238 83 3MB
English Pages 414 [433] Year 2017
GRADUATE STUDIES I N M AT H E M AT I C S
178
From Frenet to Cartan: The Method of Moving Frames Jeanne N. Clelland
American Mathematical Society
10.1090/gsm/178
From Frenet to Cartan: The Method of Moving Frames
GRADUATE STUDIES I N M AT H E M AT I C S
178
From Frenet to Cartan: The Method of Moving Frames
Jeanne N. Clelland
American Mathematical Society Providence, Rhode Island
EDITORIAL COMMITTEE Dan Abramovich Daniel S. Freed (Chair) Gigliola Staļ¬lani Jeļ¬ A. Viaclovsky 2010 Mathematics Subject Classiļ¬cation. Primary 22F30, 53A04, 53A05, 53A15, 53A20, 53A55, 53B25, 53B30, 58A10, 58A15.
For additional information and updates on this book, visit www.ams.org/bookpages/gsm-178
Library of Congress Cataloging-in-Publication Data Names: Clelland, Jeanne N., 1970Title: From Frenet to Cartan : the method of moving frames / Jeanne N. Clelland. Description: Providence, Rhode Island : American Mathematical Society, [2017] | Series: Graduate studies in mathematics ; volume 178 | Includes bibliographical references and index. Identiļ¬ers: LCCN 2016041073 | ISBN 9781470429522 (alk. paper) Subjects: LCSH: Frames (Vector analysis) | Vector analysis. | Exterior diļ¬erential systems. | Geometry, Diļ¬erential. | Mathematical physics. | AMS: Topological groups, Lie groups ā Noncompact transformation groups ā Homogeneous spaces. msc | Diļ¬erential geometry ā Classical diļ¬erential geometry ā Curves in Euclidean space. msc | Diļ¬erential geometry ā Classical diļ¬erential geometry ā Surfaces in Euclidean space. msc | Diļ¬erential geometry ā Classical diļ¬erential geometry ā Aļ¬ne diļ¬erential geometry. msc | Diļ¬erential geometry ā Classical diļ¬erential geometry ā Projective diļ¬erential geometry. msc | Diļ¬erential geometry ā Classical diļ¬erential geometry ā Diļ¬erential invariants (local theory), geometric objects. msc | Diļ¬erential geometry ā Local diļ¬erential geometry ā Local submanifolds. msc | Diļ¬erential geometry ā Local diļ¬erential geometry ā Lorentz metrics, indeļ¬nite metrics. msc | Global analysis, analysis on manifolds ā General theory of diļ¬erentiable manifolds ā Diļ¬erential forms. msc | Global analysis, analysis on manifolds ā General theory of diļ¬erentiable manifolds ā Exterior diļ¬erential systems (Cartan theory). msc Classiļ¬cation: LCC QA433 .C564 2017 | DDC 515/.63ādc23 LC record available at https://lccn. loc.gov/2016041073
Copying and reprinting. Individual readers of this publication, and nonproļ¬t libraries acting for them, are permitted to make fair use of the material, such as to copy select pages for use in teaching or research. Permission is granted to quote brief passages from this publication in reviews, provided the customary acknowledgment of the source is given. Republication, systematic copying, or multiple reproduction of any material in this publication is permitted only under license from the American Mathematical Society. Permissions to reuse portions of AMS publication content are handled by Copyright Clearance Centerās RightsLink service. For more information, please visit: http://www.ams.org/rightslink. Send requests for translation rights and licensed reprints to [email protected]. Excluded from these provisions is material for which the author holds copyright. In such cases, requests for permission to reuse or reprint material should be addressed directly to the author(s). Copyright ownership is indicated on the copyright page, or on the lower right-hand corner of the ļ¬rst page of each article within proceedings volumes. c 2017 by the American Mathematical Society. All rights reserved. The American Mathematical Society retains all rights except those granted to the United States Government. Printed in the United States of America. ā The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability. Visit the AMS home page at http://www.ams.org/ 10 9 8 7 6 5 4 3 2 1
22 21 20 19 18 17
To Rick, Kevin, and Valerie, who make everything worthwhile
Contents
Preface
xi
Acknowledgments
xv
Part 1. Background material Chapter 1. Assorted notions from diļ¬erential geometry
3
Ā§1.1. Manifolds
3
Ā§1.2. Tensors, indices, and the Einstein summation convention
9
Ā§1.3. Diļ¬erentiable maps, tangent spaces, and vector ļ¬elds
15
Ā§1.4. Lie groups and matrix groups
26
Ā§1.5. Vector bundles and principal bundles
32
Chapter 2. Diļ¬erential forms
35
Ā§2.1. Introduction
35
Ā§2.2. Dual spaces, the cotangent bundle, and tensor products
35
Ā§2.3. 1-forms on
Rn
40
Ā§2.4. p-forms on
Rn
41
Ā§2.5. The exterior derivative
43
Ā§2.6. Closed and exact forms and the PoincarĀ“e lemma
46
Ā§2.7. Diļ¬erential forms on manifolds
47
Ā§2.8. Pullbacks
49
Ā§2.9. Integration and Stokesās theorem
53
Ā§2.10. Cartanās lemma
55 vii
viii
Contents
Ā§2.11. The Lie derivative
56
Ā§2.12. Introduction to the Cartan package for Maple
59
Part 2. Curves and surfaces in homogeneous spaces via the method of moving frames Chapter 3. Homogeneous spaces
69
Ā§3.1. Introduction
69
Ā§3.2. Euclidean space
70
Ā§3.3. Orthonormal frames on Euclidean space
75
Ā§3.4. Homogeneous spaces
84
Ā§3.5. Minkowski space
85
Ā§3.6. Equi-aļ¬ne space
92
Ā§3.7. Projective space
96
Ā§3.8. Maple computations
103
Chapter 4. Curves and surfaces in Euclidean space Ā§4.1. Introduction
107
Ā§4.2. Equivalence of submanifolds of a homogeneous space Ā§4.3. Moving frames for curves in
107
E3
108 111
Ā§4.4. Compatibility conditions and existence of submanifolds with prescribed invariants
115
Ā§4.5. Moving frames for surfaces in E3
117
Ā§4.6. Maple computations
134
Chapter 5. Curves and surfaces in Minkowski space Ā§5.1. Introduction
143 143
Ā§5.2. Moving frames for timelike curves in
M1,2
Ā§5.3. Moving frames for timelike surfaces in
M1,2
144 149
Ā§5.4. An alternate construction for timelike surfaces
161
Ā§5.5. Maple computations
166
Chapter 6. Curves and surfaces in equi-aļ¬ne space Ā§6.1. Introduction Ā§6.2. Moving frames for curves in
171 A3
Ā§6.3. Moving frames for surfaces in Ā§6.4. Maple computations
171
A3
172 178 191
Contents
ix
Chapter 7. Curves and surfaces in projective space Ā§7.1. Introduction
203 203
Ā§7.2. Moving frames for curves in P2
204
P3
214
Ā§7.3. Moving frames for curves in
P3
Ā§7.4. Moving frames for surfaces in
220
Ā§7.5. Maple computations
235
Part 3. Applications of moving frames Chapter 8. Minimal surfaces in E3 and A3
251
Ā§8.1. Introduction
251
Ā§8.2. Minimal surfaces in
E3
251
Ā§8.3. Minimal surfaces in
A3
268
Ā§8.4. Maple computations
280
Chapter 9. Pseudospherical surfaces and BĀØacklundās theorem
287
Ā§9.1. Introduction
287
Ā§9.2. Line congruences
288
Ā§9.3. BĀØacklundās theorem
289
Ā§9.4. Pseudospherical surfaces and the sine-Gordon equation
293
Ā§9.5. The BĀØacklund transformation for the sine-Gordon equation
297
Ā§9.6. Maple computations
303
Chapter 10. Two classical theorems
311
Ā§10.1. Doubly ruled surfaces in R3
311
Ā§10.2. The Cauchy-Crofton formula
324
Ā§10.3. Maple computations
329
Part 4. Beyond the ļ¬at case: Moving frames on Riemannian manifolds Chapter 11. Curves and surfaces in elliptic and hyperbolic spaces Ā§11.1. Introduction
339
Ā§11.2. The homogeneous spaces Ā§11.3. A more intrinsic view of
339
Sn
Sn
Ā§11.4. Moving frames for curves in
and and S3
Hn
Hn
and
340 345
H3
348
Ā§11.5. Moving frames for surfaces in S3 and H3
351
Ā§11.6. Maple computations
357
x
Contents
Chapter 12. The nonhomogeneous case: Moving frames on Riemannian manifolds 361 Ā§12.1. Introduction
361
Ā§12.2. Orthonormal frames and connections on Riemannian manifolds
362
Ā§12.3. The Levi-Civita connection
370
Ā§12.4. The structure equations
373
Ā§12.5. Moving frames for curves in 3-dimensional Riemannian manifolds
379
Ā§12.6. Moving frames for surfaces in 3-dimensional Riemannian manifolds
381
Ā§12.7. Maple computations
388
Bibliography
397
Index
403
Preface
Perhaps the earliest example of a moving frame is the Frenet frame along a nondegenerate curve in the Euclidean space R3 , consisting of a triple of orthonormal vectors (T, N, B) based at each point of the curve. First introduced by Bartels in the early nineteenth century [Sen31] and later described by Frenet in his thesis [Fre47] and Serret in [Ser51], the frame at each point is chosen based on properties of the geometry of the curve near that point, and the fundamental geometric invariants of the curveā curvature and torsionāappear when the derivatives of the frame vectors are expressed in terms of the frame vectors themselves. In the late nineteenth century, Darboux studied the problem of constructing moving frames on surfaces in Euclidean space [Dar72a], [Dar72b], Ā“ Cartan general[Dar72c], [Dar72d]. In the early twentieth century, Elie ized the notion of moving frames to other geometries (for example, aļ¬ne and projective geometry) and developed the theory of moving frames extensively. A very nice introduction to Cartanās ideas may be found in Guggenheimerās text [Gug77]. More recently, Fels and Olver [FO98], [FO99] have introduced the notion of an āequivariant moving frameā, which expands on Cartanās construction and provides new algorithmic tools for computing invariants. This approach has generated substantial interest and spawned a wide variety of applications in the last several years. This material will not be treated here, but several surveys of recent results are available; for example, see [Man10], [Olv10], and [Olv11a].
xi
xii
Preface
The goal of this book is to provide an introduction to Cartanās theory of moving frames at a level suitable for beginning graduate students, with an emphasis on curves and surfaces in various 3-dimensional homogeneous spaces. This book assumes a standard undergraduate mathematics background, including courses in linear algebra, abstract algebra, real analysis, and topology, as well as a course on the diļ¬erential geometry of curves and surfaces. (An appropriate diļ¬erential geometry course might be based on a text such as [dC76], [OāN06], or [Opr07].) There are occasional references to additional topics such as diļ¬erential equations, but these are less crucial. The ļ¬rst two chapters contain background material that might typically be taught in a graduate diļ¬erential geometry course; Chapter 1 contains general material from diļ¬erential geometry, while Chapter 2 focuses more speciļ¬cally on diļ¬erential forms. Students who have taken such a course might safely skip these chapters, although it might be wise to skim them to get accustomed to the notation that will be used throughout the book. Chapters 3ā7 are the heart of the book. Chapter 3 introduces the main ingredients for the method of moving frames: homogeneous spaces, frame bundles, and Maurer-Cartan forms. Chapters 4ā7 show how to apply the method of moving frames to compute local geometric invariants for curves and surfaces in 3-dimensional Euclidean, Minkowski, aļ¬ne, and projective spaces. These chapters should be read in order (with the possible exception of Chapter 5), as they build on each other. Chapters 8ā10 show how the method of moving frames may be applied to several classical problems in diļ¬erential geometry. The ļ¬rst half of Chapter 8, all of Chapter 9, and the last half of Chapter 10 may be read anytime after Chapter 4; the remainder of these chapters may be read anytime after Chapter 6. Chapters 11 and 12 give a brief introduction to the method of moving frames on non-ļ¬at Riemannian manifolds and the additional issues that arise when the underlying space has nonzero curvature. These chapters may be read anytime after Chapter 4. Exercises are embedded in the text rather than being presented at the end of each chapter. Readers are strongly encouraged to pause and attempt the exercises as they occur, as they are intended to engage the reader and to enhance the understanding of the text. Many of the exercises contain results which are important for understanding the remainder of the text; these exercises are marked with a star and should be given particular attention. (Even if you donāt do them, you should at least read them!)
Preface
xiii
A special feature of this book is that it includes guidance on how to use the mathematical software package Maple to perform many of the computations involved in the exercises. (If you do not have access to Maple, rest assured that, with very few exceptions, the exercises can be done perfectly well by hand.) The computations here make use of the custom Maple package Cartan, which was written by myself and Yunliang Yu of Duke University. The Cartan package can be downloaded either from the AMS webpage www.ams.org/bookpages/gsm-178 or from my webpage at http://euclid.colorado.edu/~jnc/Maple.html. (Installation instructions are included with the package.) The last section of Chapter 2 contains an introduction to the Cartan package, and beginning with Chapter 3, each chapter includes a section at the end describing how to use Maple and the Cartan package for some of the exercises in that chapter. Additional exercises are worked out in Maple worksheets for each chapter that are available on the AMS webpage. Remark. As of Maple 16 and above, much of Cartanās functionality is now available as part of the DifferentialGeometry package, which is included in the standard Maple installation and covers a wide range of applications. The two packages have very diļ¬erent syntax, and no attempt will be made here to translateābut interested readers are encouraged to do so!
Acknowledgments
First and foremost, my deepest thanks go to Robert Bryantāmy teacher, mentor, and friendāfor inviting me to teach alongside him at the Mathematical Sciences Research Institute in the summer of 1999, when I was a mere three years post-Ph.D.; for not laughing out loud when I naively mentioned the idea of turning the lecture notes into a book (although he probably should have); and for unļ¬agging support in more ways than I can count over the years. Thanks also to Edward Dunne and Sergei Gelfand at the American Mathematical Society for expressing interest in the project early on and for extreme patience and not losing faith in me as it dragged on for many more years than I ever imagined. I am also grateful to the anonymous reviewers for the AMS who read initial drafts of the manuscript, pointed out signiļ¬cant errors, and made valuable suggestions for improvements. I am forever grateful to Bryan Kaufman and Nathaniel Bushek, who in 2009 asked if I would supervise an independent study course for them. I suggested that they work through my nascent manuscript, and they eagerly agreed, struggling through a version that consisted of little more than the original lecture notes. Their questions and suggestions were invaluable and had a major impact on the tone, content, and structure of the book. This project might have stayed forever on my to-do list if not for them. Thanks especially to Bryan for suggesting that I add the material on curves and surfaces in Minkowski space and to Sunita Vatuk for recommending the book [Cal00] on this material.
xv
xvi
Acknowledgments
Thanks to all the other students who have worked through subsequent versions of the manuscript over the last several years: Brian Carlsen, Michael Schmidt, Edward Estrada, Molly May, Jonah Miller, Sean Peneyra, Duļ¬ Baker-Jarvis, Akaxia Cruz, Rachel Helm, Peter Joeris, Joshua Karpel, Andrew Jensen, and Michael Mahoney. These independent study coursesāand the research projects that followedāhave been, hands down, the most rewarding experiences of my teaching career. I hope you all enjoyed them half as much as I did! And thanks to Sunita Vatuk and George Wilkens for sitting in on some of these courses, contributing many valuable insights to our discussions, and making great suggestions for the manuscript. I am grateful to the Mathematical Sciences Research Institute for sponsoring the 1999 Summer Graduate Workshop where I gave the lectures that were the genesis for this book; videos of the original lectures are available on MSRIās webpage at [Cle99]. I am also grateful to the National Science Foundation for research support; portions of this book were written while I was supported by NSF grants DMS-0908456 and DMS-1206272. Finally, profound thanks to my husband, Rick; his love and support have been constant and unwavering, and I count myself fortunate beyond all measure to have him as my best friend and partner in life.
Part 1
Background material
10.1090/gsm/178/01
Chapter 1
Assorted notions from diļ¬erential geometry This chapter contains some useful background material from diļ¬erential geometry. Much of this material would typically be covered in courses on multivariable analysis, algebra, and graduate-level diļ¬erential geometry and linear algebra. This chapter is not intended to be a comprehensive introduction to any of these topics; the focus will be on ideas that will be used in the remainder of the book, and details will be kept to a minimum.
