Front Cover
Contents
Preface
Acknowledgements
Chapter 1: Plane Curves: Local Properties
1.1 Parametrizations
1.2 Position, Velocity, and Acceleration
1.3 Curvature
1.4 Osculating Circles, Evolutes, and Involutes
1.5 Natural Equations
Chapter 2: Plane Curves: Global Properties
2.1 Basic Properties
2.2 Rotation Index
2.3 Isoperimetric Inequality
2.4 Curvature, Convexity, and the Four-Vertex Theorem
Chapter 3: Curves in Space: Local Properties
3.1 Definitions, Examples, and Differentiation
3.2 Curvature, Torsion, and the Frenet Frame
3.3 Osculating Plane and Osculating Sphere. 3.4 Natural EquationsChapter 4: Curves in Space: Global Properties
4.1 Basic Properties
4.2 Indicatrices and Total Curvature
4.3.1 Knots
Chapter 5: Regular Surfaces
5.1 Parametrized Surfaces
5.2 Tangent Planes and Regular Surfaces
5.3 Change of Coordinates
5.4 The Tangent Space and the Normal Vector
5.5 Orientable Surfaces
Chapter 6: The First and Second Fundamental Forms
6.1 The First Fundamental Form
6.2 Map Projections (Optional)
6.2.1 Metric Properties of Maps of the Earth
6.2.2 Azimuthal Projections
6.2.3 Cylindrical Projections. 6.2.4 Coordinate Changes on the Sphere6.3 The Gauss Map
6.4 The Second Fundamental Form
6.5 Normal and Principal Curvatures
6.6 Gaussian and Mean Curvatures
6.7 Developable Surfaces and Minimal Surfaces
6.7.1 Developable Surfaces
6.7.2 Minimal Surfaces
Chapter 7: The Fundamental Equations of Surfaces
7.1 Gauss's Equations and the Christoffel Symbols
7.2 Codazzi Equations and the Theorema Egregium
7.3 The Fundamental Theorem of Surface Theory
Chapter 8: The Gauss-Bonnet Theorem and Geometry of Geodesics
8.1 Curvatures and Torsion
8.1.1 Natural Frames
8.1.2 Normal Curvature. 8.1.3 Geodesic Curvature8.1.4 Geodesic Torsion
8.2 Gauss-Bonnet Theorem, Local Form
8.3 Gauss-Bonnet Theorem, Global Form
8.4 Geodesics
8.5 Geodesic Coordinates
8.5.1 General Geodesic Coordinates
8.5.2 Geodesic Polar Coordinates
8.6 Applications to Plane, Spherical, and Elliptic Geometry
8.6.1 Plane Geometry
8.6.2 Spherical Geometry
8.6.3 Elliptic Geometry
8.7 Hyperbolic Geometry
8.7.1 Synthetic Hyperbolic Geometry
8.7.2 The Poincaré Upper Half-Plane
8.7.3 The Poincaré Disk
8.7.4 The Pseudosphere Revisited
Chapter 9: Curves and Surfaces in n-dimensional Euclidean Space. 9.1 Curves in n-dimensional Euclidean Space9.1.1 Curvatures and the Frenet Frame
9.1.2 Osculating Planes, Circles, and Spheres
9.1.3 The Fundamental Theorem of Curves in Rn
9.2 Surfaces in n-dimensional Euclidean Space
9.2.1 Regular Surfaces in Rn
9.2.2 Intrinsic Geometry for Surfaces
9.2.3 Orientability
9.2.4 The Gauss-Bonnet Theorem
Appendix A: Tensor Notation
A.1 Tensor Notation
A.1.1 Curvilinear Coordinate Systems
A.1.2 Tensors: Definitions and Notation
A.1.3 Operations on Tensors
A.1.4 Examples
A.1.5 Symmetries
A.1.6 Numerical Tensors
Bibliography
Back Cover.

##### Citation preview

Differential Geometry of Curves and Surfaces

Differential Geometry of Curves and Surfaces Second Edition Thomas Banchoff Stephen Lovett

Contents

1

Preface

vii

Acknowledgements

xv

Plane Curves: Local Properties

1.1 1.2 1.3 1.4 1.5 2

. . . . .

. . . . .

. . . . .

1 13 25 33 40 47

Basic Properties . . . . . Rotation Index . . . . . . Isoperimetric Inequality . Curvature, Convexity, and

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . the Four-Vertex Theorem

Curves in Space: Local Properties

3.1 3.2 3.3 3.4 4

. . . . .

Plane Curves: Global Properties

2.1 2.2 2.3 2.4 3

1

Parametrizations . . . . . . . . . . . . . . . Position, Velocity, and Acceleration . . . . . Curvature . . . . . . . . . . . . . . . . . . . Osculating Circles, Evolutes, and Involutes Natural Equations . . . . . . . . . . . . . .

Definitions, Examples, and Differentiation Curvature, Torsion, and the Frenet Frame Osculating Plane and Osculating Sphere . Natural Equations . . . . . . . . . . . . .

47 52 61 64 71

. . . .

. . . .

. . . .

. . . .

. . . .

71 80 90 97

Curves in Space: Global Properties

103

4.1 4.2 4.3

103 106 115

Basic Properties . . . . . . . . . . . . . . . . . . . Indicatrices and Total Curvature . . . . . . . . . . Knots and Links . . . . . . . . . . . . . . . . . . .

v

vi

Contents

5

Regular Surfaces

5.1 5.2 5.3 5.4 5.5 6

8

. . . . .

. . . . .

. . . . .

125 133 150 155 159

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

165 180 192 198 209 221 231

The Fundamental Equations of Surfaces

247

7.1 7.2 7.3

Gauss’s Equations and the Christoffel Symbols . . Codazzi Equations and the Theorema Egregium . The Fundamental Theorem of Surface Theory . . .

248 258 268

The First and Second Fundamental Forms

The First Fundamental Form . . . . . . . . Map Projections (Optional) . . . . . . . . . The Gauss Map . . . . . . . . . . . . . . . . The Second Fundamental Form . . . . . . . Normal and Principal Curvatures . . . . . . Gaussian and Mean Curvatures . . . . . . . Developable Surfaces and Minimal Surfaces

165

The Gauss-Bonnet Theorem and Geometry of Geodesics

273

8.1 8.2 8.3 8.4 8.5 8.6

274 284 295 305 323

8.7 9

125

. . . . .

6.1 6.2 6.3 6.4 6.5 6.6 6.7 7

Parametrized Surfaces . . . . . . . . . . . . Tangent Planes and Regular Surfaces . . . . Change of Coordinates . . . . . . . . . . . . The Tangent Space and the Normal Vector Orientable Surfaces . . . . . . . . . . . . . .

Curvatures and Torsion . . . . . . . . . . . . . . . Gauss-Bonnet Theorem, Local Form . . . . . . . . Gauss-Bonnet Theorem, Global Form . . . . . . . Geodesics . . . . . . . . . . . . . . . . . . . . . . . Geodesic Coordinates . . . . . . . . . . . . . . . . Applications to Plane, Spherical, and Elliptic Geometry . . . . . . . . . . . . . . . . . . . . . . . . . Hyperbolic Geometry . . . . . . . . . . . . . . . .

335 342

Curves and Surfaces in n-dimensional Euclidean Space

355

9.1 9.2

355 364

Curves in n-dimensional Euclidean Space . . . . . Surfaces in n-dimensional Euclidean Space . . . . .

A Tensor Notation

A.1 Tensor Notation . . . . . . . . . . . . . . . . . . . Bibliography

377

377 405

Preface

What Is Differential Geometry? Differential geometry studies properties of curves, surfaces, and higher-dimensional curved spaces using tools from calculus and linear algebra. Just as the introduction of calculus expands the descriptive and predictive abilities of nearly every field of scientific study, so the use of calculus in geometry brings about avenues of inquiry that extend far beyond classical geometry. Before the advent of calculus, much of geometry consisted of proving consequences of Euclid’s postulates. Even conics, which came into vogue in the physical sciences after Kepler observed that planets travel around the sun in ellipses, arise as the intersection of a double cone and a plane, two shapes which fit comfortably within the paradigm of Euclidean geometry. One cannot underestimate the impact of geometry on science, philosophy, and civilization as a whole. The geometric proofs in Euclid’s Elements served as models of mathematical proof for over two thousand years in the Western tradition of a liberal arts education. Geometry also produced an unending flow of applications in surveying, architecture, ballistics, astronomy, and natural philosophy more generally. The objects of study in Euclidean geometry (points, lines, planes, circles, spheres, cones, and conics) are limited in what they can describe. A boundless variety of curves and surfaces and manifolds arise naturally in areas of inquiry that employ geometry. To address these new classes of objects, various branches of mathematics brought their tools to bear on the expanding horizons of geometry, each with a different bent and set of fruitful results. Techniques from calculus and analysis led to differential geometry, pure set theoretic methods led to topology, and modern algebra contributed the field of algebraic geometry.

vii

viii

Preface

The types of questions one typically asks in differential geometry extend far beyond what one can ask in classical geometry and yet the former do not entirely subsume the latter. Differential geometry questions often fall into two categories: local properties, by which one means properties of a curve or surface defined in the neighborhood of a point, or global properties, which refer to properties of the curve or surface taken as a whole. As a comparison to functions of one variable, the derivative of a function f at a point a is a local property since one only needs information f near a whereas the integral of f between a and b is a global property. Some of the most interesting theorems in differential geometry relate local properties to global ones. The culminating theorem in this book, the Gauss-Bonnet Theorem, relates global properties of curves and surfaces to the topology of a surface and leads to fundamental results in non-Euclidean spherical and hyperbolic geometry.

Using This Textbook This new edition is intended as a textbook for a single semester undergraduate course in the differential geometry of curves and surfaces, with only multivariable calculus and linear algebra as prerequisites. The interactive computer graphics applets that are provided for this book can be used for computer labs, in-class illustrations, exploratory exercises, or simply as intuitive aids for the reader. Each section concludes with a collection of exercises which range from perfunctory to challenging, suitable for daily or weekly problem sets. However, the self-contained text, the careful introduction of concepts, the many exercises, and the interactive computer graphics also make this text well-suited for self-study. Such a reader should feel free to primarily follow the textbook and use the software as supporting material; primarily follow the presentation in the software package and consult the textbook for definitions, theorems, and proofs; or try to follow both with equal weight. Either way, the authors hope that the dual nature of software applets and classic textbook structure will offer the reader both a rigorous and intuitive introduction to the field of differential geometry.

Preface

This book is the first in a pair of books which together are intended to bring the reader through classical differential geometry into the modern formulation of the differential geometry of manifolds. The second book in the pair, by Lovett, is entitled Differential Geometry of Manifolds . Neither book directly relies on the other but knowledge of the content of this book is quite beneficial for .

ix

x

Preface

The authors encourage the reader to consult the general usage instruction pages for the applets. All of the applet materials are available on-line at http://diffgeo.akpeters.com/.

Organization of Topics Chapters 1 through 4 cover alternately the local and global theory of plane and space curves. In the local theory, we introduce the fundamental notions of curvature and torsion, construct various associated objects (e.g., the evolute, osculating circle, osculating sphere), and present the fundamental theorem of plane or space curves, which is an analogue of the fundamental theorem of calculus. The global theory studies how local properties (esp. curvature) relate to global properties such as closedness, concavity, winding numbers, and knottedness. The topics in these chapters are particularly well suited for computer investigation. The authors know from experience in teaching how often students make discoveries on their own by being able to quickly manipulate curves and their associated objects and properties. Chapter 5 rigorously introduces the notion of a regular surface, the type of surface on which the techniques of differential geometry are well-defined. Here one first sees the tangent plane and the concept of orientability. Chapter 6 introduces the local theory of surfaces in R3 , focusing on the metric tensor and the Gauss map from which one defines the essential notions of principal, Gaussian, and mean curvatures. In addition, we introduce the study of surfaces that have Gaussian curvature or mean curvature identically 0. One cannot underestimate the importance of this chapter. Even a reader primarily interested in the advanced topic of differentiable manifolds should be comfortable with the local theory of surfaces in R3 because it provides many visual and tractable examples of what one generalizes in the theory of manifolds. Here again, as in Chapter 8, the use of the software applets is an invaluable aid for developing a good geometric intuition. Chapter 7 first introduces the reader to the component notation for tensors. It then establishes the famous Theorema Egregium, the celebrated classical result that the Gaussian curvature depends only

Preface

on the metric tensor. Finally, it outlines a proof for the fundamental theorem of surface theory. Another title commonly used for Chapter 8 is Intrinsic Geometry. Just as Chapter 1 considers the local theory of plane curves, Chapter 8 starts with the local theory of curves on surfaces. Of particular importance in this chapter are geodesics and geodesic coordinates. The highlight of this chapter is the famous Gauss-Bonnet Theorem, both in its local and global forms, without which no elementary course in differential geometry is complete. The chapter finishes with applications to spherical and hyperbolic geometry. The book concludes in Chapter 9 with a brief discussion on curves and surfaces in Euclidean n-space. This chapter emphasizes in what ways definitions and formulas given for objects in R3 extend and possibly change when adapted to Rn .

A Comment on Prerequisites The mathematics or physics student often first encounters differential geometry at the graduate level. Typically, at that point, one is immediately exposed to the formalism of manifolds, thereby skipping the intuitive and visual foundation that informs the deeper theory. The advent of computer graphics has added a new dimension to and renewed the interest in classical differential geometry but this pedagogical habit remains. The authors wish to provide a book that introduces the undergraduate student to an interesting and visually stimulating mathematical subject that is accessible with only the full calculus sequence and linear algebra as prerequisites. In calculus courses, students usually do not study all the analysis that underlies the theorems one uses. Similarly, in keeping with the stated requirements, this textbook does not always provide all the topological and analytical background for some theorems. The reader who is interested in all the supporting material is encouraged to consult . A few key results presented in this textbook rely on theorems from the theory of differential equations but either the calculations are all spelled out or a reference to the appropriate theorem has been provided. Therefore, experience with differential equations is

xi

xii

Preface

occasionally helpful though not necessary. In a few cases, the authors chose not to supply the full proofs of certain results but instead refer the reader to the more complete text . A few exercises require some skills with differential equations but these are clearly marked with the prefix (ODE). Problems marked with (*) indicate difficulty, which may be related to technical ability, insight, or length.

Notation As a comment on vector notation, this book and  consistently use the following conventions. A vector or vector function in a Euclidean ~ ~ vector space is denoted by ~v , X(t), or X(u, v). Often γ indicates ~ ~ ~ a curve parametrized by X(t) while writing X(t) = X(u(t), v(t)) indicates a curve on a surface. The unit tangent and the binormal vectors of a curve in space are written in the standard notation ~ T~ (t) and B(t) but the principal normal is written P~ (t), reserving ~ N (t) to refer to the unit normal vector to a curve on a surface. ~ (t) is the vector obtained by rotating T~ (t) by a For a plane curve, U positive quarter turn. Furthermore, we denote by κg (t) the curvature of a plane curve since one identifies this curvature as the geodesic curvature in the theory of curves on surfaces. In this book, we often work with matrices of functions. The functions themselves are denoted, for example, by aij , and we denote the matrix by (aij ). Furthermore, it is essential to distinguish between a linear transformation between vector spaces T : V → W and its matrix with respect to given bases in V and W . Following notation that is common in current linear algebra texts, if B is a basis in V  B 0 and B 0 is a basis in W , then we denote by T B the matrix of T with respect to these bases. If the bases are understood by context, we simply write [T ] for the matrix associated to T . Occasionally, there arise irreconcilable discrepancies in definitions or notations (e.g., the definition of a critical point for a function Rn → Rm , or how one defines θ and φ in spherical coordinates). In these instances the authors made a choice that best suits their purposes and indicated commonly used alternatives.

Preface

xiii

Section Dependency This book is primarily organized on the theme of a pendulum back and forth between local theory and global theory, first of plane curves, then space curves, and then surfaces in R3 . The authors wrote the book from the perspective that the Gauss-Bonnet Theorem (presented in Sections 8.2 and 8.3) serves as the culminating theorem of the textbook. Sections depend on each other according to the following chart. 1.1–1.5 2.1–2.4

3.1–3.4

4.1–4.3

5.1–5.5 6.1, 6.3–6.6

6.2

6.7

A.1

7.1–7.2

7.3

8.1–8.5 8.6–8.7

9.1–9.2

Note that Appendix A.1 on tensor notation is valuable for background on the indices of tensor notation but it is not essential for Chapter 7 and subsequent chapters.

Changes in the Second Edition Many faculty who adopted the first edition of the textbook gave helpful feedback on their experiences. Following this feedback, in this second edition, the authors preserved the intent and attempted to improve on the execution of the first edition. This includes the following changes:

xiv

Preface

• Reworking the presentation in many places to make it more approachable; • Adding more exercises, both introductory and advanced; • Including a section on the application of differential geometry to cartography; • Adding a few investigative projects ideas; • Significantly reorganizing the last third of the book in order to reach more quickly the Gauss-Bonnet Theorem; • Adding two sections dedicated to hyperbolic and spherical geometry as applications of intrinsic geometry; • Adding a brief chapter on curves and surfaces in n-dimension Euclidean space.

Acknowledgements

Thomas Banchoff Our work at Brown University on computer visualizations in differential geometry goes back more than forty-five years, and I acknowledge my collaborator computer scientist Charles Strauss for the first fifteen years of our projects. Since 1982, an impressive collection of students have been involved in the creation and development of the software used to produce the applets for this book. The site for my 65th birthday conference http://www.math.brown. edu/TFBCON2003 lists dozens of them, together with descriptions of their contributions. Particular thanks for the applets connected with this book belong to David Eigen, Mark Howison, Greg Baltazar, Michael Schwarz, and Michael Morris. For his work as a student, an assistant, and now as a co-author, I am extremely grateful to Steve Lovett. Special thanks for help in the Brown University mathematics department go to Doreen Pappas, Natalie Johnson, Audrey Aguiar, and Carol Oliveira, and to Larry Larrivee for his invaluable computer assistance. Finally, I thank my wife Kathleen for all her support and encouragement.

Stephen Lovett I would first like to thank Thomas Banchoff my teacher, mentor, and friend. After one class, he invited me to join his team of students on developing electronic books for differential geometry and multivariable calculus. Despite ultimately specializing in algebra, the exciting projects he led and his inspiring course in differential geometry instilled in me a passion for differential geometry. His ability to introduce differential geometry as a visually stimulating and mathe-

xv

xvi

Acknowledgements

matically interesting topic served as one of my personal motivations for writing this book. I am grateful to the students and former colleagues at Eastern Nazarene College. In particular I would like to acknowledge the undergraduate students who served as a sounding board for the first few drafts of this manuscript: Luke Cochran, David Constantine, Joseph Cox, Stephen Mapes, and Christopher Young. Special thanks are due to my colleagues Karl Giberson, Lee Hammerstrom, and John Free. In addition, I am indebted to Ellie Waal who helped with editing and index creation. The continued support from my colleagues at Wheaton College made writing this book a gratifying project. In particular, I must thank Terry Perciante, Chair of the Department of Mathematics and Computer Science, for his enthusiasm and his interest. I am indebted to Dorothy Chapell, Dean of the Natural & Social Sciences, and to Stanton Jones, Provost of the College, for their encouragement and for a grant which freed up my time to finish writing. I am also grateful to Thomas VanDrunen and Darren Craig for helpful comments. Finally, I cannot adequately express in just a few words how much I am grateful to my wife Carla Favreau Lovett and my daughter Anne. While I was absorbed in this project, they provided a loving home, they braved the significant time commitment and encouraged me at every step. They also kindly put up with my occasional geometry comments such as how to see the Gaussian curvature in the reflection of “the Bean” in Chicago.

Second Edition Acknowledgments In preparation for this second edition, we wish to thank the many people who contributed corrections, improvements, and suggestions. These include but are not limited to Teddy Parker for careful editing, Daniel Flath for many suggestions for improvements, Judith Arms for errata and suggestions, and Robert Ferr´eol for bringing to our attention an excellent on-line encyclopedia of curves at http:// www.mathcurve.com/ (in French). We would also like to thank Nate Veldt, Gary Babbatz, Nathan Bliss, Matthew McMillan, and Cole Adams.

CHAPTER 1

Plane Curves: Local Properties

Just as calculus courses introduce real functions of one variable before tackling multivariable calculus, so it is natural to study curves before addressing surfaces and higher-dimensional objects. This first chapter presents local properties of plane curves, where by local property we mean properties that are defined in a neighborhood of a point on the curve. For the sake of comparison with calculus, the derivative f 0 (a) of a function f at a point a is a local property of the function since we only need knowledge of f (x) for x in (a − ε, a + ε), where ε is any positive real number, to define f 0 (a). In contrast, the definite integral of a function over an interval is a global property since we need knowledge of the function over the whole interval to calculate the integral. In contrast to this present chapter, Chapter 2 introduces global properties of plane curves.

1.1 Parametrizations Borrowing from a physical understanding of motion in the plane, we can think about plane curves by specifying the coordinates x and y as functions of a time variable t, which give the position of a point traveling along the curve. Thus we need two functions x(t) and y(t). Using vector notation to locate a point on the curve, we often write ~ ~ X(t) = (x(t), y(t)) for this pair of coordinate functions and call X(t) 2 a vector function into R . From a mathematical standpoint, t does not have to refer to time and is simply called the parameter of the vector function. Example 1.1.1 (Lines). Euclid’s first postulate of geometry is that through two distinct points there passes exactly one line. The point slope formula gives a Cartesian equation of a line through two points.

1

2

1. Plane Curves: Local Properties

y

~v

p~ + 2~v

p~

x

Figure 1.1. A line in the plane.

In analytic geometry, the approach to describing a line through two points is slightly different. Given two distinct points p~1 = (x1 , y1 ) and p~2 = (x2 , y2 ), the vector ~v = p~2 − p~1 = (x2 − x1 , y2 − y1 ) is called a direction vector of the line because all vectors along this line are multiples of ~v . Then every point on the line can be written with a position vector p~1 + t~v for some t ∈ R. Figure 1.1 shows an example with t = 2. Therefore, we find that a line can also be described by providing a point and a direction vector. Using the coordinates of vectors, given a point p~ = (x0 , y0 ) and a direction vector ~v = (v1 , v2 ), a line through p~ in the direction of ~v is the image of the following vector function: ~ = p~ + t~v = (x0 + v1 t, y0 + v2 t) X(t)

for t ∈ R.

Note that just as the same line can be determined by two different pairs of points, so the same line may be specified by different sets of these equations. For example, using any point p~ on the line will ultimately trace out the same line as t varies through all of R. Similarly, if we replace ~v with any nonzero multiple of itself, the set of points traced out as t varies through R is the same line in R2 plane. Example 1.1.2 (Circles). The pair of functions

~ = (R cos t + a, R sin t + b) X(t)

1.1. Parametrizations

3

traces out a circle of radius R > 0 centered at the point (a, b). To see this, note that by the definition of the sin t and cos t functions, (cos t, sin t) are the coordinates of the point on the unit circle that is also on the ray out of the origin that makes an angle t with the positive x-axis. Thus, ~ 1 (t) = (cos t, sin t), X

with t ∈ [0, 2π],

traces out the unit circle in a counterclockwise manner. Multiplying both coordinate functions by R stretches the circle out by a factor of R away from the origin. Thus, the vector function ~ 2 (t) = (R cos t, R sin t), X

with t ∈ [0, 2π],

has as its image the circle of radius R centered at the origin. Notice ~ 2 (t) = (x(t), y(t)), we deduce that x(t)2 + also that by writing X 2 2 y(t) = R for all t, which is the algebraic equation of the circle. In order to obtain a vector function that traces out a circle centered ~ 2 by the vector (a, b). at the point (a, b), we must simply translate X This is vector addition, and so we get ~ X(t) = (R cos t + a, R sin t + b),

with t ∈ [0, 2π].

Two different vector functions can have the same image in R2 . For example, if ω > 0, ~ X(t) = (cos ωt, sin ωt),

with t ∈ [0, 2π/ω],

also has the unit circle as its image. Referring to vocabulary in physics, this latter vector function corresponds to a point moving around the unit circle at an angular velocity of ω. When trying to establish a suitable mathematical definition of what one usually thinks of as a curve, one does not wish to consider as a curve points that jump around or pieces of segments. We would like to think of a curve as unbroken in some sense. In calculus, one introduces the notion of continuity to describe functions without “jumps” or holes, but one must exercise a little care in carrying over the notion of continuity to vector functions. More generally, we need to define the notion of a limit of a vector function as the parameter t approaches a fixed value. First, however, we remind the reader of the Euclidean distance formula.

4

1. Plane Curves: Local Properties

Definition 1.1.3. Let ~ v be a vector in R2 with coordinates ~v = (v1 , v2 )

in the standard basis. The (Euclidean) length of ~v is given by k~v k =

q √ ~v · ~v = v12 + v22 .

If p and q are two points in Rn with coordinates given by vectors ~v and w, ~ then the Euclidean distance between p and q is kw ~ − ~v k. ~ be a vector function from a subset of R into Definition 1.1.4. Let X

~ Rn . We say that the limit of X(t) as t approaches a is a vector w, ~ and we write ~ lim X(t) = w, ~ t→a

if for all ε > 0 there exists a δ > 0 such that 0 < |t − a| < δ implies ~ kX(t) − wk ~ < ε. Definition 1.1.5. Let I be an open interval of R, let a ∈ I, and let ~ : I → R2 be a vector function. We say that X(t) ~ X is continuous at ~ a if the limit as t approaches a of X(t) exists and

~ ~ lim X(t) = X(a).

t→a

The above definitions mirror the usual definition of a limit of a real function but must use the length of a vector difference to ~ discuss the proximity between X(t) and a fixed vector w. ~ Though at the outset this definition appears more complicated than the usual definition of a limit of a real function, the following proposition shows that it is not. ~ be a vector function from a subset of R into Proposition 1.1.6. Let X R2 that is defined over an interval containing a, though perhaps not ~ at a itself. Suppose in coordinates we have X(t) = (x(t), y(t)) wher~ is defined. If w ever X ~ = (w1 , w2 ), then ~ lim X(t) =w ~

t→a

if and only if

lim x(t) = w1

t→a

and

lim y(t) = w2 .

t→a

~ = w. ~ Let ε > 0 be arbitrary Proof: Suppose first that limt→a X(t) and let δ > 0 satisfy the definition of the limit of the vector function.

1.1. Parametrizations

5

~ ~ Note that |x(t)−w1 | ≤ kX(t)− wk ~ and that |y(t)−w2 | ≤ kX(t)− wk. ~ Hence, 0 < |t − a| < δ implies |x(t) − w1 | < ε and |y(t) − w2 | < ε. Thus, limt→a x(t) = w1 and limt→a y(t) = w2 . Conversely, suppose that limt→a x(t) = w1 and limt→a y(t) = w2 . Let ε > 0 be an arbitrary positive real number. By definition, there exist √ δ1 and δ2 such that 0 < |t − a| < δ1 implies |x(t) √ − w1 | < ε/ 2 and 0 < |t − a| < δ2 implies |y(t) − w2 | < ε/ 2. Taking δ = min(δ1 , δ2 ) we see that 0 < |t − a| < δ implies that r p ε2 ε2 ~ kX(t) − wk ~ = |x(t) − w1 |2 + |y(t) − w2 |2 < + = ε. 2 2 This finishes the proof of the proposition.



Corollary 1.1.7. Let I be an open interval of R, let a ∈ I, and consider

~ : I → R2 with X(t) ~ ~ a vector function X = (x(t), y(t)). Then X(t) is continuous at t = a if and only if x(t) and y(t) are continuous at t = a. Definition 1.1.5 and Corollary 1.1.7 provide the mathematical framework for what one usually thinks of as a curve in physical intuition. This motivates the following definition. Definition 1.1.8. Let I be an interval of R. A parametrized curve (or

~ : I → R2 . parametric curve) in the plane is a continuous function X ~ If we write X(t) = (x(t), y(t)), then the functions x : I → R and y : I → R are called the coordinate functions or parametric equations ~ of the parametrized curve. We call the locus of X(t) the image of 2 ~ X(t) as a subset of R .

It is important to note the distinction in this definition between a parametrized curve and its locus. For example, in Example 1.1.1 ~ we show that the parametrization X(t) = t~v + p~ traces out a line L that goes through the point p~ with direction vector ~v . However, the ~ ~ : R → R2 and not the vector-valued function X line is the locus of X itself. The vector-valued function is the parametrized curve. Since the function t 7→ t3 −t is a bijection from R to itself, the parametrized ~ (t) = (t3 − t)~v + p~ has the same locus, the line L, but is quite curve Y different as a function.

6

1. Plane Curves: Local Properties

The following examples begin to provide a library of parametric curves and illustrate how to construct parametric curves to describe a particular shape or trajectory. Example 1.1.9 (Graphs of Functions). The graph of a continuous func-

tion f : [a, b] → R over an interval [a, b] can be viewed as a parametric curve. In order to view the graph of a continuous function as a pa~ rametrized curve, we use the coordinate functions X(t) = (t, f (t)), with t ∈ [a, b]. Example 1.1.10 (Circles Revisited). Another parametrization for the

unit circle (sometimes used in number theory) is   1 − t2 2t ~ for t ∈ R. , X = (x(t), y(t)) = 1 + t2 1 + t2

(1.1)

It is easy to see that for all t ∈ R, x(t)2 +y(t)2 = 1, which means that ~ is on the unit circle. However, this parametrization the locus of X does not trace out the entire circle as it misses the point (−1, 0). We leave it as an exercise to determine a geometric interpretation of the parameter t and to show that ~ = lim X ~ = (−1, 0). lim X

t→∞

t→−∞

This example emphasizes that it is not possible to tell what the parametrization is simply by looking at the locus. A counterclockwise circle is indistinguishable from the circle covered twice, or from the circle traced out in a clockwise fashion, or, as in this example, from a circle covered in a completely different way. Example 1.1.11 (Ellipses). Without repeating all the reasoning of the

previous exercise, it is not hard to see that ~ X(t) = (a cos t, b sin t) provides a parametrization for the ellipse centered at the origin with axes along the x- and y-axes, with respective half-axes of length |a| and |b|. Note that these coordinate functions do indeed satisfy x(t)2 y(t)2 + 2 =1 a2 b

for all t.

1.1. Parametrizations

7

Example 1.1.12 (Lissajous Figures). It is sometimes amusing to see how

cos t and sin t relate to each other if we change their respective periods. Lissajous figures, which arise in the context of electronics, are curves parametrized by ~ X(t) = (cos mt, sin nt), where m and n are positive integers. See Figure 1.2 for an example of a Lissajous figure with m = 5 and n = 3.

Figure 1.2. A Lissajous figure. Example 1.1.13 (Cycloids). We can think of a usual cycloid as the locus

traced out by a point of light affixed to a bicycle tire as the bicycle rolls forward. We can establish a parametrization of such a curve as follows. Assume the wheel of radius a begins with its center at (0, a) so that the part of the wheel touching the x-axis is at the origin. We view the wheel as rolling forward on the positive x-axis. As the wheel rolls, the position of the center of the wheel is f~(t) = (at, a), where t is the angle measuring how much (many times) the wheel has turned since it started. At the same time, the light – at a distance a from the center of the wheel and first positioned straight down from the center of the wheel – rotates in a clockwise motion around the center of the wheel. See Figure 1.3. The motion of the light with respect to the center of the wheel is    π π  ~g (t) = a cos −t − , a sin −t − = (−a sin t, −a cos t). 2 2

8

1. Plane Curves: Local Properties

y

x Figure 1.3. Cycloid.

The locus of the cycloid is the vector function that is the sum of f~(t) and ~g (t). Thus, a parametrization for the cycloid is ~ X(t) = (at − a sin t, a − a cos t). One can point out that reflectors on bicycle wheels are usually not attached directly on the tire but on a spoke of the wheel. We can easily modify the above discussion to obtain the relevant parametric equations for when the point of light is located at a distance b from the center of the rolling wheel. One obtains ~ X(t) = (at − b sin t, a − b cos t). If 0 < b < a, one obtains the curve of a realistic bicycle tire reflector, and this locus is called a curtate cycloid. In contrast, the locus obtained by letting b > a is called a prolate cycloid. Example 1.1.14 (Heart Curve). Arguably the most popular curve around

Valentine’s Day is the heart. Here are some parametric equations that trace out such a curve: ~ X(t) = ((1 − cos2 t) sin t, (1 − cos3 t) cos t). We encourage the reader to visit this example in the accompanying software and to explore ways of modifying these equations to create other interesting curves. On the website http://www.mathcurve. com/ entitled an Online Encyclopedia of Curves, the host credits this curve to Rapha¨el Laporte who designed this curve for his “petite amie” (translated “girl friend”). Example 1.1.15 (Polar Functions). Functions in polar coordinates are usually given in terms of the radius r as a function of the angle

1.1. Parametrizations

9

θ by r = f (θ). The graphs of such functions can be written as parametrized curves. Recall the coordinate transformation ( x = r cos θ, y = r sin θ. Then take θ as the parameter t, and the parametric equations for the graph of r = f (θ) are ~ X(t) = (f (t) cos t, f (t) sin t). As an example, the polar function r = sin 3θ traces out a curve that resembles a three-leaf flower. (See Figure 1.4.) As a parametric curve, it is given by ~ X(t) = (sin 3t cos t, sin 3t sin t). y

x

Figure 1.4. Three-leaf flower.

Example 1.1.16 (Cardioid). Another common polar function is the car-

dioid, which is the locus of r = 1 − cos θ. In parametric equations, we have ~ X(t) = ((1 − cos t) cos t, (1 − cos t) sin t). ~ :I→ As mentioned in Example 1.1.2, for a parametric curve X ~ | t ∈ I} as a subset of R2 does R2 , the set of points C = {X(t) not depend uniquely on the functions x(t) and y(t). In fact, it is important to make a careful distinction between the notion of a

10

1. Plane Curves: Local Properties

parametrized curve as defined above, the image of the parametrized curve as a subset of R2 (also called the locus of the curve), and the notion of a curve, eventually defined as a one-dimensional manifold. (See [24, Chapter 3].) ~ : I → R2 and any Definition 1.1.17. Given a parametrized curve X continuous functions g from an interval J onto the interval I, we can produce a new vector function ξ~ : J → R2 defined by ξ~ = ~ ◦ g. The image of ξ~ is again the set C, and ξ~ = X ~ ◦ g is called a X ~ reparametrization of X. If g is not onto I, then the image of ξ~ may be a proper subset of C. In this case, we usually do not call ξ~ a reparametrization as it ~ does not trace out the same locus of X.

Problems 1.1.1. Using linear algebra, it is possible to prove that the shortest distance between a point (x0 , y0 ) and a line with equation ax + by + c = 0 is d=

|ax0 + by0 + c| √ . a2 + b2

However, a calculus proof of this result is also possible using parametrized curves. ~ (a) Show that X(t) = (−bt − a/c, at) with t ∈ R is a parametrization of the line with equation ax + by + c = 0. (b) Find the value t0 of t that minimizes the function p ~ f (t) = kX(t) − (x0 , y0 )k = (−bt − a/c − x0 )2 + (at − y0 )2 , ~ which gives the distance between a point X(t) on the line and the point (x0 , y0 ). (c) Show that f (t0 ) simplifies to the distance formula given above. 1.1.2. Let p~ be a fixed point, and let ~l(t) = ~at + ~b be the parametric equations of a line. Prove that the distance between p~ and the line is s (~a · (~b − p~))2 k~a × (~b − p~)k , k~b − p~k2 − = k~ak2 k~ak ~ = (w1 , w2 , 0) in the plane, we where for vectors ~v = (v1 , v2 , 0) and w call ~v × w ~ = (0, 0, v1 w2 − v2 w1 ), which is the cross product between ~v and w ~ when viewed as vectors in R3 . [See the previous problem.]

1.1. Parametrizations

11

1.1.3. Equation (1.1) gave the following parametric equations for a circle:   1 − t2 2t ~ , , for t ∈ R. X = (x(t), y(t)) = 1 + t 2 1 + t2 Prove that the parameter t is equal to tan θ, where θ is the angle shown in the following picture of the unit circle.

θ

1.1.4. Let C be the circle of center (0, 2) and radius 1. Let S be the set of points in R2 that are the same distance from the outside of the circle C as they are from the x-axis. (a) Prove that S has a parametrization   3 sin t 3 cos t ~ ,2 − . X(t) = 1 + cos t 1 + cos t [Hint: Use the parameter t as the angle as shown below.] (b) Prove that S is a parabola and find the Cartesian equation for it.

(0, 2) t ~ X(t)

~ 1.1.5. Find the closest point to (16, 0.5) on the curve X(t) = (t, t2 ). 1.1.6. An epicycloid is defined as the locus of a point on the edge of a circle of radius b as this circle rolls on the outside of a fixed circle of radius a. Supposing that at t = 0, the moving point is located

12

1. Plane Curves: Local Properties

at (a, 0). Prove that the following are parametric equations for an epicycloid:      ~ = (a + b) cos t − b cos a + b t , (a + b) sin t − b sin a + b t . X(t) b b 1.1.7. A hypocycloid is defined as the locus of a point on the edge of a circle of radius b as this circle rolls on the inside of a fixed circle of radius a. Assuming that a > b and that at t = 0, the moving point is located at (a, 0), find parametric equations for a hypocycloid. 2 ~ ~ 1.1.8. Consider the parametrized  curve X : R → R with equations X(t) = 1 t −t 1 t −t (e + e ), (e − e ) . Prove that the locus is one branch of a 2 2 unit hyperbola x2 − y 2 = 1. Give a parametrization for the other branch of the hyperbola.

1.1.9. Consider the vectors ~a = (3, 3) and ~b = (−1, 1). Explain why the ~ parametric curve X(t) = (cos t)~a +(sin t)~b is an ellipse. Furthermore, give a Cartesian equation of the locus of this parametric curve. ~ 1.1.10. Consider the parametric curve X(t) = (t3 − 5t, 3t2 ). Graphing it shows that it intersects itself. By solving for t1 and t2 the equation ~ 2 ), find the parameters where the curve intersects itself ~ 1 ) = X(t X(t and give the coordinates of the point on the locus where this occurs. ~ : [0, 2π] → R2 that gives a defor1.1.11. Consider the parametrized curve X ~ mation of a cardioid X(t) = ((a + cos t) cos t, (a + cos t) sin t), where a is a real number. Show that the curve intersects itself for |a| < 1. Describe what happens for a = 0. 1.1.12. (*) Consider a curve in R2 that is the solution of a Cartesian equation of the form ax2 + 2bxy + cy 2 = d, where a, b, c, d are real numbers. Note that the left-hand side can be expressed in matrices as ax2 + 2bxy + cy 2 = x

y

  a b

  x . y

b c

Recall the Spectral Theorem from linear algebra, namely that a symmetric matrix can be orthogonally diagonalized. Prove that if the eigenvalues of the above symmetric matrix are both nonzero and have the same sign, then the curve is an ellipse. Find a parametric equation for this ellipse.

1.2. Position, Velocity, and Acceleration

1.2 Position, Velocity, and Acceleration ~ In physics, one interprets the vector function X(t) = (x(t), y(t)) as providing the location along a curve at time t in reference to some fixed frame. A frame is a (usually orthonormal) basis attached to a fixed origin. The point O = (0, 0) along with the basis {~ı, ~}, where ~ı = (1, 0)

and ~ = (0, 1),

~ form the standard reference frame. We call the vector function X(t) the position vector. When one uses the standard reference frame, it is not uncommon to write ~ X(t) = x(t)~ı + y(t)~. Directly imitating Newton’s approach to finding the slope of a curve at a certain point, it is natural to ask the question, “What is ~ at point t0 ?” If we the direction and rate of change of a curve X ~ ~ 1 ), the change in look at two points on the curve, say X(t0 ) and X(t ~ 1 ) − X(t ~ 0 ), while to specify the position is given by the vector X(t rate of change in position, we would need to scale this vector by a ~ 0) factor of t1 − t0 . Hence, the vectorial rate of change between X(t ~ 1 ) is and X(t  1 ~ ~ 0) . X(t1 ) − X(t t1 − t0 To define an instantaneous rate of change along the curve at the point t0 , we need to calculate (if it exists and if it has meaning) the limit   ~ 0 (t0 ) = lim 1 X(t ~ 0 + h) − X(t ~ 0) . X (1.2) h→0 h ~ According to Proposition 1.1.6, if X(t) = (x(t), y(t)), then the limit in Equation (1.2) exists if and only if x(t) and y(t) are both dif~ 0 (t0 ) = ferentiable at t0 . Furthermore, if this limit exists, then X (x0 (t0 ), y 0 (t0 )). This leads us to the following definition. ~ : I → R2 be a vector function with coordinates Definition 1.2.1. Let X

~ ~ is differentiable at t = t0 if x(t) X(t) = (x(t), y(t)). We say that X and y(t) are both differentiable at t0 . If J is the common domain to x0 (t) and y 0 (t), then we define the derivative of the vector function ~ as the new vector function X ~ 0 : J → R2 defined by X ~ 0 (t) = X 0 0 (x (t), y (t)).

13

14

1. Plane Curves: Local Properties

If x(t) and y(t) are both differentiable functions on their domain, ~ 0 (t) = (x0 (t), y 0 (t)) is called the vethe derivative vector function X locity vector. Mathematically, this is just another vector function and traces out another curve when placed in the standard reference frame. However, because it illustrates the direction of motion along the curve, one often visualizes the velocity vector corresponding to ~ 0 ) on the curve. t = t0 as based at the point X(t Following physics language, we call the second derivative of the ~ 00 (t) = (x00 (t), y 00 (t)) the acceleration vector related vector function X ~ to X(t). In general, our calculations often require that our vector functions can be differentiated at least once and sometimes more. Consequently, when establishing theorems, we like to succinctly describe the largest class of functions for which a particular result holds. Definition 1.2.2. Let I be an interval of R, and let f : I → R be a

~ : I → R2 a parametrized curve. We say that f or X ~ is function or X r ~ of class C on I if the rth derivative of X exists and is continuous on I. We denote by C ∞ the class of functions or parametrized curves that have derivatives of all orders on I. It is also common to write f ∈ C r (I, R) to say that f is a real-valued function of class C r and ~ ∈ C r (I, R2 ) to say that X ~ is a parametrized curve in the to write X r plane of class C . To say that a function is of class C 0 over the interval I means that it is continuous. By basic theorems in calculus, the condition that a function be of a certain class is an increasingly restrictive condition. In other words, as sets of functions defined over the same interval I, the classes are nested according to C 0 ⊃ C 1 ⊃ C 2 ⊃ · · · ⊃ C ∞. Whenever we impose a condition that a function is of class C r for some r, such a supposition is generically called a “smoothness condition.” Proposition 1.2.3. Let ~ v (t) and w(t) ~ be vector functions defined and

differentiable over an interval I ⊂ R. Then the following hold: ~ ~ 0 (t) = c~v 0 (t). 1. If X(t) = c~v (t), where c ∈ R, then X

1.2. Position, Velocity, and Acceleration

~ 2. If X(t) = c(t)~v (t), where c : I → R is a real function, then ~ 0 (t) = c0 (t)~v (t) + c(t)~v 0 (t). X ~ 3. If X(t) = ~v (t) + w(t), ~ then ~ 0 (t) = ~v 0 (t) + w ~ 0 (t). X ~ 4. If X(t) = ~v (f (t)) is a vector function and f : J → I is a real function into I, then ~ 0 (t) = f 0 (t)~v 0 (f (t)) . X 5. If f (t) = ~v (t) · w(t) ~ is the dot product between ~v (t) and w(t), ~ then f 0 (t) = ~v 0 (t) · w(t) ~ + ~v (t) · w ~ 0 (t). ~ Example 1.2.4. Consider the spiral defined by X(t) = (t cos t, t sin t) for t ≥ 0. The velocity and acceleration vectors are ~ 0 (t) = (cos t − t sin t, sin t + t cos t), X ~ 00 (t) = (−2 sin t − t cos t, 2 cos t − t sin t). X ~ and X ~ 0 as a function of If we wish to calculate the angle between X time, we use the dot product method. In this example, we calculate p t2 sin2 t + t2 cos2 t = |t|, p p ~ 0 (t)k = (cos t − t sin t)2 + (sin t + t cos t)2 = 1 + t2 , kX ~ ~ 0 (t) = t cos2 t − t2 cos t sin t + t sin2 t + t2 cos t sin t = t. X(t) ·X ~ kX(t)k =

~ ~ 0 (t) is defined for all t 6= 0 Thus, the angle θ(t) between X(t) and X and is equal to !   ~ ~ 0 (t) t X(t) · X −1 −1 √ θ(t) = cos . = cos ~ ~ 0 (t)k |t| 1 + t2 kX(t)k kX

15

16

1. Plane Curves: Local Properties

~ 0 (t0 ) X

~ 1) X(t

~ ∆X

~ 0) X(t O Figure 1.5. Arc length segment.

~ Let C be the locus of a vector function X(t) for t ∈ [a, b]. Then we approximate the length of the arc l(C) with the Riemann sum l(C) ≈

n X

~ i ) − X(t ~ i−1 )k ≈ kX(t

i=1

n X

~ 0 (ti )k∆t. kX

i=1

~ i) − (See Figure 1.5 as an illustration for the approximation kX(t 0 ~ (ti )k∆t.) Taking the limit of this Riemann sum, we ~ i−1 )k ≈ kX X(t obtain the following formula for the arc length of C: Z bp l= (x0 (t))2 + (y 0 (t))2 dt. (1.3) a

A rigorous justification for the arc length formula, using the Mean Value Theorem, can be found in most calculus textbooks. In light of Equation (1.3), we define the arc length function s : ~ [a, b] → R related to X(t) as the arc length along C over the interval [a, t]. Thus, Z tp Z t 0 ~ kX (u)k du = s(t) = (x0 (u))2 + (y 0 (u))2 du. (1.4) a

a

By the Fundamental Theorem of Calculus, we then have p ~ 0 (t)k = (x0 (t))2 + (y 0 (t))2 , s0 (t) = kX which, still following the vocabulary from the trajectory of a moving ~ particle, we call the speed function of X(t).

1.2. Position, Velocity, and Acceleration

17

~ Example 1.2.5. Consider the parametrized curve defined by X(t) = ~ 0 (t) = (2t, 3t2 ), and thus the arc (t2 , t3 ). The velocity vector is X length function from t = 0 is r Z t Z tp Z t 4 0 2 4 ~ s(t) = + u2 du kX (u)k du = 4u + 9u du = 3|u| 9 0 0 0 3/2  4 2 +t , = sign(t) 9 where sign(t) = 1 if t > 0 and −1 if t < 0 and sign(0) = 0. The ~ length of X(t) between t = 0 and t = 1 is  1  3/2 13 − 8 . s(1) − s(0) = 27 The speed function allows us to understand that a parametrized ~ : I → R2 does not only contain the information that decurve X scribes the locus of the curve but it also contains information about the speed of travel s(t) along the locus. By looking at the locus, it is impossible to discern the speed function. ~ ◦ f, Consider a reparametrization of the parametrized curve X where f : J → I is a differentiable surjective (i.e., onto) function. If ~ (u0 )) = t0 = f (u0 ) where f is a differentiable function at u0 , then X(f ~ 0 ) and X(t

d

~ 0 (f (u0 ))k ~ (u0 )) = kf 0 (u0 )X

X(f

dt

~ 0 (t0 )k = |f 0 (u0 )|s0 (t0 ). = |f 0 (u0 )| kX ~ 0 (t), by which we mean Up to a change in sign, the direction of X ~ 0 (t), does not change under repathe unit vector associated with X ~ ◦ f , at any parameter value rametrization. More precisely, if ξ~ = X 0 0 ~ (f (t)) 6= ~0, we have t where f (t) 6= 0 and X ~ 0 (f (t)) ξ~0 (t) f 0 (t) X , = 0 ~ 0 (f (t))k |f (t)| kX kξ~0 (t)k

(1.5)

where f 0 (t)/|f 0 (t)| is +1 or −1. It bears repeating that in order for the right-hand side to be well defined, one needs f 0 (t) 6= 0.

18

1. Plane Curves: Local Properties

~ : I → R2 be a parametrized curve. For a Definition 1.2.6. Let X continuously differentiable function f from an interval J onto I, we ~ ◦ f a regular reparametrization if for all t ∈ J, f 0 (t) is call ξ~ = X well defined and never 0. In addition, a regular reparametrization is called positively oriented (resp. negatively oriented ) if f 0 (t) > 0 (resp. f 0 (t) < 0) for all t ∈ J. By the Mean Value Theorem, if f 0 (t) 6= 0 for all t ∈ J, then f : J → I is an injective function. Thus since f is by definition surjective, regular reparametrizations involve a bijective function f : J → I. ~ 0 (t)k is invari~ 0 (t)/kX Equation (1.5) shows that the unit vector X ant under a positively oriented reparametrization and simply changes under a negatively oriented reparametrization. Consequently, this unit vector is an important geometric object associated to a curve at a point. ~ : I → R2 be a plane parametric curve. A point Definition 1.2.7. Let X

~ ~ t0 ∈ I is called a critical point of X(t) if X(t) is not differentiable 0 ~ (t0 ) = ~0. If t0 is a critical point, then X(t ~ 0 ) is called a at t0 or if X critical value. A point t = t0 that is not critical is called a regular ~ : I → R2 is called regular if it is of point. A parametrized curve X 1 ~ 0 (t) 6= ~0 for all t ∈ I. class C (i.e., continuously differentiable) and X ~ we define the unit tangent vector Finally, if t is a regular point of X, T~ (t) as ~ 0 (t) X . T~ (t) = ~ 0 (t)k kX It is not uncommon to call a property of a curve C or a property of a point P on a curve C a geometric property if it does not depend on a parametrization of the locus near the point. In particular, a geometric property does not depend on the speed of travel along the curve through the point. At any point of a curve where T~ (t) is defined, that is to say at any regular point, we can write the velocity vector as ~ 0 (t) = s0 (t)T~ (t). X

(1.6)

1.2. Position, Velocity, and Acceleration

19

This expresses the velocity vector as the product of its magnitude (speed) and direction (unit tangent vector). In most differential geometry texts, authors simplify their formulas by reparametrizing by arc length. From the perspective of coordinates, this means picking an “origin” O on the curve C occurring at some fixed t = t0 and using the arc length along C between O and any other point P as the parameter with which to locate P on C. In fact, this can always be done. ~ Proposition 1.2.8 (Reparametrization by Arc Length). If X(t) is a regular parametrized curve, then there is a regular reparametrization of ~ by arc length. Furthermore, if X ~ is of class C k , then the arc X length reparametrization is also of class C k . Proof: Let s = f (t) be the arc length function with s = 0 corre~ is regular over its sponding to some point on the curve. Since X ~ 0 (t)k > 0 for all t. Since f (t) is strictly indomain, then f 0 (t) = kX creasing, it has an inverse function t = h(s). By the Inverse Function Theorem, since f 0 (t) 6= 0, then h(s) is differentiable with h0 (s) =

1 1 = 0 . f 0 (h(s)) f (t)

(1.7)

~ (s) = X(h(s)) ~ Note that the composite function Y satisfies ~0 ~ 0 (s) = X ~ 0 (h(s))h0 (s) = X ~ 0 (t) 1 = X (t) . Y ~ 0 (t)k f 0 (t) kX ~ is of class C k , then s0 (t) = f 0 (t) = kX ~ 0 (t)k is of class C k−1 . If X k Hence s(t) itself is of class C . The Inverse Function Theorem states that the inverse function is also of class C k . Consequently, the arc length parametrization ~y (s) is of class C k since it is the composition ~ of X(t) of class C k and t = h(s) of class C k .  The Inverse Function Theorem as used in the above proof (and its more general multivariable counterpart) appear in most introductory texts on analysis. (See  for example.) However, the key derivative in Equation (1.7) follows from the chain rule for invertible functions and is a standard differentiation formula.

20

1. Plane Curves: Local Properties

The habit of reparametrizing by arc length has benefits and drawbacks. The great benefit of this approach is that along a curve parametrized by arc length, the speed function s0 is identically 1 and the velocity vector is exactly the tangent vector ~ 0 (s) = T~ (s). X This formulation simplifies proofs and difficult calculations. The main drawback is that, in practice, most curves do not admit a simple formula for their arc length function, let alone a formula that can be written using elementary functions. For example, given an ~ explicit parametrized curve X(t), it is often very challenging to find s(t) as defined in Equation (1.4). Furthermore, to reparametrize by arc length, it would be necessary to find the inverse function t(s), representing the original parameter t as a function of s. Even if it is possible to find s(t), determining this inverse function usually cannot be written with elementary functions. Then the parametrization by ~ ~ arc length is X(s) = X(t(s)). Even using computer algebra systems, reparametrizing by arc length remains an intractable problem. The notion of “regular” in many ways mirrors geometric properties of continuously differentiable single-variable functions. In par~ ticular, if a parametrized curve X(t) is regular at t0 , then locally the curve looks linear. As we will see again in Chapter 3, with the formalism of vector functions it becomes particularly easy to express the equation of the tangent line to a curve at a point in any number ~ of dimensions. If X(t) is a curve and t0 is not a critical point for the ~ t (t) of the tangent line at t0 is curve, then the equation L 0 ~ 0 ) + (t − t0 )X ~ 0 (t0 ), ~ t (t) = X(t L 0

with t ∈ R,

or alternatively ~ 0 ) + uT~ (t0 ), ~ t (u) = X(t L 0

with u ∈ R,

(1.8)

if we do not wish to confuse the parameter of the parametrized curve ~ X(t) and the parameter of the tangent line. On the other hand, if t0 ∈ I is a critical point for the curve ~ X(t), the curve may or may not have a tangent line at t = t0 . If the following one-sided limits exist, we can define two unit tangent

1.2. Position, Velocity, and Acceleration

21

~ 0 ): vectors at X(t ~ T~ (t+ 0 ) = lim T (t) t→t+ 0

and

~ T~ (t− 0 ) = lim T (t). t→t− 0

Consequently, we can determine the angle the curve makes with itself at the corner t0 as   ~ (t− ) . ) · T α0 = cos−1 T~ (t+ 0 0 ~ + To be more precise, the angle from T~ (t− 0 ) to T (t0 ) is the exterior angle of the curve at the corner t = t0 . One can only define a tangent ~ ~ + line to X(t) at t = t0 if T~ (t− 0 ) = T (t0 ). ~ = (t2 , t3 ). We calculate that X ~ 0 (t) = (2t, 3t2 ), Example 1.2.9. Let X(t) ~ 0 (0) = ~0. However, and therefore t = 0 is a critical point because X if t 6= 0, the unit tangent vector is T~ (t) = √

t 1 (2t, 3t2 ) = √ (2, 3t). 4 + 9t |t| 4 + 9t2

4t2

Then the right-hand and left-hand side unit tangent vectors are T~ (0− ) = (−1, 0)

and

T~ (0+ ) = (1, 0).

~ These calculations indicate that as t approaches 0, X(t) lies in the fourth quadrant but approaches (0, 0) in the horizontal direction ~ stops at (−1, 0). From an intuitive standpoint, we could say that X ◦ t = 0, spins around by 180 , and moves away from the origin in the direction (1, 0), remaining in the first quadrant. ~ ~ 0 (t) = Example 1.2.10. Let X(t) = (t, | tan t|). We calculate that X (1, sign(t) sec2 t) and deduce that t = 0 is a critical point because ~ is not differentiable there. sign(t) is not defined at 0, and hence X However, we also find that 1 T~ (0− ) = √ (1, −1) 2

and

1 T~ (0+ ) = √ (1, 1). 2

Thus, T~ (0− ) · T~ (0+ ) = 0, ~ makes a right angle with itself at t = 0. However, which shows that X ~ + one can tell from the explicit values for T~ (t− 0 ) and T (t0 ) that the π exterior angle of the curve at t = 0 is 2 .

22

1. Plane Curves: Local Properties

If t0 is a critical point, it may still happen that lim T~ (t)

t→t0

~ − exists, namely when T~ (t+ 0 ) = T (t0 ), which also means α0 = 0. Then through an abuse of language, we can still talk about the unit tangent vector at that point. As an example of this possibility, consider the ~ curve X(t) = (t3 , t4 ). We can quickly calculate that lim T~ (t) = (1, 0) = ~ı.

t→0

When this limit exists, even though t0 is a critical point, one can use Equation (1.8) as parametric equations for the tangent line at t0 , replacing T~ (t0 ) in Equation (1.8) with limt→t0 T~ (t). In geometry, physics, and other applications, we must sometimes integrate a function along a curve. To this end, we use path and line integrals of scalar functions or vector functions, depending on the particular problem. Since we will make use of them, we remind the reader of the notation for such integrals. Let C be a curve pa~ rametrized by X(t) over the interval [a, b]. Let f : R2 → R be a real (scalar) function and F~ : R2 → R2 a vector field in the plane. Then the path integral of f and the line integral of F~ over the curve C are respectively

f ds

means

f (x(t), y(t))

p (x0 (t))2 + (y 0 (t))2 dt,

a

C

Z

b

Z

Z

F~ · d~s

b

Z means

F~ (x(t), y(t)) · T~ (t) dt.

a

C

Problems 1.2.1. Calculate the velocity, the acceleration, the speed, and, where defined, the unit tangent vector function of the following parametric curves: ~ (a) The circle X(t) = (R cos ωt, R sin ωt).

1.2. Position, Velocity, and Acceleration

(b) The circle as parametrized in Equation (1.1) in Example 1.1.10. (c) The epicycloids defined in Problem 1.1.6, Section 1.1. 1.2.2. A quadratic B´ezier curve with control points p~0 , p~1 , and p~2 is a ~ ~ ~ : [0, 1] → R2 with end points X(0) = p~0 and X(1) = p~2 curve X ~ 0 (0) = p~1 − p~0 and and such that the control point p~1 satisfies X ~ 0 (1) = p~2 − p~1 . Find the parametrization for this quadratic B´ezier X curve. ~ 1.2.3. Find the tangent line to X(t) = (cos(2t), sin t) at t = π/6. Also give the components of the unit tangent vector there. ~ 1.2.4. The parametrized curve X(t) = ((1 + 2 cos t) cos t, (1 + 2 cos t) sin t) intersects itself at one point. Find this point of intersection and find the angle of self-intersection (i.e., the acute angle between the tangent lines corresponding to the two different parameters t1 and t2 of self-intersection). ~ 1.2.5. For how many points on the Lissajous curve X(t) = (cos(3t), sin(2t)) does the tangent line go through the point (3, 0)? ~ 1.2.6. What can be said about a parametrized curve X(t) that has the 00 ~ property that X (t) is identically 0? 1.2.7. Find the arc length function along the parabola y = x2 , using as the origin s = 0. ~ 1.2.8. Consider the parametrized curve X(t) = (t2 , t3 ) with t ∈ R. Parametrize this curve by arc length with s = 0 corresponding to t =  0. 3/2 ~ [Ans: X(s) = 19 (27s + 8)2/3 − 49 , sign(s) 19 (27s + 8)2/3 − 49 . This is one of the few curves for which this is a tractable problem.] 1.2.9. (*) Consider the cycloid introduced in Example 1.1.13 given by ~ X(t) = (t − sin t, 1 − cos t). Prove that the path taken by a point on the edge of a rolling wheel of radius 1 during one rotation has length 8. ~ 1.2.10. Calculate the arc length function of the curve X(t) = (t2 , ln t), defined for t > 0. ~ : I → R2 be a regular parametrized curve, and let p~ be a fixed 1.2.11. Let X ~ to p~ occurs at point. Suppose that the closest point on the curve X t = t0 , which is neither of the ends of I. Prove that the line between ~ 0 ), is perpendicular to the p~ and the point closest to p~, namely X(t ~ at t = t0 . curve X

23

24

1. Plane Curves: Local Properties

~ : I → R2 intersects 1.2.12. We say that a plane curve C parametrized by X ~ ~ itself at a point p if there exist u 6= t in I such that X(t) = X(u) = p. 2 3 ~ Consider the parametric curve X(t) = (t , t − t) for t ∈ R. Find the point(s) of self-intersection. Also determine the angle at which the curve intersects itself at those point(s) by finding the angle between the unit tangent vectors corresponding to the distinct parameters. 1.2.13. Prove the differentiation formulas in Proposition 1.2.3. ~ ~ ·X ~ 0 = 0 for all values of 1.2.14. Prove that if X(t) is a curve that satisfies X ~ is a circle. [Hint: Use kXk ~ 2=X ~ ·X ~ and apply Proposition t, then X 2 ~ 1.2.3 to calculate the derivative of ||X|| .] ~ 1.2.15. Consider the ellipse given by X(t) = (a cos t, b sin t). Find the extremum values of the speed function. 1.2.16. Consider the linear spiral of Example 1.2.4. Let n ≥ 0 be a nonnegative integer. Prove that the length of the nth derivative vector function is given by p ~ (n) (t)k = n2 + t2 . kX 1.2.17. Consider the exponential spiral ~x(t) = (aebt cos t, aebt sin t) where a and b are constants. Calculate the arc length s(t) function of ~x(t). Reparametrize the spiral by arc length. 1.2.18. Consider again the exponential spiral ~x(t) = (aebt cos t, aebt sin t), with a > 0 and b < 0. ~ (a) Prove that as t → +∞, lim X(t) = (0, 0). 0 ~ (b) Show that X (t) → (0, 0) as t → +∞ and that for any t0 , Z t ~ 0 (u)k du < ∞, kX lim t→∞

t0

i.e., any part of the exponential spiral that “spirals” in toward the origin has finite arc length. 1.2.19. Recall that polar and Cartesian coordinate systems are related as follows: ( ( p x = r cos θ, r = x2 + y 2 , and y = r sin θ, tan θ = xy Suppose C is a curve in the plane parametrized using polar coordinate functions r = r(t) and θ = θ(t) so that one has a parametrization in Cartesian coordinates as ~x(t) = (x(t), y(t)) = (r(t) cos(θ(t)), r(t) sin(θ(t))).

1.3. Curvature

25

(a) Express x0 and y 0 in terms of r, θ, r0 , and θ0 . (b) Express r0 and θ0 in terms of x, y, x0 , and y 0 . (c) Express ||~x|| and ||~x0 || in terms of polar coordinate functions. (All coordinate functions are viewed as functions of t and so r0 , for example, is a shorthand for r0 (t), the derivative of r with respect to t.) 1.2.20. (ODE) Suppose a curve C is parametrized by ~x(t) such that ~x(t) and ~x0 (t) always make a constant angle with each other. Find the shape of this curve. [Hint: Use polar coordinates and the results of Problem 1.2.19.]

1.3 Curvature ~ : I → R2 be a twice-differentiable parametrization of a curve Let X C. As we saw in the previous section, the decomposition of the ve~ 0 = s0 T~ into magnitude and unit tangent direction locity vector X separates the geometric invariant (the unit tangent T~ ) from the dynamical aspect (the speed s0 (t)) of the parametrization. Taking one more derivative, we obtain the decomposition ~ 00 = s00 T~ + s0 T~ 0 . X

(1.9)

The first component describes a tangential acceleration, while the second component describes a rate of change of the tangent direction, or in other words, how much the curve is “curving.” Since T~ is a unit vector, we always have T~ · T~ = 1. Therefore, (T~ · T~ )0 = 0 =⇒ 2T~ · T~ 0 = 0, =⇒ T~ · T~ 0 = 0. Thus, T~ 0 is perpendicular to T~ . Just as there are two unit tangent vectors at a regular point of the curve, there are two unit normal vectors as well. Given a particular parametrization, there is no naturally preferred way to define “the” unit normal vector, so we make a choice.

26

1. Plane Curves: Local Properties

~ : I → R2 be a regular parametrized curve and Definition 1.3.1. Let X

~ T~ = (T1 , T2 ) the tangent vector at a regular value X(t). The unit ~ is normal vector U ~ (t) = (−T2 (t), T1 (t)). U ~ is the vector function obtained by rotating T~ by Equivalently, U π 2 . In still other words, if we view the xy-plane in three-dimensional space with x-, y-, and z-axes oriented in the usual way, and if we call ~k = (0, 0, 1) the unit vector along the positive z-axis, then U ~ = ~k× T~ . ~ at all t. Since T~ 0 ⊥ T~ , the vector function T~ 0 is a multiple of U This leads to the definition of curvature. ~ : I → R2 be a regular twice-differentiable Definition 1.3.2. Let X parametric curve. The curvature function κg (t) is the unique realvalued function defined by ~ (t). T~ 0 (t) = s0 (t)κg (t)U

(1.10)

This definition only gives κg (t) implicitly but we can obtain a formula for it as follows. Equation (1.9) becomes ~ 00 = s00 T~ + (s0 )2 κg U ~. X

(1.11)

~ = ~k. Viewing the plane as the xy-plane in three-space, we have T~ × U Thus, ~ 00 = (s0 T~ ) × (s00 T~ + (s0 )2 κg U ~) ~0 ×X X ~ = s0 s00 T~ × T~ + (s0 )3 κg T~ × U = (s0 )3 κg~k. ~ 0 (t)k, which leads to But s0 (t) = kX κg (t) =

~ 00 (t)) · ~k ~ 0 (t) × X (X x0 (t)y 00 (t) − x00 (t)y 0 (t) = . ~ 0 (t)k3 (x0 (t)2 + y 0 (t)2 )3/2 kX

(1.12)

The reader might well wonder why in the definition of κg (t) we include the factor s0 (t). It is not hard to confirm (see Problem 1.3.15)

1.3. Curvature

27

that including the s0 (t) factor renders κg (t) independent of any regular reparametrization (except perhaps up to a change of sign). ~ One should note at this point that if a curve X(s) is parametrized by arc length, then by Equations (1.6) and (1.9), the velocity and acceleration have the simple expressions ~ 0 (s) = T~ (s), X ~ 00 (s) = κg (s)U ~ (s). X Example 1.3.3. Consider the vector function that describes a particle

moving on a circle of radius R with a nonzero and constant angular ~ velocity ω > 0 given by X(t) = (R cos ωt, R sin ωt). In order to calculate the curvature, we need ~ 0 (t) = (−Rω sin ωt, Rω cos ωt), X ~ 00 (t) = (−Rω 2 cos ωt, −Rω 2 sin ωt). X Thus, using the coordinate form (right-most expression) in Equation (1.12), we get κg (t) =

R2 ω 3 sin2 t + R2 ω 3 cos2 t R2 ω 3 1 = = . 2 R3 ω 3 R (R2 ω 2 sin t + R2 ω 2 cos2 t)3/2

The curvature of the circle is a constant function that is equal to the reciprocal of the radius for all t, regardless of the nonzero angular velocity ω. The study of trajectories in physics gives particular names for the components of the first and second derivatives of a vector func~ }. We already saw that the function s0 (t) in tion in the basis {T~ , U Equation (1.6) is called the speed. In Equation (1.11), however, the function s00 (t) is called the tangential acceleration, while the quantity s0 (t)2 κg (t) is called the centripetal acceleration. Example 1.3.3 connects the curvature function to the reciprocal of a radius, so if ~ κg (t) 6= 0, then we define the radius of curvature to X(t) at t as the 1 function R(t) = κg (t) . Using the common notation v(t) = s0 (t) for the speed function, one recovers the common formula for centripetal acceleration of v2 (s0 )2 κg = . R

28

1. Plane Curves: Local Properties

B H

B A F

C

C

G A

E D

E

G H

D F

Figure 1.6. Curvature function example.

In introductory physics courses, this formula is presented only in the context of circular motion. With differential geometry at our disposal, we see that centripetal acceleration is always equal to v 2 /R at all points on a curve where κg 6= 0, where by R one means the radius of curvature. In contrast to physics where one tends to refer to the radius of curvature, the curvature function is a more geometrically natural quantity to study. Indeed, the radius of curvature of a line segment is undefined even though line segments are such useful geometric objects. On the other hand, a radius of curvature is 0 (and hence the curvature is undefined) at degenerate curves that are points or at critical points. Example 1.3.4. To help build an intuition for the curvature function of

~ a plane curve, consider the parametric curve X(t) = (2 cos t, sin(2t)+ sin t) for t ∈ [0, 2π]. Figure 1.6 shows side by side the locus of the curve and the graph of its curvature function, along with a few labeled points to serve as references. One can see that the curvature function is positive when the curve turns to the left away from the unit tangent vector T~ and negative when the curve turns to the right. The curvature function is 0 when the curve changes from curving to the left of T~ to curving to the right of T~ , or vice versa. The curve is locally curving the “most” when κg (t) has a maximum and κg (t) > 0 or

1.3. Curvature

29

when κg (t) has a minimum and κg (t) < 0. These cases correspond respectively to curving the most to the left and to the right. Besides a situation when the curvature is 0, the curve is locally curving the “least” when κg (t) has a minimum and κg (t) > 0 or when κg (t) has a maximum and κg (t) < 0. Figure 1.6 illustrates all five of the different possibilities just described. ~ : I → R2 has curProposition 1.3.5. A regular parametrized curve X

~ is a line vature κg (t) = 0 for all t ∈ I if and only if the locus of X segment. ~ traces out a line segment, then (perhaps Proof: If the locus of X ~ after a regular reparametrization) we can write X(t) = ~a + ϕ(t)~b, 0 where ϕ(t) is a differentiable real function with ϕ (t) 6= 0. Then ~ 0 (t) = ϕ0 (t)~b, X ~ 00 (t) = ϕ00 (t)~b. X Thus, by Equation (1.12), κg (t) =

(ϕ0 (t)ϕ00 (t)~b × ~b) · ~k =0 (|ϕ0 (t)| k~bk)3/2

because ~b × ~b = ~0. ~ Conversely, if X(t) = (x(t), y(t)) is a curve such that κg (t) = 0, then x0 (t)y 00 (t) − x00 (t)y 0 (t) = 0. We need to find solutions to this differential equation or determine ~ is regular, X ~ 0 6= ~0 for all t ∈ I. how solutions are related. Since X Let I1 be an interval where y 0 (t) 6= 0. Over I1 , we have   x0 y 00 − x00 y 0 x0 d x0 = 0 =⇒ 0 = C, = 0 2 0 (y ) dt y y where C is a constant. Thus, x0 = Cy 0 for all t. Integrating with respect to t we deduce x(t) = Cy(t)+D. Similarly, over an interval I2 where x0 (t) 6= 0, we deduce that y(t) = Ax(t) + B for some constants A and B. Since I can be covered with intervals where x0 (t) 6= 0 or ~ is a piecewise linear curve. y 0 (t) 6= 0, we deduce that the locus of X ~ However, since X is regular, it has no corners, and hence its locus is a line segment. 

30

1. Plane Curves: Local Properties

~ 00 as Note that the curvature function arose from calculating X 0 0 ~ ~ the derivative of the expression X = s T , which involved finding an d ~ ~ expression for dt T (t). Taking the third derivative of X(t) and using the decomposition in Equation (1.11), we get ~ + (s0 )2 κg U ~ 000 = s000 T~ + (3s00 s0 κg + (s0 )2 κ0g )U ~ 0. X ~ 0 (t). By Consequently, we need an expression for the derivative U ~ } forms an orthonormal basis and definition, for all t, the set {T~ , U hence ~ 0 = (U ~ 0 · T~ )T~ + (U ~0 ·U ~ )U ~. U ~ (t) is a unit vector function, the identity U ~ ·U ~0 = 0 However, since U ~ · T~ = 0 for all t, we also deduce holds for all t. Furthermore, since U that ~ 0 · T~ = −U ~ · T~ 0 = −s0 κg . U Consequently, ~ 0 (t) = −s0 (t)κg (t)T~ (t). U ~ : [−a, a] → R2 is reparaIt is not hard to see that if a curve X ~ metrized by X(−t), then the modified curvature function would be −κg (t). Thus, the sign of the curvature depends on what one might call the “orientation” of the curve, a notion ultimately arbitrary in this case. Excluding this technicality, curvature has a physical interpretation that one can eyeball on particular curves. If a curve is almost a straight line, then the curvature is close to 0, but if a regular curve bends tightly along a certain section, then the curvature is high (in absolute value). Of particular interest are points where the curvature reaches a local extremum. ~ : I → R2 Definition 1.3.6. Let C be a regular curve parametrized by X with curvature function κg (t). A vertex of the curve C is a point ~ 0 ) where the corresponding curvature function κg (t) attains P = X(t an extremum. By the First Derivative Test, extrema of κg (t) occur at t = t0 , if changes sign through t0 . This obviously must occur where κ0g (t0 ) = 0 but the converse is not true. ~| From the above discussion, one should notice that since {T~ , U ~ form an orthonormal basis for all t, every higher derivative of X(t), κ0g (t)

1.3. Curvature

31

~ . The if it exists, can be expressed as a linear combination of T~ and U (n) ~ ~ ~ components of X (t) in terms of T and U involve sums of products ~ is parametrized of derivatives of s(t) and κg (t). Furthermore, if X (n) ~ by arc length, then the coefficients of X (s) only involve sums of powers of derivatives of κg (s).

Problems ~ 1.3.1. Find the curvature function of the curve X(t) = (tm , tn ) for t ≥ 0. ~ 1.3.2. Find the curvature function of the ellipse X(t) = (a cos t, b sin t). ~ 1.3.3. Find the curvature function of the cycloid X(t) = (t−sin t, 1−cos t). ~ 1.3.4. Find the curvature function of the curve X(t) = (ecos t , esin t ) with t ∈ [0, 2π]. 1.3.5. Calculate the curvature function of the flower curve parametrized by ~ X(t) = (sin(nt) cos t, sin(nt) sin t). ~ 1.3.6. Consider the curve X(t) = (t2 , t3 − at), where a is a real number. Calculate the curvature function κg (t) and determine where κg (t) has extrema. 1.3.7. Consider the graph of a function y = f (x) parametrized as a curve ~ by X(t) = (t, f (t)). Find a formula for the curvature. Show that f must be of class C 2 for the curvature to exist and to be continuous. Prove that κg (t) = 0 if t is an inflection point. [Hint: Recall that x0 is the inflection point of a function y = f (x) if f 00 (x) changes sign at x0 .] 1.3.8. Find an equation in f (t) that describes where the vertices occur for ~ a function graph X(t) = (t, f (t)). Find an example of a function 0 graph where κg (t) = 0 but f 0 (t) 6= 0 and vice versa. 1.3.9. Prove that the graph of a polar function r = f (θ) at angle θ has curvature 2f 0 (θ)2 + f (θ)2 − f (θ)f 00 (θ) κg (θ) = . (f 0 (θ)2 + f (θ)2 )3/2 1.3.10. Find the vertices of an ellipse with half-axes a 6= b and calculate the curvature at those points. 1.3.11. Calculate the curvature function for all the Lissajous figures: ~x(t) = (cos(mt), sin(nt)), with t ∈ [0, 2π].

32

1. Plane Curves: Local Properties

1.3.12. Prove by direct calculation that the following formula holds for the ~ ellipse X(t) = (a cos t, b sin t): 2π

Z

κg ds = 2π. 0

[Hint: Use a substitution involving tan θ =

a b

tan t.]

1.3.13. Calculate the curvature function for the cardioid: ~x(t) = ((1 − cos t) cos t, (1 − cos t) sin t) . ~ 1.3.14. Find the vertices of the exponential curve X(t) = (t, et ) for t ∈ R. Interpret this result in terms of radius of curvature on the curve. ~ : I → R2 be a parametrized curve and f : J → I a surjective 1.3.15. Let X ~ ◦ f a regular reparametrization of function so that f makes ξ~ = X ~ ~ 0 ). Prove that ~ X. Call t0 = f (u0 ) so that ξ(u0 ) = X(t ( κg,X~ (t0 ) =

κg,ξ~(u0 ) −κg,ξ~(u0 )

if f 0 (u0 ) > 0, if f 0 (u0 ) < 0.

1.3.16. Let I be a closed and bounded interval. We define a parallel curve ~ : I → R2 as a curve that can be paramto a parametrized curve X ~ ~ (t), where ε is a real number. Suppose + εU etrized by ~γε (t) = X(t) ~ that X is a regular curve of class C 2 and assume ε 6= 0. Prove that ~γε is regular if and only if 1ε ∈ / [κm , κM ], where κm = min{κg (t)} t∈I

and

κM = max{κg (t)}. t∈I

1.3.17. Let C be a curve in R2 defined as the solution to an equation F (x, y) = 0. Use implicit differentiation to prove that at any point on this curve, the curvature of C is given by κg =

Fxx Fy2 − 2Fxy Fx Fy + Fyy Fx2 . (Fx2 + Fy2 )3/2

1.3.18. Find the vertices of the curve given by the equation x4 + y 4 = 1.

1.4. Osculating Circles, Evolutes, and Involutes

y

33

tan x sin x π 2

x

Figure 1.7. Functions sin x and tan x have contact order 1 at the origin.

1.3.19. Pedal Curves. Let C be a regular curve and A a fixed point in the plane. The pedal curve of C with respect to A is the locus of points of intersection of the tangent lines to C and lines through A ~ perpendicular to these tangents. Given the parametrization X(t) of a curve C, provide a parametric formula for the pedal curve to C with respect to A. Find an explicit parametric formula for the pedal curve of the unit circle with respect to (1, 0).

1.4 Osculating Circles, Evolutes, and Involutes Classical differential geometry introduces the notion of order of contact to measure the degree of intersection between two curves or surfaces at a particular intersection point. See Struik in [32, p. 23] for this classical definition. In this text, we provide an alternate but equivalent definition that is more relevant to our approach to always describing curves via a parametrization. As a motivating example, consider the real functions f (x) = sin x and g(x) = tan x near x0 = 0. The graphs of f and g intersect at (0, 0). However, we notice in Figure 1.7 that not only do they intersect, but they share the same tangent line. Even further, consider their sixth-order Taylor polynomial at 0: 1 sin x ≈ x − x3 + 6 1 3 tan x ≈ x + x + 3

1 5 x 120 2 5 x . 15

34

1. Plane Curves: Local Properties

In particular, both sin x and tan x have the same second-order Taylor polynomial at x0 = 0, but their third-order Taylor polynomials differ. We say that f (x) and g(x) have contact of order 2 at x0 = 0. Definition 1.4.1. Two functions f (x) and g(x) defined on a neighbor-

hood of x0 have contact of order k if for i = 0, 1, . . . , k, the derivatives f (i) (x0 ) and g (i) (x0 ) exist and f (i) (x0 ) = g (i) (x0 ). We say that f and g have contact of strict order k at x0 if they have contact of order k but do not have contact of order k + 1. Note that a simple monomial function f (x) = xn with the x-axis at (0, 0) has contact of strict order n − 1. This is because f (i) (0) = 0 for i = 0, 1, . . . , n − 1 but f (n) (0) = n!, which is not equal to 0. As we attempt to generalize this notion to order of contact between two parametrized curves in the plane, we are immediately met with two obstacles. First, in general the parameter for one parametrized curve has no connection, geometric or otherwise, to the parameter for the second curve. In the definition of order of contact for real-valued functions over the reals, the parameter x in an open interval of R is used for both functions f (x) and g(x). Secondly, parametrized curves are usually not described as functions with respect to some frame. Consequently, we must make a choice of parameter on both curves simultaneously that has geometric meaning. The typical approach is to use one of the curves parametrized by arc length as a reference. ~ Definition 1.4.2. Suppose C1 : α ~ (t) and C2 : β(u) are two parametrized curves that intersect at a point P , which corresponds to where t = t0 and u = u0 . Reparametrize C1 by arc length and let s0 be such that P = α ~ (s0 ). Let u(s) be the function such that the projec~ tion of α ~ (s) onto C2 is located at β(u(s)). Then we say that C1 and C2 have contact of order n at P if they are of class C n over an open interval around P and di ~ α ~ (s0 ) = i β(u(s)) ds s0 (i)

for all 0 ≤ i ≤ n. Furthermore, C1 and C2 have contact of strict order n if they do not also have contact of order n + 1.

1.4. Osculating Circles, Evolutes, and Involutes

The intuitive picture for order of contact indicates that two intersecting curves with contact of order 1 have the same tangent line at the intersection point. In particular, we leave it as an exercise to prove the fact that a curve and its tangent line have order of contact 1. In contrast, an intersection point between two curves with contact of strict order 0 is said to be a transversal intersection. Note that since the concept of order of contact refers to orders of differentiation, if two curves are not both of class C n near a point P it does not make sense to discuss contact of order n. At first glance, this definition of order of contact seems asymmetrical. However, it is possible to prove that using β~ as the reference curve and the arc length of C2 as the reference parameter is equivalent to Definition 1.4.2. (In fact, for a given nonnegative integer n, on the set of parametrized curves in the plane, the relation of contact of order n is an equivalence relation, i.e., is reflexive, symmetric, and transitive. In contrast, the relation of contact of strict order n is reflexive and symmetric but not necessarily transitive.) ~ : I → R2 , and Definition 1.4.3. Let C be a curve parametrized by X let t0 be a regular value of the curve. The osculating circle to C at ~ 0) . the point t0 is a circle that has contact of order 2 with C at X(t ~ : I → R2 be a parametrized curve and t0 Proposition 1.4.4. Let C : X

~ a regular value. Suppose that X(t) is twice differentiable at t0 and that κg (t0 ) 6= 0. Then, ~ 0 ); 1. There exists a unique osculating circle to C at X(t 2. It is given by the following vector function: ~ 0 )+ ~γ (t) = X(t

 1 ~ 1  ~ (t0 ) . U (t0 )+ (sin t)T~ (t0 ) − (cos t)U κg (t0 ) κg (t0 )

~ is parameProof: Without loss of generality, we can assume that X trized by arc length s and that we are looking for the osculating circle at s = 0. Call ~γ (u) the parametrization of the proposed osculating circle, which must have the form ~γ (u) = (a + R cos(u), b + R sin(u)).

35

36

1. Plane Curves: Local Properties

For ease of the proof, we will allow R to be any nonzero real number. We assume that u = u0 at the point of contact and that near u0 , ~ the function u(s) gives the projection of X(s) onto the curve ~γ . By Definition 1.4.2, in order for there to exist an osculating circle, the parameters a, b, and R and the function u(s) must satisfy ~ X(0) = ~γ (u(0)),

~ 0 (0) = d ~γ (u(s)) , X ds 0

2 ~ 00 (0) = d ~γ (u(s)) , X ds2 0

which in coordinates is equivalent to a + R cos(u(0)) = x(0),

(1.13)

b + R sin(u(0)) = y(0),

(1.14)

0

0

(1.15)

0

0

(1.16)

−Ru (0) sin(u(0)) = x (0), Ru (0) cos(u(0)) = y (0), 00

0

2

00

(1.17)

00

0

2

00

(1.18)

−Ru (0) sin(u(0)) − Ru (0) cos(u(0)) = x (0), Ru (0) cos(u(0)) − Ru (0) sin(u(0)) = y (0).

~ Since X(s) is parametrized by arc length, x0 (0)2 +y 0 (0)2 = 1, from which we conclude that |u0 (0)| = R1 . Without loss of generality, we can assume that u0 (0) and R have the same sign, so that u0 (0) = R1 . Since the curve is parametrized by arc length, x0 (s)2 + y 0 (s)2 = 1 for all s. Taking a derivative of this equation with respect to s and evaluating at s = 0, we deduce that x0 (0)x00 (0) + y 0 (0)y 00 (0) = 0. This relation, along with Equations (1.15) through (1.18) leads to u00 (0) = 0. We obtain the value of R by noting that after calculation 1 . R

κg (0) = x0 (0)y 00 (0) − x00 (0)y 0 (0) = R2 (u0 (0))3 = Using Equations (1.15) and (1.16), we find that ~, (cos(u(0)), sin(u(0))) = (y 0 (0), −x0 (0)) = −U

~ is parametrized by arc length. Thus the center of the circle since X ~γ is ~ (a, b) = (x(0), y(0)) − R (cos(u(0)), sin(u(0))) = X(0) +

1 ~ U (0). κg (0)

1.4. Osculating Circles, Evolutes, and Involutes

Finally, a quick check that we leave for the reader is to show that given the above values for a, b, and R, Equations (1.13) through ~ (1.18) are redundant given the fact that X(s) is parametrized by arc length. These results prove part 1 of the proposition. ~ is parametrized by arc Part 2 follows easily from part 1. Since X ~ length, the unit tangent vector is just T (0) = (x0 (0), y 0 (0)), giving ~ (0) = (−y 0 (0), x0 (0)). In the solutions for Equations (1.13) also U and (1.14), we see that the center of the osculating circle is at ~ ~ (0). (a, b) = (x(0) − R cos(τ (0)), y(0) − R sin(τ (0))) = X(0) + RU Furthermore, for any curve parametrized by arc length, κg (s) = x0 (s)y 00 (s) − x00 (s)y 0 (s). Thus, R1 = κg (0).  In light of Proposition 1.4.4 and Example 1.3.3, the curvature ~ has a nice physical interpretafunction κg (t) of a plane curve X tion, namely, the reciprocal of the radius of the osculating circle. A higher order of contact indicates a better geometric approximation, and hence, since there is a unique osculating circle, it is, in a geometric sense, the best approximating circle to a curve at a point. Furthermore, in Problem 1.4.6 of this section, we prove that to a ~ at t = t0 , there does not necessarily exist a touching circle curve X with order of contact greater than 2. Thus, the curvature is the inverse of the radius of the best approximating circle to a curve at a point. (From a physics point of view, we obtain an additional confirmation of the above interpretation by considering the units of the curvature function. If we view the coordinate functions x(t) and y(t) with the unit of meters and t in any unit, Equation (1.12) gives the unit of 1/meter for κg (t).) ~ be a parametrized curve, and let t = t0 be a Definition 1.4.5. Let X regular point that satisfies the conditions in Proposition 1.4.4. The center of the osculating circle at t = t0 is called the center of curvature. Definition 1.4.6. The evolute of a curve C is the locus of the centers

of curvature.

37

38

1. Plane Curves: Local Properties

~ : I → R2 be a regular parametric plane curve Proposition 1.4.7. Let X

that is of class C 2 , i.e., has a continuous second derivative. Let I 0 be a subinterval of I over which κg (t) 6= 0. Then over the interval ~ has the following parametrization: I 0 , the evolute of X ~ ~ E(t) = X(t) +

1 ~ U (t). κg (t)

Example 1.4.8. Consider the parabola y = x2 with parametric equa-

~ tions X(t) = (t, t2 ). We calculate the curvature with the following steps: ~ 0 (t) = (1, 2t), X ~ 00 (t) = (0, 2), X p s0 (t) = 1 + 4t2 , 2 . κg (t) = (1 + 4t2 )3/2 Thus, the parametric equations of the evolute are 1 1 ~ (−2t, 1) E(t) = (t, t2 ) + (1 + 4t2 )3/2 √ 2 1 + 4t2   1 = −4t3 , + 3t2 . 2 The Cartesian equation for the evolute of the parabola is then y=

x 2/3 1 + 3 . 2 4

~ A closely related curve to the evolute of X(t) is the involute, though we leave the exact nature of this relationship to the problems. The involute is defined as follows. ~ : I → R2 be a regular parametrized curve. Definition 1.4.9. Let X

~ any parametrized curve ~ι such that for all We call an involute to X ~ at t at a right angle. t ∈ I, ~ι(t) meets the tangent line to X

1.4. Osculating Circles, Evolutes, and Involutes

39

~ E y

x Figure 1.8. Evolute of a parabola.

For all t ∈ I, the point on the involute ~ι(t) is on the tangent ~ at t, so it is natural to write the parametric equations line to X ~ as ~ι(t) = X(t) + λ(t)T~ (t). Since we will wish to calculate ~ι0 (t), which involves the derivative T~ 0 (t), we must assume that the curve ~ X(t) is of class C 2 , i.e., with a second derivative that exists and is continuous. Definition 1.4.9 requires that the vector ~ι0 (t) be in a perpendicular direction to T~ (t), so we have   ~ T~ · ~ι0 = 0 =⇒ T~ · s0 T~ + λ0 T~ + λs0 κg U =⇒ s0 + λ0 = 0 =⇒ λ(t) = C − s(t), ~ is a regular where C is some constant of integration. Therefore, if X 2 parametrized curve of class C , then the formula for the involute is ~ ~ι(t) = X(t) + (C − s(t))T~ (t).

Problems 1.4.1. Prove the claim that the tangent line to a parametrized curve at a regular point has contact of order 1. ~ 1.4.2. Consider the cubic curve X(t) = (t, t3 ) defined over R. ~ at t = 1. (a) Find the osculating circle of X ~ (b) Determine the parametric equations for the evolute of X.

