesics, and was the first to describe geodesics as the solu-
tions to certain differential equations. Motivated by physical problems, he

243
 Historical Notes Appendix

showed, in 1736, that a point mass constrained to lie on a surface and sub-
ject to no other forces must move along a geodesic.
A second major figure in the history of differential geometry was
Gaspard Monge (1746--1818).He also was motivated by practical problems-
in his case, questions of fortification. Monge started the theory of space
curves in a paper he wrote in 1771 and published in 1785. His methods were
very geometric and reflected his interest in partial differential equations. He
published the first text on differential geometry in 1807. An indication of the
importance of this book is the appearance of a fifth edition, thirty years
after his death. Monge is remembered not only for his original contribu-
tions but also for the results of his teaching. Among his pupils were Pierre
Laplace (1749-1827), Jean Meusnier (1754-1793), Joseph Fourier (1768-
1830), Michel Lancret (1774-1807), Andre Ampere (1775-1836), Etienne
Malus (1775-1812), Simeon Poisson (1781-1840), Charles Dupin (1784-
1873),Victor Poncelet (1788-1867), and Olin de Rodrigues ( 1794-1851). Today
the work of the Monge school is hard to read be.cause they thought in
terms of infinitesimals. Although Euler had introduced analytical methods to
the study of space curves just before his death, these were not even used in
Monge's text twenty-five years later. As an example of this awkward language,
we note that Lancret defined the "premiere flexion" (curvature) and "seconde
flexion" (torsion) as the differential of the angles between two consecutive
normal or osculating planes. It was Augustin Cauchy (1789-1857), one of
the founders of group theory and the theory of limits, who first expressed
curvature and torsion in terms of finite quantities as we do today.
Even in the 1840s the theory of space curves was inelegant and difficult
to find. To ameliorate this situation Barre de Saint Venant (1797-1886)
wrote in 1846 a study on space curves. In this study, he included much his-
torical material and collected the available results together in one work. He
is responsible for the term "binormal" and the first proof of Lancret's
Theorem. The theory of space curves finally became unified when F. Frenet
(1816-1868) and Joseph Serret (1819-1885) working independently came up
with what we call the Frenet-Serret equations in 1847 and 1851. Unfor-
tunately, their work was not highly regarded when it appeared. Patt of this
was due to the lack of the language of linear algebra, which we have used
extensively. For example, instead of computing the derivative of the normal
with respect to arc length, they computed the derivatives of the direction
cosines of the line in the direction of the normal. It was Gaston Darboux
(1824-1917) who first unified the theory of curves with his concept of a
moving frame ("triedre mobile"). In this he was motivated by the theory of
mechanics. This is the modern theory which in turn, under the guidance of
Elie Cartan (1869-1951), gave valuable insight into the theory of manifolds.
The contributions of Carl Friedrich Gauss (1777-1855) can be found in
his Disquisitionesgenera/escirca superficiescurvas of 1827, which has been
Appendix Historical Notes

translated into English (Gauss [1965]). Once you overcome the differences in
mathematical language, this translation is not too difficult to read, especially
with the aid of the section "How to Read Gauss" in Spivak [1970, vol. II].
Stroik [1933, p. 164] notes that Gauss "became the teacher of the entire
learned world." The reason for this goes beyond the fact that he proved new
and startling results. He was responsible for an entirely new approach to dif-
ferential geometry, one that has proved fruitful for over 150 years: intrinsic
geometry based upon the first fundamental form. The idea that a surface could
be written parametrically as a function of two variables, at least locally, was
known to Euler, but Gauss emphasized that a surface should be described
this way, rather than as a set of points in R 3 whose coordinates satisfy some
relation. The sphere (or Gauss) map was also known to Euler. Gauss's ap-
proach made heavy use of it. In fact, it is the first concept that he defines in
his book. Rodrigues had earlier found the limit of the ratio between the area
of a surface and the area of the corresponding region on the sphere (Proposi-
tion 8.6 of Chapter 4). Gauss actually was the first to recognize the impor-
tance of this limit and used it as the definition of the curvature of a surface
at a point. This is a second important aspect of Gauss's contributions: his
insight into what was useful and important.
One can say that geometers before Gauss viewed a surface as being made
up of infinitely many curves, whereas Gauss viewed the surface as an entity
in itself. To better understand this, think about computing the curvature of a
surface in two ways. One way would be extrinsic: find the principal directions
and then compute the product of the normal curvatures of the corresponding
lines of curvature, an approach essentially known to Euler. The second way
would be intrinsic: use the Theorema Egregium. The contrast between the
extrinsic and the intrinsic calculations of curvature is a very deep philo-
sophical one, of which Gauss was well aware. His statement of the Theorema
Egregium(Gauss [1965, p. 201) illustrates this, along with his interest in the
surface as a whole: "If a curved surface is developed upon any other surface
whatever, the measure of curvature in each point remains unchanged."
(Here "developed" means mapped in a one-to-one, onto, distance-preserving
fashion.) Although Gauss was making tremendous steps forward with his
intrinsic geometry, it was not completely recognized at the time. While he was
advocating the study of the Gaussian curvature because it was invariant
under a development, a mathematician as good as Sophie Germain (1776-
1831) was advocating the mean curvature as the main object of study. (This
is understandable since it came up constantly in her study of elasticity.)
If there are two key ideas which have permeated the history of geometry
from 1827 to the present, they are (I) the continuing influence of physical
problems in directing various avenues of research and (2) the desire to both
understand and enlarge upon the work of Gauss. These two ideas are not
unrelated. Indeed, Gauss was surveying the Kingdom ·of Hanover from 1821
248 Hlstoric::alNotes Appendix