1.1. Manifolds Just about all the objects that we discuss in this book will be deļ¬ned on a manifold of some sort. The simplest manifolds are things that you already know about: regular curves (1-dimensional manifolds) and regular surfaces (2-dimensional manifolds). Before making a more general deļ¬nition, letās remind ourselves of some of the fundamental properties of regular surfaces. 1.1.1. Regular surfaces. First, recall the deļ¬nition of a regular surface in R3 (see, e.g., [dC76]): Deļ¬nition 1.1. A subset Ī£ ā R3 is a regular surface if for each point q ā Ī£, there exists a neighborhood V ā R3 of q and a map x : U ā V ā© Ī£ from an open set U ā R2 onto V ā© Ī£ with the following properties: (1) x is diļ¬erentiable. (2) x is a homeomorphism. (3) For each u ā U , the diļ¬erential dxu : R2 ā R2 is one-to-one. 3
4
1. Assorted notions from diļ¬erential geometry
The mapping x is called a parametrization of Ī£, or a system of local coordinates on Ī£, in a neighborhood of q. Remark 1.2. Throughout this book, the word ādiļ¬erentiableā will be taken to mean āinļ¬nitely diļ¬erentiableā (i.e., C ā ). We will also use the word āsmoothā as a synonym. Any regular surface Ī£ ā R3 can be covered by a family of parametrizations xi : Ui ā Ī£, where each Ui is an open set in R2 and each xi is an injective, diļ¬erentiable map from Ui to R3 . The image xi (Ui ) of each parametrization is an open set in Ī£ (in the subspace topology inherited from R3 ); moreover, if the images of two parametrizations xi : Ui ā Ī£ and xj : Uj ā Ī£ have nonempty intersection V ā Ī£, then the composite maps (1.1)
ā1 ā1 xā1 j ā¦ xi : xi (V ) ā xj (V ),
ā1 ā1 xā1 i ā¦ xj : xj (V ) ā xi (V )
are diļ¬erentiable maps between open sets in R2 . (See Figure 1.1.)
................................................................................. ......................... ................................................ ......................... .................... .................... .................. .................... ........................................ ....................................................... . . . . . . . . . . . . .. . . ... .......... ....... .................. . . . . . . . . . . . . . . . ... ... . . ..... ..... .... ..... . .. . . . . . . .. . . . .... .... .. .. .. ... . . . . . . . . . . ... ... . . ... .. ... ... ... ... .. j .. i ... .. .. .. .. .. .... ..... .. .. . . ... .. .. .. .. ... . . . . . .. . ... ... . .. .... ... ... .... ... .. .... .... .... .... ... .. ..... ...... .... ..... ... . ........ ......... ............. ....... . . . . . ........... . . . . . . . .... . . . .................................................. ...................................................... .. .. .. .. .. .. .. .. . ............................................................................................................... .... .................................................................. . . .......................... ... ............... .................
Ī£
x (U )
V
x (U )
xi
@ I @ xj @ @ @ vĀÆ 6 ............................
v
6
.............................................. .............. .... ......... .............. ....... ...... ......... ..... . .... .... . .... ... ... . . . i ... ... ā1 ... ... .... .... ... i ... ... . .. .. .. ... ... .. ... ... .. . . ... ... . . . ... .... .... ... .... ..... ..... ........ ........ .............. .......... .................... ............................. ...........
xā1 j ā¦ xi
U
x
(V )
-u
xā1 i ā¦ xj
-
............... ..... ......... ............. ....... ........... ..... ....... .......... .... .... ..... . . .. .... ... . ... j ...... .. ā1 . . ... ... .... j ... ... . ... .. .. .. .. ... .. .. . . ... .. . . . . ... ... ... ... .... .... .... ... ..... ... ................... ....... . . . . .......... . . . . .................... .......................... ...........
U
x
(V )
- uĀÆ
Figure 1.1. Overlapping parametrizations
When Ī£ is a regular surface in R3 , the diļ¬erentiability of the maps (1.1) is a theorem. But regular surfaces can also be deļ¬ned as abstract, intrin-
1.1. Manifolds
5
sic objects, not living in any particular Euclidean space. In this case, the diļ¬erentiability of (1.1) becomes part of the deļ¬nition of a regular surface: Deļ¬nition 1.3. A regular surface is a set Ī£ and a family of injective mappings xi : Ui ā Ī£, where each Ui is an open set in R2 , with the following properties: (1)
xi (Ui ) = Ī£.
i
(2) For each pair i, j with V = xi (Ui ) ā© xj (Uj ) = ā
, the sets xā1 i (V ) ā1 ā1 2 and xj (V ) are open sets in R , and the mappings xj ā¦ xi and xā1 i ā¦ xj are diļ¬erentiable. The mappings xi are called parametrizations of Ī£, or systems of local coordinates on Ī£. Incorporating condition (2) as an axiom allows us to deļ¬ne objects such as diļ¬erentiable functions on surfaces and diļ¬erentiable maps between surfaces in a way that is independent of a choice of parametrization. All the usual notions of diļ¬erential calculus on R2 can then be extended to analogous notions on regular surfaces. For an abstract surface, the maps xi are rarely deļ¬ned explicitly. (This is very diļ¬erent from how we view surfaces in R3 , where the parametrizations are often used to deļ¬ne the surface!) Rather, we think of an abstract surface as a collection of open sets Ui in R2 , āglued togetherā via the transition maps 2 xā1 j ā¦ xi . It is these transition maps between open sets in R that may (or may not) be deļ¬ned explicitly; the xi should just be thought of as a means of identifying a part of the surface with a system of local coordinates (u, v). If (u, v) are local coordinates on Ui , (ĀÆ u, vĀÆ) are local coordinates on Uj , and the set V = xi (Ui ) ā© xj (Uj ) ā Ī£ is nonempty, then the transition map ā1 ā1 xā1 j ā¦ xi : xi (V ) ā xj (V )
is a local coordinate transformation of the form xā1 u(u, v), vĀÆ(u, v)). j ā¦ xi (u, v) = (ĀÆ (See Figure 1.1.)
1.1.2. Manifolds: from 2 to n. A manifold (or, more precisely, a diļ¬erentiable manifold) of dimension n ā„ 1 is simply what we get by replacing
6
1. Assorted notions from diļ¬erential geometry
the number 2 in Deļ¬nition 1.3 with n: Deļ¬nition 1.4. A diļ¬erentiable manifold of dimension n ā„ 1 is a set M and a family of injective mappings xi : Ui ā M , where each Ui is an open set in Rn , with the following properties: (1) xi (Ui ) = M . i
(2) For any pair i, j with V = xi (Ui ) ā© xj (Uj ) = ā
, the sets xā1 i (V ) and ā1 ā1 n xj (V ) are open sets in R , and the mappings xj ā¦xi and xā1 i ā¦xj are diļ¬erentiable. The mappings xi are called parametrizations of M , or systems of local coordinates on M . Remark 1.5. When studying regular surfaces in R3 , it is traditional to use (u, v) as local coordinates on R2 and to write parametrizations x : U ā R3 as ā” ā¤ x(u, v) x(u, v) = ā£y(u, v)ā¦ . z(u, v) When we graduate to manifolds of arbitrary dimension, we need to use variables with indices so that we donāt run out of letters. So, depending on the context, we will generally use either u = (u1 , . . . , un ) or x = (x1 , . . . , xn ) as local coordinates on an open set U ā Rn . Just as for regular surfaces, a manifold M of dimension n can be a subset of some Euclidean space Rk with k ā„ n, or it can be an intrinsic object not living in any ambient Euclidean space. When M is a subset of Rk , it may be deļ¬ned by explicit parametrizations xi : Ui ā Rk , where the Ui are open sets in Rn . When M is an intrinsic manifold, the maps xi are generally not deļ¬ned explicitly; we simply think of M as a collection of open sets Ui ā Rn that are glued together via the transition maps xā1 j ā¦ xi . As for surfaces, these transition maps are local coordinate transformations of the form (1.2)
1 n xā1 u1 (u1 , . . . , un ), . . . , u ĀÆn (u1 , . . . , un )). j ā¦ xi (u , . . . , u ) = (ĀÆ
1.1.3. Examples. Example 1.6 (The unit sphere). Let Sn ā Rn+1 be the set Sn = {t[x1 , . . . , xn+1 ] ā Rn+1 | (x1 )2 + Ā· Ā· Ā· + (xn+1 )2 = 1}; i.e., Sn is the set of all vectors in Rn+1 of Euclidean length 1. Remark 1.7. Vectors in Rk are assumed to be column vectors unless otherwise speciļ¬ed. The notation t[x1 , . . . , xn+1 ] in Example 1.6 denotes the transpose of the row vector [x1 , . . . , xn+1 ], which is the column vector
1.1. Manifolds
7
ā”
ā¤ x1 ā¢ .. ā„ ā£ . ā¦. We will often write column vectors in this way in order to save xn+1 space, using the transpose notation in order to maintain the distinction between column vectors and row vectors. Exercise 1.8. Sn can be covered by parametrizations in the same fashion as the 2-dimensional sphere S2 ā R3 ; it just takes a bit more bookkeeping to keep up with all the indices. For i = 1, . . . , n + 1, let Vi+ = {t[x1 , . . . , xn+1 ] ā Sn | xi > 0}, Viā = {t[x1 , . . . , xn+1 ] ā Sn | xi < 0}. Let U = {t[u1 , . . . , un ] ā Rn | (u1 )2 + Ā· Ā· Ā· + (un )2 < 1}; i.e., U is the open unit ball in Rn . Deļ¬ne maps + x+ i : U ā Vi ,
ā xā i : U ā Vi
by 1 n x+ i (u , . . . , u ) = 1 n xā i (u , . . . , u ) =
t
u1 , . . . , uiā1 ,
1 ā (u1 )2 ā Ā· Ā· Ā· ā (un )2 , ui , . . . , un ,
t
u1 , . . . , uiā1 , ā 1 ā (u1 )2 ā Ā· Ā· Ā· ā (un )2 , ui , . . . , un .
ā (a) What portion of Sn is covered by the images of x+ i and xi ? Show that every point in Sn is contained in the image of at least one of these parametrizations. + n (b) For i = j, identify the open set Vij+ = x+ i (U ) ā© xj (U ) in S and the open + ā1 + + ā1 + sets (xj ) (Vij ) and (xi ) (Vij ) in U , and compute the local coordinate ā1 ā¦ x+ between the latter two open sets. transformation (x+ j ) i
Example 1.9 (Real projective space of dimension n). Let Pn denote the set of lines through the origin in the vector space Rn+1 . Since any such line is determined by any point on the line other than the origin, we can think of Pn as the quotient space Pn = (Rn+1 \ {0})/ ā¼, where ā¼ represents the equivalence relation deļ¬ned by the condition that two points x, y ā Rn+1 \ {0} satisfy x ā¼ y if and only if x and y lie on
8
1. Assorted notions from diļ¬erential geometry
the same line through the origin. It is customary to use local coordinates (x0 , x1 , . . . , xn ) on Rn+1 ; then for any nonzero real number Ī», we have [x0 , . . . , xn ] ā¼ t[Ī»x0 , . . . , Ī»xn ].
t
The equivalence class of the point x = t[x0 , . . . , xn ] is denoted by [x] = [x0 : Ā· Ā· Ā· : xn ]; these are called the homogeneous coordinates for a point in Pn . Note that each line through the origin in Rn+1 intersects the unit sphere Sn in exactly two points, which form an antipodal pair {x, āx}. So, an alternative (but equivalent!) way to think of Pn is as the manifold obtained from Sn by identifying every point x ā Sn with its polar opposite āx. Even better, we can identify Pn with the upper half of Sn (which is topologically an n-dimensional disk), with its boundary glued together so as to identify opposite points on the boundary. Note that gluing the boundaryā which is a copy of Snā1 ātogether in this fashion produces a copy of Pnā1 . So, yet another way to think of Pn is as an open n-dimensional disk (which is topologically the same as Rn ) with a copy of Pnā1 attached along its boundary. (See Figure 1.2 for an idea of how this looks for P1 and P2 .) P2
S1
............. ..... ....... .r ...r
.................................. ........... ....... ..... ....... ..... .... ... ... . . ... .. . ... .... ... ... . ... .. ... .. ... .. ... .. ... . . . . . .. . . .. . . . . . . .. .. . . . . . . .. .. . . . . . . . .. ...... .. ........ ..... . . ....... . .... . . . ............ . . . . . . . . . ......................................................... . ... . . . ... .. .... ..... .... ....... ... .... ............ ................. ................. .....................................
P1 ........... . . . .... .... .........r.....
r
O
r K
r
r
Figure 1.2. Construction of P1 and P2
Unlike the unit sphere Sn , Pn is not obviously deļ¬ned as a subset of any Euclidean space Rk . The following exercise shows how Pn can be covered by parametrizations. Exercise 1.10. For i = 0, . . . , n, let Vi = {[x0 : Ā· Ā· Ā· : xn ] ā Pn | xi = 0}. Let U = Rn , with coordinates u = (u1 , . . . , un ). Deļ¬ne maps [xi ] : U ā Pn by [xi (u1 , . . . , un )] = [u1 : Ā· Ā· Ā· : ui : 1 : ui+1 : Ā· Ā· Ā· : un ].
1.2. Tensors, indices, and the Einstein summation convention
9
(a) What portion of Pn is covered by the image of [xi ]? Show that every point in Pn is contained in the image of at least one of these parametrizations. (b) For i = j, identify the set Vij = [xi (U )] ā© [xj (U )] in Pn and the open sets [xj ]ā1 (Vij ) and [xi ]ā1 (Vij ) in U , and compute the local coordinate transformation [xj ]ā1 ā¦ [xi ] between the latter two open sets.
1.2. Tensors, indices, and the Einstein summation convention In equation (1.2), the variables are indexed with superscripts rather than subscripts. You may be wonderingāwhy on earth would we do that? So, before we go any farther, letās discuss how (and why!) indices are used throughout this book. Some objects will be indexed with superscripts (āupper indicesā or, less formally, āup indicesā), some by subscripts (ālower indicesā or ādown indicesā), and some by a combination of both. Most of these objects will be tensors or tensor ļ¬elds. Without getting too speciļ¬c, a tensor is just an element of a certain type of vector space, and a tensor ļ¬eld on a manifold is deļ¬ned by assigning a tensor to each point of the manifold. (A precise deļ¬nition will be given in Chapter 2.) For example, tensors include objects such as vectors and linear transformations, while tensor ļ¬elds include objects such as vector ļ¬elds, metrics, and diļ¬erential forms on manifolds. A key feature of tensors is the way in which their coordinate expressions change when the basis for the underlying vector space is changed. The following examples demonstrate some typical behavior. Example 1.11. Consider an n-dimensional vector space V , and let (e1 , . . ., en ) be a basis for V . Any vector v ā V can be expressed uniquely as v = a 1 e 1 + Ā· Ā· Ā· + an e n for some real numbers a1 , . . . , an ā R. ĀÆn )? Now, what happens if we express v in terms of a diļ¬erent basis (ĀÆ e1 , . . . , e Suppose that the new basis can be written in terms of the old basis as ĀÆi = e
n
rik ek ,
i = 1, . . . , n,
k=1
or, in more compact matrix notation,
ĀÆ1 . . . e ĀÆn = e1 . . . en R, e
10
1. Assorted notions from diļ¬erential geometry
ā”
where
r11 ā¢ .. R=ā£.
ā¤ . . . rn1 .. ā„ . .ā¦
r1n . . . rnn Then we have
ĀÆ1 . . . e ĀÆn Rā1 , e1 . . . en = e
and so v = a1 e1 + Ā· Ā· Ā· + an en ā” 1ā¤ a ā¢ . ā„
= e1 . . . en ā£ .. ā¦ an
ĀÆ1 = e
ā”
ā¤ a1 ā¢ ā„ ĀÆn Rā1 ā£ ... ā¦ ... e an
ĀÆ1 + Ā· Ā· Ā· + a ĀÆn , =a ĀÆ1 e ĀÆn e where
ā”
ā¤ ā” 1ā¤ a ĀÆ1 a ā¢ .. ā„ ā1 ā¢ .. ā„ ā£ . ā¦ = R ā£ . ā¦. a ĀÆn
an
While this is a simple (and hopefully familiar!) example, it illustrates some important points: (1) A vector is an example of a rank 1 tensor. āRank 1ā means that, when the vector is expressed in terms of a given basis, the componentsāin this case, (a1 , . . . , an )āeach have one index. (2) The matrix e = [e1 . . . en ] (Careful: It looks like a row vector, but each entry is really a column vector!) and the column vector a are basis-dependent, but their product v = ea is well-deļ¬ned independently of a choice of basis: When the basis vectors (e1 , . . . , en ) are changed, the coeļ¬cients (a1 , . . . , an ) change in such a way that the changes cancel each other out in the product. (3) When the matrix e, whose entries are indexed by lower indices, is transformed by R, the vector a, whose entries are indexed by upper indices, is multiplied by Rā1 . This is not an accident! In general, the placementāup or downāof indices for the components of a tensor is dictated by how these components transform under a change of basis for the underlying vector space.