40

1. Plane Curves: Local Properties

~ be a regular parametrized curve and let t0 be an inflection 1.4.3. Let X point, i.e. a point where κg (t) = 0. Prove that the tangent line to ~ X(t) at t = t0 has contact of order 2. ~ 1.4.4. Prove that the evolute of the ellipse X(t) = (a cos t, b sin t) has parametric equations  2    2  a − b2 b − a2 3 3 ~γ (t) = cos t, sin t . a b ~ 1.4.5. Consider the catenary given by the parametric equation X(t) = (t, cosh t) for t ∈ R. (a) Prove that the curvature of the catenary is κg (t) =

1 . cosh2 t

(b) Prove that the evolute of the catenary is ~ E(t) = (t − sinh t cosh t, 2 cosh t). 1.4.6. Prove that if the osculating circle to a regular parametrized curve C at a point P has contact of order 3, then P is a vertex of C. Give an example where this does not occur, thereby proving that at a regular point on a parametrized curve, there does not necessarily exist a circle with contact of order 3. 1.4.7. Prove that an osculating circle to a curve C at a point P that is not a vertex lies on either side of the curve. [See Problem 1.4.6.] ~ : I → R2 be a parametrized curve that does not have any 1.4.8. Let X inflection points (i.e., κg (t) 6= 0 for all t ∈ I). Prove that the evolute ~ 0 ) is a vertex of the curve has a critical point at t0 if and only if X(t of the curve. 1.4.9. Let ~x(t) = (t, t2 ) be the parabola. Define a new curve ~ι as the involute of ~x such that ~ι(0) = ~x(0) = (0, 0). Calculate parametric equations for ~ι(t). [Hint: This will involve an integral.] 1.4.10. (*) Continuation of the last problem: Calculate parametric equations for the evolute to ~ι.

1.5 Natural Equations An isometry of the plane is any function F : R2 → R2 such that for any two points p~, ~q ∈ R2 , the distance between them is preserved, i.e., k~q − p~k = kF (~q) − F (~ p)k.

1.5. Natural Equations

41

A well-known result about isometries is that F is an isometry if and only if ~ F (~v ) = A~v + C, ~ is any constant vector. where A is any 2×2 orthogonal matrix and C Isometries of the plane include rotations, translations, reflections, and glide reflections. It is also not hard to show that a composition of isometries is again an isometry. The condition of orthogonality A> A = I implies that det(A) = ±1. Consequently, isometries come in two flavors depending on the sign of the determinant of A. If det(A) = 1, we call the isometry a positive isometry. These include rotations, translations, and compositions thereof. If det(A) = −1, we call the isometry a negative isometry. These include reflections and glide reflections. Positive isometries are also called rigid motions because any shape (imagining it to be a solid physical object) can be moved from its original to its image under a positive isometry without bringing the shape out of the plane. ~ : I → R2 with Recall now that for a parametrized curve X ~ X(t) = (x(t), y(t)), the curvature function is given by κg (t) =

x0 (t)y 00 (t) − x00 (t)y 0 (t) (x0 (t)2 + y 0 (t)2 )3/2

.

Since any rigid motion does not stretch distances, such a transformation should not distort a plane curve. Therefore, as one might expect, the curvature is preserved under rigid motions, a fact that we now prove. ~ : I → R2 be a regular plane curve that is of Theorem 1.5.1. Let X class C 2 . Let F : R2 → R2 be a rigid motion of the plane given by ~ where A is a rotation matrix and C ~ is any vector in F (~v ) = A~v + C, 2 ~ is a regular parametrized curve R . The vector function ξ~ = F ◦ X 2 that is of class C , and the curvature function κ ¯ g (t) of ξ~ is equal to ~ the curvature function κg (t) of X. Proof: A rotation matrix in R2 is of the form   a −b A= , b a

42

1. Plane Curves: Local Properties

~ = (e, f ). We can write the parametric where a2 + b2 = 1. Let C ~ equation for ξ as ~ = (ax(t) − by(t) + e, bx(t) + ay(t) + f ). ξ(t) Then ξ~0 (t) = (ax0 (t) − by 0 (t), bx0 (t) + ay 0 (t)), ξ~0 (t) = (ax00 (t) − by 00 (t), bx00 (t) + ay 00 (t)), and therefore the curvature function of ξ~ is

κ ¯ g (t) =

(ax0 (t) − by 0 (t))(bx00 (t) + ay 00 (t)) − (bx0 (t) + ay 0 (t))(ax00 (t) − by 00 (t)) ((ax0 (t) − by 0 (t))2 + (bx0 (t) + ay 0 (t))2 )3/2

=

abx0 x00 + a2 x0 y 00 − b2 y 0 x00 − aby 0 y 00 − abx0 x00 + b2 x0 y 00 − a2 y 0 x00 + aby 0 y 00 (a2 (x0 )2 − 2abx0 y 0 + b2 (y 0 )2 + b2 (x0 )2 + 2abx0 y 0 + a2 (y 0 )2 )3/2

=

x0 y 00 − y 0 x00 = κg (t). ((x0 )2 + (y 0 )2 )3/2



The curvature function is invariant under any positive isometry, i.e., a composition of rotations and translations. Furthermore, in Problem 1.3.15, we saw that the curvature function is invariant under any regular reparametrization, except up to a sign that depends on “the direction of travel” along the curve. Consequently, |κg | is a geometric invariant that only depends on the shape of the curve at a particular point and not how the curve is parametrized or where the curve sits in R2 . It is natural to ask whether a converse relation holds, namely, whether the curvature function is sufficient to determine the parametrized curve up to a positive plane isometry. As posed, the question is not well defined since a curve can have different parametrizations. However, we can prove the following fundamental theorem. Theorem 1.5.2 (Fundamental Theorem of Plane Curves). For a given piece-

wise continuous function κg (s), there exists a regular curve of class ~ : I → R2 with the curvature C 2 parametrized by arc length by X function κg (s). Furthermore, the curve is uniquely determined up to a rigid motion in the plane.

1.5. Natural Equations

43

~ Proof: If a regular curve X(s) is parametrized by arc length, then the curvature formula is κg (s) = x0 (s)y 00 (s) − y 0 (s)x00 (s). The proof of this theorem consists of exhibiting a parametrization that satisfies this differential equation. For a regular curve of class C 2 that is parametrized by arc length, ~ 0 = T~ , and we can view T~ as a vector function of class C 1 we have X ~ into the unit circle. Therefore, from the interval of definition of X T~ = (cos(θ(s)), sin(θ(s)))

(1.19)

~ , and for some continuous function θ(s). However, T~ 0 (s) = κg (s)U Equation (1.19) show that ~ = θ0 (s) (− sin(θ(s)), cos(θ(s))) . κg (s)U Thus, we deduce that κg (s) = θ0 (s). The above remarks lead to the following result. Suppose we are given the curvature function κg (s). Performing two integrations, we ~ see that the only curves X(s) with curvature function κg (s) must be Z  Z ~ X(s) = cos(θ(s)) ds + e, sin(θ(s)) ds + f , (1.20) where

Z θ(s) =

κg (s) ds + θ0 ,

(1.21)

and where θ0 , e, and f are constants of integration. Note that the theorem requires κg (s) to be piecewise continuous in order to be integrable. Furthermore, the trigonometric addition formulas show ~ by a rotation of θ0 and the that a nonzero constant θ0 changes X nonzero constants e and f correspond to a translation along the ~ = (e, f ). vector C  Because of the geometric nature of the curvature function κg (s) (with respect to arc length) and because there exists a unique plane curve (up to a rigid motion) for a given curvature function, we call κg (s) the natural equation of its corresponding curve. The Fundamental Theorem of Plane Curves is surprising because, a priori, we expect a curve to require two functions (coordinate functions) to define it. However, only one function, κg (s), is required to uniquely define the shape of a curve.

44

1. Plane Curves: Local Properties

~ If we assume that X(s) is of class C ∞ and is equal to its power series in an open interval around s = 0, then we can see why Theorem 1.5.2 holds using Taylor series. (This condition is called real analytic at s = 0.) ~ : I → R2 be a real analytic curve parametrized by arc Let X length and assume without loss of generality that 0 ∈ I. We can ~ about 0 to get expand the Taylor series of X 2 3 ~ 00 (0) + s X ~ 000 (0) + · · · ~ ~ ~ 0 (0) + s X X(s) = X(0) + sX 2! 3!

.

~ 0 (s) = However, since the curve is parametrized by arc length, X ~ 00 (s) = κg (s)U ~ (s), X ~ 000 (s) = κ0 (s)U ~ (s) − (κg (s))2 T~ (s), and T~ (s), X g so on. The first few terms look like     1 2 1 0 1 ~ (0). ~ ~ s + κg (0)s3 + · · · U X(s) = X(0) + s − κg (0)2 s3 + · · · T~ (0) + 6 2 6 ~ (s) is just a rotation of T~ (s) by Thus, since the normal vector U π 2 , given a function κg (s), once one chooses for initial conditions ~ the point X(0) and the direction T~ (0), the Taylor series is uniquely determined. The intersection of the intervals of convergence of the ~ (0) components is a new two Taylor series in the T~ (0) and the U interval J that contains s = 0. It is possible that J could be trivial, J = {0}, but if it is not, then the Taylor series uniquely defines ~ X(s) over J. Choosing a different T~ (0) amounts to a rotation of ~ the curve in the plane, and choosing a different X(0) amounts to a translation. Therefore, we see again that making different choices for the initial conditions corresponds to a rigid motion of the curve in the plane. For even simple functions for κg (s), it is often difficult to use the approach in the proof of Theorem 1.5.2 to explicitly solve for ~ X(s). However, using a computer algebra system (CAS) with tools for solving differential equations, it is possible to produce a picture of curves that possess a given curvature function κg (s). One can create a numerical solution using the solution in Equation (1.20) with (1.21). As an equivalent technique, using a CAS, one can solve

1.5. Natural Equations

45

Figure 1.9. Curve with κg (s) = 1 + 2/(1 + s2 ).

Figure 1.10. Curve with κg (s) = 2 + sin s.

the system of differential equations  0  x (s) = cos(θ(s)), y 0 (s) = sin(θ(s)),   0 θ (s) = κg (s), and only plot the solution for the pair (x(s), y(s)). A choice of initial conditions determines the position and orientation of the curve in the plane. Figures 1.9 and 1.10 give a few interesting examples of parametric curves with a given κg (s). The code in Maple (a common computer algebra system) for Figure 1.9 is the following. > with(DEtools): > with(plots): > kappa:=s->1+2/(1+s^2); > sys := D(x)(s)=cos(th(s)), D(y)(s)=sin(th(s)), D(th)(s)=kappa(s); > DEplot([sys],[x(s),y(s),th(s)], s=-10..10, [[x(0)=0,y(0)=0,th(0)=0]], scene=[x(s),y(s)], stepsize=0.01, iterations=10, linecolor=black, scaling=constrained, axes=normal);

Problems 1.5.1. Suppose a curve has κg (s) = A. Prove by directly expanding the Taylor series that such a curve is a circle.

46

1. Plane Curves: Local Properties

1.5.2. Suppose a curve has κg (s) = 2As. Prove by directly expanding the ~ ~ 0 (0) = (1, 0), Taylor series that if such a curve has X(0) = ~0 and X then such a curve is  Z s Z s ~ cos(As2 ) ds, sin(As2 ) ds . X(s) = 0

0

1 1.5.3. Find parametric equations for a curve satisfying κg (s) = 1+s 2 by direct integration, following the method in the proof of Theorem 1.5.2. Show that this curve corresponds to the catenary y = cosh x.

Investigative Projects Project I. What can you discover about the properties of the pedal curve based on a starting curve and a given point? Project II. Since κg (t) is the inverse of the radius of curvature, we feel the curvature function when we drive. More precisely, from 2 the centripetal acceleration formula vR where v is the speed, if we travel in a car along a curve at constant speed, then the centripetal acceleration is proportional to the curvature function. In the proof of Theorem 1.5.2, we pointed out that κg (s) = θ0 (s) for some angle of direction θ(s). Use this to relate the curvature to the position of the steering wheel of a car. Then consider the problem of turning a corner. Turning a corner means starting on a straight line, turning the steering wheel for a bit and then straightening out so that the car is now going on a straight line that is perpendicular to the original line of travel. Discuss curvature functions that make a reasonably good turn on normal streets.

CHAPTER 2

Plane Curves: Global Properties

Most of the properties of curves we have studied so far are called local properties. By definition, a local property of a curve (or surface) is a property that is related to a point on the curve based on information contained just in a neighborhood of that point. By contrast, global properties concern attributes about the curve taken as a whole. In Chapter 1, the arc length of a curve was the only notion introduced that we might consider a global property. Analytically speaking, local properties of a curve at a point involve derivatives of the parametric equations while global properties deal with integration along the curve and topological properties and geometric properties of how the curve lies in R2 . Some of the proofs of global properties rely on theorems from topology. These are supplied concisely in the appendix on topology in . (The interested reader is encouraged to consult some references on basic topology, such as Gemignani  or Armstrong .)

2.1 Basic Properties Definition 2.1.1. A parametrized curve C is called closed if there ex-

~ : [a, b] → R2 of C such that X(a) ~ ~ ists a parametrization X = X(b). A closed curve is of class C k if, in addition, all the (one-sided) deriva~ at a and at b are equal of order i = 0, 1, . . . , k; in other tives of X ~ 0 (a) = X ~ 0 (b), X ~ 00 (a) = X ~ 00 (b), words, if as one-sided derivatives X ~ (k) (a) = X ~ (k) (b). and so on up to X We recall that the left-derivative (resp. right-derivative) of a function f at a is the limit   f (a + h) − f (a) f (a + h) − f (a) resp. lim . lim h h h→0− h→0+

47

48

2. Plane Curves: Global Properties

The notion of left- and right-sided derivatives of real-valued functions naturally extends to parametrized curves. ~ in the above definition The conditions on the derivatives of X seem awkward but they attempt to establish the fact that the vector ~ behaves identically at a and at b. A more topological function X approach involves using a circle S1 rather than an interval as the ~ We define a topological circle S1 as the domain for the map X. set [0, 1] with the points 0 and 1 identified. The topology of S1 is the identification topology, which in this case means that an open neighborhood U of p contains, for some ε, the subsets ( (p − ε, p + ε) with ε < min{p, 1 − p}, if p 6= 0, [0, ε) ∪ (1 − ε, 1], if p = 0 (= 1). Then we say that a curve C is closed if there exists a continuous surjective (onto) function ϕ : S1 → C. Definition 2.1.2. A curve C is called simple if there exists a parametr-

~ : I → R2 of C such that X ~ is an injective (i.e., one-to-one) ization X function. A closed curve C is called simple if there exists a parame~ : [a, b] → R2 of C such that X(t ~ 1 ) = X(t ~ 2 ) with t1 < t2 trization X only when t1 = a and t2 = b. Using the language associated to a topological circle, we say that a closed curve C is simple if there exists a bijection ϕ : S1 → C that is continuous and such that ϕ−1 is also continuous. If a curve is not simple, we intuitively think of the curve as intersecting itself. Using parametrizations, we can give the concept of a self-intersection a precise definition. Definition 2.1.3. A curve C is said to have a self-intersection at a point

~ : I → R2 of C, there exist P ∈ C if for every parametrization X ~ 1 ) = X(t ~ 2) = P . t1 6= t2 that are not endpoints of I such that X(t ~ : Proposition 2.1.4. Let C be a closed curve with parametrization X [a, b] → R2 . Then C is bounded as a subset of R2 . ~ : [a, b] → R2 be a parametrization of C with coordiProof: Let X nate functions x(t) and y(t). By the Extreme Value Theorem, since

2.1. Basic Properties

49

x(t) and y(t) are continuous functions over an interval [a, b], then they respectively attain extrema xmin minimum and xmax maximum of x(t) and ymin minimum and ymax maximum of y(t). The param~ lies entirely in the rectangle xmin ≤ x ≤ xmax and etrized curve X ymin ≤ y ≤ ymax . Therefore, the curve lies inside a disc centered around the origin. More precisely, let Mx = max{|xmin |, |xmax |}

and

My = max{|ymin |, |ymax |}.

and

y(t)2 ≤ My2 .

Then for all t ∈ [a, b], x(t)2 ≤ Mx2 Thus, ~ kX(t)k ≤

q

Mx2 + My2 ,

and so the curve C is contained in a disk of finite radius. Hence, C is bounded.  One of the most fundamental properties of global geometry of plane curves is that a simple, closed plane curve C separates the plane into two open connected components, each with the common boundary of C. Theorem 2.1.5 (Jordan Curve Theorem). Let C be a simple closed curve

in R2 . Then C separates the plane into precisely two components W1 and W2 such that R2 − C = W1 ∪ W2 and W1 ∩ W2 = ∅. One component is bounded and the other is unbounded. This intuitive fact turns out to be rather difficult to prove. Furthermore, a rigorous proof requires a precise definition of component. We refer the reader to Munkres  for a detailed discussion. In [24, Section A.4], there is a more readable proof of the weaker statement that only assumes that the curve is regular. The component of R2 − C that is bounded is called the interior of C and the component of C that is unbounded is called the exterior. We remind the reader of the following theorem from multivariable calculus, which one may view as a global property of plane curves.

50

2. Plane Curves: Global Properties

Theorem 2.1.6 (Green’s Theorem). Let C be a simple, closed, piecewise

regular, positively oriented plane curve with interior region R, and let F~ = (P (x, y), Q(x, y)) be a differentiable vector field. Then,  ZZ  Z ∂Q ∂P − dx dy. P dx + Q dy = ∂x ∂y C R We remind the reader that a curve is positively oriented if as an object travels along the curve, the interior is to the left. With unit tangent and unit normal vectors, we can say that C is positively ~ : [a, b] → R2 such that U (t) oriented if it is parametrized by X points toward the interior of the curve. We also remind the reader ~ : [a, b] → R2 is a regular parametrization for C with X(t) ~ that if X = (x(t), y(t)), then Z Z ~ P dx + Q dy = F~ · dX C

C b

Z =

~ 0 (t) dt F~ (x(t), y(t)) · X

a

Z =

b

P (x(t), y(t))x0 (t) + Q(x(t), y(t))y 0 (t) dt.

a

Corollary 2.1.7. Let C be a simple closed regular plane curve with in-

terior region R. Then the area A of R is Z Z Z 1 A= x dy = − y dx = −y dx + x dy. 2 C C C

(2.1)

Proof: In each of these integrals, one simply chooses a vector field F~ = (P, Q) such that ∂Q ∂P − = 1. ∂x ∂y For the three integrals, these are, respectively, F~ = (0, x), F~ = (−y, 0), and F~ = 12 (−y, x). Then apply Green’s Theorem.  ~ Example 2.1.8. Consider the ellipse given by X(t) = (a cos t, b sin t) for 0 ≤ t ≤ 2π. The techniques from introductory calculus used to calculate area would lead us to evaluate the integral Z a r x2 A=4 b 1 − 2 dx. a 0

2.1. Basic Properties

51

Though one can calculate this by hand, it is not simple. Much easier would be to use Green’s Theorem, which gives Z 2π Z x dy = a cos t b cos t dt A= 0 C Z 2π Z 2π 1 1 2 + cos(2t) dt cos t dt = ab = ab 2 2 0 0 i2π h1 1 = ab t + sin(2t) 2 4 0 = πab.

Problems ~ 2.1.1. Calculate the area of the region enclosed by the cardioid X(t) = ((1 − cos t) cos t, (1 − cos t) sin t). ~ 2.1.2. Use Green’s Theorem to calculate the area of one loop of X(t) = (cos t, sin(2t)). 2.1.3. Consider the function graph of a curve in polar coordinates r = f (θ) ~ parametrized by X(t) = (f (t) cos t, f (t) sin t). Suppose that for 0 ≤ t ≤ 2π, the curve is closed and encloses a region R. Use Equation (2.1) and Green’s Theorem to prove the following area formula in polar coordinates: ZZ Area(R) = r dr dθ. R

2.1.4. The diameter of a curve C is defined as the maximum distance between two points, i.e., diam(C) = max{d(p1 , p2 ) | p1 , p2 ∈ C} . We call a diameter any chord of C whose length is diam(C). Let ~ X(t) be the parametrization of a closed differentiable curve C. Let ~ ~ f (t, u) = kX(t)− X(u)k be the distance function between two points on the curve. (a) Prove that if a chord [p1 , p2 ] of C is a diameter of C, then the line (p1 , p2 ) is simultaneously perpendicular to the tangent line to C at p1 and the tangent line to C at p2 . (b) Under what other situations is the line (p1 , p2 ) simultaneously perpendicular to the tangent line to C at p1 and the tangent line to C at p2 ?

52

2. Plane Curves: Global Properties

~ 2.1.5. Prove that the √ diameter of the cardioid X(t) = ((1 − cos t) cos t, (1 − cos t) sin t) is 3 3/2. [Hint: First show that the maximum distance ~ ~ between X(t) and X(u) occurs when u = 2π − t.] 2.1.6. (*) This problem studies the relation between the curvature κg (s) of a curve C and whether it is closed. (a) Prove that if C is closed, then its curvature κg (s), as a function of arc length, is periodic. (b) Assume that κg (s) is not constant and is periodic with smallest period p. Use the Fundamental Theorem of Plane Curves (Theorem 1.5.2) to prove that if Z p 1 κ(s) ds 2π 0 is an element of Q − Z, i.e., a rational that is not an integer, then C is a closed curve. [This problem is discussed in , though the authors’ use of complex numbers is not necessary for this problem.]

2.2 Rotation Index As a leading example of what we shall term the rotation index of ~ a curve, consider the circle C : X(t) = (R cos t, R sin t) with the defining interval of I = [0, 2π]. A simple calculation shows that for all t ∈ [0, 2π], we have a curvature of κg (t) = R1 . Then it is easy to see that Z Z 2π Z 2π 1 · R dt = 2π. κg ds = κg (t)s0 (t) dt = R 0 0 C On the other hand, suppose that we use the defining interval [0, 2πn] with the same parametrization. In that case, were we to evaluate the same integral, we would obtain Z 1 κg ds = n. 2π C ~ : I → R2 . To Now consider any regular, closed, plane curve X ~ this curve, we associate the unit tangent vector T (t). Placing the base of this vector at the origin, we see that T~ itself draws out a

2.2. Rotation Index

53

curve T~ : I → R2 . It is called the tangential indicatrix of the plane ~ curve X. The tangential indicatrix of a curve lies entirely on the unit circle in the plane but with possibly a complicated parametrization. ~ the tangential indicatrix may change Depending on the shape of X, speed, stop, and double back along a portion of the circle. Nonetheless, we can use the theory of usual plane curves to study the tangential indicatrix. We are in a position to state the main proposition for this section. Proposition 2.2.1. Let C be a closed, regular, plane curve parametrized

~ : I → R2 of class C 2 . Then the quantity by X Z 1 κg ds 2π C is an integer. This integer is called the rotation index of the curve. Proof: Rewrite the above integral as follows: Z Z Z ~ (t)k dt. κg ds = κg (t)s0 (t) dt = κg (t)s0 (t)kU C

I

I

The curvature function κg (t) of plane curves is not always posi~ tive, but since X(t) is of class C 2 , then κg (t) is at least continuous, and the intermediate value theorem applies. Therefore, define two closed subsets (unions of closed subintervals) of the interval I as follows: I + = {t ∈ I : κg (t) ≥ 0}

and

I − = {t ∈ I : κg (t) ≤ 0}.

Using the definition of curvature from Equation (1.10), we split the above integral into the following two parts: Z Z Z 0 ~ κg ds = kT (t)k dt − kT~ 0 (t)k dt. C

I+

I−

However, the two integrals on the right-hand side of the above equation are integrals of the arc length of the tangential indicatrix trav~ is eling in a counterclockwise (resp. clockwise) direction. Since X a closed curve, T~ is as well, and these two integrals represent 2π times the number of times T~ travels around the circle, with a sign indicating the direction. 

54

2. Plane Curves: Global Properties

~ Figure 2.1. The curve X(t) = (cos t, sin(2t)).

~ Example 2.2.2. Consider X(t) = (cos t, sin(2t)) as the paramtrization for the curve C depicted in Figure 2.1. The curvature function is given by κg (t)s0 (t) =

6 cos t + 2 cos(3t) 4 sin t sin(2t) + 2 cos t cos(2t) = . 2 5 − cos(2t) + 4 cos(4t) sin t + 4 cos2 (2t)

~ is Thus, the rotation index n of X 1 n= 2π

Z

1 κg (t)s (t)dt = 2π C 0

Z 0

6 cos t + 2 cos(3t) dt. 5 − cos(2t) + 4 cos(4t)

Since the integrand is periodic of period 2π, using the substitution u = t + π2 and recognizing that we can integrate over any interval of length 2π, we have Z π 6 sin u − 2 sin(3u) 1 n= du. 2π −π 5 + cos(2u) + 4 cos(4u) However, the integrand is now an odd function, so the rotation index ~ is n = 0. of X R The proof of Proposition 2.2.1 analyzed the integral C κg ds as the signed arc length of the tangential indicatrix T~ (t), which one can view as a map from an interval I to the unit circle S1 . However, the

2.2. Rotation Index

55

result of Proposition 2.2.1 follows also from a more general result that we wish to explain in detail here. Some of the concepts below come from topology and illustrate the difficulty of analyzing functions from an interval to a circle. The reader should feel free to either skip the technical details of the proofs in the rest of this section or refer to [24, Appendix A] for background material. Any path f : I → S1 on the unit circle may be described by the angle function ϕ(t), so that f (t) = (cos(ϕ(t)), sin(ϕ(t))) . However, the angle ϕ(t) is defined only up to a multiple of 2π, and hence it need not be a well-defined function, let alone a continuous function. However, since f is continuous, for all t ∈ I there exists an interval (t − ε, t + ε) and a continuous function ϕ(t) ˜ such that for all u ∈ (t − ε, t + ε), ϕ(u) ˜ differs from ϕ(u) by a multiple of 2π. If ϕ˜0 (t) exists, it is a well-defined function, regardless of any choice made in selecting ϕ(t). Finally, if I = [a, b], we define the total angle function related to f as the function Z t ϕ(t) ˜ = ϕ˜0 (u) du + ϕ(a). a

By construction, ϕ(t) ˜ is continuous, satisfies f (t) = (cos(ϕ(t)), ˜ sin(ϕ(t))) ˜ , and keeps track of how many times and in what direction the path f ˜ − ϕ(a) ˜ is called the total angle travels around S1 . The quantity ϕ(b) swept out by f . Definition 2.2.3. Let I be an interval of R. Given a continuous func-

tion f : I → S1 on the unit circle, the lifting of f is the function ϕ(t), ˜ with ϕ(a) chosen so that 0 ≤ ϕ(a) < 2π. The lifting of f is denoted by f˜ to indicate its unique dependence on f . For any continuous function between circles f : S1 → S1 , viewing as an interval [0, 2π] with the endpoints identified, we also view f as a continuous function f : [0, 2π] → S1 . Since f (0) = f (2π) are points on S1 , then f˜(2π) − f˜(0) is a multiple of 2π. This leads to the following definition. S1

56

2. Plane Curves: Global Properties

Definition 2.2.4. Let f : S1 → S1 be a continuous map between circles.

Let f˜ : [0, 2π] → R be the lifting of f . Then the degree of f is the integer  1 ˜ f (2π) − f˜(0) . deg f = 2π Intuitively speaking, the degree of a function f : S1 → S1 between unit circles is how many times around f winds its domain S1 onto its target S1 . Returning to the example of the tangential indicatrix T~ of a regular closed curve C, since C is closed, T~ can be viewed as a function S1 → S1 . ~ : Proposition 2.2.5. Let C be a regular closed curve parametrized by X I → R2 . The rotation index of C is equal to the degree of T~ . Proof: Since T~ is a map T~ : I → S1 , using Equation (2.2), we can write T~ = (cos(ϕ(t)), ˜ sin(ϕ(t))) ˜ , so  ~ (t). ˜ ϕ˜0 (t) cos(ϕ(t)) ˜ = ϕ˜0 (t)U T~ 0 = −ϕ˜0 (t) sin(ϕ(t)), ~ (t), so But T~ 0 (t) = κg (t)s0 (t)U κg (t)s0 (t) = ϕ˜0 (t).

(2.2)

Thus, if we call I = [a, b], we have Z

Z κg ds = C

which concludes the proof.

b

ϕ˜0 (t) dt = ϕ(b) ˜ − ϕ(a), ˜

a



The notion of degree of a continuous function S1 → S1 can be applied in a wider context than can the rotation index of a curve since the latter requires a curve to be regular, while the former concept, as presented here, only requires ϕ˜0 (t) to be integrable. Another closely related formula to calculate the degree of a curve is the following.

2.2. Rotation Index

57

Proposition 2.2.6. Suppose that a function f : S1 → S1 is parametrized

as ~γ : [a, b] → R2 . If ~γ (t) = (γ1 (t), γ2 (t)), then deg f =

1 2π

b

Z a

γ20 (t) 1 dt = − γ1 (t) 2π

b

Z a

γ10 (t) dt. γ2 (t)

Proof: (Left as an exercise for the reader. See Problem 2.2.7.)



In practice, since many functions f : S1 → S1 as described in the above proposition rely on a parametrization that involves cosine and sine functions, it is very common for the interval [a, b] to be [0, 2πn] where n is some positive integer. Among the many uses of the notion of degree, we can make precise the notion of how often a curve turns around a point. Definition 2.2.7. Let C be a closed, regular, plane curve parametrized ~ : I → R2 , and let p~ be a point in the plane. Since C is a closed by X ~ as a function on S1 . We define the winding curve, we may view X number w(p) of C around p~ as the degree of the function f : S1 → S1 that has the parametrization ~γ : IR2 with

~γ (t) =

~ X(t) − p~ . ~ ||X(t) − p~ ||

Proposition 2.2.6 gives us a direct method to calculate the winding number of a curve around a point. Proposition 2.2.8. Let C be a closed, regular, plane curve parametrized

~ : [a, b] → R2 with X(t) ~ by X = (x(t), y(t)), and let p~ be a point in the plane. The winding number of C around p~ is Z a

b

(x(t) − px )y 0 (t) − (y(t) − py )x0 (t) dt. ~ kX(t) − p~k2

Proof: Applying Proposition 2.2.6 to the calculation of the winding number of C around a point p~, we use the parametrization ! ~ y(t) − py X(t) − p~ x(t) − px ~γ (t) = (γ1 (t), γ2 (t)) = = , ~ ~ ~ kX(t) − p~k kX(t) − p~k kX(t) − p~k

58

2. Plane Curves: Global Properties

as a parametrization γ : [a, b] → R2 . We calculate γ20 (t) =

~ ~ ~ 0 (t) · (X(t) ~ y 0 (t)kX(t) − p~k − (y(t) − py ) 12 kX(t) − p~k−1 (2X − p~)) . 2 ~ kX(t) − p~k After simplification this becomes γ20 (t) = Then

y 0 (t)(x(t) − px )2 − (x(t) − px )(y(t) − py )x0 (t) . ~ kX(t) − p~k3

(x(t) − px )y 0 (t) − (y(t) − py )x0 (t) γ20 (t) = ~ γ1 (t) kX(t) − p~k2

and the proposition follows from the result of Proposition 2.2.6.  ~ Example 2.2.9. Consider the ellipse X(t) = (2 cos t, sin t) with 0 ≤ t ≤ 2π and consider the winding number of this ellipse around the point p~ = (1, 0). This is a simple configuration and we expect that the winding number should be 1 (or possibly −1 based on the orientation of the parametrization of the ellipse). By Proposition 2.2.8, the winding number is Z 2π (2 cos t − 1) cos t − (sin t)(−2 sin t) 1 dt 2π 0 (2 cos t − 1)2 + sin2 t Z 2π 2 − cos t 1 dt. = 2π 0 (2 cos t − 1)2 + sin2 t This integral is challenging to evaluate by hand. A computer algebra system evaluates it as exactly 1. This confirms our geometric intuition. In practice, it is often hard to explicitly calculate the winding number of a curve around a point either directly from Definition 2.2.7 or by hand from Proposition 2.2.8. With Proposition 2.2.8 the calculation becomes tractable with a computer algebra system. Interestingly enough, it is usually easy to see what the degree is by plotting out the image ϕ(t). (See Figure 2.2, where in Figure 2.2(b) we have allowed ϕ(t) to come off the circle in order to see its graph more clearly.)

2.2. Rotation Index

59

~ − p~ X(t) ~ X(0) − p~

ϕ(t)

p~

ϕ(0)

O

~ (a) X(t)

(b) ϕ(t)

Figure 2.2. Winding number of 2.

One might expect that the winding number of a curve around a point p depends only on what connected component of the curve’s complement p lies in. Proposition 2.2.10. Let C be a closed, regular, plane curve parame~ : I → R2 , and let p0 and p1 be two points in the same trized by X connected component of R2 − C. The winding number of C around p0 is equal to the winding number of C around p1 .

Proof: Let β : [0, 1] → R2 be a path such that β(0) = p0 , β(1) = p1 , ~ and β(u) 6= X(t) for any u ∈ [0, 1] and t ∈ I. Define the two-variable function H : I × [0, 1] → S1 by H(t, u) =

~ X(t) − β(u) . ~ kX(t) − β(u)k

~ The function H is continuous since X(t) − β(u) is continuous and ~ kX(t) − β(u)k is also continuous and never 0. For all u ∈ [0, 1], we ˜ u) consider H(t, u) to be a function of t from I to S1 and define H(t, ˜ : I × [0, 1] → R. But as its lifting, which is a continuous function H then the function of u given by  1 ˜ ˜ u) H(2π, u) − H(0, 2π

60

2. Plane Curves: Global Properties

is continuous and discrete. Thus, it is constant, and hence    1 ˜ ˜ 1) = 1 H(2π, ˜ ˜ 0) , H(2π, 1) − H(0, 0) − H(0, 2π 2π which means that the winding numbers of C around p0 and around p1 are equal.  The winding number of a regular curve around a point offers a strategy to prove the Jordan Curve Theorem (Theorem 2.1.5) in the case when the curve is regular. The strategy is to show that near a point p on the curve C, points on one side of C have a winding number of 0, while on the other side of C points have a winding number of 1 or −1, depending on the orientation of the parametrization given for C. This establishes that a regular, simple, closed curve has at least two connected components. The proof of the Regular Jordan Curve Theorem given in [24, Section A.4] also establishes a fact that is useful in its own right. ~ : [a, b] → R2 be a simple, closed, regular, Proposition 2.2.11. Let X ~ is ±1. plane curve. The rotation index of X

Problems ~ 2.2.1. Consider the figure-eight curve X(t) = (cos t, sin(2t)) with 0 ≤ t ≤ 2π. Show that the rotation index is 0. [Hint: When performing an appropriate integration, reparametrize by t = u + π2 and change the bounds of integration to [−π, π].] 2.2.2. Show that the lemniscate given by   a cos t a cos t sin t ~ , X(t) = 1 + sin2 t 1 + sin2 t has a rotation index of 0. 2.2.3. Calculate the rotation index of the lima¸con de Pascal given by ~ X(t) = ((1 − 2 cos t) cos t, (1 − 2 cos t) sin t) . 2.2.4. Let ~x(t) = (cos 3t cos t, cos 3t sin t) be the trefoil curve. Show that I = [0, π] is enough of a domain to make ~x into a closed curve. Prove that the rotation index of ~x is 2.

2.3. Isoperimetric Inequality

2.2.5. Let C be a regular plane curve. Suppose that p is a point in the plane such that there exists a ray from p that does not intersect C. Show that the winding number of C around p is 0. 2.2.6. Give an example of a curve C (by either a sketch or parametrization) and a point p~ ∈ R2 such that the winding number of C around p~ is 2 but that every ray based at p~ intersects C in more than two points. 2.2.7. Prove Proposition 2.2.6. 2.2.8. The curve C pictured below has the parametrization ~ X(t) = ((3 + cos(5t)) cos(3t), (3 + cos(5t)) sin(3t)) for t ∈ [0, 2π]. (a) Calculate explicitly the integral formula for the winding number of C around a point p~ as given in Proposition 2.2.8. (b) Use a computer algebra system to calculate this winding number for various points in the plane in all the different connected components of R2 − C. (c) Give an intuitive justification for your results.

2.3 Isoperimetric Inequality With the Jordan Curve Theorem at our disposal, we can use Green’s Theorem to prove an inequality between the length of a simple closed curve and the area contained in the interior. This inequality, called the isoperimetric inequality, is another example of a global theorem since it relates quantities that take into account the entire curve at once.

61

62

2. Plane Curves: Global Properties

Theorem 2.3.1. Let C be a simple, closed, plane curve with length l, and let A be the area bounded by C. Then

l2 ≥ 4πA,

(2.3)

and equality holds if and only if C is a circle. Proof: Let L1 and L2 be parallel lines that are both tangents to C and such that C is contained in the strip between them. Let S1 be a circle that is also tangent to both L1 and L2 such that S1 does not intersect C. We call r the radius of this circle, and we set up the coordinate axes in the plane so that the origin is at the center of S1 , the x-axis is perpendicular to L1 (and L2 ), and the y-axis is parallel to L1 and L2 . See Figure 2.3.

s = s0 s=0

α ~

Figure 2.3. Isoperimetric inequality.

Assume that C is parametrized by arc length α ~ (s) = (x(s), y(s)) and that the parametrization is positively oriented at the points of tangency with L1 and L2 . Also, let’s call s = 0 the parameter location for the tangency point C ∩ L1 and s = s0 the parameter for

2.3. Isoperimetric Inequality

63

C ∩ L2 . We can also assume that S1 is parametrized by ~γ (s), where ~γ (s) is the intersection of ~ (s) if 0 ≤ • the upper half of S1 with the vertical line through α s ≤ s0 ; • the lower half of S1 with the vertical line through α ~ (s) if s0 ≤ s ≤ l. Notice that in this parametrization for the circle, writing ~γ (s) = (¯ x(s), y¯(s)), we have x(s) = x ¯(s). Let’s call A the area enclosed in the curve C, and let A¯ be the area of S1 . Using Green’s Theorem, we have the following formulas for the area of the circle of the interior of the curve: Z l A= xy 0 ds. 0

We would like to apply the same formula with the parametrization ~γ (s) = (¯ x(s), y¯(s)) of the circle. However, the parametrization is in general not simple. On the other hand, using a strategy similar to that in the proof of Proposition 2.2.1, we can also use Green’s Theorem and justify the claim that the area of the circle is Z l x0 y¯ ds. A¯ = πr2 = − 0

Adding the two areas, we get Z l Z lp xy 0 − x0 y¯ ds ≤ (xy 0 − x0 y¯)2 ds A + πr2 = 0 0 Z lp Z lp 2 2 0 2 0 2 ≤ (x + y¯ )((x ) + (y ) ) ds = x ¯2 + y¯2 ds = lr. 0

0

(2.4) We now use the fact that the geometric mean of two positive real numbers is less than the arithmetic mean. Thus, √ 1 1 (2.5) A · πr2 ≤ (A + πr2 ) = lr. 2 2 Squaring both sides and dividing by r2 , we get 4πA ≤ l2 . This proves the first part of Theorem 2.3.1.

64

2. Plane Curves: Global Properties

In order for equality to hold in Equation (2.3), we need equalities to hold in both the inequalities in Equation (2.4). From Equation (2.5), we conclude that A = πr2 and l = 2πr. Furthermore, since A does not change, regardless of the direction of L1 and L2 , neither does the distance r. We also must have (xy 0 − x0 y¯)2 = (x2 + y¯2 )((x0 )2 + (y 0 )2 ) = r2 , y y 0 )2 = 0. However, which, after expanding both sides, leads to (xx0 +¯ 2 2 2 differentiating the relationship x + y¯ = r for all parameter values s, we find that xx0 + y¯y¯0 = 0 for all s. Thus, for all s, we have y 0 (s) = y¯0 (s), and thus y(s) = y¯(s) + D for some constant D. Since by construction x(s) = x ¯(s), then C is a circle – a translate of S1 in the direction (0, D).  Since nothing in the above proof uses second derivatives, we only need the curve α ~ to be C 1 , i.e., that it have a continuous first derivative. Furthermore, one can generalize the proof to apply to curves that are only piecewise C 1 .

2.4 Curvature, Convexity, and the Four-Vertex Theorem In this section, all curves are assumed to be of class C 2 . Definition 2.4.1. A subset of S of Rn is called convex if for all p and q

in S, the line segment [p, q] is a subset of S, i.e., lies entirely inside S. If C is not convex, it is called concave. (See Figure 2.4.)

Figure 2.4. A concave curve.

2.4. Curvature, Convexity, and the Four-Vertex Theorem

With the Jordan Curve Theorem, we can affirm that a simple, closed curve has an interior. Hence, we can discuss convexity properties of a plane curve if we use the following definition. Definition 2.4.2. A closed, regular, simple, parametrized curve C is

called convex if the union of C and the interior of C form a convex subset of R2 . In the context of regular curves, it is possible to provide various characterizations for when a curve is convex based on a curve’s position with respect to its tangent lines. Proposition 2.4.3. A regular, closed, plane curve C is convex if and

only if it is simple and its curvature κg does not change sign. Proof: The proof involves showing that if the curvature function κg (s) with respect to arc length changes sign at s0 , then there exist a value s0 − ε near and before s0 and a value s0 − ε near and after s0 such that the segment connecting the points corresponding to s0 and s0 − ε and the segment connecting the points corresponding to s0 and s0 + ε cannot be in the same connected component of R2 − C. Let α ~ : [0, l] → R2 be a parametrization by arc length with positive orientation for the curve C. Let T~ : [0, l] → S1 be the tangential indicatrix, and let θ : [0, l] → R be the total angle function associated to the rotation index. By Equation (2.2), κg (s) = θ0 (s), so the condition that κg (s) does not change sign is equivalent to θ(s) being monotonic. We first prove that convexity implies that C is simple. This follows from the definition of convexity, which assumes that C possesses an interior and, therefore, that R2 − C has only two components. By the Jordan Curve Theorem, since we already assume C is regular and closed, for it to possess an interior, it must be simple. Assume from now on that C is simple. Suppose that κg (s) changes sign on [0, l]. At a point s = a, define the height function for all s ∈ [0, l] by ~ (a). α(s) − α ~ (a)) · U ha (s) = (~ This measures the height of the point α ~ (s) from the tangent line La ~ (a) being considered the positive direction. The to C at α ~ (a), with U

65

66

2. Plane Curves: Global Properties

derivatives of this height function are ~ (a) = T~ (s) · U ~ (a), h0 (s) = α ~ 0 (s) · U ~ (s) · U ~ (a). h00 (s) = κg (s)U Consequently, at s = a, the height function satisfies h0 (a) = 0 and h00 (a) = κg (a). From the definition of differentiation, one deduces that in a neighboorhood of a, say over the interval (a − δ, a + δ), the projection pa : C → La of C onto La is a bicontinuous bijection (i.e., both the function and its inverse are continuous). Let pa (a − δ) = α ~ (a) − ε1 T~ (a)

and

~ (a) + ε2 T~ (a), pa (a + δ) = α

where ε1 , ε2 > 0. Define the function g : (−ε1 , ε2 ) → [0, l] as follows. Let u(s) = (~ α(a+s)−~ α(a))· T~ (a). This corresponds to the projection of α ~ (a + s) − α ~ (a) onto the tangent line, or in other words the T~ component of the curve near α ~ (a) in the frame with center α ~ (a) ~ (a)). As discussed above, this projection with ordered basis (T~ (a), U is a bijection over (−δ, δ) with its image. Let g : (ε1 , ε2 ) → (−δ, δ) be the inverse function of u(s). Finally, define f : (−ε1 , ε2 ) → R by f = ha ◦ g. The graph of the ~ (a)) with origin α function f placed in the frame (T~ (a), U ~ (a) traces out the curve C in a neighborhood of s = a (Figure 2.5). C La

~ U T~

~ U

u

f (u)

T~

(a) The curve with tangent line La .

(b) The height function.

Figure 2.5. Height function f (u).

2.4. Curvature, Convexity, and the Four-Vertex Theorem

The derivatives of f are f 0 (u) = h0 (g(u))g 0 (u), f 00 (u) = h00 (g(u))(g 0 (u))2 + h0 (g(u))g 00 (u). Since g(0) = a and h0 (a) = 0, without knowing g 0 (0), we deduce that f (0) = f 0 (0) = 0 and that f 00 (0) has the same sign as κg (a). If κg (a) < 0, then from basic calculus we deduce that f is concave down over (−ε1 , ε2 ), which implies that every line segment between points on the graph of f forms a chord that is below the graph. As an example, use the segment between (u1 , f (u1 )) and (u2 , f (u2 )). Furthermore, since the curve C has a positive orientation, the interior ~ (a), which corresponds to being above the of C is in the direction of U graph of f . Thus, the line segment between the points α ~ (g(u1 )) and α ~ (g(u2 )) is in the exterior of the curve. Consequently, this proves that κg changes sign if and only if C is not convex and the proposition follows.  This proposition and a portion of the proof leads to another more geometric characterization of convex curves. Proposition 2.4.4. A regular, simple, closed curve C is convex if and

only if it lies on one side of every tangent line to C. Proof: Suppose we parametrize C by arc length with a positive orientation. From the proof of Proposition 2.4.3, one observes that at any point p of the curve C where the curvature is negative (given the positive orientation), in a neighborhood of p, the tangent line L to C at p is in the interior of C. Consequently, since C is bounded and L is not, L must intersect the curve C at another point. By Proposition 2.4.3, this proves that C is concave if and only if there exists a tangent line that has points in the interior of C and the exterior of C.  In Section 1.3, we gave the term vertex for a point on the curve where the curvature reaches an extremum. More precisely, for any ~ : I → R2 , we defined a vertex to be regular parametrized curve X 0 ~ a point X(t0 ) where κg (t0 ) = 0. Note that since the curvature of a curve is independent of the parametrization up to a possible change of sign, vertices are independent of any parametrization of C.

67

68

2. Plane Curves: Global Properties

~ (a) X(t) = (a cos t, b sin t).

~ (b) X(t) = (ecos t , esin t ).

Figure 2.6. Two closed curves and their vertices.

The concept of a vertex is obviously a local property of the curve, but if one were to experiment with a variety of closed curves, one would soon guess that there must be a restriction on the number of vertices. In Problem 1.3.10, the reader calculated that the noncircular ellipse has four vertices. Figure 2.6(b) shows another simple closed curve with six vertices. (In each figure, the dots indicate the vertices of the curve.) Theorem 2.4.5 (Four-Vertex Theorem). Every simple, closed, regular,

convex, plane curve C has at least four vertices. Proof: Let α ~ : [0, l] → R2 be a parametrization by arc length for the curve C. Since [0, l] is compact and κg : [0, l] → R is continuous, then by the Extreme Value Theorem, κg attains both a maximum and a minimum over [0, l]. These values already assure that C has two vertices. Call these points p and q and call L the line between them. Let β be the arc along C from p to q, and let γ be the arc along C from q to p. We claim that β and γ are contained in opposite half-planes defined by the line L. Assume one of the arcs is not contained in one of the half-planes. Then C meets L at another point r. By convexity, in order for the line segments [p, q], [p, r], and [q, r] to all lie inside C,

2.4. Curvature, Convexity, and the Four-Vertex Theorem

all three points would need to have L as their tangent line to C. By Problem 2.4.1, this is a contradiction. Assume now that the arcs β and γ are contained in half-planes but in the same half-plane. Then again by convexity, the only possibility is that one of the arcs is a line segment along L. Then along that line segment, the curvature κg (s) is identically 0. However, this implies that the curvature at p and q is 0, but since p and q are extrema of κg , this would force C to be a line. This is a contradiction since C, being closed, is bounded. If there are no other vertices on the curve, then κ0g (s) does not change sign on β or on γ. If L has equation Ax + By + C = 0, then Z l dκg ds (2.6) (Ax + By + C) ds 0 is positive. However, this leads to a contradiction because for all real constants A, B, and C, the integral in Equation (2.6) is always 0 (Problem 2.4.2). This proves that there is at least one other vertex. But then, since κ0g (s) changes sign at least three times, it must change signs a fourth time as well. Hence, κ0g (s) has at least four 0s. 

Problems 2.4.1. A bitangent line L to a regular curve C is a line that is tangent to C at a minimum of two points p and q such that between p and q on C, there are points that do not have L as a tangent line. Prove that a simple, regular curve is not convex if and only if it has a bitangent line. ~ : [0, l] → R2 be a regular, closed, plane curve parametrized 2.4.2. Let X ~ by arc length and write X(s) = (x(s), y(s)). (a) Show that x00 = −κy 0 and y 00 = κx0 . (b) Prove that Z l Z l x(s)κ0g (s) ds = − κg (s)x0 (s) ds, 0

0

and do the same for y(s) instead of x(s). (c) Use the above to show that for any constants A, B, and C, we have Z l dκg ds. (Ax + By + C) ds 0

69

70

2. Plane Curves: Global Properties

2.4.3. If a closed curve C is contained inside a disk of radius r, prove that there exists a point P ∈ C such that the curvature κg (P ) of C at P satisfies 1 |κg (P )| ≥ . r 2.4.4. Let α ~ : I → R2 be a regular, closed, simple curve with positive orientation. Define the parallel curve at a distance r as ~ =α ~ (t). β(t) ~ (t) − rU Call l(~ α) the length of the curve α ~ and A(~ α) the area of the interior of the curve. Prove the following: (a) β~ is regular if and only if it is simple and if and only if κg (t) > − 1r for all t ∈ I. ~ = l(~ (b) If β~ is regular, then l(β) α) + 2πr. ~ = A(~ (c) If β~ is regular, then A(β) α) + rl(~ α) + πr 2 . ~ we have (d) At any regular point of β, κβ~ (t) =

κα~ (t) . 1 + rκα~ (t)

~ 2.4.5. Show that the lima¸con X(t) = ((1 − 2 cos t) cos t, (1 − 2 cos t) sin t) for t ∈ [0, 2π] has exactly two vertices. Explain why it does not contradict the Four-Vertex Theorem. 2.4.6. (*) Consider the curve in Figure 2.6(b). Determine the coordinates ~ of the vertices of X(t). 2.4.7. Let [A, B] be a line segment in the plane, and let l > AB. Show that the curve C that joins A and B and has length l such that, together with the segment [A, B], bounds the largest possible area is an arc of a circle passing through A and B. A

B

A

B

C

C

CHAPTER 3

Curves in Space: Local Properties The local theory of space curves is similar to the theory for plane curves, but differences arise in the richer variety of configurations available for loci of curves in R3 . In the study of plane curves, we introduced the curvature function, a function of fundamental importance that measures how much a curve deviates from being a straight line. As we shall see, in the study of space curves, we again introduce a curvature function that measures how much a space curve deviates from being linear, but we also introduce a torsion function that measures how much the curve twists away from being planar.

3.1 Definitions, Examples, and Differentiation As in the study of plane curves, one must take some care in defining what one means by a space curve. If I is an interval of R, to call a ~ : I → R3 (or the image thereof) would space curve any function X allow for separate pieces or even a set of scattered points. By a curve, one typically thinks of a connected set of points, and, just as with plane curves, the desired property is continuity. Instead of repeating the definitions provided in Section 1.1, we point out that the definition for the limit of a vector function (Definition 1.1.4), the definition for continuity of a vector function (Definition 1.1.5), Proposition 1.1.6, and Corollary 1.1.7 continue to hold for vector functions into Rn . Definition 3.1.1. Let I be an interval of R. A parametrized curve in

~ : I → Rn . If for all t ∈ I we have Rn is a continuous function X ~ X(t) = (x1 (t), x2 (t), . . . , xn (t)) ,

then the functions xi : I → R for 1 ≤ i ≤ n are called the coordinate functions or parametric equations of the parametrized curve. The

71

72

3. Curves in Space: Local Properties

~ locus is the image X(I) of the parametrized curve. If n = 3, we call ~ X a space curve. In order to develop an intuition for space curves, we present a number of examples. These illustrate only some of the great variability in the shape of parametric curves but provide a short library for examples we revisit later. Example 3.1.2 (Lines). In space or in Rn , a line is uniquely defined by

a point p~ on the line and a nonzero direction vector ~v . Parametric ~ : R → Rn with X(t) ~ equations for a line are then X = ~v t + p~. Example 3.1.3 (Planar Curves). We define a planar curve as any space ~ : curve whose image lies in a plane. A parametrized plane curve X 2 ~ I → R with X(t) = (x(t), y(t)) can be considered as a parametrized ~ space curve by setting X(t) = (x(t), y(t), 0). This becomes a planar curve. More generally, recall that if ~a and ~b are linearly independent vectors of R3 and p~ is any point in R3 , then the plane spanned by {~a, ~b} through p~ can be described by p~ + u~a + v~b, where u, v ∈ R. ~ = ~a × ~b. Consequently, Note that a normal vector to this plane is N a planar curve in this plane will be given by a parametrized curve of the form ~ X(t) = p~ + u(t)~a + v(t)~b,

where u(t) and v(t) are continuous functions I → R, where I is some interval of R. In order to properly devise parametrizations for specific curves on planes in R3 , a judicious choice of the vectors ~a and ~b may be necessary. The usual basis vectors of (1, 0) and (0, 1) of R2 are not only linearly independent but they form an orthonormal set, both mutually perpendicular and of unit length. So for example, a space curve of the form ~ X(t) = p~ + r(cos t)~a + r(sin t)~b will not in general be a circle but an ellipse. As a specific example, we can obtain the equations for a circle of radius 7 in the plane 3x − 2y + 2z = 6 with center p~ = (2, 1, 1) ~ = (3, −2, 2). Two as follows. A normal vector to the plane is N

3.1. Definitions, Examples, and Differentiation

73

vectors that are not collinear and in this plane are ~a = (2, 3, 0) and ~b = (2, 0, −3). We use the Gram-Schmidt orthonormalization process to find an orthonormal basis of Span(~a, ~b). The first vector in the orthonormal basis is   2 ~a 3 u~1 = = √ , √ ,0 . k~ak 13 13 Then calculate ~v2 = ~b − proj~u1~b = ~b − (~u1 · ~b)~u1 =



18 12 , − , −3 13 13



and get a second basis vector of   18 ~v2 12 39 = √ ~u2 = , −√ , −√ . k~v2 k 1729 1729 1729 Finally, a parametrization for the circle we are looking for is given by ~ X(t) = p~ + (7 cos t)u~1 + (7 sin t)~u2 . Example 3.1.4 (Twisted Cubic). One of the simplest nonplanar curves ~ is called the twisted cubic and has the parametrization X(t) = 2 3 (t, t , t ) (see Figure 3.1(a)). We might wonder why this should be considered the simplest nonplanar curve. After all, could we not create a nonplanar curve with just quadratic polynomials? In fact, the answer is no. A curve ~ X(t) with quadratic polynomials for coordinate functions can be written as ~ X(t) = ~at2 + ~bt + ~c,

where ~a, ~b, and ~c are linearly independent vectors in R3 . If ~a and ~b ~ are different and nonzero, then the image of X(t) lies in the plane through ~c with direction vectors ~a and ~b, in other words, the plane through ~c with normal vector ~a × ~b. To see clearly that the twisted cubic is not planar, consider ~ ~ ~ ~ X(−1) = (−1, 1, −1), X(0) = (0, 0, 0), X(1) = (1, 1, 1), and X(2) = (2, 4, 8). The first three points lie in the plane x = z but the fourth does not.

74

3. Curves in Space: Local Properties

z z

x

y

y

x (a) The twisted cubic.

(b) A helix.

Figure 3.1. Two important space curves.

Example 3.1.5 (Cylindrical Helix). A cylindrical helix is a space curve

that wraps around a circular cylinder, climbing in altitude at a constant rate. We have for equations ~ X(t) = (a cos t, a sin t, bt). ~ See Figure 3.1(b) for an example with X(t) = (2 cos t, 2 sin t, 0.2t). Example 3.1.6. The coordinate functions of the parametrized curve ~ X(t) = (at cos t, at sin t, bt) satisfy the algebraic equation

x2 y 2 z 2 + 2 − 2 =0 a2 a b ~ lies on the circular cone described by for all t. Thus, the image of X this equation. ~ Example 3.1.7. Consider the parametric curve X(t) = (a cosh t cos t, a cosh t sin t, b sinh t). It is not hard to see that the coordinate func~ satisfy tions of X x2 y 2 z 2 + 2 − 2 =1 a2 a b

3.1. Definitions, Examples, and Differentiation

75

z z

y

y

x

x (b) Space cardioid.

(a) Curve on a hyperboloid.

Figure 3.2. Examples of parametric curves.

~ lies on the hyperboloid of one sheet. See so that the image of X Figure 3.2(a) for an example with ~ X(t) = (cosh t cos(10t), cosh t sin(10t), sinh t). (Recall that hyperbolic trigonometric functions are defined as cosh t =

et + e−t 2

and

sinh t =

et − e−t 2

and that they satisfy the relation cosh2 t − sinh2 t = 1 for all t ∈ R.) Example 3.1.8 (Space Cardioid). The parametric curve with equation

~ X(t) = ((1 − cos t) cos t, (1 − cos t) sin t, sin t) is called the space cardioid (see Figure 3.2(b)). One interesting property of this curve is that it is a closed curve in R3 . Projected onto the xy-plane it gives the cardioid and (as we shall see) has no critical points. In an intuitive sense, we have stretched the cardioid out of the plane and removed its critical point (or cusp). Using the same vocabulary as in Section 1.1, we call a reparame~ : I → Rn any other continuous trization of a parametrized curve X n ~ function ξ : J → R defined by ~ ◦g ξ~ = X

76

3. Curves in Space: Local Properties

for some surjective function g : J → I. Then the image C ⊆ Rn ~ When g is not surjective, the of ξ~ is the same as the image of X. image of ξ~ could be a proper subset of C, and we do not call ξ~ a ~ is reparametrization. In addition, the reparametrization ξ~ of X • regular if g is continuously differentiable and g 0 (t) 6= 0; • positively oriented if it is regular and g 0 (t) > 0 for all t ∈ J; • negatively oriented if it is regular and g 0 (t) < 0 for all t ∈ J. The definition in Equation (1.2) of the derivative for a vector function was purposefully presented irrespective of the dimension. We restate the definition in an alternate form that is sometimes better suited for proofs. Definition 3.1.9. Let I be an interval in R, and let t0 ∈ I. If f~ : I →

Rn is a continuous vector function, we say that f~ is differentiable at t0 with derivative f~0 (t0 ) if there exists a vector function ~ε such that f~(t0 + h) = f~(t0 ) + f~0 (t0 )h + h~ε(h)

and

lim k~ε(h)k = 0.

h→0

It follows as a consequence of Proposition 1.1.6 (modified to vec~ : I → Rn is differentor functions in Rn ) that a vector function X tiable at a point if and only if all its coordinate functions are differ~ : I → Rn is a continuous entiable at that point. Furthermore, if X vector function that is differentiable at t = t0 and if the coordinate ~ are functions of X ~ X(t) = (x1 (t), x2 (t), . . . , xn (t)) , ~ 0 (t0 ) is then the derivative X  ~ 0 (t0 ) = x01 (t0 ), x02 (t0 ), . . . , x0n (t0 ) . X ~ : I → Rn , borrowing from the As in R2 , for any vector function X ~ the position language of trajectories in mechanics, we often call X 0 00 ~ ~ function, X the velocity function, and X the acceleration function. ~ as the Furthermore, we define the speed function associated to X 0 function s : I → R defined by ~ 0 (t)k. s0 (t) = k|X

3.1. Definitions, Examples, and Differentiation

77

It is also a simple matter to define equations for the tangent line ~ ~ 0 (t0 ) 6= ~0. The to a parametrized curve X(t) at t = t0 as long as X parametric equations for the tangent line are ~ 0 (t0 ). ~ ~ 0) + u X L(u) = X(t Proposition 3.1.10. Proposition 1.2.3 holds for differentiable vector functions ~v and w ~ from I ⊂ R into Rn . Furthermore, suppose that ~v (t) and w(t) ~ are vector functions into R3 that are defined on and differ~ ~ is entiable over an interval I ⊂ R. If X(t) = ~v (t) × w(t), ~ then X differentiable over I and

~ 0 (t) = ~v 0 (t) × w(t) ~ + ~v (t) × w ~ 0 (t). X Proof: (Left as an exercise for the reader. See Problem 3.1.18.)



As in Section 1.1, in this introductory section for space curves, the problems focus on properties of vectors and vector functions in R3 .

Problems 3.1.1. Calculate the velocity, acceleration, and speed of the space cardioid ~ X(t) = ((1 − cos t) cos t, (1 − cos t) sin t, sin t). 3.1.2. Calculate the velocity, acceleration, and speed of the twisted cubic ~ X(t) = (t, t2 , t3 ). 3.1.3. Calculate the velocity, acceleration, and speed of the parametrized ~ curve X(t) = (tan−1 (t), sin t, cos 2t). 3.1.4. Find a parametrization of the intersection of the cylinder x2 +z 2 = 1 with the plane x + 2y + z = 2. 3.1.5. Let ~a, ~b, and ~c be three vectors in R3 . Prove that (~a × ~b) · ~c = (~b × ~c) · ~a = (~c × ~a) · ~b. 3.1.6. Let ~a, ~b, ~c, and d~ be four vectors in R3 . Prove that ~b · ~c) (~ a · ~ c ) ( ~ = (~a × ~b) · (~c × d) ~ (~b · d) ~ . (~a · d)

78

3. Curves in Space: Local Properties

3.1.7. Let ~a, ~b, ~c be constant vectors in R3 . Calculate the derivatives ~ (a) X(t) = (2t~a + t2~b) × (~b + (4 − t3 )~c); (b) f (t) = (2t~a + t2~b) · (~b + (4 − t3 )~c). 3.1.8. Modify the parametrization of a helix to give the parametrization ~ : R → R3 of a curve that lies on the cylinder x2 + y 2 = 1 such X that the z-component has (0, ∞) as its image. ~ 3.1.9. Give parametric equations of the tangent lines to X(t) = (t2 −1, 3t− t3 , t4 − 4t2 ) at (−1, 0, 0) and also where t = 2. ~ 3.1.10. Consider the parametrized space curve X(t) = (t cos t, t sin, t) with t ∈ R. (a) Find the equation of the tangent line where t = π/3. ~ (b) Do any tangent lines of X(t) intersect the x-axis and if so, then for what values of t and at what points on the x-axis? ~ 3.1.11. Consider the parametrization space curve X(t) = (cos t, sin, sin 2t) with t ∈ [0, 2π]. Find the equations of the tangent lines at the points where the curve intersects the xy-plane. 3.1.12. Let α ~ (t) be a regular, plane, parametrized curve. View the xy-plane as a subset of R3 . Let p~ be a fixed point in the plane and ~u a fixed vector. Let θ(t) be the angle α ~ (t) − p~ makes with the direction ~u. Prove that (up to a change in sign) θ0 (t) =

α(t) − p~)k k~ α0 (t) × (~ . k~ α(t) − p~k2

Conclude that the angle function θ(t) of α ~ (t) − p~ with respect to the ~ 0 (t0 ) direction ~u has a local extremum at a point t0 if and only if α is parallel to α ~ (t0 ) − p~. ~ 3.1.13. Consider the space curve X(t) = (t cos t, t sin t, t). Find where the ~ 0 and X ~ 00 occur and determine extremal values of the angle between X whether they are maxima or minima. 3.1.14. Consider the plane in R3 given by the equation ax + by + cz + f = 0 and a point P = (x0 , y0 , z0 ). Prove that the distance between the plane and the point P is d=

|ax0 + by0 + cz0 + f | √ . a2 + b2 + c2

3.1. Definitions, Examples, and Differentiation

79

3.1.15. Determine the angle of intersection between the lines ~u(t) = (2t − 1, 3t + 2, −2t + 3) and ~v (t) = (3t − 1, 5t + 2, 3t + 3). 3.1.16. Consider the two lines given by ~u(t) = (2t − 1, 3t + 2, −2t + 3) and ~v (t) = (3t + 1, −5t − 3, 3t − 1). Find the shortest distance between these two lines. [Hint: Consider the function f (s, t) = k~u(s) −~v (t)k.] 3.1.17. Consider two nonparallel lines given by the equations ~l1 (s) = ~a + s~u

and

~l2 (t) = ~b + t~v ,

where ~u × ~v 6= ~0. Prove that the distance d between these two lines is |(~u × ~v ) · (~b − ~a)| . d= k~u × ~v k ~ be two differentiable parametrized curves R → R3 . 3.1.18. Let α ~ (t) and β(t) ~ ~ Prove that if ~γ (t) = α ~ (t)× β(t), then ~γ 0 (t) = α ~ 0 (t)× β(t)+~ α(t)× β~ 0 (t). ~ 3.1.19. Let X(t) be any parametrized curve that is of class C 3 . Prove that  d ~ ~ 0 (t) × X ~ 00 (t)) = X(t) ~ ~ 0 (t) × X ~ 000 (t)). X(t) · (X · (X dt [Hint: Use Problem 3.1.18.] 3.1.20. Consider a sphere of radius R and center p~. Prove by a geometric argument that the closest point on this sphere to a point ~q 6= p~ is p~ + R

~q − p~ . k~q − p~k

We call this closest point the projection of ~q to the sphere. Let ~ : I → R3 be a parametrized space curve that does not go through X ~ ~ the origin. Deduce that X(t)/k X(t)k parametrizes the projection of the space curve to the unit sphere. 3.1.21. Prove that no four distinct points of the twisted cubic lie in a common plane. [Hint: Prove that for all a, b, c, d ∈ R, 1 1 1 1 b−a c−a d−a 2 b − a2 c2 − a2 d2 − a2 = a2 b2 c2 d2 3 b − a3 c3 − a3 d3 − a3 a3 b3 c3 d3 a b c d and use the Vandermonde determinant identity.]

80

3. Curves in Space: Local Properties

3.2 Curvature, Torsion, and the Frenet Frame ~ : I → R3 be a differentiable space Let I be an interval of R, and let X curve. Following the setup with plane curves, we can talk about the unit tangent vector T~ (t) defined by ~ 0 (t) X T~ (t) = . ~ 0 (t)k kX

(3.1)

Obviously, the unit tangent vector is not defined at a value t = t0 ~ is not differentiable at t0 or if X ~ 0 (t0 ) = 0. This leads to the if X following definition (which can be generalized to curves in Rn ). ~ : I → Rn be a parametrized curve. We call Definition 3.2.1. Let X

~ 0 (t0 ) is not defined or if X ~ 0 (t0 ) = any point t = t0 a critical point if X ~0. A point t = t0 that is not critical is called a regular point. A ~ : I → Rn is called regular if it is of class C 1 parametrized curve X ~ 0 (t) 6= ~0 for all t ∈ I. (i.e., continuously differentiable) and if X ~ 0 (t0 ) = ~0 but X ~ 0 (t) 6= ~0 for all t in some interval In practice, if X J around t0 , it is possible for lim

t→t0

~ 0 (t) X ~ 0 (t)k kX

(3.2)

to exist. In such cases, it is common to think of T~ as completed by continuity by calling T~ (t0 ) the limit in Equation (3.2). Since T~ (t) is a unit vector for all t ∈ I, we have T~ · T~ = 1 for all t ∈ I. Therefore, if T~ is itself differentiable,  d ~ T (t) · T~ (t) = 2T~ 0 (t) · T~ (t) = 0. dt Thus, the derivative of the unit tangent vector T~ 0 is perpendicular to T~ . At any point t on the curve, where T~ 0 (t) 6= ~0, we define the principal normal vector P~ (t) to the curve at t to be the unit vector T~ 0 (t) P~ (t) = . kT~ 0 (t)k

3.2. Curvature, Torsion, and the Frenet Frame

81

~ B

~ B

T~

P~ T~

T~

P~ ~ B

~ B

P~ P~

P~

T~

~ B T~ Figure 3.3. Moving Frenet frame on the space cardioid.

Again, it is still possible sometimes to complete P~ by continuity even at points t = t0 , where T~ 0 (t0 ) = ~0. In such cases, it is common to assume that P~ (t) is completed by continuity wherever possible. Nonetheless, by Equation (3.1), the requirement that T~ be differen~ being twice differentiable. tiable is tantamount to X Finally, for a space curve defined over any interval J, where T~ (t) and P~ (t) exist (perhaps when completed by continuity), we complete the set {T~ , P~ } to an orthonormal frame by adjoining the binormal ~ vector B(t) given as ~ = T~ × P~ . B ~ : I → R3 be a continuous space curve of class Definition 3.2.2. Let X ~ C 2 (i.e., has a continuous second derivative). To each point X(t) on the curve, we associate the Frenet frame as the ordered triple of ~ vectors (T~ , P~ , B). Figure 3.3 illustrates the Frenet frame as it moves through t = 0 on the space cardioid ~ X(t) = ((1 − cos t) cos t, (1 − cos t) sin t, sin t).

(3.3)

82

3. Curves in Space: Local Properties

(In this figure, the basis vectors of the Frenet frame were scaled down by a factor of 12 to make the picture clearer.) In this example, it is interesting to see that near t = 0, though the unit tangent vector ~ rotate quickly about doesn’t change too quickly, the vectors P~ and B ~ the tangent line. The functions that measure how fast T~ (resp. B) changes are called the curvature (resp. torsion) functions of the space curve. ~ (t)} for plane curves, the Frenet Similar to the basis {T~ (t), U frame provides a geometrically natural basis in which to study local properties of space curves. We now analyze the derivatives of a space curve in reference to the Frenet frame. ~ 0 (t)k. By definition of the unit Recall that the speed s0 (t) is kX tangent vector, we have ~ 0 (t) = s0 (t)T~ (t). X If ~x is twice differentiable at t, then T~ 0 (t) exists. If, in addition, P~ is defined at t, then by definition of the principal normal vector, T~ 0 is parallel to P~ . This allows us to define the curvature of a space curve as follows. ~ be a regular parametrized curve of class C 2 . Definition 3.2.3. Let X The curvature κ : I → R+ of a space curve is κ(t) =

kT~ 0 (t)k . s0 (t)

Note that at any point where P (t) is defined, κ(t) is the nonnegative number defined by T~ 0 (t) = s0 (t)κ(t)P~ (t). We would like to determine how the other unit vectors of the ~ = Frenet frame behave under differentiation. Remember that B T~ × P~ . Taking a derivative of this cross product, we obtain ~ 0 = T~ 0 × P~ + T~ × P~ 0 = T~ × P~ 0 , B since T~ 0 is parallel to P~ . However, just as T~ 0 ⊥ T~ , we have the same for P~ 0 ⊥ P~ . Thus, since we are in three dimensions, we can write

3.2. Curvature, Torsion, and the Frenet Frame

83

~ for some continuous functions f, g : I → P~ 0 (t) = f (t)T~ (t) + g(t)B(t) R. Consequently, we deduce that ~ = f (t)T~ × T~ + g(t)T~ × B ~ = −g(t)P~ . ~ 0 = T~ × (f (t)T~ + g(t)B) B Thus, the derivative of the unit binormal vector is parallel to the principal normal vector. Definition 3.2.4. Let ~ x : I → R3 be a regular space curve of class C 2

for which the Frenet frame is defined everywhere. We define the torsion function τ : I → R as the unique function such that ~ 0 (t) = −s0 (t)τ (t)P~ (t). B We now need to determine P~ 0 (t), and we already know that it ~ has the form f (t)T~ (t) + g(t)B(t). However, we can say much more ~ form without performing any specific calculations. Since (T~ , P~ , B) an orthonormal frame for all t, we have the following equations: T~ · P~ = 0

and

~ = 0. P~ · B

Taking derivatives with respect to t, we have T~ 0 · P~ + T~ · P~ 0 = 0

and

~ + P~ · B ~ 0 = 0, P~ 0 · B

P~ 0 · T~ = −T~ 0 · P~

and

~ = −P~ · B ~ 0. P~ 0 · B

that is,

Thus, we deduce that ~ P~ 0 (t) = −s0 (t)κ(t)T~ (t) + s0 (t)τ (t)B(t).

(3.4)

~ ~ ~ If we assume, as one does  in linearalgebra, that T , P , and B are ~ as the matrix that has these column vectors and write T~ P~ B vectors as columns, we can summarize Definition 3.2.3, Definition 3.2.4, and Equation (3.4) in matrix form by   0 −s0 κ 0 d ~ ~ ~ ~ ~ ~  0 sκ 0 −s0 τ  . (3.5) T P B = T P B dt 0 0 sτ 0

84

3. Curves in Space: Local Properties

As provided, the definitions for curvature and torsion of a space curve do not particularly lend themselves to direct computations when given a particular curve. We now obtain formulas for κ(t) and τ (t) in terms of ~x(t). In order for our formulas to make sense, we assume for the remainder of this section that the parametrized curve ~x : I → R3 is regular and of class C 3 . First, to find the curvature of a curve, we take the following derivatives: ~ 0 (t) = s0 (t)T~ (t), X ~ 00 (t) = s00 (t)T~ (t) + (s0 (t))2 κ(t)P~ (t). X

(3.6) (3.7)

Taking the cross product of these two vectors, we now obtain ~ 0 (t) × X ~ 00 (t) = (s0 (t))3 κ(t)B(t). ~ X

(3.8)

However, by definition of curvature for a space curve, κ(t) is a non~ 0 (t)k, so we get negative function. Furthermore, s0 (t) = kX κ(t) =

~ 0 (t) × X ~ 00 (t)k kX . s0 (t)3

(3.9)

~ Secondly, to obtain the torsion function from X(t) directly, we will need to take the third derivative ~ 000 (t) = s000 (t)T~ + s00 (t)s0 (t)κ(t)P~ + 2s0 (t)s00 (t)κ(t)P~ X ~ + (s0 (t))2 κ0 (t)P~ + (s0 (t))3 κ(t)(−κ(t)T~ + τ (t)B), which, without writing the dependence on t explicitly, reads   ~ 000 = s000 − (s0 )3 κ2 T~ + 3s00 s0 κ + (s0 )2 κ0 P~ + (s0 )3 κτ B. ~ (3.10) X ~ 00 with X ~ 000 eliminates all the terms ~ 0 ×X Taking the dot product of X ~ 000 associated to T~ and P~ . We get of X ~0 ×X ~ 00 ) · X ~ 000 = (s0 (t))6 (κ(t))2 τ (t), (X from which we deduce a formula for τ (t) only in terms of the curve ~ X(t) as follows: τ (t) =

~ 00 (t)) · X ~ 000 (t) ~ 0 (t) × X (X . ~ 0 (t) × X ~ 00 (t)k2 kX

3.2. Curvature, Torsion, and the Frenet Frame

85

Example 3.2.5 (Helices). We wish to calculate the curvature and tor-

~ sion of the circular helix X(t) = (a cos t, a sin t, bt). We need the following: ~ 0 (t) = (−a sin t, a cos t, b), X p p s0 (t) = a2 sin2 t + a2 cos2 t + b2 = a2 + b2 , ~ 00 (t) = (−a cos t, −a sin t, 0), X ~ 0 (t) × X ~ 00 (t) = (ab sin t, −ab cos t, a2 ), X ~ 000 (t) = (a sin t, −a cos t, 0). X Note that arc length since √ one can easily parametrize the helix by √ 0 2 2 s (t) = a + b is a constant function, so s(t) = t a2 + b2 . Thus, the parametrization by arc length for the circular helix is  s s b  p ~ X(s) = a cos , a sin , s , where c = a2 + b2 . c c c We now calculate the curvature and torsion as √ ~ 00 k ~0 ×X kX |a| a2 + b2 |a| κ(t) = = 2 = 2 , 0 3 3/2 2 (s ) a + b2 (a + b ) ~ 00 ) · X ~ 000 ~0 ×X (X b a2 b τ (t) = = 2 = 2 2 . 2 ~0 ×X ~ 00 k2 a (a + b ) a + b2 kX Consequently, we find that this circular helix has constant curvature and constant torsion. The circular helix, however, is a particular case of a larger class of curves simply called helices. These are defined by requiring that the unit tangent make a constant angle with a fixed line in space. ~ Thus, X(t) is a helix if and only if for some unit vector ~u, T~ · ~u = cos α = const.

(3.11)

Taking a derivative of Equation (3.11), we obtain P~ · ~u = 0. ~ so we can Hence for all t, ~u is in the plane determined by T~ and B, write ~ or ~u = cos αT~ − sin αB ~ ~u = cos αT~ + sin αB

86

3. Curves in Space: Local Properties

κ

κ

2π t

2π t (a) κ(t).

(b) τ (t).

Figure 3.4. Curvature and torsion of the space cardioid.

for all t. Taking the derivative, we obtain 0 = s0 κ cos αP~ − s0 τ sin αP~ , which implies that κ = tan α, τ and thus, for any helix, the ratio of curvature to torsion is a constant. This ratio κτ = ab is called the pitch of the helix. One can follow the above discussion in reverse and also conclude that the converse is true. Therefore, a curve is a helix if and only if the ratio of curvature to torsion is a constant. Example 3.2.6 (Space Cardioid). We consider the space cardioid again

as a follow-up to Figure 3.3. Figure 3.3 shows that in the vicinity of t = 0, the Frenet frame twists quickly about the tangent line, even while the tangent line does not move much. This indicates that near 0, κ(t) is not large, while τ (t) is relatively large. We leave the precise calculation of the curvature and torsion functions to the space cardioid as an exercise for the reader but plot their graphs in Figure 3.4. The graphs of κ(t) and τ (t) justify the intuition provided by Figure 3.3 concerning how the Frenet frame moves through t = 0. In particular, the torsion function has a high

3.2. Curvature, Torsion, and the Frenet Frame

87

peak at t = 0, which indicates that the Frenet frame rotates quickly about the tangent line. In some proofs that we will encounter later, it is often useful to assume that a curve is parametrized by arc length. In this case, in all of the above formulas, one has s0 = 1 and s00 = 0 as functions. The transformation properties of the Frenet frame then read   0 −κ 0     d ~ ~ ~ ~ κ 0 −τ  . (3.12) T P B = T~ P~ B ds 0 τ 0 ~ If X(s) is parametrized by arc length and is of class C 3 , then Equa~ 00 (s) = κ(s)P~ (s). Hence, the curvature is given tion (3.7) gives us X by ~ 00 (s)k. κ(s) = kX Furthermore, at any point where κ(s) 6= 0, the torsion function is τ (s) =

~ 00 (s)) · X ~ 000 (s) ~ 0 (s) × X (X . ~ 00 (s)k2 kX

A key property of the curvature and torsion functions of a space curve is summarized in the following proposition. ~ : I → R3 be a regular parametric curve. Proposition 3.2.7. Let X ~ is of class C 2 . If the curvature κ(t) is identi1. Suppose that X ~ is a line segment. cally 0, then the locus of X ~ is of class C 3 . If the torsion τ (t) is identically 2. Suppose that X ~ lies in a plane. 0, then the locus of X Proof: (Left as an exercise for the reader. See Problem 3.2.11.)



Problems ~ 3.2.1. Calculate the curvature and torsion of the twisted cubic X(t) = 2 3 (t, t , t ). 3.2.2. Calculate the curvature and torsion of the space cardioid parametrized in Equation (3.3).

88

3. Curves in Space: Local Properties

~ 3.2.3. Calculate the curvature and torsion of X(t) = (t, f (t), g(t)), where 3 f and g are functions of class C . ~ = (a(t − sin t), a(1 − 3.2.4. Calculate the curvature and torsion of X cos t), bt). 1−t2 ~ 3.2.5. Calculate the curvature and torsion of X(t) = (t, 1+t t , t ).

~ 3.2.6. Consider the parametrized curve X(t) = (cos(t), sin(t), sin(2t)). ~ (a) Calculate the curvature and torsion of X(t). (b) Prove that the curvature is never 0. (c) Find the exact locations of the vertices, i.e., where κ0 (t) = 0. 3.2.7. In Chapter 5, we will encounter a surface called a torus. Curves that lie on torus are often knotted. Define a torus knot as a curve parametrized by ~ X(t) = ((a + b cos(qt)) cos(pt), (a + b cos(qt)) sin(pt), b sin(qt)) where a and b are real numbers, with a > b > 0, and p and q are relatively prime positive integers. Let a = 2 and b = 1. Calculate the curvature functions of the corresponding torus knot. 3.2.8. Let ~x be a parametrized curve, and let ξ~ = ~x ◦ g be a regular repa~ rametrization of X. (a) Prove that the curvature function κ is unchanged under a regular reparametrization, i.e., that κ~x (g(u)) = κξ~(u). (b) Prove that the torsion function τ is unchanged under a positively oriented reparametrization and becomes −τ under a negatively oriented reparametrization. ~ 3.2.9. Consider the curve X(t) = (a sin t cos t, a sin2 t, a cos t). ~ lies on a sphere. (a) Prove that the locus of X ~ (b) Calculate the curvature and torsion functions of X(t). ~ 3.2.10. If X(t) is a parametrization for a planar curve, then for some fixed vectors ~a, ~b, and ~c, with ~b and ~c not collinear, and for some real functions f (t) and g(t), we can write ~ X(t) = ~a + f (t)~b + g(t)~c. Prove that for a planar curve, its torsion function is identically 0.

3.2. Curvature, Torsion, and the Frenet Frame

89

3.2.11. (ODE) Prove Proposition 3.2.7. 3.2.12. Similar to the case of plane curves, we define the evolute to a space ~ curve X(t) as the curve 1 ~ ~ P (t). ~γ (t) = X(t) + κ(t) ~ Prove that the evolute to the circular helix X(t) = (a cos t, a sin t, bt) is another circular helix. Find the pitch of this new helix. 3.2.13. Consider the circular cone with equation x2 y2 z2 + 2 = 2. 2 a a b (a) Find parametric equations for a curve that lies on this cone such that the tangent makes a constant angle with the z-axis. (b) Prove that they project to logarithmic spirals on the xy-plane. (c) Find the curvature and torsion functions. 3.2.14. Let α ~ : I → R3 be a parametrized regular curve with κ(t) 6= 0 and τ (t) 6= 0 for t ∈ I. The curve α ~ is called a Bertrand curve if there exists another curve β~ : I → R3 such that the principal normal lines to α ~ and β~ are equal at all t ∈ I. The curve β~ is called the Bertrand mate of α ~. (a) Prove that we can write ~ =α β(t) ~ (t) + rP~ (t) for some constant r. (b) Prove that α ~ is a Bertrand curve if and only if there exists a linear relation aκ(t) + bτ (t) = 1

for all t ∈ I,

where a and b are nonzero constants and κ(t) and τ (t) are the curvature and the torsion of α ~ respectively. (c) Prove that a curve α ~ has more than one Bertrand mate if and only if α ~ is a circular helix.

90

3. Curves in Space: Local Properties

3.3 Osculating Plane and Osculating Sphere ~ As with plane curves, if X(t) is a regular space curve of class C 2 , one ~ can talk about osculating circles to X(t) at a point t = t0 . Recall that an osculating circle to a curve at a point is a circle with contact of order 2 at that point. Following the proof of Proposition 1.4.4, one determines that at any point t = t0 , where κ(t0 ) 6= 0, the osculating circle exists and a parametric formula for it is   ~ 0 ) + 1 P~ (t0 ) + 1 (sin t)T~ (t0 ) − (cos t)P~ (t0 ) . ~γ (t) = X(t κ(t0 ) κ(t0 ) (3.13) Even without reference to osculating circles, given any parametric ~ : I → R3 , the second-order Taylor approximation to X ~ at curve X t = t0 is a planar curve with contact of order 2. Furthermore, if ~ is T~ 0 (t0 ) 6= ~0, this second-order approximation f~ of X ~ 0 ) + (t − t0 )X ~ 0 (t0 ) + 1 (t − t0 )2 X ~ 00 (t0 ) f~(t) = X(t 2   1 0 00 2 ~ 0 ) + s (t0 )(t − t0 ) + s (t0 )(t − t0 ) T~ (t0 ) + 1 s0 (t0 )2 κ(t0 )(t − t0 )2 P~ (t0 ). = X(t 2 2 The vector function f~ is a planar curve that lies in the plane that ~ 0 ) and has T~ (t0 ) and P~ (t0 ) as direction goes through the point X(t vectors. (If κ(t0 ) = 0, then P~ (t0 ) is not strictly defined and might not even be defined by completing by continuity. In this case, the ~ at t0 lies on a line.) This leads second-order approximation f~ to X to the following definition. ~ : I → R3 be a parametrized curve, and let Definition 3.3.1. Let X

~ is of class C 2 over an open interval containing t0 ∈ I. Suppose that X ~ at t = t0 is the t0 and that κ(t0 ) 6= 0. The osculating plane to X ~ 0 ) spanned by T~ (t0 ) and P~ (t0 ). In other words, plane through X(t the osculating plane is the set of points ~u ∈ R3 such that ~ 0 )) = 0. ~ 0 ) · (~u − X(t B(t ~ 0 ) is parallel to X ~ 0 (t0 ) × Note that from Equation (3.8), B(t 00 ~ ~ X (t0 ), so the osculating plane also has the equation (X 0 (t0 ) × ~ 00 (t0 )) · (~u − X(t ~ 0 )) = 0 in points ~u ∈ R3 . X

3.3. Osculating Plane and Osculating Sphere

We introduced the notion of order of contact between two curves in Section 1.4, but we can also talk about the order of contact between a curve C and a surface Σ by defining this latter notion as the order of contact between C and C2 , where C2 is the orthogonal projection of C onto Σ. ~ : I → R3 be a regular parametrized curve of Proposition 3.3.2. Let X

~ at t = t0 is the unique plane class C 2 . The osculating plane to X 3 ~ is of in R of contact order 2 or greater. Furthermore, assuming X 3 class C , the osculating plane has contact order 3 or greater if and only if τ (t0 ) = 0 or κ(t0 ) = 0.

~ by arc length so that ~ 0 ) and reparametrize X Proof: Set A = X(t s = 0 corresponds to the point A. The orthogonal distance f (s) ~ between X(s) and the osculation plane P is ~ ~ ~ f (s) = |B(0) · (X(s) − X(0))|. ~ Using the Taylor approximation of X(s) near s = 0; Equations (3.6), ~ · T~ = B ~ · P~ = 0, we deduce (3.7), and (3.10); and the fact that B that 1 f (s) = κ(0)τ (0)s3 + higher order terms . 6 Thus, f (s) |κ(0)τ (0)| lim 3 = s→0 s 6 and the proposition follows.  One can now interpret the sign of the torsion function τ (t) of a parametrized curve in terms of the curve’s position with respect to its osculating plane at a point. In fact, τ (t0 ) > 0 at ~x(t0 ) when the ~ curve comes up through the osculating plane (where the binormal B defines the up direction) and τ (t0 ) < 0 when the curve goes down through the osculating plane (see Figure 3.5). The osculating plane along with two other planes form what is called the moving trihedron, which consists of the coordinate planes ~ 0 ) and spanned by the in the Frenet frame. The plane through X(t principal normal and binormal is called the normal plane and is the set of points ~u that satisfy ~ 0 )) = 0. T~ (t0 ) · (~u − X(t

91

92

3. Curves in Space: Local Properties

~ B

~ B

P~

P~ T~

T~

(a) Positive torsion.

(b) Negative torsion.

Figure 3.5. Torsion and the osculating plane.

The plane through the tangent and binormal is called the rectifying plane and is the set of points ~u that satisfy ~ 0 )) = 0. P~ (t0 ) · (~u − X(t Figure 3.6 shows the osculating plane, the normal plane, and the rectifying plane together at a point on the space cardioid. We now apply the theory of order of contact from Section 1.4 to find the osculating sphere – a sphere that has order of contact 3 or higher to a curve at a point. Suppose that a sphere has center ~c and radius r so that its points ~ satisfy the equation Z ~ − ~ck2 = r2 . kZ ~ : I → R3 . The Consider a curve C parametrized by arc length by X ~ on the curve and the sphere is distance f (s) between the point X(s) ~ − ~ck − r . f (s) = kX(s) p g(t) are equal to 0 if and only if Since derivatives of G(t) = g 0 (t) = 0, then the derivatives of f (s) are equal to 0 if and only if the derivatives of ~ ~ − ~c) − ~c) · (X(s) h(s) = (X(s)

3.3. Osculating Plane and Osculating Sphere

93

normal rectifying

osculating

Figure 3.6. Moving trihedron.

are equal to 0. The first three derivatives of h(s) lead to ~ − ~c) · T~ = 0, h0 (s) = 0 ⇐⇒ (X ~ − ~c) · κP~ + 1 = 0, h00 (s) = 0 ⇐⇒ (X ~ − ~c) · (κ0 P~ − κ2 T~ + κτ B) ~ = 0. h000 (s) = 0 ⇐⇒ (X ~ 0 ) on the curve such that κ(s0 ) 6= 0 Consequently, at any point X(s and τ (s0 ) 6= 0, the first three derivatives can be equal to 0 if we have ~ − ~c) · T~ = 0, (X

~ − ~c) · P~ = − 1 , (X κ

0 ~ − ~c) · B ~ = κ . (3.14) (X κ2 τ

The equations in (3.14) give a decomposition of the center of the ~ osculating sphere to a curve X(s) at the point s = s0 . Isolating ~c, the center is 0 ~ 0) ~ 0 ) + R(s0 )P~ (s0 ) + R (s0 ) B(s ~c = X(s τ (s0 )

(3.15)

94

3. Curves in Space: Local Properties

where R(s) =

1 κ(s) .