to 1825 and this practical work combined with his work as a theoretical
geodesist motivated much of his work in differential geometry.
We shall not go into the applications of differential geometry to physics,
mechanics, and the like except to mention those most responsible. Gabriel
Lame (1795-1871) worked in the theory of elasticity and heat, as did de Saint
Venant. Ampere and Lazare Carnot (1753-1823) were guided by their
interest in electricity. Many individuals, including Alexis Clairaut (1713-
1765), Monge, Ampere, and Henri Poincare (1854-1912), studied differential
geometry to gain insight into the theory of partial differential equations.
Malus came to differential geometry from optics (line congruences). Dupin
was interested in the applications of differential geometry to mechanics.
Albert Einstein (1879-1955) based his theory of general relativity on dif-
ferential geometry in 1915. There are still applications of differential geometry
being made in new areas. See Newsweek, January 19, 1976, pp. 54-55 for a
nontechnical account of Rene Thom's catastrophe theory. (It is actually more
topology than geometry.) Catastrophe theory is also covered in Zeeman
[1976].
The idea of the surface itself being an important entity was taken up
again in the brilliant address of Bernard Riemann (1826-1866) in 1854. A
translation of this may be found in Spivak [1970, vol. II], which we urge the
reader to try to read. We have already discussed a part of Riemann's philoso-
phy in the introductions to Chapters 4 and 7. We mentioned the philosophical
importance of his realization that space and geometry are different. Of equal
importance is his realization that a "quadratic differential" (which we now
call the Riemannian metric) is the structure to add to the notion of manifold.
This gives an infinitesimal way to measure distance and an enormous amount
of structure on manifolds. To appreciate Riemann's contribution in this area
we must realize that geometers had been putting a Riemannian metric on
surfaces in R 3 (the induced metric) without realizing that it was an extra
structure. Riemann realized that this notion was extraordinarily important
and separated it out. There are now more general notions of geometry (e.g.,
linear connections or connections on a fiber bundle), but this does not
diminish Riemann's contribution.
In the period between 1854 and 1900 geometers were mostly obsessed
with the language of differential geometry rather than the subject itself. This
tendency for symbolism culminated with the development of the tensor
calculus by Gregorio Ricci (1853-1925) and his student Tullio Levi-Civita
(1873-1941). It was upon Riemann's abstract view of s~ce and this tensor
calculus that Einstein based his theory of gravitation, commonly called gen-
eral relativity. One of the important historical developments during this pe-
riod was the use of group theory in differential geometry. This unifying
concept was introduced by Felix Klein (1849-1925) and Sophus Lie (1842-
1899). It was further developed by Cartan in a plethora of fascinating papers.
Appendix Historical Notes 247

Unfortunately, much of Cartan's work is difficult to read because of the lack

of a suitable theory of topology at the time he wrote. For a modern interpre-
tation of Klein's ideas see Millman [1977).
For those readers who are interested in what has happened more recently,
a survey by Chern (1946]is a good source. Since then there has been a move-
ment to change the notation once again and get away from the local tensor
notation of Ricci and Levi-Civita. This new invariant notation, which we
used to some degree in Chapter 7, along with the definition of a linear con-
nection due to Koszul, first appeared as recently as 1954 in an important
paper by Katsumi Nomizu.
We conclude these notes and this book with a quote from Gauss (1965,
p. 45) which is as valid now as it was in 1827:
Although geometers have given much attention to general investigations of
curved surfaces and their results cover a significant portion of the domain of
higher geometry, this subject is still so far from being exhausted, that it can be
well said that, up to this time, but a small portion of an exceedingly fruitful
field hasbeen cultivated.
 Bibliography

In addition to those books and papers referred to in the text, we have

included several other related books. Some words are in order relative to the
level and difficulty of these books. Struik [1961], Laugwitz [1965], O'Neill
[1966], Stoker [1969], Goetz [1970], and Do Carmo [1976] are all under-
graduate texts. Struik and Laugwitz are classical in outlook while O'Neill is
quite modern. Stoker, Goetz, Do Carmo, and O'Neill are at about the same
level as this text, but Stoker and Do Carmo assume more analytical and
topological background and O'Neill uses the language of differential forms.
At the crossover to graduate level, we have Spivak [1970, 1975].An introduc-
tion to differential geometry at the graduate level may be found in Hicks
[1965]or Boothby [1975].At a more advanced level is Bishop and Crittenden
[1964], and finally Kobayashi and Nomizu [1963, 1969].
The books by Matsushima [1972] and Warner [1971] are about dif-
ferentiable manifolds, not geometry (that is, the concept of a connection does
not appear). They cover material basic to both differential geometry and dif-
ferential topology. For the latter subject both Guillemin and Pollack [1974]
and Milnor [1965] are excellent and readable.
Any advanced study of differential geometry requires some background
in point set topology, as may be found in Munkres [1975], and in algebraic