1.2. Tensors, indices, and the Einstein summation convention
11
Example 1.12. Let V be an m-dimensional vector space with basis (e1 , . . ., em ) and W an n-dimensional vector space with basis (f1 , . . . , fn ). Let T : V ā W be a linear transformation. In terms of the given bases for V and W , T can be represented as an n Ć m matrix ā¤ ā” 1 c1 . . . c1m ā¢ .. ā„ . AT = ā£ ... . ā¦ n n c1 . . . cm Letās think carefully about what this means. We usually write a vector v ā V as a column vector ā” 1ā¤ a ā¢ .. ā„ a=ā£ . ā¦ am (by which we really mean that v = [e1 . . . em ] Ā· a), and we write w = T (v) as the matrix product ā” 1ā¤ ā” 1 ā¤ā” 1ā¤ b c1 . . . c1m a ā¢ .. ā„ ā¢ .. ā„ ā¢ . .. ā¦ ā£ ... ā„ b=ā£.ā¦=ā£. ā¦ n n n m b c1 . . . cm a (by which we really mean that w = [f1 . . . fn ] Ā· b). The use of diļ¬erent notations here for the vectors v and w, which are well-deļ¬ned independently of a choice of basis, and their expressions a and b in terms of speciļ¬c bases, is deliberate: The equation b = AT a is a basis-dependent expression of the basis-independent equation w = T (v). What it really means is that
w = T (v) = T
=
n j=1
m
ai e i
i=1 m
cji ai
fj .
i=1
In more compact matrix notation, ā ā” 1 ā¤ā a ā¢ . ā„ā ā (1.3) T ā e1 . . . em ā£ .. ā¦ā = f1 . . . fn AT am
ā”
ā¤ a1 ā¢ .. ā„ ā£ . ā¦. am
12
1. Assorted notions from diļ¬erential geometry
Again, we ask what happens if we change bases for V and W . Suppose that we set
ĀÆ1 . . . e ĀÆm = e1 . . . em R, e (1.4)
ĀÆf1 . . . ĀÆfn = f1 . . . fn S, where R is an invertible m Ć m matrix and S is an invertible n Ć n matrix. We want to ļ¬nd the matrix AT that satisļ¬es ā ā” 1ā¤ ā” 1 ā¤ā a ĀÆ a ĀÆ ā¢ . ā„ā ā ā¢ .. ā„ ĀÆ ĀÆ . ĀÆ1 . . . e ĀÆm ā£ . ā¦ā = f1 . . . fn AT ā£ . ā¦ . Tā e a ĀÆm a ĀÆm Using (1.3) and (1.4), we can compute: ā ā ā” 1 ā¤ā ā” 1 ā¤ā a ĀÆ a ĀÆ ā¢ . ā„ā ā¢ . ā„ā ā ā . ĀÆ1 . . . e ĀÆm ā£ . ā¦ā = T ā e1 . . . em R ā£ .. ā¦ā Tā e a ĀÆm a ĀÆm ā” 1ā¤ a ĀÆ
ā¢ .. ā„ = f1 . . . fn AT R ā£ . ā¦ a ĀÆm ā” 1ā¤ a ĀÆ
ā1 ā¢ .. ā„ ĀÆ ĀÆ = f1 . . . fn (S AT R) ā£ . ā¦ . a ĀÆm So, in terms of the new bases, the new matrix representation for T is AT = S ā1 AT R. Some observations about this example: (1) The linear transformation T is an example of a rank 2 tensor. The components (cij ) of its representation AT in terms of speciļ¬c bases for V and W have an upper index i with range 1 ā¤ i ā¤ n, corresponding to the vector space W , and a lower index j with range 1 ā¤ j ā¤ m, corresponding to the vector space V . (2) When the basis for V is transformed by a matrix R and the basis for W is transformed by a matrix S, the matrix representation AT is multiplied on the left by S ā1 and on the right by R. This illustrates the general phenomenon that, under a change of basis: ā¢ Down indices indicate that components will transform by right multiplication by the matrix for the basis transformation.
1.2. Tensors, indices, and the Einstein summation convention
13
ā¢ Up indices indicate that components will transform by left multiplication by the inverse of the matrix for the basis transformation. WARNING: When applying this guideline, one must think very carefully about which direction the transformation goes in order to get the inverses in the right places! Remark 1.13. These were fairly simple examples of tensors, dealing with a single vector and a single linear transformation. We will mostly be interested in tensor ļ¬elds, which consist of an underlying manifold with a tensor deļ¬ned at each point. It may sound complicated, but you already know several examplesāe.g., vector ļ¬elds and metrics on surfaces. The only real diļ¬erences are that: (1) The components are functions on the base manifold rather than constants. (2) The vector spaces in question are usually the tangent or cotangent spaces to the manifold at each point. (3) Basis changes to the vector spaces usually arise as derivatives (i.e., Jacobian matrices) of local coordinate transformations on the manifold. Remark 1.14. When working with partial derivatives, up indices in the denominator count as down indices, and vice versa. For example, when working on a manifold coordinates (x1 , . . . , xn ), the partial de ā with local rivative operators āx1 , . . . , āxān will often be used as a local basis for the tangent space at each point (the reason for this will be explained in Ā§1.3), and the indices on these operators should be regarded as down indices. *Exercise 1.15. A metric g on a manifold M is deļ¬ned by specifying for each point q ā M a symmetric, positive deļ¬nite bilinear form gq on the tangent space Tq M . (Tangent spaces will be deļ¬ned in Ā§1.3; all you need to know here is that Tq M is an n-dimensional vector space.) This means that gq is a bilinear function gq : Tq M Ć Tq M ā R such that for all v, w ā Tq M , we have gq (v, w) = gq (w, v) and gq (v, v) ā„ 0, with gq (v, v) = 0 if and only if v = 0. If we have a basis (e1 , . . ., en ) for the tangent space Tq M , then we deļ¬ne the components of gq with respect to this basis to be the real numbers {gij (q) = gq (ei , ej ) | 1 ā¤ i, j ā¤ n}.
14
1. Assorted notions from diļ¬erential geometry
The assumption that g is symmetric implies that gji (q) = gij (q), and the bilinearity of g implies that for any vectors n n v= ai e i , w= bj e j i=1
j=1
in Tq M , we have gq (v, w) =
n n
gij (q)ai bj .
i=1 j=1
For instance, when M is a regular surface Ī£ ā R3 with a local parametrization x : U ā Ī£, we typically use the basis e1 = xu ,
e2 = xv
for Tq M , and the bilinear form is given by the dot product in R3 : gq (v, w) = v Ā· w. Then we have g11 = E = xu Ā· xu ,
g12 = g21 = F = xu Ā· xv ,
g22 = G = xv Ā· xv .
A metric g is often represented by the symmetric matrix ā¤ ā” g11 . . . g1n ā¢ .. ā„ . Ag = ā£ ... . ā¦ gn1 . . . gnn Suppose that we make a change of basis
ĀÆ1 . . . e ĀÆ n = e1 . . . en R e for Tq M . Compute the components (ĀÆ gij (q)) of gq with respect to the new ĀÆn ). How does the matrix Ag transform? Compare the transbasis (ĀÆ e1 , . . . , e formation rule for Ag to that given in Example 1.12 for the matrix representation AT for a linear transformation T from a vector space V to itself. We close this section by introducing the Einstein summation convention. In working with tensors and tensor ļ¬elds, there are a lot of sums involved: A vector is expressed as n v= ai e i ; i=1
a metric g acting on a pair of vectors v =
n
i
a ei , w =
i=1
as g(v, w) =
n n i=1 j=1
gij ai bj ,
n j=1
bj ej is expressed
1.3. Diļ¬erentiable maps, tangent spaces, and vector ļ¬elds
15
etc. So, in order to reduce notational clutter, we omit the explicit sum notation and simply write v = ai ei , g(v, w) = gij ai bj . The Einstein summation convention says that whenever the same index appears twice in an expression, once up and once down, it indicates a sum over the entire range of that index. This takes a bit of getting used to, but itās actually extremely handy. For instance, it makes the multivariable chain rule look exactly like the single-variable version, where you can ācancelā the intermediate variables: āf āf āy j = . āxi āy j āxi It also gives a quick method for error checking, much like dimensional analysis in chemistry or physics: Once repeated indices are ācanceledā, the indices on both sides of an equation should match.
1.3. Diļ¬erentiable maps, tangent spaces, and vector ļ¬elds Let U ā Rm be an open set, and consider a function F : U ā Rn . (Such a function is often called a mapping, or simply a map, from U to Rn .) We can write ā” 1 1 ā¤ f (x , . . . , xm ) ā¢ ā„ .. F (x1 , . . . , xm ) = ā£ ā¦ . n 1 m f (x , . . . , x ) for some real-valued functions f 1 , . . . , f n : U ā R. We say that F is continuous if each of the functions f 1 , . . . , f n is continuous, and F is diļ¬erentiable if each of the functions f 1 , . . . , f n is diļ¬erentiable. Now, if a function is diļ¬erentiable, then it really ought to have a derivative. So, what is the appropriate notion for the derivative of such a map? First, consider the case m = 1. In this case, U ā R is an open interval I, and the image of F is a curve in Rn . (In this case, we usually denote the function by a lowercase Greek letter rather than by a capital Roman letter.) Given a curve Ī± : I ā Rn deļ¬ned by ā” 1 ā¤ y (t) ā¢ .. ā„ Ī±(t) = ā£ . ā¦ , y n (t)
16
1. Assorted notions from diļ¬erential geometry
the derivative of Ī± at any point t ā I is often simply deļ¬ned to be the column vector ā” dy1 ā¤ dt (t) ā¢ .. ā„ (1.5) Ī± (t) = ā£ . ā¦ , dy n dt (t)
also called the tangent vector to Ī± at the point Ī±(t). But the expression (1.5) is actually an incomplete description of the tangent vector because the base point of a vector is a crucial part of its deļ¬nition. A more accurate deļ¬nition would be āā” 1 ā¤ ā” dy1 ā¤ā y (t) dt (t) āā¢ .. ā„ ā¢ .. ā„ā (1.6) Ī± (t) = āā£ . ā¦ , ā£ . ā¦ā , dy n y n (t) dt (t) including both the base point and its derivative. While the abbreviated notation (1.5) is common, it is important to remember that it represents a vector based at the speciļ¬c point Ī±(t) in Rn . Given a point y = t[y 1 , . . . , y n ] ā Rn , the tangent space to Rn at y, denoted Ty Rn , is the set of all tangent vectors to all curves in Rn passing through y; i.e., Ty Rn = {Ī± (0) | Ī± : I ā Rn is a smooth curve with Ī±(0) = y}. Ty Rn is an n-dimensional vector space, and it is canonically isomorphic to the base space Rn . This means not only that Ty Rn is isomorphic to Rn (which is obvious, since both are n-dimensional vector spaces), but also that there is one particular isomorphism between them that is somehow the ārightā one. Speciļ¬cally, this isomorphism is deļ¬ned by which identi (1.5), n dy 1 n t ļ¬es the tangent vector Ī± (0) ā Ty R with the vector dt (0), . . . , dydt (0) ā Rn . Now let m be arbitrary, and let F : U ā Rm ā Rn be a diļ¬erentiable map. Fix a point x ā U . The derivative or diļ¬erential of F (denoted by dF or Fā , depending on the context) at x is a linear map from Tx Rm to TF (x) Rn deļ¬ned as follows: Given a tangent vector v ā Tx Rm , let Ī± : I ā Rm be a smooth curve with Ī±(0) = x and Ī± (0) = v. Then dFx (v) = (F ā¦ Ī±) (0) ā TF (x) Rn . (See Figure 1.3.)
1.3. Diļ¬erentiable maps, tangent spaces, and vector ļ¬elds
v=Ī± (0)
Fā¦Ī±
.................... ............. ......... ....... ..... . . . ... .... .... .... ... . ... .. .. .. .... .. ..
r
F
x=Ī±(0)
Ī±
-
17
rP dF (v)=(F ā¦ Ī±) (0) PP q P
............................................................ ............ ......... x ........ ...... ..... .... .... ... ... .
F (x)
Figure 1.3. Construction of dFx (v)
While it may not be obvious, it is true that: (1) dFx (v) is well-deļ¬ned, independent of the choice of the curve Ī±. (2) dFx : Tx Rm ā TF (x) Rn is a linear map. (3) The matrix representation for dFx in terms of the canonical bases for Tx Rm and TF (x) Rn is the Jacobian matrix of F at x. *Exercise 1.16. Prove the statements above: Let F : U ā Rm ā Rn be a diļ¬erentiable map. (a) Let x ā U , and let v ā Tx Rm be a tangent vector. Let Ī± : I ā Rm , Ī² : I ā Rm be two curves with Ī±(0) = Ī²(0) = x,
Ī± (0) = Ī² (0) = v.
Show by direct computation that (F ā¦ Ī±) (0) = (F ā¦ Ī²) (0) ā TF (x) Rn . This shows that the diļ¬erential dFx (v) is well-deļ¬ned. (b) Let x ā U , and let v, w ā Tx Rm be tangent vectors. Show that dFx (v + w) = dFx (v) + dFx (w), and for any real number c, dFx (cv) = c dFx (v). This shows that dFx (v) is a linear map. (Hint: Let Ī± : I ā Rm , Ī² : I ā Rm be two curves with Ī±(0) = Ī²(0) = x, Write
ā”
ā¤ x11 (t) ā¢ ā„ Ī±(t) = ā£ ... ā¦ , xm 1 (t)
Ī± (0) = v,
Ī² (0) = w. ā”
ā¤ x12 (t) ā¢ ā„ Ī²(t) = ā£ ... ā¦ . xm 2 (t)
18
1. Assorted notions from diļ¬erential geometry
Let Ī³ be the curve
ā” ā¢ Ī³(t) = ā£
1 2 1 2
1 ā¤ x1 (2t) + x12 (2t) ā„ .. ā¦. . m (xm 1 (2t) + x2 (2t))
Then Ī³ satisļ¬es Ī³(0) = x and Ī³ (0) = v + w.) (c) Let
ā”
ā¤ y 1 (x1 , . . . , xm ) ā¢ ā„ .. F (x1 , . . . , xm ) = ā£ ā¦. . y n (x1 , . . . , xm )
Show that the matrix representation for dFx in terms of the canonical bases for Tx Rm and TF (x) Rn is the Jacobian matrix i āy J= . āxj This means that if v has canonical representation a = t[a1 , . . . , am ], then the canonical representation for dFx (v) is ā” āy1 j ā¤ a āxj ā¢ .. ā„ Ja = ā£ . ā¦ . āy n j a āxj
(Hint: What is the simplest curve Ī±(t) that you can think of with Ī±(0) = x and Ī± (0) = v?) Now letās get a little more abstract and deļ¬ne the analogous notions for manifolds. Deļ¬nition 1.17. Suppose that M and N are m- and n-dimensional manifolds, respectively. A map F : M ā N is called diļ¬erentiable if for any parametrizations x : U ā Rm ā M,
y : V ā Rn ā N
on M and N such that F (x(U )) ā y(V ), the composite map yā1 ā¦ F ā¦ x : U ā V is diļ¬erentiable. (Deļ¬nition 1.4 implies that this condition is independent of the choice of parametrizations x and y.) Remark 1.18. Keeping track of all these diļ¬erent maps can be challenging, and so they are often described using diagrams. For example, the maps
1.3. Diļ¬erentiable maps, tangent spaces, and vector ļ¬elds
19
above might be depicted as follows: M x
6
Rm ā U
F
- N 6y
yā1 ā¦F ā¦x
- V ā Rn .