The radius of the osculating sphere is s r=

R(s0

 )2

+

R0 (s0 ) τ (s0 )

2 .

That the first three derivatives of the distance function f (s) are 0 implies that the curve C and the osculating sphere have contact of order 3. If τ (s0 ) = 0, then h000 (s) may still be 0 as long as κ0 (s0 ) = 0 at the same time. In that case, the curve admits a one-parameter family of ~ component of the center ~c osculating spheres at ~x(s0 ), where the B can be anything. This discussion leads to the following proposition. ~ : I → R3 be a regular parametrized curve of Proposition 3.3.3. Let X class C 3 . Let t0 ∈ I be a point on the curve where κ(t0 ) 6= 0. The ~ admits an osculating sphere at t = t0 if either (1) τ (t0 ) 6= 0 curve X 1 or (2) τ (t0 ) = 0 and κ0 (t0 ) = 0. Define R(t) = κ(t) . If τ (t0 ) 6= 0, then at t0 the curve admits a unique osculating sphere with center ~c and radius r, where ~ 0 ) + R(t0 )P~ (t0 ) + ~c = X(t

R0 (t

0)

s0 (t0 )τ (t0 )

s ~ 0) B(t

and

r=



R(t0

)2

+

R0 (t0 ) s0 (t0 )τ (t0 )

2 .

If τ (t0 ) = 0 and κ0 (t0 ) = 0, then at t0 the curve admits as an osculating sphere any sphere with center ~c and radius r where ~ 0 ) + R(t0 )P~ (t0 ) + cB B(t ~ 0) ~c = X(t

and

q r = R(t0 )2 + c2B ,

where cB is any real number. Proof: The only matter to address beyond the previous discussion is to see how the various quantities in Equation (3.15) change under a reparametrization. Let J ⊂ R be an interval, and let f : J → I be of class C 3 such that f 0 doesn’t change sign. Then ξ~ = ~x ◦ f is a regular reparametr-

3.3. Osculating Plane and Osculating Sphere

Figure 3.7. Osculating sphere.

~ It is not hard to check that ization of X. T~ξ = sign(f 0 )T~ , P~ξ = P~ , ~ ξ = sign(f 0 )B. ~ B Also, in Problem 3.2.8, the reader showed that κξ = κ, τξ = sign(f 0 )τ. ~ is invariant under Consquently, we have Rξ = R and also that τ1 B any regular reparametrization. ~ ~ On the other hand, if t = f (u) so that ξ(u) = X(t), then 0 0 0 Rξ (u) = R(f (u)) and therefore, Rξ (u) = R (f (u))f (u) = R0 (t)f 0 (u). 1 dR In particular, dR ds = s0 (t) dt . The proposition then follows from the prior discussion.  Figure 3.7 illustrates an example of a curve and its osculating sphere at a point. The figure also shows the Frenet frame at the point in question, along with the orthogonal projection of the curve onto the sphere. This projection onto the sphere has the parametric equations ~ X(t) − ~c ~γ (t) = r. ~ kX(t) − ~ck

95

96

3. Curves in Space: Local Properties

Problems ~ 3.3.1. Find the osculating plane to the space cardioid X(t) = ((1−cos t) cos t, (1 − cos t) sin t, sin t) at t = π. ~ 3.3.2. Calculate the osculating circle to twisted cubic X(t) = (t, t2 , t3 ) at t = 1. ~ 3.3.3. Consider the curve X(t) = (cos t, sin t, sin 2t) for t ∈ [0, 2π]. Note that this curve lies on the cylinder x2 + y 2 = 1. (a) Find equations for the osculating plane at t0 . (b) For what values of t is the osculating plane tangent to the cylinder (i.e., the osculating plane has a normal vector of the form (λ cos θ, λ sin θ, 0)? ~ 3.3.4. Let X(t) : I → R3 be the parametrization by arc length of a curve ~ C. Consider a circle that passes through the three points X(s), ~ + h1 ), and X(s ~ + h2 ). Prove that as (h1 , h2 ) → (0, 0), the X(s limiting position of this circle is precisely the osculating circle to C ~ at the point X(s). ~ 3.3.5. Let X(t) : I → R3 be a parametrized curve, and let t ∈ I be a fixed point where κ(t) 6= 0. Define π : R3 → R2 as the orthogonal ~ at t. Define γ = π◦X ~ projection of R3 onto the osculating plane to X ~ into the osculating as the orthogonal projection of the space curve X ~ is equal to the curvature plane. Prove that the curvature κ(t) of X κg (t) of the plane curve ~γ . ~ 3.3.6. Calculate the osculating sphere of the twisted cubic X(t) = (t, t2 , t3 ) at the point (0, 0, 0). 3.3.7. Prove that if all the normal planes to a curve C pass through a fixed point p~, then C lies on a sphere of center p~. 3.3.8. Let C be a curve whose curvature and torsion in terms of arc length 1 . Suppose that τ (s) and R0 (S) are κ(s) and τ (s). Set R(s) = κ(s) are never 0. Prove that C lies on a sphere if and only if  0 2 R (s) R(s)2 + = const. τ (s) 3.3.9. Determine a condition where the osculating circle has contact of order 3 or higher. ~ : I → R3 of class 3.3.10. Let C be a regular curve with a parametrization X 3 ~ 0 ) as the C . Define the osculating helix to C at a point P = X(t unique helix that goes through P and has curvature κ(t0 ) and τ (t0 ).

3.4. Natural Equations

(a) Find the parametric equations of the osculating helix. (b) Prove that the osculating helix has contact of order 3 with C at P . 3.3.11. Let C be a curve that lies on a sphere of radius r and suppose that its curvature and torsion functions with respect to arc length are κ(s) and τ (s). Since the curve lies on a sphere, this sphere is its osculating sphere at all points on C. Thus, by Proposition 3.3.3, if  0 2 (s) R(s) = 1/κ(s), then R(s)2 + Rτ (s) is a constant function equation to r2 . Prove the converse: that if C is a curve with κ(s) and τ (s)  0 2 (s) is constant, then C lies on a sphere. [Hint: such that R(s)2 + Rτ (s) Prove that the center of the osculating sphere does not move.]

3.4 Natural Equations In Section 1.5, we showed that the curvature (in terms of arc length) function uniquely specifies a regular curve up to its location and orientation in R2 . For space curves, one must introduce the torsion function for a measurement of how much the curve twists away from being planar, and Equation (3.10) shows how the torsion function ~ 000 in the Frenet frame. Since we know appears as a component of X ~ change with respect to t, we can express all higher how T~ , P~ , and B (n) ~ ~ derivatives X (t) of X(t) in the Frenet frame in terms of s0 (t), κ(t), and τ (t) and their derivatives. This leads one to posit that a curve is to some degree determined uniquely by its curvature and torsion functions. The following theorem shows that this is indeed the case. Theorem 3.4.1 (Fundamental Theorem of Space Curves). Given functions

κ(s) ≥ 0 and τ (s) continuously differentiable over some interval J ⊆ R containing 0, there exists an open interval I containing 0 ~ : I → R3 that parametrizes its locus and a regular vector function X by arc length, with κ(s) and τ (s) as its curvature and torsion functions, respectively. Furthermore, any two curves C1 and C2 with curvature function κ(s) and torsion function τ (s) can be mapped onto one another by a rigid motion of R3 . Proof: Let κ(s) and τ (s) be functions defined over an interval J ⊂ R with κ(s) ≥ 0. Consider the following system of 12 linear first-order

97

98

3. Curves in Space: Local Properties

differential equations: x01 (s) = t1 (s),

p01 (s) = −κ(s)t1 (s) + τ (s)b1 (s),

x02 (s) = t2 (s),

p02 (s) = −κ(s)t2 (s) + τ (s)b2 (s),

x03 (s) = t3 (s),

p03 (s) = −κ(s)t3 (s) + τ (s)b3 (s),

t01 (s) = κ(s)p1 (s),

b01 (s) = −τ (s)p1 (s),

t02 (s) = κ(s)p2 (s),

b02 (s) = −τ (s)p2 (s),

t03 (s) = κ(s)p3 (s),

b03 (s) = −τ (s)p3 (s)

(3.16)

where xi (s), ti (s), pi (s), and bi (s), with i = 1, 2, 3, are unknown functions. With the stated conditions on κ(s) and τ (s), according to the existence and uniqueness theorem for first-order systems of differential equations (see [2, Section 31.8]), there exists a solution to the above system defined for s in a neighborhood of 0. Furthermore, there exists a unique solution with specified initial conditions x1 (0) = x10 ,

x2 (0) = x20 ,

x3 (0) = x30 ,

t1 (0) = t10 ,

t2 (0) = t20 ,

t3 (0) = t30 ,

p1 (0) = p10 ,

p2 (0) = p20 ,

p3 (0) = p30 ,

b1 (0) = b10 ,

b2 (0) = b20 ,

b3 (0) = b30 .

Define the two matrices of functions  0 −κ(s) 0 0 −τ (s) A(s) = κ(s) 0 τ (s) 0

and

 t1 (s) p1 (s) b1 (s) M (s) = t2 (s) p2 (s) b2 (s) . t3 (s) p3 (s) b3 (s)

Recall from Equation (3.12) that M 0 (s) = M (s)A(s). It is possible to show that since A(s) is antisymmetric, the function f (s) = M (s)T M (s) is constant as a matrix of functions. Therefore, any solution to Equation (3.16) is such that M (s)T M (s) remains constant. In particular, if we choose initial conditions such that       t10 p10 b10 t20  , p20  , b20  (3.17) t30 p30 b30 form an orthonormal basis of R3 , then solutions to Equation (3.16)

3.4. Natural Equations

are such that, for all s in the domain of the solution, the vectors       p1 (s) b1 (s) t1 (s) p2 (s) , b2 (s) t2 (s) , (3.18) t3 (s) p3 (s) b3 (s) form an orthonormal basis. Therefore, we have shown that with a choice of initial conditions such that the vectors in Equation (3.17) form an orthonormal basis, ~ the corresponding solution to Equation (3.16) is such that X(s) = 3 (x1 (s), x2 (s), x3 (s)) is a regular curve of class C with curvature κ(s) and torsion τ (s). In addition, the vector functions in Equation (3.18) ~ vectors of the Frenet frame associated to X(s). ~ are the T~ , P~ , and B This proves existence. The existence and uniqueness theorem of systems of differential equations states that a solution is unique once (the right number of) initial conditions are specified. However, we have imposed the additional condition that the vectors in Equation (3.17) form an orthonormal set. Any different choice for the initial conditions in Equation (3.17) corresponds to a rotation in R3 . Also, two different choices of initial conditions x10 , x20 , x30 correspond to a translation in R3 . Therefore, different allowed initial conditions correspond to a ~ rigid motion of the locus of X(s) in R3 . This proves the theorem. Because of the Fundamental Theorem of Space Curves, the pair of functions κ(s) and τ (s), where κ(s) is a positive function, are called the natural equations of a curve. Taken as a pair, they define the curve uniquely up to a rigid motion in the plane. Just as in Section 1.5, the proof of Theorem 3.4.1 gives an algorithm to reconstruct a space curve from its curvature and torsion (with respect to arc length) functions. The code in Maple for Figure 3.8 is the following. > with(DEtools): > with(plots): > kappa:=s->(1-s^2)/(1+s^2): > tau:=s->sin(s)/5: > sys := D(x1)(s)=t1(s), D(x2)(s)=t2(s), D(x3)(s)=t3(s), D(t1)(s)=kappa(s)*p1(s), D(t2)(s)=kappa(s)*p2(s), D(t3)(s)=kappa(s)*p3(s), D(p1)(s)=-kappa(s)*t1(s)-tau(s)*b1(s), D(p2)(s)=-kappa(s)*t2(s)-tau(s)*b2(s),

99

100

3. Curves in Space: Local Properties

Figure 3.8. A ball of yarn by natural equations.

D(p3)(s)=-kappa(s)*t3(s)-tau(s)*b3(s), D(b1)(s)=tau(s)*p1(s), D(b2)(s)=tau(s)*p2(s), D(b3)(s)=tau(s)*p3(s): > DEplot3d({sys},{x1(s),x2(s),x3(s),t1(s),t2(s),t3(s), p1(s),p2(s),p3(s),b1(s),b2(s),b3(s)},s=0..85,[[x1(0)=0,x2(0)=0, x3(0)=0,t1(0)=1,t2(0)=0,t3(0)=0,p1(0)=0,p2(0)=1,p3(0)=0, b1(0)=0,b2(0)=0,b3(0)=1]],scene=[x1(s),x2(s),x3(s)], stepsize=0.02,linecolor=black,scaling=constrained);

This code constructs a curve with κ(s) = (1 − s2 )/(1 + s2 ) and τ (s) = 15 sin(s), but only for s ≥ 0. Note that lim κ(s) = 1,

s→∞

so as s gets large, the space curve has approximately constant curvature κ = 1 but with a torsion that oscillates between − 15 and 15 . From Proposition 3.3.3, we expect that as s grows, the curve would lie more and more on a sphere of radius 1.

Problems 3.4.1. Assume that κ(s) and τ (s) are of class C ∞ over an interval I and ~ : I → R3 that are also assume that we only consider space curves X ∞ ~ to prove Theorem of class C . Use the Taylor expansion of X(s) 3.4.1. 3.4.2. Prove that if κ(s) and τ (s) are nonzero constant functions, then the resulting curve must be a helix.

3.4. Natural Equations

Investigative Projects Project I. Problem 3.2.7 introduced the torus knot. Explore the torsion function for various values of p and q. In particular, can you create a torus knot whose torsion function is everywhere negative? Project II. Using a CAS to draw the curves from the natural equations, attempt to find some nontrivial examples of curvature and torsion functions κ(s) and τ (s) that generated closed, nonplanar space curves. Describe the reasons behind your attempts.

101

CHAPTER 4

Curves in Space: Global Properties

Paralleling our presentation of curves in the plane, we now turn from local properties of space curves to global properties. As before, global properties of curves are properties that involve the curve as a whole as opposed to properties that are defined in the neighborhood of a point on the curve. The Jordan Curve Theorem does not apply to curves in R3 , so Green’s Theorem, the isoperimetric inequality, and theorems connecting curvature and convexity do not have an equivalent for space curves. On the other hand, curves in space exhibit new types of global properties, in particular, knottedness and linking.

4.1 Basic Properties Definition 4.1.1. A parametrized space curve C is called closed if there

~ : [a, b] → R3 of C such that X(a) ~ exists a parametrization X = k ~ X(b). A closed curve is of class C if, in addition, all the (one-sided) ~ at a and at b are equal of order i = 0, 1, . . . , k; derivatives of X ~ 0 (a) = X ~ 0 (b), X ~ 00 (a) = in other words, if as one-sided derivatives X ~ 00 (b), and so on up to X ~ (k) (a) = X ~ (k) (b). A space curve C is called X simple if it has no self-intersections, and a closed curve is called simple if it has no self-intersections except at the endpoints. As discussed in Section 2.1, a closed space curve can be understood as a function f : S1 → R3 that is continuous as a function between topological spaces (see [24, Appendix A]). Again, to say that a closed curve is simple is tantamount to saying that as a continuous function f : S1 → R3 , it is injective. Our first result has an equivalent in the theory of plane curves. Proposition 4.1.2. If a regular curve is closed, then it is bounded.

103

104

4. Curves in Space: Global Properties

Proof: (Left as an exercise for the reader.)



Stokes’ Theorem presents a global property of curves in space in that it relates a quantity calculated along the curve with a quantity that depends on any surface that has that curve as a boundary. However, the theorem involves a vector field in R3 , and it does not address “geometric” properties of the curve, by which we mean properties that are independent of the curve’s location and orientation in space. Therefore, we simply state Stokes’ Theorem as an example of a global theorem and trust the reader has seen it in a previous calculus course. Theorem 4.1.3 (Stokes’ Theorem). Let S be an oriented, piecewise reg-

ular surface bounded by a closed, piecewise regular curve C. Let F~ : R3 → R3 be a vector field over R3 that is of class C 1 on S. Supposing that C is oriented according to the right-hand rule, Z ZZ ~ × F~ ) · dS. ~ (∇ (4.1) F~ · d~s = C

S

If S has no boundary curve, then the line integral on the left is taken as 0. Note as a reminder that the line element d~s stands for ~γ 0 (t)dt, ~ stands for ~ndA where where ~γ (t) is a parametrization of C, and dS ~n is the unit normal and dA is the surface element at a point on the ~ surface. If a surface is parametrized by a vector function X(u, v) 3 ~=X ~u × X ~ v du dv. into R , then dS The following is an interesting corollary to Stokes’ Theorem. Corollary 4.1.4. Let C be a simple, regular, closed space curve param-

etrized by ~γ : I → R3 . There exists no function f : R3 → R of class ~ = T~ (t) at all points of the C 2 such that the gradient satisfies ∇f curve. Proof: Recall that for all functions f : R3 → R of class C 2 , the curl ~ × ∇f ~ = ~0. If S is any of the gradient is identically 0, namely, ∇ orientable piecewise regular surface that has C as a boundary, then ZZ ZZ ~ ~ ~ ~ = ~0. ~0 · dS (∇ × ∇f ) · dS = S

S

4.1. Basic Properties

105

~ = T~ (t) at all points of the curve C, then If f did satisfy ∇f Z C

~ · d~s = ∇f

Z

T~ (t) · T~ (t)s0 (t)dt =

I

Z ds, C

which is the length of the curve. By Stokes’ Theorem, this leads to a contradiction, assuming the curve has length greater than 0.  Example 4.1.5. Stokes’ Theorem is a very powerful theorem but, as

this example illustrates, it is important to check that the conditions are satisfied. Consider the vector field F~ (x, y, z) defined by F~ (x, y, z) =

 0, −

z y , 2 2 2 y + z y + z2



~ and the curve C parametrized by X(t) = (cos(2t), cos t, sin t). The ~ ~ curl of the vector field F is ∇ × F = (0, 0, 0) so the flux through any surface S over which F~ is C 1 is ZZ ~ = 0. ∇ × F~ · dS S

On the other hand, the vector field is defined along C and the circulation of F~ along C is Z F~ · d~s = 2π. C

This may seem like a contradiction because there certainly exists a surface S whose boundary is C. However, we cannot forget to observe that F~ is well-defined on all of R3 except for the x-axis. The reason this example does not give a contradiction to Stokes’ Theorem is because there is no surface S whose boundary is C and which does not intersect the x-axis. Intuitively speaking, the curve C and the x-axis are “linked,” in the sense that there is no way to move and deform one curve continuously in such a way as to move C and the x-axis into two separate half-spaces without making the curves intersect. We discuss linked curves more in Section 4.3.

106

4. Curves in Space: Global Properties

Problems 4.1.1. Verify the flux and circulation calculations in Example 4.1.5. 4.1.2. Let α ~ : I → R3 be a regular, closed space curve, and let p~ be a point. Prove that a point t0 of maximum distance on the curve away from p~ is such that α ~ (t0 ) − p~ is in the normal plane to the curve at α ~ (t0 ). 4.1.3. Recall that the diameter of a curve α ~ is the maximum of the function f (t, u) = k~ α(t) − α ~ (u)k. Use the second derivative test in multivariable calculus to show that if (t0 , u0 ) gives a diameter of a space curve, then the following hold: ~ (t0 ) is in the intersection of the normal (a) The vector α ~ (u0 ) − α planes of α(t0 ) and α(u0 ). ~ (t0 ) is on the side of the rectifying plane (b) The vector α ~ (u0 ) − α of α ~ (t0 ) that makes an acute angle with P~ (t0 ) (and similarly for α(u0 )). ~ (u0 )k is greater than or equal to (c) The diameter k~ α(t0 ) − α max{1/κ(t0 ), 1/κ(u0 )}. [Hint: One can assume that α ~ is parametrized by arc length. Also, the extrema of f (t, u) occur at and have the same properties as the extrema of g(t, u) = f (t, u)2 .] 4.1.4. Prove Proposition 4.1.2. 4.1.5. By using the vector field F~ (x, y, z) = (0, x, 0), Stokes’ Theorem establishes the familiar Green’s Theorem formula for area of the interior of a curve C in the xy-plane: Z Z A = (0, x, 0) · d~s = x dy. C

C

Suppose that we consider a curve C on the sphere of radius R and centered at the origin. Prove that there does not exist a vector field F~ (x, y, z) in R3 that could be used to calculate the area of the “interior” of the curve as a line integral. [Hint: Recall relations among vector differential operators.]

4.2 Indicatrices and Total Curvature ~ : I → R3 , define the tangent, Definition 4.2.1. Given a space curve X principal, and binormal indicatrices respectively, as the loci of the

4.2. Indicatrices and Total Curvature

107

~ space curves on the unit sphere given by T~ (t), P~ (t), and B(t), as defined in Section 3.2. Since the vectors of the Frenet frame have a length of 1, the indicatrices are curves on the unit sphere S2 in R3 . In contrast to the tangent indicatrix for plane curves, there are more possibilities for curves on the sphere than for curves on a circle. Therefore, results such as the theorem on the rotation index or results about the winding number do not have an immediate equivalent for curves in space. Example 4.2.2 (Lines). The tangent indicatrix of a line is a single point

~v /k~v k on the sphere, where the line is parametrized by ~ X(t) = ~v t + p~. Example 4.2.3 (Helices). Consider a helix around an axis L through

the origin. The tangent indicatrix of this helix is the circle on the unit sphere S2 , given as the intersection of S2 with the plane perpendicular p to L at a distance of 1/ 1 + (κ/τ )2 from the origin, where κτ is the pitch of the helix. Notice that this result is true for general helices (see Example 3.2.5) and not just circular helices. Example 4.2.4 (Space Cardioid). The space cardioid, given by the pa-

rametrization ~ X(t) = ((1 − cos t) cos t, (1 − cos t) sin t, sin t) ,

t ∈ [0, 2π],

is a closed curve. The tangent indicatrix is again a closed curve and is shown in Figure 4.1. The figure shows that the tangent indicatrix has a double point. In Problem 4.2.3, the reader is invited to show this and a few other interesting properties of the tangent indicatrix of the space cardioid. Definition 4.2.5. The total curvature of a closed curve C parametrized

~ : [a, b] → R3 is by X Z

Z κ ds =

C

b

κ(t)s0 (t) dt.

a

This is a nonnegative real number since κ(t) ≥ 0 for space curves.

108

4. Curves in Space: Global Properties

~ 2) X(t z ~ 1) X(t T~ (t1 )

y

T~ (t2 ) x

Figure 4.1. The space cardioid and its tangent indicatrix.

Though we cannot define the concept of a rotation index, Fenchel’s Theorem gives a lower bound for the total curvature of a space curve. Since T~ 0 = s0 κP~ , the speed of the tangent indicatrix is ~ is the length of the s0 (t)κ(t). Therefore, the total curvature of X tangent indicatrix. One can also note that the tangent indicatrix has a critical point where κ(t) = 0. Theorem 4.2.6 (Fenchel’s Theorem). The total curvature of a regular closed space curve C is greater than or equal to 2π. It is equal to 2π if and only if C is a convex plane curve.

Before we prove Fenchel’s Theorem, we need to discuss the concept of distance between points on the unit sphere S2 . We can define the distance between two points p and q on S2 as d(p, q) = inf{length(Γ) | Γ is a curve connecting p and q}. Let O be the center of the unit sphere. It is not difficult to show that the path connecting p and q of shortest distance is an arc between p and q on the circle defined by the intersection of the sphere and the plane containing O, p, and q. (We will obtain this result in Example 8.4.8 when studying geodesics, but it is possible to prove this claim

4.2. Indicatrices and Total Curvature

109

B

A O Figure 4.2. Spherical distance.

without the techniques of geodesics.) Then the spherical distance between two points on the unit sphere is d(p, q) = cos−1 (p · q), where we view p and q as vectors in R3 . We will denote by AB the Euclidean distance between two points A and B as elements in R3 . If A, B ∈ S2 , we use the notation AB to denote the distance between A and B on the sphere. Since OAB forms an isosceles triangle (see Figure 4.2), we see that the spherical and Euclidean distance are related via     AB −1 AB AB = 2 sin and AB = 2 sin . 2 2 Lemma 4.2.7 (Horn’s Lemma). Let Γ be a regular closed curve on the unit sphere S2 . If Γ has length less than 2π, then there exists a great circle C on the sphere such that Γ does not intersect C.

Proof: The proof we give to this lemma is due to R. A. Horn . Let A and B be two points of Γ that divide the curve into two arcs of equal length L/2. The points A and B divide Γ into two curves Γ1 and Γ2 of equal length. Since L < 2π, the spherical distance between A and B is less than the length of Γ1 , which is strictly less than π. Consequently, P and Q are not antipodal (i.e., the segment [P, Q] is not a diameter of the sphere) and therefore there exists a

110

4. Curves in Space: Global Properties

unique point M on the minor arc from P to Q midway between P and Q. We claim that Γ does not meet the equator C that has M as the north pole, so the curve lies in the hemisphere centered at M . To show that Γ1 does not meet the equator consider a copy Γ01 of Γ1 rotated one half turn about M . The curve Γ01 is a curve from B to A of length L/2. Define the closed curve Γ00 as the curve that follows Γ1 from A and B and then follows Γ01 from B back to A. The curve Γ00 has length L. Furthermore, if Γ1 intersected C, then so would Γ01 . Hence, Γ00 would contain antipodal points R and R0 . But the spherical distance RR0 = π. Therefore, the distance from R to R0 along Γ00 would be greater than or equal to π and, by symmetry of Γ00 , similarly for the path from R0 to R. This contradicts the fact that the length of Γ00 is less than 2π. Thus any curve with length less than 2π lies in an open hemisphere.  We are now in a position to prove Fenchel’s Theorem. Proof (of Theorem 4.2.6): Let ~γ : [a, b] → R3 be a parametrization for the regular closed curve C, and let p be any point on the unit sphere. Consider the function g(t) = p · ~γ (t). Since [a, b] is a closed and bounded interval and g(t) is continuous, then it attains a maximum and minimum value in [a, b]. This value occurs where g 0 (t) = p · ~γ 0 (t) = 0. We remark that the set of points ~x ∈ S2 such that p · ~x = 0 is the great circle on S2 that is on the plane perpendicular to the line (Op), where O is the center of the sphere. Since ~γ 0 (t) and T~ (t) are collinear and since p was chosen arbitrarily, we conclude that the tangent indicatrix intersects every great circle on S2 . Consequently, by Lemma 4.2.7, the length of the tangent indicatrix is greater than or equal to 2π. Thus, since the length of the tangent indicatrix is the total curvature, Z Z b 0 κ(t)s (t) dt = κ ds ≥ 2π. a

C

To prove the second part of the theorem, first note that if ~γ (t) traces out a convex plane curve, then κ(t) = |κg (t)| for ~γ as a plane

4.2. Indicatrices and Total Curvature

111

curve. Furthermore, by Propositions 2.4.3 and 2.2.11, the total curvature is 2π. Therefore, to finish proving the theorem, we only need to prove the converse. Suppose that the total curvature of ~γ is 2π. The length of the tangent indicatrix is therefore 2π. Furthermore, since by the above reasoning the tangent indicatrix must intersect every great circle, the tangent indicatrix must itself be a great circle. Thus T~ (t), T~ 0 (t), and T~ 00 (t) are coplanar and so τ (t) = 0. Thus, by Proposition 3.2.7, ~γ (t) is planar. In this case, it is not hard to check that κ(t) = |κg (t)|, where κg (t) is the curvature of ~γ (t) as a plane curve. Thus, Z 2π = |κg | ds C

is the total distance that T~ (t) travels on the unit circle, and this is the length of the unit circle. Consequently, if t1 , t2 ∈ [a, b), with T~ (t1 ) = T~ (t2 ), then T~ (t) is constant over [t1 , t2 ] because, otherwise, the total distance T~ (t) travels on the unit circle would exceed 2π by at least Z t2

|κg (t)| ds. t1

We conclude then that ~γ has no bitangent lines, and by Problem 2.4.1, we conclude that C is a convex plane curve.  As an immediate corollary, we obtain the following result. Corollary 4.2.8. If a regular, closed space curve has a curvature func-

tion κ that satisfies 1 , R then C has length greater than or equal to 2πR. κ≤

Proof: Suppose that a regular, closed space curve has a curvature function κ(s), given in terms of arc length, that satisfies the given hypothesis. Then if L is the length of the curve, we have Z L Z L Z L ds ≥ Rκ(s) ds = R κ(s) ds ≥ 2πR, L= 0

0

0

where the last inequality follows from Fenchel’s Theorem.



112

4. Curves in Space: Global Properties

Though it is outside the scope of our current techniques, we wish to state here Jacobi’s Theorem since it is a global theorem on space curves. The proof follows as an application of the Gauss-Bonnet Theorem (Problem 8.3.6). Theorem 4.2.9 (Jacobi’s Theorem). Let α ~ : I → R3 be a closed, regular

parametrized curve whose curvature is never 0. Suppose that the principal normal indicatrix P~ : I → S2 is simple. Then the locus P~ (I) of the principal normal indicatrix separates the sphere into two regions of equal area. The reader should note that Jacobi’s Theorem is quite profound in the following sense. Given a parametrized curve ~γ : I → R3 such that k~γ (t)k = 1 for all t ∈ I, the problem of calculating the surface area of the sphere lying on one side or the other of the locus of this curve is not a tractable problem. We now present Crofton’s Theorem, which we will use in the next section to prove a theorem by Fary and Milnor on the total curvature of a knot. Let O be the center of the unit sphere S2 . Each great circle C is uniquely defined by a line L through the origin of the sphere as the intersection between S2 and the plane through O perpendicular to L. Furthermore, L is defined by two “poles,” the points of intersection of L with S2 . However, there exists a bijective correspondence between oriented great circles and points on S2 by associating to C the pole of L that is in the direction on L that is positive in the sense of the right-hand rule of motion on C. Consider a set Z of oriented great circles on S2 . We call the measure m(Z) of the set Z the area of the region traced out on S2 by the positive poles of the oriented circles in Z. Theorem 4.2.10 (Crofton’s Theorem). Let Γ be a curve of class C 1 on

S2 . The measure of the great circles of S2 that meet Γ is equal to four times the length of Γ. Proof: Suppose that ~e1 : [0, L] → S2 parametrizes Γ by arc length. Complete {~e1 (s)} to form an orthonormal basis {~e1 (s), ~e2 (s), ~e3 (s)} so that ~ei (s) for i = 2, 3 are of class C 1 . Without loss of generality, we can construct ~e2 and ~e3 so that det(~e1 , ~e2 , ~e3 ) = 1

4.2. Indicatrices and Total Curvature

113

for all s ∈ [0, L]. Using the same reasoning that established Equation (3.5) or the result of Problem 9.1.10, we have   0 a2 a3   d ~e1 ~e2 ~e3 = ~e1 ~e2 ~e3 −a2 0 a1  (4.2) ds −a3 −a1 0 for some continuous functions ai : [0, L] → R. Furthermore, since ~e1 is parametrized by arc length we know that it has unit speed, so a2 (s)2 + a3 (s)2 = 1. Consequently, there exists a function α : [0, L] → R of class C 1 such that ~e10 (s) = cos(α(s))~e2 + sin(α(s))~e3 . The set of oriented great circles that meet Γ at ~e1 (s) is parametrized by its positive poles (cos θ)~e2 (s) + (sin θ)~e3 (s) for θ ∈ [0, 2π]. Therefore, the region traced out by the positive poles of oriented great circles that meet Γ is parametrized by ~ (s, θ) = (cos θ)~e2 (s) + (sin θ)~e3 (s) Y

for (s, θ) ∈ [0, L] × [0, 2π].

We wish to determine the area element |dA| for the set of poles traced out by the set of oriented great circles meeting Γ. However, ~s × Y ~θ k dθ ds, |dA| = kY so after some calculation and using Equation (4.2), one obtains |dA| = |a2 (s) cos θ + a3 (s) sin θ| dθ ds = | cos(α(s) − θ)| dθ ds. ~ (θ, s) and Call Cθ,s the oriented great circle with positive pole Y denote by n(Cθ,s ) the number of points in Cθ,s ∩ Γ. Then the measure m of oriented great circles in S2 that meet Γ is ZZ Z L Z 2π m= n(Cθ,s )|dA| = | cos(α(s) − θ)| dθ ds. (4.3) 0

0

However, for all fixed α0 , we have Z 2π | cos(α0 − θ)| dθ = 4, 0

so we conclude that m = 4L, which establishes the theorem.



114

4. Curves in Space: Global Properties

Problems ~ 4.2.1. Consider the twisted cubic X(t) = (t, t2 , t3 ) with t ∈ R. (a) Prove that the closure of the tangential indicatrix of the twisted cubic is a closed curve, in particular, that lim T~ (t) = lim T~ (t) = (0, 0, 1).

t→∞

t→−∞

(b) Prove that the tangential indicatrix does not have a corner at (0, 0, 1), namely, that lim T~ 0 (t) = lim T~ 0 (t).

t→∞

t→−∞

(c) Prove directly that the total curvature of the twisted cubic is bounded. ~ 4.2.2. Consider the helix X(t) = (a cos t, a sin t, bt) for t ∈ R. Determine ~ the locus of the tangent indicatrix of X(t). 4.2.3. Figure 4.1 appears to show that the tangent indicatrix of the space cardioid has a double point. (a) Find the value t0 of t at which the double point occurs. ~ 0 ) is also on ~ 0 ), i.e., −X(t (b) Show that the antipodal point to X(t the tangent indicatrix of the space cardioid. (c) Show that the curve obtained by projecting the tangent indica~ 0 ) through the origin trix onto the plane perpendicular to X(t has three cusps. 4.2.4. Calculate directly the total curvature of the space cardioid ~ X(t) = ((1 − cos t) cos t, (1 − cos t) sin t, sin t)

for t ∈ [0, 2π].

4.2.5. Prove Fenchel’s Theorem as a corollary to Crofton’s Theorem. 4.2.6. (*) Let α ~ : [0, L] → R3 be a regular closed curve (parametrized by arc length) whose image lies on a sphere. Suppose also that κ(t) 6= 0. Prove that Z L τ (s) ds = 0. 0

4.3 Knots and Links The reader should be forewarned that the study of knots and links is a vast and fruitful area that one usually considers as a subbranch of topology. In this section, we only have space to give a cursory introduction to the concept of a knotted curve in R3 or two linked curves in R3 . However, the property of being knotted or linked is a global property of a curve or curves (since it depends on the curve as a whole), which motivates us to briefly discuss these topics in this chapter. This section presents two main theorems: the Fary-Milnor Theorem, which shows how the property of knottedness imposes a condition on the total curvature of a curve, and Gauss’s formula for the linking number of two curves.

4.3.1 Knots Intuitively speaking, a knot is a simple closed curve in R3 that cannot be deformed into a circle without breaking the curve and reconnecting it. In other words, a knot cannot be deformed into a circle by a continuous process without passing through a stage where it is not a simple curve. The following gives a precise definition to the above intuition. Definition 4.3.1. A simple closed curve Γ in R3 is called unknotted if

there exists a continuous function H : S1 × [0, 1] → R3 such that H(S1 × {0}) = Γ and H(S1 × {1}) = S1 and such that Γt = H(S1 × {t}) is a curve that is homeomorphic to a circle. If there does not exist such a function H, the curve Γ is called knotted . The function H described in the above definition is called a homotopy in R3 between Γ and a circle S1 . Figure 4.3 illustrates four intermediate stages of a homotopy between a space curve and a circle. Figure 4.4 shows the trefoil knot realized as a curve in space, along with a two-dimensional diagram. As it is somewhat tedious to plot general knotted curves, even with the assistance of a computer algebra system, one often uses a diagram that shows the “crossings,” i.e., which part of the curve passes above the other whenever the diagram would intersect in the given perspective. (The interested

115

116

4. Curves in Space: Global Properties

Figure 4.3. An unknotting homotopy in R3 .

Figure 4.4. A trefoil knot and its diagram.

reader is encouraged to read  for an advanced introduction to knot theory.) The following theorem gives a necessary relationship between a knotted curve and the curvature of a space curve. Theorem 4.3.2 (Fary-Milnor Theorem). The total curvature of a knot is

greater than or equal to 4π.

117

~ (s, θ) Y

α ~ (s)

T~ (s)

Figure 4.5. Proof of the Fary-Milnor Theorem.

Proof: Let ~x : [0, L] → R3 be a closed, regular space curve parametrized by arc length, and let T~ : [0, L] → S2 be the tangent indicatrix. Call C the image of ~x in R3 and Γ the image of T~ on the sphere. Recall the notations in the proof of Crofton’s Theorem (Theorem 4.2.10) and, in particular, that n(Cθ0 ,s0 ) is the number of times that the great circle Cθ0 ,s0 intersects the curve Γ (see Figure 4.5). By part of the proof of Crofton’s Theorem, n(Cθ0 ,s0 ) > 0 over the domain (θ, s) ∈ [0, 2π] × [0, L], and n(Cθ0 ,s0 ) is even. ~ (θ0 , s0 ) · ~x(s). Then the derivaDefine a height function h(s) = Y 0 ~ ~ tive is h (s) = Y (θ0 , s0 )·T (s) so that the relative maxima and minima of h(s) occur where Γ intersects Cθ0 ,s0 , so n(Cθ0 ,s0 ) is the number of critical points of h(s). Suppose that the total curvature of the curve ~x, which is also the length of Γ, is less than 4π. By Equation (4.3) in the proof of Crofton’s Theorem, ZZ n(Cθ,s )|dA| < 16π. Since the area of the sphere is 4π, there exists (θ0 , s0 ) such that n(Cθ0 ,s0 ) = 2, which means that the height function corresponding ~ (θ0 , s0 ) has two critical points, namely, one maximum and one to Y

118

4. Curves in Space: Global Properties

minimum occuring at s1 and s2 . The points ~x(s1 ) and ~x(s2 ) partition C into two curves over which the height function h(s) is monotonic, one increasing and the other decreasing. Consequently, any plane ~ (θ0 , s0 ) (that intersects ~x between the points ~x(s1 ) perpendicular to Y and ~x(s2 ) of extremal height function) intersects ~x in exactly two points. This fact allows us to construct a homotopy between C and a circle, which we describe below. Call v the height parameter for a point on the curve ~x(s). Call vmin and vmax the minimum and maximum values of h(s) over the ~ (θ0 , s0 ) is dedomain [0, L] of ~x. Every plane perpendicular to Y termined uniquely by the height parameter v as the unique plane ~ (θ0 , s0 ) and going through the point v Y ~ (θ0 , s0 ). perpendicular to Y According to the discussion in the previous paragraph, the plane Pv ~ (θ0 , s0 ) at height v intersects C in two points. perpendicular to Y Therefore, C can be expressed as the union of the locus of two continuous curves ~γ1 : [vmin , vmax ] −→ R3 , ~γ2 : [vmin , vmax ] −→ R3 , such that Pv ∩ C = {~γ1 (v), ~γ2 (v)}. Note that ~γ1 (vmax ) = ~γ2 (vmax ) = ~x(s1 ),

and

~γ1 (vmin ) = ~γ2 (vmin ) = ~x(s2 ). Call ~γ (v) = 12 (~γ1 (v) + ~γ2 (v)) the midpoint of the pair Pv ∩ C = {~γ1 (v), ~γ2 (v)} (which degenerates to a singleton set for v = vmin or vmax ). ~ (θ0 , s0 )} to an orthonormal basis {Y ~ (θ0 , s0 ), ~e2 , ~e3 }. Complete {Y In this basis, we can write ~γ (v) as ~γ (v) = (v, a(v) cos(α(v)), a(v) sin(α(v))) for some functions a(v) ≥ 0 and α(v) defined on [vmin , vmax ]. Furthermore, since ~γ (v), ~γ1 (v), and ~γ2 (v) as position vectors all lie in Pv , there exist functions b(v) and θ(v) such that we can write ~γ1 (v) = ~γ (v) + b(v) ((cos θ(v))~e2 + (sin θ(v))~e3 ) , ~γ2 (v) = ~γ (v) − b(v) ((cos θ(v))~e2 + (sin θ(v))~e3 ) .

119

Call v0 = 12 (vmin + vmax ) and R = 12 (vmax − vmin ) and define the function v(u) = v0 + R cos(u). Finally, define the function H : [0, 2π] × [0, 1] by ~ (θ0 , s0 ) H(u, t) = v(u)Y + (1 − t)a(v(u)) cos(α(v(u))) + sin u ((1 − t)r(u) + tR) cos ((1 − t)θ(v(u))) ~e2 + (1 − t)a(v(u)) sin(α(v(u))) + sin u ((1 − t)r(u) + tR) sin ((1 − t)θ(v(u))) ~e3 , where r : [0, 2π] → R≥0 is a function such that r(u)| sin u| = b(v(u)). We now leave it as an exercise to the reader to prove that H(u, t) is a homotopy between C and a circle of radius R such that for all t0 ∈ (0, 1), the image of H(u, t0 ) is homeomorphic to a circle (Problem 4.3.1). By Definition 4.3.1, the existence of H as defined above allows us to conclude that C is unknotted. Therefore, if C is a knot, we conclude that the total curvature of C is greater than or equal to 4π. 

4.3.2 Links A simple, closed space curve in itself resembles a circle. Knottedness is a property that concerns how a simple closed curve “sits” in the ambient space. The technical language for this scenario is that any simple closed curve is homeomorphic to a circle, but knottedness concerns how the curve is embedded in R3 . In a similar way, the notion of linking between two simple closed curves is a notion that considers how the curves are embedded in space in relation to each other. Intuitively speaking, we would like the linking number between two simple closed curves C1 and C2 in R3 to be the minimum number of crossings required in order to continuously move the curves in space (including deformations that do not break the curves) until they are separated by a plane in R3 . We cannot use this intuition as a definition since counting the number of times it takes to break one curve in order to pass through another is too vague. Figure 4.6 illustrates a basic scenario of linked curves. The figure on the left shows a rendering of two thick circles in space that are

(4.4)

120

4. Curves in Space: Global Properties

linked. Our intuitive picture leads us to think of this curve as having a single link. Since it is difficult to accurately sketch curves in R3 and to see which curve is in front of which, it is common to use a diagram to depict the curves. The diagram on the right is obtained by projecting the curves into a plane (perpendicular to some vector ~n), careful to show the white space along the curve diagram to communicate which curve is above (in reference to ~n) the other at the intersections that occur when projecting. The diagram is called the link diagram. In order to give a definition of the linking number, we need to give orientations to the curves. The standard definition of the linking number of two curves uses the link diagram along with the sign associated to each crossing, as shown below: −1

+1

The arrows indicate the orientation on the respective curves at a crossing point. In Figure 4.6, the picture on the left makes no reference to orientation, but the diagram on the right imposes an orientation by the depicted arrows. Definition 4.3.3. The linking number link(C1 , C2 ) of two simple closed

oriented curves C1 and C2 is half the sum of the signs of all the crossings in the diagram of the pair (C1 , C2 ). For example, in the oriented diagram on the right in Figure 4.6, both crossings have sign +1. Thus half the sum of the signs of the crossings is +1.

One of the surprising results about linking is Gauss’s formula for the linking number between two curves. Though linking and the linking number are obviously global properties of the curves, Gauss’s formula calculates the linking number in terms of the parametrizations of the curves, thereby connecting the global property to local properties. Theorem 4.3.4 (Gauss’s Linking Formula). Let C1 and C2 be two closed,

regular space curves parametrized by α ~ : I → R3 and β~ : J → R3 , respectively. The linking number between C1 and C2 is Z Z ~ det(~ α(u) − β(v), α ~ 0 (u), β~ 0 (v)) 1 du dv. (4.5) link(C1 , C2 ) = 3 ~ 4π I J k~ α(u) − β(v)k Though we cannot give a proof for this formula at this time, we briefly sketch a few of the concepts and techniques that go into it. In Section 2.2, we introduced the notion of degree for a continuous map between circles, f : S1 → S1 . Intuitively speaking, the degree of f counts how many times f covers its codomain and with what orientation. The degree of f takes into account when f might double back. For example, if q ∈ S1 has six preimages from f and, for four of the preimages, f passes through q with the same orientation as on the domain and, for the other two preimages, f passes through q in an opposite orientation, then the degree of f is 4 − 2 = 2. Similarly (and leaving many of the technical details for later), one can define the degree of a continuous map between spheres f : S2 → S2 as how often f covers S2 . Again, one should note the difficulty inherent in making this definition precise since one must take into account a form of orientation and doubling back, i.e., whether f folds back over itself over some region of the codomain. In the same way, one can define the degree of a continuous map F : S → S2 , where S is a regular surface without boundary. (We give the definition for a regular surface in Chapter 5.) The conditions that S is regular and has no boundary guarantee that F generically covers S2 by the same amount at all its points, so long as we take into account orientation and assign a negative covering value to F when F covers S2 negatively. As discussed in Definition 2.2.7 and in the following, it is more appropriate to view the parametrization of a simple closed curve C

121

122

4. Curves in Space: Global Properties

as a continuous function ~x : S1 → R3 . So consider two simple closed curves with parametrizations α ~ : S1 → R3 and β~ : S1 → R3 . The function Ψ : S1 × S1 → S2

given by

Ψ(t, u) =

~ β(u) −α ~ (t) ~ kβ(u) −α ~ (t)k

defines a continuous function from the torus S1 × S1 to the unit sphere S2 . The technical definition of the linking number of two curves is the degree of Ψ(t, u). The proof of Theorem 4.3.4 consists of calculating the degree of Ψ(t, u). Equation (4.5) is difficult to use in general, and in the exercises, we often content ourselves with using a computer algebra system to calculuate the integrals. Some basic facts about linking are not difficult to show with the appropriate topological background but are difficult using Equation (4.5). For example, if two simple closed curves can be separated by a plane, then link(C1 , C2 ) = 0. This fact is simple once one has a few facts about the degrees of maps to S2 , but using Gauss’s formula to show the same result is a difficult problem.

Problems 4.3.1. This exercise finishes the proof of the Fary-Milnor Theorem. Consider the function H(u, t) defined in Equation (4.4). (a) Prove that, for all t0 ∈ [0, 1], H(u, t0 ) is an injective function for u ∈ (0, 2π) and that H(0, t0 ) = H(2π, t0 ). (b) Show that the locus of H(u, 1) is a circle. (c) Show that the locus of H(u, 0) is the curve C. (d) Use the previous parts of the exercise to conclude that H(u, t) is a homotopy between C and a circle. 4.3.2. Consider the trefoil knot parametrized by α ~ (t) = ((3 + cos 3t) cos 2t, (3 + cos 3t) sin 2t, sin 3t) . Using a CAS, calculate the total curvature of this trefoil knot. Show how to create another simple closed curve that is homotopic to this trefoil knot, with a total curvature of 4π + ε for any ε > 0.

123

124

4. Curves in Space: Global Properties

that if a rigid motion is applied to two simple closed curves C1 and C2 , then their linking number does not change. 4.3.8. Consider the two simple closed curves α ~ (t) = (3 cos t, 3 sin t, 0), for t ∈ [0, 2π], ~ β(t) = ((3 + cos(nt)) cos t, (3 + cos(nt)) sin t, sin(nt)) , for t ∈ [0, 2π]. (a) Give the link diagram of these two space curves, including the orientation, and show that the linking number of these two curves is n. (b) Gauss’s formula in Equation (4.5) is quite difficult to use, but, using a computer algebra system, give support for the above answer.

CHAPTER 5

Regular Surfaces

5.1 Parametrized Surfaces There are many approaches that one can take to introduce surfaces. Some texts immediately build the formalism of differentiable manifolds, some texts first encounter surfaces in R3 as the solution set to an algebraic equation F (x, y, z) = 0 with three variables, and other texts present surfaces as the images of vector functions of two variables. In this section, we introduce the last of these three options and occasionally refer to the connection with the surfaces as solution sets to algebraic equations. Only later will we show why one requires more technical definitions to arrive at a workable definition that matches what one typically means by a “surface.” We imitate the definition of parametrized curves in Rn and begin our study of surfaces (yet to be defined) by considering continuous ~ : U → R3 , where U is a subset of R2 . Below are some functions X examples of such functions whose images in R3 are likely to appear in a multivariable calculus course. Example 5.1.1 (Planes). If ~a and ~b are linearly independent vectors in

R3 , then the plane through the point p~ and parallel to the vectors ~a and ~b can be expressed as the image of the following function: ~ X(u, v) = p~ + u~a + v~b for (u, v) ∈ R2 . In other words, we can write ~ X(u, v) = (p1 + a1 u + b1 v, p2 + a2 u + b2 v, p3 + a3 u + b3 v) for constants pi , ai , bi , with 1 ≤ i ≤ 3 that make (a1 , a2 , a3 ) and (b1 , b2 , b3 ) noncollinear vectors.

125

126

5. Regular Surfaces

Example 5.1.2 (Graphs of Functions). If z = f (x, y) is a real function

of two variables defined over some set U ⊂ R2 , one can obtain the ~ : U → R3 by graph of this function as the image of a function X setting ~ X(u, v) = (u, v, f (u, v)). Example 5.1.3 (Spheres). One can obtain a sphere of radius R centered

at the origin as the image of ~ X(u, v) = (R cos u sin v, R sin u sin v, R cos v), with (u, v) ∈ [0, 2π] × [0, π]. This expression should not appear too mysterious since it merely puts the equation of spherical coordinates r = R into Cartesian coordinates, with u = θ and v = φ. This method of parametrizing the sphere is the astronomical one, where v measures the colatitude, i.e., the angle down from the north pole. (Unfortunately, there is no uniformity in the parametrization used for spheres in calculus books.) Example 5.1.4 (Conics). Besides getting the sphere, one could also ob~ : U → R2 tain an ellipsoid as the image of some vector function X by modifying the coefficients in front of the x, y, and z coordinate functions of the parametrization for the sphere. In particular, the parametrization

~ X(u, v) = (a cos u sin v, b sin u sin v, c cos v), with (u, v) ∈ [0, 2π] × [0, π] has for its image the ellipsoid of equation x2 y 2 z 2 + 2 + 2 = 1. a2 b c One can obtain all of the other conic surfaces as follows. The circular cone x2 y 2 z2 + 2 = 2 2 a a b is the image of ~ X(u, v) = (au cos v, au sin v, bu),

with (u, v) ∈ R × [0, 2π].

5.1. Parametrized Surfaces

127

Figure 5.1. Hyperboloid of one sheet.

The function ~ X(u, v) = (cosh u cos v, cosh u sin v, sinh u),

with (u, v) ∈ R×[0, 2π],

traces out the hyperboloid of one sheet. (See Figure 5.1 for two perspectives of a hyperboloid of one sheet.) For the hyperboloid of two sheets, we need two separate functions ~ X(u, v) = (sinh u cos v, sinh u sin v, cosh u) and ~ X(u, v) = (sinh u cos v, sinh u sin v, − cosh u). Example 5.1.5 (Surfaces of Revolution). A surface of revolution is a set

S in R3 obtained by rotating, in R3 , a regular plane curve C in a plane P about a line L in P that does not meet C. The curve C is called the generating curve, and the line l is called the rotation axis. The circles described by the rotation locus of the points of C are called the parallels of S, and the various positions of C are called the meridians of S. A surface of revolution can be parametrized

128

5. Regular Surfaces

Figure 5.2. Surface of revolution.

naturally by the parameter on the curve C and the angle of rotation u about the axis. Let ~x : I → R2 be a parametric curve in the xy-plane, with ~x(t) = (x(t), y(t)), and suppose we want to find the surface of revolution of this curve about either the x- or y-axis. About the y-axis, we take ~ u) = (x(t) cos u, y(t), x(t) sin u), and about the x-axis, we would X(t, ~ u) = (x(t), y(t) cos u, y(t) sin u). Both of these options have take X(t, domains of I × [0, 2π]. (See Figure 5.2 for an example of a surface of revolution about the x-axis based on a curtate cycloid.) Looking more closely at Example 5.1.3, one should note that ~ though the function X(u, v) maps onto the sphere of radius R centered at the origin, this function exposes two potential problems with ~ is not injective. the intuitive approach. The first problem is that X In particular, for any point (x, y, z) on this sphere with y = 0 and x > 0, we have z v = cos−1 and u = 0 or 2π, R and so there exist two preimages for such points. Worse still, if we consider the north and south poles (0, 0, 1) and (0, 0, −1), the preimages are the points (u, 0) and (u, π), respectively, for all u ∈ [0, 2π]. With the given parametrization, the arc on the sphere defined by u = 0 is the set of points with more than one preimage. This latter remark leads to the second problem because the points on this arc have no particular geometric significance, and one would wish to avoid any formulation of a surface that confers special properties on some points that are geometrically ordinary.

5.1. Parametrized Surfaces

U

129

S

~ X ~ ) X(U

y x Figure 5.3. Coordinate patch.

Before addressing these and other issues necessary to be able to “do calculus” on a surface, we need to define the class of subsets of R3 that one can even hope to study with differential geometry. The primary criteria for such a subset is that one must be able to describe its points using continuous functions. Definition 5.1.6. A subset S ⊆ R3 is called a parametrized surface if

for each p ∈ S, there exists an open set U ⊂ R2 , an open neigh~ : U → R3 such borhood V of p in R3 , and a continuous function X ~ ) = V ∩ S. Each such X ~ is called a parametrization of a that X(U neighborhood of S. We call a parametrized surface of class C r if it ~ of class C r . can be covered by parametrizations X Note that in contrast to space curves, where we use intervals as the domain for parametrizations, this definition uses open sets. Furthermore, this definition does not insist that S be the image of a single parametrization but, rather that it be covered by images of parametrizations. ~ : U → V ∩ S, where V ∩ S is an open For any such function X ~ a parametrization of the coordinate neighborhood of p in S, we call X patch V ∩ S. (See Figure 5.3 for an illustration of the relation of the domain U ⊂ R2 and the coordinate patch on the surface V ∩ S.) Definition 5.1.6 states that, for every point p ∈ S, there is an open neighborhood of p on S that is the image of some vector function. With the tools of curves in the plane or space, one can study properties of a surface S by considering various families of curves on

130

5. Regular Surfaces

Figure 5.4. Coordinate lines on a torus.

the surface. If one wishes to study S near a point p, one could look for common properties of all the curves on S passing through p. There are two natural classes of curves on any given parametrized surface S that arise naturally. First, we can consider what are generically called the coordinate lines of the surface S. A coordinate line on S is the image of a space curve defined by fixing one of the variables in a particular parametrization of a coordinate patch of S. ~ : U → R3 parametrizes a patch of S, then a curve of Namely, if X the form ~ ~γ1 (u) = X(u, v0 ),

~ 0 , v), respectively γ~2 (v) = X(u

is called a coordinate line of the variable u, and respectively for v. (See Figure 5.4 for an example of coordinate lines of a torus ~ parametrized by X(u, v) = ((3 + cos v) cos u, (3 + cos v) sin u, sin v) for (u, v) ∈ [0, 2π] × [0, 2π].) Example 5.1.7. With the surfaces of revolution described in Example

5.1.5, the coordinate lines of t and u, respectively, are precisely the parallels and the meridians. Another family of curves we might study are the slices of S, by which we mean the intersection of S with a family of parallel planes. As an example, Figure 5.5 shows the parametric surface ~ X(u, v) = (cos v, sin 2v cos u, sin 2v sin u),

5.1. Parametrized Surfaces

131

Figure 5.5. Slices of a surface of revolution.

with domain [0, 2π] × [0, π] along with two sets of slices. Though any set of parallel planes can produce a set of slices on S, we often consider the intersection of S with planes of the form x = a, y = b, or z = c. In the figure on the left, we sliced the surface with planes x = a, which are perpendicular to the axis of revolution. These slices result in a set of circles. If we fix x = a, with −1 ≤ a ≤ 1, we get cos v = a, and there exists a value v0 ∈ [0, π] satisfying this equation. ~ Then the slice of X(u, v) intersecting the x = a plane gives a circle parallel to the yz-plane, given by (x, y, z) = (cos(v0 ), sin(2v0 ) cos u, sin(2v0 ) sin u). On the other hand, if we consider planes perpendicular to the zaxis, that is planes of the form z = c, the z = c slice corresponds to the set of points (cos v, sin(2v) cos u) where (u, v) ∈ [0, 2π] × [0, π] satisfy sin(2v) sin(u) = c. This defines a curve implicitly. In this case, the best we can do is to solve for sin u = c/ sin(2v) and then get a parametrization for the slices by (cos v, ±

q

sin2 (2v) − c2 , c).

These are shown on the right in Figure 5.5.

132

5. Regular Surfaces

Problems ~ for the cylinder {(x, y, z) ∈ R3 | x2 + y 2 = 5.1.1. Find a parametrization X ~ is 1}. Can the domain of this parametrization be chosen so that X bijective onto the cylinder? Explain why or why not. 5.1.2. Prove that the following functions provide parametrizations of the hyperbolic paraboloid x2 − y 2 = z. Provide appropriate domains of definitions. Which ones are injective? Which ones are surjective onto the surface? ~ (a) X(u, v) = (v cosh u, v sinh u, v 2 ). ~ (b) X(u, v) = ((u + v), (u − v), 4uv). ~ (c) X(u, v) = (uv, u(1 − v), u2 (2v − 1)). 5.1.3. Let α ~ : I → R2 be a regular, plane curve, and let L be a line that does not intersect the image of α ~ . Suppose that L goes through the point p~ and has direction vector ~u so that p~ + t~u is a parametrization of L. Find a formula for the parametrization of the surface of revolution obtained by rotating α ~ (I) about L. 5.1.4. Let α ~ , β~ : I → R3 be two parametrized space curves with the same ~ :I× domain. Define the secant surface by the parametrization X 3 R → R with ~ ~ u) = (1 − u)~ X(t, α(t) + uβ(t). ~ = (cos t, 2 sin t, −1). Prove (a) Let α ~ (t) = (2 cos t, sin t, 1) and β(t) that every slice by a constant u coordinate is an ellipse and give the eccentricity of the ellipse as a function of u. (Recall thatq the eccentricity of an ellipse with half-axes a > b > 0 is e=

1−

b2 a2 .)

~ (b) Now let α ~ (t) = (cos t, sin t, 1) and β(t) = (− sin t, cos t, −1). Prove that every slice by a constant u coordinate is a circle and give the radius as a function of u. ~ 5.1.5. Prove that on the unit sphere parametrized by X(θ, φ) = (cos θ sin φ, sin θ sin φ, cos φ), the curves given by (θ, φ) = (ln t, 2 tan−1 t) intersect every meridian with an angle of π4 . (Note that a meridian is a curve such that θ = const. A curve on the sphere that intersects the meridians at a constant angle is called a loxodrome.)

5.2. Tangent Planes and Regular Surfaces

5.1.6. Consider a curve in the plane α ~ (t) = (x(t), y(t)) and the surface of revolution obtained by rotating the image of α ~ about the y-axis. ~ u) = (x(t) cos(u), This surface of revolution is parametrized by X(t, y(t), x(t) sin(u)). (a) Consider u-coordinate lines, i.e., the space curves γ1 (t) = (x(t) cos(u0 ), y(t), x(t) sin(u0 )) , where u0 is fixed. Calculate the space curvature and torsion of γ1 . (b) Repeat the above question with the t-coordinate lines, i.e., the space curves γ2 (u) = (x(t0 ) cos(u), y(t0 ), x(t0 ) sin(u)), where t0 is fixed. 5.1.7. Consider the set of points S = {(x, y, z) ∈ R3 | x4 + y 4 + z 4 = 1}. Modify the usual parametrization for a sphere to find parametrizations that cover S.

5.2 Tangent Planes and Regular Surfaces In the local theory of curves, we called a curve C regular at a point p if there is a parametrization ~x(t) of C near p = ~x(t0 ) such that ~x0 (t0 ) exists and ~x0 (t0 ) 6= ~0. In Section 1.2, we saw that this definition is tantamount to requiring that lim

t→t0

~x0 (t) k~x0 (t)k

exists and, hence, that there exists a tangent line to C at p. Imitating this latter geometric property, we will eventually call a point p ∈ S regular if one can define a tangent plane to p at S, but we must delay a precise definition until after we define the tangent plane to S at p. Definition 5.2.1. Let S be a parametrized surface and p a point in S.

Consider the set of space curves ~γ : (−ε, ε) → R3 such that ~γ (0) = p, and the image of ~γ lies entirely in S. A tangent vector to S at p is any vector in R3 that can be expressed as ~γ 0 (0), where ~γ is such a space curve. From the view point of intuition, one often considers tangent vectors to S at p to be based at p. The set of tangent vectors,

133

134

5. Regular Surfaces

~γ (t)

~γ 0 (0)

Figure 5.6. Double cone.

as a subset of R3 , gives a sort of approximation of the behavior of S at p. By rescaling the parameter t, one notices that the set of tangent vectors to S at a point p contains ~0 and is closed under scalar multiplication. Hence, the set of tangent vectors will be a (usually infinite) union of lines, and, as we shall see, it is often, but not always, a plane. Example 5.2.2. As an example of a set of tangent vectors that is not

a plane, consider the cone with equation x2 + y 2 − z 2 = 0, with parametrization ~ X(u, v) = (v cos u, v sin u, v), ~ for (u, v) ∈ [0, 2π] × R (see Figure 5.6). We note that X(u, 0) = ~ (0, 0, 0) for all u ∈ [0, 2π] and consider the point p = X(u, 0) = (0, 0, 0). A curve on the cone through p has a parametric equation ~ of the form ~γ (t) = X(u(t), v(t)), where v(0) = 0 and u is any real in [0, 2π]. For such a curve, using the multivariable chain rule, we have ~γ 0 (t) =

~ dv ~ du ∂ X ~ ~ ∂X ∂X ∂X + = u0 (t) + v 0 (t) , ∂u dt ∂v dt ∂u ∂v

~ u (u, 0) = (0, 0, 0), and since X ~γ 0 (0) = v 0 (0) (cos u, sin u, 1).

5.2. Tangent Planes and Regular Surfaces

135

Since v 0 (0) and u can be any real values, we determine that the union of the endpoints the set of tangent vectors to the cone at (0, 0, 0) is precisely the cone itself. Definition 5.2.3. Let S be a parametrized surface and p a point in S. If the set of tangent vectors to S at p forms a two-dimensional subspace of R3 , we call this subspace the tangent space to S at p and denote it by Tp S. If Tp S exists, we call the tangent plane the set of points {p + ~v | ~v ∈ Tp S}.

We wish to find a condition that determines when the set of tangent vectors is a two-dimensional vector subspace. ~ : U → R3 be a Let p be a point on a surface S, and let X parametrization of a neighborhood V ∩ S of p for some U ⊆ R2 . ~ 0 , v0 ) and consider the coordinate lines ~γ1 (t) = Suppose that p = X(u ~ 0 + t, v0 ) and ~γ2 (t) = X(u ~ 0 , v0 + t) through p. Both ~γ1 and ~γ2 X(u lie on the surface and, by the chain rule, ~γ10 (0) =

~ ∂X (u0 , v0 ) ∂u

and ~γ20 (0) =

~ ∂X (u0 , v0 ). ∂v

Furthermore, for any curve ~γ (t) on the surface with ~γ (0) = p, we can write ~ ~γ (t) = X(u(t), v(t)) for some functions u(t) and v(t), with u(0) = u0 and v(0) = v0 . By the chain rule, ~γ 0 (t) = u0 (t)

~ ~ ∂X ∂X (u(t), v(t)) + v 0 (t) (u(t), v(t)), ∂u ∂v

so at p, ~γ 0 (0) = u0 (0)

~ ~ ∂X ∂X (u0 , v0 ) + v 0 (0) (u0 , v0 ) = u0 (0)~γ10 (0) + v 0 (0)~γ20 (0). ∂u ∂v ~

~

X X (u0 , v0 ) and ∂∂v (u0 , v0 ). Thus, ~γ 0 (0) is a linear combination of ∂∂u ~ is Definition 5.1.6 does not assume that the parametrization X injective. Therefore, the set of tangent vectors is a plane if and only

136

5. Regular Surfaces

if the union of all linear subspaces Span

! ~ ~ ∂X ∂X (u0 , v0 ), (u0 , v0 ) , ∂u ∂v

~ 0 , v0 ) = p is a plane. In particular, if there exists only one where X(u ~ 0 , v0 ) = p, then a tangent plane exists at p if (u0 , v0 ) such that X(u ~ ~ ∂X ∂X ∂u (u0 , v0 ) and ∂v (u0 , v0 ) are not collinear. To simplify notations, one often uses the following abbreviated notation for partial derivatives: ~ ~ u (u, v) = ∂ X (u, v) X ∂u

and

~ ~ v (u, v) = ∂ X (u, v). X ∂v

~ −1 (p), then S has One then writes succinctly that if {(u0 , v0 )} = X ~ u (u0 , v0 ) × X ~ v (u0 , v0 ) 6= ~0. a tangent plane at p if and only if X Furthermore, the tangent plane to S at p is the unique plane through ~ u (u0 , v0 ) × X ~ v (u0 , v0 ) 6= ~0. This leads us to p with normal vector X a formula for the tangent plane. Proposition 5.2.4. Let S be a surface parametrized near a point p by ~ : U → R3 . Suppose also that X ~ is injective at p (i.e., p has X ~ only one preimage) and that p = X(u0 , v0 ). Then the set of tangent vectors forms a two-dimensional subspace of R3 if and only if ~ u (u0 , v0 ) × X ~ v (u0 , v0 ) 6= ~0. In this case, the tangent plane to S at X p satisfies the following equation for position vectors ~x = (x, y, z):

~ u (u0 , v0 ) × X ~ v (u0 , v0 )) = 0. ~ 0 , v0 )) · (X (~x − X(u

(5.1)

~ is not injective at p Through an abuse of language, even if X ~ u (u0 , v0 ) × X ~ v (u0 , v0 ) 6= 0 for some (u0 , v0 ) ∈ U , we will call and X ~ at q = (u0 , v0 ). Equation (5.1) the equation of the tangent plane to X ~ 0 , v0 ) and S is the image This is an abuse of language since if p = X(u ~ of X, then p might have more than one preimage and, therefore, the set of tangent vectors to S of p might not be a plane but be a union of planes. Problem 5.2.9 gives an example of a parametrized surface that has points where the set of tangent vectors is the union of two planes.

5.2. Tangent Planes and Regular Surfaces

Figure 5.7. A tangent plane.

~ Example 5.2.5. Consider the function X(u, v) = (u, v, 4 + 3u2 − u4 − 2v 2 ). We calculate ~u × X ~ v = (1, 0, 6u − 4u3 ) × (0, 1, −4v) X = (−6u + 4u3 , 4v, 1). Figure 5.7 shows the surface along with the tangent plane at the ~ ~ point X(−1.4, −1). At this specific point, we have p = X(−1.4, −1) = ~ ~ (−1.4, −1, 4.0384) and Xu × Xv (−1.4, −1) = (−2.576, −4, 1). Since ~ is in fact the graph of an injective two-variable function, by PropoX ~ at p is sition 5.2.4, the tangent plane to X − 2.576(x + 1.4) − 4(y + 1) + (z − 4.0384) = 0 ⇐⇒ − 2.576x − 4y + z = 11.6448. Example 5.2.6. As a second example, consider the hyperboloid of one

sheet given by the parametrization ~ X(u, v) = (cosh u cos v, cosh u sin v, sinh u). A calculation produces ~ v = (− cosh2 u cos v, − cosh2 u sin v, cosh u sinh u). ~u × X X

137

138

5. Regular Surfaces

Figure 5.8. A tangent plane on a hyperboloid of one sheet.

~ Then at the point p = X(1, 0) = (cosh 1, 0, sinh 1), the tangent plane exists, and its equation is − cosh2 1(x − cosh 1) + cosh 1 sinh 1(z − sinh 1) = 0

⇐⇒

(cosh 1)x − (sinh 1)z = 1.

Figure 5.8 shows two perspectives of this hyperboloid along with its tangent plane at p. It is interesting to contrast this picture with Figure 5.7, in which the tangent plane does not intersect the surface except at p. The shape of the hyperboloid forces it to intersect its tangent plane at all points. ~ we Using the notion of the differential of the parametrization X, can summarize Proposition 5.2.4 in another way that is more convenient when we discuss parametrized surfaces in higher dimensions. (See Section 5.2 in  for a longer explanation of the differential of a function from Rm to Rn .) If F~ is a function from an open set U ⊂ Rn to Rm , we write ~ F = (F1 , F2 , . . . , Fm )T and think of F~ as a column vector of functions Fi , each in n variables. For all ~a ∈ U , we say that F is differentiable at ~a if each Fi is differentiable at ~a. Furthermore, we define the differential of F~ at ~a, written dF~~a , as the linear transformation Rn → ~ ∂F Rm that sends the jth standard vector ~ej to ∂x (~a). We denote by j [dF~~a ] the matrix of dF~~a with respect to the standard bases of Rn and

5.2. Tangent Planes and Regular Surfaces

139

Rm , which we can write explicitly as  ∂F

1 ∂x1



∂ F~ dF~~a = ∂x1 



∂ F~ ∂x2

···

∂ F~ ∂xn

   ∂F2  ∂x = 1  ..  .

∂Fm ∂x1

∂F1 ∂x2 ∂F2 ∂x2

···

.. .

..

∂Fm ∂x2

··· . ···

∂F1 ∂xn ∂F2 ∂xn

    . (5.2) ..  . 

∂Fm ∂xn

By a slight abuse of notation, we will sometimes write [dF~ ] or even dF~ , without the subscript, to mean the matrix of functions with ∂ F~ /∂xj as the jth column. We can now restate Proposition 5.2.4 as follows. Corollary 5.2.7. Let S be a surface, where p ∈ S, and suppose that

~ : in an open neighborhood of p, the surface S is parametrized by X 3 U → R . The tangent space (and tangent plane) to S at p exists ~ −1 (p) = {q} (a singleton set), the differential dX ~ q exists, and if X ~ dXq has maximal rank. Furthermore, the tangent space is given by ~ q ). Tp S = Im(dX Proof: Note that with respect to the standard bases in R2 and R3 , ~ q is the 3 × 2 matrix if q = (u0 , v0 ), the matrix of dX  ∂X1  ∂X1 ∂u (q) ∂v (q)    2  2 ~ q =  ∂X  (q) ∂X (5.3) dX ∂v (q) ,  ∂u ∂X3 ∂X3 ∂u (q) ∂v (q) ~ where we have written X(u, v) = (X1 (u, v), X2 (u, v), X3 (u, v)). First note that for the tangent plane to exist, the partial deriva~ must exist, so the differential exists. From Proposition tives of X ~ −1 (p) is a singleton 5.2.4, we know that the surface is a plane if X ~ u (q) × X ~ v (q) exists and set, which we’ll denote by {q} ⊂ U , and X ~ ~ v (q) are is nonzero. This is equivalent to saying that Xu (q) and X ~ linearly independent, which means that dXq has maximal rank.  ~ q to have maxThe condition in Corollary 5.2.7 that requires dX imal rank occurs often enough in various contexts that we give it a name, as follows.

140

5. Regular Surfaces

Definition 5.2.8. Let F be a function from an open set U ⊂ Rn to Rm .

A point q ∈ U is called a critical point if dFq does not exist or does not have maximal rank, i.e., rank dFq < min(m, n). If q is a critical point of F , we call F (q) ∈ Rm a critical value of F . If p ∈ Rm is not a critical value of F (even if p is not in the image of F ), then we call p a regular value of F . Definition 5.2.9. Let S be a parametrized surface in R3 . A point p ∈ S

is called a singular point of S if no tangent plane exists at p. In light of Corollary 5.2.7 and Definition 5.2.8 a point p of a ~ : U → R3 of surface S is a singular point if every parametrization X ~ or X ~ −1 (p) an open neighborhood of p on S is a critical value of X is not a singleton. We have belabored the issue of whether or not one can define a tangent plane to S at a point p because, in an intuitive sense, if a tangent plane does not exist locally near p, then the surface S does not “resemble” a two-dimensional plane and thus does not look like what we would expect in a surface. Just as with curves we introduced the notion of a regular curve (i.e., a curve ~x : I → Rn such that ~x0 (t) 6= ~0 for all t ∈ I), so with surfaces, we would like to define a class of surfaces that is smooth enough so that we can “do calculus” on it. The points in the above discussion lead to the following definition. Definition 5.2.10. A subset S ⊆ R3 is a regular surface if for each

p ∈ S, there exists an open set U ⊆ R2 , an open neighborhood V of ~ : U → V ∩ S such p in R3 , and a surjective continuous function X that ~ is continuously differentiable: if we write X(u, ~ 1. X v) = (x(u, v), y(u, v), z(u, v)), then the functions x(u, v), y(u, v), and z(u, v) have continuous partial derivatives with respect to u and v; ~ is a homeomorphism: X ~ is continuous and has an inverse 2. X −1 ~ −1 is continuous; ~ X : V ∩ S → U such that X ~ satisfies the regularity condition: for each (u, v) ∈ U , the 3. X ~ (u,v) : R2 → R3 is a one-to-one linear transfordifferential dX mation.

5.2. Tangent Planes and Regular Surfaces

141

~ is called a system of coordinates (in a The parametrization X neighborhood) of p, and the neighborhood V ∩ S of p in S is called a coordinate neighborhood . One should observe that a regular surface is defined as a subset of R3 and not as a function from a subset of R2 to R3 . However, one may view a regular surface as the union of the images of an appropriate set of systems of coordinates. Definition 5.2.11. A regular surface is called of class C r if all the sys-

~ are of class C r , i.e., all partial derivatives of tems of coordinates X order up to r exist and are continuous. A regular surface is called of ~ are such that all partial class C ∞ if all its systems of coordinates X derivatives of all orders exist and are continuous. Example 5.2.12. We can prove directly that the unit sphere, denoted

S2 , is a regular surface in a few different ways. We’ll use rectangular coordinates first. Consider a point p = (x, y, z) ∈ S2 , and let U = ~ (1) : U → R3 {(u, v) | u2 + v 2 < 1}. If z > 0, then the mapping X √ defined by (u, v, 1 − u2 − v 2 ) is clearly a bijection between U and ~ (1) is also a homeomorphism because it S2 ∩ {(x, y, z)|z > 0}. X ~ −1 is simply the vertical projection is continuous, and its inverse X (1) of the upper unit sphere onto R2 and since projection is a linear ~ (1) satisfies all the transformation, it is continuous. Furthermore, X conditions in Definition 5.2.10. We leave it to the reader to check Condition 1, while for Condition 3, we calculate the differential 

~ (1) dX

1  0 = u −√ 1 − u2 − v 2

 0  1 . v −√ 1 − u2 − v 2

For all q = (u, v) ∈ U , the entries in this matrix are well defined and produce a matrix that defines a one-to-one linear transformation. ~ (1)q ] is of maximal rank. Consequently, [dX ~ (1) : U → R3 so far only tells us that the open The mapping X upper half of the sphere is a regular surface, but we wish to show that the whole sphere is a regular surface. For other points p = (x, y, z) ∈ S2 , we use the following parame~ (i) : U → R3 of coordinate patches around p: trizations X

142

5. Regular Surfaces

~ (4) X

~ (1) X ~ (6) X

~ (5) X

~ (3) X ~ (2) X

Figure 5.9. Six coordinate patches on the sphere.

p ~ (1) (u, v) = (u, v, 1 − u2 − v 2 ), if z > 0, X p ~ (3) (u, v) = (u, 1 − u2 − v 2 , v), if y > 0, X p ~ (5) (u, v) = ( 1 − u2 − v 2 , u, v), if x > 0, X

p ~ (2) (u, v) = (u, v, − 1 − u2 − v 2 ), if z < 0, X p ~ (4) (u, v) = (u, − 1 − u2 − v 2 , v), if y < 0, X p ~ (6) (u, v) = (− 1 − u2 − v 2 , u, v). if x < 0, X

These six parametrizations give six coordinate patches such that for all p ∈ S2 , p is in at least one of these coordinate patches (see Figure 5.9 for an illustration of the six patches covering the sphere). Since the domain for each parametrization is open, all six patches are necessary. Example 5.2.13. Carefully using two colatitude–longitude parametr-

izations, we get another way to show that the unit sphere S2 is a regular surface. Let H1 be the closed half-plane H1 = {(x, y, z) ∈ R3 | x ≥ 0 and y = 0}. ~ (1) : U → R3 Let p ∈ S2 − H1 , and let U = (0, 2π) × (0, π). Then X defined by ~ (1) (u, v) = (cos u sin v, sin u sin v, cos v) X

5.2. Tangent Planes and Regular Surfaces

is a homeomorphism between U and S2 − H1 that satisfies the remaining conditions in the definition of a regular surface. (The reader should check this claim.) On the other hand, if p ∈ S2 − H2 , where H2 is the closed halfplane H2 = {(x, y, z) ∈ R3 | x ≤ 0 and z = 0}, then a homeomorphism between U → S2 − H2 is ~ (2) (u, v) = (− cos u sin v, cos v, − sin u sin v). X This also satisfies the differentiability and regularity conditions. Furthermore, (S2 − H1 ) ∪ (S2 − H2 ) = S2 , so the two parametrizations of coordinate patches cover the whole sphere. The three conditions in Definition 5.2.10 each eliminate various situations we do not wish to let into the class of surfaces of inter~ be differentiable est in differential geometry. The condition that X eliminates the possibility of corners or folds. A cube, for example, is not a regular surface because for whatever parametrization is used in the neighborhood of an edge where two faces meet, at least one of the partial derivatives will not exist. ~ be a homeomorphism, might iniThe second condition, that X tially appear the least intuitive. Since we assume that in the neigh~ : U → borhood of each point p ∈ S there is a parametrization X V ∩ S that is a bijection, we already eliminate a surface that intersects itself or degenerates to a curve. Figure 5.10 (see Problem 5.2.9 for the parametrization) shows the image of a parametrized surface ~ that intersects itself along a ray. The tangent surface to this surX face at any point along this ray is the union of two planes and not a single plane. ~ : U → V ∩ S of a parameRequiring not only that each patch X trized surface be a bijection but also bicontinuous (i.e., be a home~ 2 ) are arbitrarily close in ~ 1 ) and X(q omorphism) means that if X(q 3 R , then q1 and q2 are arbitrarily close. This eliminates situations similar to that depicted in Figure 5.11, where an open strip is twisted back onto itself so that it does not intersect itself but that the distance between one end and another part in the middle of the strip is 0.

143

144

5. Regular Surfaces

Figure 5.10. Not a bijection.

We now explain why this is not a homeomorphism. Consider an ~ ) open disk U inside the open strip, as shown in the figure. Then X(U 0 0 is not open because a set U is open in S if and only if U = S ∩ V for some open set V in R3 . But any open set V ⊂ R3 that contains ~ ) must contain other points of S, as shown. Thus, U open in the X(U ~ ) is open, which shows that X ~ −1 is domain does not imply that X(U ~ is not a homeomorphism. not continuous and, hence, that X

~ X

Figure 5.11. Not a homeomorphism.

5.2. Tangent Planes and Regular Surfaces

145

The third condition of Definition 5.2.10 is necessary because, even with the first two conditions satisfied, it is still possible to ~ that has a point p such that the create a parametrized surface X tangent space to the surface at p is not a plane but a line. (See Problem 5.2.12 for an example of a surface that satisfies the first two conditions for a regular surface but not the third.) From the above explanations of the criteria for a regular surface, it would seem hard to determine at the outset whether or not a given surface is regular. However, the next two propositions show under what conditions function graphs and surfaces given as solutions to one equation are regular. Proposition 5.2.14. Let U ⊂ R2 be open.

Then if a function f : U → R is differentiable, the subset S = {(x, y, x) ∈ R3 | (x, y, z) = (u, v, f (u, v))} is a regular surface.

Proof: The inverse function F −1 : S → U is the projection onto the xy-plane. F −1 is continuous, and since f is differentiable, F is also continuous. Thus, F is a homeomorphism between U and S. The regularity condition is clearly satisfied because   1 0 dF(u,v) =  0 1  fu fv is always one-to-one regardless of the values of fu and fv .



Another class of surfaces in R3 one studies arises as level surfaces of functions F : R3 → R, that is, as points that satisfy the equation f (x, y, z) = a. The following proposition involves a little more analysis than that presented in most multivariable calculus courses. However, it shows where the terminology “regular surface” comes from in light of the notion of a regular value of multivariable functions presented in Definition 5.2.8. (Most problems in this textbook involve explicit parametrizations, but the proof of this proposition illustrates the value of the Implicit Function Theorem, usually introduced in a course on mathematical analysis.) Proposition 5.2.15. Let f : U ⊂ R3 → R be a differentiable function,

and let a ∈ R be a regular value of f , i.e., a real number such that

146

5. Regular Surfaces

for all p ∈ f −1 (a), dfp is not ~0. Then the surface defined by S = {(x, y, z) ∈ R3 | f (x, y, z) = a} is a regular surface. Furthermore, the ~ (p) is normal to S at p. gradient dfp = ∇f Proof: Let p ∈ S. Since dfp 6= ~0, after perhaps relabeling the axes, we have fz (p) 6= 0. We consider the function G : R3 → R3 defined by G(x, y, z) = (x, y, f (x, y, z)) and notice that   1 0 0 dGp =  0 1 0  . fx fy fz Consequently, det(dGp ) = fz (p) 6= 0, so dGp is invertible. The Implicit Function Theorem allows us to conclude that there exists a neighborhood U 0 of p in U such that G is one-to-one on U 0 , V = G(U 0 ) is open, and the inverse G−1 : V → U 0 is differentiable. However, since (u, v, w) = G(x, y, z) = (x, y, f (x, y, z)), we will be able to write the inverse function as (x, y, z) = G−1 (u, v, w) = (u, v, g(u, v, w)) for some differentiable function g : V → R. Furthermore, if V 0 is the projection of V onto the uv-plane, then h : V 0 → R defined by ha (u, v) = g(u, v, a) is differentiable and G(f −1 (a) ∩ U 0 ) = V ∩ {(u, v, w) ∈ R3 | w = a} ⇐⇒ f −1 (a) ∩ U 0 = {(u, v, ha (u, v)) | (u, v) ∈ V 0 }. Thus, f −1 (a) ∩ U 0 is the graph of ha , and by Proposition 5.2.14, f −1 (a) ∩ U 0 is a coordinate neighborhood of p. Thus, every point p ∈ S has a coordinate neighborhood, and so f −1 (a) is a regular surface in R3 . ~ To prove that dfp is normal to S at p, suppose that X(u, v) = (x(u, v), y(u, v), z(u, v)) parametrizes regularly a neighborhood of S around p. Then, differentiating the defining equation f (x, y, z) = a with respect to u and v one obtains  ∂f ∂x ∂f ∂y ∂f ∂z    ∂x ∂u + ∂y ∂u + ∂z ∂u = 0,  ∂f ∂x ∂f ∂y ∂f ∂v   + + = 0. ∂x ∂v ∂y ∂v ∂z ∂u ~ u and X ~v Thus, dfp = (fx (p), fy (p), fz (p)) is perpendicular to X at p. 

5.2. Tangent Planes and Regular Surfaces

147

(Note: See Section 6.4 in  for a statement of the Implicit Function Theorem and some examples.) We point out that Definition 5.2.10 can be easily modified to provide a definition of regularity for a surface S in Rn . The only required changes are that each parametrization of a neighborhood of ~ : R2 → Rn and that at each point p = S be a continuous function X ~ ~ X(u, v), the differential dX(u,v) be an injective linear transformation R2 to Rn .

Problems 5.2.1. Consider the parametrized surface ~ X(u, v) = ((v 2 +1) cos u, (v 2 +1) sin u, v),

where (u, v) ∈ [0, 2π] × R.

Find the equation of the tangent plane to this surface at p = (2, 0, 1) = ~ X(0, 1). 5.2.2. Calculate the equation of the tangent plane to the torus ~ X(u, v) = ((2 + cos v) cos u, (2 + cos v) sin u, sin v)  at the point (u, v) = π6 , π3 . 5.2.3. Show that the equation of the tangent plane at (x0 , y0 , z0 ) at a regular surface given by f (x, y, z) = 0, where 0 is a regular value of f , is fx (x0 , y0 , z0 )(x−x0 )+fy (x0 , y0 , z0 )(y−y0 )+fz (x0 , y0 , z0 )(z−z0 ) = 0. 5.2.4. Determine the tangent planes to x2 + y 2 − z 2 = 1 at the points (x, y, 0) and show that they are all parallel to the z-axis. 5.2.5. Show that the cylinder {(x, y, z) ∈ R3 | x2 + y 2 = 1} is a regular surface and find parametrizations whose coordinate neighborhoods cover it. 5.2.6. Show that the two-sheeted cone with equation x2 + y 2 − z 2 = 0 is not a regular surface. 5.2.7. Surfaces of Revolution. Consider the surface of revolution S with ~ parametrization X(u, v) = (f (v) cos u, f (v) sin u, g(v)), where (f (v), g(v)) is a regular plane curve (in the xz-plane). Prove that S is regular if and only if for all v in the domain of f , f (v) 6= 0 and g 0 (v) 6= 0 or f (v) 6= 0 and f 0 (v) 6= 0.

148

5. Regular Surfaces

Figure 5.12. Parametrization of Problem 5.2.12. 5.2.8. Let f (x, y, z) = (x + 2y + 3z − 4)2 . (a) Locate the critical points and the critical values of f . (b) For what values of c is the set f (x, y, z) = c a regular surface? (c) Repeat (a) and (b) for the function f (x, y, z) = xy 2 z 3 . ~ 5.2.9. Consider the surface parametrized by X(u, v) = (uv, u2 −v 2 , u3 −v 3 ). ~ intersects itself along the ray (x, 0, 0), with x ≥ 0. Find Prove that X the points (u, v) that map to this ray. Use these to prove that the ~ at any point on the open ray (x, 0, 0), with x > 0, tangent space to X is the union of two planes. (See Figure 5.10 for the plot of a portion of this surface near (0, 0, 0).) ~ 5.2.10. Consider the parametrized X(u, v) = (u, v 3 , −v 2 ). Prove that for ~ ~ q does any point on the surface p = X(q) such that q = (u, 0), dX not have maximal rank. ~ 5.2.11. Consider X(u, v) = (v cos u, v sin u, e−v ) for (u, v) ∈ [0, 2π] × [0, ∞). ~ fails the regularity condition. Find all points where X ~ 5.2.12. Show that the parametrized surface X(u, v) = (u3 +v+u, u2 +uv, v 3 ) satisfies the first and second conditions of Definition 5.2.10 but does not satisfy the third at p = (0, 0, 0). (See Figure 5.12.) 5.2.13. Consider the hyperbolic paraboloid defined by S = {(x, y, z) ∈ R3 | z = x2 − y 2 }. Check that the following provide systems of coordinates for parts of S. ~ (a) X(u, v) = (u + v, u − v, 4uv) for (u, v) ∈ R2 . ~ (b) X(u, v) = (u cosh v, u sinh v, u2 ) for (u, v) ∈ R2 .

5.2. Tangent Planes and Regular Surfaces

149

z N

p

y π(p)

x

Figure 5.13. Stereographic projection.

5.2.14. Find a parametrization for the hyperboloid of two sheets x2 + y 2 − z 2 = 1. 5.2.15. Stereographic Projection. One way to define coordinates on the surface of the sphere S2 given by x2 + y 2 + z 2 = 1 is to use the stereographic projection of π : S2 − {N } → R2 , where N = (0, 0, 1), defined as follows. Given any point p ∈ S2 , the line (pN ) intersects the xy-plane at exactly one point, which is the image of the function π(p). If (x, y, z) are the coordinates for p in S2 , let us write π(x, y, z) = (u, v) (see Figure 5.13).   y x (a) Prove that π(x, y, z) = 1−z , 1−z . (b) Prove that π −1 (u, v) =



2u 2v u2 + v 2 − 1 , 2 , 2 2 2 2 u + v + 1 u + v + 1 u + v2 + 1

 .

(c) Prove that by using stereographic projections, it is possible to cover the sphere with two coordinate neighborhoods. 5.2.16. Let α ~ : I → R2 be a plane curve and let β~ : I → R3 be the pullback of α ~ onto the unit sphere via the stereographic projection as described ~ in Problem 5.2.15. In other words, β(t) = π −1 (~ α(t)). Prove that points on the locus of β~ where the torsion τ is zero correspond to vertices in α ~ , that is points where the plane curvature κg of α ~ satisfies κg0 (t) = 0.