248
Bibliography 249

topology, some of which can be found in Singer and Thorpe [1967] and
Massey [1967).
ALMGREN, F., Plateau's Problem. New York: Benjamin, 1966.
BERGER, M., Lectures on Geodesics in Riemannian Geometry, Tata Institute on Fundamental
Research Lectures in Mathematics, No. 33. Bombay: Tata Institute, 1965.
BERTRAND, J., "La theorie des courbes a double courbure," Journal de Mathematiques Pures
et Appliquees, 15 (1850), 332-350.
BIRKHOFF, G., and G. ROTA, Ordinary Differential Equations (2nd ed.). Waltham, Mass.:
Blaisdell, 1969.
BISHOP, R.,"There is More than One Way to Frame a Curve," American Mathematical
Monthly, 82 (1975), 246-251.
---, and R. CRITTENDEN, Geometry of Manifolds. New York: Academic Press, 1964.
BLASCHKE, w., Vorlesungen iiber Differentialgeometrie, I (3rd ed.). Berlin: Springer, 1930.
BOOTHBY,w., An Introduction to Differentiable Manifolds and Riemannian Geometry. New
York: Academic Press, 1975.
BORSUK, K.,"Sur la courbure totale des courbes fermees," Anna/es de la Societe Polonaise
de Mathematique, 20 (1948), 251-265.
BOYER, c., A History of Mathematics. New York: Wiley, 1968.
BREUER, s., and D. GOTTLIEB, "Explicit Characterizations of Spherical Curves," Proceedings
of the American Mathematical Society, 27 (1971), 126-127.
CHERN, s., "Some New Viewpoints in Differential Geometry in the Large," Bulletin of the
American Mathematical Society, 52 (1946), 1-30.
--, "Curves and Surfaces in Euclidean Space," in Studies in Global Geometry and
Analysis, ed. S. Chern, 16-56. Englewood Oiffs, N.J.: Mathematical Association of America
and Prentice-Hall, 1967.
CODDINGTON, E., and N. LEVINSON, Ordinary Differential Equations. New York: McGraw-
Hill, 1955.
COOLIDGE, J., A History of Geometrical Methods. Oxford, England: Oxford University
Press, 1940.
DO CARMO, M., Differential Geometry of Curves and Surfaces. Englewood Oiffs, N.J.:
Prentice-Hall, 1976.
FARY, 1., "Sur la courbure totale d'une courbe gauche faisant un noeud," Bulletin de la
Societe Mathematique de France, 77 (1949), 128-138.
FENCHEL, w., "Uber Kriinunung and Windung geschlossenen Raumkurven," Mathematische
Annalen, 101 (1929), 238-252.
---, "On the Differential Geometry of Global Space Curves," Bulletin of the American
Mathematical Society, 57 (1951), 44-54.
FULKS, w., Advanced Calculus (2nd ed.). New York: Wiley, 1969.
GAUSS, K.,General Investigations of Curved Surfaces, tr. A. Hiltebertel and J. Morehead.
Hewlett, N.Y.: Raven Press, 1965.
GLUCK, H., "Higher Curvatures of Curves in Euclidean Space, I, II," American Mathe-
matical Monthly, 73 (1966), 699-704; 74 (1967), 1049-1056.
 260 Bibliography

OOETZ, A., Introductionto DifferentialGeometry. Reading, Mass.: Addison-Wesley,1970.

GUJLLEMJN, v., and A. POLLACK, Differential Topology.Englewood Qiffs, N.J.: Prentice-
Hall, 1974.
HALL, T., Carl FriedrichGauss, A Biography,tr. A. Froderberg. Cambridge, Mass.: MIT
Press, 1970.
HICKS, N., Notes on DifferentialGeometry. Princeton: van Nostrand, 1965.
HILBERT, D., The Foundationsof Geometry (2nd ed.), tr. E. Townsend. Chicago: Open
Court, 1921.
--, Grundlagender Geometrie(8th ed.). Stuttgart: Teubner, 1956.
---, and s. COHN·Vos.mN,
Geometry and the Imagination,tr. P. Nemenyi. New York:
Chelsea, 1952.
HOFFMAN, K., and R. KUNZE,
Linear Algebra (2nd ed.). Englewood Qiffs, N.J.: Prentice-
Hall, 1971.
HOPF,H., "Uber die Drehung der Tangenten und Sehnen ebener Kurven," Compositio
Mathematica, 2 (1935), S0-62.
JACKSON, s., "Vertices of Plane Curves," Bulletin of the American Mathematical Society,
50 (1944), 564-578.
KNESBR, A., "Bemerkungeniiber die Anzahl de Extreme der Kriimmung auf geschlossenen
Kurven und iiber verwandte Fragen einer nicht-EuklidischenGeometric," Fest,,J'chrift
zum
siebzigstenGeburstagvon Heinrich Weber, 170-180.1..eipzig:Teubner, 1912.
KOBAYASHI, s., and K. NOMIZU,
Foundationsof Differential Geometry, I, H. New York:
lnterscience, 1963, 1969.
LAUGWITZ, D., Differentialand Riemann/an Geometry. New York: Academic Press, 1965.
MASSEY, w., AlgebraicTopology:An Introduction.New York: Harcourt, Brace and World,
1967.
MATSUSHIMA,
Y., DifferentiableManifolds. New York: Marcel Dekker, 1972.
• MILLMAN, R., "KleinianTransformation Geometry," American Mathematical Mo,,thly,
84 (1977). I

--, and A.STEHNEY,

"The Geometry of Connections,"AmericanMathematicalMonthly,
80 (1973),475-500.
MILNOR,
J., "On the Total Curvature of Knots," Annals of Mathematics,52 (1950),248-257.