This is called a commutative diagram. (Sometimes we simply say that āthe diagram commutesā.) This means that if we start with a point in U and map it to N , then we will arrive at the same point of N regardless of which path we take. Speciļ¬cally, if u ā U , then (F ā¦ x)(u) = (y ā¦ (yā1 ā¦ F ā¦ x))(u) ā N. Of course, the commutativity of the diagram is obvious in this case, but sometimes it can indicate something more substantial. In order to deļ¬ne the derivative of such a map F , we ļ¬rst need to deļ¬ne the tangent space to a manifold at a point. When M is a surface in R3 , we naturally think of the tangent plane at each point q ā M as a 2-dimensional subspace of the ambient space R3 . But if M is an abstract manifold and not a subset of some larger Euclidean space, then there may not be any natural choice of Euclidean space available for the tangent space to live in. In this case, we need a more self-contained notion for the tangent space. It turns out that the answer lies in the observation that tangent vectors act on functions via directional derivative: Suppose that x ā Rm and v ā Tx Rm . If Ī± : I ā Rm is any smooth curve in Rm with Ī±(0) = x, Ī± (0) = v and if f : Rm ā R is any diļ¬erentiable function, then v acts on f as follows: d v[f ] = f (Ī±(t)). dt t=0
In other words, v[f ] is the directional derivative of f at x in the direction of v. For an abstract manifold M , we simply turn this observation around and use it as a deļ¬nition for tangent vectors. Deļ¬nition 1.19. Let Ī± : I ā M be a smooth curve in M . The tangent vector Ī± (t0 ) to Ī± at the point Ī±(t0 ) ā M is the diļ¬erential operator Ī± (t0 ) : C ā (M ) ā R (where C ā (M ) denotes the space of all smooth, real-valued functions on M ) deļ¬ned by d Ī± (t0 )[f ] = f (Ī±(t)) dt t=t0
20
1. Assorted notions from diļ¬erential geometry
for f ā C ā (M ). The tangent space Tq M to M at a point q ā M is the set of all tangent vectors to all curves in M passing through q; i.e., Tq M = {Ī± (0) | Ī± : I ā M is a smooth curve with Ī±(0) = q}. Example 1.20. Let Ī£ be a regular surface in R3 and x : U ā Ī£ a parametrization of Ī£ deļ¬ned on some open set U ā R2 , with local coordinates (u, v) on U . The tangent vectors (xu , xv ) form a basis for the tangent plane Tq Ī£ at each point q ā x(U ). In order to see how these vectors act as diļ¬erential operators, let f : Ī£ ā R be a diļ¬erentiable, real-valued function on Ī£. Let q = x(u0 , v0 ) ā x(U ), and let Ī±(t) be the curve Ī±(t) = x(u0 + t, v0 ) in Ī£. Then Ī± (0) = xu , and so
d xu [f ] = f (Ī±(t)) dt t=0 d = f (x(u0 + t, v0 )) dt t=0 ā(f ā¦ x) = . āu (u,v)=(u0 ,v0 ) Similarly, xv [f ] = ā(fāvā¦x) . So, if we think of the parametrization (u,v)=(u0 ,v0 )
x as identifying its image x(U ) ā Ī£ with the open set U ā R2 and identifying a function f ā C ā (M ) withthe composition f ā¦x ā C ā (U ), then the partial ā ā derivative operators āu form a natural basis for the tangent space Tq Ī£ , āv at each point q ā x(U ). *Exercise 1.21. Let M be a manifold with local coordinates x = (x1 , . . ., xm ) on some open set V ā M . (The parametrization x : U ā Rm ā V ā M is implicit in this statement.) Let q ā V , and write q = x(x10 , . . . , xm 0 ). The tangent space Tq M can be given the structure of an m-dimensional vector space, with the diļ¬erential operators āxā 1 , . . . , āxām as a basis, almost exactly as in Example 1.20 (a) Let v = Ī± (0) ā Tq M , where Ī±(t) = x(x1 (t), . . . , xm (t)). Show that for any f ā C ā (M ), we have ā(f ā¦ x) i v[f ] = (x ) (0) . āxi (x1 ,...,xm )=(x1 ,...,xm ) 0
0
Therefore, any tangent vector v ā Tq M may be regarded as a linear combi ā ā nation of the diļ¬erential operators āx1 , . . . , āxm .
1.3. Diļ¬erentiable maps, tangent spaces, and vector ļ¬elds
21
ā 1 m ā R, then (b) Show that if ci āx i = 0 for some real numbers c , . . . , c c1 = Ā· Ā· Ā· = cm = 0. This shows that the diļ¬erential operators āxā 1 , . . . , āxām ā are linearly independent. (Hint: The statement āci āx i = 0ā means that
ā¦x) ci ā(f = 0 for every diļ¬erentiable function f : M ā R. Try to concoct āxi real-valued functions fi on V (which can then be extended to all of M via standard techniques) with the property that 1, i = j, ā(fi ā¦ x) = j āx 0, i = j.
*Exercise 1.22. (a) Suppose that we have a local coordinate transformation of the form xi = xi (ĀÆ x1 , . . . , x ĀÆm ),
i = 1, . . . , m,
on M . Use the deļ¬nition of tangent vectors and the multivariable chain rule to show that
ā ā ā ā = J, Ā· Ā· Ā· Ā· Ā· Ā· m m 1 1 āx ĀÆ āx āx ĀÆ āx where J is the Jacobian matrix of the coordinate transformation, i.e., the i matrix whose (i, j)th entry is āāxxĀÆj . (b) Write out the ith component of the vector equation in part (a) using the Einstein summation convention, and make sure that the indices on both sides of the equation match up correctly. The union of all the tangent spaces Tq M to a manifold M forms an object called the tangent bundle T M of M : TM = Tq M. qāM
If M has dimension m, then T M can be given the structure of a smooth manifold of dimension 2m. Before we consider the general case, recall how this works for surfaces in R3 : Example 1.23 (The tangent bundle of a surface). Let Ī£ ā R3 be a regular surface. At each point q ā Ī£, the tangent plane Tq Ī£ at q consists of all tangent vectors to Ī£ at q. The tangent bundle T Ī£ of Ī£ is simply the union of all these tangent planes: TĪ£ = Tq Ī£. qāĪ£
22
1. Assorted notions from diļ¬erential geometry
Note that the tangent bundle is characterized by (1) the base space Ī£ and (2) for each point q ā Ī£, a vector space Tq Ī£ associated to q, called the ļ¬ber at q. Moreover, the vector spaces associated to each point all have the same dimensionā2, in this case. T Ī£ is called the total space of the tangent bundle, and it can be given the structure of a manifold of dimension 4, as follows. If x = (x1 , x2 ) is a system of local coordinates on an open set V ā Ī£, it can be canonicallyextended to a system of local coordinates (x, y) = (x1 , x2 , y 1 , y 2 ) on T V = qāV Tq V ā T Ī£ by associating to any tangent vector v ā Tq V its unique representation as ā ā (1.7) v = y1 1 + y2 2 . āx āx *Exercise 1.24. Let T Ī£ be the tangent bundle of a regular surface, and let ĀÆ = (ĀÆ x = (x1 , x2 ) and x x1 , x ĀÆ2 ) be two overlapping systems of local coordinates on Ī£, related by a coordinate transformation of the form (1.8)
x1 = x1 (ĀÆ x1 , x ĀÆ2 ),
x2 = x2 (ĀÆ x1 , x ĀÆ2 ).
Show that the local coordinate transformation between the canonically assoĀÆ ) = (ĀÆ ciated local coordinates (x, y) = (x1 , x2 , y 1 , y 2 ) and (ĀÆ x, y x1 , x ĀÆ2 , yĀÆ1 , yĀÆ2 ) on T Ī£ is given by equations (1.8) together with the equations y1 yĀÆ1 = J , y2 yĀÆ2 where J is the Jacobian matrix of the transformation (1.8). (Hint: The result of Exercise 1.22 should be helpful here.) Tangent bundles for manifolds of arbitrary dimension work exactly the same way. If M is a manifold of dimension m and x : U ā Rm ā M is a parametrization of M , then the associated canonical parametrization of T M is the map (x, y) : U Ć Rm ā T M given by ā (x, y)(x1 , . . . , xm , y 1 , . . . , y m ) = y i i ā Tx(x1 ,...,xm ) M. āx In other words, for each point q ā x(U ), the tangent space Tq M isidentiļ¬ed with the vector space Rm by identifying the basis āxā 1 , . . ., āxām for Tq M with the standard basis for Rm . The same calculation as in Exercise 1.24 shows that the transition map between any two canonical parametrizations ĀÆ ) is diļ¬erentiable and, in fact, has the form (x, y) and (ĀÆ x, y xi = xi (ĀÆ x1 , . . . , x ĀÆm ),
yi =
āxi j yĀÆ ; āx ĀÆj
1.3. Diļ¬erentiable maps, tangent spaces, and vector ļ¬elds
23
therefore, T M is a smooth manifold of dimension 2m. T M also comes equipped with the base-point projection map Ļ : T M ā M deļ¬ned by the condition that for any v ā Tq M , Ļ(v) = q. Hereās the good news: Now that we know how to deļ¬ne tangent spaces for manifolds, the deļ¬nition of the derivative of a diļ¬erentiable map F : M ā N is exactly the same as before. Deļ¬nition 1.25. Let M be a manifold of dimension m, N a manifold of dimension n, and let F : M ā N be a diļ¬erentiable map. Fix a point q ā M . The derivative or diļ¬erential of F at q is the linear map dFq : Tq M ā TF (q) N deļ¬ned as follows: Given a tangent vector v ā Tq M , let Ī± : I ā M be a smooth curve with Ī±(0) = q and Ī± (0) = v. Then dFq (v) = (F ā¦ Ī±) (0) ā TF (q) N. If we happen to have local coordinates x = (x1 , . . . , xm ) on M and y = (y 1 , . . . , y n ) on N , then we can compute exactly as if M and N were Rm and Rn , respectively. The matrix representation for dFq will still look like the Jacobian matrix of F at q, provided that the natural coordinate bases ! ā ā ā ā , . . . , āxm and āy1 , . . . , āyn are used for the tangent spaces Tq M and āx1 TF (q) N , respectively. The main diļ¬erence is that the isomorphisms between Tq M, TF (q) N and Rm , Rn deļ¬ned by these coordinate bases are no longer canonical: Changing local coordinates changes the natural bases for the tangent spaces as well. This is the main reason why we need to know how tensors transform under changes of basis for the underlying vector spaces. *Exercise 1.26. Let F : M ā N be a diļ¬erentiable map. Let x : U ā Rm ā M,
y : V ā Rn ā N
be parametrizations, and let H = yā1 ā¦ F ā¦ x : U ā V. We can write H(x1 , . . . , xm ) = (y 1 (x1 , . . . , xm ), . . . , y n (x1 , . . . , xm )). (a) Suppose that q ā x(U ), and let v = aj āxā j ā Tq M . Show that " dFq (v) =
āy i a āxj j
#
ā ā TF (q) N ; āy i
24
1. Assorted notions from diļ¬erential geometry
in other words, ā ā dFq ā āxā 1
ā¤ā a1 ā¢ . ā„ā ā ā . āxm ā£ . ā¦ā = āy 1 ā”
Ā·Ā·Ā·
Ā·Ā·Ā·
ā āy n
am
ā¤ a1 ā¢ ā„ J ā£ ... ā¦ , ā”
am
where J is the n Ć m Jacobian matrix of H. BONUS: What happens to this formula if you perform local coordinate transformations on M and N ? (b) The diļ¬erential of F can be extended to a map dF : T M ā T N in the obvious way: For v ā Tq M , deļ¬ne dF (v) = dFq (v). Show that dF is a diļ¬erentiable map from T M to T N . *Exercise 1.27. Let M , N , and P be manifolds, and suppose that F : M ā N and G : N ā P are diļ¬erentiable maps. (a) Prove the chain rule: For q ā M , d(G ā¦ F )q = dGF (q) ā¦ dFq . (b) Give an explicit interpretation of the chain rule in terms of local coordinates and Jacobian matrices. (What are the sizes of each of the matrices involved, in terms of the dimensions m, n, p of M, N , and P ?) Write out the (i, j)th component of this matrix equation using the Einstein summation convention, and make sure that the indices on both sides of the equation match up correctly. We will often need to consider vector ļ¬elds on manifolds. Intuitively, a vector ļ¬eld on a manifold M is simply a choice of a single tangent vector in each tangent space Tq M , but we also need to know what it means for a vector ļ¬eld to be diļ¬erentiable. Fortunately, we now have all the tools that we need in order to make this idea precise: Deļ¬nition 1.28. A (smooth) vector ļ¬eld v on a manifold M is a diļ¬erentiable map v : M ā TM with the property that for any q ā M , we have v(q) ā Tq M. (Equivalently, the composition Ļ ā¦ v : M ā M is the identity map.)
1.3. Diļ¬erentiable maps, tangent spaces, and vector ļ¬elds
25
A vector ļ¬eld v on a manifold M may be regarded as a ļ¬rst-order, linear, homogeneous diļ¬erential operator v : C ā (M ) ā C ā (M ) deļ¬ned as follows: For f ā C ā (M ), the function v[f ] : M ā R is given by v[f ](q) = v(q)[f ]. This means that the value of v[f ] at q ā M is just the value of the tangent vector v(q) ā Tq M acting on f at q. In terms of local coordinates (x1 , . . . , xm ) on M , a vector ļ¬eld v is generally expressed as ā (1.9) v(x1 , . . . , xm ) = ai (x1 , . . . , xm ) i . āx Then for f ā C ā (M ), the function v[f ] is expressed in terms of these local coordinates as āf (1.10) v[f ](x1 , . . . , xm ) = ai (x1 , . . . , xm ) i (x1 , . . . , xm ). āx Conversely, any (smooth) ļ¬rst-order, linear, homogeneous diļ¬erential operator L : C ā (M ) ā C ā (M ) given in local coordinates by (1.10) deļ¬nes a vector ļ¬eld v on M described in local coordinates by (1.9). We close this section by deļ¬ning four important types of diļ¬erentiable maps between manifolds. Deļ¬nition 1.29. Let M be a manifold of dimension m, N a manifold of dimension n, and let F : M ā N be a diļ¬erentiable map. F is called (1) a diļ¬eomorphism if F is bijective and the inverse map F ā1 : N ā M is also diļ¬erentiable; if such a map F exists, then we say that M and N are diļ¬eomorphic; (2) an immersion if dFq : Tq M ā TF (q) N is injective at every point q ā M (note that this requires m ā¤ n); (3) an embedding if F is an injective immersion that is also a homeomorphism from M onto its image F (M ) ā N , where F (M ) is given the subspace topology inherited from the topology of N (note that this requires m ā¤ n); (4) a submersion if dFq : Tq M ā TF (q) N is surjective at every point q ā M (note that this requires m ā„ n). Remark 1.30. The deļ¬nition of an embedding looks rather technical, but in practice, an embedding is usually just an injective immersion. The image of an immersion may have self-intersections, but the restriction of an immersion to a suļ¬ciently small neighborhood of any point in its domain is always an embedding.