150

5. Regular Surfaces

5.2.17. Cones. Let C be a regular planar curve parametrized by α ~ : (a, b) → R3 lying in a plane P that does not contain the origin O. The cone Σ over C is the union of all the lines passing through O and p for all points p on C. ~ of the cone Σ over C. (a) Find a parametrization X (b) Find the points where Σ is not regular. 5.2.18. Central Projection. The central projection used in cartography involves a function f : S2 → (0, 2π) × R defined geometrically as follows. View (0, 2π) × R as a cylinder wrapping around the unit sphere S2 in such a way that the slit in the cylinder falls in the halfplane P with y = 0 and x ≥ 0. The central projection f sends a point p ∈ S2 − P to f (p) on the cylinder by defining f (p) as the unique point on the cylinder on ray [O, p), where O = (0, 0, 0) is the origin. (a) Let (x, y, z) be the coordinates of p ∈ S2 −P and call f (x, y, z) = (u, v). Find a formula for f . (b) Calculate a formula for f −1 . (c) If one uses central projections for coordinate neighborhoods on the sphere, how many are necessary to cover the sphere?

5.3 Change of Coordinates A particular patch of a surface can be parametrized in a variety of ways. In particular, we could use a different coordinate system in ~ R2 to describe the domain U of X. Example 5.3.1 (Two Parametrizations of the Sphere). As an example, consider the following two parametrizations of the upper half of the unit sphere in R3 : √ ~ 1. X(u, v) = (u, v, 1 − u2 − v 2 ), with (u, v) ∈ U where the domain U is the closed unit disk, i.e., U = {(u, v) ∈ R2 | u2 +v 2 ≤ 1}. To be more specific (if we needed to provide intervals for (u, v), say to calculate integrals), we could say p p U = {(u, v) ∈ R2 | − 1 − u2 ≤ v ≤ 1 − u2 and −1 ≤ u ≤ 1}.

~ (θ, ϕ) = (cos θ sin ϕ, sin θ sin ϕ, cos ϕ), with (θ, ϕ) ∈ U 0 where 2. Y we have U 0 = [0, 2π] × [0, π2 ].

5.3. Change of Coordinates

151

~ and Y ~ parametrize In this case, it’s relatively easy to verify that X the same surface patch by setting the change of coordinates (u, v) = F (θ, ϕ), with F :

U0

−→ U

(θ, ϕ) 7−→ (cos θ sin ϕ, sin θ sin ϕ). We notice that F (U 0 ) = U (i.e., F is a surjective function) and that, ~ =X ~ ◦ F. as vector functions, Y 2 The function F : R → R2 is not a bijection between U and U 0 , but it is a bijection between U2 = {(u, v) ∈ U | u2 + v 2 < 1 and v = 0 implies u < 0} and

 π , U20 = (0, 2π) × 0, 2

and the inverse function (θ, ϕ) = F −1 (u, v) is given by    tan−1 uv , if u > 0 and v ≥ 0,    π   if u = 0 and v > 0, 2,  v −1 θ = tan if u < 0, u + π,   3π  , if u = 0 and v < 0,   2   −1 v  tan u + 2π, if u > 0 and v ≤ 0. and p ϕ = sin−1 ( u2 + v 2 ). It is tedious but not hard to verify that both F and F −1 are differentiable on the open domains U2 and U20 . Figure 5.14 illustrates F . We now consider a general change of coordinates F in R2 . Con~ : U → R3 and Y ~ = sider two parametrized surfaces defined by X 0 3 0 0 2 ~ ◦ F : U → R , where F : U → U for subsets U and U in R . The X above example illustrates three progressively stringent conditions on the change-of-coordinates function F . First, a sufficient condition ~ =X ~ ◦ F and X ~ to have the same image in R3 is for F to be for Y surjective. Second, if one wishes for the coordinates on a surface to ~ being a homeomorphism, be unique or if we are concerned with Y

152

5. Regular Surfaces

ϕ

v F

π 2

U20 π

1 u 2π θ

Figure 5.14. Coordinate change by F .

we might wish to require that F be a bijection, in which case, if F is already surjective, one merely must restrict the domain of F to make it a bijection. Finally, requiring that both F and F −1 be differentiable adds a “smoothness” condition to how the coordinates transform. This latter condition has a name. Definition 5.3.2. Let U and V be subsets of Rn . We call a function

F : U → V a diffeomorphism if it is a bijection, and both F and F −1 are differentiable. Proposition 5.3.3. Let U and V be subsets of Rn , and let F : U →

V be a diffeomorphism. Suppose that q = F (q 0 ). Then the linear transformation dFq0 is invertible, and dFq−1 = d(F −1 )q . 0 Proof: Suppose that F : Rm → Rn and G : Rn → Rs are differentiable, multivariable, vector-valued functions. (For what follows, one can also assume that these functions have small domains, just as long as the range of F is a subset of the domain of G.) Using the notion of the differential defined in Equation (5.2), it is an easy exercise to show that the formula for the chain rule on G ◦ F in a multivariable context can be written as d(G ◦ F )a = dGF (a) dFa , where we mean matrix multiplication.

5.3. Change of Coordinates

153

Applying this to the situation of this proposition, where F −1 ◦ F = idU , we get d(F −1 )F (q0 ) ◦ dFq0 = In , where In is the n × n identity matrix. Furthermore, the same reasoning holds with F ◦ F −1 , and we conclude that dFq0 is invertible and that (dFq0 )−1 = d(F −1 )q .  ~ : U → R3 Suppose again that we have parametrized surfaces X 0 3 0 ~ ~ and Y = X ◦ F : U → R , where F : U → U is a diffeomorphism between open subsets U 0 and U in R2 . Let q 0 ∈ U 0 and define q = F (q 0 ). Then, just as in the above proof, the chain rule in multiple variables gives ~q0 = d(X ~ ◦ F )q0 = dX ~ q ◦ dFq0 . dY If for the change of coordinates F we write (u, v) = F (s, t) = (f (s, t), g(s, t)), then the matrix of the differential of this coordinate change is     ∂u ∂u ∂f ∂f  ∂s ∂t   ∂s ∂t     [dF ] =   ∂v ∂v  =  ∂g ∂g  . ∂s ∂t ∂s ∂t With matrices, the chain rule is written as  ∂u  ∂s ~(s,t) ] = [dX ~ F (s,t) ] ·  [dY  ∂v ∂s

 ∂u ∂t   ∂v  . ∂t

(5.4)

~ = (Y1 , Y2 , Y3 ) if necessary, we find ~ = (X1 , X2 , X3 ) and Y Writing X that Equation (5.4) is equivalent to the relations ~ ~ ∂u ∂ X ~ ∂v ∂Y ∂X ∂u ~ = + = Xu + ∂s ∂u ∂s ∂v ∂s ∂s ~ ∂u ∂ X ~ ∂v ~ ∂X ∂u ~ ∂Y = + = Xu + ∂t ∂u ∂t ∂v ∂t ∂t

∂v ~ Xv , ∂s ∂v ~ Xv . ∂t

(5.5)

154

5. Regular Surfaces

Problems 5.3.1. Prove the claim in Example 5.3.1 that both F and F −1 are differentiable on the open domains U2 and U20 . 5.3.2. Consider the change of coordinates from polar to Cartesian described by (x, y) = F (r, θ) = (r cos θ, r sin θ). (a) Calculate the matrix of the differential [dF ]. (b) Determine F −1 and calculate [dF −1 ]. (c) Verify Proposition 5.3.3 that [d(F −1 )(x,y) ] = [dF(r,θ) ]−1 . 5.3.3. Consider the coordinate change in two variables F (s, t) → (s2 − t2 , 2st). Let U = {(s, t) | s2 + t2 ≤ 1 and t ≥ 0}. Show p that F (U ) is the ~ whole unit disk. Prove that if X(x, y) = (x, y, 1 − x2 − y 2 ), then ~ ◦ F : U → R3 is a parametrization for the upper half of the unit X sphere. 5.3.4. Consider the coordinate change in the previous problem but with domain U = {(s, t) | t > 0}. (a) Determine the set F (U ) and show that F : U → F (U ) is a bijection. (b) Determine the function F −1 explicitly. (c) Show in what sense [d(F −1 )] = [dF ]−1 . 5.3.5. Consider spherical and cylindrical coordinate systems. Consider the coordinate change function from spherical to cylindrical defined by (r, θ, z) = F (ρ, θ, ϕ) = (ρ sin ϕ, θ, ρ cos ϕ). (Note that ϕ is the angle down from the positive z-axis.) Show that with U 0 = (0, ∞) × (0, 2π) × (0, π) U = (0, ∞) × (0, 2π) × R the coordinate change function F : U 0 → U is a diffeomorphism. 5.3.6. Prove that F (x, y) = (x3 , y 3 ) is a bijection from R2 to R2 but not a diffeomorphism. Prove also that F (x, y) = (x3 + x, y 3 + y) is a diffeomorphism of R2 onto itself.

5.4. The Tangent Space and the Normal Vector

155

5.3.7. Prove that the function F : R2 → R2 defined by F (x, y) = x cos(x2 + y 2 ) − y sin(x2 + y 2 ), x sin(x2 + y 2 ) + y cos(x2 + y 2 ) is a diffeomorphism of R2 onto itself. 5.3.8. Find an example of a diffeomorphism of R2 onto an open square and show why it is a diffeomorphism.

5.4 The Tangent Space and the Normal Vector Let S be a regular surface parametrized in the neighborhood of a ~ : U → R3 . Suppose that p = X(q). ~ point p by X Since S is regular, the set of tangent vectors to S at p is a subspace Tp S, and we saw in ~ q ). The condition that S be regular Corollary 5.2.7 that Tp S = Im(dX ~ q are linearly independent. Thus, at p ensures that the columns of dX ~ given the parametrization X of the neighborhood of p, the set of ~ v (q)} forms a basis of Tp S. ~ u (q), X vectors {X Comparing surfaces to curves, recall that at every regular point on a curve, the unit tangent vector to a curve at a point is invariant under a positively oriented reparametrization of the curve, and the tangent line to the curve at a point is an entirely geometric object, completely unchanged under reparametrizations. Implicit in Definition 5.2.1, the set of tangent vectors to a surface at a point, and hence the tangent space, is a geometric object, invariant under reparametrization. In contrast, the change of basis formulas Equations (5.5) and ~ =X ~ ◦ F and if q = F (q 0 ) and p = X(q), ~ (5.4) show that when Y the 0 0 ~ ~ ~ ~ sets of vectors {Xu (q), Xv (q)} and {Ys (q ), Yt (q )} are not necessarily equal. Consequently, though Tp S is invariant under a reparametrization, the basis induced from the parametrization is not invariant. ~ : U → R3 be a parametrization of a coProposition 5.4.1. Let X ordinate patch on a regular surface S. Consider a diffeomorphism ~ =X ~ ◦ F is a regular parametrization F : U 0 ⊂ R2 → U such that Y ~ of an open subset of S. Set q = F (q 0 ) and p = X(q). If ~a ∈ Tp S has coordinates ~s + a ~t = a1 X ~ u + a2 X ~ v, ~a = a ¯1 Y ¯2 Y



156

5. Regular Surfaces

then       a ¯1 a −1 a1 −1 = [dFq0 ] = [d(F )q ] 1 . a ¯2 a2 a2 Proof: By the chain rule in Equations (5.5) and (5.4), we have     ∂u ~ ∂u ~ ∂v ~ ∂v ~ ~ ~ a ¯1 Ys + a Xu + Xv + a Xu + Xv ¯2 ¯ 2 Yt = a ¯1 ∂s ∂s ∂t ∂t     ∂u ∂u ~ ∂v ∂v ~ = a ¯1 +a ¯2 Xu + a +a ¯2 Xv . ¯1 ∂s ∂t ∂s ∂t ~ u, X ~ v } is a basis of Tp S, we deduce ~ u + a2 X ~ v and {X Since ~a = a1 X that   ∂u ∂u      ∂s ∂t  a a1  ¯1 . =  a  ∂v ∂v a2 ¯2 ∂s ∂t The last claim of the proposition follows in light of Proposition 5.3.3.  Using the language of linear algebra, Proposition 5.4.1 shows that [d(F −1 )q ] is the change of coordinates matrix between the basis ~ and the basis corresponding corresponding to the parametrization X ~ =X ~ ◦ F. to the reparametrization Y At this point, we wish to restate Proposition 5.4.1 using notations that we will adopt later, in particular in Chapter 7. We label the coordinates (u, v) as (x1 , x2 ), and we label the coordinates (s, t) as (¯ x1 , x ¯2 ). We write the diffeomorphism that relates the coordinates as ¯2 ) = F (x1 , x2 ) (¯ x1 , x

and

(x1 , x2 ) = F −1 (¯ x1 , x ¯2 ).

Then according to Proposition 5.4.1, the change of coordinates on Tp S satisfies   ¯1 ∂x ¯1 ∂ x      ∂x1 ∂x2  a1 a ¯1   =  ∂x (5.6) ¯2 ∂ x ¯2  a2 . a ¯2 ∂x1 ∂x2

5.4. The Tangent Space and the Normal Vector

157

The matrix product in Equation (5.6) is often written in summation notation as 2 X ∂x ¯i ak . a ¯i = ∂xk k=1

~ is a regular parametrization According to Proposition 5.2.4, if X ~ of a neighborhood of p = X(q) of a regular surface S, the tangent ~ u (q) × X ~ v (q) as a normal vector. Since the tangent plane plane has X is invariant under reparametrization, this cross product must only be rescaled under reparametrization. ~ : U → R3 be a parametrization of a coorProposition 5.4.2. Let X

dinate patch on a regular surface S. Consider a function F : U 0 ⊂ ~ =X ~ ◦ F is a regular parametrization of an open R2 → U such that Y ~ subset of S. Then if q = F (q 0 ) and p = X(q), ~s (q 0 ) × Y ~t (q 0 ) = det(dFq0 )(X ~ u (q) × X ~ v (q)). Y Proof: (Left as an exercise for the reader.)



We remind the reader that the determinant det(dFq0 ) of the differential matrix of the change of coordinate function is called the Jacobian of F at q 0 . (The reader may recall that the Jacobian appears in the change of variables formula for multiple integrals.) Proposition 5.4.2 motivates us to define the unit normal to a ~ is a parametrization of regular surface S at point p as follows. If X ~ a neighborhood of p and p = X(q), then the unit normal vector is ~ ~ ~ (q) = Xu × Xv (q). N ~u × X ~ vk kX

(5.7)

This vector is invariant under reparametrization except up to a change in sign, namely, the sign of the Jacobian of the corresponding change of coordinates. Definition 5.4.3. With the setup as in Proposition 5.4.2, we say that a

~ =X ~ ◦F is a positively (resp. negatively) regular reparametrization Y oriented reparametrization if det(dFq0 ) > 0 (resp. det(dFq0 ) < 0) for all q 0 ∈ U 0 .

158

5. Regular Surfaces

This unit normal vector, among other things, allows one to easily answer questions that involve the angle at which two surfaces intersect at a point. In other words, let S1 and S2 be two regular surfaces, and let p ∈ S1 ∩ S2 . Then S1 intersects with S2 at p, with an angle ~1 · N ~ 2 = cos θ. Of course, the particular parametrization θ, where N might change the sign of one of the unit normal vectors, but the acute angle between Tp (S1 ) and Tp (S2 ) is well defined, regardless of signs.

Problems ~ 5.4.1. Calculate the unit normal vector for function graphs X(u, v) = (u, v, f (u, v)). 5.4.2. Suppose that a coordinate neighborhood of a regular surface can be parametrized by ~ ~ X(u, v) = α ~ (u) + β(v), where α ~ and β~ are regular parametrized curves defined over intervals I and J, respectively. Prove that along coordinate lines (either u = u0 or v = v0 ) all the tangent planes to the surface are parallel to a fixed line. ~ : U → R3 be a parametrization of an open set of a regular 5.4.3. Let X surface, and let p~ be a point in R3 that is not in S. Consider the function F : U → R defined by ~ F (u, v) = kX(u, v) − p~k. Prove that if ~q = (u0 , v0 ) is a critical point of F , then the vector ~ q ) − p~ is normal to the surface S at X(~ ~ q ). X(~ 5.4.4. Tangential Surfaces. Let α ~ : I → R3 be a regular parametrized curve with curvature κ(t) = 6 0. Call the tangential surface to α ~ the parametrized surface ~ u) = α X(t, ~ (t) + u~ α0 (t),

with t ∈ I and u 6= 0.

~ 0 , u), the tangent Prove that for any fixed t0 ∈ I, along any curve X(t planes are all equal. 5.4.5. Tubes. Let α ~ : I → R3 be a regular parametrized curve. Let r be a positive real constant. Define the tube of radius r around α ~ as the parametrized surface ~ u) = α ~ X(t, ~ (t) + r(cos u)P~ (t) + r(sin u)B(t), with (t, u) ∈ I × [0, 2π].

5.5. Orientable Surfaces

159

~ (a) Prove that a necessary (though not sufficient) condition for X to be a regular surface is that r < 1/(max κ(t)). t∈I

~ is regular, the unit normal vector is (b) Show that when X ~ (t, u) = − cos uP~ (t) − sin uB(t). ~ N 5.4.6. Prove Proposition 5.4.2. 5.4.7. Let f and g be real functions such that f (v) > 0 and g 0 (v) 6= 0. Consider a parametrized surface given by ~ X(u, v) = (f (v) cos u, f (v) sin u, g(v)) . Show that all the normal lines to this surface pass through the z-axis. [Hint: See Problem 5.2.7.] 5.4.8. Two regular surfaces S1 and S2 intersect transversely if for all p ∈ S1 ∩ S2 , Tp (S1 ) 6= Tp (S2 ). Prove that if S1 and S2 intersect transversely, then the set S1 ∩ S2 is a disjoint union of regular curves.

5.5 Orientable Surfaces Let S be a regular surface. Each point p of S has a coordinate neigh~ : Uα → R3 borhood Vα ⊂ S parametrized by some vector function X satisfying the differentiability, homeomorphism, and regularity conditions of Definition 5.2.10. Thus S is covered by a collection of coordinate patches, and if S is compact, then we can cover S by a finite such collection. The concept of an orientable surface encapsulates the notion of being able to define an inside and an outside to the surface. At any point p of S, there are two unit normal vectors to the surface. If it makes sense to distinguish between “two sides” of the surface, then ~ p at p eliminates all the options for all specifying the unit normal N ~ p0 for all other the other points on the surface and uniquely defines N points of S. Definition 5.5.1. Let S be a regular surface in R3 . We say that S is

orientable if it can be covered by a collection of coordinate patches

160

5. Regular Surfaces

~ (i) : Ui → Ri , with i ∈ I, given as the images of parametrizations X 3 2 R , where Ui ⊂ R , such that if ~ (i) ~ (i) ∂ X ∂X × ∂v (q), ~ (i) (q) = ∂u N

∂X ~ (i) ~ ∂X (i)

×

∂u ∂v ~ (i) (q) = N ~ (j) (q) for all q ∈ Ui ∩ Uj , for all i, j ∈ I. If for the then N surface S it is impossible to find such a covering by parametrizations, then we call S nonorientable. By Proposition 5.4.2, we can provide an alternative formulation of this definition. Proposition 5.5.2. A regular surface S is orientable if and only if it is

possible to cover it with coordinate patches Ri , given as the images ~ (i) : Ui → R3 , such that if X ~ (j) = X ~ (i) ◦ F for of parametrizations X a change of coordinates F : Ui ∩ Uj → Ui ∩ Uj , then det(dFq ) > 0 for all q ∈ Ui ∩ Uj . Example 5.5.3. The commonly given example of a nonorientable sur-

face in R3 is the M¨obius strip M . Intuitively, the M¨obius strip is a surface obtained by taking a long and narrow strip of paper and gluing the short ends together as though to make a cylinder but putting one twist in the strip before gluing (see Figure 5.15).

Figure 5.15. M¨obius strip.

5.5. Orientable Surfaces

By looking at Figure 5.15, one can imagine a normal vector pointing outward on the surface of the M¨obius strip, but if one follows the direction of this normal vector around on the surface, when it comes back around, it will be pointing inward this time instead of outward. We wish to be more specific in this case. One can parametrize most (an open dense subset) of the M¨obius strip by   u u u ~ sin u, 2 − v sin cos u, v cos , X(u, v) = 2 − v sin 2 2 2 with (u, v) ∈ (0, 2π) × (−1, 1). This coordinate neighborhood omits the boundary of the closed M¨obius strip as well as the coordinate line with u = 0.  ~ u, 1 = (0, 2, 1 ). Let p be the point on M given by limu→0 X 2 2 To obtain the same point p as a limit with u → 2π, we must have ~ u, − 1 . With some a different value for v, namely, p = limu→2π X 2 calculations, one can find that     ~u × X ~ v = − v cos u − 2 − v sin u sin u cos u ~i X 2 2  2   v u u ~ + sin u − 2 − v sin cos u cos j 2 2 2   u u ~ − 2 − v sin sin k 2 2 and

 2 2 ~u × X ~ v k2 = v + 2 − v sin u . kX 4 2 From this, it is not hard to show that     1 8 1 ~ lim N u, = −√ , −√ , 0 u→0 2 65 65     ~ u, − 1 = √1 , √8 , 0 . and lim N u→2π 2 65 65 Consequently, no collection of systems of coordinates of M that in~ can satisfy the conditions of orientability. However, the cludes X ~ is not the problem. Using any colspecific system of coordinates X lection of coordinate patches to cover M , as soon as one tries to “go all the way around” M , there are two overlapping coordinate patches for which the conditions of Definition 5.5.1 fail.

161

162

5. Regular Surfaces

The M¨obius strip has a boundary at v = ±1 in the above parametrization. Then there are many examples of nonorientable regular surfaces that have a boundary. For example, one can attach a “handle” to the M¨obius strip, a tube connecting one part of the strip to another. However, for reasons that can only be explained by theorems in topology, there is not “enough room” in R3 to fit a closed nonorientable surface that does not have a boundary and does not intersect itself. However, because there is “more room” in higher dimensions, there exist many other closed nonorientable surfaces without boundary in R4 that do not intersect themselves. See Chapter 9.2 for a few examples. Any particular choice of unit normal vector at each point on a regular surface defines a function n : S → S2 , where S2 is the twodimensional sphere. (The notation Sk refers to the k-dimensional sphere. From a topological perspective, the dimension of the ambient space is irrelevant, but in basic differential geometry, one often pictures Sk as the unit k-dimensional sphere in Rk+1 .) We can now state a new criterion for orientability. Proposition 5.5.4. Let S be a regular surface in R3 . S is orientable if

and only if there exists a continuous function n : S → S2 such that n(p) is normal to S at p for all p ∈ S.

Proof: (⇒) Suppose that S is a regular orientable surface. Suppose that S is covered by a collection of coordinate patches Ri , with i ∈ ~ (i) : Ui → R3 that I, given as the images of parametrizations X satisfy the condition of Definition 5.5.1. Suppose that p is in the ~ (k) (q). Define n : S → S2 as coordinate patch Rk on S and p = X ~ (k) (q). The criteria for Definition 5.5.1 ensure that n is n(p) = N well defined, regardless of the chosen coordinate patch Rk . Since n is continuous over each Ri and the collection {Ri }i∈I cover, S, n is continuous. (⇐) Let n : S → S2 be a continuous function such that n(p) is normal to S at p for all p ∈ S. Let {Ri }i∈I be a collection of ~ (i) : Ui → R3 that covers coordinate patches of S parametrized by X S and satisfies the definition for a regular surface. Over each open

5.5. Orientable Surfaces

163

set Ri , the function ~ (i) ∂ X ~ (i) ∂X × ∂u ∂v ~ (i) (u, v) = N

∂X ~ ~ (i) ∂X (i)

×

∂u ∂v ~ (i) (q) = p for some is defined and continuous. Now for all i ∈ I, if X ~ (i) (q) may be equal to n(p) or −n(p). p ∈ Ri , the normal vector N ~ i and n are both continuous functions and Furthermore, since N kn(p) − (−n(p))k = 2, ~ (i) (q) = n ◦ X ~ (i) (q) or N ~ (i) (q) = −n ◦ X ~ (i) (q) for all we deduce that N q ∈ Ui . Define a new collection of coordinate patches as follows: For all ~ (i) = n ◦ X ~ (i) over Ui , then set Y ~i = X ~ i . If N ~ (i) = −n ◦ X ~ (i) i ∈ I, if N ~(i) (u, v) = X ~ (i) (−u, v), with the domain of Y ~(i) over Ui , then define Y modified accordingly. Then ~~ N Y

(i)

~(i) =n◦Y

~(i) }i∈I for all i ∈ I, and thus the collection of parametrizations {Y satisfies the Definition 5.5.1 for orientability.  Definition 5.5.5. Let S be an orientable regular surface in R3 . A choice

of continuous function n : S → S2 of unit normal vectors on S is called an orientation on S. An orientable regular surface equipped with a choice of such a function n is called an oriented surface. If S is an oriented regular surface, a pair of vectors (~v , w) ~ of the tangent plane Tp S is called a positively oriented basis if the vectors form a basis and (~v × w) ~ · n(p) > 0. ~ : U → R3 of an A parametrization X positively oriented if the ordered pair oriented basis of the tangent plane of S

open set of S is also called ~ u, X ~ v ) forms a positively (X ~ at X(u, v) for all (u, v) ∈ U .

164

5. Regular Surfaces

If we consider surfaces in Rk , the cross product is not available, so Definition 5.5.1 no longer applies. However, Proposition 5.5.2 has no explicit dependency on the dimension of the ambient space and therefore is taken as the definition of orientability for regular surfaces in Rk . In diagrams depicting orientable regular surfaces, if the regular surface S is in R3 , one can often sketch the surface and indicate a unit normal vector at a point on the surface. Because the surface is orientable, this choice of unit normal vector uniquely determines the unit normal vectors at all points on the surface. However, in a diagram that depicts a neighborhood of a regular surface S in Rk , where k is not necessarily 3, we indicate the orientation of S with an oriented loop. The orientation of the loop indicates that in this diagram, any parametrization of S is such that one must sweep that ~ u and X ~ v is in the plane in this direction so that the angle between X interval (0, π). See Figure 5.16. ~v X

~u X

~v X

~u X

Figure 5.16. Orientation diagram.

Problems 5.5.1. Supply the details for Example 5.5.3. 5.5.2. Let S be a regular surface in R3 given as the set of solutions to the equation F (x, y, z) = a, where F : U ⊂ R3 → R is differentiable and a is a regular value of F . Prove that S is orientable.

CHAPTER 6

The First and Second Fundamental Forms

Recall that the local geometry of space curves is completely determined by two geometric invariants: the curvature and the torsion. Similarly, as we shall see, the local geometry of a regular surface S in R3 is determined by the first and second fundamental forms. The value of restricting attention to regular surfaces is that at all points on a regular surface, there is an open neighborhood reg~ Thus, at a ularly homeomorphic to R2 via a parametrization X. ~ ~ point p ∈ S, with p = X(q), the differential dXq provides a natural isomorphism between R2 and Tp S. Whenever we consider vectors on S based at the point p, we must consider them as elements of Tp S, and we can “do geometry” locally on S by identifying Tp S with R2 . The first fundamental form establishes a natural direct product on Tp S that ultimately leads to formulas for length of vectors in Tp S, arc length, angle between vectors in Tp S, and area formulas on S. The second fundamental form provides a measurement of how the normal vector changes as one moves over the surface S, thereby, in an intuitive sense, describing how S sits in R3 .

6.1 The First Fundamental Form Definition 6.1.1. Let S be a regular surface and p ∈ S. The first

fundamental form Ip (·, ·) is the restriction of the usual dot product ~ q (R2 ), in R3 to the tangent plane Tp S. Namely, for ~a, ~b in Tp S = dX Ip (~a, ~b) = ~a · ~b. Note that for each point p ∈ S, the first fundamental form Ip (·, ·) is defined only on the tangent space. There exists a unique matrix

165

166

6. The First and Second Fundamental Forms

that represents Ip (·, ·) with respect to the standard basis on Tp S, but it is important to remember that this matrix is a matrix of functions with components depending on the point p ∈ S. For any other point p2 ∈ S, the first fundamental form is still defined the same way, but the corresponding matrix is most likely different. This definition draws on the ambient space R3 for an inner product on each tangent plane. However, if around p the surface S can ~ there is a natural basis on Tp S, namely X ~u be parametrized by X, ~ and Xv . We wish to express the inner product Ip (·, ·) in terms of this basis on S. Let ~ u (q) + a2 X ~ v (q) ~a = a1 X

~ u (q) + b2 X ~ v (q). and ~b = b1 X

Then we calculate that ~ u (q) + a1 b2 X ~ v (q) ~ u (q) · X ~ u (q) · X ~a · ~b = a1 b1 X ~ v (q) + a2 b2 X ~ v (q). ~ u (q) · X ~ v (q) · X + a2 b1 X This proves the following proposition. Proposition 6.1.2. Let Ip (·, ·) : Tp S 2 → R be the first fundamental

form at a point p on a regular surface S. Given a regular parametri~ : U → R3 of a neighborhood of p, the matrix associated with zation X ~ u, X ~ v } is the first fundamental form Ip (·, ·) with respect to the basis {X   g g g = 11 12 , g21 g22 where ~u · X ~u g11 = X ~v · X ~u g21 = X

and and

~u · X ~ v, g12 = X ~v · X ~ v, g22 = X

~ 0 , v0 ). evaluated at (u0 , v0 ) when p = X(u The quantity g is a matrix of real functions from the open domain ~ U ⊂ R2 . With this notation, if p = X(q), then the first fundamental form Ip (·, ·) at the point p can be expressed as the bilinear form    b T ~ ~ Ip (~a, b) = ~a g(q)b = a1 a2 g(q) 1 b2 for all ~a, ~b ∈ Tp S.

6.1. The First Fundamental Form

167

One should remark immediately that for any differentiable func~ : U → R3 , one has X ~u · X ~v = X ~v · X ~ u . Thus, g12 = g21 and tion X the matrix g is symmetric. Example 6.1.3 (The xy-Plane). As the simplest possible example, consider the xy-plane. It is a regular surface parametrized by a single ~ system of coordinates X(u, v) = (u, v, 0). Obviously, we obtain

  1 0 . g= 0 1 This should have been obvious from the definition of the first fundamental form. The xy-plane is its own tangent space for all p, and ~ induces the basis the parametrization X ~ u, X ~ v } = {(1, 0, 0), (0, 1, 0)}. {X  ~= Therefore, for any ~v = vv12 and w terms of the standard basis, we have ~v · w ~ = v1 w1 + v2 w2 = v1

w1 w2

 , with coordinates given in

    1 0 w1 v2 . 0 1 w2

Example 6.1.4 (Cylinder). Consider the right circular cylinder param-

~ etrized by X(u, v) = (cos u, sin u, v), with (u, v) ∈ (0, 2π) × R. We calculate ~ u = (− sin u, cos u, 0) X

and

~ v = (0, 0, 1), X

and thus,   1 0 g= . 0 1 Obviously a cylinder and the xy-plane are not the same surface. As we will see later, a plane and a cylinder share many properties. However, the second fundamental form, which carries information about how the normal vector behaves on the cylinder, will distinguish between the cylinder and the xy-plane.

168

6. The First and Second Fundamental Forms

z

~u X

~v X y

x

Figure 6.1. Coordinate basis on the sphere.

Example 6.1.5 (Spheres). Consider the longitude–colatitude parame~ v) = (cos u sin v, sin u sin v, cos v), with trization on the sphere X(u, ~ are (u, v) ∈ (0, 2π) × (0, π). The first derivatives of X

~ u = (− sin u sin v, cos u sin v, 0), X ~ v = (cos u cos v, sin u cos v, − sin v). X We deduce that for this parametrization   2   sin v 0 g11 g21 = . g= 1 0 g12 g22 Figure 6.1 shows the sphere with this parametrization along with ~ ~ v } at the point X(π/2, ~ u, X π/4). the basis {X Some classical differential geometry texts use the letters E = g11 , F = g12 , G = g22 . The classical notation was replaced by the “tensor notation” gij as the notion of a tensor became more prevalent in differential geometry. A discussion of tensor notation is provided in the Appendix A.1 since it is not essential in this book. As a first application of how the first fundamental form allows one to do geometry on a regular surface, consider the problem of calculating the arc length of a curve on the surface. Let S be a ~ : regular surface with a coordinate neighborhood parametrized by X

6.1. The First Fundamental Form

169

U → R3 , where U ⊂ R2 is open. Consider a curve on S given ~ by ~γ (t) = X(u(t), v(t)), where (u(t), v(t)) = α ~ (t) is a differentiable parametrized plane curve in the domain U . The arc length formula over the interval [t0 , t] is Z t s(t) = k~γ 0 (τ )k dτ. t0

After some reworking, we can rewrite the formula as Z tp g11 (u0 (τ ))2 + 2g12 u0 (τ )v 0 (τ ) + g22 (v 0 (τ ))2 dτ s(t) =

(6.1)

t0

or, more explicitly,

s(t) =

Z tp g11 (u(τ ), v(τ ))(u0 (τ ))2 + 2g12 (u(τ ), v(τ ))u0 (t)v 0 (τ ) + g22 (u(τ ), v(τ ))(v 0 (τ ))2 dτ . t0

Using the first fundamental form, we can rewrite the arc length formula as Z tq I~γ (t) (~ α0 (τ ), α ~ 0 (τ )) dτ . (6.2) s(t) = t0

The reader should note that this is an abuse of notation since, for all p ∈ S, Ip (·, ·) is a bilinear form on the tangent space Tp S, but α ~ 0 (t) is a vector function in R2 . However, the justification behind this notation is that for all t, the coordinates of α ~ 0 (t) in the standard basis of R2 are precisely the coordinates of ~γ 0 (t) in the basis ~ u, X ~ v } based at the point ~γ (t). (To be more precise, one could {X write Iγ(t) ([~ α0 (t)], [~ α0 (t)]) but this notation can become rather burdensome.) In essence, Equation (6.2) relates the geometry in the particular coordinate neighborhood of S to the plane geometry in the tangent plane Tp S. More precisely, while doing geometry in U , by using Ip (·, ·) instead of the usual dot product in R2 , one obtains information about what happens on the regular surface S as opposed to simply what happens on the tangent plane Tp S. This approach does not work only for calculating arc length. ~ ~ Consider two plane curves α ~ (t) and β(t) in the domain of X, ~ 0 ) = q. Also let ~γ = X ~ ◦α ~ ◦ β~ be the with α ~ (t0 ) = β(t ~ and ~δ = X

170

6. The First and Second Fundamental Forms

corresponding curves on the regular surface S. Then at the point ~ p = X(q), the curves ~γ (t) and ~δ(t) form an angle of θ, with α0 (t0 ), β~ 0 (t0 )) Ip (~ q . 0 0 0 0 ~ ~ Ip (~ α (t0 ), α ~ (t0 )) Ip (β (t0 ), β (t0 )) (6.3) (Again, the comment following Equation (6.2) applies here as well.) ~ u, X ~ v }, the vectors X ~ u and X ~ v have the coordiIn the basis {X nates (1, 0) and (0, 1), respectively. Therefore, Equation (6.3) implies that the angle ϕ of the coordinate curves of a parametrization ~ X(u, v) is given by cos θ =

~γ 0 (t0 ) · ~δ0 (t0 ) =p k~γ 0 (t0 )k k~δ0 (t0 )k

~v ~u · X Ip ((1, 0), (0, 1)) X p =p ~ u k kX ~ vk Ip ((1, 0), (1, 0)) Ip ((0, 1), (0, 1)) kX g12 =√ . g11 g22

cos ϕ =

Thus, all the coordinate curves of a parametrization are orthogonal ~ If X ~ if and only if g12 (u, v) = 0 for all (u, v) in the domain of X. ~ satisfies this property, we call X an orthogonal parametrization. Example 6.1.6 (Loxodromes). Consider the longitude–colatitude para-

metrization of the sphere, given in Example 6.1.5. Recall that a meridian of the sphere is any curve on the sphere with u fixed and that the coefficients for the first fundamental form are g11 = sin2 v,

g12 = g21 = 0,

g22 = 1.

A loxodrome on the sphere is a curve that makes a constant angle β with every meridian. Let α ~ (t) = (u(t), v(t)) be a curve in ~ α(t)). If ~γ (t) makes the the domain (0, 2π) × (0, π) and ~γ (t) = X(~ same angle β with all the meridians, then by Equation (6.3),     sin2 v 0 0 0 1 1 v0 p = . cos β = p √ (u0 )2 sin2 v + (v 0 )2 1 (u0 )2 sin2 v + (v 0 )2 u0

v0

6.1. The First Fundamental Form

171

Then (u0 )2 sin2 v cos2 β + (v 0 )2 cos2 β = (v 0 )2 ⇐⇒(v 0 )2 sin2 β = (u0 )2 sin2 v cos2 β ⇐⇒(u0 )2 sin2 v = (v 0 )2 tan2 β =⇒ ± u0 cot β =

v0 , sin v

where the ± makes sense in that a loxodrome can travel either toward the north pole or toward the south pole. Now integrating both sides of the last equation with respect to t and then performing a substitution, one obtains Z

0

Z

±u cot β dt =

v0 dt ⇐⇒ sin v

Z

Z

1 dv sin v v  ⇐⇒ (± cot β)u + C = ln tan , 2 ± cot β du =

where we have chosen a form of antiderivative of sin1 v that suits our calculations best. This establishes an equation between u and v that any loxodrome must satisfy. From this, one can deduce the following parametrization for the loxodrome:  (u(t), v(t)) = (± tan β)(ln t − C), 2 tan−1 t . (Figure 6.2 was plotted using tan β = 4 and C = 0.)

Figure 6.2. Loxodrome on the sphere.

172

6. The First and Second Fundamental Forms

Not only does the first fundamental form allow us to talk about lengths of curves and angles between curves on a regular surface, but it also provides a way to calculate the area of a region on the regular surface. ~ : U ⊂ R2 → S be the parametrization for a Proposition 6.1.7. Let X coordinate neighborhood of a regular surface. Then if Q is a compact ~ subset of U and R = X(Q) is a region of S, then the area of R is given by ZZ p det(g) du dv. (6.4) A(R) = Q

Proof: Recall the formula for surface area introduced in multivariable calculus that gives the surface area of the region R on S as ZZ ZZ ~u × X ~ v k du dv. A(R) = dA = kX (6.5) R

Q

(See [25, Section 7.4] or [30, Section 17.7] for an explanation of Equation (6.5).) By Problem 3.1.6, for any two vectors ~v and w ~ in R3 , the following identity holds: (~v × w) ~ · (~v × w) ~ = (~v · ~v )(w ~ · w) ~ − (~v · w) ~ 2. Therefore, 2 ~u × X ~ v k2 = (X ~u · X ~ u )(X ~v · X ~ v ) − (X ~u · X ~ v )2 = g11 g22 − g12 kX , (6.6)

and the proposition follows because g12 = g21 .



Proposition 6.1.7 is interesting in itself as it leads to another property of the first fundamental form. It is obvious from the definition of the first fundamental form that g11 (u, v) ≥ 0 and g22 (u, v) ≥ 0 for all ~ u ×X ~ v k2 = det(g) and since X ~ u ×X ~ v is (u, v) ∈ U . However, since kX never ~0 on a regular surface, we deduce that det(g) = g11 g22 −g12 > 0 for all (u, v) ∈ U . In other words, the matrix of functions g is always a positive definite matrix.

6.1. The First Fundamental Form

173

S

R

Figure 6.3. Surface area patch for Example 6.1.8.

Example 6.1.8. Consider the regular surface of revoluation S param~ etrized by X(u, v) = (v cos u, v sin u, ln v), with (u, v) ∈ [0, 2π] × (0, ∞) = U . (From a practical standpoint, it is not so important that U is not an open subset of R2 , but it should be understood that S may require two coordinate neighborhoods to satisfy the criteria for a regular surface. See Figure 6.3.) Let Q = [0, 2π] × [1, 2], and ~ let’s calculate the area of R = X(Q). It’s not hard to show that   2 v 0 (gij ) = . 0 1 + v12

Then Proposition 6.1.7 gives Z 2π Z 2 p h1 p i2 1 v 2 + 1 dv du = 2π v v 2 + 1 + sinh−1 v A(R) = 2 2 1 0 1 √ !! √ √ 2+ 5 √ = π 2 5 − 2 + ln . 1+ 2 As we have done for many quantities on curves and surfaces, we wish to see how the functions in the first fundamental form change under a coordinate transformation. In order to avoid confusion with

174

6. The First and Second Fundamental Forms

various coordinate systems, we introduce some notations that will become more common in future topics. Suppose that an open subset V of a regular surface S can be parametrized by two different sets of ~ 1 , x2 ) and Y ~ (¯ coordinates (x1 , x2 ) and (¯ x1 , x ¯2 ). If we write X(x x1 , x ¯2 ) ~ −1 ◦ X ~ is for the specific parametrizations, then the function F = Y a diffeomorphism between two open subsets of R2 and ¯2 ) = F (x1 , x2 ). (¯ x1 , x For convenience, we will often simply write ( ¯1 (x1 , x2 ), x ¯1 = x x ¯2 = x ¯2 (x1 , x2 ) to indicate the dependency of one set of variables on the other. Vice versa, we can write ( x1 , x ¯2 ), x1 = x1 (¯ (x1 , x2 ) = F −1 (¯ x1 , x ¯2 ) or simply x2 = x2 (¯ x1 , x ¯2 ). ~ 1 (¯ ~ (¯ ¯2 ) = X(x x1 , x ¯2 ), x2 (¯ x1 , x ¯2 )) for the respective parameNow Y x1 , x trizations. Let gij be the coefficient functions in the first fundamental form for S parametrized by x1 , x2 , and let g¯ij be the coefficient functions for S parametrized by x ¯1 , x ¯2 . Then by the chain rule,     ∂x1 ∂x2 ∂x1 ∂x2 ~ ~ ~ ~ ~ ~ g¯ij = Yx¯i · Yx¯j = Xx1 · X x1 . + Xx2 + Xx2 ∂x ¯i ∂x ¯i ∂x ¯j ∂x ¯j After some reorganization, we find 2 X 2 X ∂xk ∂xl g¯ij = gkl . ∂x ¯i ∂ x ¯j

(6.7)

k=1 l=1

This type of transformation of coordinates is our first (nontrivial) encounter with tensors. Because the gij functions satisfy this particular identity under a change of coordinates, we call the matrix of functions g = (gij ) the components of a tensor of type (0, 2). The collection of the quantities gij defined at each point of p ∈ S is referred to as the components of metric tensor of S.

6.1. The First Fundamental Form

The value of the first fundamental form is that, given the metric tensor, one can use the appropriate formulas for arc length, for angles between curves, and for areas of regions without knowing the specific parametrization of the surface. An additional benefit of the first fundamental form and the metric tensor is that one can still use them to study the geometry of surfaces in higher dimensions, as we will see in Section 9.2. Because of Equation (6.2), a few mathematicians and most physicists use an alternative notation for the metric tensor. These authors describe the metric tensor by saying that the “line element” in a coordinate patch is given by ds2 = g11 (u, v) du2 + 2g12 (u, v) du dv + g22 (u, v) dv 2 . This notation has the advantage that one can write it concisely on a single line instead of in matrix format. However, one should not forget that it not only leads to Equation (6.2) for arc length of a curve on a surface but that it also leads to Equation (6.3) for angles between curves on the surface and the formula in Proposition 9.2.2 that gives the area of regions on the surface. Example 6.1.9 (Intrinsic Geometry). If we are given the metric coefficients gij (u, v) of a surface as functions of u and v, we do not know the actual equations of the surface in space in order to compute quantities like the length of a curve or the area of a coordinate patch in the domain of the metric coefficients. Such computations are an example of “intrinsic geometry of a surface,” namely the study of geometric computations that only depend on the metric coefficients. One of the most important examples of intrinsic geometry is given by the metric coefficient functions g11 (u, v) = 1/v 2 , g12 (u, v) = 0 = g21 (u, v), and g22 (u, v) = 1/v 2 defined on the domain consisting of all points in the (u, v) plane with v > 0, the “upper half-plane.” For example, we can compute the length of the curve given by (u(t), v(t)) = (t, c) where a ≤ t ≤ b by calculating the integral Z br 1 0 2 1 u (t) + 2(0)u0 (t)v 0 (t) + 2 v 0 (t)2 dt 2 c c a Z b Z br 1 1 + 0 + 0 dt = (1/c) dt = (b − a). = 2 c c a a

175

176

6. The First and Second Fundamental Forms

We can also calculate the length of the curve (u(t), v(t)) = (a, t) where 0 < c ≤ t ≤ d by Z d Z dr 1 1 0 + 0 + 2 (1) dt = 1/tdt = [ln(t)]dc = ln(d) − ln(c). t2 t c c Note that the limit of this length as d goes to infinity is infinity and the same is true as c goes to 0. It follows that the length of the boundary of the domain a ≤ u ≤ b, and c ≤ v ≤ d, is given by (b − a)(1/c) − (b − a)(1/d) + 2(ln(d) − ln(c)). Since the area of this coordinate patch depends only on the metric the area of thep coordinate patch coefficient gij (u, v), we can compute p in this metric by integrating g1 1(u, v)g2 2(u, v) = (1/v 2 )(1/v 2 ) = 1/v 2 to get   Z bZ d 2 1 1 1v − . dv du = (b − a) c d a c Note that the limit of the area of this domain using these metric coefficients goes to (b − a)(1/c) as d goes to infinity, and as c goes to 0, the area becomes infinitely large. The upper half-plane with this alternative metric tensor is called the Poincar´e Upper Half-Plane.

Problems ~ t) 6.1.1. Calculate the metric tensor gij for the plane parametrized by X(s, 3 = ~u + s~v + tw ~ with (s, t) ∈ R . 6.1.2. Calculate the metric tensor gij for the surface with parametrization ~ X(u, v) = (u2 − v, uv, u). 6.1.3. Let a be a nonzero real constant. Calculate the metric tensor for the ~ helicoid parametrized by X(u, v) = (v cos u, v sin u, au). 6.1.4. Calculate the metric tensor for the following families of conic surfaces: ~ (a) X(u, v) = (a cos u sin v, b sin u sin v, c cos v), ellipsoid. ~ (b) X(u, v) = (av cos u, bv sin u, v 2 ), elliptic paraboloid.

6.1. The First Fundamental Form

~ (c) X(u, v) = (au cosh v, bu sinh v, u2 ), hyperbolic paraboloid. ~ (d) X(u, v) = (a cosh v cos u, b cosh v sin u, c sinh v), hyperboloid of one sheet. ~ (e) X(u, v) = (a cos u sinh v, b sin u sinh v, c cosh v), hyperboloid of two sheets. 6.1.5. Let 0 < r < R be real numbers and consider the torus parametrized by ~ X(u, v) = ((R + r cos v) cos u, (R + r cos v) sin u, r sin v). ~ is U = (0, 2π) × (0, 2π), then the (a) Show that if the domain of X parametrization is regular and that if U = [0, 2π] × [0, 2π] the parametrization is surjective. (b) Calculate the metric tensor of this torus. (c) Use the metric tensor to calculate the area of the torus. 6.1.6. Let U be an open subset of R2 , and let f : U → R be a two-variable function. Explicitly calculate the metric tensor for the graph of ~ ~ : U → R3 , with X(u, v) = f , which can be parametrized by X (u, v, f (u, v)). Prove that the coordinate lines are orthogonal if and only if fu fv = 0. 6.1.7. Let γ be a loxodrome on the sphere as described in Example 6.1.6. Prove that the arc length of the loxodrome between the north and south pole is π sec β (regardless of the constant of integration C). 6.1.8. Consider the surface of revolution parametrized by ~ X(u, v) = (f (v) cos u, f (v) sin u, g(v)), with f and g chosen so that the surface is regular (see Problem 5.2.7). Calculate the metric tensor. 6.1.9. Let α ~ (t) be a regular space curve and consider the tangential surface ~ u) = α S parametrized by X(t, ~ (t) + uT~ (t). Calculate the metric ~ tensor for X. 6.1.10. Tubes. Let α ~ (t) be a regular space curve. We call the tube of ~ u) = radius r around α ~ the surface that is parametrized by X(t, ~ ~ α ~ (t) + (r cos u)P (t) + (r sin u)B(t). Calculate the metric tensor for ~ Supposing that the tube is regular, prove that the area of the X. tube is 2πr times the length of α ~. ~ 6.1.11. Consider the right circular cone parametrized by X(u, v) = (v cos u, v sin u, v).

177

178

6. The First and Second Fundamental Forms

(a) Let α ~ (t) be a curve in the plane such that for all t, the vectors α ~ (t) and α ~ 0 (t) make a constant angle of β. Prove that α ~ (t) = (f (t) cos t, f (t) sin t), where f (t) = Re(cot β)t for any real R > 0. ~ f (t)) on the cone. Cal(b) Consider the spiral curve ~γ (t) = X(t, culate the arc length of ~γ for 0 ≤ t ≤ 2π. (It will depend on R and β.) ~ : R2 → R3 of the unit sphere by 6.1.12. Consider the parametrization Y stereographic projection (see Problem 5.2.15 and use the parame~ (u, v) = π −1 (u, v)). trization Y (a) Calculate the corresponding metric tensor component functions gij . ~ of the sphere (as pre(b) Consider the usual parametrization X sented in Example 6.1.5.) Give appropriate domains for and explicitly determine the formula for the change of coordinate ~ −1 ◦ X ~ = π ◦ X. ~ system given by F −1 = Y (c) Explicitly verify Equation (6.7) in this context. ~ : R2 − {(0, 0)} → R3 defined by 6.1.13. Consider the parametrization X 1 2 2 ~ X(x, y) = (x, y, 2 ln(x + y )). ~ is another parametrization for the Log Trumpet (a) Show that X discussed in Example 6.1.8. (b) Calculate the components of the metric tensor associate to this parametrization. (c) Use this and the parametrization given in Example 6.1.8 to illustrate how your result exemplifies the coordinate change properties of the metric tensor components as described in Equation 6.7. 6.1.14. Suppose that the first fundamental form has a matrix of the form   1 0 g= . 0 f (u, v) Prove that all the v-coordinate lines have equal arc length over any interval u ∈ [u1 , u2 ]. In this case, the v-coordinate lines are called parallel. 6.1.15. Let (gij ) be the metric tensor of some surface S parametrized by the coordinates (x1 , x2 ) ∈ U , where U is some open subset in R2 .

6.1. The First Fundamental Form

Suppose that we reparametrize the surface with coordinates (¯ x1 , x ¯2 ) in such a way that ( x1 = f (¯ x1 , x ¯2 ), ¯2 , x2 = x where f : V → U is a function with V ⊂ R. Let (¯ gkl ) be the metric tensor to S under this parametrization. (a) Prove that (¯ gkl ) is a diagonal matrix if and only if the function f satisfies the differential equation x1 , x ¯2 ), x ¯2 ) ∂f g12 (f (¯ . =− ∂x ¯2 g11 (f (¯ x1 , x ¯2 ), x ¯2 ) (b) (ODE) Use the above result and the existence theorem for differential equations to prove that every regular surface admits an orthogonal parametrization. 6.1.16. Let C be a regular value of the function F (x, y, z) and consider the regular surface S defined by F (x, y, z) = C. Without loss of generality, suppose that the variables x and y can be used to parametrize a neighborhood U of some point p ∈ S. Use implicit differentiation to calculate the metric tensor functions gij in terms of derivatives of F. ~ 6.1.17. Consider the parametrization X(u, v) = (cos u sin v, sin u sin v, cos v) of the unit sphere. It is easy to show that the unit normal vector ~ to the sphere at (u, v) is again the vector N ~ (u, v) = X(u, ~ N v). Let f (u, v) be a nonnegative real function in two variables such that f (u, 0) is constant, f (u, π) is constant, and for all fixed v0 , f (u, v0 ) is periodic 2π. A normal variation to a given surface S is a surface created by going out a distance of f (u, v) along the normal vector ~ of S, given by N ~ (u, v) = X(u, ~ ~ (u, v). Y v) + f (u, v)N A normal variation of the unit sphere is therefore given by ~ (u, v) = (1 + f (u, v))X(u, ~ Y v). ~. (a) Calculate the metric tensor for the parametrization Y (b) Use the explicit function f (u, v) = cos2 (2u) cos2 (2v − π/2). ~. Calculate the metric tensor for Y 6.1.18. Consider the metric coefficients g11 (u, v) = 1/v 2 , g12 (u, v) = 0 = g21 (u, v), and g22 (u, v) = 1/v 2 on the upper half-plane in Example 6.1.9.

179

180

6. The First and Second Fundamental Forms

√ √ (a) Compute the length of the curve (u, v) = ( 2 cos t, 2 sin t) with π/4 ≤ t ≤ 2π/4 from (1, 1) to (−1, 1) using these metric coefficients. (b) Show that the length of the curve is less than the length of the segment (u(t), v(t)) = (t, 1) with −1 ≤ t ≤ 1 using these metric coefficients. (c) Show that the length of the curve (u(t), v(t)) = (R cos t, R sin t) where R is a constant and pi/4 ≤ t ≤ 3π/4 is independent of R. 6.1.19. Consider the metric defined in the previous problem on the upper half-plane. (a) Compute the area above the curve (u(t), v(t)) = (R cos t + m, R sin t) with π/4 ≤ t ≤ 3π/4 and below the line v = d, for some d > R. Observe that the area is independent of the constant m. What happens to the area of this region as d approaches infinity? (b) Repeat the same question with (u, v) = (R cos t + m, R sin t) but with 0 ≤ t ≤ π. [Hint: This will involve an improper integral.] 6.1.20. Consider the general metric coefficients g11 (u, v) = g22 (u, v) = f (v) and g12 (u, v) = g21 (u, v) = 0 defined on the upper half-plane. Find an expression for the length of the curve (u(t), v(t)) for a ≤ t ≤ b.

6.2 Map Projections (Optional) 6.2.1 Metric Properties of Maps of the Earth In order to illustrate properties and uses of the first fundamental form, we propose to briefly discuss maps of the Earth. Through history, maps of portions of the Earth have helped political leaders understand the geography of regions over which they exert influence. Starting particularly in the age of exploration, traders and navigators needed maps that describe large regions of the Earth. For navigation purposes, it would be ideal if a map accurately represented distances between points, angles between directions, and areas of regions. Distances, angles, and areas are the metric properties of a map. However, because a map is typically drawn on a flat piece of paper whereas the Earth is (very close to) spherical, no map can accurately reflect all metric properties at the same time.

6.2. Map Projections (Optional)

181

Points on the Earth are usually located using coordinates of latitude then longitude. In other places in this text (Example 6.1.5), we introduced the mathematicians’ customary way of parametrizing the unit sphere: the polar coordinates angle θ, which is essentially the longitude, and then the colatitude ϕ, which is the angle down from the positive z-axis. To cover the sphere, we use 0 ≤ θ < 2π and 0 ≤ ϕ ≤ π. There is no absolute consensus on which coordinates is listed first. If we denote geographer’s latitude by ϕg , then ϕ + ϕg = π/2. Hence, the North Pole has ϕg = π/2, the equator is at ϕg = 0, and the South Pole has ϕg = −π/2. (Of course, latitude and longitude are typically described with degrees.) For the purposes of our discussion, we will use the parametrization of the sphere ~ X(ϕ, θ) = (R sin ϕ cos θ, R sin ϕ sin θ, R cos ϕ)

(6.8)

with colatitude 0 ≤ ϕ ≤ π and longitude −π < θ ≤ π and where R is the radius of the Earth. Recall that coordinate lines with θ constant are called meridians. Note that colatitude lines are precisely latitude lines. Points on a map are given by a pair of coordinates, say x and y, which are standard Cartesian coordinates of the plane. A map projection corresponds to a function from (ϕ, θ) coordinates to (x, y) coordinates, M (ϕ, θ) = (x(ϕ, θ), y(ϕ, θ)). (6.9) Note that maps generally do not show the x and y grid but rather the curves in the xy-plane that correspond to constant θ and ϕ coordinates. Using the standard parametrization of the Earth (6.8), the metric properties of the Earth are encapsulated by the first fundamental form given by its metric tensor !   ~ϕ X ~θ ~ϕ · X ~ϕ · X R2 0 X gEarth = ~ ~ ~ θ = 0 R2 sin2 ϕ . ~θ · X Xθ · Xϕ X From this, we deduce that the line element ds satisfies ds2 = R2 (dϕ2 + sin2 ϕdθ2 ),

182

6. The First and Second Fundamental Forms

which means in other words dsEarth

q = R (ϕ0 )2 + sin2 ϕ(θ0 )2 dt.

The surface area element is p dSEarth = det g dϕ dθ = R2 sin ϕ dϕ dθ. For the purposes of calculations, some authors set R = 1 so that the radius of the Earth becomes the unit length in length and area calculations. By keeping R explicit, we can give it any necessary value in any given system of units. Now the map function M in (6.9) is a function M : [0, π] × [−π, π] → R2 . This can be understood as an alternate parametrization of a portion of the plane using the coordinates ϕ and θ. It has a first fundamental form defined by   Mϕ · Mϕ Mϕ · Mθ , gMap = Mθ · M ϕ Mθ · M θ where the dot products require us to view M (ϕ, θ) and its derivatives as a vector. Note that at each point p on a surface, the first fundamental form is an inner product Ip ( , ) on the tangent plane Tp S to the surface at p. However, in the case of a map, the surface is a subset of R2 and the tangent plane is R2 itself. Definition 6.2.1. A conformal map is a map in which the angles of intersection of paths on the map are the same as the angles of intersection of paths on the Earth.

Conformal maps are particularly useful for navigation since they can tell a pilot that at least he or she is heading in a desired direction. Proposition 6.2.2. A map is conformal if there is a positive function

w(ϕ, θ) such that gMap = w(ϕ, θ)gEarth . Proof: Let ~a and ~b be vectors in the tangent plane to the Earth at a ~ ϕ, X ~ θ) point p = (ϕ, θ), with components given with respect to the (X

6.2. Map Projections (Optional)

183

ordered basis of the tangent plane. The angle α between them has a cosine of Ip (~a, ~b) q . cos α = p Ip (~a, ~a) Ip (~b, ~b) If there is a positive function w(ϕ, θ) as described in the proposition, then the first fundamental form of the map is IpMap ( , ) = w(p)Ip ( , ). Then cosine of the angle between directions in the map is w(p)Ip (~a, ~b) IpMap (~a, ~b) q q =p w(p)Ip (~a, ~a) w(p)Ip (~b, ~b) IpMap (~a, ~a) IpMap (~b, ~b)

q

Ip (~a, ~b) q = cos α. =p Ip (~a, ~a) Ip (~b, ~b) The proposition follows.



Maps that preserve area are maps such that there exists a constant D so that the area of any region measured in the map is a factor D times that surface area of the corresponding region on the Earth. Hence, for all regions R in the ϕθ-plane, we have ZZ ZZ dSMap = D dSEarth R Z ZR ZZ p p ⇐⇒ det gMap dϕ dθ = D det gEarth dϕ dθ. R

R

The Mean Value Theorem for integrals gives the following proposition. Proposition 6.2.3. A map of the Earth is area-preserving if and only

if there is a positive constant D such that det(gMap ) = D2 det(gEarth ). p The arc length element in a map is ds = (x0 (t))2 + (y 0 (t))2 dt, which, in particular, does not depend on the position in the plane

184

6. The First and Second Fundamental Forms

but only on the components of the velocity vector. The explicit dependence of dsEarth on ϕ shows that no map can calculate arc lengths of all paths correctly. In other words, there is no map with a given proportionality factor such that the arc length of the path in the map is that factor times the arc length of the corresponding path on the Earth. However, it is still possible that certain sets of paths in a given map calculation have the same relative arc lengths as on the Earth.

6.2.2 Azimuthal Projections A first category of map projection is the azimuthal projections. One point P on the Earth is chosen as the center of the map and serves as the origin in the xy-coordinate system. If P is the North Pole, then the longitude lines through P appear on the map as equally spaced rays radiating from the origin, and the latitude circles appear on the map as circles centered on the map with radii determined by the particular projection. (If P is the South Pole, longitude and latitude lines have a similar representation on the map. However, if P is not one of the poles, then the radial lines in the map and circles at given radii around the origin have more complicated interpretations in latitude and longitude on the Earth.) Choosing P as the North Pole, any azimuthal projection maps the (ϕ, θ) coordinates to polar coordinates (r, θ) = (r(ϕ), θ) on the map. Hence, the map function has the form (x, y) = M (ϕ, θ) = (r(ϕ) cos θ, r(ϕ) sin θ), where r is a nonnegative increasing function with r(0) = 0. The first fundamental form of any azimuthal projection is  0 2  r (ϕ) 0 gMap = . 0 r(ϕ) Different choices of the function r(ϕ) correspond to different azimuthal projections. Even if we always pick the center of the map to be the North Pole, there are a variety of classical choices. Some give a map for the whole Earth (except for the South Pole), while others only map the northern hemisphere.

6.2. Map Projections (Optional)

Example 6.2.4 (Orthographic Projection). One of the geometrically sim-

plest azimuthal projections is the orthographic projection. It maps only the northern hemisphere. We imagine the map as the tangent plane T to the North Pole. A point Q in the northern hemisphere of the Earth is mapped to the point Q0 in the tangent via orthogonal projection onto the tangent plane T . The resulting map fills a circular disc in the plane. This disc is then scaled uniformly by a factor c to any reasonable radius for a map. (If the disc were not scaled, it would have a radius equal to the radius of the Earth, which would make for a very large map!) In this case, the function r(ϕ) is simply r(ϕ) = cR sin ϕ for 0 ≤ ϕ ≤ π/2. Example 6.2.5 (Gnomic Projection). Another simple azimuthal projec-

tion is the gnomic projection. It also maps only the northern hemisphere and this time excludes the equator. We imagine the map as the tangent plane T to the North Pole. Let O be the center of the Earth. A point Q in the northern hemisphere of the Earth is mapped to the point Q0 in the tangent plane where Q0 is the intersection of −−→ the ray OQ with the tangent plane T . The resulting map fills the whole plane. As usual, the tangent plane is then scaled uniformly by a reasonable factor c. In practice, one only depicts a certain bounded portion of the map. In this case, the function r(ϕ) is simply r(ϕ) = cR tan ϕ for 0 ≤ ϕ < π/2. One interesting property about the gnomic projection is that every arc of a great circle on the Earth is mapped to a straight line segment. Indeed, a great circle on a sphere corresponds to the intersection of the sphere and a plane P through the origin. Via the gnomic projection, the arc of such a great circle maps to the intersection of P ∩ T , which is a line. Example 6.2.6 (Stereographic Projection). Problem 5.2.15 introduced

stereographic projection as an alternate parametrization of the sphere. However, as presented in that problem, it defines an azimuthal projection with the South Pole as the center. In this example, we make a few adjustments to the presentation in Problem 5.2.15 to conform to the presentation here.

185

186

6. The First and Second Fundamental Forms

Like the gnomic projection, the stereographic projection assumes the map is the tangent plane T to the North Pole. (This a variation from the presentation in Problem 5.2.15, where the map was assumed to be the plane through the equator.) A point Q on the Earth (except for the South Pole) is mapped to the point Q0 , which is the intersection of T and the ray out of the South Pole through Q. As usual, the tangent plane is then scaled uniformly by a reasonable factor c. The stereographic projection maps the whole Earth except for the South Pole and covers the whole plane. We leave it as an exercise to show that in this case, the function r(ϕ) is 2cR sin ϕ r(ϕ) = for 0 ≤ ϕ < π. 1 + cos ϕ Example 6.2.7 (Area-Preserving Azimuthal). Suppose that we want to

devise an area-preserving azimuthal projection. Then by Proposition 6.2.3, the radial function r(ϕ) must be such that det(gMap ) = r(ϕ)2 (r0 (ϕ))2 = D2 R4 sin2 ϕ for some positive constant D. Taking a square root of both sides, we get r(ϕ)r0 (ϕ) = DR2 sin ϕ. Integrating both sides with respect to ϕ and using substitution, we find that Z Z 1 2 0 −DR cos ϕ = r(ϕ)r (ϕ) dϕ = r dr = r2 + C, 2 for some constant of integration. Since we want r(0) with ϕ = 0, we have C = −DR2 and thus p √ ϕ r(ϕ) = R 2D(1 − cos ϕ) = 2 DR sin 2 for 0 ≤ ϕ < π. Hence, there exists an area-preserving azimuthal projection that covers the whole √ Earth except for the South Pole and its map is a disc of radius 2 DR.

6.2.3 Cylindrical Projections A cylindrical projection map is one based off of the intuition of taking a globe, wrapping a sheet of paper as a cylindrical tube around the

6.2. Map Projections (Optional)

Figure 6.4. Cylindrical projection.

equator of the globe, projecting the surface of the globe onto the sheet of paper, and then unwrapping the cylinder of paper to lay it flat. See Figure 6.4 for a visual. More precisely, one point P on the Earth is chosen as a pole. If P is the North Pole, then longitude lines through P appear on the map as equally spaced vertical parallel lines and latitude circles appear on the map as horizontal lines, with spacing determined by the particular projection. The choice of point P corresponds to a choice of an axis for the cylinder because we assume the cylinder has ←→ axis OP , where O is the center of the sphere. We always scale the map horizontally so that a unit in the x direction corresponds to a unit in θ. With P as the North Pole, then (ϕ, θ) on the Earth maps to (x, y) = M (ϕ, θ) = (θ, h(ϕ)), where h is a decreasing function with h(π/2) = 0. That h is decreasing corresponds to setting the North Pole ϕ = 0 as up on the map. The first fundamental form of such maps is  0  (h (ϕ))2 0 gMap = . 0 1 Different choices of the function h(ϕ) correspond to different cylindrical projections. Also h(ϕ) may be scaled linearly simply to adjust the vertical size of the map. As with the azimuthal projections, even setting P as the North Pole, there are a number of classical choices.

187

188

6. The First and Second Fundamental Forms

Example 6.2.8 (Radial Cylindrical). Let O represent the center of the

Earth. The radial cylindrical projection is a geometrically simple projection in which each point Q on the Earth, except for the North Pole and the South Pole, is mapped to the point Q0 on the cylinder ←−− that is the intersection with the ray OQ. The cylinder is then unrolled and, as usual, scaled by a reasonable factor c. The map then consists of the strip [0, 2π] × R. In this projection, we have h(ϕ) = c cot ϕ, where c is a constant. Example 6.2.9 (Mercator). The well-known Mercator projection is a cylindrical projection that is conformal. By Proposition 6.2.2, the Mercator projection must be such that there is a function w(ϕ, θ) such that   2   0 R 0 (h (ϕ))2 0 = w(ϕ, θ) . 0 1 0 R2 sin2 ϕ

Obviously, we must have w(ϕ, θ) = 1/R2 sin2 ϕ and then we also need π  1 h0 (ϕ) = − with h = 0. R sin ϕ 2 The choice of the negative sign for h0 (ϕ) comes from the requirement that h(ϕ) is decreasing. Integrating, we find that   Z ϕ  1 + cos ϕ dπ π = ln = ln cot , h(ϕ) = − sin ϕ 2 π/2 sin ϕ for 0 < ϕ < π/2. Hence,  the map for the Mercator projection is M (ϕ, θ) = θ, ln cot π2 . Many other classes of map projections exist. For example, pseudocylindrical projections attempt to fix the distortions that invariably occur near the poles in cylindrical projections. Indeed, on a sphere, latitude circles decrease in radius the closer one gets to the poles. However, under any cylindrical projection every latitude circle becomes a horizontal line segment of the same length. A pseudocylindrical projection has for its map   w(ϕ)θ (x, y) = M (ϕ, θ) = , h(θ) 2π where

6.2. Map Projections (Optional)

189

• as for cylindrical projections, h(ϕ) is a decreasing function with h(π/2) = 0; • w(ϕ) is a nonnegative function that gives the width of the map at a given latitude ϕ. For many maps, the function w(ϕ) satisfies w(0) = w(π) = 0 and w(π/2) = 1, so the width is 0 at the North and South Pole and 1 at the equator.

6.2.4 Coordinate Changes on the Sphere With various maps for the Earth (sphere) at our disposal, we take the opportunity to illustrate the coordinate change transformation property of the metric tensor as given in Equation (6.7). To locate points on the sphere (Earth), we will consider the standard latitude and longitude (ϕ, θ) coordinates and the (x, y) coordinates as given by the stereographic projection described in Example 6.2.6. For simplicity, let us use R = c = 1. Then the stereographic map is   2 sin ϕ cos θ 2 sin ϕ sin θ (x, y) = M (ϕ, θ) = , . 1 + cos ϕ 1 + cos ϕ Using similar geometry as required by Problem 5.2.15, we can show that the (x, y) coordinates parametrize the unit sphere by   4x 4y 4 − x2 − y 2 ~ . , , Y (x, y) = 4 + x 2 + y 2 4 + x 2 + y 2 4 + x2 + y 2 In order to verify Equation (6.7), let us call the (x1 , x2 ) coordinates the (x, y) coordinates and we call the (¯ x1 , x ¯2 ) coordinates the (ϕ, θ) coordinates. In these labels, we already know that   1 0 (¯ gij ) = . 0 sin2 ϕ ~ (x, y), we find that Using the parametrization Y  16 ! ~ ~ ~ ~ Y ·Y Y ·Y  (4 + x2 + y 2 )2 (gij ) = ~x ~x ~x ~y =  Yy · Yx Yy · Yy 0

0

 . 16 (4 + x2 + y 2 )2 (6.10)

190

6. The First and Second Fundamental Forms

Furthermore, the differential [dM(ϕ,θ) ], also known as the Jacobian matrix of the coordinate change for (ϕ, θ) coordinates to (x, y) coordinates, is  ∂x  ∂ϕ [dM(ϕ,θ) ] =   ∂y ∂ϕ

∂x   2 cos θ ∂θ   1 + cos ϕ   ∂y  =  2 sin θ ∂θ 1 + cos ϕ

2 sin ϕ sin θ  1 + cos ϕ   2 sin ϕ cos θ  . 1 + cos ϕ

Recall that Equation (6.7) states that 2 X 2 X ∂xk ∂xl gkl . g¯ij = ∂x ¯i ∂ x ¯j k=1 l=1

This can be rewritten more explicitly as g¯ij =

∂x1 ∂x2 ∂x2 ∂x1 ∂x2 ∂x2 ∂x1 ∂x1 g11 + g12 + g21 + g22 . ∂x ¯i ∂ x ¯j ∂x ¯i ∂ x ¯j ∂x ¯i ∂ x ¯j ∂x ¯i ∂ x ¯j

We consider our specific example. However, note that as a simplification, we have g12 = g21 = 0. Also, we point out ahead of time that     2 sin ϕ cos θ 2 2 sin ϕ sin θ 2 8 2 2 . + +4= x +y +4= 1 + cos ϕ 1 + cos ϕ 1 + cos ϕ Thus

16 (1 + cos ϕ)2 . = (x2 + y 2 + 4)2 4

So we get the following confirmation: ∂x1 ∂x1 ∂x2 ∂x2 g11 + g22 ∂x ¯1 ∂ x ¯1 ∂x ¯1 ∂ x ¯1  2   2 cos θ 2 2 sin θ 16 16 = + 1 + cos ϕ (4 + x2 + y 2 )2 1 + cos ϕ (4 + x2 + y 2 )2 4 sin2 θ (1 + cos ϕ)2 4 cos2 θ (1 + cos ϕ)2 + = (1 + cos ϕ)2 4 (1 + cos ϕ)2 4 = 1;

g¯11 =

6.2. Map Projections (Optional)

191

∂x1 ∂x1 ∂x2 ∂x2 g11 + g22 ∂x ¯1 ∂ x ¯2 ∂x ¯ ∂x ¯  1 2      16 2 sin ϕ cos θ 16 2 cos θ 2 sin θ 2 sin ϕ sin θ − = + 2 2 2 1 + cos ϕ 1 + cos ϕ (4 + x + y ) 1 + cos ϕ 1 + cos ϕ (4 + x2 + y 2 )2 = 0;

g¯12 =

we can verify that g¯21 = g¯12 = 0; and finally ∂x1 ∂x1 ∂x2 ∂x2 g11 + g22 ∂x ¯2 ∂ x ¯2 ∂x ¯2 ∂ x ¯2 2    2 sin ϕ cos θ 2 16 16 2 sin ϕ sin θ = − + 2 2 2 1 + cos ϕ (4 + x + y ) 1 + cos ϕ (4 + x2 + y 2 )2 4 sin2 ϕ (1 + cos ϕ)2 = (1 + cos ϕ)2 4 2 = sin ϕ.

g¯22 =

These calculations illustrate the coordinate change transformation described by Equation (6.7)

Problems 6.2.1. Find the possible maps for cylindrical projections that are areapreserving. 6.2.2. Verify the calculations for the metric tensor given in Equation (6.10). 6.2.3. Show that the metric tensor of the map for the general pseudocylindrical projection is  2 0  1 4π h (ϕ) + θ2 w0 (ϕ) θw(ϕ)w0 (ϕ) . gMap = θw(ϕ)w0 (ϕ) w(ϕ)2 4π 2 6.2.4. The Sinusoidal Projection is a pseudocylindrical map projection that  uses w(ϕ) = sin ϕ and h(ϕ) = c π2 − ϕ , where c is a positive constant. (a) Prove that the lengths of the latitude lines on the map are a fixed multiple of the lengths of the latitude lines on the Earth. (b) Prove that this map also preserves areas. (c) Decide if the map is conformal. (d) Describe the shape of the image of the map.

192

6. The First and Second Fundamental Forms

6.3 The Gauss Map In Section 2.2, given any closed, simple, regular curve γ : I → R2 in the plane, we defined the tangential indicatrix to be the curve given by T~ : I → R2 , the unit tangent vector of γ(t). The image of T~ lies on the unit circle, and though T~ is a closed curve, it need not be regular as its locus may stop and double back. Regardless of the parametrization, the tangential indicatrix is well defined up to a change in sign. On any regular surface S in R3 , the tangent plane Tp S at a point p ∈ S is a two-dimensional subspace of R3 , and hence the vectors that are normal to S at p form a one-dimensional subspace. Thus, there exist exactly two possible choices for a unit normal vector. Proposition 5.5.4 implies that if the surface is orientable, we can specify its orientation by a continuous function n : S → S2 , where S2 means the unit sphere in R3 . Definition 6.3.1. Let S be an oriented regular surface in R3 with an

orientation n : S → S2 . In the classical theory of surfaces, the function n is also called the Gauss map. It is important to remain aware of the distinction between the ~ . Let S be an oriented surface with orientation functions n and N 2 ~ : U → R3 , where U is an open n : S → S . Suppose that X 2 subset of R , parametrizes a neighborhood of S, then the function ~ : U → S2 is defined in reference to X ~ by Equation (5.7). The N ~ ~ is a positively functions n and N are related by the fact that if X oriented parametrization, then ~ = n ◦ X. ~ N Example 6.3.2. Consider as an example a sphere S in R3 equipped

with the orientation n, with the unit vectors normal to S pointing away from the center ~c of the sphere. The sphere can be given as the solution to the equation k~x − ~ck2 = R2 .

(6.11)

Recall that k~v k2 = ~v · ~v . If ~x(t) is any curve on the sphere, then by differentiating the relationship in Equation (6.11), one obtains 0

~x (t) · (~x(t) − ~c) = 0.

6.3. The Gauss Map

193

z

y

x

A B

D

B D

A

x y

Figure 6.5. Gauss map of the elliptic paraboloid.

The tangent plane to the sphere S at a point p~ consists of all possible vectors ~x0 (t0 ), where ~x(t) is a curve on S, with ~x(t0 ) = p~. Therefore, at any point p~ on the sphere, there are two options for unit normal vectors: p~ − ~c p~ − ~c or − . k~ p − ~ck k~ p − ~ck However, it is the former that provides the outward-pointing orientation n. Thus the Gauss map for the sphere of radius R and center ~c is explicitly p~ − ~c . n(~ p) = R Furthermore, if S is itself the unit sphere centered at the origin, then the Gauss map is the identity function.

Example 6.3.3. Consider the elliptic paraboloid S defined by the equa-

tion z = x2 + y 2 . This is an orientable surface, and suppose it is oriented with the unit normal always pointing in a positive zdirection. (Figure 6.5 shows this is possible.) Using the function F (x, y, z) = z − x2 − y 2 , the elliptic paraboloid is given by the equation F (x, y, z) = 0. By Proposition 5.2.15, for all (x, y, z) ∈ S, the ~ (x, y, z) is a normal vector, so (since the z-direction is gradient ∇F

194

6. The First and Second Fundamental Forms

positive) the Gauss map is (−2x, −2y, 1) . n(x, y, z) = p 4x2 + 4y 2 + 1 Figure 6.5 shows how the Gauss map acts on the patch {(x, y, z) ∈ S | − 1 ≤ x ≤ 1 and − 1 ≤ y ≤ 1}. It is not hard to show that the image of the Gauss map on the unit sphere is the upper hemisphere. Indeed, using cylindrical coordinates, the Gauss map is n(r, θ, z) =

(−2r cos θ, −2r sin θ, 1) √ . 4r2 + 1

√ The function f (r) = 1/ 4r2 + 1 is a bijection from [0, +∞) to (0, 1] and it is decreasing. Thus, for any z0 ∈ (0, 1], there exists a unique r0 ∈ [0, +∞) with f (r0 ) = z0 . Then, with θ ∈ [0, 2π], the image of n(r0 , θ, z0 ) is a circle on the unit sphere at height z0 . Thus, the image of n is n(S) = {(x, y, z) ∈ S2 | z > 0}. Example 6.3.4. In contrast to Example 6.3.3, consider the hyperbolic

~ : R2 → R3 , with paraboloid given by the parametrization X ~ X(u, v) = (u, v, u2 − v 2 ). We calculate ~ u = (1, 0, 2u) X

and

~ v = (0, 1, −2v), X

and thus, ~ ~ ~ = Xu × Xv = √(−2u, 2v, 1) . N ~u × X ~ vk 4u2 + 4v 2 + 1 kX Like Figure 6.5, Figure 6.6 shows the Gauss map for a “square” on the hyperbolic paraboloid. These two examples manifest a central behavior of the Gauss map. On the elliptic paraboloid, the Gauss map preserves the orientation of the square in the sense that if one travels along the boundary in a clockwise sense on the surface, one also travels along the boundary of the image of the square mapped by n in a clockwise sense. On the other hand, with the hyperbolic paraboloid, the Gauss map reverses the orientation of the square.

6.3. The Gauss Map

195

z D

x

y

z

A

D

B

C A

B y

Figure 6.6. Gauss map on the hyperbolic paraboloid.

~ :U → Proposition 6.3.5. Let S be a regular surface in R3 , and let X

~ ), where U R3 be a parametrization of a coordinate patch V = X(U 2 ~ is an open set in R . Define the vector function N : U → R3 by ~ ~ ~ = Xu × Xv . N ~u × X ~ vk kX ~ is of class C r−1 . ~ is of class C r , then N If X

~ : Proof: By Problem 3.1.18, we deduce that if F~ : U → R3 , G 3 1 U → R , and h : U → R are of class C , then the three functions ~ : U → R, F~ × G ~ : U → R3 , and hF~ : U → R3 are also of class F~ · G C 1 , and their partial derivatives follow appropriate product rules. By definition, ~v ~u × X X ~ =q . N ~u × X ~ v ) · (X ~u × X ~ v) (X ~ will involve One can see that any partial derivative of any order of N the partial derivative of a combination of the above three types of products and the derivative of an expression of the form  −k/2 ~u × X ~ v ) · (X ~u × X ~ v) f (u, v) = (X ,

196

6. The First and Second Fundamental Forms

~u × X ~ v k is where k is an odd positive integer. Since S is regular, kX ~ never 0, so a particular higher derivative of N will exist if that higher ~ u and for X ~ v . The proposition follows. derivative exists for X  One would be correct to think that the different behaviors in Figures 6.5 and 6.6 can be illustrated and quantified by a function between tangent spaces to S and S2 , respectively. This perspective is encapsulated in the notion of the differential of functions between regular surfaces. We develop the more general theory of differentials of functions between manifolds in Chapter 11 of . However, the differential of the Gauss map, though only a particular case of the differentials of functions between surfaces, serves a central role in the rest of this chapter. Let S be a regular oriented surface with orientation n, and let ~ : U → R3 be a regular positively oriented parametrization of a X ~ ) of S. Suppose also that S is of class C 2 , coordinate patch X(U ~ is of class C 1 . which, according to Proposition 6.3.5, implies that N ~ ·N ~ = 1 for all (u, v) ∈ U , we have Since N ~ ·N ~u = 0 N

and

~ ·N ~v = 0 N

(6.12)

~ v are vector ~ u and N for all (u, v) ∈ U . Therefore, both derivatives N ~ ~ ~ functions such that Nu (q) and Nv (q) are in Tp S when X(q) = p. 3 ~ The vector function N : U → R is itself a parametrized surface, with its image lying in the unit sphere, though not necessarily giving a regular parametrization of S2 . The simple fact in Equation (6.12) ~ admits a tangent plane at q ∈ U , then indicates that if N Tp S = TN~ (q) (S2 )

(6.13)

~ u, X ~ v } and as subspaces of R3 . In other words, the sets of vectors {X ~ u, N ~ v } span the same subspace Tp S. {N The differential of the Gauss map at a point p ∈ S is a linear transformation dnp : Tp S → Tn(p) (S2 ), but by virtue of Equation (6.13), one identifies it as a linear transformation dnp : Tp S → Tp S. For all ~v ∈ Tp S, one defines dnp (~v ) = w ~ if there exists a curve

6.3. The Gauss Map

197

α ~ : I → U such that  ~  α ~ (0) = q, with X(q) = p,      d ~ X ◦α ~ (t) t=0 = ~v , and dt       d N ~ ◦α ~ (t) t=0 = w. ~ dt

(6.14)

~ ~ ◦α ~ (t) = N ~ ◦α In other words, by writing X(t) =X ~ (t) and N ~ (t), the 0 0 ~ ~ differential of the Gauss map satisfies dnp (X (t)) = N (t). ~ v0 ) or X(u ~ 0 , t) for the curve α Using coordinate lines X(t, ~ (t) in the above definition, it is easy to see that ~ u) = N ~ u, dnp (X ~ v) = N ~v. dnp (X

(6.15)

Though the Gauss map n : S → S2 is independent of any coordinate systems on S, the differential dnp requires reference to a regular positively oriented parametrization of a neighborhood of p. However, though different parametrizations of a neighborhood of p induce different coordinate bases on Tp S, the definition in Equation (6.14) remains unchanged and consequently, as a linear transformation of Tp S to itself, is independent of the parametrization.

Problems 6.3.1. Describe the region of the unit sphere covered by the image of the Gauss map for the following surfaces: (a) The hyperbolic paraboloid given by z = x2 − y 2 . (b) The hyperboloid of one sheet given by x2 + y 2 − z 2 = 1. 6.3.2. Describe the region of the unit sphere covered by the image of the Gauss map for the following surfaces: (a) The cone with opening angle α given by z 2 tan2 α = x2 + y 2 . (b) The right circular cylinder x2 + y 2 = R2 , with R a constant. 6.3.3. What regular surface S has a single point on S2 as the image of the Gauss map?

198

6. The First and Second Fundamental Forms

6.3.4. Show that the graph of a differentiable function z = f (x, y) has a Gauss map that lies inside either the upper hemisphere or the lower hemisphere. 6.3.5. Prove that the image of the Gauss map of a regular surface S is an arc of a great circle if and only if S is a generalized cylinder. (See Example 6.7.2.) 6.3.6. Consider the surface S obtained as a one-sheeted cone over a regular plane curve C (see Problem 5.2.17). Prove that the image of the Gauss map for S is a curve on the unit sphere. Find a parametrization for the image of the Gauss map in terms of a parametrization of C. [Hint: Without loss of generality, suppose that C lies in the plane z = a.]

6.4 The Second Fundamental Form The examples in the previous section illustrate how the differential of the Gauss map dnp qualifies how much the surface S is curving at the point p. Despite the abstract definition for dnp , it is not difficult to calculate the matrix for dnp : Tp S → Tp S. However, in order for the differential dnp to exist at all points p ∈ S, we will need to assume that S is a surface of class C 2 . Let S be a regular oriented surface of class C 2 with orientation ~ : U → R3 be a positively oriented parametrization of a n, and let X ~ satisfies neighborhood V of a point p on S. Since N ~ ·X ~i = 0 N

for i = 1, 2,

then by differentiating with respect to another variable, the product rule gives ~j · X ~i + N ~ ·X ~ ij = 0 N for i, j = 1, 2. ~ ij = X ~ ji , by interchanging i and j, one deduces Note that since X that ~ j = −N ~ ·X ~ ij = N ~j · X ~ i. ~i · X (6.16) N (Recall that the notation f~i in Equation (6.16) and in what follows refers to taking the derivative of the multivariable vector function f~ with respect to the ith variable.) This leads to the following proposition.

6.4. The Second Fundamental Form

199

Proposition 6.4.1. Using the above setup, the linear map dnp is a selfadjoint operator with respect to the first fundamental form Ip (·, ·).

~ u + v2 X ~ v and w ~u + ~ = w1 X Proof: Let ~v , w ~ ∈ Tp S and write ~v = v1 X ~ ~ is the dot w2 Xv . Recall that the first fundamental form Ip (~v , w) product ~v · w ~ when viewing Tp S as a subset of R3 . Also recall from ~ u) = N ~ u and similarly for v. Hence, Equation (6.15) that dnp (X ~ u + v2 N ~ v , w) ~ = Ip (v1 N ~ Ip (dnp (~v ), w) ~ u + v2 N ~ v ) · (w1 X ~ u + w2 X ~ v) = (v1 N =

2 X

~j ~i · X v i wj N

i,j=1

=

2 X

~i ~j · X v i wj N

(by Equation (6.16))

i,j=1

~ u + v2 X ~ v ) · (w1 N ~ u + w2 N ~v) = (v1 X ~ = Ip (~v , dnp (w)).



Definition 6.4.2. Let S be an oriented regular surface of class C 2 with

orientation n, and let p be a point of S. We define the second fundamental form as the quadratic form on Tp S defined by IIp (~v ) = −dnp (~v ) · ~v = −Ip (dnp (~v ), ~v ). The first fundamental form allows one to measure lengths, angles, and area of regions on a parametrized surface. The second fundamental form provides a measure for how much the normal vector changes if one travels away from p in a particular direction ~v , with ~v ∈ Tp S. Recall from linear algebra that every quadratic form Q on a vector space V of dimension n is of the form Q(~v ) = ~v t M v for some n × n matrix. Define the functions Lij : U → R by   T L11 (q) L12 (q) IIp (~v ) = ~v ~v , L21 (q) L22 (q)

200

6. The First and Second Fundamental Forms

~ where p = X(q), for all ~v ∈ Tp S. Since IIp (~v ) = −dnp (~v ) · ~v = −~v T dnp (~v ),  ~ v }, one obtains ~ u, X by writing ~v = ab in the coordinate basis {X   a ~ 1 + bN ~ 2 ) · (aX ~ 1 + bX ~ 2) = −(aN IIp b ~ 1 + abN ~1 · X ~ 2 + abN ~2 · X ~ 1 + b2 N ~ 2 ). ~1 · X ~2 · X = −(a2 N Therefore, ~j · X ~i Lij = −N

for all 1 ≤ i, j ≤ 2,

(6.17)

and by Equation (6.16), ~ ·X ~ ij . Lij = N

(6.18)

This provides a convenient way to calculate the Lij functions and, hence, the second fundamental form. By Proposition 6.4.1, the matrix (Lij ) is a symmetric matrix, so L12 = L21 . In classical differential geometry texts, authors often refer to the coefficients of the second fundamental form using the letters e, f , and g, as follows: e = L11 ,

f = L12 = L21 ,

g = L22 .

Example 6.4.3 (Spheres). Let S be the sphere of radius R with outward orientation and consider the coordinate patch parametrized by ~ X(u, v) = (R cos u sin v, R sin u sin v, R cos v). The unit normal vector is ~ = (cos u sin v, sin u sin v, cos v), N

and the second derivatives are

~ 12 X

~ 11 = (−R cos u sin v, −R sin u sin v, 0), X ~ 21 = (−R sin u cos v, R cos u cos v, 0), =X ~ 22 = (−R cos u sin v, −R sin u sin v, −R cos v). X

Thus, the matrix for the second fundamental form is   −R sin2 v 0 (Lij ) = . 0 −R

6.4. The Second Fundamental Form

One should have expected the negative signs since along any curve ~ changes through a point p with direction ~v , the unit normal vector N with a differential in the direction of ~v and not opposite to it, so by Definition 6.4.2, the second fundamental form is always negative. At a point p ∈ S in a coordinate neighborhood parametrized ~ ~ positively by X(u, v), if X(q) = p, the function   s 1 1 = (L11 (q)s2 + 2L12 (q)st + L22 (q)t2 ) f (s, t) = IIp t 2 2 is called the osculating paraboloid . This paraboloid provides the second-order approximation of the surface near p in reference to the ~ (q) in the following sense. Setting q = (u0 , v0 ), the normal vector N ~ is second-order Taylor approximation of X ~ u (u0 , v0 )(u − u0 ) + X ~ v (u0 , v0 )(v − v0 ) ~ ~ 0 , v0 ) + X X(u, v) ≈X(u 1~ 2 ~ + X uu (u0 , v0 )(u − u0 ) + Xuv (u0 , v0 )(u − u0 )(v − v0 ) 2 1~ 2 + X (6.19) vv (u0 , v0 )(v − v0 ) . 2 ~ v , setting s = u − u0 and ~ is perpendicular to X ~ u and X Since N t = v − v0 ,   ~ + u0 , t + v0 ) − X(u ~ 0 , v0 ) · N ~ (u0 , v0 ). f (s, t) = X(s Furthermore, from the second derivative test in multivariable calculus, one knows that the point (s, t) = (0, 0) is a local extremum if L11 L22 − L212 > 0 and is a saddle point if L11 L22 − L212 < 0. This leads to the following definition. (See also Figure 6.7.) Definition 6.4.4. Let S be a regular orientable surface of class C 2 . A

point p on S is called: 1. elliptic if det(Lij ) > 0; 2. hyperbolic if det(Lij ) < 0; 3. parabolic if det(Lij ) = 0 but not all Lij = 0; and 4. planar if Lij = 0 for all i, j.