---, Morse Theory, Annals of Mathematics Studies No. 51. Princeton: Princeton
University Press, 1963.
--, Topologyfrom the DifferentiableViewpoint.Charlottesville,Va.: University Press
of Virginia, 1965.
MISNER, c., K. THORNE,
and J. WHEELER, Gravitation.SanFrancisco: Freeman, 1973. /
MUKHOPADHYAYA, s., "New Methods in the Geometry of a Plane Arc," Bulletinof the Cul-
cutta Mathematical Society, I (1909), 31-37.
MUNKRES, J., Topology,A First Course.Englewood Qiffs, N.J.: Prentice-Hall, 1975.
o'NEILL,e., Elementary DifferentialGeometry.New York: AcadeinicPress, 1966.
RUTISHAUSER,
H., and u. SAM.ELSON,
"Sur le rayon d'une sphere dont la surface contient une
 Bibliography 261

courbe fermee," Comptes Rendus Hebdomadaires des Seances de l'Academie de Science,

227 (1948), 755-757.
SALKOWSKI, F., "Zur Transformation von Raumkurven," Mathematische Annalen, 66 (1909),
517-557.
SAMELSON, H., "Orientability of Hypersurfaces in R"," Proceedings of the American Mathe-
matical Society, 22 (1969), 301-302.
SCHERRER, w., "Eine Kennzeichnung der Kugel," Vierteljahresschrift Naturforscher Gesell-
schaft in Zurich, 85 (1940), 40-46.
SCHMIDT, E., "Uber das isoperimetrische Problem in Raum von n Dimensionen," Mathe-
matische Zeitschrift, 44 (1939), 689-788.
SEGRE, s., "Sui circoli geodetici di una superficie a curvature totale contante, che contengono
nell'interno una linea assegnata," Bollettino de/la Unione Matematica ltaliana, 13 (1934),
279-283.
---, "Una nuova caratterizzazione della sfera," Atti de/la Accademia Nazionale dei
Lincei, Rendiconti, 3 (1947), 420-422.
---, "Sulla torsione integrate delle curve chiuse sghembe," Atti del/a Accademia
Nazionale dei Lincei, Rendiconti, 3 (1947), 422-426.
SINGER, 1., and J. THORPE, Lecture Notes on Elementary Topology and Geometry. Glenview,
Ill.: Scott, Foresman, 1967.
SPIVAK, M.,A Comprehensive Introduction to Differential Geometry, I, II, Ill, IV, V. Boston:
Publish or Perish, 1970, 1975.
STERNBERG, s., Lectures on Differential Geometry. Englewood Qiffs, N.J.: Prentice-Hall,
1964.
STOKER, J., Differential Geometry. New York: Interscience, 1969.
/STRUIK, o., "Outline of a History of Differential Geometry," Isis, 19 (1933), 92-120; 20
(1933), 161-191.
---, Lectures on Classical Differential Geometry (2nd ed.). Reading, Mass.: Addison-
Wesley, 1961.
--, A Concise History of Mathematics (3rd ed.). New York: Dover, 1967.
THOMAS, G., Calculus and Analytic Geometry (4th ed.). Reading, Mass.: Addison-Wesley,
1968.
voss, A., "Uber Kurvenpaare in Raume," Sitzungsberichte, Akadamie der Wissenschaften
zu Munchen, 39 (1909), 106.
voss, K., "Eine Bemerkung iiber die Totalkriimmung geschlossenen Raumkurven," Archiv
der Mathematik, 6 (1955), 259-263.
WARNER, F.,Foundations of Differentiable Manifolds and Lie Groups. Glenview, Ill.: Scott,
Foresman, 1971.
WONG, Y., "A Global Formulation of the Condition for a Curve to Lie in a Sphere,''
Monatschefte Fur Mathematik, 67 (1963), 363-365.
---, "On an Explicit Characterization of Spherical Curves," Proceedings of the American
Mathematical Society, 34 (1972), 239-242.
ZEEMAN, E., "Catastrophe Theory," Scientific American, 234 (1976), 65-83.
 I.

Notational Index

B (binormal ofa regular <I>* (differential of a ma~

curve),26 ping),219

g (determinant of metric
cosh(t) (hyperbolic cosine), 12 coefficients), 95
g,, (metric coefficients), 93,
223
dA (area element), 130 gk' (inverse of metric co-
du; (dual of a basic tangent efficients), 95
vector), 100 ruk (Christoffel symbol), 104
v (connection), 224
VxY (covariant derivative),
116 (5.14), 224 H (hyperbolic half plane),
c5voc (angularvariation), 182 225
c5ii (Kronecker symbol), 3 H (meancurvature), 130
a;ax' (basic tangent vector),
212
11 (second fundamental
form), 123
«J(M) (set of differentiable func- /(V) ( total index of a vector
tions on a manifold), 209 field), 196