26
1. Assorted notions from diļ¬erential geometry
1.4. Lie groups and matrix groups We begin with the following deļ¬nition. Deļ¬nition 1.31. A Lie group is a set G that is both a group and a diļ¬erentiable manifold, with the additional property that the map Ī¼ : G Ć G ā G given by Ī¼(g, h) = ghā1 is diļ¬erentiable. This is a concise way of stating that group multiplication and group inverse are both diļ¬erentiable operations on G, as the following exercise shows. Exercise 1.32. Let G be a Lie group. Show that the multiplication map G Ć G ā G deļ¬ned by (g, h) ā gh and the inverse map G ā G deļ¬ned by g ā g ā1 are both diļ¬erentiable. Conversely, show that diļ¬erentiability of both of these maps implies that the map Ī¼ in Deļ¬nition 1.31 is diļ¬erentiable. (Hint: Consider the inverse map ļ¬rst.) To any Lie group G is associated a Lie algebra g. The Lie algebra g is simply the tangent space to G at the identity element e ā G; thus g is a vector space of the same dimension as the manifold G. There is also a product structure on g known as the Lie bracket. Deļ¬ning this structure will require a bit of eļ¬ort. For each element g ā G, the left translation or left multiplication map Lg : G ā G is deļ¬ned by Lg (h) = gh. (Similarly, the right translation map Rg : G ā G is deļ¬ned by Rg (h) = hg.) For any element h ā G, the diļ¬erential (dLg )h of Lg at h is a linear map from Th G to Tgh G. In particular, the diļ¬erential of Lg at the identity element e is a linear map from the Lie algebra g to Tg G. We can use this map to Ė on G deļ¬ned associate to any element v ā g a left-invariant vector ļ¬eld v at each point g ā G by (1.11)
Ė (g) = (dLg )e (v). v
1.4. Lie groups and matrix groups
27
This terminology is explained in the following exercise: *Exercise 1.33. Let G be a Lie group with Lie algebra g, let v ā g, and Ė be the vector ļ¬eld on G deļ¬ned by (1.11). let v Ė at gh ā G is given (a) Show that for any elements g, h ā G, the value of v by Ė (gh) = dLg (Ė v v(h)), where dLg : Th G ā Tgh G is the diļ¬erential of the left multiplication map Lg . This is why the vector Ė is called left-invariant: It is invariant under the action of left multiļ¬eld v plication by any element of G. (It is also true that any left-invariant vector Ė on a Lie group G is smooth; see, e.g., [Lee13].) ļ¬eld v (b) Conversely, suppose that a smooth vector ļ¬eld V on G has the property that for any g, h ā G, (1.12)
V(gh) = dLg (V(h)).
Ė , where v = V(e). Show that V = v Recall that a vector ļ¬eld on G is really a ļ¬rst-order, linear, homogeneous diļ¬erential operator on smooth functions f : G ā R. Given any two smooth vector ļ¬elds V, W on G (or indeed, any two smooth vector ļ¬elds on any manifold), the Lie bracket [V, W] is the smooth vector ļ¬eld deļ¬ned by the property that for any smooth function f : G ā R, [V, W][f ] = V[W[f ]] ā W[V[f ]]. Wait a minuteāa vector ļ¬eld is a ļ¬rst-order diļ¬erential operator, but that looks like a second-order diļ¬erential operator! The following exercise should allay your concerns: *Exercise 1.34. Let V, W be two vector ļ¬elds on a manifold M . There is no harm in working within a parametrization, so without loss of generality we may assume that V(x) = ai (x)
ā , āxi
W(x) = bj (x)
ā āxj
for some functions (ai (x), bj (x)). Show by direct computation that the Lie bracket [V, W] is a vector ļ¬eldāi.e., a ļ¬rst-order, linear, homogeneous differential operatorāon M . How are its coeļ¬cients related to (ai (x), bj (x))? (Hint: Compute the action of the operator [V, W] on a smooth function f : M ā R. And since āsmoothā means āinļ¬nitely diļ¬erentiableā, you can safely assume that mixed partial derivatives commute.)
28
1. Assorted notions from diļ¬erential geometry
Ė, w Ė to Finally, we use the Lie bracket of the left-invariant vector ļ¬elds v deļ¬ne the Lie bracket operation on the Lie algebra g: Deļ¬nition 1.35. Let G be a Lie group with associated Lie algebra g. For any two elements v, w ā g, the Lie bracket [v, w] of v and w is the unique element z ā g such that Ė =z Ė, [Ė v, w] where tildes denote the extensions to left-invariant vector ļ¬elds on G, as deļ¬ned above. In order for this deļ¬nition to make sense, we need to know that the Lie bracket of two left-invariant vector ļ¬elds on G is also left-invariant. Ė, w Ė be left-invariant vector ļ¬elds on a Lie group G. *Exercise 1.36. Let v Ė satisļ¬es the condition (1.12) and that it Show that the Lie bracket [Ė v, w] Ė for some z ā g. Conclude that the Lie bracket is a is therefore equal to z well-deļ¬ned operation on g. While this all sounds fairly abstract, the most common examples of Lie groupsāand the only ones that we will encounter in this bookāare subgroups of the group GL(n) of invertible n Ć n matrices. The diļ¬erentiable structure on GL(n) is simply that inherited by regarding GL(n) as an open 2 set in Rn . The Lie algebra of any such Lie group is a subspace of the vector space of n Ć n matrices. Remark 1.37. If it seems a little strange to think of a matrix as a vector, let G be any subgroup of GL(n), and let Ī±(t) be a curve in G with Ī±(0) = I. Then we can write Ī±(t) = g(t), where g(t) is an n Ć n matrix. The tangent vector to this curve at t = 0 is, of course, Ī± (0) = g (0), 2
which, while it could be thought of as a vector in Rn , is most naturally regarded as an n Ć n matrix. The following exercise shows that the Lie bracket on the Lie algebra of any subgroup of GL(n) is simply the matrix commutator: [A, B] = AB ā BA. *Exercise 1.38. Let G be a subgroup of GL(n). (a) Let g ā G, and let Lg : G ā G be the left multiplication map. Show that the diļ¬erential dLg : g ā Tg G is given explicitly by dLg (A) = g Ā· A,
A ā g,
1.4. Lie groups and matrix groups
29
where the right-hand side is just the matrix product of the two nĆn matrices g and A. Therefore, the vector ļ¬eld AĖ on G deļ¬ned by Ė A(g) =gĀ·A is left-invariant. (Hint: Consider a curve h(t) in G with h(0) = I, h (0) = A.) (b) Letās get explicit about exactly what diļ¬erential operator the vector ļ¬eld AĖ represents on G: If we use local coordinates (xij ) on the manifold of n Ć n matrices, then a general element g ā G is represented by a matrix [xij ]. If we write A = [aij ], then the tangent vector A ā g represents the diļ¬erential operator ā A = aij i , āxj and the matrix product AĖ = g Ā· A, whose entries are AĖij = xir arj , represents the diļ¬erential operator ā AĖ = xir arj i . āxj Now, let A = [aij ], B = [bkl ] ā g. Then we can write ā AĖ = xir arj i , āxj
Ė = x k bs ā . B s l āxkl
Use the formula that you computed in Exercise 1.34 (and some very careful index manipulation!) to show that ! ā Ė B] Ė = x i a r bl ā b r al [A, . r l j l j āxij (c) Conclude from part (b) that Ė B] Ė = g Ā· (AB ā BA) = (AB [A, ā BA) and that the Lie bracket on g is therefore just the matrix commutator: [A, B] = AB ā BA. The following examples describe some of the most commonly encountered Lie groups; we will be seeing all of these groups later on in the book. Example 1.39. The general linear group GL(n), consisting of all invertible n Ć n matrices. As a manifold, GL(n) is an open subset of the space MnĆn of all n Ć n matrices; speciļ¬cally, GL(n) = {g ā MnĆn | det(g) = 0}.
30
1. Assorted notions from diļ¬erential geometry
Thus, its Lie algebra gl(n) is the tangent space to the space of all n Ć n matrices at the identity In . This tangent space is naturally isomorphic to MnĆn . Example 1.40. The special linear group SL(n). SL(n) is deļ¬ned as SL(n) = {A ā GL(n) | det(A) = 1}. A standard computation using the implicit function theorem (which we omit here) shows that SL(n) is a manifold of dimension n2 ā 1. The Lie algebra sl(n) consists of all tangent vectors to curves in SL(n) passing through the identity In at t = 0. So, suppose that
A(t) = aij (t) is a curve of matrices in SL(n) with A(0) = In . Since A(t) ā SL(n) for all t, we have det(A(t)) = sgn(Ļ)a1Ļ(1) (t) Ā· Ā· Ā· anĻ(n) (t) = 1, ĻāSn
where the sum is over all permutations Ļ in the symmetric group Sn and sgn(Ļ) = Ā±1 is the sign of Ļ. In order to determine what conditions this imposes on the tangent vector A (0), we need to diļ¬erentiate this equation and evaluate the result at t = 0. Fortunately, this computation is not as bad as it looks: The fact that A(0) = In means that aij (0) = 0 unless i = j, in which case aij (0) = 1. So most of the terms in the derivative will drop out at t = 0, leaving the equation (aii ) (0) = 0. (Donāt forget the summation convention!) In other words, A (0) must have trace equal to zero. Exercise 1.41. Do this computation explicitly for SL(2): Let 1 a1 (t) a12 (t) A(t) = , a21 (t) a22 (t) with A(0) = I and det(A(t)) = a11 (t)a22 (t) ā a12 (t)a21 (t) = 1. Diļ¬erentiate this equation and set t = 0, and show that (a11 ) (0) + (a22 ) (0) = 0; i.e., tr(A (0)) = 0.
1.4. Lie groups and matrix groups
31
Aside from the trace condition, there are no other restrictions on A (0). This is most easily veriļ¬ed by comparing dimensions: SL(n) has dimension n2 ā 1, and the set of n Ć n matrices with trace equal to zero is a vector space of dimension n2 ā 1. Since sl(n) is contained in this vector space and has the same dimension as the entire space, it must be equal to the entire space. Therefore, sl(n) = {B ā MnĆn | tr(B) = 0}. Example 1.42. The orthogonal group O(n). O(n) is deļ¬ned as O(n) = {A ā GL(n) | tA A = In }, where tA denotes the transpose of the matrix A. A standard computation using the implicit function theorem (which we omit here) shows that O(n) is a manifold of dimension 12 n(n ā 1). In order to compute its Lie algebra o(n), suppose that A(t) is a curve of matrices in O(n) with A(0) = In . Since A(t) ā O(n) for all t, we have t
A(t) A(t) = In .
Diļ¬erentiating this equation and setting t = 0 yields t
A (0) + A (0) = 0;
in other words, A (0) must be a skew-symmetric matrix. Comparing dimensions as in the previous example shows that there are no other restrictions on A (0), and so o(n) = {B ā MnĆn | tB + B = 0}. Example 1.43. The special orthogonal group SO(n). SO(n) is deļ¬ned as SO(n) = {A ā GL(n) | tA A = In and det(A) = 1}. Clearly, SO(n) is a subgroup of O(n); it consists of those matrices in O(n) having determinant equal to 1. Now, it follows from the deļ¬ning equation for O(n) that every element of O(n) has determinant equal to either 1 or ā1. Moreover, the function det : O(n) ā R is continuous, with image equal to the set {1, ā1}. Therefore, the subsets SO(n) = {A ā O(n) | det(A) = 1}, O(n) \ SO(n) = {A ā O(n) | det(A) = ā1} are disconnected. In fact, O(n) consists of two connected components, one of which is SO(n) and the other of which is the coset of SO(n) consisting of those elements with determinant ā1. It follows that the Lie algebra so(n) is, in fact, equal to the Lie algebra o(n). This Lie algebra is usually denoted by so(n).
32
1. Assorted notions from diļ¬erential geometry
For a more thorough discussion of Lie groups and Lie algebras, see, e.g., [Lee13].
1.5. Vector bundles and principal bundles The tangent bundle T M of a manifold M is an example of a vector bundle. The formal deļ¬nition of a vector bundle is a bit complicated, but the general idea is much like this example. Deļ¬nition 1.44. A rank k vector bundle consists of a manifold B called the base space, a manifold E called the total space, and a diļ¬erentiable projection map Ļ : E ā B such that for each point q ā B, the inverse image Ļ ā1 (q) ā E is a vector space of dimension k. This vector space is called the ļ¬ber of E at q. Moreover, every point q ā B must have a neighborhood U ā B such that the set Ļ ā1 (U ) ā E is diļ¬eomorphic to U Ć Rk via a Ė ā U , maps the ļ¬ber Ļ ā1 (Ė diļ¬eomorphism that, for every q q) linearly onto k {Ė q} Ć R . Such a diļ¬eomorphism is called a local trivialization of the vector bundle. It is important to note that, while vector bundles always have local trivializations, they may not have global trivializations. That is to say, the local trivializations may patch together in such a way that their union becomes ātwistedā in a way that cannot be straightened out. So, a vector bundle Ļ : E ā B is not necessarily diļ¬eomorphic to B Ć Rk . For example, the tangent bundle of a compact surface has no global trivialization unless the surface has Euler characteristic equal to zero. Next, we introduce the notion of a section of a vector bundle. Deļ¬nition 1.45. A (smooth) section of a vector bundle Ļ : E ā B is a diļ¬erentiable map Ļ : B ā E with the property that the composition Ļ ā¦ Ļ : B ā B is the identity map on B. A section Ļ : B ā E is kind of like a vector-valued function on the base space B, but not quite. A function has a domain space and a range space; in particular, the function takes values in the same range space at each point of the domain. A section Ļ has a domain space B, but at each point q ā B, the value of Ļ(q) must lie in the ļ¬ber of E at q. Intuitively, this means that the range space is diļ¬erent at each point of the domain space. We may consider both local sections, whose domain consists of some open set in B, and global sections, whose domain is the entire space B. A section is often identiļ¬ed with its image in E, which is a submanifold of E consisting of exactly one point in each ļ¬ber. This may sound complicated, but again,
1.5. Vector bundles and principal bundles
33
you already know an example: A vector ļ¬eld on a surface Ī£ is a section of the tangent bundle T Ī£. Any vector bundle has a distinguished section called the zero section; this is the section that assigns to each point q ā B the zero vector in the ļ¬ber of E at q. Other than the zero section, there is no canonical way of choosing āconstantā sections unless an additional structure known as a connection is introduced. (And even then, it may only be possible to choose a section which is āconstantā along a curve, but not on any open set.) A connection allows sections of vector bundles to be diļ¬erentiated through a process called covariant diļ¬erentiation. The tangent bundle of a regular surface in R3 carries a canonical connection called the Levi-Civita connection, which is used to deļ¬ne parallelisms, geodesics, and so forth. (These concepts will be discussed in more detail in Chapters 11 and 12.) A section is called nonvanishing if its image does not intersect the image of the zero section. If a vector bundle does not have a global trivialization, then it may be that there do not exist any nonvanishing global sections. If this sounds surprising, then recall the PoincarĀ“e-Hopf theorem: Any smooth vector ļ¬eld on a compact surface with nonzero Euler characteristic must vanish at some point on the surface. This is equivalent to the statement that the tangent bundle has no nonvanishing global sections. As if vector bundles werenāt enough, we will also frequently encounter principal bundles. A principal bundle is similar to a vector bundle, in that it consists of a manifold B called the base space, a manifold P called the total space, and a diļ¬erentiable projection map Ļ : P ā B. The main diļ¬erence is that the ļ¬ber Ļ ā1 (q) at any point q ā B is a manifold diļ¬eomorphic to some Lie group G rather than to a vector space. However, this diļ¬eomorphism is generally not a group isomorphism in any canonical way. In particular, there is usually no well-deļ¬ned āidentity sectionā of P , and a principal bundle generally has no distinguished section similar to the zero section of a vector bundle. (In fact, it is possible for a principal bundle to have no global sections whatsoever.) Rather, each ļ¬ber Ļ ā1 (q) is a manifold on which G acts freely and transitively. A principal bundle Ļ : P ā B whose ļ¬ber at each point is diļ¬eomorphic to a Lie group G is often represented by the diagram
G
- P Ļ
?
B.
34
1. Assorted notions from diļ¬erential geometry
The arrow G ā P doesnāt really represent a map here; it just indicates that each ļ¬ber of P is diļ¬eomorphic to G and that G acts freely and transitively on the ļ¬bers of P . We will see some examples of principal bundles beginning in Chapter 3. In many cases, the total space P will itself be a Lie group, and the ļ¬bers will all be diļ¬eomorphic to some subgroup G of P . Exercise 1.46. Let Ī£ be the unit sphere S2 ā R3 . Let F (S2 ) denote the orthonormal frame bundle of S2 ; this is the set of all orthonormal bases (a.k.a. āframesā) e = (e1 , e2 ) for the tangent space Tq S2 at each point q ā S2 . Given any orthonormal frame e at a point q ā S2 , any other frame at q can be obtained from e by composition of a rotation and (possibly) a reļ¬ection, i.e., an element of the Lie group O(2). Conversely, every element of O(2) acts on e to produce a new orthonormal frame at q, and distinct elements of O(2) will produce diļ¬erent frames. Thus, the ļ¬ber of F (S2 ) over any point q ā S2 is diļ¬eomorphic to O(2), and F (S2 ) is a principal bundle over S2 with ļ¬ber group O(2). Prove that F (S2 ) has no global sections. (Hint: Use the PoincarĀ“e-Hopf theorem.)