201

202

6. The First and Second Fundamental Forms

(a) Elliptic point.

(b) Hyperbolic point.

(c) Parabolic point.

Figure 6.7. Elliptic, hyperbolic, and parabolic points.

It is not hard to check (Problem 6.4.7) that if (x1 , x2 ) and (¯ x1 , x ¯2 ) are two coordinate systems for the same open set of an oriented surface S, then 2  ∂(x , x ) 1 2 ¯ kl ) = det(L det(Lij ). (6.20) ∂(¯ x1 , x ¯2 ) Furthermore, the change of coordinates between (x1 , x2 ) and (¯ x1 , x ¯2 ) is a diffeomorphism between two open sets in R2 . Therefore, the Jacobian in Equation (6.20) is never 0 and Definition 6.4.4 is independent of any particular parametrization of S. Example 6.4.5. Consider the torus parametrized by

~ X(u, v) = ((2 + cos v) cos u, (2 + cos v) sin u, sin v) . The unit normal vector and the second derivatives of the vector ~ are function X ~ = (− cos u cos v, − sin u cos v, − sin v), N ~ 11 = (−(2 + cos v) cos u, −(2 + cos v) sin u, 0), X ~ 12 = (sin u sin v, − cos u sin v, 0), X ~ 22 = (− cos v cos u, − cos v sin u, − sin v). X Thus, L11 = (2 + cos v) cos v,

L12 = 0,

L22 = 1.

6.4. The Second Fundamental Form

203

z

Parabolic points

bolic points Hyper

Ellip

tic points y

x Figure 6.8. Torus.

Thus, in this example it is easy to see that det(Lij ) = (2+cos v) cos v. Since −1 ≤ cos v ≤ 1, we know 2 + cos v ≥ 1, so the sign of det(Lij ) is the sign of cos v. Hence, the parabolic points on the torus are where v = ± π2 , the elliptic points are where − π2 < v < π2 , and the hyperbolic points are where π2 < v < 3π 2 (see Figure 6.8). As Figure 6.7 implies, the only quadratic surface possessing at least one planar point is a plane. However, similar to the “undecided” case in the second derivative test from multivariable calculus, on a general surface S, the existence of a planar point does not imply that S is a plane; it merely implies that third-order behavior, as opposed to second-order, governs the local geometry of the surface with respect to the normal vector. In this case, a variety of possibilities can occur. Example 6.4.6 (Monkey Saddle). The simplest surface that illustrates ~ v) = third-order behavior is the monkey saddle parametrized by X(u, 2 3 (u, v, u − 3uv ). We calculate that

~ uu = (0, 0, 6u), X ~ uv = (0, 0, −6v), X ~ vv = (0, 0, −6u), X

204

6. The First and Second Fundamental Forms

and hence, even without calculating the unit normal vector function, we deduce that Lij (0, 0) = 0 for all i, j. Consequently, (0, 0, 0) is a planar point (see Figure 6.9 for a picture). Near (0, 0, 0), the best approximating quadratic to the surface is in fact the plane z = 0, but obviously such an approximation describes the surface poorly. (The terminology “monkey saddle” comes from the shape of the surface having room for two legs and a tail.) Example 6.4.5 illustrates a general fact about the local shape of a surface encapsulated in the following proposition. Proposition 6.4.7. Let S be a regular surface of class C 2 , and let V be

a coordinate neighborhood on S. 1. If p ∈ V is an elliptic point, then there exists a neighborhood V 0 of p such that Tp S does not intersect V 0 − {p}. 2. If p ∈ V is a hyperbolic point, then Tp S intersects every deleted neighborhood V 0 of p. ~ : U → R3 be a parametrization of the coordinate Proof: Let X ~ neighborhood V and suppose that X(0, 0) = p. Consider the realvalued function on U ~ ~ ~ (0, 0). h(u, v) = (X(u, v) − X(0, 0)) · N

6.4. The Second Fundamental Form

205

~ (0, 0) is a unit vector perpendicular to Tp S, then h(u, v) Since N is the height function of signed distance between the surface S at ~ p0 = X(u, v) and the tangent plane Tp S. ~ near (0, 0) is given by EquaThe second-order Taylor series of X ~ tion (6.19), with a remainder function R(u, v) such that ~ R(u, v) ~ = 0. 2 u + v2 (u,v)→(0,0) lim

Thus 1~ 2 ~ ~ ~ h(u, v) = X uu (0, 0) · N (0, 0)u + Xuv (0, 0) · N (0, 0)uv 2 1~ 2 ~ ~ ~ + X vv (0, 0) · N (0, 0)v + R(u, v) · N (0, 0) 2 1 = (L11 (0, 0)u2 + 2L12 (0, 0)uv + L22 (0, 0)v 2 ) 2 ~ ~ (0, 0) + R(u, v) · N   u 1 ~ + R(u, v) · n(p). = IIp v 2 Solving the quadratic equation L11 u2 + 2L12 uv + L22 v 2 = 0

(6.21)

for u in terms of v leads to   p L11 u = −L12 ± (L12 )2 − L11 L22 v. If we assume that (u, v) 6= (0, 0), Equation (6.21) has no solutions if det(Lij ) > 0 and  has solutions if det(Lij ) < 0. Hence, if p is hyperbolic, IIp ( uv ) changes sign in a neighborhood of (0, 0), and if p is elliptic, it does not. ¯ 2 by ¯ 1 and R Define functions R ¯ i (u, v) = Ri (u, v) R u2 + v 2

for i = 1, 2.

Then solving h(u, v) = 0 amounts to solving ¯ 1 (u, v))u2 + 2L12 uv + (L22 + R ¯ 2 (u, v))v 2 = 0. (L11 + R

206

6. The First and Second Fundamental Forms

¯ 1 and R ¯ 2 have a limit of 0 as (u, v) approaches (0, 0), Since both R lim

¯ 1 (u, v))(L22 (0, 0) + R ¯ 2 (u, v)) − L12 (0, 0)2 (L11 (0, 0) + R

(u,v)→(0,0)

= L11 (0, 0)L22 (0, 0) − L12 (0, 0)2 Thus, if p is an elliptic point, there is a neighborhood of (0, 0) in which h(u, v) = 0 does not have solutions except for (u, v) = (0, 0), and if p is hyperbolic, every neighborhood of (0, 0) has points through which h(u, v) changes sign.  Let us return now to considering the differential of the Gauss map. With Equation (6.18), it is now possible to explicitly calculate the matrix for dnp : Tp S → Tp S in terms of the oriented basis ~ v ), where X ~ : U → R3 is a positively oriented parametrization ~ u, X (X ~ u and N ~ v lie in Tp S, there exist of a neighborhood around p. Since N i functions aj (u, v) defined on U such that 2~ ~ ~ u = a1 X N 1 u + a1 X v , 1 2 ~ ~ ~v = a X N 2 u + a2 X v .

(6.22)

Using numerical indices to represent corresponding derivatives, Equation (6.22) can be written as ~j = N

2 X

~ i, aij X

(6.23)

i=1

~ u and X ~2 = X ~ v . (It is important to remem~1 = X where we denote X i ber that the superscripts in aj also correspond to indices and not to powers. This notation, though perhaps awkward at first, is the standard notation for components of a tensor. It is useful to remember that the superscript represents a row index while the subscript is a column index.  1 ~ For any vector w ~ ∈ Tp S with coordinates w ~ = w w2 = w1 Xu + ~ v , the differential of the Gauss map satisfies w2 X ~ 1 + w2 N ~2 ~ = w1 N dnp (w) ~ 1 + (a21 w1 + a22 w2 )X ~2 = (a11 w1 + a12 w2 )X  1 1   w1 a1 a2 . = a21 a22 w2

6.4. The Second Fundamental Form

207

However, from Equations (6.16), (6.18), and (6.23), ! 2 X k ~ j, ~i · X ~j = ~k · X ai X −Lij = N k=1

and so −Lij =

2 X

aki gkj =

k=1

2 X

gjk aki .

(6.24)

k=1

In matrix notation, Equation (6.24) means that     1 1  g11 g12 a1 a2 L11 L12 = . − L21 L22 g21 g22 a21 a22

(6.25)

Since the metric tensor is a positive definite matrix at any point on a regular surface, one can multiply both sides of Equation (6.25) by the inverse of (gij ), and conclude that the matrix for dnp given in ~ v } is ~ u, X terms of the basis {X  −1    1 1 g11 g12 L11 L12 a 1 a2 =− . (6.26) a21 a22 g21 g22 L21 L22 This matrix formula is more common in modern texts but classical differential geometry texts refer to the individual component equations implicit in the above formula as the Weingarten equations. Note that the (gij ) and (Lij ) matrices are symmetric but that the (aij ) matrix need not be symmetric. Since all the above matrices are 2 × 2, det(Lij ) = det(−Lij ) and since det(gij ) > 0, the determinant det(Lij ) has the same sign as det(aij ). Therefore, one deduces the following reformulation of Definition 6.4.4. Proposition 6.4.8. Let S be an oriented regular surface of class C 2

with orientation n. Then, a point p ∈ S is called 1. elliptic if det(dnp ) > 0; 2. hyperbolic if det(dnp ) < 0; 3. parabolic if det(dnp ) = 0 but dnp 6= 0; 4. planar if dnp = 0.

208

6. The First and Second Fundamental Forms

Problems 6.4.1. Calculate the second fundamental form (i.e., the matrix of functions ~ (Lij )) for the ellipsoid X(u, v) = (a cos u sin v, b sin u sin v, c cos v). 6.4.2. Calculate the second fundamental form (i.e., the matrix of functions ~ v) = (au, bv, uv). (Lij )) for the parabolic hyperboloid X(u, 6.4.3. Calculate the second fundamental form (i.e., the matrix of functions ~ (Lij )) for the catenoid X(u, v) = (cosh v cos u, cosh v sin u, v). 6.4.4. Calculate the second fundamental form (i.e., the matrix of functions (Lij )) using the following parametrizations of the right cylinder: ~ (a) X(u, v) = (cos u, sin u, v). ~ (u, v) = (cos(u + v), sin(u + v), v). (b) Y ~ , find all vectors ~v ∈ Tp S such that Using the parametrization Y IIp (~v ) = 0. Show that these correspond to the straight lines on the cylinder. 6.4.5. Consider Enneper’s surface parametrized by   u3 v3 2 2 2 2 ~ X(u, v) = u − + uv , v − + vu , u − v . 3 3 Show that (a) the coefficients of the first fundamental form are g11 = g22 = (1 + u2 + v 2 )2 ,

g12 = 0;

(b) the coefficients of the second fundamental form are L11 = 2,

L12 = 0,

L22 = −2.

6.4.6. Suppose that an open set V of a regular oriented surface has two x1 , x ¯2 ) and suppose that V is systems of coordinates (x1 , x2 ) and (¯ ~ 1 , x2 ) in terms of (x1 , x2 ) ∈ U and by Y ~ (¯ parametrized by X(x x1 , x ¯2 ) ¯2 ). Call Lij the terms of the second fundamental in terms of (¯ x1 , x ¯ ij the terms of form in terms of the (x1 , x2 ) coordinates and call L ¯2 ) coordinates. the second fundamental form in terms of the (¯ x1 , x Prove that 2 X ∂xi ∂xj ¯ kl = Lij . L ∂ x ¯k ∂ x ¯l i,j=1

6.5. Normal and Principal Curvatures

209

6.4.7. Use the previous exercise to show that under the same conditions ¯ kl ) = det(L



∂(x1 , x2 ) ∂(¯ x1 , x ¯2 )

2 det(Lij ).

6.4.8. Suppose we are under the same conditions as in Problem 6.4.6. Call [dF ] the 2 × 2 matrix of the differential of the coordinate change, that is   ∂xi . [dF ] = ∂x ¯j i,j=1,2 Call a ¯ji the coefficients of dnp in terms of the coordinate system ¯2 ). Prove that, as matrices, (¯ x1 , x (¯ aji ) = (dF ) (alk ) (dF )−1 . 6.4.9. Calculate the second fundamental form (Lij ) and the matrix for ~ dnp for function graphs X(u, v) = (u, v, f (u, v)). Determine which points are elliptic, hyperbolic, or parabolic. (This exercise coupled with Proposition 6.4.8 is a proof of the second derivative test from multivariable calculus.) 6.4.10. Prove that all points on the tangential surface of a regular space curve are parabolic. 6.4.11. Give an example of a surface with an isolated parabolic point. 6.4.12. Consider a regular surface S given by the equation F (x, y, z) = 0. Use implicit differentiation to provide a criterion for determining whether points are elliptic, hyperbolic, parabolic, or planar. [Hint: Assume that z can be expressed as a function of x and y. Then by implicit differentiation, Fx ∂z =− ∂x Fz

and

Fy ∂z =− . ∂y Fz

~ Then using the parametrization of S by X(x, y) = (x, y, z(x, y)) and ~ and then also these implicit derivatives, it is possible to calculate N the Lij matrix.]

6.5 Normal and Principal Curvatures One way to analyze the shape of a regular surface S near a point p is to consider curves on S through p and analyze the normal component

210

6. The First and Second Fundamental Forms

of their principal curvature vector. In order to use the techniques of differential geometry, in this section we will always assume that the surface S is of class C 2 and that the curves on S are regular curves also of class C 2 . ~ : Definition 6.5.1. Let S be a regular surface of class C 2 , and let X U → R3 be a parametrization of a coordinate neighborhood V of S. Let ~γ : I → R3 be a parametrization of class C 2 for a curve C that lies on S in V . The normal curvature of S along C is the function κn (t) =

1 ~0 ~ ~ ) = κ cos θ, T · N = κ(P~ · N s0

where θ is the angle between the principal normal vector P~ of the ~ of the surface. curve and the normal vector N Interestingly enough, though the curvature of a space curve in general depends on the second derivative of the curve, once one has the second fundamental form for a surface, the normal curvature at a point depends only on the direction. Proposition 6.5.2. Let p be a point on S in the neighborhood V and

suppose that ~γ is the parametrization of a curve C on S such that ~γ (0) = p and T~ (0) = w ~ ∈ Tp S. Then at p, we write κn (0) = IIp (w). ~ ~ ◦α Proof: Suppose that ~γ = X ~ , where α ~ (t) = (u(t), v(t)) is a curve in the domain U . The normal vector of S along the curve is given ~ (t) = N ~ (~ ~ (t) = 0 for all t ∈ I, by the function N α(t)). Since T~ (t) · N then ~ = −T~ · N ~ 0, T~ 0 · N and, hence, 1 ~ ~0 T ·N . s0 (t) ~ (t) = N ~ (u(t), v(t)), then N ~ 0 (t) = N ~ u u0 (t) + N ~ v v 0 (t), and Since N 0 (0) u ~ 0 (0) = dnp 0 therefore, N ~ Thus, at the point p, = dnp (s0 (0)w). v (0) the normal curvature of S along C is κn (t) = −

κn = −

1 ~ T (0) · dnp (s0 (0)w) ~ = −w ~ · dnp (w) ~ = IIp (w). ~

s0 (0)



6.5. Normal and Principal Curvatures

211

Corollary 6.5.3. All curves C on S passing through the point p with direction w ~ ∈ Tp S have the same normal curvature at p.

Because of Corollary 6.5.3, given a point p on a regular surface S, one knows everything about the local change in the Gauss map by knowing the values of the second fundamental form on the unit circle in Tp S around p. In particular, most interesting are the optimal values of IIp (w) ~ for kwk ~ = 1 in Tp S. To find these optimal values,  ~ u, X ~ v } and optimize set w ~ = ab in the coordinate basis {X IIp (w) ~ = L11 a2 + 2L12 ab + L22 b2 , with variables a, b subject to the constraint Ip (w, ~ w) ~ = g11 a2 + 2g12 ab + g22 b2 = 1. Using Lagrange multipliers, one finds that there exists λ such that the optimization problem is solved when ( L11 a + L12 b = λ(g11 a + g12 b), L21 a + L22 b = λ(g21 a + g22 b), or, in matrix form,       a a g11 g12 L11 L12 . =λ b b L21 L22 g21 g22

(6.27)

This remark leads to the following fundamental proposition. Proposition 6.5.4. The maximum and minimum values κ1 and κ2 of

IIp (w) ~ restricted to the unit circle are the negatives of the eigenvalues of dnp . Furthermore, there exists an orthonormal basis {~e1 , ~e2 } of Tp S such that dnp (~e1 ) = −κ1~e1 and dnp (~e2 ) = −κ2~e2 . Proof: By multiplying on the left by the inverse g −1 of the metric tensor, Equation (6.27) becomes     a a . =λ g −1 L b b   Since the matrix of the differential of the Gauss map is dnp = −g −1 L, the first part of the proposition follows.

212

6. The First and Second Fundamental Forms

Since the first fundamental form Ip (·, ·) is positive definite and since, by Proposition 6.4.1, dnp is self-adjoint with respect to this form, then by the Spectral Theorem, dnp is diagonalizable. (Many linear algebra textbooks discuss the Spectral Theorem exclusively using the dot product as the positive definite form. See [21, Section 7, Chapter XV] for a more general presentation of the Spectral Theorem, which we use here.) Now, since dnp is self-adjoint with respect to the first fundamental form, Ip (dnp (~e1 ), ~e2 ) = Ip (−κ1~e1 , ~e2 ) = −κ1 Ip (~e1 , ~e2 ), k Ip (~e1 , dnp (~e2 )) = Ip (~e1 , −κ2~e2 ) = −κ2 Ip (~e1 , ~e2 ), and thus, (κ1 − κ2 )Ip (~e1 , ~e2 ) = 0. Hence, if κ1 6= κ2 , then Ip (~e1 , ~e2 ) = ~e1 · ~e2 = 0, where in the latter dot product we view ~e1 and ~e2 as vectors in R3 , and if κ1 = κ2 , then by the Spectral Theorem, any orthonormal basis satisfies the claim of the proposition.  Definition 6.5.5. Let S be a regular surface, and let p be a point on

S. The maximum and minimum normal curvatures κ1 and κ2 at p are called the principal curvatures of S at p. The corresponding directions, i.e., unit eigenvectors ~e1 and ~e2 with dnp (~ei ) = −κi~ei , are called principal directions at p. In a plane, the second fundamental form is identically 0, all normal curvatures including the principal curvatures are 0, and hence, all directions at all points are principal directions. Similarly, it is not hard to show that at every point of the sphere the normal curvature in every direction is the same, and hence, all directions are principal. Example 6.5.6 (Ellipsoids). As a contrast to the plane or sphere, con-

sider the ellipsoid parametrized by ~ X(u, v) = (a cos u sin v, b sin u sin v, c cos v). It turns out that the formulas for κ1 and κ2 as functions of (u, v) are quite long so we shall calculate the principal curvatures at the point

6.5. Normal and Principal Curvatures

213

 corresponding to (u, v) = π3 , π6 . It is not too hard to calculate the coefficients of the first and second fundamental forms abc sin2 v , L11 = √ det g = (b2 − a2 ) sin u cos u sin v cos v, L12 = 0, abc . = a2 cos2 u cos2 v + b2 sin2 u cos2 v + c2 sin2 v, L22 = √ det g

g11 = a2 sin2 u sin2 v + b2 cos2 u sin2 v, g12 g22

At (u, v) = (aji ) =

π π 3, 6

  8abc −3a2 − 9b2 − 4c2 −12a2 + 12b2 , −3a2 + 3b2 −12a2 − 4b2 (12a2 b2 + 3a2 c2 + b2 c2 )3/2

and, after some algebra simplifications, the eigenvalues of this matrix are   p 4abc −15a2 − 13b2 − 4c2 ± (9a2 − 5b2 − 4c2 )2 + 36(a2 − b2 )2 λ= . (12a2 b2 + 3a2 c2 + b2 c2 )3/2 From this we see that at this particular point for (u, v), the eigenvalues are equal if and only if a2 − b2 = 0 and 9a2 − 5b2 − 4c2 = 0, which is equivalent to a2 = b2 = c2 . ~ If a curve on the surface given by γ(t) = X(u(t), v(t)) is such 0 that at γ(t0 ) its direction γ (t0 ) is a principal direction with principal curvature κi , then by Equation (6.27), with λ = −κi , we have ( L11 u0 + L12 v 0 = −κi (g11 u0 + g12 v 0 ), L21 u0 + L22 v 0 = −κi (g21 u0 + g22 v 0 ). Eliminating κi from these two equations leads to the relationship (L11 g21 − L21 g11 )(u0 )2 + (L11 g22 − L22 g11 )u0 v 0 + (L12 g22 − L22 g12 )(v 0 )2 = 0, or equivalently, 0 2 (v ) −u0 v 0 (u0 )2 = 0. g11 g g 12 22 L11 L12 L22

(6.28)

214

6. The First and Second Fundamental Forms

Also, in light of the Weingarten equations in Equation (6.26), we can summarize the above two formulas by  0  0  u v · dnp 0 = 0. 0 −u v Definition 6.5.7. Any regular curve on a regular oriented surface given

~ by γ(t) = X(u(t), v(t)) in a coordinate neighborhood parametrized 3 ~ by X : U → R that satisfies Equation (6.28) for all t is called a line of curvature. The lines of curvature on a surface form an orthogonal family of curves on the surface, that is, two sets of curves intersecting at right angles. With an appropriate change of variables and in an interval 0 dv of t where u0 (t) 6= 0, using the chain rule du = uv 0(t) (t) , (6.28) can be rewritten as  2 dv dv +(L11 g22 −L22 g11 ) +(L11 g21 −L21 g11 ) = 0. (L12 g22 −L22 g12 ) du du (6.29) 0 If in the neighborhood we are studying u (t) = 0 for some t, then since we assume the curve is regular, v 0 (t) cannot also be 0, and hence, we can rewrite Equation (6.28) with u as a function of v. In general, solving the above differential equation is an intractable problem. Of course, as long as the coefficients of the first and second fundamental form are not proportional to each other, one can easily dv solve algebraically for du in Equation (6.29) and obtain two distinct dv solutions of du as a function of u and v. Then according to the theory of differential equations (see [2, Section 9.2] for a reference), dv given any point (u0 , v0 ) and for each of the two solutions of du as a function of u and v, there exists a unique function v = f (u) solving Equation (6.29). Consequently, a regular oriented surface of class C 2 can be covered by lines of curvature wherever the coefficients of the first and second fundamental form are continuous and are not proportional to each other. Points where this cannot be done have a special name. Definition 6.5.8. Let S be a regular oriented surface of class C 2 . An

umbilical point is a point p on S such that, given a parametriza~ of a neighborhood of p, the corresponding first and second tion X

6.5. Normal and Principal Curvatures

fundamental forms have coefficients that are proportional, namely, L11 g12 − L12 g11 = 0, L11 g22 − L22 g11 = 0, and L22 g12 − L12 g22 = 0. Proposition 6.5.9. Let S be a regular oriented surface. A point p on S

is an umbilical point if and only if the eigenvalues of dnp are equal. ~ be a parametrization of a neighborhood of p. With Proof: Let X ~ ~ respect to this parametrization  and  thej associated basis {Xu , Xv } on Tp S, the matrix of dnp is dnp = (ai ). Proposition 6.5.4 implies that (aji ) is diagonalizable. The matrix (aji ) has equal eigenvalues if and only if there exists an invertible matrix B such that   λ 0 j B −1 , (ai ) = B 0 λ and then (aji ) = B(λI)B −1 = λBB −1 = λI. Consequently (aji ) has equal eigenvalues if and only if it is already diagonal, with elements on the diagonal being equal. Then λI = (aji ) = −g −1 L, and therefore, L = −λg, which is tantamount to saying that the coefficients of the first and second fundamental forms are proportional. The proposition follows.  The proof of Proposition 6.5.4 shows that if κ1 6= κ2 , then the orthonormal basis of principal directions is unique up to signs, while if κ1 = κ2 , any orthogonal basis is principal. Therefore, in light of Proposition 6.5.9, at umbilical points, there is no preferred basis of principal directions, and thus it makes sense that the only solutions to Equation (6.28) at umbilical points have u0 (t) = v 0 (t) = 0, that is, (u(t), v(t)) = (u0 , v0 ). Note that at an umbilical point, Equation (6.28) degenerates to the trivial equation 0 = 0. Solving Equation (6.28) gives the lines of curvature on a surface. At every point p on the surface that is not umbilical, there are two lines of curvature through p, and they intersect at right angles. This simple remark leads to the following nice characterization. ~ of a Proposition 6.5.10. The coordinate lines of a parametrization X surface are curvature lines if and only if g12 = L12 = 0.

215

216

6. The First and Second Fundamental Forms

Proof: Suppose that coordinate lines are curvature lines. Since any two lines of curvature intersect at a point at a right angle, we know ~u · X ~ v = 0. Furthermore, coordinate lines are given by that g12 = X ~ ~ 0 , t). If ~γ (t) = X(t, ~ v0 ), then u0 (t) = 1 ~γ (t) = X(t, v0 ) or ~γ (t) = X(u 0 ~ 0 , t), then u0 (t) = 0 and v 0 (t) = 1. and v (t) = 0, and if ~γ (t) = X(u Then Equation (6.28) implies that both of the following hold: L11 g21 − L21 g11 = 0

and

L12 g22 − L22 g12 = 0.

Since g12 = 0, we have L21 g11 = 0 and L12 g22 = 0. However, since 2 > 0, then at all points on S the functions g det(g) = g11 g22 − g12 11 and g22 cannot both be 0 at the same time. Since L12 = L21 , we deduce that L12 = 0. Conversely, if g12 = L12 = 0, then (gij ) and (Lij ) are both di~ u and agonal matrices, making (aij ) a diagonal matrix, and hence, X ~ v are eigenvectors of dnp . Hence, coordinate lines are curvature X lines.  As a linear transformation from Tp S to itself, dnp is independent of a parametrization of a neighborhood of p. Therefore, if p is not an umbilical point, the principal directions {~e1 , ~e2 } as eigenvectors of dnp provide a basis of Tp S that possesses more geometric meaning than the coordinate basis of any particular parametrization of a neighborhood of p. In the orthonormal basis {~e1 , ~e2 }, a unit vector w ~ ∈ Tp S is written as w ~ = cos θ~e1 + sin θ~e2 for some angle θ. Using these coordinates, the normal curvature of S at p in the direction of w ~ is IIp (w) ~ = −w ~ · dnp (w) ~ = −(cos θ~e1 + sin θ~e2 ) · dnp (cos θ~e1 + sin θ~e2 ) = −(cos θ~e1 + sin θ~e2 ) · (cos θdnp (~e1 ) + sin θdnp (~e2 )) = −(cos θ~e1 + sin θ~e2 ) · (− cos θκ1~e1 − sin θκ2~e2 ) ~ = (cos2 θ)κ1 + (sin2 θ)κ2 . IIp (w)

(6.30)

Equation (6.30) is called Euler’s curvature formula. Another useful geometric characterization of the behavior of S near p is called the Dupin indicatrix . The Dupin indicatrix consists of all vectors w ~ ∈ Tp S such that IIp (w) ~ = ±1. If w ~ = (w1 , w2 ) =

6.5. Normal and Principal Curvatures

217

(ρ cos θ, ρ sin θ) are expressions of w ~ in Cartesian and polar coordinates referenced in terms of the orthonormal frame {~e1 , ~e2 }, then Euler’s curvature formula gives ±1 = IIp (w) ~ = ρ2 IIp (w) ~ = κ1 ρ2 cos2 θ + κ2 ρ2 sin2 θ. Thus, the Dupin indicatrix satisfies the equation κ1 w12 + κ2 w22 = ±1.

(6.31)

Since the principal curvatures at a point p are the negatives of the eigenvalues of dnp , Proposition 6.4.8 provides a characterization of whether a point is elliptic, hyperbolic, parabolic, or planar in terms of the principal curvatures. More precisely, a point p ∈ S is 1. elliptic if κ1 and κ2 have the same sign; 2. hyperbolic if κ1 and κ2 have opposite signs; 3. parabolic if exactly one of κ1 and κ2 is 0; 4. planar if κ1 = κ2 = 0. The Dupin indicatrix justifies this terminology because p is elliptic or hyperbolic if and only if Equation (6.31) is the equation for a single ellipse or two hyperbolas, respectively. Furthermore, the half-axes for the corresponding ellipse or hyperbola are s s 1 1 and . |κ1 | |κ2 | See Figure 6.10 for an illustration of the Dupin indicatrix in the elliptic and hyperbolic cases. Also observe that this figure has |κ1 | > |κ2 |. If a point is parabolic, then the Dupin indicatrix is simply a pair of parallel lines equidistant from one of the principal direction lines, and if a point is planar, the Dupin indicatrix is the empty set. When a point is hyperbolic, it is possible to find the asymptotes of the Dupin indicatrix without referring directly to the principal curvatures.

218

6. The First and Second Fundamental Forms

~e2

~e2 ρ θ

p

θ

~e1

(a) Elliptic point.

p

~e1

(b) Hyperbolic point.

Figure 6.10. The Dupin indicatrix.

Definition 6.5.11. Let p be a point on a regular oriented surface S of class C 2 . An asymptotic direction of S at p is a unit vector w ~ in Tp S such that the normal curvature is 0. An asymptotic curve on S is a regular curve C such that at every point p ∈ C, the unit tangent vector at p is an asymptotic direction.

Since the principal curvatures at a point p are the maximum and minimum normal curvatures, at an elliptic point, there exist no asymptotic directions, which one can see from the Dupin indicatrix. If, on the other hand, p is hyperbolic, then the principal curvatures have opposite signs. If a unit vector w ~ is an asymptotic direction at p and makes an angle θ with ~e1 , then by Euler’s formula we deduce that κ2 cos2 θ = − . κ1 − κ2 The right-hand side is always positive and less than 1, so this equation leads to four solutions for w, ~ i.e., two mutually negative pairs, each representing one of the asymptotes of the Dupin indicatrix hyperbola. As we see in Problem 6.5.11, the asymptotic curves in fact provide a more subtle description of the behavior of a surface near a point than the Dupin indicatrix does since it describes more than second-order phenomena. As with the lines of curvature, it is not difficult to find a differential equation that characterizes asymptotic curves on a surface. ~ : U → R3 parametrizes a neighborhood of p ∈ S Suppose that X ~ ◦α and that ~γ : I → S is a curve on S defined by ~γ = X ~ , with

6.5. Normal and Principal Curvatures

219

α ~ : I → U and α ~ (t) = (u(t), v(t)). By Proposition 6.5.2, if we call κn (t) the normal curvature of S at ~γ (t) in the direction of ~γ 0 (t), then  0    1 u ~ . II~γ (t) κn (t) = II~γ (t) T (t) = 0 2 v0 (s (t)) Thus, since an asymptotic curve must satisfy κn (t) = 0 for all t, then any asymptotic curve must satisfy the differential equation L11 (u0 )2 + 2L12 u0 v 0 + L22 (v 0 )2 = 0. Example 6.5.12 (Surfaces of Revolution). Consider a surface of revolu-

tion parametrized by ~ X(u, v) = (f (v) cos u, f (v) sin u, h(v)), with f (v) > 0, u ∈ (0, 2π), and v ∈ (a, b). It is not hard to show (see Problem 6.6.11) that the coefficients of the first fundamental form are g11 = f (v)2 ,

g12 = g21 = 0,

g22 = (f 0 (v))2 + (h0 (v))2 ,

and the coefficients of the second fundamental form are f h0 f 00 h0 − f 0 h00 L11 = − p , L12 = L21 = 0, L22 = p . (f 0 )2 + (h0 )2 (f 0 )2 + (h0 )2 By Problem 6.5.6, we deduce that the meridians (where u = const.) and the parallels (where v = const.) are lines of curvature. We can also conclude that the principal curvatures are κ1 (u, v) = −

h0 (v) f (v)

p

(f 0 (v))2 + (h0 (v))2

,

f 00 (v)h0 (v) − f 0 (v)h00 (v) , κ2 (u, v) = (f 0 (v))2 + (h0 (v))3/2

(6.32)

though, as written, one can make no assumption that κ1 > κ2 . Obviously, the first and second fundamental forms depend only on the coordinate v, and hence, all properties of points, such as whether they are elliptic, hyperbolic, parabolic, planar, or umbilical, depend only on v. Setting κ1 = κ2 in Equation (6.32) produces an equation that determines for what v the points on the surface are umbilical points.

220

6. The First and Second Fundamental Forms

Problems 6.5.1. Find the principal directions and the principal curvatures of the quadric surface z = ax2 + 2bxy + cy 2 at (0, 0, 0) in terms of the constants a, b, c. 6.5.2. Find the principal directions and the principal curvatures of the surface z = 4x2 − x4 − y 2 at its critical points (as a function in two variables). 6.5.3. Provide the details for Example 6.5.6. 6.5.4. Consider the ellipsoid with half-axes a, b, and c. (a) Prove the ellipsoid has four umbilical points when a, b, and c are distinct. (b) Calculate the coordinates of the umbilical points when all halfaxes have different length. (c) What happens when two of the half-axes are equal? 6.5.5. (ODE) Determine the asymptotic curves and the lines of curvature of the helicoid ~ X(u, v) = (v cos u, v sin u, cu). ~ be the parametrization for a neighborhood of S. Prove that if 6.5.6. Let X (gij ) and (Lij ) are diagonal matrices, then the lines of curvature are the coordinate curves (curves on S where u =const. or v =const.). 6.5.7. Let the L22

~ be the parametrization for a neighborhood of S. Prove that X coordinate curves are asymptotic curves if and only if L11 = = 0.

6.5.8. (ODE) Determine the asymptotic curves of the catenoid ~ X(u, v) = (cosh v cos u, cosh v sin u, v). 6.5.9. Consider Enneper’s surface. Use the results of Problem 6.4.5 to show the following: (a) The principal curvatures are κ1 =

2 (1 +

u2

+

v 2 )2

,

κ2 = −

2 (1 +

u2

+ v 2 )2

.

(b) The lines of curvature are the coordinate curves. (c) The asymptotic curves are u + v =const. and u − v =const.

6.6. Gaussian and Mean Curvatures

221

6.5.10. Find equations for the lines of curvature when the surface is given by z = f (x, y). 6.5.11. Consider the monkey saddle given by the graph of the function z = x3 − 3xy 2 . Prove that the set of asymptotic curves that possesses (0, 0, 0) as a limit point consists of three straight lines through (0, 0, 0) with equal angles between them. 6.5.12. Let S1 and S2 be two regular surfaces that intersect along a regular curve C. Let p be a point on C, and call λ1 and λ2 the normal curvatures of S at p in the direction of C. Prove that the curvature κ of C at p satisfies κ2 sin2 θ = λ21 + λ22 − 2λ1 λ2 cos2 θ, where θ is the angle between S1 and S2 at p (calculated using the normals to S1 and S2 at p). 6.5.13. Let S be a regular oriented surface and p a point on S. Two nonzero vectors ~u1 , ~u2 ∈ Tp S are called conjugate if Ip (dnp (~u1 ), ~u2 ) = Ip (~u1 , dnp (~u2 )) = 0.

(6.33)

Prove the following: (a) A curve C on S parametrized by ~γ : I → S is a line of curvature if and only if the unit tangent vector T~ and any normal vector to T~ in T~γ (t) (S) are conjugate to each other. (b) Let ~γ1 (t) and ~γ2 (t) be regular space curves, and define a sur~ face S by the parametrization X(u, v) = ~γ1 (u) + ~γ2 (v). Surfaces constructed in this manner are called translation surfaces. Show that the coordinate lines of S are conjugate lines.

6.6 Gaussian and Mean Curvatures We now arrive at two fundamental geometric invariants that encapsulate a considerable amount of useful information about the local shape of a surface. Again, in this section, we must assume that S is a regular surface of class C 2 . Definition 6.6.1. Let κ1 and κ2 be the principal curvatures of a regular

oriented surface S at a point p. Define 1. the Gaussian curvature of S at p as the product K = κ1 κ2 ;

222

6. The First and Second Fundamental Forms

2. the mean curvature of S at p as the average H = principal curvatures.

κ1 +κ2 2

of the

By Problem 6.4.8, the matrix for the Gauss map in terms of any particular coordinate system on a neighborhood of p on S is conjugate to the corresponding matrix  for a different coordinate system via the change of basis matrix ∂∂xx¯ji . Therefore, since det(BAB −1 ) = det(A) and Tr(BAB −1 ) = Tr(A), which is a standard result in linear algebra, the eigenvalues of the Gauss map, the principal curvatures, the Gaussian curvature, and the mean curvature are invariant under coordinate changes. Furthermore, since det(−A) = det(A) when A is a square matrix with an even number of rows, then K = κ1 κ2 = det(aji ) = det(dnp ), H=

κ1 + κ2 1 1 = − Tr(aji ) = − Tr(dnp ). 2 2 2

Consequently, as claimed above, the Gaussian curvature and the mean curvature (up to a sign) of S at p are geometric invariants in that they do not depend on the orientation or position of S in space, and they do not depend on any particular coordinate system on S in a neighborhood of the point p. In order to calculate the mean curvature H(u, v), one has no choice but to calculate the matrix of the differential of the Gauss map (aij ) = [dnp ]. However, from a computational perspective, Equation (6.26) leads to a much simpler formula for the Gaussian curvature function K(u, v). Recall that det(AB) = det(A) det(B) for all square matrices of the same size and also that det(A−1 ) = 1/ det(A). Then Equation (6.26) implies that K=

det(Lij ) L11 L22 − L212 = 2 . det(gij ) g11 g22 − g12

(6.34)

This equation lends itself readily to calculations since one does not need to fully compute the matrix of the Gauss map, let alone find its eigenvalues. However, we can also obtain an alternate characterization of the Gaussian curvature.

6.6. Gaussian and Mean Curvatures

223

Proposition 6.6.2. Let S be a regular oriented surface of class C 2 , and

~ : U ⊂ R2 → R3 . Define let V be a neighborhood parametrized by X ~ as the unit normal vector N ~ ~ ~ = Xu × Xv . N ~u × X ~ vk kX Then over the domain U , the Gaussian curvature is the unique function K(u, v) satisfying ~ v = K(u, v)X ~u × X ~ v. ~u × N N Proof: From the definition of the (aij ) functions in Equation (6.22), we have ~u × N ~ v = (a11 X ~ u + a21 X ~ v ) × (a12 X ~ u + a22 X ~ v ). N It is then easy to see that ~u × N ~ v = det(aij )X ~u × X ~ v. N The result follows since K = det(aij ).



Corollary 6.6.3. If V is a region of a regular oriented surface S of

class C 2 , then ZZ

ZZ K dS =

V

dS, n(V )

where the latter integral is the signed area on the unit sphere of the image of V under the Gauss map. Example 6.6.4 (Spheres). Consider the sphere parametrized by the vec-

tor function ~ X(u, v) = (R cos u sin v, R sin u sin v, R cos v), with (u, v) ∈ (0, 2π) × (0, π). Example 6.1.5 and Example 6.4.3 gave us   2 2   R sin v 0 −R sin2 v 0 (gij ) = and (Lij ) = . 0 R2 0 −R

224

6. The First and Second Fundamental Forms

Using Equation (6.34), it is then easy to see that the Gaussian curvature function on the sphere is K(u, v) =

R2 sin2 v 1 = 2, 2 4 R R sin v

which is a constant function. (Note that we assumed that v 6= 0, π, which correspond to the north and south poles of the parametrization. However, another parametrization that includes the north and south poles would show that at these points as well we have K = 1/R2 .) The matrix of the Gauss map is then   1   −R 0 −1 , dnp = −g L = 0 − R1 from which one immediately deduces that the principal curvatures are κ1 (u, v) = κ2 (u, v) = R1 . This shows that all points on the sphere are umbilical points. In addition, the mean curvature is also a constant function H(u, v) = R1 . Example 6.6.5 (Function Graphs). Consider the graph of a function f :

~ : U → R3 , with U ⊂ R2 → R. The graph can be parametrized by X ~ X(u, v) = (u, v, f (u, v)). Problem 6.4.9 asks the reader to calculate the matrix (Lij ). It is not hard to show that   1 + (fu )2 fu fv (gij ) = and fv fu 1 + (fv )2   1 fuu fuv (Lij ) = p , 1 + fu2 + fv2 fvu fvv where fu is the typical shorthand to mean ∂f ∂u . But then det(g) = 2 2 1 + fu + fv , and we find that the Gaussian curvature function on a function graph is 2 fuu fuv 1 = fuu fvv − fuv . K(u, v) = (1 + fu2 + fv2 )2 fvu fvv (1 + fu2 + fv2 )2

6.6. Gaussian and Mean Curvatures

This result allows us to rephrase the second derivative test in the calculus of a function f from R2 to R as follows: If f has continuous second partial derivatives and (u0 , v0 ) is a critical point (i.e., fu (u0 , v0 ) = fv (u0 , v0 ) = 0), then 1. (u0 , v0 ) is a local maximum if K(u0 , v0 ) > 0 and fuu (u0 , v0 ) < 0; 2. (u0 , v0 ) is a local minimum if K(u0 , v0 ) > 0 and fuu (u0 , v0 ) > 0; 3. (u0 , v0 ) is a saddle point if K(u0 , v0 ) < 0; 4. the test is inconclusive if K(u0 , v0 ) = 0. In the language we have introduced in this chapter, local minima and local maxima of the function z = f (u, v) are elliptic points, saddle points are hyperbolic points, and points where the second derivative test is inconclusive are either parabolic or planar points. Example 6.6.6 (Pseudosphere). A tractrix is a curve in the plane with

parametric equations α ~ (t) = (sech t, t − tanh t), and it has the y-axis as an asymptote. The pseudosphere is defined as half of the surface of revolution of a tractrix about its asymptote. More precisely, we can parametrize the pseudosphere by ~ X(u, v) = (sech v cos u, sech v sin u, v − tanh v), with u ∈ [0, 2π) and v ∈ [0, ∞). (See Figure 6.11 for a picture of the tractrix and the pseudosphere.) We leave it as an exercise for the reader to prove that   sech2 v 0 (gij ) = and 0 tanh2 v   − sech v tanh v 0 . (Lij ) = 0 sech v tanh v The Gaussian curvature for the pseudosphere is K = −1, which motivates the name “pseudosphere” since it is analogous to the unit sphere but with a constant Gaussian curvature of −1 instead of 1.

225

226

6. The First and Second Fundamental Forms

Figure 6.11. Tractrix and pseudosphere. Proposition 6.6.7. Let S be a regular oriented surface of class C 2 with

Gauss map n : S → S2 , and let p be a point on S. Call K(p) the Gaussian curvature of S at p, and suppose that K(p) 6= 0. Let Bε be the ball of radius ε around p and define Vε = Bε ∩ S. Then Area(n(Vε )) . ε→0 Area(Vε )

|K(p)| = lim

In other words, the absolute value of the Gaussian curvature at a point p is the limit around p of the ratio of the surface area on S2 mapped under the Gauss map to the corresponding surface area on S. ~ : U → R3 be a regular parametrization of a neighborProof: Let X ~ 0 , v0 ), where U is an open subset of R2 . Since X(U ~ ) hood of p = X(u ~ ) for all ε small enough. Furthermore, is an open set, Vε ⊂ X(U since we assumed that K(p) 6= 0, by the continuity of the Gaussian curvature function K : U → R, we deduce that K does not change sign for ε chosen small enough. Therefore, from now on, we assume that K does not change sign. ~ −1 (Vε ), the preimage of the neighborhood Vε under Define Uε = X ~ Since X ~ is bijective as a regular parametrization, {(u0 , v0 )} is X. the unique point in all Uε for all ε > 0. Then by the formula for surface area, ZZ ZZ ~u × X ~ v k du dv. dA = kX Area(Vε ) = Vε

6.6. Gaussian and Mean Curvatures

227

Similarly, on the unit sphere, ZZ

~u × N ~ v k du dv. kN

Area(n(Vε )) = Uε

However, by Proposition 6.6.2, we also have ZZ ~u × X ~ v k du dv. Area(n(Vε )) = |K(u, v)|kX Uε

By the Mean Value Theorem for double integrals, for every ε > 0, there exist points (uε , vε ) and (u0ε , vε0 ) in the open set Uε such that ZZ ~u × X ~ v k du dv = kX ~ u (uε , vε ) × X ~ v (uε , vε )k kX Uε

and ZZ

~u × N ~ v k du dv = |K(u0ε , vε0 )| kX ~ u (u0ε , vε0 ) × X ~ v (u0ε , vε0 )k. kN

Thus, since limε→0 (uε , vε ) = limε→0 (u0ε , vε0 ) = (u0 , v0 ), we have ~ u (u0 , v 0 ) × X ~ v (u0 , v 0 )k Area(n(Vε )) |K(u0ε , vε0 )| kX ε ε ε ε = lim ~ u (uε , vε ) × X ~ v (uε , vε )k ε→0 Area(Vε ) ε→0 kX lim

=

~ u (u0 , v0 ) × X ~ v (u0 , v0 )k |K(u0 , v0 )|kX ~ u (u0 , v0 ) × X ~ v (u0 , v0 )k kX

= |K(u0 , v0 )| = |K(p)|.



Problems 6.6.1. In Example 6.6.5 we calculated the Gaussian curvature of function graphs. Calculate the mean curvature. 6.6.2. Using an appropriate parametrization, find the Gaussian curvature of the hyperboloid of one sheet, x2 + y 2 − z 2 = 1. 6.6.3. Calculate the Gaussian and mean curvature functions of the general ellipsoid ~ X(u, v) = (a cos u sin v, b sin u sin v, c cos v).

228

6. The First and Second Fundamental Forms

6.6.4. Calculate the Gaussian curvature of the torus parametrized by ~ X(u, v) = ((a + b cos v) cos u, (a + b cos v) sin u, b sin v) where a > b are constants and (u, v) ∈ (0, 2π) × (0, 2π). 6.6.5. Consider a regular space curve ~γ : I → R3 and the tangential surface defined by ~ u) = ~γ (t) + u~γ 0 (t) X(t, for (t, u) ∈ I × R. Calculate the mean and Gaussian curvature functions. ~ 6.6.6. Let α ~ (t) and β(t) be differentiable vector functions with common domain I. Define the secant surface between the two resulting curves by ~ ~ u) = (1 − u)~ X(t, α(t) + uβ(t) for (t, u) ∈ I × R. Assume that the corresponding surface is regular. (a) Prove that K(u, v) = 0 for any point with u = 12 . (b) Prove that K(u, v) = 0 if and only if u = in the plane spanned by α ~ 0 (t) and β~ 0 (t).

1 2

~ −α or (β(t) ~ (t)) is

6.6.7. If M is a nonorientable surface, one can define the Gaussian curva~ =X ~u × X ~ v kX ~u × X ~ vk ture on a coordinate patch of M by using N and Equation (6.34) without reference to dnp , which is not well defined since no Gauss map n : M → S2 exists. Consider the M¨obius strip M depicted in Figure 5.15. Show that, using the parametrization of the M¨obius strip in Example 5.5.3, the Gaussian curvature is 1 K=−  . 1 2 u 2 2 4 v + (2 − v sin 2 ) 6.6.8. Tubes. Let ~γ : I → R3 be a regular space curve and let r be a positive real number. Consider the tube of radius r around ~γ (t) parametrized by ~ ~ X(u, v) = ~γ (u) + (r cos v)P~ (u) + (r sin v)B(u). (a) Calculate the second fundamental form (Lij ) and the matrix for dnp . (b) Calculate the Gaussian curvature function K(u, v) on the tube. (c) Prove that all the points with K = 0 are either points on ~ curves ~γ (t) ± rB(t) for all t ∈ I or points on circles ~γ (u0 ) + ~ 0 ), where u0 satisfies κ(u0 ) = 0. (r cos v)P~ (u0 ) + (r sin v)B(u

6.6. Gaussian and Mean Curvatures

6.6.9. Consider the pseudosphere and the parametrization provided in Example 6.6.6. (a) Prove the statements about the pseudosphere in Example 6.6.6. (b) Find the mean curvature function on the pseudosphere. (c) Modify the given parametric equations to find a parametrization of a surface with constant Gaussian curvature K = − R12 . (d) Determine the lines of curvature on the pseudosphere. ~ 6.6.10. Consider the monkey saddle X(u, v) = (u, v, u3 − 3uv 2 ). Calculate ~ and the points where K > 0, the Gaussian curvature function of X K < 0, or K = 0. 6.6.11. Surfaces of revolution. Consider the surface of revolution defined by revolving the parametrized curve (f (t), h(t)) in the xy-plane about the y-axis. Its paramtrization is ~ X(u, v) = (f (v) cos u, f (v) sin u, h(v)) for u ∈ [0, 2π) and v ∈ I, where I is some interval. Assume that f and h are such that the surface of revolution is a regular surface. (a) Calculate the second fundamental form (Lij ). (b) Calculate the coefficients of the matrix for dnp . (c) Calculate the Gaussian curvature function. (d) Determine which points are elliptic, hyperbolic, parabolic, or planar. (e) Prove that the lines of curvature are the coordinate lines. ~ 6.6.12. Normal Variations. Let X(u, v) be the parametrization for a coordinate patch V of a regular surface S. Consider the normal variation of S over V that is parametrized by ~ ~ (u, v) ~r (u, v) = X(u, v) + rN Y ~ is the unit normal vector for some constant r ∈ R and where N ~ associated to X. ~ is everywhere equal to N ~ ~ Y to Y (a) Prove that the unit normal N (except perhaps up to a sign). ~r and K the Gaussian (b) Call KY the Gaussian curvature for Y ~ Prove that curvature of X. ~ ~ ~u × X ~ v ) = KY ∂ Y r × ∂ Y r . K(X ∂u ∂v

229

230

6. The First and Second Fundamental Forms

(c) Call VY the corresponding coordinate patches on the normal variation. Conclude that ZZ ZZ K dA = KY dA. V

VY

~ = (β1 (t), β2 (t)) 6.6.13. Consider plane curves α ~ (t) = (α1 (s), α2 (s)) and β(t) both parametrized by arc length. Assume we are in R3 with standard basis {~i, ~j, ~k}. Consider the parametrized surface S given by ~ (s) + β2 (t)~k, ~ t) = α X(s, ~ (s) + β1 (t)U ~ (s) is the usual unit normal vector for plane curves. where U (a) Prove that if S is regular, then 1 − β1 (t)κα (s) 6= 0 for all (s, t), where κα (s) is the curvature of the plane curve α ~. (b) Calculate the Gaussian curvature of S. (c) Prove that κα (s) = 0 or κβ (t) = 0 imply that K(s, t) = 0, but explain why the converse is not true. 6.6.14. Let S be a regular oriented surface and p a point of S. Let ~u be any fixed unit vector in Tp S. Show that the mean curvature H at p is given by Z 1 π κn (θ) dθ, H= π 0 where κn (θ) is the normal curvature of S along a direction making an angle θ with ~u. 6.6.15. (*) Theorem of Beltrami-Enneper . Prove that the absolute value of the torsion at any point on an asymptotic curve with nonzero curvature is given by √ |τ | = −K, where K is the Gaussian curvature of the surface at that point. ~ and consider the 6.6.16. Let S be a regular surface in R3 parametrized by X, linear transformation T : R3 → R3 given by T (~x) = A~x with respect to the standard basis, where A is an invertible matrix. It is usually an intractable problem to determine the Gaussian or mean curvature of the image surface S 0 = T (S) from those of S. Nonetheless, it is possible to answer the following question: Prove that T preserves the sign of the Gaussian curvature of any surface in R3 , more precisely, if p ∈ S and q = T (p) is the corresponding point on S 0 , then K(p) = 0 ⇔ K(q) = 0 and sign K(p) = sign K(q). [Hint: Use Equation (6.18).]

6.7. Developable Surfaces and Minimal Surfaces

6.6.17. (*) Consider a regular surface S ∈ R3 defined by F (x, y, z) = 0 and a point p on S. Use implicit differentiation to find a formula for Gaussian curvature K at the point p. (The formula is not particularly pretty but can be written concisely using determinants.)

6.7 Developable Surfaces and Minimal Surfaces When studying plane curves, we showed in Proposition 1.3.5 that if the curvature κg (t) of a regular plane curve is always 0, then the curve is a line segment. In this section, we wish to study properties of surfaces with either Gaussian curvature everywhere 0 or mean curvature everywhere 0. Since there exist formulas for the mean curvature and Gaussian curvature in terms of a particular parametrization, the equations K = 0 and H = 0 are partial differential equations. However, since they are nonlinear differential equations that a priori involve three unknown functions in two variables, they are intractable in general. Nonetheless, we shall study various classes of surfaces that satisfy K = 0 or H = 0. Planes satisfy K = 0, but other surfaces do as well, for example, cylinders and cones. In a geometric sense, cylinders and cones in fact resemble a plane because, as any elementary school student knows, one can create a cylinder or a cone out of a flat paper without folding, stretching, or crumpling. In the first half of this section, we introduce ruled surfaces and determine the conditions for these ruled surfaces to have Gaussian curvature everywhere 0. As of yet, we have neither seen a particularly intuitive interpretation of the mean curvature nor have presented surfaces that satisfy H = 0. However, as we shall see, surfaces that satisfy H = 0 have minimal surface area in the following sense. Given a simple closed curve C in R3 , among surfaces S that have C = ∂S as a boundary, a surface with minimal surface area will have mean curvature H everywhere 0. Consequently, a surface that satisfies H = 0 is called a minimal surface. Many articles and books are devoted to the study of minimal surfaces (see , , or  to name a few; an Internet search will reveal many more), so in the interest of space, the second half of this section gives only a brief introduction to minimal surfaces.

231

232

6. The First and Second Fundamental Forms

6.7.1 Developable Surfaces Geometrically, we define a ruled surface as the union of a (differentiable) one-parameter family of straight lines in R3 . One can specify each line in the family by a point α ~ (t) and by a direction given by another vector w(t). ~ The adjective “differentiable” means that both α ~ and w ~ are differentiable vector functions over some interval I ⊂ R. We parametrize a ruled surface by ~ u) = α X(t, ~ (t) + u w(t), ~

with (t, u) ∈ I × R.

(6.35)

~ (t) with direction w(t) ~ the We call the lines Lt passing through α rulings, and the curve α ~ (t) is called the directrix of the surface. We should note that this definition does not insist that ruled surfaces be regular; in particular, we allow singular points, that is, points where ~t × X ~ u = ~0. X Definition 6.7.1. A developable surface is a surface S such that at each point P on S, there is a line (called generator) through P that lies on S and such that S has the same tangent plane at all points on this generator.

A developable surface is a surface that can be formed by bending a portion of the plane into space. Intuitively speaking, each bend line corresponds to a generator. The set of generator lines on the surface define a one-parameter family of lines in R3 that sweep out the surface. Hence, every developable surface is a ruled surface. Now suppose we can choose a directrix α ~ (t) of the developable with the generators as the rulings. Then the normal vector is constant along each ruling and hence the partial derivative in that direction is 0. Then by Proposition 6.6.2, a developable surface has a Gaussian curvature that is identically 0. Consequently, an equivalent definition of a developable surface is that it is a ruled surface that has Gaussian curvature that is constantly 0. Example 6.7.2 (Cylinders). The simplest example of a ruled surface is

a cylinder. The general definition of a cylinder is a ruled surface that can be given as a one-parameter family of lines {~ α(t), w(t)}, ~ where α ~ (t) is planar and w(t) ~ is a constant vector in R3 . It is easy to show

6.7. Developable Surfaces and Minimal Surfaces

(a) Cone.

233

(b) Hyperboloid.

Figure 6.12. Ruled surfaces.

that this has Gaussian curvature that is identically 0, so the cylinder is a developable surface. Example 6.7.3 (Cones). The general definition of a cone is a surface that can be given as a one-parameter family of lines {~ α(t), w(t)}, ~ where α ~ (t) lies in a plane P and the rulings Lt all pass through some common point p ∈ / P . Therefore, the cone over α ~ (t) through p can be parametrized by

~ u) = α X(t, ~ (t) + u(p − α ~ (t)). Again, it is possible to show that the Gaussian curvature is identically 0. Example 6.7.4. As a perhaps initially surprising example, the hyper-

boloid of one sheet is also a ruled surface. The standard parametrization for the hyperboloid of one sheet is ~ X(u, v) = (cosh v cos u, cosh v sin u, sinh v). Consider now the ruled surface with directrix α ~ (t) = (cos t, sin t, 0) 0 ~ and with ruling directions w(t) ~ =α ~ (t) + k = (− sin t, cos t, 1) (see Figure 6.12(b)). We obtain the parametrized surface ~ (t, u) = (cos t − u sin t, sin t + u cos t, u). Y

234

6. The First and Second Fundamental Forms

~ as a reparametrizaThough it is not particularly easy to express Y ~ tion of X, it is not hard to see that both of these parametrizations satisfy the following usual equation that gives the hyperboloid as a conic surface: x2 + y 2 − z 2 = 1. This ruled surface is not developable since the Gaussian curvature is not identically 0. In order to identify developable surfaces, we determine the Gaussian curvature for ruled surfaces generally. However, in order to simplify subsequent calculations, we make two additional assumptions that do not lose generality. First, note that one can impose the condition that kw(t)k ~ = 1 without changing the definition or the image of the parametrization in Equation (6.35). In defining particular ruled surfaces, it is usually easier not to make this assumption, but it does simplify calculations since kw(t)k ~ = 1 for all t ∈ I implies 0 = 0. Second, note that different curves α ~ (t) can that w(t) ~ · w(t) ~ serve as the directrix for the same ruled surface, so we wish to em~ ploy a curve β(t) as a directrix, which will simplify the algebra in ~ our calculations. We choose a curve β(t) satisfying β~ 0 (t) · w ~ 0 (t) = 0. ~ ~ we can write Since β(t) lies on X, ~ =α β(t) ~ (t) + u(t)w(t). ~

(6.36)

Then ~ 0 (t) + u0 (t)w(t) ~ + u(t)w ~ 0 (t), β~ 0 (t) = α and since β~ 0 · w ~ 0 = 0, we have ~ 0 (t) + u(t)w ~ 0 (t) · w ~ 0 (t) = 0. α ~ 0 (t) · w Thus, we determine that u(t) = −

~ 0 (t) α ~ 0 (t) · w . kw ~ 0 (t)k2

(6.37)

Furthermore, it is easy to prove (see Problem 6.7.4) that, with our ~ ~ present assumptions, β(t) is unique. We call the curve β(t) the line of stricture of the ruled surface.

6.7. Developable Surfaces and Minimal Surfaces

235

~ ~ = β(t)+ We now use the parametrization for the ruled surface X uw(t) ~ and proceed to calculate the first and second fundamental forms. Understanding that β~ and w ~ are functions of t, we find that ~ t = β~ 0 + u w ~ 0, X ~ u = w, X ~

and

~0

0

~t × X ~u = β × w X ~ + uw ~ × w, ~

~ tt = β~ 00 + u w X ~ 00 , ~ tu = w X ~ 0, ~ uu = ~0. X

One can already notice that if w(t) ~ is constant, then L12 = L21 = L22 = 0, which leads to K = 0 and proves that all cylinders have Gaussian curvature K = 0. We will assume now that w ~ 0 (t) is not identically 0. Because of the conditions w ~ ·w ~ 0 = 0 and β~ 0 · w ~ 0 = 0, we can 0 0 0 ~ ~ conclude that β × w ~ is parallel to w ~ . Thus, β × w ~ is perpendicular to w ~ ×w ~ 0 , so ~t × X ~ u k2 = kβ~ 0 × wk kX ~ 2 + u2 kw ~ 0 k2 . Furthermore, since β~ 0 × w ~ is parallel to w ~ 0 it is its own projection onto w ~ 0 . Hence, kβ~ 0 × wk ~ = |(β~ 0 × w) ~ ·w ~ 0 |/kw ~ 0 k and hence, ~t × X ~ u k2 = (β~ 0 w kX ~w ~ 0 )2 /kw ~ 0 k2 + u2 kw ~ 0 k2  1  ~0 0 2 2 0 4 ( β , w ~ w ~ ) + u k w ~ k = kw ~ 0 k2

(6.38)

~w ~ 0 ) for (β~ 0 × w) ~ ·w ~ 0 , the triple-vector where we use the notation (β~ 0 w product in R3 . Consequently, we can write the unit normal vector ~ as N kw ~ 0k

~ (t, u) = q N

  β~ 0 × w ~ + uw ~0 × w ~ ,

(β~ 0 w ~w ~ 0 )2 + u2 kw ~ 0 k4

and therefore, again using the conditions that w ~ ·w ~ 0 = 0 and β~ 0 · w ~0 = 0, we get ! kβ~ 0 k2 + u2 kw ~ 0 k2 β~ 0 · w ~ (gij ) = ~ 1 β~ 0 · w and

236

6. The First and Second Fundamental Forms

(Lij ) = q

kw ~ 0k (β~ 0 w ~w ~ 0 )2 + u2 kw ~ 0 k4

! (β~ 00 + uw ~ 00 ) · (β~ 0 × w ~ + uw ~ 0 × w) ~ (β~ 0 w ~w ~ 0) . ~w ~ 0) 0 (β~ 0 w

We cannot say much for the entry L11 , but thanks to Equation ~u × X ~ v k2 = (6.34) for the Gaussian curvature and the fact that kX det(gij ) (see Equation (6.6)), we can calculate the Gaussian curvature for a ruled surface as K=

det(Lij ) ~w ~ 0 )2 kw ~ 0 k4 (β~ 0 w = − 2 . det(gij ) (β~ 0 w ~w ~ 0 )2 + u2 kw ~ 0 k4

(6.39)

This formula for the Gaussian curvature of a ruled surface makes a few facts readily apparent. First, K ≤ 0 for all points of a ruled surface. Second, by Equation (6.38), all singular points of a ruled ~ u = ~0, must have u = 0 and therefore ~ t ×X surface, i.e., points where X occur on the line of stricture. Finally, by Equation (6.39), a ruled surface satisfies K(t, u) = 0 if and only if (β~ 0 w ~w ~ 0 ) = 0 for all t ∈ I. Let us return now to the general definition of a ruled surface from Equation (6.35) with regular curves α ~ (t) and w(t) ~ with no conditions. Since both w ~ and w ~ 0 are perpendicular to w ~ ×w ~ 0, (~ α0 w ~w ~ 0) = α ~0 · w ~ ×w ~ 0 = β~ 0 · w ~ ×w ~ 0 = (β~ 0 w ~w ~ 0 ). Furthermore, if we call w ˆ = w/k ~ wk, ~ we remark that (~ α0 w ~w ~ 0 ) = kwk ~ 2 (~ α0 w ˆw ˆ 0 ) = kwk ~ 2 (β~ 0 w ˆw ˆ 0 ). Consequently, (β~ 0 w ˆw ˆ 0 ) = 0 if and only if (~ α0 w ~w ~ 0 ) = 0. We have proven the following proposition. Proposition 6.7.5. A ruled surface in R3 with directrix α ~ (t) and such

that each ruling has direction w(t) ~ is a developable surface if and only if (~ α0 w ~w ~ 0 ) = 0. Essentially by definition, every developable surface has Gaussian curvature that is identically 0. Surprisingly, the converse is true. Theorem 6.7.6. A regular surface S of class C 2 in R3 has K = 0

identically if and only if the surface is a developable surface.

6.7. Developable Surfaces and Minimal Surfaces

Proof: Suppose that S is a regular surface of class C 2 such that ~ : U → R3 be a parametrization X(u, ~ K = 0. Let X v) of a coordinate patch of S. The following reasoning will occur on this coordinate patch but the result with extend to the whole surface. ~ u, N ~v} By Proposition 6.6.2, K(u, v) = 0 if and only if the set {N are everywhere linearly dependent. Also notice that K = 0 if and only if L11 L22 − L212 = 0. Obviously, in order for this equality to hold L11 and L22 must have equal signs everywhere. Without loss of generality in what follows, we assume that they are nonnegative. Consider the asymptotic curves on the surface whose parametrization (u(t), v(t)) satisfy the differential equation L11 (u0 )2 + 2L12 u0 v 0 + L22 (v 0 )2 = 0 p ⇐⇒L11 (u0 )2 + 2 L11 L22 u0 v 0 + L22 (v 0 )2 = 0 p p ⇐⇒( L11 u0 + L22 v 0 )2 = 0. Not both L11 and L22 can be zero for L11 L22 −L212 = 0 to hold. If we suppose that L22 6= 0, then by substitution, this gives the differential equation r dv L11 . =− du L22 By the Theorem of Existence and Uniqueness for differential equations, for each point (u0 , v0 ) ∈ U , there is an asymptotic curve through (u0 , v0 ). Consequently, at each point p ∈ S, there is a neighborhood of p for which it is possible to change parametrizations to use coordinates (¯ u, v¯) such that the asymptotic curves corresponding to v¯ are constant. Using the (¯ u, v¯) coordinates, the equation for asymptotic curves ¯ 22 (¯ is simply v¯0 = 0 or more precisely L v 0 )2 = 0. From this, we deduce that the components of the second fundamental form have ¯ 11 = L ¯ 12 = 0. Thus L ~ u¯ · N ~ u¯ = X ~ v¯ · N ~ u¯ = 0. X ~ u¯ = 0 at all points on the surface and hence the normal vector Thus N depends on only one parameter, v¯. Now suppose that the tangent planes of S depend on only one parameter v. Then there exist a differentiable vector function ~a(v)

237

238

6. The First and Second Fundamental Forms

Figure 6.13. A developable surface.

and a differential real-valued function f (v) such that the tangent spaces of S (possibly on some open coordinate neighborhood of S) satisfy the equation ~x · ~a(v) = f (v),

where ~x = (x, y, z).

Given a fixed v0 we have ~x · ~a(v0 ) = f (v0 ) as well as ~x · ~a(v0 + h) = f (v0 + h) as h → 0. In particular, this leads to the fact the normal vectors to the surface S that are also in the tangent plane to S at a point that has v = v0 must also satisfy ~x · ~a0 (v0 ) = f 0 (v0 ), which is another plane. Hence, such normal vectors are perpendicular to two planes. Consequently, the surface is a ruled surface, such that along the rulings, the normal vectors are constant. Thus the surface S is developable.  It is interesting to remark that a ruled surface is a cone if α ~ 0 (t) = 0 and is a cylinder if and only if w ~ 0 (t) = 0, showing again that both cones and cylinders are developable and have K = 0. The exercises present examples of developable surfaces that are neither cones nor cylinders. Figure 6.13 also illustrates a developable surface that has α ~ = (0, t, cos t). Developable surfaces are particularly interesting for design and manufacturing. Because developable surfaces are ruled and have Gaussian curvature 0, they can be created by bending a region of the plane, without stretching it. Consequently, developable surfaces can be made out of sheet metal or any inelastic material that starts

6.7. Developable Surfaces and Minimal Surfaces

out flat. Some postmodern architecture exemplifies the use of developable surfaces (such as Frank Gehry’s Guggenheim Museum in Bilbao or City of Wine Hotel Marqu`es de Riscal).

6.7.2 Minimal Surfaces Definition 6.7.7. A minimal surface is a parametrized surface of class C 2 that satisfies the regularity condition and for which the mean curvature is identically 0.

We first wish to justify the name “minimal.” ~ : U → R3 be a coordinate neighborhood of a regular Let X parametrized surface of class C 2 . Let D0 be a connected compact ~ 0 ). Let h : U → R be a differentiable set in U , and let D = X(D ~ over D0 determined by h is the function. A normal variation of X family of surfaces with t ∈ (−ε, ε) defined by ~ t : D0 −→ R3 X ~ ~ (u, v). (u, v) 7−→ X(u, v) + th(u, v)N ~ t be a normal variation of X ~ over a compact Proposition 6.7.8. Let X region D0 and determined by some function h. For ε small enough, ~ t satisfies the regularity condition for all t ∈ (−ε, ε). In this case, X ~ t is the area of X ZZ p q A(t) = 1 − 4thH + t2 R det(gij ) du dv (6.40) D0

for some function R(u, v, t) that is polynomial in t. t as the coefficients of the metric tensor. The surProof: Denote gij faces of the normal variation have

~ u + thu N ~ + thN ~ u, ~ ut = X X ~ vt = X ~ v + thv N ~ + thN ~v. X Thus, we calculate, t ~u · N ~ u + t2 h2 N ~ u + t2 h2u , ~u · N g11 = g11 + 2thX t ~u · N ~v + X ~v · N ~ u ) + t2 h2 N ~ v + t2 hu hv , ~u · N g12 = g12 + th(X t ~v · N ~ v + t2 h2 N ~ v + t2 h2v . ~v · N g22 = g22 + 2thX

239

240

6. The First and Second Fundamental Forms

However, by Equation (6.17), one can summarize the above equations as t ~ j + t2 hi hj , ~i · N gij = gij − 2thLij + t2 h2 N where we use the notation h1 (resp. h2 ) to indicate the partial derivative hu (resp. hv ). Therefore, we calculate that t ¯ ) = det(gij ) − 2th(g11 L22 − 2g12 L12 + g22 L11 ) + t2 R, det(gij

¯ v, t) is a function of the form A0 (u, v) + tA1 (u, v) + where R(u, 2 t A2 (u, v) for continuous functions Ai defined over D0 . However, the mean curvature is   1 g11 L22 − 2g12 L12 + g22 L11 H= , 2 2 g11 g22 − g12 which leads to t ¯ = det(gij )(1 − 4thH + t2 R), det(gij ) = det(gij )(1 − 4thH) + t2 R (6.41) ¯ det(gij ). Since D0 is compact, the functions h, H, where R = R/ and R are bounded over D0 , which shows that t ) = det(gij ) lim det(gij

t→0

for all (u, v) ∈ D0 .

t ) 6= 0 for all t ∈ (−ε, ε), Hence, if ε > 0 is small enough, then det(gij and thus, all normal variations satisfy the regularity condition. The rest of the proposition follows from Equation (6.41) since

A(t) =

ZZ q D0

t ) du dv. det(gij



~ : U → R3 be a parametrized surface of class Proposition 6.7.9. Let X

C 2 , and let D0 ⊆ U be a compact set. Let A(t) be the area function ~ parametrizes a minimal surface defined in Equation (6.40). Then X 0 ~ if and only if A (0) = 0 for all D0 and all normal variations of X 0 over D .

6.7. Developable Surfaces and Minimal Surfaces

Proof: We calculate ZZ q −4hH + 2tR + t2 Rt √ det(gij ) du dv, A0 (t) = 2 1 − 4thH + t2 R D0

which implies that 0

ZZ

A (0) = −2

hH

q det(gij ) du dv.

D0

~ parametrizes a minimal surface, which means that Obviously, if X H = 0 for all (u, v) ∈ U , then A0 (0) = 0 regardless of the function h(u, v) or the compact set D0 . To prove the converse, suppose that A0 (0) = 0 for all continuous functions h(u, v) and all compact D0 ⊂ U . Choosing h(u, v) = H(u, v), since det(gij ) > 0, we have ZZ q 0 2 A (0) = −2 H det(gij ) du dv, D0

so A0 (0) ≤ 0 for all choices of D0 . Since H(u, v) is continuous, if H(u0 , v0 ) 6= 0 for any point (u0 , v0 ), there is a compact set D0 containing (u0 , v0 ) such that H(u, v)2 > 0 over D0 . Thus, if H is anywhere nonzero, there exists a compact subset D0 such that A0 (0) < 0. This is a contradiction, so we conclude that A0 (0) = 0 for all h and all D0 implies that H(u, v) = 0 for all (u, v) ∈ U .  ~ that satProposition 6.7.9 shows that a parametrized surface X ~ ~ isfies the regularity condition kXu × Xv k 6= 0 is minimal (has H = 0 everywhere) if it is a surface such that over every patch of surface there is no way to deform it along normal vectors to obtain a surface of lesser surface area. Minimal surfaces have also enjoyed considerable popular attention, especially in museums of science, as soap films on a wire frame. When a wire frame is dipped into a soapy liquid and pulled out, the surface tension on the soap film pulls the film into the state of least potential energy, which turns out to be the surface such that no normal variation can lead to a surface with less surface area. Of course,

241

242

6. The First and Second Fundamental Forms

when experimenting with soap films, one can experiment with wire frames that are nonregular curves or not even curves at all (the skeleton of a cube, for example) and investigate what kinds of soap film surfaces result. The problem of determining a minimal surface with a given closed regular space curve C as a boundary was raised by Lagrange in 1760. However, the mathematical problem became known as Plateau’s problem, after Joseph Plateau who specifically studied soap film surfaces. The 19th century saw a few specialized solutions to the problem, but it was not solved until 1930. (See  for a more complete history of Plateau’s problem.) Though one can easily construct a minimal surface with a soap film on a wire frame, either checking that a surface is minimal or finding a parametrization for a minimal surface is quite difficult. In fact, the study of minimal surfaces continues to provide new areas of research and connections with other branches of analysis. Perhaps among the most interesting results is a connection between complex analytic functions and minimal surfaces [11, p. 206] or the use of elliptic integrals to parametrize special minimal surfaces [27, Chapter 1], topics that go beyond the scope of this book. However, it is quite likely that having both the simple experimentative aspect and connections to advanced mathematics has fueled interest in minimal surfaces.

Problems 6.7.1. Prove that the surface given by z = kxy, where k is a constant, is a ruled surface and give it as a one-parameter family of lines. 6.7.2. Figure 6.13 shows a developable surface that has the line of stricture α ~ (t) = (0, t, cos t). Suppose the rulings are w(t) ~ = (1, w2 (t), w3 (t)). Find the equations or differential equation required by w2 (t) and w3 (t) to produce a developable surface. [In Figure 6.13, the rulings have w(t) ~ = (1, −0.3t, −0.3 cos t).] 6.7.3. The tangential surface to a space curve α ~ (t) was presented in Problem 6.6.5. Show that the curve α ~ is the line of stricture for the tangential surface. Show that the tangential surface to a regular space curve is a developable surface. ~ and {~ α2 (t), w(t)}, ~ where kw(t)k ~ = 1, are 6.7.4. Suppose that {~ α1 (t), w(t)} two one-parameter families of lines that trace out the same ruled

6.7. Developable Surfaces and Minimal Surfaces

243

surface. Prove that Equation (6.36) with (6.37), using either α ~ 1 or α ~ 2 , produce the same line of stricture. 6.7.5. Suppose that a ruled surface has α ~ (t) = (cos t, sin t, 0) as the line of stricture. Suppose also that w(t) = (w1 (t), w2 (t), 1). Find algebraic or differential equations that w1 (t) and w2 (t) must satisfy so that the ruled surface is a developable surface. 6.7.6. Let S be an orientable surface, and let α ~ (s) be a curve on S parametrized by arc length. Assume that α ~ is nowhere tangent to the ~ (s) as the unit normal vector asymptotic direction on S, and define N to S along α ~ (s). Consider the ruled surface ~ (s) × N ~ 0 (s) N ~ u) = α X(s, ~ (s) + u . ~ 0 (s)k kN The assumption that α ~ 0 (s) is not an asymptotic direction ensures 0 ~ ~ v) is a developable surface. (This ~ that N (s) 6= 0. Prove that X(s, kind of surface is called the envelope of a family of tangent planes along a curve of a surface.) ~ v) = α 6.7.7. Let X(t, ~ (t) + v w(t) ~ be a developable surface. Prove that at a regular point ~v · X ~t = 0 N

and

~v · X ~ v = 0. N

Use this result to prove that the tangent plane of a developable surface is constant along a line of ruling. 6.7.8. Show that a surface F (x, y, z) = 0 is developable if Fxx Fxy Fxz Fx Fyx Fyy Fyz Fy Fzx Fzy Fzz Fz = 0. Fx Fy Fz 0 ~ 6.7.9. Consider the helicoid parametrized by X(u, v) = (v cos u, v sin u, cu), where c is a fixed number. (a) Determine the asymptotic curves. (b) Determine the lines of curvature. (c) Show that the mean curvature of the helicoid is 0. 6.7.10. Prove that the catenoid parametrized by ~ X(u, v) = (a cosh v cos u, a cosh v sin u, av) is a minimal surface.

244

6. The First and Second Fundamental Forms

~ 6.7.11. Prove that Enneper’s minimal surface X(u, v) = (u − v3 2 2 2 + vu , u − v ) deserves its name. 3

u3 3

+ uv 2 , v −

6.7.12. Show that for all constants D1 and D2 , the surface parametrized by the function graph ~ X(u, v) = (u, v, − ln(cos(u + D1 )) + ln(cos(−v + D2 ))) where it is defined is a minimal surface. 6.7.13. (ODE) Prove that the only nonplanar minimal surfaces of the form ~ X(u, v) = (u, v, h(u) + k(v)) have h(u) = −

1 ln(cos(Cu + D1 )) C

and

k(v) =

1 ln(cos(−Cv + D2 )) C

for some nonzero constant C and any constants D1 and D2 . ~ : U → R3 and Y ~ : U → R3 6.7.14. Let U ⊂ R2 , and suppose that X parametrize two minimal surfaces defined over the same domain. ~ t (u, v) = Prove that for all t ∈ [0, 1], the surfaces parametrized by Z ~ ~ (u, v) are also minimal surfaces. (1 − t)X(u, v) + tY 6.7.15. (ODE) Prove that the only surface of revolution that is a minimal surface is a catenoid (see Problem 6.7.10). 6.7.16. Suppose that S is a minimal regular surface with no planar points. Prove that at all points p ∈ S, the Gauss map n : S → S2 satisfies ~ 1 ) · dnp (w ~ 2 ) = −K(p)w ~1 · w ~2 dnp (w for all w ~ 1, w ~ 2 ∈ Tp S, where K(p) is the Gaussian curvature of S at p. Use this result to show that on a minimal surface the angle between two intersecting curves on S is the angle between their images on S2 under the Gauss map n. ~ 6.7.17. Let X(u, v) be a parametrization of a regular, orientable surface, ~ (u, v) = X ~u ×X ~ v /kX ~u ×X ~ v k be the orientation. A parallel and let N ~ is a surface parametrized by surface to X ~ (u, v) = X(u, ~ ~ (u, v). Y v) + aN ~ prove (a) If K and H are the Gaussian and mean curvatures of X, that ~v = (1 − 2Ha + Ka2 )X ~u × X ~ v. ~u × Y Y

6.7. Developable Surfaces and Minimal Surfaces

~ is (b) Prove that at regular points, the Gaussian curvature of Y K , 1 − 2Ha + Ka2 and the mean curvature is H − Ka . 1 − 2Ha + Ka2 ~ is a surface with constant (c) Use the above to prove that if X mean curvature H = c 6= 0, then there is a parallel surface to ~ that has constant Gaussian curvature. X

245

CHAPTER 7

The Fundamental Equations of Surfaces In the previous chapter, we began to study the local geometry of a surface and saw how many computations depend on the coefficients of the first and second fundamental forms. In fact, in Section 6.1 we saw that many metric calculations (angles between curves, arc length, area, etc.) depend only on the first fundamental form. We first approached such concepts from the perspective of objects in R3 , but with the first fundamental form, one can perform all the calculations by using only the coordinates that parametrize the surface, without referring to the ambient space. In fact, it is not at all difficult to imagine a regular surface as a subset of Rn , with n > 3, and the formulas that depend on the first fundamental form would remain unchanged. See Chapter 9. Concepts that rely only on the first fundamental form are called intrinsic properties of a surface. On the other hand, concepts such as the second fundamental form, the Gauss map, Gaussian curvature, and principal curvatures were defined in reference to the unit normal vector, and these are not necessarily intrinsic properties. To illustrate this point, consider a parametrized surface S that is a subset of R4 , and let p be a point of S. One can still define the tangent plane as the span of two linearly independent tangent vectors, but there no longer exists a unique (up to sign) unit normal vector to S at p. In this situation, one cannot define the Gauss map. Properties that we presented as depending on the Gauss map either cannot be defined in this situation or need to be defined in an alternate way. This chapter studies what kind of information about a surface one can know from just the first fundamental form versus that which can be determined from knowledge of both the first and the second fundamental forms.

247

248

7. The Fundamental Equations of Surfaces

Section 7.1 introduces the Christoffel symbols and studies relations between the coefficients of the first and second fundamental forms. In Section 7.2, we present the famous Theorema Egregium, which proves that the Gaussian curvature of a surface is in fact an intrinsic property. More precisely, the theorem expresses the Gaussian curvature in terms of the gij functions. In Section 7.3, we present another landmark result for surfaces in R3 , the Fundamental Theorem of Surface Theory, which proves that under appropriate conditions, the coefficients of the first and second fundamental forms determine the surface up to position and orientation in space. Some of the quantities we encounter in this chapter involve multiple indices. Some of these quantities represent a new mathematical object called a tensor. Tensors simultaneously generalize vectors, matrices, inner products, and many other objects that arise in linear algebra. Furthermore, just as it is possible to define a vector field that to each point in a region of the plane (or space) one associates a vector, so it is possible to define a tensor field. An introduction to tensor notation and the classical description of a tensor is given in the appendix in Section A.1. The only notational convention we underscore here is the Einstein summation convention: In a tensor notation expression, whenever a superscript index also appears as a subscript index, it is understood that we sum over α that index. For example, suppose that Aij kl and Bβγ are different collections of quantities with each index showing running from 1 to n, where n is the dimension of the ambient space Rn . Then by the k expression Aij kl Bim we mean the set of quantities def

j k Clm = Aij kl Bim =

n n X X

k Aij kl Bim .

i=1 k=1

Aij kl

The collection of quantities consists of n4 quantities, the collecα consists of n3 quantities, and the collection C j consists tion of Bβγ lm again of n3 quantities.

7.1 Gauss’s Equations and the Christoffel Symbols We now return to the study of surfaces. As we shall see, the results of this section assume that one can take the third derivative of a

7.1. Gauss’s Equations and the Christoffel Symbols

coordinate parametrization. Therefore, in this section and in the remainder of the chapter, unless otherwise stated we consider only regular surfaces of class C 3 . If one compares the theory of surfaces developed so far to the theory of curves, one will point out one major gap in the presentation of the former. In the theory of space curves, we discussed natural equations, namely the curvature κ(s) and torsion τ (s) with respect to arc length, and we proved that these two functions locally define a unique curve up to its position in space. Implicit in the proof of this result for natural equations of curves was the fact that in general there do not exist algebraic relations between the functions κ(s) and τ (s). In the theory of surfaces, the problem of finding natural equations cannot be quite so simple. For example, even when restricting our attention to the first fundamental form, we know that given any three functions E(u, v), F (u, v), and G(u, v), there does not necessarily exist a surface with     g11 g12 E F = F G g21 g22 because we need the nontrivial requirement that EG − F 2 > 0. Furthermore, one might suspect that, because of the smoothness ~ uuv = X ~ uvu = X ~ vuu , the (gij ) and (Lij ) conditions that imply that X coefficients may satisfy some inherent relations. Also, when discussing natural equations for space curves, one often uses the Frenet frame as a basis for R3 , with origin at a point p of the curve. In particular, one must use the Frenet frame to obtain a parametrization of a curve in the neighborhood of p from the knowledge of the curvature κ(s) and torsion τ (s). When performing calculations in the neighborhood of a point p on a surface ~ : U → R3 , the basis {X ~ 1, X ~ 2, N ~ } is the most S parametrized by X natural, but, unlike the Frenet frame, this basis is not orthonormal. ~ 1, X ~ 2, N ~ } depends significantly on the paIn addition, the basis {X ~ rametrization X, while a reparametrization of a curve can at most ~ change the sign of T~ and B. ~ }, where ~e1 and ~e2 are the principal One might propose {~e1 , ~e2 , N directions, as a basis more related to the geometry of a surface. The orthonormality has advantages, but calculations using this basis

249

250

7. The Fundamental Equations of Surfaces

quickly become intractable. Furthermore, the eigenvectors ~e1 and ~e2 are not well-defined at umbilic points since the eigenspaces of dnp at umbilic points consist of the whole tangent space Tp S. For these ~ 2, N ~ } for R3 ~ 1, X reasons, it remains more natural to use the basis {X in a neighborhood of p and study the relations that arise between the (gij ) and (Lij ) coefficients. Recapping earlier definitions, we have     ~1 X ~2 X ~ ~1 · X ~1 · N ~1 · X X g11 g12 0 ~ ~ ~2 X ~ ~2 · X ~2 · N X2 · X1 X  = g21 g22 0 . ~ ·X ~1 N ~ ·X ~2 N ~ ·N ~ 0 0 1 N Every vector in R3 can be expressed as a linear combination in this basis. In particular, one would like to express the second ~ 12 , X ~ 21 , and X ~ 22 as linear combinations of this ~ 11 , X derivatives X ~ ij · N ~ , and since basis. From Equation (6.18), we know that Lij = X ~ ij along basis ~ i = 0, we deduce that Lij is the coordinate of X ~ ·X N ~ . However, we do not know the coordinates of X ~ ij along vector N ~ 2. ~ 1 and X the other two basis vectors X Definition 7.1.1. The collection of eight functions Γijk : U → R with

indices 1 ≤ i, j, k ≤ 2 are defined as the unique functions that satisfy 2 ~ ~ jk = Γ1 X ~ ~ X jk 1 + Γjk X2 + Ljk N .

(7.1)

Definition 7.1.1 names the functions Γijk implicitly, but we now proceed to find formulas for them in terms of the metric coefficients. Note that though we have eight combinations for three indices, each ranging between 1 and 2, we do not in fact have eight distinct ~ jk = X ~ kj . Thus, we already know that functions because X Γijk = Γikj . Before determining formulas for the Γijk functions, we will first ~ ij · X ~ k that are easier to find. Since these establish expressions for X quantities occur frequently, there is a common shorthand symbol for them, namely, ~ k. ~ ij · X [ij, k] = X

7.1. Gauss’s Equations and the Christoffel Symbols

251

Begin by fixing j = k = 1. Using the formulas ∂g11 ∂ ~ ~ ~ 11 · X ~ 1, = (X1 · X1 ) = 2X 1 ∂x ∂x1 ∂ ~ ~ ∂g12 ~ 11 · X ~2 + X ~ 21 · X ~ 1, = (X1 · X2 ) = X ∂x1 ∂x1 ∂ ~ ~ ∂g11 ~ 12 · X ~ 1, = (X1 · X1 ) = 2X 2 ∂x ∂x2 we deduce that ∂g12 1 ∂g11 1 ∂g11 and [11, 2] = − . [11, 1] = 1 2 ∂x ∂x1 2 ∂x2 Pursuing the calculations for the remaining cases, using the fact that (gij ) is symmetric, and rewriting the results in a convenient manner, one can prove the following lemma. Lemma 7.1.2. For all indices 1 ≤ i, j, k ≤ 2, we have

~ ij · X ~k = [ij, k] = X

1 2



∂gjk ∂gki ∂gij + − ∂xi ∂xj ∂xk

 .

(7.2)

Using this lemma, one can easily establish a formula for the functions Γijk . We remind the reader that we shall often use the Einstein summation convention when an index is repeated in one superscript and one subscript position. (See Section A.1 for a more accurate explanation of the convention.) Before we give the proof, we also mention the useful artifice in tensorial notation called the Kronecker delta δij with 1 ≤ i ≤ n and 1 ≤ j ≤ n defined as ( 1 if i = j j δi = 0 if i 6= j. The Kronecker delta is a way to represent the identity matrix. ~ : U → R3 be the parametrization of a regular Proposition 7.1.3. Let X ~ surface of class C 2 in the neighborhood of some point p = X(q). i Then the coefficients Γjk satisfy Γijk

=

2 X l=1

(g ij ),

where (gij )−1 .

il

g [jk, l] =

2 X l=1

g

il 1

2



∂gjk ∂gkl ∂glj + − , j k ∂x ∂x ∂xl

 (7.3)

with the indices in superscript, is the inverse matrix

252

7. The Fundamental Equations of Surfaces

~ ij · X ~ k. Proof: We have already determined formulas for [ij, k] = X However, from Equation (7.1), taking a dot product with respect to ~ i , we obtain the derivative vectors X [ij, k] = Γlij glk . Multiplying the matrix of the first fundamental form (glk ) by its inverse, we can write glk g kα = δlα where δlα is the Kronecker delta (defined in Equation (A.3)). Then we get 2 X g kα [ij, k] = Γlij glk g kα = Γlij δlα = Γαij . k=1

The proposition follows from the symmetry of the (g ij ) matrix.