266
258 Index

i,. (rotation index of a R/ik (coefficients of the

curve), 55 Riemannian curvature
ip(V) (index of a vector field), tensor), 141,242(8.9)
195 R(X,Y)Z (Riemann-Christoffel
curvature tensor), 238
p (radius of curvature), 34
J(f) (Jacobian of/), 79

s (Ricci curvature), 239

K (Gaussian curvature), s (intrinsic normal of a
130 surface curve), 102
Kp(II) (sectional curvature), 239
si (unit circle), 58
k (curvature ofa plane
s2 (unit sphere), 87 (1.10)
curve), 52 S" (n-sphere), 203
K (curvature of a regular sech(t) (hyperbolic secant), 231
curve),24 sinh(t) (hyperbolic sine), 12
K1,K2 (principal curvatures), (1 (radius of torsion), 34
129
Kii (geodesic curvature), 103
Kn (normal curvature), 103 T (unit tangent of a regular
curve), 15
T« ( tangent ofa manifold
curve),225
L (Weingarten map), 125
TPM (tangent space), 93, 213
Lu (coefficients of the second T2 (torus), 92 (2.2)
fundamentalform),104 T" (n-torus), 208 (2.3)
.I!k (coefficients of the tanh(t) (hyperbolic tangent), 231
Weingarten map), 125
t (unittangent ofa P!ane
curve), 52
't (torsion of a regular
N (principal normal of a curve),26
regular curve), 26
n (unit normal of a plane
curve),52 x«p (tangent ofa manifold
n (unit normal of a sur- curve),211
face), 81 X(M) (set of differentiable
vector fields on a mani-
fold), 216
p« (parallel translation X; (basic tangent vector), 81
along a curve), 229 Xu (derivative of a basic
pn (projective n-space), 200 tangent vector), 102
 I I.

Topical Index

Angle between vectors, 3, 181 Basis,2

Angular excess, 187 orthonormal, 3
Angular variation, 182 Beltrami-Enneper Theorem, 139
Angular velocity vector, 35 (4.1 l) (8.35)
Antipodal points, 162 Bertrand curves, 38, 41 (4.41), 48
Arc length, 20 (6.12)
reparametrization by, 21 characterization, 38 (4.3 l)
of a surface curve, 95 Bertrand mate, 38
Area, 130, 135 ofa circular helix, 46 ( 5.6)
Area element, 130 Binormal, 26
Asymptotic curve, 134 spherical image, 36
ofa developable surface, 140 (8.53) Bisect a curve, 163
Asymptotic direction, 134, 138 Bonnet's rigidity theorem, 15l
Atlas,204 Bound, 181
Axis: Boundary, 181
of a helix, 32 Bounded set, 174
of revolution, 86 ( 1.2) Bump function, 215 (3.6)

Ball of radius E, 200 Catastrophe theory, 246

Barbier's Theorem, 70 Catenoid, 137 (8.15), 146, 147, 148

267
268 lnde.

Cauchy-Schwarzlnequality, 3 Coordinate transformation (cont.)

Chern-Hopf Conjecture, 241 effect on Li,, 108 (4.4)
Christoffel symbol, I 04, 105 effect on I! 1 , 127 (7.6)
of a connection, 224 effect on tangent vectors, 85
intrinsic formula, 105 Covariant derivative, 116 (5.14), 224,
Circle of latitude, 86 ( 1.2) 225
Class ct, 11, 209 Covariant transformation, 98
Closed set, 174 Cover, 89,204
Codazzi-Mainardi equations, 142 Crofton'sFormula, 168
Coefficients of the Riemannian Cross product, 6
metric,93 Curvaturaintegra, 189
Compact, 174 Curvature:
Complete, 112, 238 center of, 39(4.34),41 (4.44)
Components of a vector, 2 of a regular curve, 24
Cone,207 motivation, 24
asadevelopable surface, 141 (8.56) Gaussian, 127, 130, 174, 242 (8.12)
gil' 101 (3.7) formula, 137 (8.6, 8.7)
Conjugatedirections, 134, 138 intrinsic formula, 143
Conjugate family of curves, 139 (8.31) geodesic, 103
Connection, 224 extrinsic formula, 106
metrical, 234 intrinsicformula, 106
Riemannian, 232, 236 mean, 127, 130,242(8.12)
forasubmanifold,241 (8.7) formula, 137 (8.3, 8.11)
symmetric, 235 normal, 103
torsion-free, 235 of a plane curve, 25, 52
Contact: principal, 129
spherical, 39 radius of, 34, 39 (4.34)
ofa surface, 138 (8.29) ofan oval, 72(5.6, 5.9)
Continuous function: Ricci,239
in a metric space, 201 scalar,240
from a surface to R 2 , 89 sectional, 239
Contravariant transformation, 98 total, 161, 189
Convex curve, 60 Curve:
Convex surfa~, 193 asymptotic, 134
Coordinate chart, 203 binormal to, 26
Coordinate function, 212 closed, 53
Coordinate patch, 77 convex, 60, 63 (3.1)
Coordinate transformation, 79 curvature, 24 (see also Curvature)
effect on du',101 evolute,40
effect on g, 97 Fundamental Theorem, 42
effectonr/, 108(4.11) geometric, 14
effect on gil, 96 involute, 40
effect on g"',97 knotted, 169
Index 269

Curve (cont.) Curve (cont.)