10.1090/gsm/178/02
Chapter 2
Diļ¬erential forms
2.1. Introduction In calculus, you may have seen the diļ¬erential or exterior derivative df of a function f : R3 ā R deļ¬ned to be df =
āf āf āf dx + dy + dz. āx āy āz
The expression df is called a 1-form and you probably learned various ways of manipulating such thingsāfor instance, how to integrate a 1-form along a parametrized curve in R3 . But what is a 1-form, really? In this chapter, we will give formal deļ¬nitions for 1-forms and then more generally for p-forms with p ā„ 0 on Rn , and we will explore a bit about how they work. Once we understand how diļ¬erential forms behave in Rn , we will see how to deļ¬ne them more generally on manifolds. But before we launch into all that, it will be helpful to go into a bit more detail about certain types of tensors.
2.2. Dual spaces, the cotangent bundle, and tensor products Deļ¬nition 2.1. Let V be an n-dimensional vector space. The dual space V ā is the set of all linear mappings Ī± : V ā R. It should be clear that V ā is closed under addition and scalar multiplication, and so V ā is a vector space. The following exercise shows that V ā has 35
36
2. Diļ¬erential forms
dimension n: *Exercise 2.2. Let (e1 , . . . , en ) be a basis for V . Deļ¬ne linear functions eāi : V ā R, 1 ā¤ i ā¤ n, by the property that 1, i = j, āi i e (ej ) = Ī“j = 0, i = j. (Note that requiring eāi to be linear and specifying the value of eāi on all of the basis vectors ej completely determines the function eāi on V .) (a) Show that the functions (eā1 , . . . , eān ) are linearly independent. (Hint: The hypothesis āci eāi = 0ā means that for every v ā V , ci eāi (v) = 0.) (b) Show that the functions (eā1 , . . . , eān ) span V ā . (Hint: Let Ī± : V ā R be a linear mapping, and let ci = Ī±(ei ). Consider the function ci eāi : V ā R.) (c) Show that (V ā )ā ā¼ = V . (Hint: Associate to any vector v ā V the function ā Ė : V ā R deļ¬ned by v Ė (Ī±) = Ī±(v).) v The basis (eā1 , . . . , eān ) for V ā constructed in Exercise 2.2 is called the dual basis to the basis (e1 , . . . , en ) for V . Also, note that the isomorphism (V ā )ā ā¼ = V is canonical: It is completely independent of any choice of basis for V . Thus it is accurateāand commonāto regard (V ā )ā as being equal to V . The most common example of a dual space that we will encounter is the following: Example 2.3. Let M be a manifold of dimension m, let q ā M , and let V = Tq M . The dual space V ā = Tqā M is called the cotangent space to M at q. The cotangent bundle of M is the union of all these cotangent spaces: T āM = Tqā M. qāM
T āM
has the structure of a smooth manifold of dimension 2m: Given any parametrization x : U ā Rm ā M of M , there is an associated canonical parametrization (x, p) : U Ć Rm ā T ā M of T ā M , deļ¬ned in a manner analogous to the canonical parametrizations of the tangent bundle T M . ā Speciļ¬cally, the element Ī± = (x, p)(x1 , . . . , xm , p1 , . . . , pm ) ā Tx(x 1 ,...,xm ) M ā is deļ¬ned by the condition that for any v = y i āx i ā Tx(x1 ,...,xm ) M , we have
Ī±(v) = pi y i ā R. Remark 2.4. More generally, given a vector space V , the dual space V ā is sometimes referred to as the covector space of V , and the elements of V ā as
2.2. Dual spaces, the cotangent bundle, and tensor products
37
covectors. If the elements of V are represented by column vectors, then the elements of V ā are represented by row vectors, and vice versa. Most of the tensor ļ¬elds that commonly appear in geometry are sections of vector bundles that are constructed from the tangent and cotangent bundles of a manifold. The construction used to build more complicated bundles from these two is the tensor product. The oļ¬cial deļ¬nition starts by deļ¬ning tensor products for dual spaces: Deļ¬nition 2.5. Let V be a vector space of dimension m and W a vector space of dimension n. Let Ī± ā V ā , Ī² ā W ā . The tensor product of Ī± and Ī², denoted Ī± ā Ī², is the bilinear function Ī±āĪ² :V ĆW āR deļ¬ned by (Ī± ā Ī²)(v, w) = Ī±(v)Ī²(w). The tensor product of V ā and W ā , denoted V ā ā W ā , is the vector space consisting of all linear combinations of such tensor products; i.e., V ā ā W ā = span{Ī± ā Ī² | Ī± ā V ā , Ī² ā W ā }. *Exercise 2.6. (a) Show that Ī± ā Ī² is indeed a bilinear function on V Ć W , i.e., that for any vectors v, v1 , v2 ā V , w, w1 , w2 ā W , and real numbers a, b, (Ī± ā Ī²)(av1 + bv2 , w) = a(Ī± ā Ī²)(v1 , w) + b(Ī± ā Ī²)(v2 , w), (Ī± ā Ī²)(v, aw1 + bw2 ) = a(Ī± ā Ī²)(v, w1 ) + b(Ī± ā Ī²)(v, w2 ). (b) Show that V ā ā W ā is a vector space of dimension mn. (Hint: Let (e1 , . . . , em ) be a basis for V , with dual basis (eā1 , . . . , eām ) for V ā , and let (f1 , . . . , fn ) be a basis for W , with dual basis (f ā1 , . . . , f ān ) for W ā . Show that {eāi ā f āj | 1 ā¤ i ā¤ m, 1 ā¤ j ā¤ n} is a basis for V ā ā W ā .) So, how do we deļ¬ne the tensor product V ā W ? We simply take advantage of the canonical identiļ¬cation V = (V ā )ā that we proved in Exercise 2.2: Deļ¬nition 2.7. Let V be a vector space of dimension m and W a vector Ė ā (V ā )ā , space of dimension n. For any vectors v ā V, w ā W , deļ¬ne v Ė ā (W ā )ā by w Ė (Ī±) = Ī±(v), v
Ī± ā V ā,
Ė w(Ī²) = Ī²(w),
Ī² ā W ā.
38
2. Diļ¬erential forms
The tensor product of v and w, denoted v ā w, is given by Ėāw Ė ā (V ā )ā ā (W ā )ā . vāw =v The tensor product of V and W , denoted V ā W , is the vector space V ā W = (V ā )ā ā (W ā )ā . Of course, this deļ¬nition is much too convoluted to use in practice! For practical purposes, if (e1 , . . . , em ) is a basis for V and (f1 , . . . , fn ) is a basis for W , then V ā W is simply a vector space of dimension mn, with basis given by the formal symbols {ei ā fj | 1 ā¤ i ā¤ m, 1 ā¤ j ā¤ n}. Remark 2.8. The tensor product is associative; i.e., (v ā w) ā x = v ā (w ā x), and so it makes sense to write v ā w ā x. But it is not commutative; even when V = W , v ā w is, in general, not equal to w ā v. Deļ¬nition 2.9. A rank k tensor is an element of a tensor product of the form V1 ā Ā· Ā· Ā· ā Vk , where V1 , . . . , Vk are vector spaces, some or all of which may be dual spaces. In particular, a rank 1 tensor is simply an element of a vector space V (or V ā ). Example 2.10 (Cf. Example 1.12). Let V be an m-dimensional vector space with basis (e1 , . . . , em ) and dual basis (eā1 , . . . , eām ) for V ā , and let W be an n-dimensional vector space with basis (f1 , . . . , fn ) and dual basis (f ā1 , . . . , f ān ) for W ā . Let T : V ā W be a linear transformation, represented in terms of the given bases for V and W by the n Ć m matrix ā¤ ā” 1 c1 . . . c1m ā¢ .. ā„ . AT = ā£ ... . ā¦ n n c1 . . . cm Then T is a rank 2 tensor; it is an element of the vector space W ā V ā , and it can be written in terms of the given bases as T = cij fi ā eāj . Two important categories of tensors are the symmetric and skew-symmetric tensors of a vector space V .
2.2. Dual spaces, the cotangent bundle, and tensor products
39
Deļ¬nition 2.11. Let V be an n-dimensional vector space, and let V ā2 = V ā V . (Similarly, let V āk denote the tensor product of k copies of V .) The space of symmetric 2-tensors of V is the subspace S 2 V of V ā2 deļ¬ned by S 2 V = span{v ā v | v ā V }. The space of skew-symmetric 2-tensors of V is the subspace Ī2 V of V ā2 deļ¬ned by Ī2 V = span{v ā w ā w ā v | v, w ā V }. Exercise 2.12. Show that for any vectors v, w ā V , v ā w + w ā v ā S 2 V. (Hint: Consider (v + w) ā (v + w).) Deļ¬nition 2.13. The symmetric product v ā¦ w and wedge product v ā§ w are deļ¬ned by v ā¦ w = 12 (v ā w + w ā v), v ā§ w = v ā w ā w ā v. (Note that some authors insert a factor of 12 into the deļ¬nition of the wedge product as well.) More generally, if v1 , . . . , vk ā V , then 1 v1 ā¦ Ā· Ā· Ā· ā¦ vk = v ā Ā· Ā· Ā· ā vĻ(k) , k! Ļ Ļ(1) v1 ā§ Ā· Ā· Ā· ā§ vk = sgn(Ļ)vĻ(1) ā Ā· Ā· Ā· ā vĻ(k) , Ļ
where the sum is over all permutations Ļ in the symmetric group Sk and sgn(Ļ) = Ā±1 is the sign of Ļ. The spaces of symmetric and skew-symmetric k-tensors of V are the subspaces S k V , Īk V of V āk deļ¬ned by S k V = span{v1 ā¦ Ā· Ā· Ā· ā¦ vk | v1 , . . . , vk ā V }, Īk V = span{v1 ā§ Ā· Ā· Ā· ā§ vk | v1 , . . . , vk ā V }. Exercise 2.14. (a) Show that V ā V = S 2 V ā Ī2 V ; i.e., every element of V ā V can be uniquely expressed as the sum of a symmetric product and a skew-symmetric product. (b) Show that the analogous decomposition does not hold for V ā3 . (Hint: Compute the dimensions of V ā3 , S 3 V , and Ī3 V in terms of the dimension n of V .) All these tensor products can be used to form tensor bundles on a manifold M . A tensor bundle on M is simply a vector bundle on M where each ļ¬ber is isomorphic to a ļ¬xed tensor product of vector spaces. In the most common examples, the ļ¬ber over each point q ā M is a tensor product of
40
2. Diļ¬erential forms
some numbers of copies of the tangent space Tq M and the cotangent space Tqā M , or some natural subspace of such a tensor product. Finally, this allows us to give a formal deļ¬nition of tensor ļ¬elds: Deļ¬nition 2.15. A rank k tensor ļ¬eld on a manifold M is a section of a vector bundle Ļ : E ā M , where the ļ¬ber Eq of E over each point q ā M has the form Eq = V1 ā Ā· Ā· Ā· ā Vk , where V1 , . . . , Vk are vector spaces. (In most cases, each of the vector spaces V1 , . . . , Vk is equal to either Tq M or Tqā M .) Remark 2.16. In practice, tensor ļ¬elds on a manifold M are almost always referred to as ātensors on M ā. For example, a Riemannian manifold M has a āmetric tensorā g and a āRiemann curvature tensorā R. Strictly speaking, these are both tensor ļ¬elds on M , but you will probably never encounter the phrase āRiemann curvature tensor ļ¬eldā. Example 2.17 (Cf. Exercise 1.15). Let M be a manifold of dimension n, and let S 2 (T ā M ) = S 2 (Tqā M ). qāM
S 2 (T ā M )
is a subbundle of the tensor bundle T ā M ā T ā M , and it is a bundle of symmetric 2-tensors on M . (Note that I didnāt say ātheā bundle of symmetric 2-tensors on M , because there are other bundles that would satisfy the same description, e.g., S 2 (T M ).) A metric g on M is a section of S 2 (T ā M ) because, at every point q ā M , g deļ¬nes a symmetric, bilinear function gq : Tq M Ć Tq M ā R. Now weāre ready to start talking about diļ¬erential forms!
2.3. 1-forms on Rn Deļ¬nition 2.18. A (smooth) 1-form Ļ on Rn is a (smooth) section of the cotangent bundle T ā Rn . Equivalently, Ļ : T Rn ā R is a real-valued function on the set of all tangent vectors to Rn , with the properties that: (1) For each x ā Rn , the restriction Ļx : Tx Rn ā R of Ļ to Tx Rn is a linear map.
2.4. p-forms on Rn
41
(2) For any smooth vector ļ¬eld v on Rn , the function Ļ(v) : Rn ā R is smooth. In particular, the 1-forms (dx1 , . . . , dxn ) are deļ¬ned by the property that for any vector v = t[v 1 , . . . , v n ] ā Tx Rn , (2.1)
dxi (v) = v i .
*Exercise 2.19 (Cf. Exercise 2.2). In this exercise, we will show directly that for any point x ā Rn , the 1-forms (dx1 , . . . , dxn ) deļ¬ned by (2.1) form a basis for the cotangent space Txā Rn . (a) In order to show that the 1-forms (dx1 , . . . , dxn ) are linearly independent, suppose that ci dxi = 0. This means that ci dxi (v) = 0 for every vector v ā Tx Rn . Show that this implies that c1 = Ā· Ā· Ā· = cn = 0. (b) In order to show that the 1-forms (dx1 , . . . , dxn ) span Txā Rn , let Ļx ā Txā Rn be arbitrary. Let (e1 , . . . , en ) be the standard basis for Tx Rn , and let ci = Ļx (ei ),
i = 1, . . . , n.
Show that Ļx = ci dxi . (It suļ¬ces to show that Ļx (v) = ci dxi (v) for every v ā Tx Rn .) It follows from Exercise 2.19 that the 1-forms (dx1 , . . . , dxn ) form a basis for the 1-forms on Rn as a module over C ā (Rn ); this means that any 1-form Ļ on Rn can be expressed uniquely as (2.2)
Ļ = fi (x) dxi
for some smooth, real-valued functions f1 , . . . , fn : Rn ā R. If a vector ļ¬eld v on Rn has the form v(x) = t[v 1 (x), . . . , v n (x)], then we have Ļ(v(x)) = fi (x) v i (x).
2.4. p-forms on Rn The 1-forms on Rn are the building blocks of an algebra, called the algebra of diļ¬erential forms on Rn . The multiplication in this algebra is the wedge product of Deļ¬nition 2.13. The wedge product is multilinear and skew-
42
2. Diļ¬erential forms
symmetric; i.e., for any two 1-forms Ļ, Ļ, we have Ļ ā§ Ļ = āĻ ā§ Ļ. In particular, Ļ ā§ Ļ = 0 for any 1-form Ļ. If each summand of a diļ¬erential form Ī¦ is a wedge product of p 1-forms, then the form is called a p-form. Scalar-valued functions are considered to be 0-forms, and any form on Rn of degree p > n must be zero due to the skew-symmetry. Exercise 2.20. Prove this last statement. (Hint: Let Ļ1 , . . . , Ļp be 1-forms. According to (2.2), we can write Ļj = fji (x) dxi ,
j = 1, . . . , p.
Show by direct computation that if p > n, then Ļ1 ā§ Ā· Ā· Ā· ā§ Ļp = 0.) More formally, we have the following deļ¬nition: Deļ¬nition 2.21. A p-form on Rn is a section of the tensor bundle Īp (T ā Rn ). The algebra of diļ¬erential forms on Rn consists of all sections of the bundle $ $ Īp (T ā Rn ) = Īp (Txā Rn ), xāRn pā„0
pā„0
with multiplication given by the wedge product ā§ : Īp (T ā Rn ) Ć Īq (T ā Rn ) ā Īp+q (T ā Rn ). The space of p-forms on Rn is generally denoted by Ī©p (Rn ), and the algebra of all diļ¬erential forms on Rn is denoted by $ Ī©ā (Rn ) = Ī©p (Rn ). pā„0
A basis for the p-forms on Rn (as a module over C ā (Rn )) is given by the set {dxi1 ā§ Ā· Ā· Ā· ā§ dxip | 1 ā¤ i1 < i2 < Ā· Ā· Ā· < ip ā¤ n}; this simply means that any p-form Ī¦ on Rn can be expressed uniquely as Ī¦= fI (x) dxi1 ā§ Ā· Ā· Ā· ā§ dxip , |I|=p
where I ranges over all increasing multi-indices I = (i1 , . . . , ip ) of length p and the coeļ¬cients fI (x) represent smooth functions fI : Rn ā R.