Definition 7.1.4. The symbols [ij, k] are called the Christoffel symbols

of the first kind, while the functions Γijk are called the Christoffel symbols of the second kind or, more simply, just the Christoffel symbols. The formula in Proposition 7.1.3 along with Definition 7.1.1 is called Gauss’s formula for surfaces. One of the first uses of Gauss’s formula is that knowing the functions Ljk and Γijk allows us to write not just the second partial ~ with coordinates with respect derivatives of the parametrization X ~ ~ ~ to the basis ordered (X1 , X2 , N ), but also all higher derivatives of ~ 1 , x2 ). We point out that knowing gij in the neighborhood of a X(x ~1 point allows us to determine the lengths of and angle between X ~ and X2 . Furthermore, as we shall see later, knowing the three distinct functions Ljk and the six distinct functions Γijk for a particular surface S in the neighborhood of p allows one to use a Taylor series expansion in the two variables x1 and x2 to write an infinite sum that provides a parametrization of S in a neighborhood of p once ~ 1 , and N ~ are given. However, we still have not identified any p, X relations that must exist between Lij and gij , so we cannot yet state an equivalent to the natural equations theorem for space curves.

7.1. Gauss’s Equations and the Christoffel Symbols

253

We point out that though we used a superscript and subscripts for the Christoffel symbol Γijk , these functions do not form the components of a tensor but rather transform according to the following proposition. Proposition 7.1.5. Let (x1 , x2 ) and (¯ x1 , x ¯2 ) be two coordinate systems

for a neighborhood of a point p on a surface S. If we denote by Γm ij ¯ µ the Christoffel symbols in the respective coordinate systems, and Γ αβ then they are related by i j ¯µ m ¯µ ∂ 2 xm ∂ x ¯ µ = ∂x ∂x ∂ x Γ + . Γ αβ ∂x ¯α ∂ x ¯β ∂xm ij ∂ x ¯α ∂ x ¯β ∂xm

Proof: We leave some of the details of this proof for the reader but present an outline here. x1 , x ¯2 ) for a Consider two systems of coordinates (x1 , x2 ) and (¯ neighborhood of a point p on a surface S. We know that the metric coefficients change according to g¯αβ =

∂xi ∂xj gij ∂x ¯α ∂ x ¯β

and

g¯αβ =

¯β ij ∂x ¯α ∂ x g . ∂xi ∂xj

(7.4)

¯2 ) coordinate system, we denote the Christoffel symbols In the (¯ x1 , x of the first kind by   gβν gαβ 1 ∂¯ ∂¯ gνα ∂¯ , (7.5) [αβ, ν] = + − 2 ∂x ¯α ∂x ¯ν ∂x ¯β and we must first relate this to the Christoffel symbols [ij, k] in the (x1 , x2 ) coordinate system. Note that in this proof, we make indices (i, j, k, m) in the (x1 , x2 ) coordinate system correspond to indices (α, β, ν, µ). Using Equation (7.4), the first term in Equation (7.5) transforms according to ∂¯ gβν ∂ 2 xj ∂xk ∂xj ∂ 2 xk ∂xj ∂xk ∂xi ∂gjk = g + g + . jk jk ∂x ¯α ¯ν ¯α ∂ x ¯ν ¯ν ∂ x ¯α ∂xi ∂x ¯α ∂ x ¯β ∂ x ∂x ¯β ∂ x ∂x ¯β ∂ x (7.6) Equation (7.6) and the corresponding results of the other two terms in the sum in Equation (7.5) produce an expression for the transformation property of the Christoffel symbol of the first kind.

254

7. The Fundamental Equations of Surfaces

The expression is not pleasing, especially since it involves the coefficients gij explicitly. One must now calculate ¯µ = Γ αβ

2 X

g¯µν [αβ, ν] =

ν=1

2 X ∂x ¯µ ∂ x ¯ν ml g [αβ, ν], ∂xm ∂xl ν=1

where we replace [αβ, ν] with the terms found in Equation (7.6) and similar equalities. After appropriate simplifications, we find that  µ  ¯ ∂ 2 xj ∂x ¯µ ∂xi ∂xj m 1 ∂ x ∂x ¯ µ ∂ 2 xi µ ¯ Γαβ = Γ + + ∂xm ∂ x ¯α ∂ x ∂xi ∂ x ¯β ij 2 ∂xj ∂ x ¯α ∂ x ¯β ¯α ∂ x ¯β   ¯µ ∂xj ∂ 2 xk ml 1 ∂x ¯ν ∂ x ∂xj ∂ 2 xi ml + g gjk − α α ν g gij 2 ∂xl ∂xm ∂ x ¯α ∂ x ¯α ∂ x ¯ν ∂x ¯ ∂x ¯ ∂x ¯  i  ν µ 2 k i 2 j 1 ∂x ∂x ∂ x ¯ ¯ ∂x ∂x ∂ x ml ml + g gki − α ν β g gji . 2 ∂xl ∂xm ∂ x ¯α ∂ x ∂x ¯ ∂x ¯β ∂ x ¯ν ¯ ∂x ¯ (7.7) However, it is important to remember that when one sums over an index, the actual name of the index does not change the result of the summation. Applying this observation to Equation (7.7) and remembering that the components gij are symmetric in their indices finishes the proof of the proposition.  One should not view the fact that the Christoffel symbols Γijk do not form the components of a tensor as just a small annoyance; it is of fundamental importance in the theory of manifolds. From an intuitive perspective, one could understand tensors as objects that are related to the tangent space to a surface at a point. However, since the Γijk functions explicitly involve the second derivatives of a parametrization of a neighborhood of a point on a surface, one might have been able to predict that the Γijk functions would not necessarily form the components of a tensor. Example 7.1.6 (Sphere). As a simple example, consider the usual pa-

rametrization of the sphere, ~ 1 , x2 ) = (R cos x1 sin x2 , R sin x1 sin x2 , R cos x2 ). X(x

7.1. Gauss’s Equations and the Christoffel Symbols

255

In spherical coordinates, the variable names we used correspond to x1 = θ as the meridian and x2 = ϕ the latitude, measured as the angle down from the positive z-axis. A simple calculation leads to !   2 2 2 1 0 1 R sin (x ) 0 2 2 . and g ij = 2 sin (x ) gij = 0 R2 R 0 1 By Equation (7.2), we find that [11, 2] = −R2 sin(x2 ) cos(x2 ),

[11, 1] = 0, [12, 1] = R2 sin(x2 ) cos(x2 ), 2

2

[12, 2] = 0,

2

[21, 1] = R sin(x ) cos(x ),

[21, 2] = 0,

[22, 1] = 0,

[22, 2] = 0.

Then, using Equation (7.3), we get Γ111 = 0, Γ112 = Γ121 = cot(x2 ), Γ122 = 0,

Γ211 = − sin(x2 ) cos(x2 ), Γ212 = Γ221 = 0, Γ222 = 0.

As an example of an application of Gauss’s equation, recall from Equation (6.33) that two vectors ~u1 and ~u2 in the tangent space Tp S to a surface S at p are called conjugate if Ip (dnp (~u1 ), ~u2 ) = Ip (~u1 , dnp (~u2 )) = 0.

(7.8)

Intuitively speaking, Equation (7.8) states that conjugate directions ~ are such that the direction of change of the unit normal vector N along ~u1 is perpendicular to ~u2 and vice versa. Given a parametri~ of a regular surface in the neighborhood of a point p, we zation X ~ produces a conjugate set of coordinate lines if say that X ~ 1) · X ~ 2 = 0 = dnp (X ~ 2) · X ~1 dnp (X ~ One can state this alternatively for all (x1 , x2 ) in the domain of X. as ~2 = 0 = N ~2 · X ~ 1, ~1 · X N

256

7. The Fundamental Equations of Surfaces

~ produces a set of conjugate coordinate lines if and only if L12 = so X ~ 12 reduces to 0. In this situation, Gauss’s equation for X ~ 12 = Γ112 X ~ 1 + Γ212 X ~ 2, X

(7.9)

~ has a conjugate set of coordinate lines if so the parametrization X and only if Equation (7.9) holds. In addition, Equation (7.9) encompasses three independent linear differential equations in the rectangular coordinate functions of the parametrization. Therefore, given functions P (x1 , x2 ) and Q(x1 , x2 ), any three linearly independent solutions to the equation ∂2f ∂f ∂f − P (x1 , x2 ) 1 − Q(x1 , x2 ) 2 = 0, 1 2 ∂x ∂x ∂x ∂x

(7.10)

when taken as the coordinate functions, give a parametrization of a surface with conjugate coordinate lines in which Γ112 = P (x1 , x2 ) and Γ212 = Q(x1 , x2 ). Equation (7.10) is, in general, not simple to ~ does not appear in Equation (7.10), the coordinate solve but since N function solutions are independent of each other. Example 7.1.7 (Intrinsic Geometry). Since the Christoffel symbols are

defined using the metric coefficients and their partial derivatives, these functions are intrinsic. We can use the metric coefficients g11 (u, v) = 1/v 2 , g12 (u, v) = 0 = g21 (u, v), and g22 (u, v) = 1/v 2 from Example 6.1.9 to compute the Christoffel symbols for the points in the Poincar´e upper half-plane. Note that 0=

∂g12 ∂g22 ∂g11 = = ∂u ∂u ∂u

since the coefficients only depend on v. Also ∂g11 ∂g22 2 ∂g12 = = − 3 and = 0. ∂v ∂v v ∂v From this information we can calculate the Christoffel symbols for this metric.

7.1. Gauss’s Equations and the Christoffel Symbols

257

Using Equation (7.2) we find [11, 1] = 0, 1 , v3 1 [21, 1] = − 3 , v

[12, 1] = −

[22, 1] = 0,

[11, 2] =

1 , v3

[12, 2] = 0, [21, 2] = 0, [22, 2] = −

1 . v3

Then using Equation (7.2), we get Γ111 = 0,

Γ211 = 1/v,

Γ112 = −1/v,

Γ212 = 0,

Γ121 = −1/v,

Γ221 = 0,

Γ122 = 0,

Γ222 = −1/v.

Problems 7.1.1. Fill in the details of the proof for Proposition 7.1.5. 7.1.2. Calculate the Christoffel symbols for the torus parametrized by ~ X(u, v) = ((a + b cos v) cos u, (a + b cos v) sin u, b sin v) , where we assume that b < a. 7.1.3. Calculate the Christoffel symbols for functions graphs, i.e., surfaces ~ parametrized by X(u, v) = (u, v, f (u, v)), where f is a function from U ⊂ R2 to R. 7.1.4. Calculate the Christoffel symbols for surfaces of revolution (see Problem 5.2.7). 7.1.5. Calculate the Christoffel symbols for the pseudosphere (see Example 6.6.6). 7.1.6. Tubes. Let α ~ (t) be a regular space curve and let r be small enough that the tube ~ u) = α ~ X(t, ~ (t) + (r cos u)P~ (t) + (r sin u)B(t) ~ is a regular surface. Calculate the Christoffel symbols for the tube X.

258

7. The Fundamental Equations of Surfaces

~ of a neighborhood on a surface S is 7.1.7. Prove that a parametrization X such that all the coordinate lines x2 = c are asymptotic lines if and only if L11 = 0. 7.1.8. Defining the function D(x1 , x2 ) as D2 = det(gij ), show that ∂ (ln D) = Γ111 + Γ212 ∂x1

and

∂ (ln D) = Γ112 + Γ222 . ∂x2

7.1.9. Suppose that S is a surface with a parametrization that satisfies Γ112 = Γ212 = 0. Prove that S is a translation surface. 7.1.10. Using D as defined in Problem 7.1.8, calling θ(x1 , x2 ) the angle between the coordinate lines, prove that ∂θ D 2 D 2 =− Γ − Γ ∂x1 g11 11 g22 11

and

g22 Γ211 + g11 Γ112 = 0

and

∂θ D 2 D 1 =− Γ − Γ . ∂x2 g11 12 g22 22 (7.11) Show that if the parametrization is orthogonal, i.e., g12 = 0, Equation (7.11) becomes g22 Γ212 + g11 Γ122 = 0.

7.1.11. As in Example 7.1.7, calculate the Christoffel symbols of the first and second kind for a general metric of the form g11 (u, v) = f (v), g22 (u, v) = f (v), and g12 (u, v) = 0.

7.2 Codazzi Equations and the Theorema Egregium Gauss formula in Equation (7.1) define expressions for any second ~ ij in terms of the coefficients of the first and second funderivative X ~ 2, N ~ }. We remind the reader ~ 1, X damental forms and the basis {X that the symmetry of the dot product imposes gij = gji . Further~ ji , we also deduced that Lij = Lji . ~ ij = X more, from X Definition 5.2.10 of a regular surface imposes the condition that ~ of a coordinate all the higher derivatives of any parametrization X patch be continuous. Therefore, for a regular parametrization, the order in which one takes derivatives with respect to given variables is irrelevant. In particular, for the mixed third derivatives, we have ~ 112 = X ~ 121 and X ~ 221 = X ~ 122 , which can be listed in the following X slightly more suggestive manner: ~ 11 ~ 12 ∂X ∂X = ∂x2 ∂x1

and

~ 22 ~ 12 ∂X ∂X = . ∂x1 ∂x2

7.2. Codazzi Equations and the Theorema Egregium

259

Consequently, because of the nontrivial expressions in Gauss’s equations, we obtain ∂ ∂x2 ∂ ∂x2

  ~ = ∂ ~ 1 + Γ211 X ~ 2 + L11 N Γ111 X ∂x1   ~ 1 + Γ212 X ~ 2 + L12 N ~ = ∂ Γ112 X ∂x1

  ~ , ~ 1 + Γ212 X ~ 2 + L12 N Γ112 X   ~ 1 + Γ222 X ~ 2 + L22 N ~ . Γ122 X (7.12)

It is natural to equate the basis components of each equation. However, when we take the partial derivatives, we will obtain ex~ and derivatives of N ~ pressions involving second derivatives of X ~ ~ ~ that we need to put back into the usual {X1 , X2 , N } basis using the Weingarten equations given in Equation (6.26) or Gauss’s formula in Equation (7.1). Doing this transforms Equation (7.12) into six distinct equations from which we deduce two significant theorems in the theory of surfaces. Theorem 7.2.1 (Codazzi Equations). For any parametrized surface of

class C 3 , the following hold: ∂L11 ∂L12 − = L11 Γ112 + L12 Γ212 − L12 Γ111 − L22 Γ211 , ∂x2 ∂x1 ∂L12 ∂L22 − = L11 Γ122 + L12 Γ222 − L12 Γ112 − L22 Γ212 . ∂x2 ∂x1

(7.13)

We can summarize these two equations in one by ∂Lij ∂Lil − Γkil Lkj = − Γkij Lkl l ∂xj ∂x for all 1 ≤ i, j, l ≤ 2, where Einstein summation is implied. ~ coefficient functions in both equations from Proof: Equate the N Equation (7.12).  The classical formulation of the above relationships in Equation (7.13) are collectively called the Codazzi equations or sometimes the Mainardi-Codazzi equations. These equations present a relationship that must hold between the coefficients of the first and second fundamental forms that must hold for all regular surfaces. They are a

260

7. The Fundamental Equations of Surfaces

crucial part of the Fundamental Theorem of Surface Theory that we will present in Section 7.3. The following theorem is another result that stems from equating ~ 2 in Equation (7.12). ~ 1 and X components along X Theorem 7.2.2 (Theorema Egregium). The Gaussian curvature of a surface is an intrinsic property of the surface, that is, it depends only on the coefficients of the metric tensor and higher derivatives thereof.

~ 2 coefficient functions in the first equa~ 1 and X Proof: Equating the X tion of Equation (7.12), we obtain ∂Γ111 ∂Γ112 1 1 2 1 1 + Γ Γ + Γ Γ + L a = + Γ112 Γ111 + Γ212 Γ112 + L12 a11 , 11 2 11 12 11 22 ∂x2 ∂x1 ∂Γ211 ∂Γ212 1 2 2 2 2 + Γ Γ + Γ Γ + L a = + Γ112 Γ211 + Γ212 Γ212 + L12 a21 . 11 11 12 11 22 2 ∂x2 ∂x1 P Since aij = − 2l=1 Ljl g li , where (g ij ) = (gij )−1 , after some simplification, we can write these two equations as g21 −g11

L11 L22 − (L12 )2 ∂Γ112 ∂Γ111 = − + Γ212 Γ112 − Γ211 Γ122 , det g ∂x1 ∂x2

(7.14)

L11 L22 − (L12 )2 ∂Γ212 ∂Γ211 = − + Γ112 Γ211 + Γ212 Γ212 − Γ111 Γ212 − Γ211 Γ222 . det g ∂x1 ∂x2

(7.15)

Since K = det(Lij )/ det(gij ), we can use either one of these equations to obtain a formula for K in terms of the function gij and Γijk . However, since the Christoffel symbols Γijk are themselves determined by the coefficients of the metric tensor, then these equations provide formulas for the Gaussian curvature exclusively in terms of the first fundamental form.  Gauss coined the name “Theorema Egregium,” which means “an excellent theorem,” and indeed this result should seem rather surprising. A priori, the Gaussian curvature depends on the components of the first fundamental form and the second fundamental form. After all, the Gaussian curvature at a point p on the surface is K = det(dnp ), where dnp is the differential of the Gauss

7.2. Codazzi Equations and the Theorema Egregium

261

map. Since the coefficients of the matrix dnp with respect to the ~ 2 } involve the Lij coefficients, one would naturally ex~ 1, X basis {X pect that we would need the second fundamental form to calculate K. However, the Theorema Egregium shows that only the coefficient functions of the first fundamental form are necessary. In the study of surfaces, we have described a geometric quantity as a “local” or “global property” of a curve or surface if it does not change under the orientation and position of the curve or surface in space. On the other hand, topology studies properties of point sets that are invariant under any homeomorphism – a bijective function that is continuous in both directions. The concept of an intrinsic property lies somewhere between these two extremes. Any bijective function between two regular surfaces preserving the metric tensor defines an isometry between the two surfaces. Then intrinsic geometry will treat these two surfaces as the same. In the above proof, Equations (7.14) and (7.15) motivate the following definition of the Riemann symbols: l Rijk

∂Γljk ∂Γlik l m l = − + Γm ik Γmj − Γjk Γmi . ∂xj ∂xi

(7.16)

As it turns out (see Problem 7.2.3), the Riemann symbols form the components of a (1, 3)-tensor. Then, a closely associated tensor denoted by m Rijkl = Rijk gml (7.17) has the interesting property that R1212 = det(Lij ), and hence, the Gaussian curvature of a surface at a point is given by the component function R1212 K= . (7.18) det(gij ) l is called the Riemann The tensor associated to the components Rijk curvature tensor and plays an important role in the analysis on manifolds.

Example 7.2.3 (Cone). As a simple example, consider the right circular cone defined by the parametrization

~ X(u, v) = (v cos u, v sin u, v),

262

7. The Fundamental Equations of Surfaces

where u ∈ [0, 2π] and v > 0, which we know, as a developable surface, has Gaussian curvature K = 0 everywhere. Simple calculations give, for the metric tensor, g11 = v 2 ,

g12 = 0,

g22 = 2.

For the Christoffel symbols, one obtains 1 Γ211 = − v, 2

Γ112 = Γ121 =

1 v

with all the other symbols of the second kind Γijk = 0. With these data, an application of Equation (7.16) shows that all but two of the Riemann symbols vanish simply because all the terms vanish. However, for the two nontrivial terms, one calculates ∂Γ112 ∂Γ122 1 1 − + Γ112 Γ112 + Γ212 Γ122 − Γ122 Γ111 − Γ222 Γ121 = − 2 + 2 = 0, ∂v ∂u v v    2 2 1 1 ∂Γ11 ∂Γ12 1 − + Γ111 Γ212 + Γ211 Γ222 − Γ121 Γ211 − Γ221 Γ221 = − − − v = 0. = ∂v ∂u 2 v 2

1 = R122 2 R121

Consequently, though the Christoffel symbols are not identically 0, the Riemann curvature tensor is identically 0. One finds then that R1212 = 0 is a function of u and v, and hence, using Equation (7.18), one recovers the fact that the cone has Gaussian curvature identically 0 everywhere. The Theorema Egregium (Theorem 7.2.2) excited the mathematics community in the middle of the 19th century and sparked a search for a variety of alternative formulations for the Gaussian curvature of a surface as a function of the gij metric coefficients. We leave a few such formulas for K for the exercises, but we present a few here. ~ ij · N ~ , det(gij ) = kX ~u × X ~ v k2 , and N ~ = Recall that Lij = X ~ ~ ~ ~ Xu × Xv /kXu × Xv k. From these facts, one can easily see that ~ vv X ~ uv X ~ uX ~ v )(X ~ uX ~ v ) − (X ~ uX ~ v )2 ~ uu X (X L11 L22 − L212 = , det(gij ) det(gij )2 (7.19) ~B ~ C) ~ is the triple-vector product in R3 . However, the triplewhere (A vector product is a determinant and we know that for all matrices M1 K=

7.2. Codazzi Equations and the Theorema Egregium

263

and M2 , one has det(M1 M2 ) = det(M1 ) det(M2 ), and so Equation (7.19) becomes   X X ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ · X · X · X · X · X · X X X X X uv uu vv uu u uu v uv uv u uv v 1  ~ ~ vv X ~u X ~v − X ~ uv X ~u X ~ v  ~u · X ~u · X ~u · X ~u · X ~u · X K= .  Xu · X 2 ~ det(gij ) ~ vv X ~u X ~ v X ~ uv X ~u X ~ v ~v · X ~v · X ~v · X ~v · X ~v · X Xv · X (7.20) This equation involves only the metric tensor coefficients and Γijk ~ uu ·X ~ vv and X ~ uv ·X ~ uv . At first symbols, except in the inner products X glance, this problem seems insurmountable, but if one performs the Laplace expansion on the determinants along the first row, factors terms involving det(gij ), and collects into determinants again, one can show that Equation (7.20) is equivalent to K=   X ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ 0 X · X − X · X · X · X · X · X X X X uu vv uv uv uu u uu v uv u uv v 1  ~ vv ~u X ~ v − X ~ uv X ~u X ~ v  ~u · X ~u · X ~u · X ~u · X ~u · X ~u · X X X .  2 ~ det(gij ) ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ Xv · Xvv Xv · Xu Xv · Xv Xv · Xuv Xv · Xu Xv · Xv ~ uu · X ~ vv or The value in this is that though one cannot express X ~ ~ Xuv · Xuv using only the metric coefficients, it can be shown (Problem 7.2.1) that ~ vv − X ~ uv · X ~ uv = − 1 g11,22 + g12,12 − 1 g22,11 , ~ uu · X X 2 2 where by gij,kl we mean the function

(7.21)

∂ 2 gij . ∂xk ∂xl This produces the following formula for K that is appealing on the grounds that it illustrates symmetry of the metric tensor coefficients in determining K:  1 − g11,22 + g12,12 − 1 g11,22 1 g11,1 g12,1 − 1 g11,2 0 2 2 2 2 1 1 1 − g11,2  g − g g g K= 22,1 11 12 12,2 2 12 det(gij )2 1 g22,1 g g g 22,2 21 22 2 2

1 2 g11,2 g11

g21

1 2 g22,1 g12  .

g22 (7.22)

264

7. The Fundamental Equations of Surfaces

In the situation where one considers an orthogonal parametrization of a neighborhood of a surface, namely, when g12 = 0, Equation (7.22) takes on the particularly interesting form of  √  √  1 ∂ g22 1 ∂ g11 ∂ + . √ √ g11 g22 g11 ∂x1 ∂x2 g22 ∂x2 (7.23) By historical habit, Equations (7.22) and (7.23) are referred to as Gauss’s equations. We now prove a theorem that we could have introduced much earlier, but it arises in part as an application of the Codazzi equations. K = −√

1



∂ ∂x1



Definition 7.2.4. A set S ⊂ Rn is called path-connected if for any pair

of points p, q ∈ S, there exists a continuous path (curve) α : [0, 1] → Rn , with α(0) = p, α(1) = q, and α([0, 1]) ⊂ S. Proposition 7.2.5. If S is a path-connected regular surface in which all

its points are umbilical, then S is either contained in a plane or in a sphere. ~ : U → R3 be the Proof: Let U ⊂ R2 be an open set, and let X ~ ) of S. Since all points parametrization of a neighborhood V = X(U of S are umbilical, then the eigenvalues of dnp are equal to a value λ that a priori is a function of the coordinates (u, v) of the patch V . We first show that λ(u, v) is a constant over U . By Proposition 6.5.4, there always exist two linearly independent eigenvectors to dnp , so dnp is always diagonalizable. Thus, at any point p of S, the whole tangent plane Tp S is the eigenspace of the eigenvalue λ. Thus, in particular we have ~ u ) = λX ~ u, ~ u = dnp (X N ~ v = dnp (X ~ v ) = λX ~ v. N ~ vu , a fact that is equivalent to the Codazzi equations ~ uv = N Since N (see Problem 7.2.2), we have ~ uv = λu X ~ vu . ~ u + λX ~ v + λX λv X

7.2. Codazzi Equations and the Theorema Egregium

~ u and X ~ v are linearly independent, we deduce that λu = Since X λv = 0, and therefore λ is a constant function. ~u = N ~ v = ~0, so the unit normal vector N ~ is a If λ = 0, then N ~ constant vector N0 over the domain U . Then ∂ ~ ~ ~u · N ~ 0 + ~0 = ~0 (X · N 0 ) = X ∂u ∂ ~ ~ 0 ) = ~0, which shows that X ~ ·N ~ 0 = ~0, and similarly for ∂v (X · N ~ ) lies in the plane through p which further proves that V = X(U ~ perpendicular to N0 . On the other hand, if λ 6= 0, then consider the vector function

1~ ~ (u, v) = X(u, ~ (u, v). Y v) − N λ ~ (u, v) By a similar calculation as above, we see that the function Y is a constant vector. Then we deduce that

1 1 ~ ~k= ~ kX(u, v) − Y ,

N

= λ |λ| ~ ) lies on a sphere of center Y ~ and of which shows that V = X(U 1 radius |λ| . So far, the above proof has established that any parametrization of a neighborhood V of S either lies on a sphere or lies in a plane but not that all of S necessarily lies on a sphere or in a plane. The assumption that S is path-connected extends the result to the whole surface as follows. By Example A.1.10, if V1 and V2 are two coordinate patches of S with respective coordinate systems (x1 , x2 ) and (¯ x1 , x ¯2 ) such that V1 ∩ V2 6= ∅, then on V1 ∩ V2 the coefficients of the Gauss map with respect to the coordinate systems are related by a ¯ij =

∂x ¯i ∂xl k a . ¯j l ∂xk ∂ x

This is tantamount to saying that the matrices for the Gauss map relative to different coordinate systems are similar matrices. Hence, if V1 and V2 are two overlapping coordinate patches, then the eigenvalue λ is constant over V1 ∪ V2 , and then all of V1 ∪ V2 either lies in a plane or on a sphere.

265

266

7. The Fundamental Equations of Surfaces

Since S is path-connected, given any two points p and q on S, there exists a path α : [0, 1] → S, with α(0) = p and α(1) = q. From basic facts in topology, we know that α([0, 1]) is a compact set and that any open cover of a compact set has a finite subcover. Thus, given a collection of coordinate patches that cover α([0, 1]), then only a finite subcollection of these coordinate patches is necessary to cover α([0, 1]). Call this collection {V1 , V2 , . . . , Vr }. By the reasoning in the above paragraph, we see that V1 ∪ V2 ∪ · · · ∪ Vr lies either in a plane or on a sphere. Thus all points of S either lie in the same plane or lie on the same sphere. 

Problems 7.2.1. Prove Equation (7.21). 7.2.2. Show that one can obtain the Codazzi equations from the equation ~ 12 = N ~ 21 and by using the Gauss equations in Equation (7.1). N 7.2.3. Prove that the Riemann symbols defined in Equation (7.16) form the components of a (1, 3)-tensor. [Requires Appendix A.] 7.2.4. Prove Equation (7.18) and the claim that supports it. 7.2.5. Suppose that g12 = L12 = 0 (i.e., the coordinate lines are lines of curvature), so that the principal curvatures satisfy κ1 =

L11 g11

and

κ2 =

L22 . g22

Prove that the Codazzi equations are equivalent to ∂κ1 g11,2 = (κ2 − κ1 ) ∂x2 2g11 where by g11,2 we mean

and

∂κ2 g22,1 = (κ1 − κ2 ), ∂x1 2g22

∂g11 and similarly for g22,1 . ∂x2

l 7.2.6. Prove that the Riemann symbols Rijk defined in Equation (7.16) are l antisymmetric in the indices {i, j}. Conclude that for surfaces, Rijk represent at most four distinct functions.

7.2.7. Prove Equation (7.23).

7.2. Codazzi Equations and the Theorema Egregium

7.2.8. Prove the following formula by Blaschke for the Gaussian curvature: g11 g12 g22 1 g11,1 g12,1 g22,1 K =− 4 det(gij )2 g11,2 g12,2 g22,2 ! !! ∂ ∂ g22,1 − g12,2 g12,1 − g11,2 1 p p − . − p ∂x2 2 det(gij ) ∂x1 det(gij ) det(gij ) 7.2.9. Consider the two surfaces parametrized by ~ X(u, v) = (v cos u, v sin u, ln v), ~ (u, v) = (v cos u, v sin u, u). Y Prove that these two surfaces have equal Gaussian curvature functions over the same domain but that they do not possess the same metric tensor. (This gives an example of two surfaces with given coordinate systems over which different metric tensor components lead to the same Gaussian curvature function.) 7.2.10. Since the Gaussian curvature is intrinsic, depending only on the metric coefficients and their derivatives, compute that Gaussian curvature for the metric with g11 (u, v) = g22 (u, v) = 1/v 2 and g12 (u, v) = 0 on the Poincar´e upper half-plane, as in Example 6.1.9 and Example 7.1.7. 7.2.11. Carry out the same computation as in the previous problem for the general metric g11 (u, v) = g22 (u, v) = f (v) and g12 (u, v) = 0 on the upper half-plane. [See Problems 6.1.20 and 7.1.11.] 7.2.12. Consider a surface with metric coefficients g11 (u, v) = 1, g12 (u, v) = 0, and g22 (u, v) = f (u, v), for some positive function f . Find the Christoffel symbols of both kinds for this metric and compute the Gaussian curvature K. 7.2.13. Liouville surface. Consider a surface with metric coefficients defined by g11 (u, v) = g22 (u, v) = U (u)+V (v) and g12 (u, v) = 0, where U (u) is a function of u and V (v) is a function of v. Find the Christoffel symbols of both kinds for this metric and compute the Gaussian curvature K. Such a surface is called a Liouville surface. ~ 7.2.14. The coordinate curves of a parametrization X(u, v) form a Tchebysheff net if the lengths of the opposite sides of any quadrilateral formed by them are equal.

267

268

7. The Fundamental Equations of Surfaces

(a) Prove that a necessary and sufficient condition for a parametrization to be a Tchebysheff net is ∂g11 ∂g22 = = 0. ∂v ∂u (b) (ODE) Prove that when a parametrization constitutes a Tchebysheff net, there exists a reparametrization of the coordinate neighborhood so that the new components of the metric tensor are g11 = 1, g12 = cos θ, g22 = 1, where θ is the angle between the coordinate lines at the given point on the surface. (c) Show that in this case, K=−

θuv . sin θ

7.3 The Fundamental Theorem of Surface Theory The original proof of this theorem was provided by Bonnet in 1855 in . In more recent texts, one can find the proof in the Appendix to Chapter 4 in  or in Chapter VI of . We will not provide a complete proof of the Fundamental Theorem of Surface Theory here since it involves solving a system of partial differential equations, but we will sketch the main points behind it. Suppose we consider a regular oriented surface S and coordinate ~ : U → R3 . We have seen that the copatch V parametrized by X efficients (gij ) and (Lij ) of the first and second fundamental forms satisfy det(gij ) > 0 and the Gauss-Codazzi equations. That given these conditions, there exists an essentially unique surface with specified first and second fundamental forms is a profound result, called the Fundamental Theorem of Surface Theory. Theorem 7.3.1. If E, F , G and e, f , g are sufficiently differentiable functions of (u, v) that satisfy the Gauss-Codazzi equations (7.22) ~ and (7.13) and EG − F 2 > 0, then there exists a parametrization X of a regular orientable surface that admits

g11 = E,

g12 = F,

g22 = G,

L11 = e,

L12 = f,

L22 = g.

7.3. The Fundamental Theorem of Surface Theory

Furthermore, this surface is uniquely determined up to its position in space. The setup for the proof is to consider nine functions ξi (u, v), ϕi (u, v), and ψi (u, v) with 1 ≤ i ≤ 3, and think of these functions ~ u, X ~ v , and N ~ so that as the components of the vector functions X ~ ~ ~ Xu = (ξ1 , ξ2 , ξ3 ), Xv = (ϕ1 , ϕ2 , ϕ3 ), and N = (ψ1 , ψ2 , ψ3 ). With this setup, the equations that define Gauss’s and Weingarten equations, namely, Equations (7.1) and (6.26), become the following system of 18 partial differential equations: for i = 1, 2, 3, ∂ξi = Γ111 ξi + Γ211 ϕi + L11 ψi , ∂u ∂ϕi = Γ121 ξi + Γ221 ϕi + L21 ψi , ∂u ∂ψi = a11 ξi + a21 ϕi , ∂u

∂ξi = Γ112 ξi + Γ212 ϕi + L12 ψi , ∂v ∂ϕi = Γ122 ξi + Γ222 ϕi + L22 ψi , ∂v (7.24) ∂ψi = a12 ξi + a22 ϕi . ∂v

In general, when a system of partial differential equations involving n functions ui (x1 , . . . , xm ) has n < m, the solutions may involve not only constants of integration but also unknown functions that can be any continuous function from R to R (or some appropriate interval). However, when n > m, i.e., when there are more functions in the system than there are independent variables, the system may be “overdetermined” and may either have less freedom in its solution set or have no solutions at all. In fact, one cannot expect the above system to have solutions if the mixed partial derivatives ~ = (ψ1 , ψ2 , ψ3 ) are not of ξ~ = (ξ1 , ξ2 , ξ3 ), ϕ ~ = (ϕ1 , ϕ2 , ϕ3 ), and ψ equal. This is usually called the compatibility condition for systems of partial differential equations, and, as we see in the above system, this condition imposes relations between the functions Γijk (u, v) and Ljk (u, v). The key ingredient behind the Fundamental Theorem of Surface Theory is Theorem V in Appendix B of  that, applied to our context, states that if all second derivatives of the Γijk and Ljk functions are continuous and if the compatibility condition holds in Equation (7.24), solutions to the system exist and are unique once values for ~ 0 , v0 ), ϕ ~ 0 , v0 ) are given, where (u0 , v0 ) is a point ξ(u ~ (u0 , v0 ), and ψ(u

269

270

7. The Fundamental Equations of Surfaces

in the common domain of Γijk (u, v) and Ljk (u, v). The compatibility condition required in this theorem is satisfied if and only if the functions g11 = E, g12 = F , g22 = G, L11 = e, L12 = f , and L22 = g satisfy the Gauss-Codazzi equations. Solutions to Equation (7.24) can be chosen in such a way that ~ 0 , v0 ) = E(u0 , v0 ), ϕ ~ 0 , v0 ) · ξ(u ~ (u0 , v0 ) · ϕ ~ (u0 , v0 ) = G(u0 , v0 ), ξ(u ~ 0 , v0 ) · ϕ ~ 0 , v0 ) · ψ(u ~ 0 , v0 ) = 1, ξ(u ~ (u0 , v0 ) = F (u0 , v0 ), ψ(u ~ 0 , v0 ) · ξ(u ~ 0 , v0 ) = 0, ~ 0 , v0 ) · ϕ ψ(u ψ(u ~ (u0 , v0 ) = 0, ~ 0 , v0 ) × ϕ ξ(u ~ (u0 , v0 ) ~ 0 , v0 ). = ψ(u ~ 0 , v0 ) × ϕ kξ(u ~ (u0 , v0 )k

(7.25)

The next step of the proof is to show that, given the above initial conditions, the following equations hold for all (u, v) where the solutions are defined: ~ v) · ξ(u, ~ v) = E(u, v), ξ(u, ~ v) · ϕ ξ(u, ~ (u, v) = F (u, v),

ϕ ~ (u, v) · ϕ ~ (u, v) = G(u, v), ~ v) · ψ(u, ~ v) = 1, ψ(u,

~ v) · ξ(u, ~ v) = 0, ψ(u,

~ v) · ϕ ψ(u, ~ (u, v) = 0,

~ v) × ϕ ξ(u, ~ (u, v) ~ v). = ψ(u, ~ v) × ϕ kξ(u, ~ (u, v)k ~ ϕ ~ we form the new system of From the solutions for ξ, ~ , and ψ, differential equations ( ~ ~ u = ξ, X ~v = ϕ ~. X ~ over appropriate One easily obtains a solution for the function X (u, v) by Z v Z u ~ ~ ϕ ~ (u0 , v) dv. (7.26) ξ(u, v) du + X(u, v) = u0

v0

~ is defined over an open set U ⊂ R2 The resulting vector function X ~ containing (u0 , v0 ), and X parametrizes a regular surface S. By construction, the coefficients of the first fundamental form for this surface are g11 = E, g12 = F, g22 = G.

7.3. The Fundamental Theorem of Surface Theory

271

One then proves that it is also true that the coefficients of the second fundamental form satisfy L11 = e,

L12 = f,

L22 = g.

It remains to be shown that this surface is unique up to a rigid motion in R3 . It is not hard to see that the equalities in Equation (7.25) imposed on the initial conditions still allow one the freedom ˆ 0 , v0 ) = ξ(u ~ 0 , v0 )/kξ(u ~ 0 , v0 )k and the to choose the unit vector ξ(u ~ ˆ vector ψ(u0 , v0 ), which must be perpendicular to ξ(u0 , v0 ). The vectors ˆ 0 , v0 ), ψ(u ~ 0 , v0 ), ξ(u ˆ 0 , v0 ) × ψ(u ~ 0 , v0 ) ξ(u form a positive orthonormal frame, so any two choices allowed by Equation (7.25) differ from each other by a rotation in R3 . Finally, the integration in Equation (7.26) introduces a constant vector of integration. Thus, two solutions to Gauss’s and Weingarten’s equations differ from each other by a rotation and a translation, namely, any rigid motion in R3 .

Problems 7.3.1. (ODE) Consider solutions to Gauss’s and Weingarten’s equations for which the coefficients of the first and second fundamental forms are constant. (a) Let E, F , and G be constants such that EG − F > 0 and view them as constant functions. Prove that the Gauss-Codazzi equations impose Ljk = 0. (b) Prove that all solutions to the Gauss-Weingarten equations in this situation are planes. 7.3.2. (ODE) Find all regular parametrized surfaces that have g11 = 1,

g12 = 0,

g22 = cos2 u,

L11 = 1,

L12 = 0,

L22 = cos2 u.

~ 7.3.3. Does there exist a surface X(u, v) with g11 = 1,

g12 = 0,

g22 = cos2 u,

L11 = cos2 u,

L12 = 0,

L22 = 1?

CHAPTER 8

The Gauss-Bonnet Theorem and Geometry of Geodesics Historians of mathematics often point to Euclid as the inventor, or at least the father in some metaphorical sense, of the synthetic methods of mathematical proofs. Euclid’s Elements, which comprises 13 books, treats a wide variety of topics but focuses heavily on geometry. In his Elements, Euclid presents 23 geometric definitions along with five postulates (or axioms), and in Books I-VI and XI-XIII, he proves around 250 propositions about lines, circles, angles, and ratios of quantities in the plane and in space. Most popular texts about the nature of mathematics agree that Euclid’s geometric methods held a foundational importance in the development of mathematics (see, e.g., [12, pp. 76–77] or [18, p. 7]). Many such texts also retell the story of the discovery in the 19th century of geometries that remain consistent yet do not satisfy the fifth and most debatable of Euclid’s postulates (, [9, pp. 214– 227], , [18, pp. 217–223]). One can readily list elliptical geometry and hyperbolic geometry as examples of such geometries. At this point in this book, with the methods from the theory of curves and surfaces now at our disposal, we stand in a position to study geometry on any regular surface. As we shall see, on a general regular surface the notion of “straightness” is not intuitive, and one might debate whether such a notion should exist at all. Hence, instead of only trying to consider lines and circles, we first study regular curves on a regular surface in general and only later define notions of shortest distance, straightness, and parallelism. Arguably, the most important theorem of this book is the Gauss-Bonnet Theorem, which relates the Gaussian curvature on a region R of the surface to a specific function along the boundary ∂R. One of the central themes of this chapter is to present the Gauss-Bonnet Theorem and

273

274

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

introduce geodesics (curves that generalize the concept of shortest distance and straightness). In the last two sections, we discuss applications of the Gauss-Bonnet Theorem to non-Euclidean geometry.

8.1 Curvatures and Torsion 8.1.1 Natural Frames Throughout this chapter, we let S be a regular surface of class C 2 and let V be an open set of S parametrized by a vector function ~ : U → R3 , where U is an open set in R2 . We consider a curve C X ~ ◦α of class C 2 of the form ~γ = X ~ , where α ~ : I → U is a curve in the ~ with α domain of X, ~ (t) = (u(t), v(t)). Let p = ~γ (t0 ) be a point on the curve and on the surface. Using the theory of curves in R3 , one typically performs calculations on ~ (Since order quantities related to ~γ in the Frenet frame (T~ , P~ , B). matters when discussing coordinates, we use the triple notation as opposed to set notation to describe an ordered basis.) This frame is orthonormal and hence has many nice properties. On the other hand, using the local theory of surfaces, one would typically perform ~ u, X ~ v, N ~ ) reference frame. This frame is not calculations in the (X orthonormal but is often the most practical. If a point p on a surface ~ ), where the vectors ~ei is not an umbilical point, the frame (~e1 , ~e2 , N are principal directions (well defined up to a change in sign), is a natural orthonormal frame associated to S at p. We could construct ~ u /kX ~ u k and yet another orthonormal frame by first taking w ~1 = X ~ then using the Gram-Schmidt algorithm on {w ~ 1 , Xv } to obtain a unit ~ ) is an orthonormal frame. vector w ~ 2 so that (w ~ 1, w ~ 2, N ~ u, X ~ v, N ~ ), the frame Though more geometric in nature than (X ~ (~e1 , ~e2 , N ) does not lend itself well to calculations with specific parametrizations. Studying curves on surfaces, one could choose between these three reference frames, but it turns out that a combination will be most helpful. Borrowing first from surface theory, we use the unit normal vec~ , which is invariant up to a sign under parametrization of S. tor N ~ (t) to be a single-variable vector function We will often consider N ~ (t) = N ~ ◦α over the interval I, by which we mean explicitly N ~ (t). Borrowing from the theory of space curves, we use T~ , the unit tangent vector to the curve C at p, which is also invariant up to a sign

8.1. Curvatures and Torsion

275

~ and T~ are under reparametrization of the curve. By construction, N perpendicular to each other, and we just need to choose a third vector to complete the frame. We cannot use P~ as a third vector in the frame because there is no guarantee that the principal normal vector ~ . In fact, Section 6.5 discusses how the P~ (t) is perpendicular to N ~ and second fundamental form measures the relationship between N ~ P . Consequently, to complete a natural orthonormal frame related to a curve on a surface, we define the new vector function ~ (t) = N ~ (t) × T~ (t). U We remind the reader that in the theory of plane curves, the unit ~ plane to a curve at a point is defined as the counterclockwise normal U rotation of the unit tangent T~ . One can rephrase this definition as ~ plane = ~k × T~ . U Therefore, our definition of a normal vector in the theory of plane curves matches the above definition if one considers the xy-plane a surface in R3 , with ~k as the unit surface normal vector and, in this ~ plane we now denote as U ~. context, what we called U From now on, for calculations related to a curve on a surface, ~,N ~ ), which is often called the Darboux we will use the frame (T~ , U frame.

8.1.2 Normal Curvature As in the theory of space curves, we begin the study of curves on surfaces with the derivative T~ 0 = s0 κP~ . Since the principal normal vector is perpendicular to T~ , this vector T~ 0 , often called the curva~,N ~ ) frame, ture vector , is perpendicular to T~ , and thus, in the (T~ , U ~ ~ decomposes into a U component and an N component. (Alternately, the “curvature vector” sometimes refers to just κP~ .) In Definition 6.5.1, we already introduced the notion of the normal curvature κn (t) ~ . We now define an additional funcas the component of T~ 0 along N tion κg : I → R such that ~ + s0 (t)κn (t)N ~ T~ 0 = s0 (t)κ(t)P~ = s0 (t)κg (t)U for all t ∈ I. We call κg the geodesic curvature of C at p on S.

(8.1)

276

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

~ , T~ , U ~ ) is an orthonormal frame, we can already provide Since (N one way to calculate these curvatures, namely, using ~ κg = κP~ · U

and

~. κn = κP~ · N

(8.2)

If one is given parametric equations for the curve ~γ and the surface ~ then Equation (8.2) is often the most direct way of calculating X, κn . If we are given the second fundamental form to the surface S, there is an alternative way to calculate κn . We remind the reader of Proposition 6.5.2, which relates the second fundamental form of S to the normal curvature by κn (t0 ) = IIp (T~ (t0 )). Therefore, the second fundamental form of a unit vector T~ is the component of a curve’s curvature vector T~ 0 in the normal direction. If T~ (t0 ) makes an angle θ with ~e1 , the principal direction associated to the maximum principal curvature, then by Euler’s curvature formula in Equation (6.30), κn (t0 ) = κ1 cos2 θ + κ2 sin2 θ. We remind the reader that T~ at a point p on a regular curve C may change sign under a reparametrization and thus the principal ~ , which is normal P~ may as well. For surfaces, the unit normal N calculated by ~ ~ ~ = Xu × Xv , N ~u × X ~ vk kX may change sign under a reparametrization. Thus, in Equation (8.2) one notices that κn and κg are unique only up to a possible sign change under a reparametrization of the curve or of the surface. On the other hand, none of these vectors changes direction under positively oriented reparametrizations. ~,N ~) Figure 8.1 illustrates the principal normal P~ in the (T~ , U reference frame of a circle on the surface of a sphere. In this figure, however, since the surface is a sphere, all the points are umbilical, and there does not exist a unique basis {~e1 , ~e2 } of principal directions (even up to sign).

8.1. Curvatures and Torsion

277

~ U P~

T~

~ N ~ , T~ , U ~ , and P~ of a curve on a surface. Figure 8.1. N

As an application of Euler’s formula in the context of curves on surfaces, let us consider any two curves ~γ1 and ~γ2 on the surface S, each parametrized by arc length in such a way that they intersect orthogonally at the point p at s = s0 . If we write ~γ10 (s0 ) = (cos θ)~e1 + (sin θ)~e2 , then ~γ20 (s0 ) = ±((sin θ)~e1 − (cos θ)~e2 ) (1)

(2)

for either − or + signs as a choice on ±. If we denote κn and κn as the normal curvatures of ~γ1 and ~γ2 , respectively, at p, then one obtains the interesting fact that (2) 2 2 2 2 κ(1) n + κn = cos θκ1 + sin θκ2 + sin θκ1 + cos θκ2

= κ1 + κ2 = 2H. In other words, for any two orthogonal unit tangent directions ~v1 and ~v2 at a point p, the average of the associated normal curvatures is equal to the mean curvature and does not depend on the particular directions of ~v1 and ~v2 . ~ , and ~ = 0, it follows that N ~ 0 · T~ = −T~ 0 · N Note that since T~ · N hence, we deduce that ~ 0 · T~ = −s0 (t)κn (t). N

278

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

We remind the reader that an asymptotic curve on a surface is a curve that satisfies κn (t) = 0 (see Definition 6.5.11) at all points on the curve. Geometrically, this means that an asymptotic curve is a ~ 0 has no T~ component or, vice versa, a curve on curve on which N 0 ~ component. Also, from an intuitive perspective, which T~ has no N though we introduced κn as a measure of how much T~ changes in the ~ , −κn gives a measure of how much N ~ changes normal direction N in the tangent direction T~ .

8.1.3 Geodesic Curvature The geodesic curvature κg of a curve on a surface corresponds to the component of the curvature vector occurring in the tangent plane. Equation (8.2) states this relationship as ~, κg = κP~ · U which is equivalent to ~ = T~ 0 · (N ~ × T~ ) = (T~ 0 N ~ T~ ) = (T~ T~ 0 N ~ ), s0 (t)κg = T~ 0 · U

(8.3)

where (~u~v w) ~ is the triple-vector product in R3 . This already provides a formula to calculate κg (t) in specific situations. It turns out, however, that the geodesic curvature is an intrinsic quantity, a fact that we show now. ~ ◦~ Let us write α ~ (t) = (u(t), v(t)) and ~γ = X α for the curve on the surface. (To simplify the expression of certain formulas as we did in Chapter 7, we will refer to the coordinates u and v as x1 and x2 , and we will use the expressions x˙ 1 , x˙ 2 , x ¨1 , x ¨2 to refer to the derivatives of u0 , v 0 , u00 , v 00 .) For the tangent vector, we have ~ u u0 (t) + X ~ v v 0 (t) = s0 (t)T~ = X

2 X

~ i. x˙ i X

(8.4)

i=1

Taking another derivative of this expression, we get s00 T~ + s0 T~ 0 ~ uu (u0 )2 + 2X ~ uv u0 v 0 + X ~ vv (v 0 )2 + X ~ u u00 + X ~ v v 00 =X =

2 2 X X i=1 j=1

~ ij + x˙ i x˙ j X

2 X k=1

~ k. x ¨k X

(8.5)

8.1. Curvatures and Torsion

279

~. ~ ) can also be written as (T~ × T~ 0 )· N The triple-vector product (T~ T~ 0 N ~ ~ Therefore, since T × T = ~0, taking the cross product of the two expressions in Equations (8.4) and (8.5) and then taking the dot ~ leads to product with N h ~ uu )(u0 )3 + (2X ~u × X ~ uv + X ~v × X ~ uu )(u0 )2 v 0 ~ ) = (X ~u × X (s0 )2 (T~ T~ 0 N i ~ ~u × X ~ vv + 2X ~v × X ~ uv )u0 (v 0 )2 + (X ~v × X ~ vv )(v 0 )3 · N + (X ~ v) · N ~ (u0 v 00 − u00 v 0 ). ~u × X + (X

(8.6)

It is possible to express all the coefficients of the terms (u0 )3 , (u0 )2 v 0 , and so forth using the metric tensor coefficients gij or their derivatives. For example, using the result of Problem 3.1.6, ~v ~ ×X ~u × X ~ uu ) · N ~ = (X ~u × X ~ uu ) · X pu (X det(g) ~ u )(X ~ uu · X ~ v ) − (X ~u · X ~ v )(X ~ uu · X ~ u) ~u · X (X p = det(g) g11 [11, 2] − g12 [11, 1] p = det(g) det(g)g 22 [11, 2] + det(g)g 21 [11, 1] p det(g) p 2 = Γ11 det(g). =

Repeating the calculations for all p ~ uu ) · N ~ = Γ2 det(g), ~u × X (X 11 p ~ uv ) · N ~ = Γ212 det(g), ~u × X (X p ~ vv ) · N ~ = Γ2 det(g), ~u × X (X 22

the relevant terms, we get p ~v × X ~ uu ) · N ~ = −Γ1 det(g), (X 11 p ~v × X ~ uv ) · N ~ = −Γ112 det(g), (X p ~v × X ~ vv ) · N ~ = −Γ1 det(g). (X 22

~ v) · N ~ = ~u ×X Putting these expressions, along with (X Equation (8.6) and using Equation (8.3) gives

p

det(g), into

(s0 )3 κg = Γ211 (u0 )3 + (2Γ212 − Γ111 )(u0 )2 v 0 + (Γ222 − 2Γ112 )u0 (v 0 )2 p −Γ122 (v 0 )3 + u0 v 00 − u00 v 0 g11 g22 − (g12 )2 . (8.7)

280

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

This shows that, as opposed to the normal curvature, the geodesic curvature is an intrinsic quantity, depending only on the metric tensor and the parametric equations α ~ (t) = (u(t), v(t)). Though complete, Equation (8.7) is not written as concisely as it could be using tensor notation. We define the permutation symbols   +1, if h1 h2 · · · hn is an even permutation of 1, 2, . . . , n, εh1 h2 ···hn = −1, if h1 h2 · · · hn is an odd permutation of 1, 2, . . . , n,   0, if h1 h2 · · · hn is not a permutation of 1, 2, . . . , n. (See also Section A.1 for more details on this tensor notation.) We shall use this symbol in the simple case with n = 2, so with only two indices. Then our indices satisfy 1 ≤ i, j ≤ 2 and εii = 0, ε12 = 1, and ε21 = −1. We can now summarize (8.7) as ! p 2 2 X X det(g) κg = εil Γljk x˙ i x˙ j x˙ k + εij x˙ i x ¨j (s0 )3 i,j,k,l=1

i,j=1

or, in other words, p κg =

 det(g)  l i j k i j ε , Γ x ˙ x ˙ x ˙ + ε x ˙ x ¨ ij il jk (s0 )3

where we use the Einstein summation notation convention. ~, U ~,N ~ ) frame off the Example 8.1.1 (Plane Curves). We modeled the (T ~ , ~k), viewing the plane as a surface in R3 with normal vecframe (T~ , U tor ~k. Consider then curves in the plane as curves on a surface. Using an orthonormal basis in the plane, i.e., the trivial parametrization ~ X(u, v) = (u, v), we get g11 = g22 = 1 and g12p= 0. In this case, all the Christoffel symbols are 0 and of course det(g) = 1. Then Equation (8.7) gives u0 v 00 − u00 v 0 , κg = (s0 )3 which is precisely the Equation (1.12) we obtained for the curvature of a plane curve in Section 1.3. It is an interesting fact, the proof of which we leave as an exercise to the reader (Problem 8.1.6), that the geodesic curvature at a point p on a curve C on a surface S is equal to the geodesic curvature of the

8.1. Curvatures and Torsion

281

plane curve obtained by projecting C orthogonally onto Tp S. From an intuitive perspective, this property indicates why the geodesic curvature should be an intrinsic quantity. Intrinsic properties depend on the metric tensor, which is an inner product on the tangent space Tp S, so any quantity that measures something within the tangent space is usually an intrinsic property.

8.1.4 Geodesic Torsion Similar to how we viewed the Frenet frame in Chapter 3, we consider ~,N ~ ) as a moving frame based at a point ~γ (t). Since the triple (T~ , U ~ ~ ~ T , U , and N , are unit vectors, using the same reasoning as we did to establish Equation (3.5), we can determine that d ~ T dt

~ U

  ~ = T~ N

~ U

 ~ A(t), N

where A(t) is an antisymmetric matrix. We already know most of the coefficient functions in A(t). In Equation (8.1), our definitions of the normal and geodesic curvature gave ~ + s0 (t)κn (t)N ~ T~ 0 = s0 (t)κg (t)U

(8.8)

and using the usual dot product relations among the vectors in an orthonormal basis, namely T~ · T~ = 1, ~ · T~ = 0, N

~ ·U ~ = 1, U ~ ·U ~ = 0, N

~ ·N ~ = 1, N ~ = 0, T~ · U

we deduce the following equalities: ~0 ·U ~ = 0, ~0 ·N ~ = 0, T~ 0 · T~ = 0, U N ~ 0 · T~ = −T~ 0 · N ~, ~0 ·U ~ = −U ~0 ·N ~, ~ = −U ~ 0 · T~ . N N T~ 0 · U Consequently, from Equation (8.8), ~ 0 · T~ = −s0 κn N

and

U~ 0 · T~ = −s0 κg .

Therefore, in order to describe the coefficients of A(t), and thereby ~ 0, N ~ 0 in the (T~ , U ~,N ~ ) frame, we need express the derivatives T~ 0 , U

282

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

to label one more coefficient function. Define the geodesic torsion τg : I → R to be the unique function such that ~ = s0 (t)τg (t) ~0 ·N U

for all t ∈ I.

With this function defined, we can imitate Equation (3.5) for the change of the Frenet frame, and write   0 −κg (t) −κn (t)     d ~ ~ ~ ~ N ~ s0 (t)  κg (t) 0 −τg (t)  . T U N = T~ U dt κn (t) τg (t) 0 The normal and geodesic curvatures both possess fairly intuitive geometric explanations as to what they measure. However, it is harder to say precisely what the geodesic torsion measures. Using the formula ~ 0 = s0 (−κn T~ + τg U ~ ), N we could say that the geodesic torsion measures the rate of change ~ in the U ~ direction, which would mean the rate of change of N ~ of N twisting around the direction of motion along the curve (the direction in the tangent plane perpendicular to T~ ). This intuition is not particularly instructive. In Section 8.4, we study geodesic curves of a surface, which in some sense generalize the notion of straight lines on the surface. Problem 8.4.1 shows that the geodesic torsion is the torsion function of the geodesic curves.

Problems ~ = (R cos u sin v, R sin u sin v, R cos v), with (u, v) ∈ [0, 2π] × 8.1.1. Let X [0, π], be a parametrization for a sphere. Consider the circle on the ~ ϕ0 ), where ϕ0 is a fixed constant. Calcusphere given by ~γ (t) = X(t, late the normal curvature, the geodesic curvature, and the geodesic ~ (see Figure 8.1 for an illustration with ϕ0 = 2π/3). torsion of ~γ on X 8.1.2. Consider the cylinder x2 + y 2 = 1, and consider the curve C on the cylinder obtained by intersecting the cylinder with the plane through the x-axis that makes an angle of θ with the xy-plane. (a) Show that C is an ellipse. (b) Compute the normal curvature, geodesic curvature, and geodesic torsion of C on the cylinder. (c) Is the geodesic torsion an intrinsic quantity?

8.1. Curvatures and Torsion

283

8.1.3. Let ~γ (s) be a curve parametrized by arc length on a surface param~ etrized by X(u, v). Prove that κg

p

det(g) =

∂ ~ ~ ∂ ~ ~ (T · Xv ) − (T · Xu ). ∂u ∂v

~ 8.1.4. Consider the torus parametrized by X(u, v) = ((a+b sin v) cos u, (a+ b sin v) sin u, b cos v), where a > b. The curve ~γ (mt, nt), where m and n are relatively prime, is called the (m, n)-torus knot. Calculate the geodesic curvature of the (m, n)-torus knot on the torus. 8.1.5. Consider the surface that is a function graph z = f (x, y). Calculate the normal curvature, geodesic curvature, and geodesic torsion of a level curve, i.e., a curve of the form f (x, y) = c. ~ be the parametrization for a coordinate patch of a regular 8.1.6. Let X ~ ◦α surface S, and let ~γ = X ~ be the parametrization for a curve C on S. Consider a point p on S, and let C 0 be the orthogonal projection of C onto the tangent plane Tp S. Prove that the geodesic curvature κg at p of C on S is equal to the curvature κg of C 0 at p as a plane curve in Tp S. 8.1.7. Bonnet’s Formula. Suppose that a curve on a surface is given by ϕ(u, v) = C, where C is a constant. Prove that the geodesic curvature is given by      g12 ϕv − g22 ϕu g12 ϕu − g11 ϕv ∂ ∂ 1 κg = p + . ∂v g11 ϕ2v − 2g12 ϕu ϕv + g22 ϕ2u det(g) ∂u g11 ϕ2v − 2g12 ϕu ϕv + g22 ϕ2u 8.1.8. Liouville’s Formula. (Comment: This result is of vital importance in the proof of the Gauss-Bonnet Theorem in the next section.) Let ~ X(u, v) be an orthogonal parametrization of a patch on a regular surface S. Let C be a curve on this patch parametrized by arc ~ length by ~γ (s) = X(u(s), v(s)). Let θ(s) be a function defined along ~ u . Prove that the geodesic C that gives the angle between T~ and X curvature of C is given by κg =

dθ + κ(u) cos θ + κ(v) sin θ, ds

where κ(u) is the geodesic curvature along the u-parameter curve (i.e., v = v0 ) and similarly for κ(v) . ~ 8.1.9. Let S be a regular surface parametrized by X(u, v), and let C be ~ ~ a regular curve on S parametrized by X(t) = X(u(t), v(t)) for t ∈

284

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

[a, b]. Consider the normal tube around C parametrized by ~ ~ (t) + r sin v N ~ (t) ~r (t, v) = X(t) + r cos v U Y for some r > 0. ~ (t, v), and conclude that (a) Find the metric tensor associated to Y det g = (s0 )2 r2 (1 − rκg cos v − rκn sin v)2 . ~r is a (b) Show that if r is small enough but still positive, then Y regular parametrization. ~r is regular, calculate the coefficients Lij of (c) Assuming that Y the second fundamental form of the normal tube around C. Prove that the Gaussian curvature K satisfies s0 (t) (κn (t) sin v + κg (t) cos v). K(t, v) = − p det(g)

8.2 Gauss-Bonnet Theorem, Local Form No course on classical differential geometry is complete without the Gauss-Bonnet Theorem, arguably the most profound theorem in the differential geometry study of surfaces. The Gauss-Bonnet Theorem simultaneously encompasses a total curvature theorem for surfaces, the total geodesic curvature formula for plane curves, and other famous results, such as the sum of angles formula for a triangle in plane, spherical, or hyperbolic geometry. Other applications of the theorem extend much further and lead to deep connections between topological invariants and differential geometric quantities, such as the Gaussian curvature. In his landmark paper , Gauss proved an initial version of what is now called the Gauss-Bonnet Theorem. The form in which we present this theorem was first published by Bonnet in 1848 . Various alternative proofs exist (e.g., ) and a search of the literature turns up a large variety of generalizations (e.g., to polyhedral surfaces and to higher-dimensional manifolds). Despite the far-reaching consequences of the theorem, the difficulty of the proof resides in one essentially topological property of curves on surfaces. When we present this theorem, we will simply provide a reference.

8.2. Gauss-Bonnet Theorem, Local Form

285

The theorem then follows easily from Green’s Theorem for simple closed curves in the plane. First, however, we motivate the Gauss-Bonnet Theorem with the following intuitive example. Example 8.2.1 (The Moldy Potato Chip). We particularly encourage the

reader to consult the demo applet for this example. Consider a region R on a regular surface S such that the boundary curve ∂R is a regular curve that is simple and simply connected (can be shrunk continuously to a point) on the surface S. We now create the “moldy potato chip” as the surface that consists of taking the region R and spreading it out over every possible normal direction by a distance of r, where r is a fixed real number. As the demo shows, the surface of the moldy potato chip (MPC) consists of three pieces: two pieces that are the parallel surfaces of S “above” and “below” R, and one region that consists of a half-tube around ∂R. (In the applet, these portions are colored by magenta and green, respectively.) RR The total Gaussian curvature MPC K dS of the MPC is the same as the area of the portion of the unit sphere covered by the Gauss map of the MPC. Consider this area as the boundary ∂R is shrunk continuously to a point; it must vary continuously with the shrinking of ∂R. However, this area must always be an integral multiple of 4π because the Gauss map of a surface without boundary covers the unit sphere a fixed number of times. The only function that is continuous and discrete is a constant function. In the limit, the moldy surface around a point is simply a sphere whose Gauss map image is the unit sphere with area 4π. Hence, the total curvature of the moldy potato chip is always 4π. However, Problem 6.6.12 showed that if r > 0 is small enough, the total Gaussian curvature for each of the two parallel surfaces RR above and below R is R K dS. Furthermore (and we leave the full details of this until Example 8.2.6), the total Gaussian curvature of R the half-tube around ∂R is 2 ∂R κg ds. Hence, we conclude that ZZ ZZ Z K dS = 4π ⇐⇒ 2 K dS + 2 κg ds = 4π MPC ZZ R Z ∂R ⇐⇒ K dS + κg ds = 2π. (8.9) R

∂R

286

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

The applet for this example colors the magenta and green regions in a lighter shade when the Gaussian curvature of the MPC is positive and in a darker shade when the Gaussian curvature is negative. In this way, the user can see how, even if certain regions of the Gauss map are covered more than once, the signed area cancels portions out so that the total signed area covered is 4π.) In what follows, we work to establish Equation (8.9) rigorously and expand the hypotheses under which it holds, the end result being the celebrated global Gauss-Bonnet Theorem. Furthermore, we provide an intrinsic proof of the Gauss-Bonnet Theorem, which establishes it as long as we have a metric tensor and without the assumption that the surface is in an ambient Euclidean three-space. Certain types of regions on surfaces play an important role in what follows. We call a region R on a regular surface S simple if its boundary can be parametrized by a simple, closed, piecewise regular curve. Using theorems of existence and uniqueness to certain differential equations, there are a variety of ways to see that near every point p on S there exists a neighborhood of p that is parametrized by an orthogonal parametrization. In Section 8.5 we will see such an orthogonal parametrization using what are called geodesic coordinate systems on S. On a regular surface of class C 3 in R3 , we can utilize extrinsic properties to realize an orthogonal parametrization. As discussed in ~ 0 , v0 ) is not an umbilical point of S, then Section 6.5, if p = X(u the lines of curvature provide orthogonal coordinate lines for a parametrization of a neighborhood of p. More precisely, by the theorem on existence and uniqueness of solutions to differential equations, it is possible to find a solution to the first-order differential equation that results from Equation (6.29) with the initial condition of (u0 , v0 ) and for each point, say (u0 + h, v0 ) with −ε < h < ε, along the resulting line of curvature, we can calculate the perpendicular line of curvature as the solution to the other branch resulting from Equation (6.29). This is not a tractable problem for specific surfaces, but what matters for what follows is that at every point p, we can parametrize an open neighborhood of p on S with an orthogonal parametrization.

8.2. Gauss-Bonnet Theorem, Local Form

287

~ : U → R3 Let S be a regular surface of class C 3 , and let X ~ ) in be an orthogonal parametrization of a neighborhood V = X(U S. Consider a regular curve ~γ (s) parametrized by arc length whose image lies in V . Liouville’s Formula (see Problem 8.1.8) states that the geodesic curvature of ~γ (s) satisfies κg (s) =

dϕ + κ(u) cos ϕ + κ(v) sin ϕ, ds

where κ(u) is the geodesic curvature along the u-parameter curve (i.e., v = v0 ) and similarly for κ(v) and ϕ(s) is the angle ~γ 0 (s) makes ~ u . Using Equation (8.7) to calculate κ(u) and κ(v) and writwith X ing cos ϕ and sin ϕ in terms of the metric tensor, one can rewrite Liouville’s Formula as κg (s) =

∂g22 dv ∂g11 du 1 1 dϕ + √ − √ . ds 2 g11 g22 ∂u ds 2 g11 g22 ∂v ds

(8.10)

Now consider a simple region R in V with a boundary ∂R that is parametrized by arc length by a regular curve α ~ (s) defined over ~ the interval [0, `]. Additionally, suppose that α ~ (s) = X(u(s), v(s)) for coordinate functions u(s) and v(s). Finally, suppose that U 0 is ~ 0 ) = R. the subset of U such that X(U Integrating the geodesic curvature around the curve ∂R, we get  Z Z ` Z ` 1 ∂g22 dv ∂g11 du dϕ 1 − √ ds + ds, κg ds = √ 2 g g ∂u ds 2 g g ∂v ds 11 22 11 22 0 0 ds ∂R where ` is the length of C. Applying Green’s Theorem (Theorem 2.1.6) to the first term, we get      Z Z ` ZZ 1 1 1 ∂ ∂g22 ∂g11 dϕ ∂ + du dv + ds. κg ds = √ √ ∂u g11 g22 ∂u ∂v g11 g22 ∂v 0 ds ∂R U0 2 By Equation (7.23), this becomes Z ZZ Z √ κg ds = − K g11 g22 du dv + U0

∂R

0

ZZ κg ds = − ∂R

`

U0

dϕ ds, ds

√ K g11 g22 du dv + 2π.

(8.11)

288

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

The last term in Equation (8.11) is the content of the Theorem of Turning Tangents, proved by H. Hopf in , which we state without proving. Lemma 8.2.2 (Theorem of Turning Tangents, Regular). Let C be a sim-

ple, closed, regular curve on a regular surface S of class C 3 parametrized by α ~ . Let ϕ(s) be defined as above. Then Z dϕ ds = ±2π, C ds where the sign of 2π is positive if the orientation of α ~ is such that ~ the normal vector U (s) to α ~ (s) points into the region enclosed by the curve and negative otherwise. Putting together Equation (8.11) and Lemma 8.2.2 establishes a first local version of the Gauss-Bonnet Theorem. ~ : U → Theorem 8.2.3 (Local Gauss-Bonnet Theorem, Regular). Let X ~ ) of R3 be an orthogonal parametrization of a neighborhood V = X(U an oriented surface S of class C 3 . Let R ⊂ V be a simple region of S, and suppose that the boundary is ∂R = α ~ ([0, `]) for some simple, closed, regular, positively oriented curve α ~ : [0, `] → S of class C 3 parametrized by arc length. Then Z ZZ κg ds + K dS = 2π. ∂R

R

We have called the above formulation of the Gauss-Bonnet Theorem a local version since, as stated, it requires that the region R of S be inside a coordinate neighborhood of S that admits an orthogonal parametrization. Furthermore, the above theorem assumes that ∂R is a regular curve. By Problem 6.1.15, we know that if S is a regular surface of class C 2 , then at every point p ∈ S there exists a neighborhood of p that can be parametrized by a regular orthogonal parametrization. We will first generalize this theorem to include regions whose boundaries satisfy a looser condition than being regular. Then, to obtain a global version of the theorem, we will “piece together” local instances of the above theorem. The above proof of the local Gauss-Bonnet Theorem is almost an intrinsic proof but not quite. In order to use Liouville’s Formula,

8.2. Gauss-Bonnet Theorem, Local Form

289

we needed to refer to an orthogonal parametrization of a patch of the surface. At that point, we chose to refer to a parametrization of a patch that has curvature lines as coordinate lines. However, curvature lines are along the direction of principal curvatures, which are extrinsic properties. Proposition 8.5.5 proves the existence of geodesic coordinate systems on patches of a regular surface. Such coordinate systems do depend entirely on the components of the metric tensor and its derivatives, and they induce orthogonal parametrizations. Citing this proposition that we will encounter later establishes the Gauss-Bonnet Theorem as a purely intrinsic result. We now extend the local Gauss-Bonnet Theorem to the broader class of piecewise regular curves in order to set the scene for the global theorem. In order to present the Gauss-Bonnet Theorem for piecewise regular curves, we need to first establish a few definitions about the geometry of such curves on surfaces. We call a set of points {ti }i∈K in R discrete if inf{|ti − tj | i, j ∈ K and i 6= j} > 0, i.e., if any two distinct points are separated by at least some fixed, positive real number. The indexing set K may be finite, say K = {1, . . . , k}, or, if it is infinite, it may be taken as either the set of nonnegative integers N or the set of integers Z. ~ : I → Rm is regular near t0 ∈ I, i.e., on Suppose that a curve X the interval (t0 − ε, t0 + ε), where ε > 0, and also suppose that it is regular on (t0 − ε, t0 ) and (t0 , t0 + ε). Then we can consider the limits of the unit tangent T~ (t) as t approaches t0 from the right and from the left. Call ~ T~ (t− 0 ) = lim T (t) t→t− 0

and

~ T~ (t+ 0 ) = lim T (t). t→t+ 0

To be precise, we say that a curve α ~ : I → Rn is piecewise regular if • α ~ is continuous over I; ~ is reg• there exists a discrete set of points {ti }i∈K such that α ular over each open interval in the set difference I − {ti }i∈K ;

290

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

~ + • T~ (t− i ) and T (ti ) exist for all ti ∈ {ti }i∈K . A regular curve is also piecewise regular, by virtue of considering the situation in which the set {ti }i∈K is empty. If α ~ is regular over an interval I 0 , then the trace α ~ (I 0 ) is called a regular arc of the curve. ~ + If T~ (t− ~ (ti ) is called a cusp. Otherwise, if i ) = −T (ti ), then a point α − T~ (ti ) and T~ (t+ ) are not collinear, then α ~ (ti ) is called a corner . i We point out that our definition for piecewise regular curves ap~ has plies to any curve in Rm . However we now also assume that α the properties that will interest us for the Gauss-Bonnet Theorem, namely that α ~ : I → R3 is a simple, closed, piecewise regular curve on a regular oriented surface S with orientation n. In this case, I is a closed and bounded interval, and there can be at most a finite number of vertices. Furthermore, we impose the criterion that the curve α ~ traces out the image α ~ (I) in the same direction as the orientation of S. See the surface and curve orientations in Figure 8.2.

α ~ (t1 )

θ1 < 0

θ2 > 0

α ~ (t2 )

Figure 8.2. A regular piecewise curve.

~ (ti ), we define the external angle θi of the vertex For all corners α ~0 + as the angle −π < θi < π swept out from T~ (t− i ) to T (ti ) in the − ~ (ti ) and spanned by T~ 0 (ti ) and T~ 0 (t+ plane through the point α i ) (see Figure 8.2). Note that this external angle may be positive or negative.

8.2. Gauss-Bonnet Theorem, Local Form

291

We are now in a position to approach the local Gauss-Bonnet Theorem for piecewise regular curves. Let S be a regular surface, and let V be a neighborhood of S ~ : U → R3 with U ⊆ R2 . We assume parametrized orthogonally by X ~ : U → V is a homeomorphism. Consider a simple region R in that X the neighborhood V on S such that its boundary ∂R is parametrized by a simple, closed, piecewise regular curve α ~ : I → R3 . Suppose that α ~ is parametrized by arc length so that I = [0, `] and has vertices at s1 < s2 < · · · < sk . Set s0 = 0 and sk+1 = `. Also call Ci the image of α ~ ([si−1 , si ]) for 1 ≤ i ≤ k + 1; these are the regular arcs of ~ ∂R . Suppose, additionally, that α ~ (s) = X(u(s), v(s)). Let U 0 be 0 ~ the subset of U such that R = X(U ). As in the previous case, integrating the geodesic curvature around the curve C, we get k Z X

Z κg ds = ∂R

κg ds =

i=0 Ci k Z si+1 X

=

i=0

k Z X i=0



si

si+1

κg (s) ds

si

1 ∂g22 dv ∂g11 du 1 − √ √ 2 g11 g22 ∂u ds 2 g11 g22 ∂v ds

 ds +

k Z X i=0

si+1

si

dϕ ds. ds

Applying Green’s Theorem (generalized to simple closed piecewise regular curves) to the first term, we get ZZ

Z κg ds = ∂R

U0

1 2



∂ ∂u



1

∂g22 √ g11 g22 ∂u

 +

∂ ∂v



1

∂g11 √ g11 g22 ∂v



By Equation (7.23), this becomes Z

k

ZZ κg ds = −

∂R

U0

X √ K g11 g22 du dv + i=0

Z

si+1

si

dϕ ds, ds

ZZ κg ds = − ∂R

K dS + ~ 0) X(U

k X i=0

(ϕ(si+1 ) − ϕ(si )) .

(8.12)

du dv +

k Z X i=0

si+1

si

dϕ ds. ds

292

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

It remains for us to interpret the last term in Equation (8.12). As before, this is precisely the more general version of the Theorem of Turning Tangents . Lemma 8.2.4 (Theorem of Turning Tangents). Let α ~ be a simple, closed,

piecewise regular curve on a regular surface S of class C 3 . Let α ~ (si ) be the vertices with external angles θi , and let ϕ(s) be as defined above. Then k X

(ϕ(si+1 ) − ϕ(si )) = ±2π −

i=0

k X

θi ,

i=1

where the sign of 2π is positive if the orientation of α ~ is such that ~ (s) to α the normal vector U ~ (s) points into the region enclosed by the curve and negative otherwise. Putting together Equation (8.12) and Lemma 8.2.4 establishes a local version of the Gauss-Bonnet Theorem. ~ : U → R3 be an Theorem 8.2.5 (Local Gauss-Bonnet Theorem). Let X

~ ) of an oriented orthogonal parametrization of a region V = X(U 3 surface S of class C . Let R ⊂ V be a simple region of S, and suppose that the boundary is ∂R = α ~ ([0, `]) for some simple, closed, piecewise regular, positively oriented curve α ~ : [0, `] → S of class C 2 , parametrized by arc length. Let α ~ (si ), with 1 ≤ i ≤ k, be the vertices of ∂R, and let θi be their external angles. Call Ci the regular arcs of ∂R. Then Z

ZZ κg ds +

K dS + R

∂R

k X

θi = 2π.

i=1

Note that in this statement of the theorem, there are k vertices and k + 1 regular arcs because the formulation assumes that α ~ (0) is not a vertex. Consequently, we must remember to interpret the integral on the left as an integral over k + 1 regular arcs as Z κg ds = ∂R

k Z X i=0

Ci

κg ds.

8.2. Gauss-Bonnet Theorem, Local Form

We typically call the more general Theorem 8.2.5 the Local GaussBonnet Theorem and understand that Theorem 8.2.3 is a subcase of this theorem. We illustrate this theorem in a similar vein as the Moldy Potato Chip example but now with a patch of surface bounded by a piecewise regular curve. Example 8.2.6 (The Moldy Patch). The motivating example, Example 8.2.1 with the moldy potato chip, falls just shy of giving an extrinsic (assumes we have/know a normal vector to the surface) proof of the local Gauss-Bonnet Theorem. We provide the details here. Consider a simply connected region R on a regular surface S as described in Theorem 8.2.5. Suppose that R is parametrized by ~ the associated normal ~ : U → R3 for some U ⊂ R2 , and call N X vector. Define the surface Tr as the tubular neighborhood of R with radius r. This means that Tr consists of

1. two pieces for the normal variation to R parametrized respec~ +rN ~ and X ~ −rN ~ over U (which we call respectively tively by X U(+r) and U(−r) ); 2. k half-tubes of radius r around the smooth pieces of the boundary ∂R pointing “away” from the region R; 3. k lunes of spheres (of radius r) at the k vertices of ∂R. Figure 8.3 shows the pieces of Tr for a patch on a torus. We assume from now on that r is small enough so that each of the pieces of Tr is a regular surface. Let K be the Gaussian curvature of S over R, and call KT the Gaussian curvature of the moldy patch Tr . By Proposition 6.6.2, the quantity KT dST is a signed area element of the image on the unit sphere of Tr under its Gauss map. So in calculating the total curvature of the moldy patch, one is adding or subtracting area of the sphere depending on the sign of KT . We now reason why ZZ KT dST = 4π. (8.13) Tr

293

294

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

Figure 8.3. A “moldy patch.”

Since Tr has no boundary, the Gauss map of Tr covers the unit sphere an integer number of times, where we add area for positive curvature and subtract for negative curvature. Thus, the integral in Equation (8.13) must be equal to 4πh, where h is an integer. Suppose that the region R is contractible, which means that it can be shrunk continuously to a point while remaining on the surface. Now if R is a point, then Tr is a sphere of radius r. In this case, the Gauss map for Tr is a bijective map onto the unit sphere, and hence, the integral in Equation (8.13) gives precisely the surface of the unit sphere and so is equal to 4π. However, as one “uncontracts” from a point to R, the integral in Equation (8.13) must vary continuously. However, a continuous function to the set of integer multiples of 4π is a constant function. Though we have not spelled out the whole topological background, this reasoning justifies Equation (8.13). We now break down Equation (8.13) according to various pieces of Tr : the normal variation patches, the half-tubes, and the lunes of spheres. By Problem 6.6.12, if we call K(+r) and K(−r) the respective ~ + rN ~ and X ~ − rN ~ over U , then Gaussian curvatures of X ZZ ZZ ZZ K dS = K(+r) dS + K(−r) dS. 2 U

U(+r)

U(−r)

In Problem 8.1.9, one calculates the Gaussian curvature of the normal tube around a curve on the surface. Consider a regular arc

8.3. Gauss-Bonnet Theorem, Global Form

295

C of ∂R, and assume that it is parametrized by t ∈ [a, b]. As a consequence of Problem 8.1.9, over the normal half-tube HT pointing ~ (i.e., away from the inside of R), the total Gaussian away from U curvature is ZZ Z b Z 3π/2 K dS = −s0 (t) ((sin v)κn (t) + (cos v)κg (t)) dt a

HT

π/2 b

Z =2

Z

0

s (t)κg (t) dt = 2 a

κg ds. C

Finally, we consider the lunes of Tr that are around the vertices of ∂R. Under the image of the Gauss map, the lunes map to the same corresponding lune on the unit sphere. Thus, in Equation (8.13), a lune around a vertex with exterior angle θi contributes (θi /2π)4π = 2θi to the surface of the unit sphere. Combining each of these results, we find that ZZ ZZ k Z k X X KT dST = 2 K dS + 2 κg ds + 2 θi , Tr

U

i=0

Ci

i=1

and the local Gauss-Bonnet Theorem follows immediately from Equation (8.13). In our presentation of the local Gauss-Bonnet Theorem, we did not allow corners to be cusps. The main difficulty lies in deciding whether to assign a value of π or −π to the angle θi of any given cusp in order to retain the validity of the local Gauss-Bonnet Theorem. We can allow for cusps on the boundary ∂R if we employ the following sign convention for the angles of cusps: If the cusp α ~ (ti ) − points into the interior of the closed curve (i.e., α ~ (ti ) points into the interior) then θi = −π; and if the cusp α ~ (ti ) points away from the interior of the closed curve (i.e., α ~ (t− i ) points away from the interior) then θi = π. (See Figure 8.4.) The orientation of the surface matters since it determines the direction of travel around the boundary curve ∂R.)

8.3 Gauss-Bonnet Theorem, Global Form In order to extend the Gauss-Bonnet Theorem to a global presentation (i.e., outside of a single coordinate patch on the surface), we

296

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

θ2 = π

θ1 = −π

Figure 8.4. External angles at cusps.

need to briefly discuss triangulations of surfaces, the classification of orientable surfaces, and the Euler characteristic of regions of surfaces in R3 . These topics are typically considered in the area of topology, but we summarize the results that we need in order to give a full treatment to the global Gauss-Bonnet Theorem without insisting that the reader have mastery of the supporting topology. (The authors include technical details behind these concepts in Appendices A.5 and A.6 in . Otherwise, the interested reader could consult Chapters 6 and 7 of .) In intuitive terms, a triangulation of a surface consists of a network of a finite number of regular curve segments on the surface such that any point on the surface either lies on one of the curves or lies in a region that is bounded by precisely three curve segments. The first picture in Figure 8.5 depicts a triangulation on a torus. As an additional technical requirement, one should be able to continuously deform the surface with its triangulation so that each “triangle” becomes a true triangle without changing the topological nature of the surface. (Compare the two pictures in Figure 8.5.) In a triangulation, a vertex is an endpoint of one of the curve segments on the surface. We call the curve segments edges, and the regions enclosed by edges (the “triangles”) we call faces. An interesting and useful result, first proved by Rado in 1925, is that every regular compact surface admits a triangulation. A result of basic topology is that given a compact regular surface S, the quantity #(vertices) − #(edges) + #(faces)

8.3. Gauss-Bonnet Theorem, Global Form

297

Figure 8.5. Torus triangulation.

is the same regardless of any triangulation of S. This number is called the Euler characteristic of S and is denoted by χ(S). Furthermore, from the definition of triangulation, one can deduce that the Euler characteristic does not change if the surface is deformed continuously (no cutting or pinching). One often restates this last property by saying that the Euler characteristic is a topological invariant. For example, the torus triangulation in Figure 8.5 has 16 vertices, 48 edges, and 32 faces. Thus, the Euler characteristic of the torus is 0. As another example consider the tetrahedron, which is homeomorphic to the sphere. A tetrahedron has four vertices, six edges, and four faces, so its Euler characteristic, and therefore the Euler characteristic of the sphere, is χ = 4 − 6 + 4 = 2. A profound theorem in topology, the Classification Theorem of Surfaces states that every orientable surface without boundary is homeomorphic to a sphere or to a sphere with a finite number of “handles” added to it. Figure 8.6(a) shows a torus while Figure 8.6(b) shows a sphere with one handle added to it. These two surfaces are in fact the same under a continuous deformation, i.e., they are homeomorphic. Figure 8.6(c) depicts a two-holed torus that, in the language of the Classification Theorem of Surfaces, is called a sphere with two handles. It is not hard to show that the Euler characteristic of a sphere with g handles added is χ(S) = 2 − 2g.

(8.14)

298

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

(a) Torus.

(b) Sphere with one handle.

(c) Two-holed torus.

Figure 8.6. Tori.

The notion of the Euler characteristic applies equally well to a surface with boundary as long as the boundary is completely covered by edges and vertices of the triangulation. For example, we encourage the reader to verify that a sphere with a small disk removed has Euler characteristic of 1. We must now discuss orientations on a triangulation. When considering adjacent triangles, we can think of the orientation of a triangle as a direction of travel around the edges. Two adjacent triangles have a compatible orientation if the orientation of the first leads one to travel along the common edge in the opposite direction of the orientation on the second triangle (see Figure 8.7). It turns out that if a surface is orientable, then it is possible to choose an orientation of each triangle in the triangulation such that adjacent triangles have compatible orientations.

(a) compatible

(b) incompatible

8.3. Gauss-Bonnet Theorem, Global Form

299

A compact regular surface S is covered by a finite number of coordinate neighborhoods given by regular parametrizations. In general, if R is a regular region of S, it may not lie entirely in one coordinate patch. However, it is possible to show that not only does a regular region R admit a triangulation, but every regular region R admits a triangulation such that each triangle is contained in a coordinate neighborhood. This comment and two lemmas show that it makes sense to talk about the surface integral over the whole region R ⊂ S. ~ 1 : U1 → R3 and X ~ 2 : U2 → R3 are two Lemma 8.3.1. Suppose that X systems of coordinates of a regular surface S. Call (ui , vi ) the coor~ i , and call g (i) the corresponding metric tensor. Suppose dinates of X ~ ~ 2 (U2 ). Then for any function f : T → R, we that T ⊂ X1 (U1 ) ∩ X have ZZ ZZ q q f (u1 , v1 ) det(g (1) ) du1 dv1 = f (u2 , v2 ) det(g (2) ) du2 dv2 . ~ −1 (T ) X 1

~ −1 (T ) X 2

Proof: By Equation (6.7), one deduces that   ∂(u2 , v2 ) 2 det(g (2) ). det(g (1) ) = ∂(u1 , v1 ) 2 ,v2 ) However, ∂(u ∂(u1 ,v1 ) is the Jacobian of the coordinate transformation ~ −1 (T ) → X ~ −1 (T ) defined by X ~ −1 ◦ X ~ 1 restricted to X ~ −1 (T ). F :X 1 2 2 1 The result follows as an application of the change of variables formula in double integrals. 

Lemma 8.3.2. Let S be a regular oriented surface, and let R be a regu-

lar compact region of S, possibly with a boundary. Given a collection ~ i }i∈I of coordinate neighborhoods that cover S, {Tj }j∈J triangles {X of a triangulation of R, and i : J → I such that Tj is in the image ~ i(j) , define the sum of X X ZZ j∈J

~ −1 (Tj ) X i(j)

q f (ui(j) , vi(j) ) det g (i(j)) dui(j) dvi(j) .

This sum is independent of the choice of triangulation of R, collection of coordinate patches, and function i.

300

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

Proof: That the sum is independent of the collection of coordinate neighborhoods follows from Lemma 8.3.1. That the sum does not depend on the choice of triangulation is a little tedious to prove and is left as an exercise for the reader.  This leads to a definition of a surface integral over any region on a regular surface. Definition 8.3.3. Let S be a regular oriented surface, and let R be a

regular compact region of S, possibly with a boundary. We call the common sum described in Lemma 8.3.2 the surface integral of f over R and denote it by ZZ f dS. R

~ : U → R3 is a regular parametrization of a We point out that if X ~ ) is dense in R, then region of S such that X(U ZZ ZZ ~ ~u × X ~ v k dA, f dS = f (X(u, v))kX R

U

where the right-hand side is the usual double integral. We can now state the main theorem of this section. Theorem 8.3.4 (Global Gauss-Bonnet Theorem). Let S be a regular ori-

ented surface of class C 3 , and let R be a compact region of S with boundary ∂R. Suppose that ∂R is a simple, closed, piecewise regular, positively oriented curve. Suppose that ∂R has k regular arcs Ci of class C 2 , and let θi be the external angles of the vertices of ∂R. Then ZZ k Z k X X κg ds + K dS + θi = 2πχ(R), i=1

Ci

R

i=1

where χ(R) is the Euler characteristic of R. Proof: Let R be covered by a collection of coordinate patches. Let {Tj }j∈J be the triangles of a triangulation of R in which all the triangles Tj on S are subsets of some coordinate patch. Suppose also that every triangle in the set {Tj } is equipped with an orientation that is compatible with the orientation of S.