length,20 vertex of, 41 (4.44), 66
length minimizing, 112 Curvilinear coordinate system, 84
maximally straight, 120 Cusp, 71 (5.5)
parametric, 84 Cylinder, 115(5.10), 140, 152(10.3)
piecewise regular, 58 curvature of, 136 (8.1)
plane, 29 (3.6, 3.7), 41 (4.44), as a developable surface, 141 (8.56)
45(5.1,5.3) L, 127 (7.2)
Bertrand mate, 38 (4.32) R/ik• 144(9.4)
characterization, 31, 34, 36 (4.13,
4.16), 37 (4.18, 4.20), 48 (6.11)
curvature, 25, 52 Darbouxvector, 35(4.l l)
evolute, 41 (4.43) Developablesurface, 140
involute,41 (4.40,4.41) characterization, 140(8.49)
normal vector field, 52 Gaussian curvature, 140 (8.52)
rotationindex,55,58 Diffeomorphism, 223 (5.4)
spherical contact, 40 (4.36) Differentiable function:
tangent circular image, 58 of manifolds, 209
tangent vector field, 52 of a manifold to R, 209
polar axis, 39 (4.34) ofsurfaces, 146
principal normal, 26 ofasurfacetoR, 124, 145(9.7)
radius of curvature, 34, 39 (4.34) Differentiable vector field, 117
radius of torsion, 34 Differential of a mapping, 219
regular, 15 Dimension:
reparametrization, 17 of a manifold, 204
by arc length, 21 of a vector space, 2
segment,20 Directional derivative, 124
reparametrization, 23 (2.7) Distance, 237
simple, 54 Drag curve, 158
speed,15 Dual basis, 100
sphere, 19 (I. 7), 34, 37, 45 ( 5.3) Dual space, 100
characterization, 33, 37 (4.22, du;as a linear functional, 100
4.25, 4.26), 38 (4.27) Dupin indicatrix, 131, 138
normalcurvature, 107, 108(4.9) asymptotes, 134
torsion, 171
spherical contact, 39
tangent developable surface, 88 €-neighborhood, 88
(1.14) Eigenvalue,5,128
tangent line, 16 Eigenvector,5, 128
tangent vector field, 15 Elliptic integral, 23
unit speed, 22, 24 Elliptic paraboloid, 88 (1.11 ),
unknotted, 169 108(4.5)
velocity vector, 15 Elliptic point, 132, 138 (8.22)
280 Index

Ellipsoid, 88 (1.12), 91, 92 (2.4) Gauss1Dap,131,180

Embedding, 221 Gauss'sequations, 141
Euclid's Fifth Postulate, 188 (4.5), 231 Gauss'sformulas, 104,241 (8.7)
Euler characteristic, 188, 190 Gauss's Theorema Egregium, 143,
Euler's Theorem, 129 149
Evolute of a curve, 40 Genus, 190
Existence and uniqueness theorem: Geodesic, 109
for curves, 42 as an asymptotic curve, 138 (8.28)
for geodesics, 111, 231 characterization, 109, 110
for parallel vector fields, 118, 228 differentialequation, 109, 115(5.7,
for Riemannian connections, 236 5.8)
for surfaces, 151 existence and uniqueness theorem,
111
image under isometrics, 152 ( 10.5)
on a manifold, 230
Fary-MilnorTheorem, 73, 169
Geodesic coordinate patch, 115 ( 5.5),
Fenchel'sTheorem, 72, 165, 170(2.3)
176
Field of metrics, 233
Geodetic torsion, 137 (8.13)
Field of vectors, 216
GL(n, R), 205
First fundamental form, 94
Graph of a function, 78
Flat Euclidean space:
Green's Theorem, 51
connection of, 224, 227 (6.1)
covariant derivatives, 226
curvature, 242 (8.11)
Hadamard'sTheorem, 193
geodesics, 232 (7 .1)
Helicoid, 87(1.4), 138 (8.16), 146, 147,
parallel vector fields, 228
148, 153 (10.10)
Riemannian connection, 241 (8.4)
line of striction, 140 (8.41)
Flat point (Planar point), 132
as a ruled surface, 139 (8.36)
Four-Vertex Theorem, 66
Helix,46(5.5),48(6.9), ll5(5.10)
Frenet-Serret apparatus, 26
axis, 32, 35 (4.9)
Frenet-Serret equations, 30
characterization, 35 (4.3), 37 (4.21)
Frenet-Serret Theorem, 30
circular, 17, 23 (2.1, 2.3), 26, 32
Fundamental Lemma ofRiemannian
Bertrand mate, 46(5.6)
Geometry, 231
characterization, 46 (5.4, 5.6)
Fundamental Theorem of Curves, 42
general,32
Fundamental Theorem ofSurfaces,
generalform,44-45
151
involute,41 (4.40)
Laucret's characterization, 32
pitch, 32, 35 (4.9)
ru,,,105 Hemisphere, 162
Gauss-BonnetFormula, 173, 185 Hilbert's Theorem, 238
Gauss-Bonnet Theorem, 188 Homogeneous coordinates, 200
generalization, 191 Hopf-Rinow Theorem, 238
Index 281