2.5. The exterior derivative
43
Just as 1-forms act on vector ļ¬elds to give real-valued functions, so p-forms act on p-tuples of vector ļ¬elds to give real-valued functions. For instance, if Ļ, Ļ are 1-forms and v, w are vector ļ¬elds, then (Ļ ā§ Ļ)(v, w) = Ļ(v)Ļ(w) ā Ļ(w)Ļ(v). More generally, if Ļ1 , . . . , Ļp are 1-forms and v1 , . . . , vp are vector ļ¬elds, then (Ļ1 ā§ Ā· Ā· Ā· ā§ Ļp )(v1 , . . . , vp ) = sgn(Ļ) Ļ1 (vĻ(1) ) Ļ2 (vĻ(2) ) Ā· Ā· Ā· Ļn (vĻ(n) ), ĻāSp
where the sum is over all permutations Ļ in the symmetric group Sp and sgn(Ļ) = Ā±1 is the sign of Ļ. Exercise 2.22. Show that the wedge product is related to the determinant as follows: If Ļ1 , . . . , Ļp are 1-forms and v1 , . . . , vp are vector ļ¬elds, then ā” ā¤ Ļ1 (v1 ) Ā· Ā· Ā· Ļ1 (vp ) ā¢ ā„ ā¢ .. ā„ . (Ļ1 ā§ Ā· Ā· Ā· ā§ Ļp )(v1 , . . . , vp ) = det ā¢ ... . ā„ ā£ ā¦ Ļp (v1 ) Ā· Ā· Ā·
Ļp (vp )
2.5. The exterior derivative The exterior derivative is an operator that takes p-forms to (p + 1)-forms. We will deļ¬ne it ļ¬rst for functions and then extend this deļ¬nition to higher degree forms. Deļ¬nition 2.23. If f : Rn ā R is diļ¬erentiable, then the exterior derivative of f is the 1-form df with the property that for any x ā Rn , v ā Tx Rn , dfx (v) = v[f ]; i.e., dfx (v) is the directional derivative of f at x in the direction of v. *Exercise 2.24. Show that dxi as deļ¬ned by equation (2.1) really is the exterior derivative of the ith coordinate function xi on Rn . (So the notation is consistent!) It is not diļ¬cult to show that, as one might expect, āf i dx . āxi The exterior derivative also obeys the Leibniz rule (2.3)
(2.4)
df =
d(f g) = g df + f dg
44
2. Diļ¬erential forms
for functions f, g : Rn ā R and the chain rule d(h ā¦ f ) = (h ā¦ f ) df
(2.5)
for functions f : Rn ā R, h : R ā R. *Exercise 2.25. Verify equations (2.3), (2.4), and (2.5). We extend the deļ¬nition of the exterior derivative to p-forms on Rn as follows: Deļ¬nition 2.26. Given a p-form Ī¦ = fI (x) dxi1 ā§ Ā· Ā· Ā· ā§ dxip on Rn , the |I|=p
exterior derivative dĪ¦ of Ī¦ is the (p + 1)-form (2.6) dĪ¦ = dfI ā§ dxi1 ā§ Ā· Ā· Ā· ā§ dxip . |I|=p
If Ī¦ is a p-form and ĪØ is a q-form, then the Leibniz rule takes the form (2.7)
d(Ī¦ ā§ ĪØ) = dĪ¦ ā§ ĪØ + (ā1)p Ī¦ ā§ dĪØ.
*Exercise 2.27. Prove the Leibniz rule (2.7) in the case p = q = 1: If Ļ, Ļ are 1-forms on Rn , then d(Ļ ā§ Ļ) = dĻ ā§ Ļ ā Ļ ā§ dĻ. The following theorem is possibly the most important one in this entire book! Theorem 2.28. d ā¦ d = 0; i.e., for any diļ¬erential form Ī¦ on Rn , d(dĪ¦) = 0. Proof. First, suppose that f is a function (i.e., a 0-form). Then " # āf i d(df ) = d dx āxi n ā 2f = dxj ā§ dxi āxi āxj i,j=1 # " ā 2f ā2f dxi ā§ dxj = ā āxj āxi āxi āxj i Form(omega=1); declares the object omega to be a 1-form. You can declare multiple forms with a single command; e.g., the command > Form(omega=1, theta=1, Omega=2);
60
2. Diļ¬erential forms
declares omega and theta to be 1-forms and Omega to be a 2-form. You can declare functions (i.e., 0-forms); e.g., > Form(f=0); but it isnāt necessary; anything which hasnāt been declared with a Form command is automatically assumed to be a 0-form. You can also declare constants by assigning them to have degree ā1 (this is purely a Maple convention and has nothing to do with their degree as diļ¬erential forms!); e.g., the command > Form(a=-1, b=-1); declares a and b to be constants. Finally, if you want to declare an object to be a 1-form (certainly the most common use of this command), the ā=1ā can be omitted; e.g., the command > Form(omega, theta, eta); declares the objects omega, theta, and eta to be 1-forms. The &Ė command. This is the command for wedge product. For instance, the commands > omega:= x*d(y); > theta:= y*d(z); > omega &Ė theta; can be used to compute the wedge product of the 1-forms x dy and y dz. Maple will choose a default ordering of forms for wedge products; e.g., the command > d(y) &Ė d(x); returns ā(d(x) &Ė d(y)) If you donāt like the order that Maple chooses, you can change it with the Forder command; see Mapleās help page for the Cartan package for more details. Furthermore, if you donāt like this symbol for the wedge product, you can change it with the WedgeProduct command. The d command. This is the all-purpose exterior derivative command. (WARNING: When youāre using the Cartan package, donāt use the letter d as a variable or Maple will get hopelessly confused!) It can be used both for computation and for assignment, and it knows how to use the chain rule
2.12. Introduction to the Cartan package for Maple
61
for functions. For example, the command > d(f(x,y,z)); returns "
# " # " # ā ā ā f (x, y, z) d(x) + f (x, y, z) d(y) + f (x, y, z) d(z) āx āy āz
If you like, you can clean this up a little bit by using the PDETools[declare] command to tell Maple that f is a function of the variables (x, y, z): > PDETools[declare](f(x,y,z)); > d(f(x,y,z)); returns fx d(x) + fy d(y) + fz d(z) (Unfortunately, you still have to type out f(x,y,z) wherever it appears in the input.) This works on forms of any degree; e.g., > d(xĖ2*d(y) + y*d(z)); returns 2x(d(x) &Ė d(y)) + (d(y) &Ė d(z)) Note that for coordinate 1-forms, you must type, e.g., d(x) rather than dx; Maple will regard dx as a completely diļ¬erent variable having nothing to do with the variable x or its exterior derivative. An important feature of the Cartan package is that exterior derivatives can be assigned as well as computed, and no explicit local coordinates are required. (The value of this feature for the method of moving frames will become apparent in later chapters!) For instance, the commands > Form(omega, theta, eta); > d(omega):= theta &Ė eta; declare that omega, theta, and eta are 1-forms and that the exterior derivative of omega is the 2-form (theta &Ė eta). The Simf command. This is the all-purpose simpliļ¬cation command for diļ¬erential forms, and you should use it liberally. (Maple fails to make some fairly obvious simpliļ¬cations without it.) For instance, the command > d(xĖ2*d(y) + yĖ2*d(x)); returns ā2y (d(x) &Ė d(y)) + 2x (d(x) &Ė d(y))
62
2. Diļ¬erential forms
and then the command > Simf(%); returns (ā2y + 2x)(d(x) &Ė d(y)) (The % operator refers to the output of the previous command.) Of course, these two commands can be combined into the single command > Simf(d(xĖ2*d(y) + yĖ2*d(x))); The pick command. The command > pick(bigform, omega); where omega is a 1-form, ļ¬nds all the terms in the expression bigform that involve wedge products with omega. It then writes bigform as bigform = form1 &Ė omega + form2 where form2 contains no terms involving omega and returns form1. For instance, the command > pick(x*d(y) &Ė d(z) + y*d(z) &Ė d(x) + z*d(x) &Ė d(y), d(x)); returns āz dy + y dz This command can also be invoked with additional 1-forms as arguments in order to ļ¬nd all the terms in bigform that involve speciļ¬c wedge products of higher degree. So, for example, the command > pick(bigform, omega1, omega2); where omega1, omega2 are 1-forms, ļ¬nds all the terms in the expression bigform that involve wedge products with omega1 &Ė omega2. (Note that this command is equivalent to > pick(pick(bigform, omega2), omega1); and pay attention to the order of the 1-forms in both versions!) For instance, the command > pick(x*d(y) &Ė d(z) + y*d(z) &Ė d(x) + z*d(x) &Ė d(y), d(x), d(y)); simply returns the coeļ¬cient z.
2.12. Introduction to the Cartan package for Maple
63
The ScalarForm command. This command takes the scalar coeļ¬cient of each summand in a diļ¬erential form and returns them all as a list. For instance, the command > ScalarForm(x*d(y) &Ė d(z) + y*d(z) &Ė d(x) + z*d(x) &Ė d(y)); returns [z, y, āx] (The signs and the order of the elements in this list may vary, depending on what order Maple has decided to assign to all the 1-forms involved.) This command is particularly useful when you have a diļ¬erential form that you want to set equal to zero (often something that you computed as d ā¦ d of something else), and you therefore want to set all its coeļ¬cients equal to zero. The ScalarForm command has an optional second argument, which must be a string. If this argument is present, then the command assigns to it a list of the decomposable forms in each summand, in order corresponding to the order of the coeļ¬cients of these forms produced by the main command. For instance, the command > ScalarForm(x*d(y) &Ė d(z) + y*d(z) &Ė d(x) + z*d(x) &Ė d(y), ātermsā); returns [z, y, āx] as before, but now if you type > terms; it returns [(d(x) &Ė d(y)), (d(z) &Ė d(x)), (d(z) &Ė d(y))] Again, the order may vary, but the order of the coeļ¬cients in the ļ¬rst line will match the order of the forms in the second line. The makebacksub command. It often happens that we have two different bases for the 1-forms on a manifold, and itās handy to be able to go back and forth between them. We usually do this using the subs command to make substitutions. (And you should pretty much always follow up a subs command with a Simf command.) For instance, suppose that we are working on R5 with coordinates (x, y, z, p, q) and we want to use the basis Īø = dz ā p dx ā q dy, Ļ1 = dp ā ez dy,
Ļ 1 = dx,
Ļ 2 = dy,
Ļ2 = dq ā ez dx
64
2. Diļ¬erential forms
for the 1-forms on R5 . (Donāt worry about why you might want to do such a thing; it has to do with an exterior diļ¬erential system representing the partial diļ¬erential equation zxy = ez , but thatās well beyond the scope of this book!) You could simply make assignments such as > theta:= d(z) - p*d(x) - q*d(y); etc., but then everything that follows would be expressed in terms of the coordinate basis. If you want to do something like computing dĪø and expressing it in terms of the basis (Īø, Ļ 1 , Ļ 2 , Ļ1 , Ļ2 ), itās more eļ¬ective to set up substitutions to go back and forth between the two bases. You can do this as follows. First, you need to declare the 1-forms in your new basis: > Form(theta, omega1, omega2, pi1, pi2); (You donāt have to declare the coordinate basis; the functions x, y, z, p, q are automatically assumed to be 0-forms, so their exterior derivatives are 1-forms.) Then deļ¬ne the substitution: > sub1:= [theta = d(z) - p*d(x) - q*d(y), omega1 = d(x), omega2 = d(y), pi1 = d(p) - exp(z)*d(y), pi2 = d(q) - exp(z)*d(x)]; You can now use the substitution sub1 to go from the basis (Īø, Ļ 1 , Ļ 2 , Ļ1 , Ļ2 ) to the coordinate basis; e.g., the command > Simf(subs(sub1, theta)); yields d(z) ā p d(x) ā q d(y) Where the makebacksub command comes in is when you want to go the other way. The command > backsub1:= makebacksub(sub1); produces a substitution backsub1 that is the inverse of the substitution sub1; thus, it will go from the coordinate basis to the basis (Īø, Ļ 1 , Ļ 2 , Ļ1 , Ļ2 ). So, e.g., the command > Simf(subs(backsub1, d(z))); yields Īø + p Ļ1 + q Ļ2 One warning about the makebacksub command: It only works properly when the substitution sub1 is a complete list of the elements of one basis of
2.12. Introduction to the Cartan package for Maple
65
1-forms expressed in terms of another basis. So, for instance, if you deļ¬ne > sub2:= [theta = d(z) - p*d(x) - q*d(y), pi1 = d(p) - exp(z)*d(y), pi2 = d(q) - exp(z)*d(x)]; without including omega1 and omega2 in the list, the command > makebacksub(sub2); returns a less than helpful backwards substitution, and it might not even do it consistently if you execute it more than once. (Try it and see what happens!) You can use these two substitutions to express dĪø in terms of the basis (Īø, Ļ 1 , Ļ 2 , Ļ1 , Ļ2 ) via the following sequence of commands: > Simf(subs(sub1, theta)); d(z) ā p d(x) ā q d(y) > Simf(d(%)); (d(x)) &Ė (d(p)) + (d(y)) &Ė (d(q)) > Simf(subs(backsub1, %)); Ļ1 &Ė Ļ1 + Ļ2 &Ė Ļ2 And, of course, these commands can be combined into a single command: > Simf(subs(backsub1, Simf(d(Simf(subs(sub1, theta)))))); (Itās probably not really necessary to put all of those Simf commands in there, but it doesnāt hurt, and sometimes it prevents problems.) Having done all this, you can computeāand then assignāthe exterior derivatives of the new basis in terms of this basis, via the commands > d(theta):= Simf(subs(backsub1, Simf(d(Simf(subs(sub1, theta)))))); > d(omega1):= Simf(subs(backsub1, Simf(d(Simf(subs(sub1, omega1)))))); > d(omega2):= Simf(subs(backsub1, Simf(d(Simf(subs(sub1, omega2)))))); > d(pi1):= Simf(subs(backsub1, Simf(d(Simf(subs(sub1, pi1)))))); > d(pi2):= Simf(subs(backsub1, Simf(d(Simf(subs(sub1, pi2))))));
66
2. Diļ¬erential forms
Exercise 2.57. Use commands such as > Simf(d(d(theta))); to check that all the 1-forms in this basis, with their assigned exterior derivatives, satisfy the identity dā¦d = 0. What goes wrong with d(dĻ1 ) and d(dĻ2 )? Why does this happen, and what can you do to ļ¬x it?
Part 2
Curves and surfaces in homogeneous spaces via the method of moving frames
10.1090/gsm/178/03
Chapter 3
Homogeneous spaces
3.1. Introduction In the late nineteenth century, there were several diļ¬erent types of geometry under investigation: There was the classical Euclidean geometry with its standard notions of lengths and angles, various non-Euclidean geometries in which the parallel postulate was replaced by alternative versions and lengths were measured diļ¬erently than in Euclidean geometry, and even aļ¬ne and projective geometries, where lengths and angles werenāt well-deļ¬ned notions. In 1872, Felix Klein published a revolutionary treatise on geometry ([Kle93b]; an English translation is available in [Kle93a]), in which he proposed that the most useful way to study a geometric structure is to study its group of symmetries, i.e., the group of transformations that preserve the key features of the structure. This approach revolutionized the study of geometry, and it continues to inļ¬uence the development of the subject today. For instance, when studying curves in R3 (with the standard Euclidean metric on R3 ), you probably learned the following theorem: Theorem 3.1 (Fundamental Theorem of Space Curves). Given any smooth functions Īŗ(s), Ļ (s) on an interval I ā R with Īŗ(s) > 0, there exists a smooth, unit-speed curve Ī± : I ā R3 with curvature Īŗ(s) and torsion Ļ (s). Moreover, Ī± is unique up to rigid motion: Any other such curve Ī² diļ¬ers from Ī± by a translation and rotation in R3 . The rigid motionsātranslations and rotationsāare the symmetries of the Euclidean space R3 : They are exactly the transformations of R3 that preserve the Euclidean metric. Curvature and torsion are the invariants of 69
70
3. Homogeneous spaces
smooth curves in R3 : They are exactly the properties of smooth curves that remain unchanged when a curve is transformed by a rigid motion. Remark 3.2. Technically, translations and rotations are the orientationpreserving symmetries of Euclidean space. Reļ¬ections are also symmetries of Euclidean space, but they reverse orientation. (They also reverse certain other quantities, such as the sign of the torsion of a curve.) It is generally advantageous to restrict consideration to orientation-preserving symmetries, mainly because doing so typically allows us to work with connected Lie groups. (Think SO(n) vs. O(n).) Fundamental to Kleinās approachāand to the remainder of this bookāis the notion of a homogeneous space. We will look at curves, surfaces, etc., as submanifolds of homogeneous spaces, and our primary tool for studying such submanifolds will be the method of moving frames, which was introĀ“ Cartan in 1935 ([Car35]). In this chapter, we will start with duced by Elie a detailed discussion of Euclidean space as a homogeneous space; we will then give some general deļ¬nitions and explore several other homogeneous spaces (Minkowski space, equi-aļ¬ne space, and projective space) that are commonly studied in geometry. In subsequent chapters, we will develop the theory of curves and surfaces in each of these spaces.