8.3. Gauss-Bonnet Theorem, Global Form

301

R

Figure 8.8. A triangulation of a region R.

For each triangle Tj , for 1 ≤ l ≤ 3, call Ejl the edges of Tj (as curves on S), call Vjl the vertices of Tj , and let βjl be the interior angle of Tj at Vjl . For this triangulation, call a0 the number of vertices, a1 the number of edges, and a2 the number of triangles. By construction, the local Gauss-Bonnet Theorem applies to each triangle Tj on S, so on each Tj , we have 3 Z X l=1

ZZ K dS = 2π−

κg ds+ Ejl

Tj

3 X

(π−βjl ) = −π+

3 X

βjl . (8.15)

l=1

l=1

Now consider the sum of Equation (8.15) over all the triangles Tj . Since each triangle has an orientation compatible with the orientation of S, then whenever two triangles share an edge, the edge is traversed in opposite orientations on the adjacent triangles (see Figure R 8.8). Consequently, in the sum of Equation (8.15), each integral Ejl κg ds cancels out another similar integral along any edge that is not a part of the boundary of R. Therefore, applying Lemma 8.3.2, the left-hand side of the sum of Equation (8.15) is precisely k Z X i=1

ZZ K dS.

κg ds + Ci

R

(8.16)

302

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

The right-hand side of the sum of Equation (8.15) is −πa2 +

3 XX j

βjl .

i=1

PP

In the double sum βjl , the sum of interior angles associated to a vertex on the interior of R contributes 2π, and the sum of angles associated to a vertex V on the boundary ∂R contributes π − (exterior angle of V ). Thus, X j

−π +

3 X

! βjl

=−

k X

θi − πa2 + 2π(#interior vertices) + π(#exterior vertices).

i=1

l=1

Since there are as many vertices on the boundary as there are edges, we rewrite this as ! 3 k X X X −π + βjl = − θi − πa2 + 2πa0 − π(#exterior edges). j

i=1

l=1

(8.17) However, since each triangle has three edges, −a2 = 2a2 − 3a2 = 2a2 − 2(#interior edges) − (#exterior edges). (8.18) Consequently, taking the sum of Equation (8.15) and combining Equations (8.16), (8.17), and (8.18) one obtains XZ i

Ci

ZZ K dS = −

κg ds + R

k X

θi + 2π(a2 − a1 + a0 ).

i=1

By definition of the Euler characteristic, χ(R) = a2 − a1 + a0 . The theorem follows.  The global version of the Gauss-Bonnet Theorem directly generalizes the local version so when one refers to the Gauss-Bonnet Theorem, one means the global version. Even at a first glance, the Gauss-Bonnet Theorem is profound because it connects local properties of curves on a surface (the geodesic curvature κg , the Gaussian

8.3. Gauss-Bonnet Theorem, Global Form

303

curvature K, and angles associated to vertices) with global properties (the Euler characteristic of a region of a surface). In fact, since the Euler characteristic is a topological invariant, the Gauss-Bonnet Theorem connects local geometric properties to a topological property. Let S be a compact regular surface (without boundary). An interesting particular case of the global Gauss-Bonnet Theorem occurs when we consider R = S, which implies that ∂R is empty. This situation leads to the following strikingly simple and profound corollary. Corollary 8.3.5. Let S be an orientable, compact, regular surface of

class C 3 without boundary. Then ZZ K dS = 2πχ(S). S

~ : [0, 2π]2 → Example 8.3.6. Consider the torus T parametrized by X R3 , with ~ X(u, v) = ((a + b cos v) cos u, (a + b cos v) sin u, b sin v), ~ where a > b. We note that X((0, 2π)2 ) covers all of T except for two ~ curves on the torus, so X((0, 2π)2 ) is dense in T . By the comment after Definition 8.3.3, we can use the usualRR surface integral over this one coordinate neighborhood to calculate S K dS directly. It is not hard to calculate that over the coordinate neighborhood described above, we have K(u, v) =

cos v b(a + b cos v)

Thus, ZZ

Z

Z

K dS = S

0

0

~u × X ~ v k = b(a + b cos v). and kX

cos v · b(a + b cos v) du dv = b(a + b cos v)

Z

Z

cos v du dv = 0, 0

which proves that χ(T ) = 0. This agrees with the calculation provided by the triangulation in Figure 8.5. We shall now present a number of applications of the GaussBonnet Theorem, as well as leave a few as exercises for the reader.

0

304

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

Proposition 8.3.7. Let S be a compact, regular, orientable surface of class C 3 without boundary and with positive curvature everywhere. Then S is homeomorphic to a sphere.

Proof: Since K > 0 over the whole surface S, then for the total Gaussian curvature we have ZZ K dS > 0. S

By Corollary 8.3.5, this integral is 2πχ(S). However, by the Classification Theorem of Surfaces and Equation (8.14), since χ(S) > 0, we have χ(S) = 2, and therefore S is homeomorphic to a sphere. 

Problems 8.3.1. Provide all the details that establish Equation (8.10). 8.3.2. Provide the details for the proof of Lemma 8.3.2. 8.3.3. Verify directly the Gauss-Bonnet Theorem for the rectangular region R on a torus specified by u1 ≤ u ≤ u2 and v1 ≤ v ≤ v2 , where the u-coordinate lines are the parallels and the v-coordinate lines are the meridians of the torus, as a surface of revolution. 8.3.4. Consider the surface of revolution parametrized by ~ X(u, v) = ((2 + sin v) cos u, (2 + sin v) sin u, v) with (u, v) ∈ [0, 2π] × R. (a) Show that the geodesic curvature is constant along the coordinate line v = v0 . (b) Determine the geodesic curvature along a coordinate line u = u0 . (c) Use the RR above results and the Gauss-Bonnet Theorem to determine R K dS over a region R defined by 0 ≤ u ≤ 2π and v1 ≤ v ≤ v2 , for constants v1 and v2 . 8.3.5. Let S be a regular, orientable, compact surface without boundary that has positive Gaussian curvature. Prove that the surface area of S is less than 4π/Kmin , where Kmin > 0 is the minimum Gaussian curvature.

8.4. Geodesics

305

8.3.6. Jacobi’s Theorem. Let α ~ : I → R3 be a closed, regular, parametrized curve. Suppose also that α ~ (t) has a curvature function κ(t) that is never 0. Suppose also that the principal normal indicatrix, i.e., the curve P~ : I → S2 , is simple, that is, that it cuts the sphere into only two regions. By viewing P~ (I) as a curve on the sphere S2 , use the Gauss-Bonnet Theorem to prove that P~ (I) separates the sphere into two regions of equal area. ~ 8.3.7. Consider the surface parametrized X(u, v) = (u, v, uv) with (u, v) ∈ R2 . (a) Show that the Gaussian curvature is K = −1/(1 + u2 + v 2 )2 . (b) Find the formula for the geodesic curvature κg (t)s0 (t)3 in terms of the coordinate functions (u(t), v(t)) for any curve on the surface. (c) Show that κg (t) = 0 along the coordinate lines. (d) Use the Gauss-Bonnet Theorem for this surface and a region R defined as u1 ≤ u ≤ u2 and v1 ≤ v ≤ v2 to prove the double integral formula Z

u2

u1

Z

v2

v1

−u1 v1

du dv = cos−1 (1 + u2 + v 2 )3/2

!

p (1 + u21 )(1 + v12 ) −1

+ cos

−1

+ cos

u2 v1

! −1

+ cos

p (1 + u22 )(1 + v12 ) u1 v2 p (1 + u21 )(1 + v22 )

! − 2π.

8.4 Geodesics Classical geometry in the plane studies in great detail relationships between points, straight lines, and circles. One could characterize our theory of surfaces until now as a local theory in that we have concentrated our attention on the behavior of curves on surfaces at a point. Consequently, the notion of straightness (a global notion) and the concept of a circle (also a curve defined by a global property) do not yet make sense in our theory of curves on surfaces. Euclid defines a line as a “breadthless width” and a straight line as “a line which lies evenly with the points on itself.” These

−u2 v2 p (1 + u22 )(1 + v22 )

!

306

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

definitions do not particularly help us generalize the concept of a straight line to a general surface. However, it is commonly known that given any two points P and Q in Rn , a line segment connecting P and Q provides the path of shortest distance between these two points. On a regular surface S ⊂ R3 that is not planar, even the notion of distance in S between two points P and Q poses some difficulty since we cannot assume a straight line in R3 connecting P and Q lies in S. However, since it is possible to talk about the arc length of curves, we can define the distance on S between P and Q as inf{arc length of C | C is a curve on S connecting P and Q}. Therefore, one might wish to take as a first intuitive formulation of straightness on a regular surface S the following definition: A curve C on S is “straight” if for all pairs of points on the curve, the arc length between those two points P and Q is equal to the distance P Q between them. For general regular surfaces, this proposed definition turns out to be unsatisfactory, but it does lead to a more sophisticated way of generalizing straightness to surfaces. Let S be a regular surface, and let P and Q be points of S. Suppose that P and Q are in a coordinate patch that is parametrized ~ : U → R3 , where U ⊂ R2 . Consider curves on S parametrized by X ~ by ~γ (t) = X(u(t), v(t)) such that ~γ (0) = P and ~γ (1) = Q. According to Equation (6.2), the arc length of such a curve is 1p

Z

g11 (u0 (t))2 + 2g12 u0 (t)v 0 (t) + g22 (v 0 (t))2 dt,

s=

(8.19)

0

where we understand that the gij coefficients are functions of u and v, which are in turn functions of t. To find a curve that connects P and Q with the shortest arc length, one must find parametric equations (u(t), v(t)) that minimize the integral in Equation (8.19). Such problems are studied in calculus of variations, a brief introduction to which is presented in Appendix B of . According to the Euler-Lagrange Theorem in calculus of variations, if we set p f = g11 (u0 (t))2 + 2g12 u0 (t)v 0 (t) + g22 (v 0 (t))2 ,

8.4. Geodesics

307

then the parametric equations (u(t), v(t)) that optimize the arc length s in Equation (8.19) must satisfy   d ∂f def ∂f − =0 and Lu (f ) = ∂u dt ∂u0   d ∂f def ∂f Lv (f ) = − = 0. ∂v dt ∂v 0 We call Lu the Euler-Lagrange operator with respect to u and similarly for v. For a function with two intermediate variables u and v, as we have in this instance, let us also define the operator def

L(f ) = (Lu (f ), Lv (f )) . Proposition 8.4.1. The Euler-Langrange operator of

f=

p g11 (u0 (t))2 + 2g12 u0 (t)v 0 (t) + g22 (v 0 (t))2

satisfies L(f ) =

p

 det(g)κg (t) v 0 (t), −u0 (t) .

Proof: (The proof is left as an exercise for the persistent reader and relies on the careful application of Euler-Lagrange equations and Equation (8.7). )  Corollary 8.4.2. The parametric equations (u(t), v(t)) optimize the in-

tegral in Equation (8.19) if and only if κg (t) = 0 or (v 0 (t), u0 (t)) = ~0. Obviously, (v 0 (t), −u0 (t)) = ~0 integrates to (u, v) = (c1 , c2 ), where c1 and c2 are constants, so the curve degenerates to a point. The other part of Corollary 8.4.2 motivates the following definition, which generalizes to a regular surface the notion of a straight line. Definition 8.4.3. A geodesic is a curve on a surface with geodesic cur-

vature κg (t) identically 0. As a first remark, one notices that on a geodesic C, the curvature ~ , so at each point on the surface, κ = vector is T~ 0 = s0 κP~ = s0 κn N ±κn and the curve’s principal normal vector is equal to the surface

308

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

unit normal vector up to a possible change of sign. The transition ~ , T~ , U ~ } frame becomes matrix between the Frenet frame and the {N       1 0 0 ~ N ~ = T~ P~ B ~ 0 0 ε  , T~ U 0 ε 0 where ε = ±1. Consequently, along a geodesic, the osculating plane of the curve and the tangent plane to the surface are normal to each other. As a second remark, the defining property that the geodesic curvature κg is identically 0 along a geodesic simplifies applications involving the Gauss-Bonnet Theorem. For example, suppose that S is a surface as in the conditions of the Gauss-Bonnet Theorem (Theorem 8.3.4) and that R is a region on S whose boundary consists of arcs of geodesics. Then the integration of κg along the regular arcs vanishes and the Gauss-Bonnet Theorem becomes ZZ K dS + R

k X

θi = 2πχ(R),

i=1

where θi are the exterior angles where the geodesic arcs meet. The following proposition gives another example of how the Gauss-Bonnet Theorem can provide profound geometric properties about geodesics just from this simplification. Proposition 8.4.4. Let S be a compact, connected, orientable, regular

surface without boundary and of positive Gaussian curvature. If there exist two simple, closed geodesics γ1 and γ2 on S then they intersect. Proof: By Proposition 8.3.7, S is homeomorphic to a sphere. Suppose that γ1 and γ2 do not intersect. Then they form the boundary of region R that is homeomorphic to a cylinder with boundary. It is not hard to verify by supplying R with a triangulation that χ(R) = 0. However, applying the Gauss-Bonnet Theorem to this situation, we obtain ZZ K dS = 0, R

which is a contradiction since K > 0.



8.4. Geodesics

309

We now address the problem of finding geodesics. From Equation ~ ◦α (8.7), a curve ~γ = X ~ , with α ~ (t) = (u(t), v(t)), is a geodesic if and only if Γ211 (u0 )3 + (2Γ212 − Γ111 )(u0 )2 v 0 + (Γ222 − 2Γ112 )u0 (v 0 )2 − Γ122 (v 0 )3 + u0 v 00 − u00 v 0 = 0. This formula holds for any parameter t and not just when ~γ is parametrized by arc length. We can approach the task of finding equations for geodesics in an alternative way. Suppose now that we consider curves on the surface parametrized by arc length so that s0 = 1 and s00 = 0. Since T~ 0 is ~ , we conclude that parallel to N ~u = 0 T~ 0 · X

and

~ v = 0. T~ 0 · X

Then using Equation (8.5) we deduce that ~ u (u0 )2 + 2X ~ uv · X ~ u u0 v 0 + X ~ vv · X ~ u (v 0 )2 + X ~u · X ~ u u00 + X ~v · X ~ u v 00 = 0, ~ uu · X X ~ v (u0 )2 + 2X ~ uv · X ~ v u0 v 0 + X ~ vv · X ~ v (v 0 )2 + X ~u · X ~ v u00 + X ~v · X ~ v v 00 = 0. ~ uu · X X Solving algebraically for u00 and v 00 in the above two equations, we obtain the following classical equations for a geodesic curve which we will express using tensor notation for simplicity with variables x1 = u and x2 = v: 2 j k X d2 xi i dx dx =0 + Γ jk ds2 ds ds

for i = 1, 2.

(8.21)

j,k=1

Since i can be 1 or 2, this equation represents a system of two differential equations. At first sight, one might see a discrepancy between Equation (8.21), which involves two equations and Equation (8.20), which involves only one. However, it is essential to point out that the system in Equation (8.21) holds for geodesics parametrized by arc length. Furthermore, Equation (8.21) is equivalent to the system of equations κg (t) = 0

and

(s0 )2 = g11 (u0 )2 + 2g12 u0 v 0 + g22 (v 0 )2 .

(8.20)

310

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

Finding explicit parametric equations of geodesic curves on a surface is often difficult since it involves solving a system of nonlinear second-order differential equations. Interestingly enough, there is a common strategy in differential equations that allows us to transform this system of second-order equations in two functions x1 (s) and x2 (s) into a system of first-order differential equations in four functions. (See the proof of Theorem 8.4.10.) There are common computational techniques to solve systems of first-order differential equations numerically, even if they are nonlinear. Consequently, there are algorithms to solve Equation (8.21) numerically. In specific situations, there sometimes exist simplifications for Equation (8.21) that allow one to explicitly compute the geodesics on a particular surface. Example 8.4.5 (The Plane: Cartesian Coordinates). Consider the xy-plane

parametrized with the usual Cartesian coordinates. Then the Christoffel symbols are all Γijk = 0. Using Equation (8.21), the equations for a geodesic in the plane are simply d2 u =0 ds2

and

d2 v = 0. ds2

Integrating both equations twice, one obtains u(s) = as + c and v(s) = bs + d. Furthermore, since (u0 (s))2 + (v 0 (s))2 = 1, these constants must satisfy a2 +b2 = 1. Therefore, geodesics parametrized by arc length in the plane are given as ~γ (s) = p~ + s~u, where p~ is a point and ~u is a unit vector. Example 8.4.6 (The Plane: Polar Coordinates). In contrast to the pre-

vious example, consider the xy-plane parametrized with polar coordinates so that as a surface in R3 , the xy-plane is given by ~ θ) = (r cos θ, r sin θ, 0). X(r, A short calculation gives Γ212 = Γ221 =

1 r

and

Γ122 = −r

8.4. Geodesics

311

and the remaining five other symbols are 0. Equations (8.21) become  2 dθ d2 r −r = 0, (8.22) 2 ds ds d2 θ 2 dr dθ + = 0. (8.23) ds2 r ds ds We transform this system to obtain a differential equation relating r and θ as follows. Note that by repeatedly using the chain rule, we get   dr dr dθ d2 r d2 r dθ 2 dr d2 θ = and = 2 + . ds dθ ds ds2 dθ ds dθ ds2 Putting these two into Equations (8.22) and (8.23) leads to !    2 dθ d2 r 2 dr 2 − − r = 0, ds dθ2 r dθ which breaks into the pair of equations   dθ d2 r 2 dr 2 = 0 or − − r = 0. ds dθ2 r dθ

(8.24)

One notices that the first equation in Equation (8.24) is satisfied when θ is a constant, which corresponds to a line through the origin. The other equation in Equation (8.24) does not appear particularly tractable but a substitution simplifies it greatly. Using the new variable u = 1r , one can check that   d2 r 2 du 2 1 d2 u = − , dθ2 u3 dθ u2 dθ2 and hence the second equation in Equation (8.24) reduces to d2 u + u = 0. (8.25) dθ2 Using standard techniques in ordinary differential equations, the general solution to Equation (8.25) can be written as u(θ) = C cos(θ−θ0 ) where C and θ0 are constants. Therefore, in polar coordinates, equations for geodesics in the plane (i.e., lines) are given by θ=C

or

r(θ) = C sec(θ − θ0 ).

312

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

q

p

Figure 8.9. Two geodesics on a cylinder.

Example 8.4.7 (Cylinder). Consider a right circular cylinder. A pa-

~ rametrization for this cylinder is X(u, v) = (cos(u), sin(u), v), with (u, v) in [0, 2π] × R. An easy calculation shows that Γijk = 0 for all i, j, k, and hence, that geodesics on a cylinder are curves of the form ~ ◦α γ=X ~ where α ~ (t) = (at + b, ct + d) for a, b, c, d that are constant. As a result, the geodesics on a cylinder are either straight lines parallel to the axis of the cylinder, circles in planes perpendicular to the axis of the cylinder, or helices around that axis. Figure 8.9 illustrates how on a surface, two different geodesics may connect two distinct points, a situation that does not occur in the plane. In fact, on a cylinder, there is an infinite number of geodesics that connect any two points. The difference between each geodesic connecting p and q is how many times the geodesic wraps around the cylinder and in which direction it wraps. Example 8.4.8 (Sphere). Consider the parametrization of the sphere

given by ~ 1 , x2 ) = (R cos x1 sin x2 , R sin x1 sin x2 , R cos x2 ), X(x

8.4. Geodesics

313

where x1 = u is the longitude θ in spherical coordinates, and x2 = v is the colatitude angle ϕ down from the positive z-axis. In Example 7.1.6, we determined the Christoffel symbols for this parametrization. Equations (8.21) for geodesics on the sphere become 1 2 d2 x1 2 dx dx = 0, + 2 cot(x ) ds2 ds ds  1 2 dx d2 x2 2 2 − sin(x ) cos(x ) = 0. ds2 ds

(8.26)

A geodesic on the sphere is now just a curve of the form ~γ (s) = ~ 1 (s), x2 (s)), where x1 (s) and x2 (s) satisfy the system of differX(x ential equations in Equation (8.26). Taking a first derivative of ~γ (s) gives  dx1 dx2 0 ~γ (s) = R − sin x1 sin x2 + cos x1 cos x2 , ds ds  dx1 dx2 dx2 + sin x1 cos x2 , − sin x2 , cos x1 sin x2 ds ds ds and the second derivative, after simplification using Equation (8.26), is  1 2  2 2 i h dx dx d2~γ 2 2 ~γ (s). = − sin (x ) + 2 ds ds ds h  1 2  2 2 i 2 2 2 However, the term R sin (x ) dx is the first funda+ dx ds ds mental form applied to ((x1 )0 (s), (x2 )0 (s)), which is precisely the square of the speed of ~γ (s). However, since the geodesic is parametrized by arc length, its speed is identically 1. Thus, Equations (8.26) lead to the differential equation ~γ 00 (s) +

1 ~γ (s) = 0. R2

Standard techniques with differential equations allow one to show that all solutions to this differential equation are of the form s s ~γ (s) = ~a cos + ~b sin , R R

314

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

where ~a and ~b are constant vectors. Note that ~γ (0) = ~a and that ~γ 0 (0) = R1 ~b. Furthermore, to satisfy the conditions that ~γ (s) lie on the sphere of radius R and be parametrized by arc length, we deduce that ~a and ~b satisfy k~ak = R,

k~bk = R,

and

~a · ~b = 0.

Therefore, we find that ~γ (s) traces out a great arc on the sphere that is the intersection of the sphere and the plane through the center of the sphere spanned by ~γ (0) and ~γ 0 (0). Example 8.4.9 (Surfaces of Revolution). Consider a surface of revolu-

tion about the z-axis given by the parametric equations ~ X(u, v) = (f (v) cos u, f (v) sin u, h(v)) , where f and h are functions defined over a common interval I such that over [0, 2π] × I. We assume that over the open interval (0, 2π) × ~ is a regular parametrization, which implies that I the function X f (v) > 0. A simple calculation reveals that the Christoffel symbols of the second kind are Γ111 = 0, Γ211 = −

Γ112 = ff0 , + (h0 )2

(f 0 )2

f0 , f

Γ212 = 0,

Γ122 = 0, Γ222 =

f 0 f 00 + h0 h00 . (f 0 )2 + (h0 )2

Equations (8.21) for geodesics parametrized by arc length on a surface of revolution become d2 u 2f 0 du dv =0 + ds2 f ds ds  2   du d2 v ff0 f 0 f 00 + h0 h00 dv 2 − 0 2 + 0 2 = 0. ds2 (f ) + (h0 )2 ds (f ) + (h0 )2 ds

(8.27)

As complicated as Equations (8.27) appear, it is possible to find a “solution” to this system of differential equations for u in terms of v. However, before establishing a general solution, we will determine which meridians (u =const.) and which parallels or latitude lines (v =const.) are geodesics.

8.4. Geodesics

315

Consider first the meridian lines that are defined by u = C, where C is a constant. Notice that the first equation in the system in Equation (8.27) is trivially satisfied for meridians and that the second equation becomes   d2 v f 0 f 00 + h0 h00 dv 2 + 0 2 = 0. (8.28) ds2 (f ) + (h0 )2 ds It is easy to check that the first fundamental form on (u0 (t), v 0 (t)) on the surface of revolution is f (v)2 u0 (t)2 + (f 0 (v)2 + h0 (v)2 )v 0 (t)2 .

(8.29)

However, assuming that we have a meridian that is parametrized by arc length, then (f 0 (v)2 + h0 (v)2 )v 0 (s)2 = 1 since the speed function of a curve parametrized by arc length is 1. Consequently, 1 v 0 (s)2 = 0 2 , f (v) + h0 (v)2 and taking a derivative of this equation with respect to s, one obtains 2v 0 v 00 = −

2f 0 (v)f 00 (v)v 0 + 2h0 (v)h00 (v)v 0 (f 0 (v)2 + h0 (v)2 )2

= −2

f 0 (v)f 00 (v) + h0 (v)h00 (v) 0 3 (v ) . f 0 (v)2 + h0 (v)2

Since v 0 (s) 6= 0 on the meridian parametrized by arc length, then   f 0 (v)f 00 (v) + h0 (v)h00 (v) dv 2 d2 v =− , ds2 f 0 (v)2 + h0 (v)2 ds which shows that Equation (8.28) is identically satisfied on all meridians. Now consider the parallel curves on a surface of revolution, which are defined by v = v0 , a real constant. (See Figure 8.10.) In Equation (8.27), the first equation leads to u0 (s) = C, a constant, and the second equation becomes  2 du ff0 = 0. 0 2 0 2 (f ) + (h ) ds

316

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

geodesic geodesic geodesic

not a geodesic

Figure 8.10. Geodesics on a surface of revolution.

Since the parallel curves, parametrized by arc length, are regular curves, then C 6= 0, and the condition that the surface of revolution be a regular surface implies that f (v) 6= 0. Thus, the second equation is satisfied for parallels v = v0 if and only if f 0 (v0 ) = 0. Even if a geodesic is neither a meridian nor a parallel, the first equation in Equation (8.27) is simple enough that one can nonetheless deduce some interesting consequences. Note that by taking a derivative with respect to the arc length parameter s, we have d 2 0 (f u ) = 2f f 0 v 0 u0 + f 2 u00 . ds Multiplying the first equation of Equation (8.27) by f 2 , we see that it can be written as (8.30) f (v)2 u0 (s) = C, where C is a constant. Note that when a curve on a surface is parametrized by arc length with coordinate functions (u0 (s), v 0 (s)), the angle θ it makes with any given parallel curve satisfies Ip ((u0 , v 0 ), (1, 0)) p cos θ = p = u0 f. 0 0 0 0 Ip ((u , v ), (u , v )) Ip ((1, 0), (1, 0))

8.4. Geodesics

317

As a geometric interpretation, since f is the radius r of surface of revolution at a given point, Equation (8.30) leads to the relation r cos θ = C

(8.31)

for all nonparallel geodesics on a surface of revolution, where θ is the angle between the geodesic and the parallels. Equation (8.31), which is equivalent to Equation (8.30) for nonparallel curves, is often called the Clairaut relation. Note that a curve satisfying Clairaut’s relation is a meridian if and only if C = 0. With the relation u0 = C/f 2 , since the speed is equal to 1, Equation (8.29) leads to  2 dv C2 ((f 0 )2 + (h0 )2 ) = 1 − 2 . (8.32) ds f Taking the derivative of this equation with respect to s, one obtains  3 dv dv d2 v C 2 f 0 dv 0 2 0 2 0 00 0 00 2 ((f ) + (h ) ) + 2(f f + h h ) = 2 ds ds2 ds f 3 ds which is equivalent to   2    du dv d2 v ff0 f 0 f 00 + h0 h00 dv 2 ((f ) + (h ) ) = 0. − 0 2 + 0 2 ds ds2 (f ) + (h0 )2 ds (f ) + (h0 )2 ds 0 2

0 2

Therefore, if a geodesic is not a parallel, the first equation of (8.27), which is equivalent to Clairaut’s relation, implies the second equation. Assuming that a geodesic is not a meridian, then from Clairaut’s relation, we know that u0 (s) is never 0 so one can define an inverse function to u(s), namely s(u), and then v can be given as a function of u by v = v(s(u)). Then in Equation (8.32), replacing dv ds with dv du dv C = , we obtain du ds du f 2 

dv du

2

C2 C2 0 2 0 2 ((f ) + (h ) ) = 1 − , f4 f2

and hence, dv f = du C

s

f 2 − C2 . (f 0 )2 + (h0 )2

318

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

Over an interval where this derivative is not 0, it is possible to take an inverse function of u with respect to v, and, by integration, this function satisfies s Z f 0 (v)2 + h0 (v)2 1 dv + D u=C f (v) f (v)2 − C 2 for some constant of integration D. The constants C and D lead to a two-parameter family of solutions that parametrize segments of geodesics. Theorem 8.4.10. Let S be a regular surface of class C 3 . For every

point p on S and every unit vector w ~ ∈ Tp S, there exists a unique geodesic on S through p in the direction of w. ~ ~ : U → R3 be a regular parametrization of a neighborProof: Let X hood of p on S, and let gij and Γijk be the components of the metric tensor and Christoffel symbols of the second kind, respectively, in ~ Note that since S is of class C 3 , all of the functions relation to X. i Γjk are of class C 1 over their domain. The proof of this proposition is an application of the existence and uniqueness theorem for first-order systems of differential equations (see [2, Section 31.8]), which states that if F : Rn → Rn is of class C 1 , then there exists a unique function ~x : I → Rn that satisfies ~x0 = F (~x)

and

~ ~x(t0 ) = C,

~ is a constant vector. where I is an open interval containing t0 , and C Consider now the system of differential equations given in Equation (8.21), where we do not assume that a solution is parametrized by arc length. Setting the dependent variables v 1 = (x1 )0 and v 2 = (x2 )0 , then Equation (8.21) is equivalent to the system of first-order differential equations   (x1 )0 = v 1 ,    (x2 )0 = v 2 , (8.33) 1 )0 1 (v 1 )2 − 2Γ1 v 1 v 2 − Γ1 (v 2 )2 ,  (v = −Γ  11 12 22   (v 2 )0 = −Γ2 (v 1 )2 − 2Γ2 v 1 v 2 − Γ2 (v 2 )2 , 11 12 22

8.4. Geodesics

319

where the functions Γijk are functions of x1 and x2 . Therefore, according to the existence and uniqueness theorem, there exists a unique solution to Equation (8.21) with specific values given for ~ 1 (s), x2 (s)). x1 (s0 ), x2 (s0 ), (x1 )0 (s0 ), and (x2 )0 (s0 ). Set α ~ (s) = X(x Now Equation (8.21) is the formula for arc length parametrizations of geodesics. However, it is not hard to prove that if f (s) = g11 (x1 (s), x2 (s))(x1 )0 (s)2 + 2g12 (x1 (s), x2 (s))(x1 )0 (s)(x2 )0 (s) + g22 (x1 (s), x2 (s))(x2 )0 (s)2 , where x1 and x2 satisfy Equation (8.21), then f 0 (s) = 0, so f (s) is a constant function over its domain. Consequently, if (x1 )0 (s0 ) and (x2 )0 (s0 ) are given so that α ~ 0 (s0 ) = w, ~ then k~ α(s)k = 1 for all s. If, 1 2 in addition, x (s0 ) and x (s0 ) are chosen so that α ~ (s0 ) = p, then α ~ (s) is an arc length parametrization of a geodesic passing through p with direction w ~ and is unique.  The proof of the above theorem establishes an additional fact concerning the equation for a geodesic curve. ~ be the parametrization of a neighborhood of Proposition 8.4.11. Let X a regular surface of class C 3 . Any solution (x1 (s), x2 (s)) to Equation (8.21), where one makes no prior assumptions on the parameter s, is ~ 1 (s), x2 (s)), whose locus such that the parametric curve ~γ (s) = X(x is a geodesic, has constant speed. Example 8.4.12. Since the geodesic curvature is intrinsic, only de-

pending on the metric coefficients gij (u, v), we can find the expression for the geodesic curvature of a curve in the Poincar´e upper half-plane with the metric g11 (u, v) = g22 (u, v) = 1/v 2 and g12 (u, v) = 0, using the computations from Example 7.1.7. Using Formula (8.7), we find that   1 0 3 1 1 0 0 3 0 2 0 00 00 0 u (t) + u (t)v (t) + u (t)v (t) − u (t)v (t) . s (t) κg (t) = v(t)2 v(t) v(t)2

Problems 8.4.1. Let S be a regular surface, and let ~γ be a geodesic on S. Prove that the geodesic torsion of γ is equal to ±τ , where τ is the torsion

(8.34)

320

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

function of ~γ (t) as a space curve. [Hint: The possible difference in sign stems from the possible change in sign of κn which may come from a reparametrization of the surface. This property of τg justifies its name as geodesic torsion.] 8.4.2. (ODE) Consider a right circular cone with opening angle α, where 0 < α < π/2. Consider the coordinate patch parametrized by ~ X(u, v) = (v sin α cos u, v sin α sin u, v cos α), where we assume v > 0. Determine equations for the geodesics on this cone. [Hint: Find a dv in terms of u and v and then differential equation that expresses du solve this to find an equation for u in terms of v.] ~ 8.4.3. Consider the torus parametrized by X(u, v) = ((a+b cos v) cos u, (a+ b cos v) sin u, b sin v) where a > b. Show that the geodesics on a torus satisfy the differential equation p dr 1 p 2 = r r − C 2 b2 − (r − a)2 , du Cb where C is a constant and r = a + b cos v. 8.4.4. Find the differential equations that determine geodesics on a function graph z = f (x, y). ~ : U → R3 is a parametrization of a coordinate patch on a 8.4.5. If X regular surface S such that g11 = E(u), g12 = 0 and g22 = G(u) show that (a) the u-parameter curves (i.e., over which v is a constant) are geodesics; (b) the v-parameter curve u = u0 is a geodesic if and only if Gu (u0 ) = 0; (c) the curve ~x(u, v(u)) is a geodesic if and only if p Z C E(u) p du, v=± p G(u) G(u) − C 2 where C is a constant. 8.4.6. Fill in the details in the proof of Proposition 8.4.10, namely, prove that if x1 (s) and x2 (s) satisfy Equation (8.21), then the function ~ 1 (s), x2 (s)), is constant. f (s), which is the square of the speed of X(x 8.4.7. Liouville Surface. A regular surface is called a Liouville surface if it can be covered by coordinate patches in such a way that each patch ~ can be parametrized by X(u, v) such that g11 = g22 = U (u) + V (v)

and

g12 = 0,

8.4. Geodesics

321

where U is a function of u alone and V is a function of v alone. (Note that Liouville surfaces generalize surfaces of revolution.) Prove the following facts: (a) Show that the geodesics on a Liouville surface can be given as solutions to an equation of the form Z Z du dv p =± p + c2 , U (u) − c1 V (v) + c1 where c1 and c2 are constants. (b) Show that if ω is the angle a geodesic makes with the curve v =const., then U sin2 ω − V cos2 ω = C for some constant C. 8.4.8. Let α ~ : I → R3 be a regular space curve parametrized by arc length with nowhere 0 curvature. Consider the ruled surface parametrized by ~ t) = α ~ X(s, ~ (s) + tB(s), ~ is the binormal vector of α ~ is defined over where B ~ . Suppose that X I × (−ε, ε), with ε > 0. ~ is a regular (a) Prove that if ε is small enough, the image S of X surface. (b) Prove that if S is a regular surface, α ~ (I) is a geodesic on S. (This shows that every regular space curve that has nonzero curvature is the geodesic of some surface.) 8.4.9. Consider the elliptic paraboloid given by z = x2 + y 2 . Note that this is a surface of revolution. Consider the geodesics on this surface that are not meridians. (a) Suppose that the geodesic intersects the parallel at z = z0 with an angle of θ0 . Find the lowest parallel that the geodesic reaches. (b) Prove that any geodesic that is not a meridian intersects itself an infinite number of times. 8.4.10. Consider the hyperboloid of one sheet given by the equation x2 + y 2 − z 2 = 1 and let p be a point in the upper half-space defined by z > 0. Consider now geodesic curves that go through p and make an acute angle of θ0 with the parallel of the hyperboloid passing through p. Call r0 the distance from p to the z-axis.

322

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

r0

r0 p

p

θ0

θ0

Figure 8.11. Geodesics on the hyperboloid of one sheet.

(a) Prove that if cos θ0 > 1/r0 , then the geodesic remains in the upper half-space z > 0. (b) Prove that if cos θ0 < 1/r0 , then the geodesic crosses the z = 0 plane, and descends indefinitely in the negative z-direction. (c) Prove that if cos θ0 = 1/r0 , then the geodesic, as it descends from p, asymptotically approaches the parallel given by x2 + y 2 = 1 at z = 0. (d) Using the parametrization ~ X(u, v) = (cosh v cos u, cosh v sin u, sinh v) , suppose that the initial conditions for the geodesic are u(0) = 0 ~ and v(0) = v0 , so that p = X(0, v0 ). Find the initial conditions u0 (0) and v 0 (0) so that the geodesic is parametrized by arc length and satisfies the condition in (c). (See Figure 8.11 for examples of nonasymptotic behavior.) 8.4.11. Using the calculations of the formula for geodesic curvature from Example 8.4.12, show that the curve u(t) = (R cos t + m, R sin t), where R > 0 and m are constants, is a geodesic in the Poincar´e upper half-plane. Also show that any curve (a, t), where a is a constant, is a geodesic for this metric.

8.5. Geodesic Coordinates

323

8.4.12. Let S be a regular, orientable surface of class C 3 in R3 that is homeomorphic to the sphere. Let γ be a simple closed geodesic in S. The curve γ separates S into two regions A and B that share γ as their boundary. Let n : S → S2 be the Gauss map induced from a given orientation of S. Prove that n(A) and n(B) have the same area. [Hint: Use the Gauss-Bonnet Theorem.] 8.4.13. Let S be an orientable surface with Gaussian curvature K ≤ 0, and let p ∈ S. (a) Let γ1 and γ2 be two geodesics that intersect at p. Prove that γ1 and γ2 do not intersect at another point q in such a way that γ1 and γ2 form the boundary of a simple region R. (b) Prove also that a geodesic on S cannot intersect itself in such a way as to enclose a simple region.

8.5 Geodesic Coordinates 8.5.1 General Geodesic Coordinates Definition 8.5.1. Let S be a regular surface of class C 3 . A system of

~ of geodesic coordinates is an orthogonal regular parametrization X S such that, for one of the coordinates, all the coordinate lines are geodesics. There are many ways to define geodesic coordinates on a regular surface of class C 3 . We introduce one general way first. Suppose that α ~ (t) for t ∈ [a, b] is a regular curve with image C on 2 S of class C . According to Theorem 8.4.10, for each t0 ∈ [a, b], there exists a unique geodesic on S through α ~ (t0 ) perpendicular to C. (See Figure 8.12.) Furthermore, one can parametrize the geodesics ~γt0 (s) by arc length such that ~γt0 (0) = α ~ (t0 ) and t 7→ ~γt0 (0) is continuous in t. Define the function ~ t) = ~γt (s). X(s,

(8.35)

We wish to show that over an open set containing C, the function ~ X(s, t) is a regular parametrization of class C 2 . (One desires class C 2 since this is required for the first and second fundamental forms to exist.) However, we must assume that S is of class C 5 .

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

α ~ (v) u-coord inate li nes

324

Figure 8.12. Geodesic coordinate system generated by α ~ (v). Proposition 8.5.2. If S is of class C 5 , there exists an ε > 0 such that

~ t), as defined in Equation (8.35), is a regular parametrization X(s, of class C 2 over (−ε, ε) × (a, b). Proof: The proof relies on standard theorems of existence and uniqueness for differential equations as well as on some basic topology. ~ 1 , x2 ) be a regular parametriLet p be a point on C, and let X(x 5 ~ p ) of S containing p. zation of class C of a neighborhood Vp = X(U ~ 1 (t), α2 (t)), ~ (t) = X(α Let α1 (t) and α2 (t) be functions such that α the given parametrization of C on S in the neighborhood Vp . Call Ip the domain of α ~ such that α ~ (Ip ) ⊂ Vp , and suppose that p = α ~ (t0 ). 1 2 Let x (s, t) and x (s, t) be functions that map onto the geodesics ~γt (s) via ~ 1 (s, t), x2 (s, t)) = ~γt (s). X(x By Theorem 8.4.10, for all parameters t, over an interval s ∈ (−εt , εt ), there exists a unique solution for x1 (s, t) and for x2 (s, t) to Equation (8.21) given the initial conditions x1 (0, t) = α1 (t), ∂x1 (0, t) = u(t), ∂s

x2 (0, t) = α2 (t), ∂x2 (0, t) = v(t), ∂s

8.5. Geodesic Coordinates

325

where u and v are any functions of class C 1 . Furthermore, these solutions are of class C 2 in the variable s. Now since for each t we want ~γt (s) to be an arc length parametrization of a geodesic perpendicular to C, we impose the following two conditions on u and v for all t: (i)

g11 u2 + 2g12 uv + g22 v 2 = 1,

dα1 dα2 dα1 dα2 + g12 u + g12 v + g22 v = 0, dt dt dt dt where (i), for example, means that (ii) g11 u

g11 (α1 (t), α2 (t))u(t)2 + 2g12 (α1 (t), α2 (t))u(t)v(t) + g22 (α1 (t), α2 (t))v(t)2 = 1 for all t. The requirement to parametrize each ~γt (s) so that ~γt0 (0) is continuous in t is equivalent to requiring that u and v be continuous. These conditions completely specify u(t) and v(t) for all t. Notice that for (i) and (ii) to both be satisfied, ((α1 )0 (t), (α2 )0 (t)) and (u(t), v(t)) cannot be linear multiples of each other, and hence, 1 0 (α ) (t) u(t) (α2 )0 (t) v(t) 6= 0 for all t. Now write the equations for geodesics as a first-order system, as ~ is of class C 5 , then all the functions Γi in Equation (8.33). Since X jk are of class C 3 . Then according to theorems of dependency of solutions of differential equations on initial conditions (see [2, Theorem 32.4]), solutions to the system from Equation (8.33) are of class C 2 in terms of initial conditions, which implies that ∂ 2 x1 , ∂s∂t

∂ 2 x2 , ∂s∂t

∂ 2 x1 , ∂t2

and

∂ 2 x2 ∂t2

are continuous over a possibly smaller open neighborhood Up0 of (0, t0 ). Thus, we conclude that x1 (s, t) and x2 (s, t) are of class C 2 over Up0 . ~ t), note that To prove regularity of X(s, 1 0 (α ) (t0 ) u(t0 ) ∂(x1 , x2 ) , (α2 )0 (t0 ) v(t0 ) = ∂(s, t) (0,t0 )

326

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

the Jacobian of the change of variables x1 (s, t) and x2 (s, t) at the point (0, t0 ). Since this Jacobian is a continuous function and is nonzero at (0, t0 ), there is an open neighborhood Up00 of (0, t0 ) such that

∂(x1 ,x2 ) ∂(s,t)

6= 0. By Proposition 5.4.2, ~ ~ ∂X ∂X ∂(x1 , x2 ) × = ∂s ∂t ∂(s, t)

~ ~ ∂X ∂X × ∂x1 ∂x2

! ,

~ t) is regular. Consequently, so over Up00 the parametrization X(s, there exists an εp > 0 such that ¯p def = (−εp , εp ) × (t0 − εp , t0 + εp ) ⊂ Up0 ∩ Up00 , U ~ t) is regular and of class C 2 . ¯p , the parametrization X(s, and over U ~ U ¯p ). Call V¯p = X( Finally, consider the whole curve C on S. The curve C can be covered by open sets of the form V¯p for various p ∈ C. Let p and q ¯p → V¯p and X ~q : U ¯q → V¯q ~p : U be two points on C, and let us write X for associated parametrizations of V¯p and V¯q . If ¯p = (−εp , εp ) × Ip U

and

¯q = (−εq , εq ) × Iq U

~ p (s, t) = X ~ q (s, t) by the overlap, since for all t ∈ Ip ∩ Iq we have X ~ ~ uniqueness of geodesics, then we can extend Xp and Xq to a function ~ over X (− min(εp , εq ), min(εp , εq )) × (Ip ∪ Iq ). Since [a, b] is compact, [a, b] can be covered by a finite number of sets ¯p . Therefore, there exists ε > 0 and a single function X ~ defined U ~ over (−ε, ε) × (a, b) such that X(0, t) = α ~ (t), and for t ∈ (a, b), ~ t) parametrizes a geodesic by arc length. X(s,  Proposition 8.5.2 leads to the following general theorem. Theorem 8.5.3. Let S be a regular surface of class C 5 , and let α ~ :

[a, b] → S be a regular parametrization of a simple curve of class C 2 . ~ Then there exists a system of geodesic coordinates X(u, v) of class 2 ~ C defined over −ε < u < ε and a < v < b such that X(0, v) = α ~ (v), and the u-coordinate lines parametrize geodesics by arc length.

8.5. Geodesic Coordinates

327

~ Proof: Proposition 8.5.2 constructs the function X(u, v) as desired. ~ However, it remains to be proven that X(u, v) is an orthogonal parametrization. Consider the derivative of g12 with respect to u g12,1 =

∂ ~ ~ ∂g12 ~ 11 · X ~2 + X ~1 · X ~ 12 . = (X1 · X2 ) = X ∂u ∂u

~1 = ~1 ·X Since each u-coordinate line is parametrized by arc length, X ~1 · 1, so by differentiating with respect to v, we find that g11,2 = 2X ~ X12 = 0. By the same reasoning, over the u-coordinate lines, in the ~ , T~ , U ~ } frame one has {N ~ ∂X = T~ ∂u

and

~ ∂2X ~ + κn N ~. = κP~ = κg U ∂u2

~ 11 = ~ But since X(u, v) with v fixed is a geodesic, κg = 0. Thus, X ~ 2 = 0. ~ and, in particular, X ~ 11 · X κn N Consequently, g12,1 = 0, and therefore, g12 is a function of v only. We can write g12 (u, v) = g12 (0, v). However, by construction ~ of X(u, v) in Proposition 8.5.2, g12 (0, v) = 0 for all v. Thus, g12 is ~ identically 0, and hence, X(u, v) is an orthogonal parametrization. The class of geodesic coordinate systems described in Theorem 8.5.3 is of a particular type. Not every geodesic coordinate system needs to be defined in reference to a curve C on S as done ~ above. Let X(u, v) be any system of geodesic coordinates where the u-coordinate lines are geodesics. A priori, we know only that g12 = 0. However, much more can be said. ~ :U →V Proposition 8.5.4. Let S be a surface of class C 3 , and let X be a regular parametrization of class C 3 of a neighborhood V of S. ~ The parametrization X(u, v) is a system of geodesic coordinates in which the u-coordinate lines are geodesics if and only if the metric tensor is of the form   E(u) 0 g= . 0 G(u, v)

328

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

~ is an orthogonal if and only if Proof: First, the parametrization X g12 = 0. By Equation (8.7), along the u-coordinate lines, the geodesic curvature is  3 du √ Γ211 g11 g22 . κg = ds Since the metric tensor must be positive definite everywhere, g11 g22 is never 0. Since du/ds 6= 0, along a u-coordinate line, κg = 0 if and ∂g11 only if Γ211 = − 12 g 22 = 0. The result follows.  ∂v ds = du p E(u) is independent of v. Regardless of v, the arc length formula between u = u0 and u is Z up E(u) du. s(u) = Along u-coordinate lines, since v 0 = 0, the speed function

u0

Therefore, it is possible to reparametrize u along the u-coordinate lines by arc length, with u(s) = s−1 (u). This leads to the following proposition. ~ u, v) Proposition 8.5.5. Let S be a regular surface of class C 3 , and let X(¯ be a system of geodesic coordinates in a neighborhood V of S. Over the same neighborhood V , there exists a system of geodesic coor~ dinates X(u, v) such that the u-coordinate lines are geodesics parametrized by arc length. Furthermore, if this is the case, then the coefficients of the metric tensor are of the form g11 (u, v) = 1,

g12 (u, v) = 0,

g22 (u, v) = G(u, v),

and the Gaussian curvature of S is given by √ 1 ∂ 2 g22 . K = −√ g22 ∂u2

(8.36)

Proof: The fact that g11 (u, v) = 1 follows from the u-coordinate lines being parametrized by arc length and Equation (8.36) is a direct application of Equation (7.23). 

8.5. Geodesic Coordinates

329

8.5.2 Geodesic Polar Coordinates Let p be a point on a regular surface of class C 3 . By Theorem 8.4.10, for every ~v ∈ Tp S, there exists a unique geodesic ~γp,~v : (−ε, ε) → S, with ~γp,~v (0) = p. Furthermore, by Proposition 8.4.11, we know that 0 (t)k = k~ k~γp,~ v k. We remark then that for any constant scalar λ, v ~γp,~v (λt) = ~γp,λ~v (t) for all t ∈ (−ε/λ, ε/λ). Definition 8.5.6. Let S be a regular surface of class C 3 , and let p ∈ S

be a point. For any tangent ~v ∈ Tp S, we define the exponential map at p as ( p, if ~v = ~0, expp (~v ) = ~γp,~v (1), if ~v 6= ~0, whenever ~γp,~v (1) is well defined. The map expp : U → S, where U is a neighborhood of ~0 in Tp S corresponds to traveling along the geodesic through p with direction ~v over the distance k~v k. We will show that the exponential map can lead to some nice parametrizations of a neighborhood of p on S, but we first need to prove the following two propositions. Proposition 8.5.7. For all p ∈ S, the exponential map is defined over an open neighborhood U of ~0. Furthermore, if S is of class C 4 , then expp is differentiable over U as a function from Tp S into R3 .

Proof: We first show that expp is defined over some open disk centered at ~0. Let C1 ⊂ Tp S be the set of all vectors of unit length. Set R to be a large positive real number. For each ~v ∈ C1 , let ε(~v ) be the largest positive real ε ≤ R such that ~γp,~v : (−ε, ε) → S parametrizes a geodesic, or in other words, solves the system of equations in Equation (8.21). The theorems of existence and uniqueness of solutions to differential equations tell us that the solutions to Equation (8.21) depend continuously on the initial conditions, so ε(~v ) is continuous over C1 . (Setting R as an upper bound ensures that ε(~v ) is defined for all ~v ∈ C1 .)

330

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

Since C1 is a compact set, as a function to R, ε(~v ) attains a minimum ε0 on C1 . However, since ε(~v ) > 0 for all ~v ∈ C1 , then ε0 > 0. Consequently, expp is defined over the open ball Bε0 (~0) and is continuous. ~ : U 0 → S be a regular parametrization of a neighborhood Let X of p on S. According to the theorem of dependence of solutions to differential equations on initial conditions, the function ~γp (~v , t) : U × (−ε, ε) −→ R3 is of class C 1 in the initial conditions ~v if the Christoffel symbol ~ needs to be of class functions are of class C 2 , which means that X 4 C . The result follows.  Proposition 8.5.8. Let S be a regular surface of class C 4 . There is a

neighborhood U of ~0 in Tp S such that expp is a homeomorphism onto V = expp (U ), which is an open neighborhood of p on S. Proof: The tangent space Tp S is isomorphic as a vector space to R2 , and one can therefore view expp as a function from U 0 ⊂ R2 into R3 . Let ~v ∈ Tp S be a nonzero vector, and let α ~ (t) = t~v be defined for t ∈ (−ε, ε) for some ε > 0. Consider the curve on S defined by expp (~ α(t)) = expp (t~v ) = ~γp,t~v (1) = ~γp,~v (t). According to the chain rule,  d~ α d expp (t~v ) = d(expp )0 = d(expp )0 (~v ), dt dt 0 0 where ~v in this expression is viewed as an element in Tp S, and hence, by isomorphism, in R2 . However, by construction of the geodesics, 0 (0) = ~ ~γp,~ v , where we view ~v as an element of R3 . This result proves, v in particular, that d(expp )0 is nonsingular. Using the Implicit Function Theorem from analysis (see Theorem 8.27 in ), the fact that d(expp )0 is invertible implies that there exists an open neighborhood U of ~0 such that expp : U → expp (U ) is a bijection. Furthermore, setting V = expp (U ), by the Implicit Function Theorem, the inverse function exp−1 p : V → U is at least of class C 1 , and hence, it is continuous. Hence, expp is a homeomorphism between U and expp (U ). 

8.5. Geodesic Coordinates

331

The identification of Tp S with R2 allows one to define parametrizations of neighborhoods of p with some nice properties. Definition 8.5.9. Let S be a surface of class C 4 . A system of Riemann

normal coordinates of a neighborhood of p is a parametrization defined by ~ ~ 1 + vw ~ 2 ), X(u, v) = expp (uw where {w ~ 1, w ~ 2 } is an orthonormal basis of Tp S. Propositions 8.5.7 and 8.5.8 imply that Riemann normal coordinates provide a regular parametrization of a neighborhood V of p. Furthermore, the theorem of the dependence of solutions to differential equations on initial conditions shows that Riemann normal coordinates form a parametrization of class C r if S is a surface of class C r+3 . Definition 8.5.10. Let S be a surface of class C 4 . The geodesic polar

coordinates of a neighborhood of p give a parametrization defined by ~ θ) = expp ((r cos θ)w ~ 1 + (r sin θ)w ~ 2) , X(r, where {w ~ 1, w ~ 2 } is an orthonormal basis of Tp S. A curve on S that can be parametrized by ~γ (t) = expp ((R cos t)w ~ 1 + (R sin t)w ~ 2) ,

for t ∈ [0, 2π]

is called a geodesic circle of center p and radius R. Figure 8.13 gives an example of coordinate lines of a geodesic polar coordinate system on a torus. It is important to note that a geodesic circle is neither a geodesic curve on the surface nor a circle in R3 . We present the following propositions about the above coordinate systems but leave the proofs as exercises for the reader. Theorem 8.5.11. Let S be a regular surface of class C 5 , and let p be

a point on S. There exists an open neighborhood U of (0, 0) in R2 such that the Riemann normal coordinates defined in Definition 8.5.9 form a system of geodesic coordinates.

332

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

Figure 8.13. Geodesic polar coordinate lines on a torus.

Proof: This theorem follows immediately from Theorem 8.5.3, where one uses the curve α ~ (t) = expp (tw ~ 1 ), where w ~ 1 is a unit vector in Tp S.



Proposition 8.5.12. Let S be a regular surface of class C 5 , and let p

~ be a parametrization of a Riemann normal be a point on S. Let X ~ coordinate system in a neighborhood of p so that X(0, 0) = p, and ~ The let gij be the coefficients of the metric tensor associated to X. j coefficients satisfy gij (0, 0) = δi , and all the first partial derivatives of all the gij functions vanish at (0, 0). Proof: (Left as an exercise for the reader. See Problem 8.5.3.)



The next theorem discusses the existence of geodesic polar coordinate systems in a neighborhood of a point p ∈ S. Theorem 8.5.13. Let S be a regular surface of class C 5 , and let p be

a point on S. There exists an ε > 0 such that the parametrization ~ θ) in Definition 8.5.10, with 0 < r < ε and 0 < θ < 2π, is a X(r, regular parametrization and defines a geodesic coordinate system in a neighborhood V whose closure contains p in its interior. ~ θ) is defined as a function for −r0 < The parametrization X(r, r < r0 and θ ∈ R for some r0 > 0. However, similar to usual polar

8.5. Geodesic Coordinates

333

coordinates in R2 , one must restrict one’s attention to r > 0 and ~ is a homeomorphism. The image 0 < θ < 2π to ensure that X ~ ((0, ε) × (0, 2π)) is then a “geodesic disk” on S centered at p with X a “geodesic radius” removed. By definition, the parametrization ~ θ) is such that all the coordinate lines for one of the variables X(r, are geodesics, but Theorem 8.5.13 asserts that, within a small enough ~ θ) is regular and orthogonal. radius, the parametrization X(r, The existence of geodesic polar coordinates at any point p on a surface S of high enough class leads to interesting characterizations of the Gaussian curvature K of S at p. First, we remind the reader ~ θ) is a system of geodesic polar coordinates, then the that if X(r, coefficients of the associated metric tensor are of the form g11 (r, θ) = 1,

g12 (r, θ) = 0,

g22 (r, θ) = G(r, θ)

(8.37)

~ for some function G(r, θ) defined over the domain of X. ~ θ) be Proposition 8.5.14. Let S be a surface of class C 5 , and let X(r, a system of geodesic polar coordinates at p on S. Then the function G(r, θ) in Equation (8.37) satisfies p 1 G(r, θ) = r − K(p)r3 + R(r, θ), 6 where K(p) is the Gaussian curvature of S at pm and R(r, θ) is a function that satisfies lim

r→0

R(r, θ) = 0. r3

Proof: (Left as an exercise for the reader. See Problem 8.5.5.)



Proposition 8.5.14 shows that the perimeter of a geodesic circle at p of radius R is Z 2π p 1 G(R, θ) dθ = 2πR − K(p)πR3 + F (R) C= 3 0 where F (r) is a function of the radial variable r and satisfies limr→0 F (r)/r3 = 0. This, and a similar consideration of the area of geodesic disks around p, leads to the following geometric characterization of the Gaussian curvature at p.

334

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

Theorem 8.5.15. Let S be a surface of class C 5 , and let p ∈ S be a

point. Define C(r) (resp. A(r)) as the perimeter (resp. the area) of the geodesic circle (resp. disk) centered at p and of radius r. The Gaussian curvature K(p) of S at p satisfies     3 2πr − C(r) 12 πr2 − A(r) K(p) = lim and K(p) = lim . r→0 π r→0 π r3 r4 Proof: (Left as an exercise for the reader. See Problem 8.5.6.)



Theorem 8.5.15 is particularly interesting because, like the Theorema Egregium, it establishes the Gaussian curvature to a surface as a point as an intrinsic property of the surface, but in an original way. Geodesics and geodesic coordinate systems are intrinsic properties of the surface, and hence geodesic circles around a point are intrinsically defined. Some authors go so far as to use the second formula in Theorem 8.5.15 as a definition of the Gaussian curvature.

Problems 8.5.1. Let S be the unit sphere and suppose that p is a point on S. (a) Find a formula (using vectors) for the exponential map expp : Tp S → S. (b) Use this to give a formula for the geodesic polar coordinates (Definition 8.5.10) of a patch of S around p in terms of some ~ 2. w ~ 1 and w (c) Show that if p is the north pole, then with a proper choice of w ~ 1 and w ~ 2 we recover our usual colatitude–longitude parametrization of the sphere. ~ : U → V is a parametrization in which both fami8.5.2. Prove that if X lies of coordinate lines are families of geodesics, then the Gaussian curvature satisfies K(u, v) = 0. 8.5.3. Prove Proposition 8.5.12. [Hint: First use Equation (8.21) to prove that all Γijk vanish at (0, 0), i.e., at p.] 8.5.4. (*) Prove Theorem 8.5.13. 8.5.5. Prove Proposition 8.5.14. [Hint: Use a Taylor series expansion of G(r, θ) and Equation (8.36).] 8.5.6. Prove Theorem 8.5.15.

8.6. Applications to Plane, Spherical, and Elliptic Geometry

8.5.7. Consider the usual parametrization of the sphere S of radius R ~ X(u, v) = (R cos u sin v, R sin u sin v, R cos v). ~ Define the new parametrization Y (r, θ) = X(θ, r/R). (a) Prove that Y (r, θ) is a geodesic polar coordinate system of S at p = (R, 0, 0). (b) Prove the result of Proposition 8.5.14 directly and determine the corresponding remainder function R(r, θ).

8.6 Applications to Plane, Spherical, and Elliptic Geometry 8.6.1 Plane Geometry In plane geometry, we consider the plane to be a surface with Gaussian curvature identically 0 and curves in the plane we consider to be curves on a surface. Consider now a region R in the plane such that the boundary ∂R is a piecewise regular, simple, closed curve. It is easy to calculate that the Euler characteristic of a region that is homeomorphic to a disk is χ(R) = 1. As a first case, let ∂R be a polygon. Since the regular arcs are straight lines, the geodesic curvature of each regular arc of ∂R is identically 0. The Gauss-Bonnet formula then reduces to the wellknown fact that the sum of the exterior angles {θ1 , θ2 , . . . , θk } around a polygon is 2π. Also, since the exterior angle θi at any corner is π − αi , where αi is the interior angle, we deduce that the sum of the interior angles of an n-sided polygon is n X

αi = (n − 2)π.

i=1

This statement is in fact equivalent to a theorem that occurs in every high school geometry curriculum, namely, that the sum of the interior angles of a triangle is π radians. As a second case of the Gauss-Bonnet Theorem in plane geometry, suppose that ∂R is a simple, closed, regular curve of class C 2 . In this case, ∂R has no corners or cusps and hence no exterior angles.

335

336

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

Figure 8.14. Triangle on a sphere.

Assuming the boundary is oriented so that the interior R is in the ~ direction, then the Gauss-Bonnet Theorem reduces to the positive U formula I κg ds = 2π. ∂R

Therefore, the Gauss-Bonnet Theorem subsumes the proposition that the rotation index of a simple, closed, regular curve is 1 (see Propositions 2.2.1 and 2.2.11).

8.6.2 Spherical Geometry Ever since navigators confirmed that the Earth is (approximately) spherical, geometry of the sphere became an important area of study. Improper calculations could send explorers and navigators in drastically wrong directions. A key result particularly valuable to navigators is that the sum of the interior angles of a triangle is not equal to two right angles. Figure 8.14 gives an example of a triangle on a sphere in which each vertex has a right angle, and hence, the sum of the interior angles is 3π/2. Example 8.4.8 showed that a geodesic on a sphere is a great arc of the sphere, i.e., an arc on the circle of intersection between the sphere and a plane through the center of the sphere. Therefore, any

8.6. Applications to Plane, Spherical, and Elliptic Geometry

337

two geodesic lines intersect in at least two points: the common intersection of the sphere and the two planes supporting the geodesics. This remark, along with the fact that geometry on the sphere satisfies the first four postulates of Euclid’s Elements, justifies the claim that geometry of the sphere is an elliptic geometry. Consider now a triangle R on a sphere of radius R. By definition of a geometric triangle, ∂R has three vertices, and its regular arcs are geodesic curves. The Gaussian curvature of a sphere of radius R is R12 . In this situation, the Gauss-Bonnet Theorem gives (θ1 + θ2 + θ3 ) +

A = 2π, R2

(8.38)

where A is the surface area of R, and θi are the exterior angles of the vertices. With the interior angles αi = π − θi , then Equation (8.38) reduces to   A A α1 + α2 + α3 = π + 2 = π + 4π . R 4πR2 This shows that in a spherical triangle, the excess sum of the angles, i.e., α1 + α2 + α3 − π, is 4π times the ratio of the surface area of the triangle to the surface area of the whole sphere. As an example, the spherical triangle in Figure 8.14 has an excess of π/2 and covers 1/8 of the sphere. The quantity α1 + α2 + α3 − π is often referred to as the angle excess of a triangle on sphere. For navigators, the Gauss-Bonnet Theorem matters significantly because if they traveled in a circuit, generally following straight lines and only making sharp turns at discrete specific locations, the sum of the exterior angles of the circuit does not add up to 2π but to a quantity that is strictly less than 2π. We propose to study (geodesic) circles on a sphere of radius R. Recall that a circle on a sphere is the set of points that are at a fixed distance r from a point, where distance means geodesic distance. Call D the closed disk inside this circle. An example of a geodesic circle is any latitude line, since points on a latitude line are equidistant from the north pole. Using the parametric equations for the sphere, ~ X(u, v) = (R cos u sin v, R sin u sin v, R cos v),

338

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

a circle of radius r around the north pole (0, 0, R) is given by ~γ (t) = ~ v0 ). Note that the geodesic radius is the length of an arc of X(t, radius R and angle equal to the colatitude, so r = Rv0 . It is not hard to tell by symmetry that the geodesic curvature κg of a circle on a sphere is a constant. The Gauss-Bonnet Theorem then gives κg (perimeter of ∂D) +

1 (surface area of D) = 2π. R2

Using the surface area formula for a surface of revolution from singlevariable calculus, it is possible to prove (Problem 8.6.1) that the area of D is   r  surface area of D = 2πR2 (1 − cos v0 ) = 2πR2 1 − cos . R (8.39) As a circle in R3 , the radius is R sin v0 , so the perimeter of D is r perimeter of D = 2πR sin v0 = 2πR sin . R Consequently, the Gauss-Bonnet Theorem gives the geodesic curvature of a circle on a sphere as r 1 κg = cot . R R Using a Maclaurin series for the even function x cot x, we can show that this geodesic curvature is approximately   1  r 3 1  r −1 1  r  κg ≈ − − − ··· R R 3 R 45 R 1 1 1 3 ≈ − r− r − ··· 2 r 3R 45R4 In this expression, as r → 0, then κg behaves asymptotically like 1 1 r , where r is the usual curvature of a circle of radius r. Also, as R → ∞, which corresponds to a sphere with a radius that grows so large that the sphere approaches planar, we have r 1 1 lim κg = lim cot = , R→∞ R R→∞ R r which again recovers the familiar curvature of a circle of radius r.

8.6. Applications to Plane, Spherical, and Elliptic Geometry

8.6.3 Elliptic Geometry As mentioned in the introduction to this chapter, in his Elements, Euclid proves all his propositions from 23 definitions and five postulates. The fifth postulate reads as follows: If a straight line crossing two straight lines makes the interior angles on the same side less than two right angles, the two straight lines, if extended indefinitely, meet on that side on which are the angles less than the two right angles. This postulate is much wordier than the other four, and many mathematicians over the centuries attempted to prove the fifth postulate from the others, but never succeeded. Studying Euclid’s Elements, it appears that Euclid himself uses the fifth postulate sparingly. Proposition I.29 is the first proposition to cite Postulate 5. Furthermore, some proofs could be simplified if Euclid cited Postulate 5. One could speculate that Euclid avoided a liberal use of Postulate 5 so that, were someone to prove it as a consequence of the other four, then the proofs of theorems would still be given in a minimal form. Indeed, many commonly known properties about plane geometry hold without reference to the fifth postulate. However, examples of a few commonly known theorems that rely on it are: 1. Playfair Axiom: Given a line L and a point P not on L, there exists a unique line through P that does not intersect L. 2. The sum of the interior angles of a triangle is π. 3. Pythagorean Theorem: the squared length of the hypotenuse of a right triangle is the sum of the squares of the lengths of the sides adjacent to the right angle. In fact, in the above list, (1) is equivalent to the fifth postulate under the assumption that the first four postulates hold. That is why this theorem became known as the Playfair Axiom. In the 19th century, mathematicians Lobachevsky and Bolyai independently considered geometries that retained the first four postulates of Euclid’s Elements but assuming alternatives to the Playfair Axiom. The two logical alternatives to the Playfair Axiom are:

339

340

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

Elliptic Given a line L and a point P not on L, there does not exist a line through P that does not intersect L. Hyperbolic Given a line L and a point P not on L, there exists more than one line through P that does not intersect L. Depending on whether we use one or the other of the above two alternatives we obtain alternate geometries that are respectively called elliptic geometry and hyperbolic geometry. Collectively, elliptic and hyperbolic geometries were first called non-Euclidean geometry but then this label soon encompassed all types of geometry where the metric differs from the flat Euclidean metric. It is possible to prove many theorems in these different geometries using synthetic techniques, that is involving proofs that avoid the use of coordinates. (See , , or  for a comprehensive synthetic treatment of elliptic or hyperbolic geometry.) One well-known result in these geometries is that the sum of the interior angles of a triangle is greater than π in the case of elliptic geometry and less than π in the case of hyperbolic geometry. When Lobachevsky first investigated alternatives to Euclidean he considered hyperbolic geometry but rejected the elliptic hypothesis. Spherical geometry satisfies the elliptic hypothesis. However, in geometry on the sphere, any two great circles intersect at two points, and this does not satisfy the first axiom of Euclid. Consequently, it is possible that he believed that since spherical geometry has the elliptic hypothesis but fails Euclid’s first axiom, the elliptic geometry was inconsistent. However, in the late nineteenth century, it was realized that on a real projective plane obtained by identifying opposite points on the sphere, all of the first four Euclidean axioms remain valid and that the elliptic hypothesis holds as well. A number of the theorems that were known for spheres were valid in this elliptic geometry on a non-orientable surface. (Because a proper topological introduction to the real projective plane would take us too far afield, we do not describe it explicitly in this textbook but encourage the reader to discover it in subsequent studies in geometry.) We would like to now consider applications of the Gauss-Bonnet Theorem to elliptic and hyperbolic geometry. We first need to clarify the use of the terms “line,” “right angle,” and “parallel.” Let us assume that we have a metric. In the geometry on a surface of class

8.6. Applications to Plane, Spherical, and Elliptic Geometry

C 3 , the word “line” (or straight line) means a geodesic. Two lines meet at a right angle at a point p if the direction vectors of the lines at p have a first fundamental form that vanishes. The word “parallel” is more problematic. In Problem 1.3.16, we defined a parallel curve to a given curve ~γ as a curve whose locus is always a fixed orthogonal distance r from the locus of ~γ . This is true of parallel lines in the plane. However, in this sense of parallelism, a parallel curve to a line (geodesic) need not be another line. Consequently, we do not directly use the word parallel but prefer to discuss whether two lines intersect or not. Elliptic and hyperbolic geometry involve the notion of congruence. Recall that in any geometry that has a notion of congruence, the concept of area must satisfy the following three axioms: Axiom 1 The area of any set must be nonnegative. Axiom 2 The area of congruent sets must be the same. Axiom 3 The area of the union of disjoint sets must equal the sum of the areas of these sets. From Axiom 2 and Theorem 8.5.15, we can deduce that the Gaussian curvature is constant in elliptic and hyperbolic geometries. Consider a triangle R in a space of constant Gaussian curvature K. Because the edges of the triangle are geodesics, by the GaussBonnet Theorem, ZZ K dS + θ1 + θ2 + θ2 = 2πχ(R), R

where θi is the exterior angle at the ith vertex. Note that θi = π −αi , where αi are the interior angles of the triangle. Furthermore, the Euler characteristic of any triangle is χ(R) = 3 − 3 + 1 = 1. Thus K · Area(R) + (π − α1 ) + (π − α2 ) + (π − α3 ) = 2π =⇒ α1 + α2 + α3 = π + K · Area(R). We mentioned earlier that in elliptic (resp. hyperbolic) geometry, the interior sum of angles of a triangle is greater (resp. less) than pi. We deduce that in elliptic geometry the Gaussian curvature is

341

342

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

positive and constant, whereas in hyperbolic geometry, the Gaussian curvature is negative and constant. It is possible to prove from synthetic geometry in either elliptic or hyperbolic geometry) that the sum of the interior angles of a triangle satisfies α1 + α2 + α3 = π + c · Area(R), where c is a positive constant for elliptic geometry and a negative constant for hyperbolic geometry. However, the Gauss-Bonnet Theorem shows that this constant c is precisely the Gaussian curvature.

Problems 8.6.1. Calculate the surface area of a disk (Equation (8.39)) on a sphere using the surface area formula for a surface of revolution. Calculate directly the geodesic curvature function of a circle on a sphere. 8.6.2. Using Equation (8.39) for the surface area of a disk on a sphere and Theorem 8.5.15 recover the familiar result that the Gaussian curvature of a sphere of radius R is 1/R2 . 8.6.3. Suppose that a sphere of radius R has the longitude–colatitude (u, v) system of coordinates. Suppose that a simply connected region R of this sphere does not include the north or south pole and suppose that the boundary ∂R is a single regular, closed curve of class C 2 parametrized by (u(t), v(t)) for t ∈ [a, b] in its coordinate plane. Prove that the area of the region is 2πR2 + R2

Z

b

(cos v)u0 +

a

(cos v)u0 + u00 v 0 − u0 v 00 dt. (sin2 v)(u0 )2 + (v 0 )2

8.6.4. Using Euclidean style constructions, describe a procedure for finding the midpoint of a given great circle arc on the unit sphere with endpoints A and B.

8.7 Hyperbolic Geometry 8.7.1 Synthetic Hyperbolic Geometry Recall from the previous section that hyperbolic geometry is a geometry that involves the undefined terms of points, lines, and circles and that satisfies the first four postulates of Euclid’s Elements along

8.7. Hyperbolic Geometry

343

P `2

α

β

`1

` Q Figure 8.15. Left and right sensed-parallels.

with the fifth hyperbolic axiom: (H) for any line ` and any point P not on `, there pass more than one line that does intersect `. Synthetic hyperbolic geometry is like Euclidean geometry in its style of proof, in that it does not use coordinate systems. Various texts (see Chapter 2 in  for example) explore this geometry in detail. Some theorems are identical to Euclidean geometry if they only involve the first four postulates and some theorems differ considerably from Euclidean geometry. We mention a few results without proof in order to motivate the applications of the Gauss-Bonnet Theorem to hyperbolic geometry. Let ` be a line and let P be a point not on `. We can invoke Proposition I.12 in the Elements because its proof does not rely on the fifth postulate. Hence, we can construct a line through P that is ←→ perpendicular to `. In Figure 8.15, this is the line P Q. (Figure 8.15 only shows the segment P Q.) Obviously, the diagram in Figure 8.15 is sketched in the Euclidean plane as a “local approximation” and we must understand that as `1 and `2 are extended indefinitely, they do not intersect `. Consider the lines through P as they sweep out angles away from ←→ ←→ P Q. There must exist a first line on the right side of P Q that does not intersect ` and similarly on the left side. These lines are called the right (resp. left) sensed-parallel of ` through P ; they are indicated in Figure 8.15 by `1 (resp. `2 ). The angle α (resp. β) is called the angle of right (resp. left) parallelism. It is not hard to prove that

344

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

α = β and that this common angle is acute. All the lines through P that lie in the shaded region in Figure 8.15 are called ultraparallel to ` through P . In addition to usual triangles, in hyperbolic geometry, one also considers asymptotic triangles. If `1 and `2 are right (resp. left) sensed-parallels, then we say that they meet at an ideal point Ω. If −→ A is on `1 and B is on `2 , then the figure consisting of the ray AΩ, −→ the segment AB, and the ray BΩ is called an asymptotic triangle and denoted 4ABΩ. It is also possible to have asymptotic triangles with two or three ideal points as we shall see in models below. It is natural to try to prove theorems in hyperbolic geometry that mirror the development of Euclidean geometry as presented in The Elements. A number of surprising results arise. For example, in any triangle, the sum of the interior angles is less than two right angles. For a triangle 4ABC, the defect of 4ABC is the difference between π and the sum of the interior angles. Furthermore, with a few basic axioms of how area works, it is possible to prove that there exists a positive constant c such that for all triangles 4ABC, the area-defect formula holds Area(4ABC) = c2 defect(4ABC). In the previous section, we offered a reason why the Gaussian curvature in elliptic and in hyperbolic geometry should be constant. Under this assumption, we saw that because the sum of the interior angles of the triangle is less than π, then the Gaussian curvature in hyperbolic geometry must be negative. Let us suppose that K = − c12 in imitation of what happens for a sphere. Then using the GaussBonnet Theorem applied to triangle R in hyperbolic geometry, we get −

1 Area(R) + (π − α1 ) + (π − α2 ) + (π − α3 ) = 2π, c2

where αi are the interior angles of the triangle. Thus Area(R) = c2 (π − (α1 + α2 + α3 )), which recovers the area-defect formula. The Gauss-Bonnet approach is not a synthetic approach but it leads to the same relationship.

8.7. Hyperbolic Geometry

345

As mentioned earlier, the diagram offered in Figure 8.15 is only of limited value because if we extend `1 and ` for example, they intersect. The integrity of a synthetic proof cannot rely on a diagram but it is unfortunate that such a diagram fails to capture key properties, such as the angle defect of a triangle. This made it difficult for some mathematicians to initially accept hyperbolic geometry since it deviated from sense perception. However, it was soon discovered that by using different metrics, it is possible to depict the hyperbolic plane more effectively.

8.7.2 The Poincar´e Upper Half-Plane As early as Section 6.1, we introduced the alternate metric  0 1 v2

1  v2 gij =  0

on the upper half-plane H = {(u, v) ∈ R2 | v > 0}. We called this the Poincar´e upper half-plane. In Example 6.1.9, we calculated certain lengths of curves and also showed the area of the region a ≤ u ≤ b and c ≤ v ≤ d is   1 1 (b − a) − . c d The area of any region R in the plane is ZZ Area(R) = R

1 du dv. v2

It is possible to prove that, given this metric, the Gaussian curvature is constant with K = −1. Building on other examples in the text, in Exercise 8.4.11 we proved that geodesics in this space are either vertical lines or semicircles of radius R and with origin on the uaxis. Consequently, the Poincar´e upper half-plane offers a model of hyperbolic geometry in which “lines” consist of these geodesics. Interestingly enough, unlike for the cylinder in which there exist points such that there are many geodesics connecting them, given

346

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

Q2 Q P P2 A2

A

B

Figure 8.16. Geodesics in the Poincar´e half-plane.

any two points in the Poincar´e, there exists only one geodesic connecting them. In Problem 8.7.3, the reader is asked to prove a distance formula between points by calculating the length between the points along the unique geodesic that connects them. Finally, consider two vectors ~a and ~b based at a point p with coordinates (u0 , v0 ). We can understand ~a and ~b as vectors in the tangent plane to p. The angle θ between ~a and ~b satisfies Ip (~a, ~b) q cos θ = p Ip (~a, ~a) Ip (~b, ~b) 1 a b v02 1 1

+ v12 a2 b2 q0 =q 1 2 1 2 1 2 a + a a + v2 1 v2 2 v2 1 0

0

0

1 2 a v02 2

a1 b1 + a2 b2 p =p 2 . a1 + a22 b21 + b22 This is the usual formula for the cosine of an angle between two vectors in the plane. Consequently, the Poincar´e upper half-plane model of hyperbolic geometry accurately reflects angles as the angles measured in the usual sense. The following diagram depicts three lines (geodesics) in the Poincar´e upper half-plane model of hyperbolic geometry that meet to form a triangle. This particular triangle is a usual and not an asymptotic one.

8.7. Hyperbolic Geometry

347

C

B

A v=0

The reader might have guessed that there is a right angle at A and that would be correct. The circles in the above diagram were chosen specifically so that the two smaller circles meet at a right angle. Because of the above comment, since they meet at right angles as circles in the model shown in the Euclidean plane, the corresponding lines in the hyperbolic geometry meet at right angles as well. (It is easy to prove in Euclidean geometry that two circles of radius r and R √ respectively intersect orthogonally if the distance between them is r2 + R2 .) On the other hand, in the diagrams below, the first figure depicts an asymptotic triangle with one ideal point, the second figure depicts an asymptotic triangle with two ideal points for vertices, and the third figure depicts an asymptotic triangle with three ideal points.

A

A B Ω1

Ω1

Ω2

Ω3

Ω2

348

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

8.7.3 The Poincar´e Disk Another model for hyperbolic geometry is to model the points in the hyperbolic plane as the points in the open disk of radius R, namely DR = {(u, v) ∈ R2 | u2 + v 2 < R2 } but to use the metric tensor   4R4 0   gij =  (R2 − x2 − y 2 )2 . 4 4R 0 (R2 −x2 −y 2 )2 The disk DR equipped with this metric is called the Poincar´e disk. If (u(t), v(t)) parametrizes a curve in DR , then the length of the curve with t1 ≤ t ≤ t2 is Z t2 p 2R2 (u0 (t))2 + (v 0 (t))2 dt. 2 2 2 t1 (R − u(t) − v(t) ) Consequently, as the curve approaches the boundary of the disk DR , the length becomes arbitrarily large. Hence, we can intuitively think of the boundary ∂DR as heading far away, becoming unbounded. We leave it as an exercise for the reader to compute the Christoffel symbols and to prove that the Gaussian curvature of the Poincar´e disk is −1/R2 . Consequently, the Poincar´e disk has constant negative Gaussian curvature. (If R is unspecified, we assume that R = 1 and that D is the unit disk.) Using the so called M¨obius transformations from complex analysis (a technique that is just beyond the scope of this book) from the Poincar´e upper half-plane to the Poincar´e disk, it is possible to prove that the geodesics on the Poincar´e are either diameters of the disk or arcs of circles that meet the boundary ∂D at right angles.

8.7. Hyperbolic Geometry

Figure 8.17. Tiling of the Poincar´e disk with regular pentagons.

The diagram above depicts a (non-asymptotic) triangle in the Poincar´e disk. An asymptotic triangle in the Poincar´e disk involves geodesic edges that meet at the boundary of D. Because of the relationship between the defect of triangles and by extension the defect of other polygons, it is possible to tessellate (tile) the hyperbolic plane with regular polygons in ways that are impossible in the Euclidean plane. For example, it is possible to tessellate the hyperbolic plane with pentagons in such a way that at each vertex of the tessellation, four regular pentagons meet. (See Figure 8.17.) Since four pentagons meet at a vertex, each interior angle of the pentagon is π2 . Each pentagon is the union of five isosceles triangles, each congruent to each other, meeting at the center of the pentagon. Each of these triangles has for interior angles π4 , π4 , π 2 π and 2π 5 . Hence the defect of each triangle is 10 and its area is c 10 , 1 when the Gaussian curvature is a constant K = − c2 . Thus, in order to tessellate the hyperbolic plane when K = −1 with pentagons such that four meet at each vertex, we simply must choose regular pentagons of area π2 . Artist M. C. Escher (1891–1972) was known for artwork that explores mathematical symmetries in creative ways. Among a few of his more esoteric works are symmetry patterns based on tessellations of the hyperbolic plane.

349

350

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

Figure 8.18. A geodesic triangle on a pseudosphere.

8.7.4 The Pseudosphere Revisited The Poincar´e upper half-plane and Poincar´e disk offer effective models of the hyperbolic plane. One may feel still somewhat dissatisfied because these models do not correspond to regular surfaces in R3 . The pseudosphere introduced in Example 6.6.6 is a regular surface in R3 that has constant Gaussian curvature equal to K = −1. Consequently, the pseudosphere offers another model of hyperbolic space. Figure 8.18 depicts a triangle (the edges are geodesics) on the pseudosphere. Problem 8.7.8 guides the reader to explore some of the properties of the pseudosphere starting from its metric.

Problems 8.7.1. In the Poincar´e upper half-plane, consider the isosceles triangle whose vertices are where the following three lines (geodesics) meet: x2 + y 2 = 36, (x − 3)2 + y 2 = 16, and (x + 3)2 + y 2 = 16. Calculate the area of this triangle. 8.7.2. Consider the Poincar´e upper half-plane. (a) Find the geodesic curvature of the line (u(t), v(t)) = (t, mt), with t > 0, where m is some constant. R (b) Explicitly evaluate the line integral ∂R κg ds, where R is the region bounded by three Euclidean straight lines from (1, 1) to (2, 2) to (1, 2) and back to (1, 1).

8.7. Hyperbolic Geometry

351

RR R

K dS.

(d) Directly evaluate the double integral in the previous part to confirm your result. [Recall that K = −1.] 8.7.3. Consider the Poincar´e half-plane. Let P and Q be two points on H. We determine the distance d(P, Q) between these two points in the Poincar´e metric on H (see Figure 8.16). (a) If P and Q lie on a geodesic that is a half-circle, prove that the distance between them is |P A|/|P B| d(P, Q) = ln , |QA|/|QB| where A and B are points as shown in Figure 8.16 and |P A| is the usual Euclidean distance between P and A and similarly for all the others. (b) If P and Q lie on a geodesic that is a vertical line, prove that the distance between them is |P A | 2 2 d(P, Q) = ln , |Q2 A2 | where A2 is again as shown in Figure 8.16. 8.7.4. Describe a procedure for finding the midpoint of a given hyperbolic segment on the Poincar´e upper half-plane. In particular,  find the 1 1 midpoint of the hyperbolic segment from (0, 1) to √2 , √2 . [Hint: See Problem 8.7.3.] 8.7.5. Show that the set of points at a fixed distance d from the vertical line u = 0 in the hyperbolic plane consists of the points on two Euclidean lines of the form y = mx. 8.7.6. Using the Poincar´e upper half-plane, Figure 8.19 depicts two lines `1 ←→ and `2 that are parallel (right-sensed). The line AB is perpendicular ←→ to `1 so the acute angle between AB and `2 is the angle of parallelism. Suppose that in this diagram `1 is a circle of radius 1 and let h be the distance between A and B. Calculate the angle of parallelism in terms of the distance h. Show that as h → 0, the angle of parallelism goes to π/2, and as h → ∞, the angle of parallelism goes to 0. 8.7.7. Prove that in the Poincar´e disk, the angle at which curves meet as observed in the model in R2 is precisely the angle at which the curves meet in the Poincar´e disk metric.

352

8. The Gauss-Bonnet Theorem and Geometry of Geodesics

B h

`2

A `1

Ω Figure 8.19. Angle of parallelism.

8.7.8. Consider the upper half-plane H = {(x, y) ∈ R2 | y > 0} with the metric tensor  −2y  e 0 (gij ) = . 0 1 (a) Prove that the Gaussian curvature of such a surface has K = −1 everywhere. ~ : H → R3 defined by (b) Show that the parametrization X   p p ~ X(x, y) = e−y cos x, e−y sin x, ln(1 + 1 − e−2y ) + y − 1 − e−2y has the above metric coefficients. ~ (c) Prove that X(x, y) is a regular reparametrization of the pseudosphere as described in Example 6.6.6. (d) (ODE,*) Find parametric equations (u(t), v(t)) for the geodesic curves on H with this given metric. 8.7.9. Consider the Poincar´e half-plane. We can consider R2 as the set of complex numbers C. In this context, H = {z ∈ C | Im(z) > 0}. Consider a fractional linear transformation of C of the form w=

az + b , cz + d

where a, b, c, d ∈ R and ad − bc = 1. Write z = u + iv and w = x + iy. (a) Prove that the function w = f (z) sends H into H bijectively.

8.7. Hyperbolic Geometry

(b) Prove that the metric tensor for H is unchanged under this transformation, more precisely, that the metric coefficients of H in the w-coordinate are   −2 0 y . (¯ gkl ) = 0 y −2 [We say that the Poincar´e metric on the upper half-plane is invariant under the action of SL(2, R).] (c) Show explicitly that this fractional linear transformation sends geodesics to geodesics.

353

CHAPTER 9

Curves and Surfaces in n-dimensional Euclidean Space

Up to this point, this text emphasized curves and surfaces in R3 . However, our only reason to do so was based on a discrimination by dimensionality: As beings living in three-dimensional space, we are accustomed to visualizing in R3 . In this chapter, we generalize a number of definitions and theorems about curves and surfaces to Euclidean n-dimensional space.

9.1 Curves in n-dimensional Euclidean Space In Chapters 1 and 3, we studied local properties of curves with an emphasis on notions of velocity, acceleration, speed, curvature, and torsion. Along the way, we introduced the natural orthonormal frames ~ ) for a regular plane attached to curves at a point, namely the (T~ , U ~ for a regular space curve of curve and the Frenet frame (T~ , P~ , B) 2 3 class C in R . Many local properties and definitions have immediate generalizations to curves in n-dimensional Euclidean space, Rn . Let I be an ~ : I → Rn be a parametrized curve. Then interval of R and let X ~ 0 (t); • the velocity is the vector function X ~ 0 (t)k; • the speed is real-valued function s0 (t) = kX • the arc length function with origin t = t0 , is the antiderivative of s0 (t) with s(t0 ) = 0; ~ 00 (t); • the acceleration is the vector function X

355

356

9. Curves and Surfaces in n-dimensional Euclidean Space

~ is regular at t = t0 if X ~ 0 (t0 ) exists and • the parametrization X is nonzero; ~ is regular if it is regular at all t0 in its • the parametrization X domain; • a curve C in Rn is regular if it is the locus (image) of a regular parametrized curve; • at any regular point t = t0 , the unit tangent vector is ~ 0 (t0 ) X T~ (t0 ) = . ~ 0 (t0 )k kX One hitch that occurs in generalizing our theory from curves in space to curves in Rn comes from the fact that for n 6= 3 there does not exist a cross product in Rn . So any formula or reasoning must avoid that nice property about R3 .