Hyperboliccosine, 12 Isomorphism, 4
Hyperbolic disk (see Poincare disk) lsoperimetric Inequality, 64
Hyperbolic half plane, 179 (2.4),
188 (4.2, 4.3. 4.5)
connection, 225 Jacobian, 79, 201, 220
covariant derivatives, 227 (6.3) Jacobi's identity, 218
distance, 241 (8.8) Jacobi's Theorem, 162
geodesics, 231 Jordan Curve Theorem, 59
parallel translation, 229
parallel vector fields, 228
Riemannian connection, 241 (8.5) Kronecker symbol, 3
Riemannian metric, 234
Hyperbolic paraboloid, 87 (1. 5), 91,
111 Lancret's Theorem, 32
asymptotic curves, 138 (8.27) Left invariant vector field, 223 (5.5)
geodesics, 115(5.9), 116(5.12) Length of a curve, 20, 237
as a ruled surface, 139 (8.37) Length of a vector, 2
Hyperbolicpoint, 132, 138(8.22,8.23) Length minimizing curves, 112
Hyperbolic secant, 231 Levelsurface,91
Hyperholicsine, 12 orientability, 181
Hyperbolic space form, 242 (8.13) Lie bracket, 144(9.5), 217
Hyperbolic tangent, 231 Lie group, 209, 223 (5.5)
Hyperboloid of one sheet,.91 Line:
geodesics, 115(5.11), 116(5.13) as an asymptotic curve, 138 (8.26)
as a ruled surface, 139 (8.38) characterization, 28, 33, 36 (4.12,
Hyperboloid of two sheets, 92 4.16), 48 (6.8)
Hypersurface defined by f, 203, 206, equation,8
221, 241 (8.7) asageodesic, 115(5.4)
tangentspace,222 Line integral, 50
Line of curvature, 129
asanasymptoticcurve, 139(8.33)
Immersion, 223 ofadevelopable surface, 140(8.53)
Implicit Function Theorem, 202 ofa surface ofrevolution, 137 (8.5)
Index of a vector field, 195 Line ofstriction, 140 (8.40)
Inner product, 2 Linear functional, 100
Intrinsicnormal, 103 Linear transformation, 4
Inverse Function Theorem, 202 eigenvalues and eigenvectors, 5
Involute of a curve, 40 Linearly dependent and independent
Isometric, 147 vectors, 2
locally, 147 Logarithmic spiral, 19 (1.6), 45 (5.1)
Isometry, 147, 234
local, 147
orientation preserving, 152 (10.8) Manifold, 204
282 Index

Manifold (cont.) Normal spherical image (cont.)

dimension, 204 of a surface, 131
product, 208 (2.2) Normal vector field:
tangentspace,213 of a plane curve, 52
Maximally straight curve, 120 of a regular curve, 26
Measure of a set, 167 of a surface, 81
Meridian, 86(1.2) n-sphere, 203, 205, 208(2.l), 215(3.2)
Metric: curvature, 242 (8.11)
induced, 234, 241 (8. 7) n-torus, 208 (2.3)
nondegenerate, 240 Null-homotopic, 182
Riemannian, 233
topological, 200
Metric coefficients, 93
O(n), 208 (2.5)
Metric space, 200
Open set:
Meusnier's Theorem on umbilics 175
in a metric space, 200
Minimal surface, 130, 137 '
in the plane, 76
characterization, 138 (8.17)
in a surface, 123
Mixed scalar product, 7
Opposite point:
M(n), 205, 208 (2.4)
on an oval, 69
Mobius band, 87 (I .9)
on an ovaloid, 194(6. l)
Iineofstriction, 140(8.41)
Orientable surface, 180
nonorientability, 87(1.9), 181
Orientation of a vector space, 6
as a ruled surface, 139 (8.36)
right handed, 6
as a surface, 92 (2.5)
Orientedgreatcircle, 167
Monge patch, 78
Orthogonal net, 101(3.4)
area formula, 136
Orthogonal vectors, 3
r/, 106-101
Orthonormal basis, 3
gf'94 Osculatingcircle, 39(4.34)
g 1, IOI (3.3)
Osculating paraboloid, 138 (8.29)
£11,106-107 Osculating plane, 31
Monotone, 60
equation, 35(4.7)
Moving frame, 28
Osculating sphere, 39 (4.35)
Moving trihedron, 28
Oval, 66, 71 (5.1), 72(5.6, 5.7, 5.8, 5.9)
constant width, 69
Barbier's Theorem, 70
Neighborhood, 200
width,69
Nonhomogeneous coordinates 205
Ovaloid (Convex surface), 193
Non-unit speed curves, 46-48 '
Normal plane, 31, 37(4.22)
Normalspace,103
ofasubmanifold, 241 (8.7) Parabolic point, 132, 138(8.22)
Normal spherical image: Parallelism defined by a curve, 122
of a curve, 36, 162 (6.5)
Index 283

Parallelizable manifold, 218 (4.6), Proper coordinate patch, 89

232(7.3) Pseudo-sphere, 158
Parallel translation, 118
on a manifold, 229
and metrical connections, 234 Raisinganindex, 126
Parallel vector field: Rectifying plane, 31
along a curve, II 7, 228 Reflection through a point, 103
differential equation, 117 Region, 181
existence and uniqueness, 118 Regularity condition for surfaces,
on a manifold, 229 77
Parametric curves, 84 Reparametrization:
as asymptotic curves, 138 (8.24) by arc length, 21
as a conjugate family of curves, ofacurve, 17
139(8.32) of a curve segment, 23 (2. 7)
Period of a closed curve, 53 of a surface, 79
Perpendicular (see Orthogonal net, Riemann-Christoffel curvature ten-
Orthogonal vectors) sor:
Picard's Theorem, 42 type (I, 3), 238
Planar point (Flat point), 132 type (0, 4), 239
Plane: Riemannian connection, 232, 236
curvature of, 136 (8.1) Riemannian curvature tensor, 141,
equation of, 9 143
geodesic curvature in, 108 (4.6) coefficients, 141, 242(8.9)
gli, 101 (3.6) invariant definition, 145 (9. I0)
normal, 31 symmetry properties, 144(9.2),
osculating, 31 145(9.11)
parallel vector fields in, 117 Riemannian manifold, 233
R,'ik• 144(9.l) Riemannian metric, 233
rectifying, 31 Right handed orientation, 6
Plethora, 246 Rigid motion, 150
Poincare-BrouwerTheorem, 196 Rigidly equivalent, 150
Poincare disk, 179 (2.5), 188 (4.2, 4.4, Rotation, 150
4.5), 242(8.13) Rotation index, 55, 58
Polar axis, 39 (4.34) Rotation Index Theorem, 56, 161
Polygon, 187(4.1), 188, 191 (5.2) Ruled surface, 139
Principal curvatures, 129 curvature, 139 (8.39)
Principal directions, 129 line of striction, 140 (8.40)
Principal normal vector field, 26 noncylindrical, 140
Product manifold, 208 (2.2) singular point, 139
Projectiven-space, 200, 203 (I.I, 1.2),
205, 215 (3.2) '
Projective plane, 200, 223 (5.2, 5.6) Second fundamental form, 123
Proper coordinate chart, 203 coefficients, 104
284 Index

Second fundamental tensor of a sub- Surface (cont.)

manifold, 241 (8.7) convex, 193
Self-adjoint linear transformation, first fundamental form, 94
127 Fundamental Theorem, 151
Simply connected, 185 metric coefficients, 93
SL(n),201 minimal, 130
Span,2 normal to, 81
Special orthogonal matrix, 150 normal space, I 03
Speed ofa curve, 15 orientable, 180
Sphere, 87(1.10), 88(1.13), 108 (4.7, simple, 77
4.10), 185(3.2), 187,209, tangent plane, 81
215 (3.1), 223 (5.2) tangent space, 93
angular excess, 187 Surface of revolution, 86 ( 1.2), 179
angularvariation, 184 axis, 86 (1.2)
area formula, 136 circle oflatitude, 86 ( 1.2)
curvatureof, 136(8.1) of constant curvature, 153
Dupinindicatrix, 132, 138(8.20) curvature, 127(7.5), 137(8.10),
equation, 10 159(11.1)
Euler characteristic, 190 developable, 140 (8.48)
g,96 geodesics, 110, 115 (5.1, 5.6)
geodesics, 110, 115(5.3),232(7.5) gu, 101 (3.1)
gr,, 94, 98, 101 (3.2) Lt,, 108(4.2,4.3)
g ,96 1!1,127(7.5)
L, 127(7.1) linesofcurvature, 137(8.5)
normal to, 102, 108 (4.8) meridian, 86 ( 1.2)
orientability, 180 minimal, 138 (8.18)
paralleltranslationon, 118-119, principal curvatures, 137 (8.10)
121 as a surface, 92 (2.1)
parallel vector fields, 117
R,/i1., 144(9.3)
simply connected, 185 Tangent circular image, 58
stereographic projection, 87 (1.10) Tangent developable surface, 88
as a surface, 90, 92 (2.3) (l.14), 140(8.46)
Sphere of dimension n (seen-sphere) as a developable surface, 141 (8.56)
Spherical image, 36 gu,140(8.47)
Stereographic projection, 87 (1.10) line of striction, 140 (8.41)
String involute, 71 (5.5) as a ruled surface, 139 (8.36)
Submanifold, 221 Tangentlinetoacurve, 16
Surface,89 Tangent plane, 81
arc length on, 95 Tangent space:
complete, 112 to a manifold, 213
ofconstantcurvature,178 to a surface, 93
of constant width, 194(6.1) basis, 84
Index

Tangent spherical image, 36 Unit circle, 58, I 69

curvature and torsion, 48 (6.10) Unit disk, 169
Tangent vector field, 15 Unit normal to a surface, 81
invariance under reparametriza- Unit speed curve, 22, 24, 237
tion, 18
of a plane curve, 52
Tangent vectors: Vector field :
on a manifold, 210 along a curve, 116, 225
on a surface, 83 on a manifold, 216
as a vector space 84, 213 existence, 218 (4.2)
Taylor's Theorem, 213 Vector product, 6
Tensoroftype(l, I), 126 Vector space, I
TheoremaEgregium, 143, 149 basis,2
Torsion: dimension, 2
ofa curve, 26, 139 (8.35) orientation, 6
geodetic, 137 (8.13) Vectors:
radius,34 angle between, 3
ofa sphere curve, 171 components, 2
Torus, 86(1.1), 152(10.4), 181, 197 linearly dependent and indepen-
(7.1) dent, 2
curvature of, 136 (8.1) orthogonal, 3
L, 127(7.3) Velocityvector, 15
not simply connected, 185 Vertex,41 (4.44),66
orientability, 180 Four-Vertex Theorem, 66
R/jt, 144(9.4)
as a surface, 92 (2.2)
Total angular variation, 182
Totalcurvatureofacurve, 161 Weingartenmap, 125
Total curvature ofa surface, 189 coefficients, 125
Total index ofa vector field, 196 Weingarten'sequations, 126
Totaltorsion, 170 Whitehead's Theorem, 238
Tractrix, 158 Whitney Embedding Theorem, 222
Triple scalar product, 7 Width:
ofan oval, 69
ofan ovaloid, 194 (6.1)
u;-curve, 84
Umbilic, 129
Meusnier's Theorem, 175 Zero of a vector field, 195