9.1.1 Curvatures and the Frenet Frame ~ : I → Rn satisfies Y ~ (t)· Y ~ (t) = Recall that any unit vector function Y 1 so after differentiating this expression, for all t ∈ I, ~ (t) = 0 =⇒ Y ~ 0 (t) ⊥ Y ~ (t). ~ 0 (t) · Y 2Y

(9.1)

Thus, if the derivative exists, T~ 0 (t) is perpendicular to T~ (t), though T~ 0 (t) is not necessarily of unit length. ~ Suppose that X(t) is a regular parametrized curve of class C 2 . ~ We define the first curvature κ1 (t) of X(t) implicitly as the unique nonnegative function such that T~ 0 (t) = s0 (t)κ1 (t)P~1 (t), and where P~1 (t) is a unit vector function. This implicit definition might not produce a well-defined vector at P~1 (t0 ) if κ1 (t0 ) = 0, i.e., if T~ 0 (t0 ) = ~0. This problem can be remedied if T~ 0 (t) 6= ~0 for 0 < |t − t0 |ε, for some positive ε. In this case, P~1 (t) is a welldefined and continuous vector function on the unit sphere in Rn for 0 < |t − t0 | < ε. Then a value for P~1 (t0 ) can be assigned by

9.1. Curves in n-dimensional Euclidean Space

357

completing P~1 (t) by continuity. On the other hand, if κ1 (t) = 0 over an interval of t, then for the points in this interval, the vector function P~1 (t) is undefined. As with curves in R3 , assuming the derivatives exist, we have ~ 0 (t) = s0 (t)T~ (t) X ~ 00 (t) = s00 (t)T~ (t) + s0 (t)2 κ1 (t)P~1 (t). X The vector function P~1 (t) is called the first normal or first principal ~ normal vector function. As we consider higher derivatives of X(t) in reference to this type of decomposition, we need to make sense of the derivative of P~1 (t) and the derivatives of subsequent vector functions that arise in a similar way. ~ (t) and Z(t) ~ Now if Y are unit vector functions with the same domain interval I and are everywhere perpendicular to each other, then for all t ∈ I, ~ (t) · Z(t) ~ ~ 0 (t) · Z(t) ~ +Y ~ (t) · Z ~ 0 (t) = 0 Y = 0 =⇒ Y ~ 0 (t) · Z(t) ~ ~ (t) · Z ~ 0 (t). =⇒ Y = −Y

(9.2)

Now consider the vector function P~10 (t). From Equation (9.1), we know that P~10 (t) · P~1 (t) = 0 and from (9.2), we have P~10 (t) · T~ (t) = ~ −s0 (t)κ1 (t). If n > 3, we now define the second curvature of X(t) as ≥0 the unique nonnegative function κ2 : I → R such that P~10 (t) = −s0 (t)κ1 (t)T~ (t) + s0 (t)κ2 (t)P~2 (t), for some unit vector function P~2 (t) that is perpendicular to both T~ and P~1 . As long as κ2 (t) is not zero, P~2 (t) is well-defined. It is called the second normal or second principal normal vector function. By repeating this process, we recursively define the ith curvature function and the associated ith unit normal vector function, for 1 ≤ i ≤ n−1. However, an exception is made for the (n−1)th “curvature” and normal vector. In defining T~ , P~1 , up to P~n−1 as above, we establish an orthonormal set of vectors (unit length and mutually perpendicular). By construction, the n-tuple of vectors organized into a matrix as   A(t) = T~ (t) P~1 (t) · · · P~n−1 (t) (9.3)

358

9. Curves and Surfaces in n-dimensional Euclidean Space

defines an orthogonal matrix. Orthogonal matrices can have a determinant that is 1 or −1. Knowing the vectors T~ through P~n−2 along with the condition that A(t) be orthogonal leaves exactly two possibilities for P~n−1 . For calculations and other reasons, it is desirable that A(t) be a positive orthogonal matrix, i.e., satisfy det A(t) = 1. Thus, with the requirement that det A(t) = 1, the (n − 1)th normal vector P~n−1 is uniquely determined from T~ through P~n−2 . However, because we use T~ through P~n−2 to determine P~n−1 (t), then we cannot define the (n − 1)th curvature function κn−1 (t) to be a nonnegative function. ~ : I → Rn be a parametrized curve. The nDefinition 9.1.1. Let X

tuple of vector functions (T~ (t), P~1 (t), · · · , P~n−1 (t)) as constructed ~ above is called the Frenet frame of X. As with Equation (3.5), we can summarize the derivatives of the Frenet frame vectors by   0 −s0 κ1 0 ··· 0 0  s0 κ1 0 −s0 κ2 · · · 0 0   0   s κ2 0 ··· 0 0 d   0 A(t) = A(t)  . . . . . . . . . . . . .   . dt . . . . .   0  0 0 0 ··· 0 −s κn−1  0 0 0 · · · s0 κn−1 0 where A(t) is the positive orthogonal matrix defined in Equation (9.3). See  for the original introduction to the notion of higher curvatures of curves in Rn . Equation (3.9) for a formula of the curvature of a space curve relied on taking the cross product of two vectors. However, this is not a valid operation in Rn for n > 3. Consequently, we need another formula for κ1 (t). Note that as before, we have ~ 0 = s0 T~ X ~ 00 = s00 T~ + (s0 )2 κ1 P~1 . X The vector formula of Problem 3.1.6 shows ~ ~a · ~a ~a · b k~a × ~bk2 = ~ . b · ~a ~b · ~b

9.1. Curves in n-dimensional Euclidean Space

359

~0×X ~ 00 k, we are inspired to consider Since Formula (3.9) involves kX the following calculation: ~ 0 ~ 0 ~ 0 ~ 00 0 2 · X · X s0 s00 X X (s ) ~ 00 ~ 0 ~ 00 ~ 00 = 0 00 00 2 0 4 2 ss (s ) + (s ) κ1 X · X X · X = (s0 )2 (s00 )2 + (s0 )6 κ21 − (s0 )2 (s00 )2 = (s0 )6 κ21 . This leads to the following formula for first curvature: v u q ~0 X ~ 00 ~0 ·X ~0 ·X 1 1 u X t κ1 (t) = det(B(t)> B(t)), ~ 00 ~ 0 ~ 00 ~ 00 = ~ 0 k3 X ~ 0 k3 · X X · X kX kX   ~ 0 (t) X ~ 00 (t) . In order to prove where B(t) is the n × 2 matrix X the more general formula, we must use the Cauchy-Binet formula. Theorem 9.1.2 (Cauchy-Binet). Define [m] = {1, 2, . . . , m}. If M is an n × m matrix and S ⊆ [n] and S 0 ⊆ [m], use the notation MS,S 0 to mean the submatrix consisting of the entries from the rows taken from the set S with columns taken from the set S 0 . If A is an m × n matrix and B an n × m matrix with m ≤ n, then X det(A[m],S ) det(BS,[m] ). det(AB) = S⊆[n] |S|=m

Proof: (A proof can be found in the appendices of .)



~ : I → Rn be a regular parametrization of Proposition 9.1.3. Let X n class C n of a curve in   R for n ≥ 2. Define the matrix Bm (t) = ~0 X ~ (m) for any integer m with 1 ≤ m ≤ n. Then if ~ 00 · · · X X 1 ≤ m ≤ n − 2, q s0 (t)(m+1)(m+2)/2 κ1 (t)m κ2 (t)m−1 · · · κm (t) = det(Bm+1 (t)> Bm+1 (t)).

and for the n − 1 curvature function κn−1 (t), we have s0 (t)n(n+1)/2 κ1 (t)n−1 κ2 (t)n−2 · · · κn−1 (t) = det(Bn (t)).

360

9. Curves and Surfaces in n-dimensional Euclidean Space

Proof: Since the standard frame and the Frenet frame are orthonormal bases, there exists an orthogonal matrix M (t) that gives the transition matrix from coordinates in the Frenet frame to standard coordinates. By the definition of the Frenet frame, Bm (t) = M (t)Dm (t) where Dm (t) is an n × m upper triangular matrix (in the sense that entries are 0 below the main diagonal). For example, if n = 4 then, using Equations (3.6), (3.7), and (3.10) as generalized to R4 , we have  0  0   s s s00 s00 s000 − (s0 )3 κ21 0 2 00 0 2 0 2 0  0 (s0 )2 κ1    , D3 =  0 (s ) κ1 3s (s ) κ1 + (s ) κ1  . D2 =  0 3 0    0 0 0 (s ) κ1 κ2 0 0 0 0 0 By the manner in which we defined the curvatures, we can see that the (1, 1) entry of Dm (t) is s0 and that the jth diagonal entry of ~ (j) · P~j−1 . By repeatedly using the orthonormality propDm (t) is X erty of the Frenet frame, we obtain ~ (j) · P~j−1 = s0 T~ (j−1) · P~j−1 X (j−2) ~ = (s0 )2 κ1 P~1 · Pj−1 (j−3) ~ = (s0 )3 κ1 κ2 P~2 · Pj−1 .. . =

= (s0 )j κ1 κ2 · · · κj−1 . Since M (t) is an orthogonal matrix, Bm (t)> Bm (t) = Dm (t)> M (t)> M (t)Dm (t) = Dm (t)> Dm (t). Writing simply D for Dm (t), by the Cauchy-Binet Theorem, X X det(D> D) = det((DS,[m] )> ) det(DS,[m] ) = det(DS,[m] )2 . S⊆[n] |S|=m

S⊆[n] |S|=m

However, since D is upper triangular, if S 6= [m], then DS,[m] contains a row of 0s and hence its determinant vanishes. Thus det(D> D) = det(D[m],[m] )2 , which is the square of the product of its diagonal

9.1. Curves in n-dimensional Euclidean Space

361

elements. Having determined the diagonal elements above, and since all the functions involved are nonnegative, q Pm det(Bm (t)> Bm (t)) = (s0 ) j=1 j κm−1 κm−2 · · · κm−1 1 2 = (s0 )m(m+1)/2 κm−1 κm−2 · · · κm−1 . 1 2 The first part of the proposition follows. We can use the same recursive formula for κn−1 simply because κn−1 (t) is not necessarily a nonnegative function. However, in this case Bn (t) is a square matrix. Also, Bn (t) = M (t)Dn (t) and since M (t) is orthogonal det(Bn (t)) = det(Dn (t)). However, Dn (t) is upper triangular so its determinant is the product of its diagonal elements. Thus det(Bn (t)) = (s0 )n(n+1)/2 κn−1 κn−2 · · · κn−1 . 1 2



Proposition 9.1.3 gives a recursive formula for the higher curvatures and consequently it is a straightforward exercise, though sometimes tedious to calculate the curvature functions of a parametrized curve in Rn . The reader should also note that Proposition 9.1.3 directly generalizes the formula for curvature of a plane curve (Equation (1.12)), the formula for the curvature of a space curve (Equation (3.9)), and the formula for the torsion of a space curve (Equation (3.2)).

9.1.2 Osculating Planes, Circles, and Spheres Following the local theory of plane curves and space curves as developed earlier in the text, for curves in Rn we can consider osculating k-planes or osculating k-spheres, where k is any integer with 1 ≤ k ≤ n − 1. ~ : I → Rn be a regular parametrized curve Definition 9.1.4. Let X

~ at t = t0 is the k-plane of class C n . The osculating k-plane to X ~ 0 ) and spanned by the vectors T~ (t0 ), P~1 (t0 ), through the point X(t ~ . . . , Pk−1 (t0 ). Note that the osculating line to a curve at a point is simply the tangent line.

362

9. Curves and Surfaces in n-dimensional Euclidean Space

We call a k-sphere in Rn , any k-dimensional sphere in a (k + 1) dimensional plane in Rn . Note that in this indexing, a circle is a 1sphere (in a 2-plane), a usual sphere is a 2-sphere (in a 3-plane), and so forth. We leave as exercises to the reader formulas for osculating k-spheres.

9.1.3 The Fundamental Theorem of Curves in Rn The Fundamental Theorem of Space Curves immediately generalizes to an n dimensional case. The proof of this generalization involves no new strategy so we omit the proof. Theorem 9.1.5 (Fundamental Theorem of Curves). Given curvature func-

tions κi (s) ≥ 0, for 1 ≤ i ≤ n − 2 and κn−1 (s) continuously differentiable over some interval J ⊆ R containing 0, there exists an open ~ : I → Rn interval I containing 0 and a regular vector function X that parametrizes its locus by arc length, with κi (s) as the ith curvature functions. Furthermore, any two curves C1 and C2 with the same (n − 1)-tuple of curvature functions can be mapped onto one another by a rigid motion of Rn . For the same reason as in the case of space curves, we call the (n − 1)-tuple of curvature functions (κ1 (s), κ2 (s), . . . , κn−1 (s)) the natural equations of the curve in Rn .

Problems ~ 9.1.1. In R4 , consider the twisted quartic X(t) = (t, t2 , t3 , t4 ), for t ∈ R. Calculate the three curvature functions, namely κ1 (t), κ2 (t), and κ3 (t). [This “twisted quartic” is usually called the rational normal curve of dimension 4.] 9.1.2. Calculate all three curvature functions κ1 (t), κ2 (t), and κ3 (t) asso~ ciated to the regular curve X(t) = (R cos t, R sin t, r cos t, r sin t) in 4 R , where R, r are positive real constants. 9.1.3. Calculate all three curvature functions κ1 (t), κ2 (t), and κ3 (t) asso~ ciated to the regular curve X(t) = (cos t, sin t, cos(2t), sin(2t)). 9.1.4. Let m and n be positive integers. Calculate all three curvature functions κ1 (t), κ2 (t), and κ3 (t) associated to the regular curve ~ X(t) = (cos(mt), sin(mt), cos(nt), sin(nt)).

9.1. Curves in n-dimensional Euclidean Space

9.1.5. Consider regular curves in Rn . Use Proposition 9.1.3 to show that the curvature functions are invariant under a positively oriented reparametrization. Prove that under a negatively oriented reparametrization the curvatures κ1 (t), . . . , κn−2 (t) remain invariant and that the n − 1 curvature changes according to κ ¯ n−1 (t) = (−1)n κn−1 (t). 9.1.6. Proposition 9.1.3 defines curvatures in a recursive way but does not show any simplifications if such exist. Show an explicit formula for κ3 (t) for curves in Rn , first when n = 4 and then for n ≥ 5. ~ 9.1.7. Calculate the osculating 2-sphere to X(t) = (t, t2 , t3 , t4 ) in R4 at t = 0. How is this similar or different from the osculating 2-sphere ~ to X(t) = (t, t2 , t3 ) in R3 ? [See Problem 3.3.6.] 9.1.8. Follow the ideas in Section 3.3 to find the center and the radius of the osculating 3-sphere to a curve in R4 . [Note that the osculating 3-sphere to a curve C at a point P will be a 3-sphere with contact of order 4 to C at P .] 9.1.9. Let A(t) and B(t) be an n × n matrix of differentiable real-valued functions defined in an interval I ⊆ R. If A(t) = (aij (t)), then denote by A0 (t) the n × n matrix (a0ij (t)). Prove that: (a)

d > dt (A(t) )

= (A0 (t))> ;

(b)

d dt (A(t)

(c)

d dt (A(t)B(t))

(d)

d −1 ) dt (A(t)

+ B(t)) = A0 (t) + B 0 (t); = A0 (t)B(t) + A(t)B 0 (t);

= −A(t)−1 A0 (t)A(t)−1 , if A(t) is invertible over I.

9.1.10. Suppose that A(t) is an n × n matrix of differentiable real-valued functions defined in an interval I ⊆ R that is everywhere orthogonal, i.e. A(t)> A(t) = I. Use the previous exercise to prove that A0 (t) = A(t)M (t), where M (t) is an antisymmetric matrix (M (t)> = −M (t)). 9.1.11. Prove that a curve in Rn lies in a k-dimensional “plane” (a subspace of Rn of the form p~ + W , where W is a k-dimensional subspace) if and only if κk (s) = · · · = κn−1 (s) = 0. 9.1.12. (*) Find a parametrized curve in R4 that has constant and nonzero curvature functions κ1 , κ2 , and κ3 . [Hint: This will be a fourdimensional equivalent of a helix.]

363

364

9. Curves and Surfaces in n-dimensional Euclidean Space

9.2 Surfaces in n-dimensional Euclidean Space 9.2.1 Regular Surfaces in Rn In Chapter 5, we introduced the concept of a regular surface in R3 . As an overarching intuition, a regular surface in R3 is a set of points S obtained from a parametrization such that at each point p ∈ S the set of tangent vectors forms a plane. In order to make all the notions in the previous sentence precise we needed to discuss parametrizations of surfaces, which in turn made sense of tangent vectors to surfaces at a point, and address the conditions under which the set of tangent vectors forms a tangent plane. This culminated in Definition 5.2.10. Subsequent to establishing a workable definition for a regular surface, we proceeded to define orientability, the metric tensor, and then a variety of intrinsic and extrinsic properties of surfaces. Our investigations culminated in the Gauss-Bonnet Theorem, from which we gave a variety of applications. Most of the presentation given for regular surfaces in R3 can directly generalize to define regular surfaces in Rn with n > 3, with one main exception. The difference between the case of surfaces in R3 and surfaces in Rn comes from two reasons that are intimately related: • there is no way to naturally define a vector cross product between two vectors in Rn ; • the vector space of perpendicular (normal) vectors to a plane is not one-dimensional. Given any two-dimensional plane of vectors in Rn , the set of vectors perpendicular to the plane has dimension n − 2. So if n > 3, then n − 2 > 1. ~ : In Proposition 5.2.4, we proved that a parametrized surface X 3 2 U → R , where U is an open set in R , has a tangent plane at ~ −1 (p) is a single point q = (u0 , v0 ) ∈ U and X ~ u (u0 , v0 ) × p if X ~ Xv (u0 , v0 ) 6= ~0. Subsequently, we gave an alternate characterization which led to the definition of a regular surface, Definition 5.2.10. Happily, that alternate characterization generalizes with hardly any changes to a definition in Rn .

9.2. Surfaces in n-dimensional Euclidean Space

Definition 9.2.1. A subset S ⊆ Rn is a regular surface if for each p ∈ S,

there exists an open set U ⊆ R2 , an open neighborhood V of p in ~ : U → V ∩ S such that Rn , and a surjective continuous function X ~ is continuously differentiable: if we write X(u, ~ 1. X v) = (x1 (u, v), x2 (u, v), . . . , xn (u, v)), then for i = 1, 2, . . . , n, the functions xi (u, v) have continuous partial derivatives with respect to u and v; ~ is a homeomorphism: X ~ is continuous and has an inverse 2. X −1 ~ ~ −1 is continuous; X : V ∩ S → U such that X ~ satisfies the regularity condition: for each (u, v) ∈ U , the 3. X ~ (u,v) : R2 → Rn is a one-to-one linear transfordifferential dX mation. Furthermore, a regular surface is said to be of class C n (resp. C ∞ ) ~ : U → Rn has continuous n derivatives (resp. if each function X continuous derivatives of all orders). ~ u (u, v) and The regularity condition can be restated to say that X ~ Xu (u, v) are linearly dependent over the domain. The concepts of a regular, positively oriented, or negatively oriented reparametrization of a surface carry over in an identical fashion from surfaces in R3 to surfaces in Rn .

9.2.2 Intrinsic Geometry for Surfaces Recall that intrinsic geometry is a property of surfaces that depend entirely on the metric coefficients (and the partial derivatives thereof) of the regular. Since intrinsic properties do not depend on a normal vector to a surface at a point, then such properties apply to any surface as long as we have a metric tensor. For a regular surface in Rn , we define the first fundamental form in precisely the same way as we did for surfaces in R3 . The first fundamental form Ip ( , ) to a regular surface S at p is the inner product on Tp S obtained as the restriction of the dot product in Rn to the tangent plane Tp S.

365

366

9. Curves and Surfaces in n-dimensional Euclidean Space

Suppose that a coordinate patch of S has a regular parame~ : U → Rn . Suppose also that p = X(u ~ 0 , v0 ) with trization X (u0 , v0 ) ∈ U is a point on the surface. Consider the standard or~ u (u0 , v0 ), X ~ v (u0 , v0 )) of the tangent space Tp S. dered basis B = (X Suppose also that two vectors ~a, ~b ∈ Tp (S) are expressed with the following components with respect to B,   a1 [~a]B = a2

and

[~b]B =

  b1 . b2

Then the first fundamental form has Ip (~a, ~b) = ~a · ~b ~ u (u0 , v0 ) + a2 X(u ~ 0 , v0 )) · (b1 X ~ u (u0 , v0 ) + b2 X(u ~ 0 , v0 )) = (a1 X !  ~ u (u0 , v0 ) X ~ u (u0 , v0 ) · X ~ v (u0 , v0 ) ~ u (u0 , v0 ) · X  X b1 = a1 a2 ~ ~ ~ ~ b2 Xv (u0 , v0 ) · Xu (u0 , v0 ) Xv (u0 , v0 ) · Xv (u0 , v0 )    b = a1 a2 g(u0 , v0 ) 1 , b2 where we define the metric tensor g = (gij ) as the matrix of functions defined on U ⊆ R2 ! ~ u (u, v) X ~ u (u, v) · X ~ v (u, v) ~ u (u, v) · X X g(u, v) = ~ ~ u (u, v) X ~ v (u, v) · X ~ v (u, v) . Xv (u, v) · X Since the metric tensor depends on dot products, which exist in any Rn , the usual definition of the metric tensor remains unchanged in any number of dimensions. Furthermore, the uses and interpretation of the metric remain unchanged. Equations (6.1) and (6.2) for the arc length of a curve on a regular surface remain identical for surfaces in Rn with exactly the same proof. This is also true for Equation (6.3) to calculate angles between tangent vectors. However, Proposition 6.1.7 concerning the area is ~ u ×X ~v still true but its original proof referred to the cross product of X in order to find the area element.

9.2. Surfaces in n-dimensional Euclidean Space

Proposition 9.2.2. Let S be a regular surface in Rn , with n ≥ 3, and

~ : U → Rn be a parametrization of a coordinate patch of S, let X where U is an open subset of R2 . Let Q be a compact subset of U ~ and call R = X(Q) the compact subset of S. The area of R is given by ZZ p det(g) du dv. (9.4) A(R) = Q

Proof: The proof of Equation (6.4) relied on the fact that S is a ~u regular surface in R3 and made reference to the cross product of X ~ and Xv , namely, using the formula ZZ ~u × X ~ v k du dv kX Q

from multivariable calculus. The Riemann sum behind this inte~ over gral involves approximating the surface area traced out by X [ui , ui+1 ] × [vi , vi+1 ] by a parallelogram spanned on two sides by ~ u (u∗ , v ∗ ) and X ~ v (u∗ , v ∗ ), where (u∗ , v ∗ ) is a selection point in the X subrectangle [ui , ui+1 ] × [vi , vi+1 ]. In order to find a formula for the surface area of a regular surface in Rn , one cannot refer to any formula using the cross product since the cross product is only defined in R3 . However, the approach used by the Riemann sum is still the right one. Therefore, one must find a convenient formula for the area of the parallelogram spanned by ~ u and X ~ v in Rn . X Regardless of the dimension of the ambient Euclidean space, we ~ v k sin θ, ~ u k kX can calculate the area of these parallelograms as kX where θ is the angle between the two vectors. One cannot get sin θ ~u · X ~ v = kX ~ u k kX ~ v k cos θ. directly but one can obtain cos θ from X Thus, ~ v )2 = (X ~u · X ~ u )(X ~v · X ~ v ) cos2 θ ~u · X (X ~u · X ~ v )2 = (X ~u · X ~ u )(X ~v · X ~ v )(1 − sin2 θ) =⇒ (X ~u · X ~ u )(X ~v · X ~ v ) − (X ~u · X ~ v )2 = (X ~u · X ~ u )(X ~v · X ~ v ) sin2 θ =⇒ (X q ~u · X ~ u )(X ~v · X ~ v ) − (X ~u · X ~ v )2 = kX ~ u k kX ~ v k sin θ. ⇐⇒ (X

367

368

9. Curves and Surfaces in n-dimensional Euclidean Space

p ~ u k kX ~ v k sin θ. Thus, at each point (u, v) ∈ U , we have det(g) = kX Taking the limit of the Riemann sum as the norm of a mesh that goes to 0 gives Equation (9.4).  Another proof for the above result relies on the geometric fact that the area of a parallelogram spanned by ~a, ~b ∈ Rn is v u q u ~a · ~a ~a · ~b t det(C > C) = ~ , b · ~a ~b · ~b   where C is the n × 2 matrix C = ~a ~b with the vectors ~a and ~b (expressed in standard coordinates) as columns. In Chapter 7, we introduced the Christoffel symbols and used them to prove the Theorema Egregium, which affirms that the Gaussian curvature is an intrinsic property. In particular, Christoffel symbols and the Gaussian curvature can be computed for regular surfaces of class C 2 in Rn in precisely the same way that we computed them in Chapter 7. Example 9.2.3 (Flat Torus). The topological definition of a torus is any

set that is homeomorphic to S1 × S1 , where S1 is a circle. In other parts of this book, we discuss the torus in R3 given by the parametrization ~ X(u, v) = ((b + a cos v) cos u, (b + a cos v) sin u, sin v)

for (u, v) ∈ [0, 2π] × [0, 2π],

and where 0 < a < b. This torus has a non-identity metric tensor and in Example 6.4.5, we saw that this torus has elliptic points whenever 0 < v < π, hyperbolic points when π < v < 2π, and parabolic points when v is a multiple of π. In particular, these are regions where the Gaussian curvature is respectively positive, negative, and zero. Consider the following parametrization of the torus in R4 : ~ X(u, v) = (cos u, sin u, cos v, sin v).

(9.5)

~ over [0, 2π] × [0, 2π] The set traced out in R4 by the function X satisfies the topological definition of a torus. It is an easy exercise (Problem 9.2.3) to prove that the metric tensor is constant with   1 0 gij (u, v) = . 0 1

9.2. Surfaces in n-dimensional Euclidean Space

369

This is the same metric tensor as a plane equipped with an orthonormal basis. Furthermore, it is obvious that the Christoffel symbols of the first kind will be identically 0, since they involve partial derivatives of the gij , and hence the Christoffel symbols of the second kind and the Gaussian curvature will be 0. This surface in Rn is called the flat torus. Speaking intuitively, we can say that there is “enough room” in R4 to embed a torus by bending but without stretching. A common way to imagine creating a torus in R3 , is to take a piece of paper and first roll it into a cylinder as shown below.

If the cylinder can be stretched, we can imagine bending it around so that the two circular ends meet up. This bending is required when we are in R3 . In four dimensions, suppose that Π1 and Π2 are planes that meet orthogonally at one point P . Suppose also that there is a line segment L1 in Π1 and another line segment L2 in Π2 that are perpendicular to each other that also meet at P . Let Π be a plane parallel and unequal to the plane spanned by L1 and L2 . We can first roll Π around L1 along a circle that lies in Π2 and the resulting cylinder can be rolled without bending around L2 along a circle that lies in Π2 . The resulting torus never bends the plane Π. We remind the reader that the formulas for the Gaussian curvature arise in Equations (7.16), (7.17), and (7.18). In particular, K=

 1 R1212 1 2 = R121 g12 + R121 g22 , det(gij ) det(gij )

(9.6)

l are defined in Equation (7.16). The Riemann where the symbols Rijk symbols satisfy certain symmetry relations which imply that R1212 =

370

9. Curves and Surfaces in n-dimensional Euclidean Space

R2121 . Hence, we also have K=

 R2121 1 1 2 = R121 g11 + R121 g21 . det(gij ) det(gij )

9.2.3 Orientability The property of orientability becomes more challenging to define for surfaces in Rn precisely because in Rn with n > 3, there does not exist a one-dimensional normal line to a regular surface at a ~ (u, v). point. Consequently, we cannot talk about a normal vector N From an intuitive perspective, we would like to say that a surface is orientable if it is possible to cover the surface continuously with a sense of clockwise rotation. Without a normal vector, we cannot refer to a positive rotation around a normal vector. Instead, we refer to the sign of a determinant. Recall that the definition of a regular surface implied that S is covered by a collection of coordinate patches. Recall that for any two ¯ with coordinates (x1 , x2 ) and (¯ coordinate patches U and U x1 , x ¯2 ), the change of coordinates corresponds to a regular reparametrization of the surface. Consequently, the Jacobian 1 ∂x ¯ ¯2 ) ∂x1 ∂(¯ x1 , x = ¯2 ∂(x1 , x2 ) ∂ x ∂x1

∂x ¯1 ∂x2 ∂x ¯2 ∂x2

is never zero. The sign of a 2 × 2 determinant is the sign of the angle between the first column and the second column vector. Suppose ~ :U ¯ → Rn , ~ : U → Rn and also by Y that surface is parametrized by X then over subsets of the domains that correspond to the intersec~ ) and Y ~ (U ¯ ), the sign of the determinant corresponds to tion of X(U ~ x1 to X ~ x2 has the same sign as the angle whether the angle from X ~x¯1 to Y ~x¯2 . It is this sign that is used to determine orientability. from Y Definition 9.2.4. A regular surface is called orientable if it is covered

by a collection of coordinate patches such that for any two over¯ with coordinates (x1 , x2 ) and lapping coordinate patches U and U

9.2. Surfaces in n-dimensional Euclidean Space

371

(¯ x1 , x ¯2 ), the Jacobian satisfies ¯2 ) ∂(¯ x1 , x > 0. ∂(x1 , x2 ) Whenever this condition holds between two coordinate patches, we say they have a compatible orientation. This condition is challenging to check for surfaces in general. However, for some surfaces that arise as the image of a single parametrized function, not necessarily a bijection, it may be easy to check if the surface is orientable. First, if a single parametrization ~ : U → S provides the required homeomorphism between U and X S, then the surface is automatically orientable. Example 9.2.5 (Flat Torus). As a second example, consider the flat

torus described in Example 9.2.3. The parametrization ~ X(u, v) = (cos u, sin u, cos v, sin v) with (u, v) ∈ R2 has the flat torus as its locus, though not bijectively. ~ 1 , v1 ) = X(u ~ 2 , v2 ), Now any point p on the flat torus has p = X(u where u2 − u1 and v2 − v1 are integer multiples of 2π. However, we have ~ u (u, v) = (− sin u, cos u, sin v, cos v) X ~ v (u, v) = (cos u, sin u, − sin v, cos v) X

and

and for any two pairs (u1 , v1 ) with (u2 , v2 ) with u2 −u1 , v2 −v1 ∈ 2πZ, we have ~ u (u2 , v2 ) ~ u (u1 , v1 ) = X X

and

~ v (u1 , v1 ) = X ~ v (u2 , v2 ). X

We can cover the flat torus with two coordinate patches U = (0, 2π)× ¯ = (π, 3π) × (π, 3π), each using the same function X. ~ (0, 2π) and U Over regions where these coordinate patches overlap, the Jacobian of the change of coordinates is constant. Hence the flat torus is orientable.

372

9. Curves and Surfaces in n-dimensional Euclidean Space

Example 9.2.6 (Klein Bottle). An example of a non-orientable surface in R4 is the Klein bottle. This is similar to the torus but with a twist. ~ : [0, 2π] × [0, 2π] → R4 defined by Consider the parametrization X

 u   u ~ , b sin v sin , X(u, v) = (b + a cos v) cos u, (b + a cos v) sin u, b sin v cos 2 2 with 0 < a < b. The first two components of this should remind the reader of the usual torus in R3 . (See Problem 6.6.4.) However, the last two components are in the x3 and x4 direction and are both perpendicular to the internal ring of the torus (b cos u, b sin u, 0, 0). In these components, we trace out a small ring as it goes around the larger internal ring, but making a half-turn twist in the process. It is ~ over the domain [0, 2π] × [0, 2π] not hard to see that the image of X defines a regular surface without boundary. We leave it as an exercise to check that over the domain [0, 2π) × [0, 2π), the parametrized surface is a bijection. Now consider the coordinate patch with this parametrization but over the domain U = (0, 2π) × (0, 2π). Consider the point ~ ~ X(0, π/2) = (b, 0, b, 0) = X(2π, 3π/2), ~ ). We can calculate that which is not in X(U    π   b ~ v 0, π = (−a, 0, 0, 0) ~ = 0, b, 0, and X Xu 0, 2 2 2 and also that     b 3π ~ = 0, b, 0, Xu 2π, 2 2

  ~ v 2π, π = (a, 0, 0, 0). and X 2

~ v (u, v)) vary continuously on the sur~ u (u, v), X The pair of vectors (X face. Assume there does exist a cover of the Klein bottle using coordinate patches that make it an orientable surface. Then the Jacobian of the coordinate change with respect to the coordinates defined by ~ and that of any other patch must have the same sign. Consider X ~ :U ¯ → S such that p ∈ Y ~ (U ¯ ) and that Y ~ (U ¯) a coordinate patch Y ~ ¯ ~ is connected. The image Y (U ) ∩ X(U ) consists of two connected components V1 and V2 with V1 ∩ V2 = ∅. At p, the coordinate change

9.2. Surfaces in n-dimensional Euclidean Space

matrix on Tp S over V1 and the coordinate change matrix on Tp S   ~ u 0, π = X ~ u 2π, 3π , over V2 cannot have the same sign because X 2 2   ~ v 2π, 3π . ~ v 0, π = −X whereas X 2 2

9.2.4 The Gauss-Bonnet Theorem In Chapter 8, we approached the Gauss-Bonnet Theorem starting from the perspective of curves on surfaces. After introducing the concept of geodesic curvature, we were able to prove the Gauss-Bonnet Theorem. One key part of the proof relied on using an orthogonal coordinate system. We originally cited the orthogonal coordinate system involving lines of curvature; however, lines of curvature are extrinsic properties and hence cannot be naturally generalized to surfaces in Rn . On the other hand, geodesics are intrinsic properties of surfaces and in Section 8.5 we proved the existence of geodesic coordinate systems that are orthogonal. More precisely, around every point p on the surface S there exists a neighborhood U of p that is parametrized by a geodesic coordinate system. Such a coordinate system, which is an intrinsic property, leads to the proof of the global Gauss-Bonnet Theorem. Consequently, Theorem 8.3.4 is an intrinsic theorem and holds for regular surfaces in Rn . The Gauss-Bonnet Theorem requires the surface to be orientable in order to put a compatible orientation on the boundary curve C. This orientation on C affects the sign of Z κg ds, C

which in turns affects the Gauss-Bonnet formula. If R is a region of the surface without boundary, then the orientation of the surface is irrelevant for the formula and we deduce the following corollary. Corollary 9.2.7. If R is a regular surface without boundary of class C 2

in Rn , then

ZZ K dS = χ(R). R

Problems 9.2.1. Consider the parametrized surface in R4 given by ~ X(u, v) = (u cos v sin v, u sin2 v, u cos v, u)

373

374

9. Curves and Surfaces in n-dimensional Euclidean Space

with (u, v) ∈ R × [0, 2π]. (a) Prove that this surface is regular everywhere except at (0, 0, 0, 0). (b) Find the tangent plane at (u0 , v0 ) = (1, π/2). (c) Find the normal plane to the surface at (u0 , v0 ) = (1, π/2). 9.2.2. Consider the parametrized surface in R4 given by ~ X(u, v) = (a cos u sin v, a sin u sin v, a cos v, bu) for (u, v) ∈ R × [0, π], where a and b are positive constants. We can view this as a generalization of a helix in the sense that, instead of parametrizing a circle that rises along a perpendicular axis, it parametrizes a sphere that rises along an axis perpendicular to the plane of the sphere. (a) Prove that this surface is regular. (b) Find the tangent plane at (u0 , v0 ) = (π/2, π/2). (c) Calculate the metric tensor components. 9.2.3. Prove the claim concerning the metric tensor of the flat torus in Example 9.2.3. 9.2.4. Consider the following generalization to the flat torus. Let a, b, c, d be positive real constants and consider the surface in R4 parametrized by ~ X(u, v) = (a cos u, b sin u, c cos v, d sin v) with (u, v) ∈ [0, 2π] × [0, 2π]. Show that in general the coefficients of the metric tensor are non-constant functions but that the Gaussian curvature is identically 0. ~ 9.2.5. A Veronese surface in R5 is a surface parametrized by X(u, v) = 2 2 2 a(u, v, u , uv, v ) for (u, v) ∈ R . Calculate the metric tensor, the Christoffel symbols, and the Gaussian curvature. 9.2.6. Calculate the metric tensor and the Gaussian curvature for the surface in R4 parametrized by ~ X(u, v) = (u3 , u2 v, uv 2 , v 3 )

for (u, v) ∈ R2 .

9.2.7. The graph of a change of coordinates in F : R2 → R2 is a parametrized surface in R4 . In particular, if F (u, v) = (f (u, v), g(u, v)) over a domain U , then the graph of f is parametrized by ~ X(u, v) = (u, v, f (u, v), g(u, v))

for (u, v) ∈ U.

9.2. Surfaces in n-dimensional Euclidean Space

(a) Calculate the components of the metric tensor of this function graph. (b) Using Equation (9.6), calculate the Gaussian curvature. (c) As an application, give the metric tensor and the Gaussian curvature for polar coordinate transformation, i.e., when u = r, v = θ, f (r, θ) = r cos θ, and g(r, θ) = r sin θ. 9.2.8. Let α ~ : I → R2 and β~ : J → R2 be regular curves in the plane with ~ = (β1 (t), β2 (t)). component functions α ~ (t) = (α1 (t), α2 (t)) and β(t) Calculate the metric tensor and the Gaussian curvature function of the surface parametrized by ~ X(u, v) = (α1 (u), α2 (u), β1 (v), β2 (v)) for (u, v) ∈ I × J. 9.2.9. Consider the flat torus with the parametrization given in Equation (9.5). Use Equation (8.7) to show that curves of the form (u, v) = (at + c, bt + d), for a, b, c, d constants and t ∈ R, are geodesics. ~ : [0, 2π) × [0, 2π) → R4 is 9.2.10. Justify the claim in Example 9.2.6 that X bijective with its image. 9.2.11. Calculate the Gaussian curvature of the Klein bottle using the parametrization given in Example 9.2.6. 9.2.12. There are many ways to define a two-dimensional tube around a curve ~γ : I → R4 . Suppose that we define the tube as ~ X(u, v) = ~γ (u) + (r cos v)P~1 (u) + (r sin v)P~2 (u), with (u, v) ∈ I × [0, 2π], and for some radius r chosen small enough so that the surface is regular. (a) Determine the metric tensor of this tube. (b) Calculate the Gaussian curvature function K(u, v). (c) Find all the points where K(u, v) = 0. [Compare to Problem 6.6.8.]

375

APPENDIX A

Tensor Notation

A.1 Tensor Notation In Chapter 6, we gave to the first fundamental form the alternate name of metric tensor and delayed the explanation of what a tensor is. Mathematicians and physicists often present tensors and the tensor product in very different ways, sometimes making it difficult for a reader to see that authors in different fields are talking about the same thing. In this section, we introduce tensor notation in what one might call the “physics style,” which emphasizes how components of objects change under a coordinate transformation. Readers of mathematics who are well aquainted with tensor algebras on vector spaces might find this approach unsatisfactory, but physicists should recognize it. (The reader who wishes to understand the full modern mathematical formulation of tensors and see how the physics style meshes with the mathematical style should consult Appendix C in .) The description of tensors we introduce below relies heavily on transformations between coordinate systems. Though we discuss coordinates and transformations between them generally, one can keep in mind as running examples Cartesian or polar coordinates for regions of the plane or Cartesian, cylindrical, and spherical coordinates for regions of R3 . Ultimately, our discussion will apply to changes of coordinates between overlapping coordinate patches on regular surfaces. (The reader should be aware that , which follows the present text, provides a rigorous introduction to these concepts, whereas our presentation here is a little more based on intuition.)

377

378

A. Tensor Notation

A.1.1 Curvilinear Coordinate Systems Let S be an open set in Rn . A continuous surjective function f : U → S, where U is an open set in Rn , defines a coordinate system on S by associating to every point P ∈ S an n-tuple x(P ) = (x1 (P ), x2 (P ), . . . , xn (P )), such that f (x(P )) = P . In this notation, the superscripts do not indicate powers of a variable x but the ith coordinate for that point in the given coordinate system. Though a possible source of confusion at the beginning, differential geometry literature uses superscripts instead of the usual subscripts in order to mesh properly with subsequent tensor notation. One also sometimes says that f : U → S parametrizes S. As with polar coordinates, where (r0 , θ0 ) and (r0 , θ0 + 2π) correspond to the same point in the plane, the n-tuple need not be uniquely associated to the point P . Let S be an open set in Rn , and consider two coordinate systems on S relative to which the coordinates of a point P are denoted by (x1 , x2 , . . . , xn ) and (¯ x1 , x ¯2 , . . . , x ¯n ). Suppose that the open set U ⊂ Rn parametrizes S using the ntuple of coordinates (x1 , x2 , . . . , xn ) and that the open set V ⊂ Rn parametrizes S using the coordinates (¯ x1 , x ¯2 , . . . , x ¯n ). We assume that there exists a bijective change-of-coordinates function F : U → V so that we can write (¯ x1 , x ¯2 , . . . , x ¯n ) = F (x1 , x2 , . . . , xn ).

(A.1)

Again using the superscript notation, we might write explicitly   ¯1 = F 1 (x1 , x2 , . . . , xn ),  x .. .   x n ¯ = F n (x1 , x2 , . . . , xn ), where F i are functions from U to R. We will assume from now on that the change of variables function F is always of class C 2 , i.e., that all the second partial derivatives are continuous. Unless it becomes necessary for clarity, one often abbreviates the notation and writes x ¯i = x ¯i (xj ), ¯2 , . . . , x ¯n ) funcby which one understands that the coordinates (¯ x1 , x

A.1. Tensor Notation

379

tionally depend on the coordinates (x1 , x2 , . . . , xn ). Thus, we write ∂x ¯i ∂F i for , ∂xj ∂xj and the matrix of the differential dFP (see Equation (5.2)) is given by  i ∂x ¯ [dFP ]ij = . ∂xj Just as the functions x ¯i = x ¯i (xj ) represent the change of variables j j k F , we write x = x (¯ x ) to indicate the component functions of the −1 inverse F : V → U . In this notation, Proposition 5.3.3 states that, as matrices,  j   i −1 ∂x ¯ ∂x = , (A.2) ∂x ¯i ∂xj where we assume that the functions in the first matrix are evaluated ¯n )-coordinates, while the functions in the secat p in the (¯ x1 , . . . , x ond matrix are evaluated at p in the (x1 , . . . , xn )-coordinates. One can express the same relationship in an alternate way by writing xi = xi (¯ xj (xk )) and applying the chain rule when differentiating with respect to xk as follows: n

∂xi ¯1 ¯2 ¯n X ∂xi ∂ x ¯j ∂xi ∂ x ∂xi ∂ x ∂xi ∂ x = + + · · · + = . ∂x ¯1 ∂xk ∂x ¯2 ∂xk ∂x ¯n ∂xk ∂x ¯j ∂xk ∂xk j=1

However, by definition of a coordinate system in Rn , there must be no function dependence of one variable on another, so ∂xi = δki , ∂xk where δki is the Kronecker delta symbol defined by ( 1, if i = j, δji = 0, if i 6= j.

(A.3)

Therefore, since δji are essentially the entries of the identity matrix, we conclude that n X ∂xi ∂ x ¯j = δki (A.4) ∂x ¯j ∂xk j=1

and hence recover Equation (A.2).

380

A. Tensor Notation

z

ϕ

P

O θ

y

x Figure A.1. Spherical coordinates.

The space Rn is a vector space, so to each point P , we can asso−−→ ciate the position vector ~r = OP . Let (x1 , x2 , . . . , xn ) be a coordinate system of an open set containing P . The natural basis of Rn at a point P associated to this coordinate system is the set of vectors n ∂~r ∂~r ∂~r o , , . . . , , ∂x1 ∂x2 ∂xn where all the derivatives are evaluated at P . Note that one often expresses the vector ~r in terms of the Cartesian coordinate system, and this expression is precisely the transformation functions between the given coordinate system and the Cartesian coordinate system. Example A.1.1 (Spherical Coordinates). Spherical coordinates for points in R3 consist of a triple (ρ, θ, ϕ), where ρ is the distance of P to the origin, θ is the angle between the xz-plane and the vertical plane containing P , and ϕ is the angle between the positive z-axis and the ray [OP ) (see Figure A.1). In Cartesian coordinates, the position vector ~r for a point with spherical coordinates (ρ, θ, ϕ) is

~r = (ρ cos θ sin ϕ, ρ sin θ sin ϕ, ρ cos ϕ).

(A.5)

A.1. Tensor Notation

This corresponds to the coordinate transformation from spherical to Cartesian coordinates. Then ∂~r = (cos θ sin ϕ, sin θ sin ϕ, cos ϕ), ∂ρ ∂~r = (−ρ sin θ sin ϕ, ρ cos θ sin ϕ, 0), ∂θ ∂~r = (ρ cos θ cos ϕ, ρ sin θ cos ϕ, −ρ sin ϕ). ∂ϕ It is interesting to note that these three vectors are orthogonal for all triples (ρ, θ, ϕ). When this is the case, we say that the coordinate system is orthogonal . In this case, one obtains a natural orthonormal basis at P associated to this coordinate system by simply dividing by the length of each vector (or its negative). For spherical coordinates, the associated orthonormal basis at any point consists of the three vectors ~eρ = (cos θ sin ϕ, sin θ sin ϕ, cos ϕ), ~eθ = (− sin θ, cos θ, 0), ~eϕ = (cos θ cos ϕ, sin θ cos ϕ, − sin ϕ). The factors 1, ρ sin ϕ, and ρ between ∂~r/∂xi and its normalized vector are sometimes called the scaling factors for each coordinate. Again, it is important to note that unlike the usual bases in linear algebra, the basis {~eρ , ~eθ , ~eϕ } depends on the coordinates (ρ, θ ϕ) of a point in Rn . We are now in a position to discuss how components of various quantities defined locally, namely in a neighborhood U of a point p ∈ R3 , change under a coordinate transformation on U . As mentioned before, our definitions do not possess the usual mathematical flavor, and the supporting discussion might feel like a game of symbols. However, it is important to understand the transformational properties of tensor components even before becoming familiar with the machinery of linear algebra of tensors. We begin with the simplest situation.

381

382

A. Tensor Notation

Definition A.1.2. Let p ∈ Rn , and let U be a neighborhood of p. Sup-

pose that (x1 , x2 , . . . , xn ) and (¯ x1 , x ¯2 , . . . , x ¯n ) are two systems of co1 n ordinates on U . A function f (x , . . . , x ) given in the (x1 , x2 , . . . , xn )coordinates is said to be a scalar if its expression f¯(¯ x1 , . . . , x ¯n ) in 1 n the (¯ x ,...,x ¯ )-coordinates has the same numerical value. In other words, if ¯n ) = f (x1 , . . . , xn ). f¯(¯ x1 , . . . , x

In Definition A.1.2, it is understood that the coordinates (x1 , x2 , . . . , xn ) and (¯ x1 , x ¯2 , . . . , x ¯n ) refer to the same point p in Rn . This definition might appear at first glance not to hold much content in that every function defined in reference to some coordinate system should possess this property, but that is not true. Suppose that f is a scalar. The quantity that gives the derivative of f in the first coordinate is not a scalar, for though f¯(¯ x1 , . . . , x ¯n ) = 1 n f (x , . . . , x ), we have ∂ f¯ ∂f ∂x1 ∂f ∂x2 ∂f ∂xn = + + · · · + . ∂x ¯1 ∂x1 ∂ x ¯1 ∂x2 ∂ x ¯1 ∂xn ∂ x ¯1 As a second example of how quantities change under coordi~ of a differentiable nate transformations, we consider the gradient ∇f scalar function f . Recall that   ∂f ∂f ∂f ~ , ,..., n ∇f = ∂x1 ∂x2 ∂x in usual Cartesian coordinates. The gradient is a vector field, or we ~ P , which is a may simply consider the gradient of f at P , namely ∇f vector. We highlight the transformational properties of the gradient. The chain rule gives n

X ∂xi ∂f ∂ f¯ = . ∂x ¯j ∂x ¯j ∂xi

(A.6)

i=1

However, it turns out that this is not the only way components of what we usually call a “vector” can change under a coordinate transformation.

A.1. Tensor Notation

383

Again, consider two coordinate systems (x1 , x2 , . . . , xn ) and (¯ x1 , n n ~ on an open set of R . Let A be a vector in R , which we ~ in the respecconsider based at P . The components of the vector A 1 n 1 ¯ tive coordinate systems is (A , . . . , A ) and (A , . . . , A¯n ), where

x ¯2 , . . . , x ¯n )

~= A

n X i=1

n

Ai

X ∂~r ∂~r = A¯j j . i ∂x ∂x ¯ j=1

P ∂~r ∂ x¯j ∂~ r ~ in the two , we find that the components of A Since ∂x i = j ∂x ¯j ∂xi systems of coordinates are related by A¯j =

n X ∂x ¯j i=1

∂xi

Ai .

(A.7)

Example A.1.3 (Velocity in Spherical Coordinates). Consider a space

curve that is parametrized by ~r(t) for t ∈ I. The chain rule allows us to write the velocity vector in spherical or Cartesian coordinates as ∂~r ∂~r ∂~r + θ0 (t) + ϕ0 (t) ∂ρ ∂θ ∂ϕ 0 ~ 0 ~ 0 ~ = x (t)i + y (t)j + z (t)k.

~r0 (t) = ρ0 (t)

However, differentiating Equation (A.5), assuming ρ, θ, and ϕ are functions of t, allows us to identify the Cartesian coordinates of the velocity vector as:  0    0 x cos θ sin ϕ −ρ sin θ sin ϕ ρ cos θ cos ϕ ρ y 0  =  sin θ sin ϕ −ρ cos θ sin ϕ ρ sin θ cos ϕ   θ0  . cos ϕ 0 −ρ sin ϕ z0 ϕ0 If we label the spherical coordinates as (¯ x1 , x ¯2 , x ¯3 ) and the Cartesian coordinates as (x1 , x2 , x3 ), then the transition matrix between components of the velocity vector  spherical coordinates to Carte from ∂xi sian coordinates is precisely ∂ x¯j . Thus, the velocity vector of any curve ~γ (t) through a point P = ~γ (t0 ) does not change according to Equation (A.6) but according to Equation (A.7).

384

A. Tensor Notation

Though we have presented the notion of curvilinear coordinates in general, one should keep in mind the linear coordinate changes  1  1 x ¯ x 2 x  x2  ¯      ..  = M  ..   .   .  x ¯n

xn

where M is an n × n matrix. If either of these coordinate systems is given as coordinates in a basis of Rn , then the other system simply corresponds to a change of basis in Rn . It is easy to show that the transition matrix is then  i ∂x ¯ = M, ∂xj  i ∂x ¯ is constant the usual basis transition matrix. Furthermore, ∂x j over all Rn .

A.1.2 Tensors: Definitions and Notation The relations established in Equations (A.6) and (A.7) show that there are two different kinds of vectors, each following different transformational properties under a coordinate change. This distinction is not emphasized in most linear algebra courses but essentially corresponds to the difference between a column vector and a row vector, which in turn corresponds to vectors in Rn and its dual (Rn )∗ . (The ∗ notation denotes the dual of a vector space. See Appendix C.3 in  for background.) We summarize this dichotomy in the following two definitions. Definition A.1.4. Let (x1 , . . . , xn ) and (¯ x1 , . . . , x ¯n ) be two coordinate

systems in a neighborhood of a point p ∈ Rn . An n-tuple of real numbers (A1 , A2 , . . . , An ) is said to constitute the components of a contravariant vector at a point p if these components transform according to the relation A¯j =

n X ∂x ¯j i=1

∂xi

Ai ,

where we assume the partial derivatives are evaluated at p.

A.1. Tensor Notation

385

Definition A.1.5. Under the same conditions as above, an n-tuple (B1 ,

B2 , . . . , Bn ) is said to constitute the components of a covariant vector at a point p if these components transform according to the relation n X ∂xi Bj = Bi , ∂x ¯j i=1

where we assume the partial derivatives are evaluated at p. A few comments are in order at this point. Though the above two definitions are unsatisfactory from the modern perspective of set theory, these are precisely what one is likely to find in a classical mathematics text or a physics text presenting differential geometry. Nonetheless, we will content ourselves with these definitions and with the more general Definition A.1.6. We defer until Chapter 4 in  what is considered the proper modern definition of a tensor on a manifold. Next, we point out that the quantities (A1 , A2 , . . . , An ) in Definition A.1.4 or (B1 , B2 , . . . , Bn ) in Definition A.1.5 can either be constant or be functions of the coordinates in a neighborhood of p. If the quantities are constant, one says they form the components of an affine vector. If the quantities are functions in the coordinates, then the components define a different vector for every point in an open set, and thus, one views these quantities as the components of a vector field over a neighborhood of p. Finally, in terms of notation, we distinguish between the two types of vectors by using subscripts for covariant vectors and superscripts for contravariant vectors. This convention of notation is consistent throughout the literature and forms a central part of tensor calculus. This convention also explains the use of superscripts for the coordinates since (xi ) represents the components of a contravariant vector, namely, the position vector of a point. As a further example to motivate the definition of a tensor, recall the transformational properties of the components of the first fundamental form given in Equation (6.7): g¯ij =

2 X 2 X ∂xk ∂xl gkl , ∂x ¯i ∂ x ¯j k=1 l=1

386

A. Tensor Notation

where gkl (resp. g¯ij ) represents the coefficients of the first fundamental form in the (x1 , x2 )- (resp. (¯ x1 , x ¯2 )-) coordinates. This formula mimics but generalizes the transformational properties in Definition A.1.4 and Definition A.1.5. Many objects of interest that arise in differential geometry possess similar properties and lead to the following definition of a tensor. Definition A.1.6. Let (x1 , . . . , xn ) and (¯ x1 , . . . , x ¯n ) be two coordinate

systems in a neighborhood of a point p ∈ Rn . A set of nr+s quantities ···ir Tji11ji22···j is said to constitute the components of a tensor of type s (r, s) if under a coordinate transformation these quantities transform according to

k k ···kr T l11l22···ls

=

n X n X i1 =1 i2 =1

···

n X n X n X ir =1 j1 =1 j2 =1

n X ∂x ¯kr ∂xj1 ∂xj2 ∂xjs i1 i2 ···ir ∂x ¯ k1 ∂ x ¯ k2 ··· · · · · · · T , ∂xi1 ∂xi2 ∂xir ∂ x ¯l1 ∂ x ¯ l2 ∂x ¯ls j1 j2 ···js js =1

(A.8) where we assume the partial derivatives are evaluated at p. The rank of the tensor is the integer r + s. From the above definition, one could rightly surmise that basic calculations with tensors involve numerous repeated summations. In order to alleviate this notational burden, mathematicians and physicists who use the tensor notation as presented above utilize the Einstein summation convention. In this convention of notation, one assumes that one takes a sum from 1 to n (the dimension of Rn or the number of coordinates) over any index that appears both in a superscript and a subscript of a product. Furthermore, for this ∂x ¯i convention, in a partial derivative ∂x j , the index i is considered a superscript, and the index j is considered a subscript. For example, if Aij form the components of a (0, 2)-tensor and B k constitutes the components of a contravariant vector, with the Einstein summation convention, the expression Aij B j means n X

Aij B j .

j=1

As another example, with the Einstein summation convention, the transformational property in Equation (A.8) of a tensor is written

A.1. Tensor Notation

387

as k k ···kr

T l11l22···ls =

¯ k2 ∂x ¯k1 ∂ x ∂x ¯ks ∂xj1 ∂xj2 ∂xjr i1 i2 ···ir · · · · · · T , ∂xi1 ∂xi2 ∂xis ∂ x ¯l1 ∂ x ¯ l2 ∂x ¯lr j1 j2 ···js

where the summations from 1 to n over the indices i1 , i2 , . . . ir , j1 , ij j2 , . . . js is understood. As a third example, if Ckl is a tensor of type ij (2, 2), then Ckj means n X ij Ckj . j=1

On the other hand, with this convention, we do not sum over the index i in the expression Ai + B i or even in Ai + Bi . In fact, as we shall see, though the former expression has an interpretation, the latter does not. In the rest of this book, we will use the Einstein summation convention when working with components of tensors.

A.1.3 Operations on Tensors It is possible to construct new tensors from old ones. (Again, the reader is encouraged to consult Appendix C.4 in  to see the underlying algebraic meaning of the following operations.) i1 i2 ···ir r First of all, if Sji11 ji22···i ···js and Tj1 j2 ···js are both components of tensors of type (r, s), then the quantities i1 i2 ···ir ···ir r Wji11ji22···j = Sji11 ji22···i ···js + Tj1 j2 ···js s

form the components of another (r, s)-tensor. In other words, tensors of the same type can be added to obtain another tensor of the same type. The proof is very easy and follows immediately from the transformational properties and distributivity. k1 k2 ···kt r Secondly, if Sji11 ji22···i ···js and Tl1 l2 ···lu are components of tensors of type (r, s) and (t, u), respectively, then the quantities obtained by multiplying these components as in, ···ir k1 k2 ···kt k1 k2 ···kt r Wji11ji22···j = Sji11 ji22···i ···js Tl1 l2 ···lu , s l1 l2 ···lu

form the components of another tensor but of type (r + t, s + u). Again, the proof is very easy, but one must be careful with the

388

A. Tensor Notation

plethora of indices. One should note that this operation of tensor product works also for multiplying a tensor by a scalar since a scalar is a tensor of rank 0. Finally, another common operation on tensors is the contraction between two indices. We illustrate the contraction with an example. Let Aijk rs be the components of a (3, 2)-tensor, and define the quantities Brij

=

Aijk rk

=

n X

Aijk rk

by Einstein summation convention.

k=1

It is not hard to show (left as an exercise for the reader) that Brij constitute the components of a tensor of type (2, 1). More generally, starting with a tensor of type (r, s), if one sums over an index that appears both in the superscript and in the subscript, one obtains the components of a (r − 1, s − 1)-tensor. This is the contraction of a tensor over the stated indices.

A.1.4 Examples Example A.1.7. Following the terminology of Definition A.1.6, a co-

variant vector is often called a (0, 1)-tensor, and similarly, a contravariant vector is called a (1, 0)-tensor. Example A.1.8. In Problem 6.4.6, one showed that the coefficients

Lij of the second fundamental form constitute the components of a (0, 2)-tensor, just as the metric tensor does. Example A.1.9 (Inverse of a (0, 2)-tensor). As a more involved exam-

ple, consider the components Aij of a (0, 2)-tensor in Rn . Denote by Aij the quantities given as the coefficients of the inverse matrix of (Aij ). We prove that Aij form the components of a (2, 0)-tensor. Suppose that the coefficients Aij given in a coordinate system with variables (x1 , . . . , xn ) and A¯rs are given in the (¯ x1 , . . . , x ¯n )coordinate system. That they are the inverse to the matrices (Aij ) and (A¯rs ) means that Aij and A¯rs are the unique quantities such that Aij Ajk = δki , A¯rs A¯st = δtr ,

and (A.9)

A.1. Tensor Notation

389

where the reader must remember that we are using the Einstein summation convention. Combining Equation (A.9) and the transformational properties of Ajk , we get i

j

∂x ∂x A¯rs s t Aij = δtr . ∂x ¯ ∂x ¯ Multiplying both sides by

∂x ¯t ∂xα

and summing over t, we obtain

∂xi ∂xj ∂x ¯t ∂x ¯t A¯rs s t Aij α = δtr α ∂x ¯ ∂x ¯ ∂x ∂x i r ∂x ∂x ¯ ⇐⇒A¯rs s δαj Aij = ∂x ¯ ∂xα ∂xi ∂x ¯r ⇐⇒A¯rs s Aiα = . ∂x ¯ ∂xα Multiplying both sides by Aαβ and then summing over α, we get ∂xi ∂xβ ∂x ¯r αβ A¯rs s δiβ = A¯rs s = A . ∂x ¯ ∂x ¯ ∂xα Finally, multiplying the rightmost equality by β, one concludes that

∂x ¯s ∂xβ

and summing over

¯s αβ ∂x ¯r ∂ x A . A¯rs = ∂xα ∂xβ This shows that the quantities Aij satisfy Definition A.1.6 and form the components of a (2, 0)-tensor. By a similar manipulation, one can show that if B ij are the components of a (2, 0)-tensor, then the quantities Bij corresponding to the inverse of the matrix of B ij form the components of a (0, 2)tensor. Example A.1.10 (Gauss Map Coefficients). In differential geometry, one denotes by g ij the coefficients of the inverse of the matrix associated to the first fundamental form. Example A.1.9 shows that g ij is a (2, 0)-tensor. Furthermore, recall that the Weingarten equations in Equation (6.26) give the components (associated to the standard

390

A. Tensor Notation

basis on Tp (S) given by a particular parametrization) of the Gauss map as aij = −g ik Lkj . By tensor product and contraction, we see that the functions aij form the components of a (1, 1)-tensor. Example A.1.11 (Metric Tensors). It is important to understand some standard operations of vectors in the context of tensor notation. Consider two vectors in a vector space V of dimension n. Using tensor notation, one refers to these vectors as affine contravariant vectors with components Ai and B j , with i, j = 1, 2, . . . , n. We have seen that addition of the vectors or scalar multiplication are the usual operations from linear algebra. Another operation between vectors in V is the dot product, which was originally defined as n X

Ai B i ,

i=1

but this is not the correct way to understand the dot product in the context of tensor algebra. The very fact that one cannot use the Einstein summation convention is a hint that we must adjust our notation. The use of the usual dot product for its intended geometric purpose makes an assumption of the given basis of V , namely, that the basis is orthonormal. When using tensor algebra, one makes no such assumption. Instead, one associates a (0, 2)-tensor gij , called the metric tensor, to the basis of V with respect to which coordinates are defined. Then the first fundamental form (or scalar product) between Ai and B j is gij Ai B j

(Einstein summation).

One immediately notices that because of tensor multiplication and contraction, the result is a scalar quantity, and hence, will remain unchanged under a coordinate transformation. In this formulation, the assumption that a basis is orthonormal is equivalent to having ( 1, if i = j, gij = 0, if i 6= j.

A.1. Tensor Notation

391

A.1.5 Symmetries The usual operations of tensor addition and scalar multiplication were explained above. We should point out that, using distributivity and associativity, one notices that the set of affine tensors of type (r, s) in Rn form a vector space. The (r + s)-tuple of all the indices can take on nr+s values, so this vector space has dimension nr+s . However, it is not uncommon that there exist symmetries within the components of a tensor. For example, as we saw for the metric tensor, we always have gij = gji . In the context of matrices, we said that the matrix (gij ) is a symmetric matrix, but in the context of tensor notation, we say that the components gij are symmetric in the ···ir indices i and j. More generally, if Tji11ji22···j are the components of a s tensor of type (r, s), we say that the components are symmetric in a set S of indices if the components remain equal when we interchange any two indices from among the indices in S. For example, let Aijk rs be the components of a (3, 2)-tensor. To say that the components are symmetric in {i, j, k} affirms the equalities ikj jik jki kij kji Aijk rs = Ars = Ars = Ars = Ars = Ars

for all i, j, k ∈ {1, 2, . . . , n}. Note that, because of the additional conditions, the dimension of the space of all (3, 2)-tensors that are symmetric in their contravariant indices is smaller than n5 but not simply n5 /6 either. We can find the dimension of this vector space by determining the cardinality of  I = (i, j, k) ∈ {1, 2, . . . , n}3 | 1 ≤ i ≤ j ≤ k ≤ n .  We will see shortly that |I| = n+2 3 , and therefore, the dimension of the vector space of (3, 2)-tensors that are symmetric in their con 2 travariant indices is n+2 n . We provide the following proposition 3 for completeness. j ···j

Proposition A.1.12. Let Ak11 ···krs be the components of a tensor over Rn

that is symmetric in a set S of its indices. Assuming that all the indices are fixed except for the indices of S, the number of independent components of the tensor is equal to the cardinality of  I = (i1 , . . . , im ) ∈ {1, 2, . . . , n}m | 1 ≤ i1 ≤ i2 ≤ · · · ≤ im ≤ n .

392

A. Tensor Notation

This cardinality is   n−1+m (n − 1 + m)! = . m (n − 1)!m! Proof: Since the components are symmetric in the set S of indices, one gets a unique representative of equivalent components by imposing that the indices in question be listed in nondecreasing order. This remark proves the first part of the proposition. To prove the second part, consider the set of integers {1, 2, . . . , n + m} and pick m distinct integers {l1 , . . . , lm } that are greater than 1 from among this set. We know from the definition of combinations that there are n−1+m ways to do this. Assuming that l1 < l2 < · · · < lm , define m it = lt − t. It is easy to see that the resulting m-tuple (i1 , . . . , im ) is in the set I. Furthermore, since one can reverse the process by defining lt = it + t for 1 ≤ t ≤ m, there exists a bijection between I and the m-tuples (l1 , . . . , lm ) described above. This establishes that   n−1+m . |I| = m  Another common situation with relationships between the components of a tensor is when components are antisymmetric in a set of indices. We say that the components are antisymmetric in a set S of indices if the components are negated when we interchange any two indices from among the indices in S. This condition imposes a number of immediate consequences. Consider, for example, the components of a (0, 3)-tensor Aijk that are antisymmetric in all its indices. If k is any value but i = j, then Aijk = Aiik = Ajik = −Aijk , and so Aiik = 0. Given any triple (i, j, k) in which at least two of the indices are equal, the corresponding component is equal to 0. As another consequence of the antisymmetric condition, consider the component A231 . One obtains the triple (2, 3, 1) from (1, 2, 3) by first interchanging 1 and 2 to get (2, 1, 3) and then interchanging the last two to get (2, 3, 1). Therefore, we see that A123 = −A213 = A231 .

A.1. Tensor Notation

393

In modern algebra, a permutation (a bijection on a finite set) that interchanges two inputs and leaves the rest fixed is called a transposition. We say that we used two transpositions to go from (1, 2, 3) to (2, 3, 1). The above example illustrates that the value of the component of a tensor indexed by a particular m-tuple (i1 , . . . , im ) of distinct indices determines the value of any component involving a permutation (j1 , . . . , jm ) of (i1 , . . . , im ), as Aj1 ...jm = ±Ai1 ...im , where the sign ± is + (resp. −) if it takes an even (resp. odd) number of interchanges to get from (i1 , . . . , im ) to (j1 , . . . , jm ). A priori, if one could get from (i1 , . . . , im ) to (j1 , . . . , jm ) with both an odd and an even number of transpositions, then Ai1 ...im and all components indexed by a permutation of (i1 , . . . , im ) would be 0. However, a fundamental fact in modern algebra (see Theorem 5.5 in ) states that given a permutation σ on {1, 2, . . . , m}, if we have two ways to write σ as a composition of transpositions, σ = τ1 ◦ τ2 ◦ · · · ◦ τa = τ10 ◦ τ20 ◦ · · · ◦ τb0 , then a and b have the same parity. Definition A.1.13. We call a permutation even (resp. odd ) if this com-

mon parity is even (resp. odd) and the sign of σ is ( 1, if σ is even, sign(σ) = −1, if σ is odd. The above discussion leads to the following proposition about the components of an antisymmetric tensor. j ···j

Proposition A.1.14. Let Ak11 ···krs be the components of a tensor over Rn

that is antisymmetric in a set S of its indices. If any of the indices in S are equal, then r Ajk11···j ···ks = 0. If |S| = m, then fixing all but the indices in S, the number of independent components of the tensor is equal to   n n! = . m m!(n − m)!

394

A. Tensor Notation k1 ···kr r Finally, if the indices of Aij11···i ···js differ from Al1 ···ls only by a permutation σ on the indices in S, then ···kr r = sign(σ) Aij11···i Akl11···l ···js . s

A.1.6 Numerical Tensors As a motivating example of what are called numerical tensors, note that the quantities δji form the components of a (1, 1)-tensor. To see this, suppose that the quantities δji are expressed in a system of coordinates (x1 , . . . , xn ) and suppose that δ¯lk are its transformed coefficients in another system of coordinates (¯ x1 , . . . , x ¯n ). Obviously, i for all fixed i and j, the values of δj are constant, and therefore ( δ¯lk

=

1, 0,

if k = l, if k = 6 l.

But using the properties of the δji coefficients and the chain rule, ∂x ¯k ∂x ¯k ∂xi ∂x ¯k ∂xj i δj = = = δ¯lk . i i l l ∂x ∂ x ∂x ∂ x ¯ ¯ ∂x ¯l Therefore, δji is a (1, 1)-tensor in a tautological way. A numerical tensor is a tensor of rank greater than 0 whose components are constant in the variables (x1 , . . . , xn ) and hence also (¯ x1 , . . . , x ¯n ). The Kronecker delta is just one example of a numerical tensor and we have already seen that it plays an important role in many complicated calculations. The Kronecker delta is the simplest case of the most important numerical tensor, the generalized Kronecker delta. The generalized Kronecker delta of order r is a r tensor of type (r, r), with components denoted by δji11 ···i ···jr defined as the following determinant:

r δji11 ···i ···jr

i1 δj i1 δ 2 j = .1 .. δ ir j1

δji12 δji22 .. . δjir2

· · · δji1r · · · δji2r . . .. . .. · · · δ ir jr

(A.10)

A.1. Tensor Notation

395

r It is not obvious from Equation (A.10) that the quantities δji11 ···i ···jr form the components of a tensor. However, one can write the components of the generalized Kronecker delta of order 2 as

ij δkl = δki δlj − δli δkj , ij which presents δkl as the difference of two (2, 2)-tensors, which shows ij indeed constitute a tensor. More generally, that the coefficients δkl expanding out Equation (A.10) gives the generalized Kronecker delta of order r as a sum of r! components of tensors of type (r, r), proving r that δji11 ···i ···jr are the components of an (r, r)-tensor. r Properties of the determinant imply that δji11 ···i ···jr is antisymmetric in both the superscript indices and the subscript indices. That is to r say, δji11 ···i ···jr = 0 if any of the superscript indices are equal or if any of the subscript indices are equal. Hence, the value of a component is negated if any two superscript indices are interchanged and similarly for subscript indices. We also note that if r > n where we assume i1 ···ir n r δji11 ···i ···jr is a tensor in R , then δj1 ···jr = 0 for all choices of indices since at least two superscript (and at least two subscript) indices would be equal. We introduce one more symbol related to the generalized Kronecker delta, namely the permutation symbol. Define

i1 ···in εi1 ···in = δ1···n ,

. εj1 ···jn = δj1···n 1 ···jn

(A.11)

Note that the use of the maximal index n in Equation (A.11) as opposed to r is intentional. Because of the properties of the determinant, it is not hard to see that εi1 ···in = εi1 ···in is equal to 1 (resp. −1) if (i1 , . . . , in ) is an even (resp. odd) permutation of (1, 2, . . . , n) and is equal to 0 if (i1 , . . . , in ) is not a permutation of (1, 2, . . . , n). We are careful, despite the notation, not to call the permutation symbols the components of a tensor, for they are not. Instead, we have the following proposition.

396

A. Tensor Notation

Proposition A.1.15. Let (x1 , . . . , xn ) and (¯ x1 , . . . , x ¯n ) be two coordi-

nate systems. The permutation symbols transform according to ε¯j1 ···jn = J

∂x ¯j1 ∂x ¯jn i1 ···in · · · ε , ∂xi1 ∂xin

ε¯k1 ···kn = J −1

∂xh1 ∂xhn · · · εh ···h , ∂x ¯k1 ∂x ¯kn 1 n

 i ∂x ¯ is the Jacobian of the transformation of coorwhere J = det ∂x j dinates function. Proof: (Left as an exercise for the reader.)



Example A.1.16 (Cross Product). Consider two contravariant vectors Ai and B j in R3 . If we define Ck = εijk Ai B j , we easily find that

C1 = A2 B 3 − A3 B 2 ,

C2 = A3 B 1 − A1 B 3 ,

C3 = A1 B 2 − A2 B 1 .

The values Ck are precisely the terms of the cross product of the vectors Ai and B j . However, a quick check shows that the quantities Ck do not form the components of a covariant tensor. One explanation in relation to standard linear algebra for the fact that Ck does not give a contravariant vector is that if ~a and ~b are vectors in R3 given with coordinates in a certain basis and if M is a coordinate-change matrix, then (M~a) × (M~b) 6= M (~a × ~b). In many physics textbooks, when one assumes that we use the usual metric, (gij ) being the identity matrix, one is not always careful with the superscript and subscript indices. This is because one can obtain a contravariant vector B j from a covariant vector Ai simply by defining B j = g ij Ai , and the components (B 1 , B 2 , B 3 ) are numerically equal to (A1 , A2 , A3 ). Therefore, in this context, one can define the cross product as the vector with components C l = g kl εijk Ai B j .

(A.12)

However, one must remember that this is not a contravariant vector since it does not satisfy the transformational properties of a tensor.

A.1. Tensor Notation

397

The generalized Kronecker delta has a close connection to determinants, which we will elucidate here. Note that if the superscript r indices are exactly equal to the subscript indices, then δji11 ···i ···jr is the determinant of the identity matrix. Thus, the contraction over all ···jr indices δjj11···j counts the number of permutations of r indices taken r from the set {1, 2, . . . , n}. Thus, ···jr = δjj11···j r

n! . (n − r)!

(A.13)

Another property of the generalized Kronecker delta is that ···jn , εj1 ···jn εi1 ···in = δij11···i n

the proof of which is left as an exercise for the reader (Problem A.1.10). Now let aij be the components of a (1, 1)-tensor, which we can view as the matrix of a linear transformation from Rn to Rn . By definition of the determinant, det(aij ) = εj1 ···jn a1j1 · · · anjn . Then, by properties of the determinant related to rearranging rows or columns, we have εi1 ···in det(aij ) = εj1 ···jn aij11 · · · aijnn . Multiplying by εi1 ···in and summing over all the indices i1 , . . . , in , we have ···jn i1 a · · · aijnn . εi1 ···in εi1 ···in det(aij ) = δij11···i n j1 Since εi1 ···in εi1 ···in counts the number of permutations of {1, . . . , n}, we have ···jn i1 a · · · aijnn . (A.14) n! det(aij ) = δij11···i n j1

Problems A.1.1. Prove that (a) δji δkj δlk = δli , (b) δji δkj δik = n.

398

A. Tensor Notation

A.1.2. Let Bi be the components of a covariant vector. Prove that the quantities ∂Bj ∂Bk Cjk = − k ∂x ∂xj form the components of a (0, 2)-tensor. ···ir be the components of a tensor of type (r, s). Prove A.1.3. Let Tji11ji22···j s ···ir that the quantities Tijii22···j , obtained by contracting over the first s two indices, form the components of a tensor of type (r − 1, s − 1). Explain why one still obtains a tensor when one contracts over any superscript and subscript index.

A.1.4. Let Sijk be the components of a tensor, and suppose they are antisymmetric in {i, j}. Find a tensor with components Tijk that is antisymmetric in {j, k} satisfying −Tijk + Tjik = Sijk . A.1.5. If Ajk is antisymmetric in its indices and Bjk is symmetric in its indices, show that the scalar Ajk Bjk is 0. A.1.6. Consider Ai , B j , C k the components of three contravariant vectors. ~B ~ C) ~ = A·( ~ B ~× Prove that εijk Ai B j C k is the value triple product (A ~ C), which is the volume of the parallelopiped spanned by these three vectors. rs . Assume that we use a metric gij that A.1.7. Prove that εijk εrsk = δij ~ of two conis the identity matrix, and define the cross product C i j k ~ ~ travariant vectors A = (A ) and B = (B ) as C = g kl εijl Ai B j . Use what you just proved to show that

~ × (B ~ × C) ~ = (A ~ · C) ~ B ~ − (A ~ · B) ~ C. ~ A ~ B, ~ C, ~ and D ~ be vectors in R3 . Use the εijk symbols to prove A.1.8. Let A, that ~ × B) ~ × (C ~ × D) ~ = (A ~B ~ D) ~ C ~ − (A ~B ~ C) ~ D. ~ (A A.1.9. Prove Proposition A.1.15. n A.1.10. Prove that εi1 ···in εj1 ···jn = δji11 ···i ···jn .

A.1.11. Let Aij be the components of an antisymmetric tensor of type (0, 2), and define the quantities Brst =

∂Ast ∂Atr ∂Ars + + . ∂xr ∂xs ∂xt

A.1. Tensor Notation

399

(a) Prove that Brst are the components of a tensor of type (0, 3). (b) Prove that the components Brst are antisymmetric in all the indices. (c) Determine the number of independent components of antisymmetric tensors of type (0, 3) over Rn . (d) Would the quantities Brst still be the components of a tensor if Aij were symmetric? A.1.12. Let A be an n × n matrix with coefficients A = (Aji ), and consider the coordinate transformation x ¯j =

n X

Aji xi .

i=1

Recall that this transformation is called orthogonal if AAT = I, where AT is the transpose of A and I is the identity matrix. The orthogonality condition implies that det(A) = ±1. An orthogonal transformation is called special or proper if, in addition, det(A) = 1. ···ir is called a proper tensor of type (r, s) if A set of quantities Tji11···j s it satisfies the tensor transformation property from Equation (A.8) for all proper orthogonal transformations. (a) Prove that the orthogonality condition is equivalent to requiring that ηij = ηhk Ahi Akj , where

( ηij =

1, 0,

if i = j, . if i = 6 j.

(b) Prove that the orthogonality condition is also equivalent to saying that orthogonal transformations are the invertible linear transformations that preserve the quantity (x1 )2 + (x2 )2 + · · · + (xn )2 . (c) Prove that (1) the space of proper tensors of type (r, s) form a vector space over R, (2) the product of a proper tensor of type (r1 , s1 ) and a proper tensor of type (r2 , s2 ) is a proper tensor of type (r1 + r2 , s1 + s2 ), and (3) contraction over two indices of a proper tensor of type (r, s) produces a proper tensor of type (r − 1, s − 1). (d) Prove that the permutation symbols are proper tensors of type (n, 0) or (0, n), as appropriate.

400

A. Tensor Notation

(e) Use this to prove that the cross product of two contravariant vectors in R3 as defined in Equation (A.12) is a proper tensor of type (1, 0). [Hint: This explains that the cross product of two vectors transforms correctly only if we restrict ourselves to proper orthogonal transformations on R3 .] (f) Suppose that we are in R3 . Prove that trix  cos α − sin α A =  sin α cos α 0 0

the rotation with ma 0 0 1

is a proper orthogonal transformation. (g) Again, suppose that we are in R3 . Prove that the linear transformation with matrix given with respect to the standard basis   cos β sin β 0 B =  sin β − cos β 0 0 0 1 is an orthogonal transformation that is not proper. A.1.13. Consider the vector space Rn+1 with coordinates (x0 , x1 , . . . , xn ), where (x1 , . . . , xn ) are called the space coordinates and x0 is the time coordinate. The usual connection between x0 and time t is x0 = ct, where c is the speed of light. We equip this space with the metric ηµν where η00 = −1, ηii = 1 for 1 ≤ i ≤ n, and ηij = 0 if i 6= j. (The quantities ηµν do not give a metric in the sense we have presented in this text so far because it is not positive definite. Though we do not provide the details here, this unusual metric gives a mathematical justification for why it is impossible to travel faster than the speed of light c.) This vector space equipped with the metric ηµν is called the n-dimensional Minkowski spacetime, and ηµν is called the Minkowski metric. Let L be an (n+1)×(n+1) matrix with coefficients Lα β , and consider the linear transformation x ¯j =

n X

Lji xi .

i=0

A Lorentz transformation is an invertible linear transformation on Minkowski spacetime with matrix L such that ηαβ = ηµν Lµα Lνβ .

A.1. Tensor Notation

401

···ir Finally, a set of quantities Tji11···j , with indices ranging in {0, . . . , n}, s is called a Lorentz tensor of type (r, s) if it satisfies the tensor transformation property from Equation (A.8) for all Lorentz transformations.

(a) Show that a transformation of Minkowski spacetime is a Lorentz transformation if and only if it preserves the quantity −(x0 )2 + (x1 )2 + (x2 )2 + · · · + (xn )2 . (b) Suppose we are working in three-dimensional Minkowski spacetime. Prove that the rotation matrix   1 0 0 0 0 cos α − sin α 0  A= 0 sin α cos α 0 0 0 0 1 represents a Lorentz transformation. (c) Again, suppose we are working in three-dimensional Minkowski spacetime. Consider the matrix   γ −βγ 0 0 −βγ γ 0 0 , L=  0 0 1 0 0 0 0 1 where β is a p positive real number satisfying −1 < β < 1 and γ = 1/( 1 − β 2 ). Prove that L represents a Lorentz transformation. (d) Prove that (1) the space of Lorentz tensors of type (r, s) form a vector space over R, (2) the product of a Lorentz tensor of type (r1 , s1 ) and a Lorentz tensor of type (r2 , s2 ) is a Lorentz tensor of type (r1 + r2 , s1 + s2 ), and (3) contraction over two indices of a Lorentz tensor of type (r, s) produces a Lorentz tensor of type (r − 1, s − 1). A.1.14. Let aij be the components of a (1, 1)-tensor, or in other words, the matrix of a linear transformation from Rn to Rn given with respect to some basis. Recall that the characteristic equation for the matrix is (A.15) det(aij − λδji ) = 0. [Hint: The solutions to this equation are the eigenvalues of the matrix.] Prove that Equation (A.15) is equivalent to n X (−1)r a(r) λn−r = 0 λ + n

r=1

402

A. Tensor Notation

where a(r) =

1 i1 ···ir i1 δ a · · · aijrr . r! j1 ···jr j1

A.1.15. Moment of Inertia Tensor. Suppose that R3 is given a basis that is not necessarily orthonormal. Let gij be the metric tensor corresponding to this basis, which means that the scalar product between two (contravariant) vectors Ai and B j is given by ~ Bi ~ = gij Ai B j . hA, In the rest of the problem, call (x1 , x2 , x3 ) the coordinates of the position vector ~r. Let S be a solid in space with a density function ρ(~r), and suppose that it rotates about an axis ` through the origin. The angular velocity vector ω ~ is defined as the vector along the axis `, pointing in the direction that makes the rotation a right-hand corkscrew motion, and with magnitude ω that is equal to the radians per second swept out by the motion of rotation. Let (ω 1 , ω 2 , ω 3 ) be the components of ω ~ in the given basis. The moment of inertia of the solid S about the direction ω ~ is defined as the quantity ZZZ 2 I` = ρ(~r)r⊥ dV, S

where r⊥ is the distance from a point ~r with coordinate (x1 , x2 , x3 ) to the axis `. The moment of inertia tensor of a solid is often presented using cross products, but we define it here using a characterization that is equivalent to the usual definition but avoids cross products. We define the moment of inertia tensor as the unique (0, 2)-tensor Iij such that ωj (Iij ω i ) = I` ω, (A.16) ω p where ω = k~ ω k = h~ ω, ω ~ i. (a) Prove that 2 r⊥ = gij xi xj −

(gkl ω k xl )2 . grs ω r ω s

(b) Prove that, using the metric gij , the moment of inertia tensor is given by ZZZ Iij = ρ(x1 , x2 , x3 )(gij gkl − gik gjl )xk xl dV. S

A.1. Tensor Notation

403

(c) Prove that Iij is symmetric in its indices. (d) Prove that if the basis of R3 is orthonormal (which means that (gij ) is the identity matrix), one recovers the following usual formulas one finds in physics texts: ZZZ ZZZ I11 = ρ((x2 )2 + (x3 )2 ) dV, I12 = − ρx1 x2 dV, S

S

ZZZ

ZZZ

I22 = Z SZ Z I33 = S

ρ((x1 )2 + (x3 )2 ) dV, ρ((x1 )2 + (x2 )2 ) dV,

I13 = − Z SZ Z I23 = −

ρx1 x3 dV, ρx2 x3 dV.

S

(We took the relation in Equation (A.16) as the defining property of the moment of inertia tensor because of the theorem that I` ω is the component of the angular moment vector along the axis of rotation j that is given by (Iij ω i ) ωω . See [13, pp. 221–222], and in particular, Equation (9.7) for an explanation. The interesting point about this approach is that it avoids the use of an orthonormal basis and provides a formula for the moment of inertia tensor when one has an affine metric tensor that is not the identity. Furthermore, since it avoids the cross product, the above definitions for the moment of inertia tensor of a solid about an axis are generalizable to solids in Rn .) A.1.16. This problem considers formulas for curvatures of curves or surfaces defined implicitly by one equation. (a) In Problem 1.3.17, the reader was asked to prove a formula for the geodesic curvature κg at a point p on a curve given implicitly by the equation F (x, y) = 0. Show that the formula found there can be written as 1 εi1 i2 εj1 j2 Fi1 Fj1 Fi2 j2 , (A.17) κg = (Fx2 + Fy2 )3/2 ∂F where, in the Einstein summation convention, F1 means ∂x 1 = ∂F ∂F ∂F and F means = , and where all the functions are 2 2 ∂x ∂x ∂y evaluated at the point p. (b) Prove that Equation (A.17) can be written as Fxx Fxy Fx 1 Fyx Fyy Fy . κg = − 2 2 3/2 (Fx + Fy ) Fx Fy 0

404

A. Tensor Notation

(c) Problem 6.6.17 asked the reader to do the same exercise but to find a formula for the Gaussian curvature at a point on a surface given implicitly by the equation F (x, y, z) = 0. Show that the Gaussian curvature K at a point p on a curve given implicity by the equation F (x, y, z) = 0 can be written as K=

1 εi1 i2 i3 εj1 j2 j3 Fi1 Fj1 Fi2 j2 Fi3 j3 (Fx2 + Fy2 + Fz2 )2

(A.18)

with the same conventions as used for a curve defined implicitly. (d) Prove that Equation (A.18) can be written as Fxx Fxy Fxz Fyx Fyy Fyz 1 K=− 2 2 2 2 (Fx + Fy + Fz ) Fzx Fzy Fzz Fx Fy Fz

Fx Fy . Fz 0

(A.19)

Bibliography

 M. A. Armstrong. Basic Topology. Undergraduate Texts in Mathematics. New York: Springer-Verlag, 1983.  Vladimir I. Arnold. Ordinary Differential Equations. Cambridge, MA: MIT Press, 1973.  Josu Arroyo, Oscar J. Garay, and Jose J. Menc´ıa. “When Is a Periodic Function the Curvature of a Closed Plane Curve?” American Mathematical Monthly 115:5 (2008), 405–414. ´  Pierre Ossian Bonnet. “La th´eorie g´en´erale des surfaces.” J. Ecole Polytechnique 19 (1848), 1–146.  Pierre Ossian Bonnet. “Sur quelques propri´et´es des lignes g´eod´esiques.” C. R. Acad. Sci. Paris 40 (1855), 1311–1313.  Roberto Bonola. Non-Euclidean Geometry – A Critical and Historical Study of its Developments. New York: Dover Publications, 1955.  Andrew Browder. Mathematical Analysis: An Introduction. Undergraduate Texts in Mathematics. New York: Springer-Verlag, 1996.  Judith N. Cederberg. A Course in Modern Geometries, Second edition. Undergraduate Texts in Mathematics. New York: SpringerVerlag, 2001.  Richard Courant, Herbert Robbins, and Ian Stewart. What Is Mathematics?, Second edition. Oxford, U.K.: Oxford University Press, 1996.  H. S. M. Coxeter. Non-Euclidean Geometry, sixth edition. Washington, DC: Mathematical Association of America, 1998.  Manfredo do Carmo. Differential Geometry of Curves and Surfaces. Upper Saddle River, NJ: Prentice Hall Inc., 1976.  William Dunham. The Mathematical Universe. New York: John Wiley and Sons, Inc., 1994.

405

406

Bibliography

 Grant R. Fowles. Analytical Mechanics, Fourth edition. Philadelphia, PA: Saunders College Publishing, 1986.  Joseph A. Gallian. Contemporary Abstract Algebra, sixth edition. New York: Houghton Mifflin, 2006.  Carl Friedrich Gauss. “Disquisitiones generales circa superficies curvas.” Comm. Soc. G¨ ottingen 6 (1823), 99–146.  Michael C. Gemignani. Elementary Topology, Second edition. New York: Dover Publications, 1972.  Herman Gluck. “Higher Curvatures of Curves in Euclidean Space.” The American Mathematical Monthly 73:7 (1966), 699–704.  Reuben Hersh and Phillip J. Davis. The Mathematical Experience. Boston, MA: Birkh¨auser, 1981. ¨  Heinz Hopf. “Uber die Drehung der Tangenten und Sehnen ebener Kurven.” Compositio Mathematica 2 (1935), 50–62.  R. A. Horn. “On Fenchel’s Theorem.” Monthly 78:4 (1971), 381–382.

American Mathematical

 Serge Lang. Algebra, Third edition. Reading, MA: Addison-Wesley, 1993.  H. Blaine Lawson. Lectures on Minimal Submanifolds. Rio de Janeiro, Brazil: Monografias de Matem´atica, IMPA, 1973.  W. B. Raymond Lickorish. An Introduction to Knots, Graduate Texts in Mathematics, 175. New York: Springer-Verlag, 1997.  Stephen T. Lovett. Differential Geometry of Manifolds. Wellesley, MA: A K Peters, Ltd., 2010.  Ib H. Masden and J¨ urgen Tornehave. From Calculus to Cohomology. Cambridge, U.K.: Cambridge University Press, 1997.  James Munkres. Topology. Englewood Cliffs, NJ: Prentice Hall, 1975.  Johannes C. C. Nitsche. Lectures on Minimal Surfaces, Volume 1. Cambridge, U.K.: Cambridge University Press, 1989.  Robert Osserman. A Survey of Minimal Surfaces. New York: Dover Publications, 1969.  Patrick J. Ryan. Euclidean and Non-Euclidean Geometry. Cambridge, U.K.: Cambridge University Press, 1986.  James Stewart. Calculus, sixth edition. Brooks / Cole, 2003.

Belmont, CA: Thomson

Bibliography

 J. J. Stoker. Differential Geometry. New York: Wiley-Interscience, 1969.  Dirk J. Struik. Lectures on Classical Differential Geometry, Second edition. New York: Dover Publications, 1961.  E. R. van Kampen. “The theorems of Gauss-Bonnet and Stokes.” Am. Jour. Math. 60:1 (1938), 129–138.

407