3.2. Euclidean space 3.2.1. Inner products. Deļ¬nition 3.3. An inner product on the vector space Rn is a function Ā·, Ā· : Rn Ć Rn ā R with the following properties: (1) Symmetry: For any vectors v, w ā Rn , v, w = w, v. (2) Bilinearity: For any vectors v, w, z ā Rn and any scalars a, b ā R, av + bw, z = av, z + bw, z and z, av + bw = az, v + bz, w. (3) Positive deļ¬niteness: For any vector v ā Rn , v, v ā„ 0, with equality if and only if v = 0. Deļ¬nition 3.4. The vector space Rn endowed with an inner product Ā·, Ā· is called Euclidean space and is denoted En .
3.2. Euclidean space
71
An inner product provides a means for measuring lengths of vectors and angles between vectors in Euclidean space: Given vectors v, w ā En , the length of v is
|v| = v, v, and the angle between v and w is the angle Īø, 0 ā¤ Īø ā¤ Ļ, satisfying v, w cos(Īø) = . |v||w| Exercise 3.5. Let (e1 , . . . , en ) be any basis for Rn , and let Ā·, Ā· be any inner product on Rn . Deļ¬ne constants {gij , 1 ā¤ i, j, ā¤ n} by gij = ei , ej . (a) Show that for any vectors v = ai ei , w = bj ej , v, w = gij ai bj . (b) (Cf. Exercise 1.15) Let Ag be the matrix ā¤ ā” g11 . . . g1n ā¢ .. ā„ , Ag = ā£ ... . ā¦ gn1 . . . gnn and let v, w be represented by the column vectors ā” 1ā¤ ā” 1ā¤ a b ā¢ .. ā„ ā¢ .. ā„ a = ā£ . ā¦, b = ā£ . ā¦, n a bn respectively. Show that v, w = taAg b, where we identify the 1 Ć 1 matrix on the right-hand side with its single real-valued entry. Deļ¬nition 3.4 makes it sound as though there might be many diļ¬erent Euclidean spaces corresponding to diļ¬erent inner products, but in fact they are all equivalent, as the following exercise shows. Exercise 3.6. Let Ā·, Ā· be any inner product on Rn , and construct a basis (e1 , . . . , en ) for Rn inductively as follows: (Initial step) Choose any nonzero vector v1 ā Rn , and set v1 e1 =
. v1 , v1 Let (e1 )ā„ denote the orthogonal complement of e1 with respect to Ā·, Ā·; i.e., (e1 )ā„ = {v ā Rn | v, e1 = 0}.
72
3. Homogeneous spaces
(a) Show that (e1 )ā„ ā Rn is a vector space of dimension n ā 1. (Hint: Consider the linear map f : Rn ā R deļ¬ned by f (v) = v, e1 , and compute the dimension of ker(f ).) (Inductive step) Suppose that e1 , . . . , ek have been chosen. Let (e1 , . . . , ek )ā„ = {v ā Rn | v, e1 = Ā· Ā· Ā· = v, ek = 0}. Let vk+1 be any nonzero vector in (e1 , . . . , ek )ā„ , and set vk+1 ek+1 =
. vk+1 , vk+1 (b) Show that ek+1 is linearly independent from (e1 , . . . , ek ). (c) Show that (e1 , . . . , ek )ā„ ā Rn is a vector space of dimension nāk. (Hint: Consider the linear map F : Rn ā Rk deļ¬ned by F (v) = t [v, e1 , . . . , v, ek ] , and compute the dimension of ker(F ). Note that, by part (b), the vectors (e1 , . . . , ek ) are linearly independent.) (d) By parts (b) and (c), this process results in the construction of a basis (e1 , . . . , en ) for Rn . Show that with respect to this basis, the (gij ) of Exercise 3.5 are 1, i = j, gij = ei , ej = 0, i = j. (We say that the basis (e1 , . . . , en ) is orthonormal with respect to the inner product Ā·, Ā·.) Therefore, the isomorphism Ļ : Rn ā En that identiļ¬es (e1 , . . . , en ) with the standard basis (e1 , . . . , en ) for En identiļ¬es Ā·, Ā· with the standard inner product on En . Exercise 3.7. Let (e1 , . . . , en ) be an orthonormal basis for En . Show that the dual basis (eā1 , . . . , eān ) for the dual space (En )ā consists of the linear mappings eāi : En ā R deļ¬ned by eāi (v) = ei , v for v ā En . 3.2.2. Symmetries and isotropy groups. Now we consider the issue of orientation-preserving symmetries (cf. Remark 3.2) of Euclidean space: What kinds of orientation-preserving transformations Ļ : En ā En preserve the fundamental properties of lengths of vectors and angles between vectorsāspeciļ¬cally, lengths and angles of vectors tangent to Rn and based at the same point of Rn ? The answer (which hopefully you already know) is
3.2. Euclidean space
73
translations and rotations, collectively known as ārigid motionsā. Any such transformation has the form Ļ(x) = Ax + b, where A is an element of the special orthogonal group SO(n) and b ā En . The set of such transformations forms a Lie group, called the Euclidean group E(n). This group can be represented as a group of (n + 1) Ć (n + 1) matrices as follows: Let ( ) 1 t0 n E(n) = : A ā SO(n), b ā E . b A Here A is an n Ć n matrix, b is an n Ć 1 column vector, and t0 represents a 1 Ć n row of 0ās, i.e., the transpose of the n Ć 1 column vector 0. If we 1 n represent a vector x ā E by the (n + 1)-dimensional column vector , x then elements of E(n) act by matrix multiplication: 1 t0 1 1 = . b A x Ax + b Remark 3.8. In the literature on Lie groups, it is common to write the A b x matrix as t and the vector as . We have chosen to write them in 0 1 1 this way so that, once we get around to deļ¬ning moving frames, the order of the columns in the matrix will correspond to the order of the vectors in the associated moving frame. Given a point x ā En , it is natural to ask: Which elements of E(n) leave x ļ¬xed? In other words, which transformations Ļ ā E(n) have the property that Ļ(x) = x? The set of such transformations is a subgroup of E(n), called the isotropy group of the point x ā En and denoted Hx . 1 t0 This question is easiest to answer when x = 0: An element ļ¬xes the b A point x = 0 if and only if b = 0. Thus, the isotropy group H0 of the point 1 t0 0 ā En consists of all elements of E(n) of the form . This subgroup 0 A is clearly isomorphic to SO(n). What about other points x ā En ? It seems reasonable to expect that there shouldnāt be anything special about 0 because we can move any point to any other point via a translation. And, in fact, the isotropy group Hx of any point x ā En is also isomorphic to SO(n). We can deļ¬ne an explicit 1 t0 isomorphism Ļ : H0 ā Hx as follows. Let tx = (where I represents x I
74
3. Homogeneous spaces
the nĆn identity matrix), so that tx represents the translation tx (y) = y+x. Then for any element h ā H0 , deļ¬ne Ļ(h) = tx htā1 x ā E(n). The following exercise shows that Ļ is indeed an isomorphism from H0 to Hx : Exercise 3.9. (a) Show that for any h ā H0 , we have Ļ(h) ā Hx ; i.e., Ļ(h)(x) = x. Therefore, Ļ is a map from H0 to Hx . (b) Show that Ļ : H0 ā Hx is a homomorphism; i.e., show that for any two elements h1 , h2 ā H0 , we have Ļ(h1 h2 ) = Ļ(h1 )Ļ(h2 ). (c) Show that Ļ : H0 ā Hx is injective; i.e., show that if Ļ(h) is the identity element in Hx , then h must be the identity element in H0 . Ė ā Hx , (d) Show that Ļ : H0 ā Hx is surjective; i.e., show that for any h Ė there exists h ā H0 with Ļ(h) = h. The isomorphism deļ¬ned by Ļ is called conjugation; we describe it using the notation Hx = tx H0 tā1 x , and we say that all the isotropy groups Hx ā E(n) are conjugate in E(n). Moreover, they are all isomorphic to SO(n). Deļ¬nition 3.10. The left coset tx H0 is the subset of E(n) deļ¬ned by tx H0 = {tx h | h ā H0 }. Note that this set is not a subgroup of E(n) unless x = 0. The following exercise shows that E(n) is the disjoint union of all the left cosets tx H0 , x ā En : *Exercise 3.11. (a) Show that ) ( 1 t0 tx H0 = : A ā SO(n) . x A (b) Conclude from part (a) that: (1) If x = y, then the left cosets tx H0 and ty H0 are disjoint. (2) Every transformation Ļ ā E(n) belongs to some left coset tx H0 .
3.3. Orthonormal frames on Euclidean space
75
In general, if G is a group and H is a subgroup of G, then the set of left cosets of H in G is denoted by G/H. Since H0 is isomorphic to SO(n), Exercise 3.11 shows that we have a natural correspondence En ā¼ = E(n)/SO(n). The set E(n)/SO(n) can be given a manifold structure so that this correspondence becomes a diļ¬eomorphism.
3.3. Orthonormal frames on Euclidean space 3.3.1. The orthonormal frame bundle. Another way to look at all this is in terms of orthonormal frames on En . Deļ¬nition 3.12. An (oriented) orthonormal frame f on En is a list of vectors f = (x; e1 , . . . , en ), where x ā En and (e1 , . . . , en ) is an oriented, orthonormal basis for the tangent space Tx En . Alternatively, we may say that (e1 , . . . , en ) is a orthonormal frame based at x. If we regard the vectors (e1 , . . . , en ) as the columns of a matrix A ā SO(n), then we see that there is a one-to-one correspondence between the set of frames on En and the Euclidean group E(n): The vector x represents the translation component, and the matrix A represents the rotation component. Regarded in this way, we can deļ¬ne a projection map Ļ : E(n) ā En by Ļ(x; e1 , . . . , en ) = x. This map is diļ¬erentiable, and the ļ¬ber over any point x ā En is the set of all oriented, orthonormal frames based at x. SO(n) acts freely and transitively on each ļ¬ber, and so this map gives an explicit description of E(n) as a principal bundle over En with ļ¬ber group SO(n): SO(n) - E(n) Ļ
?
En ā¼ = E(n)/SO(n). In this context, E(n) is also called the (oriented) orthonormal frame bundle of En , and it is denoted F (En ). 3.3.2. Dual forms, connection forms, and structure equations. A guiding principle as we proceed is that the power of the method of moving frames lies in expressing the derivatives of a frame in terms of the frame itself. (Remember how well this worked when you studied Frenet frames for curves?) So, our next step is to consider the derivativesāspeciļ¬cally, the
76
3. Homogeneous spaces
exterior derivativesāof the components (x, e1 , . . . , en ) of a frame. These components may all be thought of as En -valued functions on F (En ). (For example, the function x : F (En ) ā En is just the projection map Ļ.) Remark 3.13. In fact, this requires some abuse of notation. The function x is legitimately En -valued, but the functions (e1 , . . . , en ) actually take values in the tangent bundle T En , and their derivativesāwhich we will get to shortlyātake values in the bundle T (T En ). This distinction is sometimes important, but for the most part we will ignore it by making use of the canonical isomorphisms Tx En ā¼ = En , which allow us to regard (e1 , . . . , en ) n as E -valued functions. Now, consider the exterior derivatives dx, dei of the functions x, ei on F (En ). (Recall from Remark 2.39 that these are simply the diļ¬erentials of the maps x, ei : F (En ) ā En .) For any point f = (x; e1 , . . . , en ) ā F (En ), these are maps dx : Tf F (En ) ā Tx En , dei : Tf F (En ) ā Tei En ā¼ = Tei (Tx En ) ā¼ = Tx En . Moreover, since for any frame f = (x; e1 , . . . , en ) the vectors (e1 , . . . , en ) form a basis for the tangent space Tx En , the vector-valued 1-forms dx and dei can be expressed as linear combinations of (e1 , . . . , en ) whose coeļ¬cients are ordinary scalar-valued 1-forms (cf. Exercise 2.40). These considerations lead us to deļ¬ne scalar-valued 1-forms (Ļ i , Ļji ) on F (En ) by the equations (3.1)
dx = ei Ļ i , dei = ej Ļij ,
where 1 ā¤ i, j, ā¤ n. Remark 3.14. While it may look strange to write the scalar-valued 1-forms after the vectors in these equations, order is important here: For example, the ļ¬rst equation may be written as the matrix product ā” 1ā¤ Ļ
ā¢ . ā„ (3.2) dx = e1 Ā· Ā· Ā· en ā£ .. ā¦ , Ļn and it wouldnāt make sense in the other order without interchanging the roles of row and column vectors.
3.3. Orthonormal frames on Euclidean space
77
In other words, the 1-forms (Ļ i , Ļji ) on F (En ) are deļ¬ned by the property that for any f = (x; e1 , . . . , en ) ā F (En ), v ā Tf F (En ), dx(v) = ei Ļ i (v) ā Tx En , dei (v) = ej Ļij (v) ā Tei (Tx En ) ā¼ = Tx En . Remark 3.15. Note that the 1-forms (Ļ i , Ļji ) are not well-deļ¬ned on the base space En since they are deļ¬ned relative to a particular choice of frame (e1 , . . . , en ) for the tangent space Tx En . We can actually describe these 1-forms fairly explicitly. Given a point f = (x; e1 , . . . , en ) ā F (En ), let A ā SO(n) denote the matrix
A = e1 Ā· Ā· Ā·
en .
Then equation (3.2) can be written as ā”
ā¤ Ļ1 ā¢ ā„ dx = A ā£ ... ā¦ . Ļn Therefore, ā” (3.3)
ā¤ ā” 1ā¤ Ļ1 dx ā¢ .. ā„ ā1 ā1 ā¢ .. ā„ ā£ . ā¦ = A dx = A ā£ . ā¦ . Ļn
dxn
From this expression, it is clear that the 1-forms (Ļ 1 , . . . , Ļ n ) are linearly independent and form a basis for the 1-forms on En . Remark 3.16. This, too, requires some abuse of notation: While (Ļ 1 , . . ., Ļ n ) are not well-deļ¬ned on En , they are linearly independent, linear combinations of the 1-forms (dx1 , . . . , dxn )āor, more precisely, of the pullbacks (Ļ ā (dx1 ), . . ., Ļ ā (dxn )) of (dx1 , . . . , dxn ) to F (En ). Therefore, (dx1 , . . . , dxn ) can be expressed as linear combinations of (Ļ 1 , . . . , Ļ n ), with coeļ¬cients that are functions on F (En )āand indeed, the ļ¬rst equation in (3.1) does just that.
78
3. Homogeneous spaces
Similarly, the equations for dei in (3.1) can be combined into the matrix equation ā¤ ā” 1 Ļ1 Ā· Ā· Ā· Ļn1
ā¢ .. ā„ dA = de1 Ā· Ā· Ā· den = e1 Ā· Ā· Ā· en ā£ ... . ā¦ n Ļ1 Ā· Ā· Ā· Ļnn ā” 1 ā¤ Ļ1 Ā· Ā· Ā· Ļn1 ā¢ .. ā„ . = A ā£ ... . ā¦ Ļ1n Ā· Ā· Ā·
Ļnn
Therefore, ā”
ā¤ Ļn1 .. ā„ = Aā1 dA. . ā¦
Ļ11 Ā· Ā· Ā· ā¢ .. ā£ .
(3.4)
Ļ1n Ā· Ā· Ā·
Ļnn
Example 3.17. Consider the case n = 2. Let x = (x1 , x2 ) denote the coordinates of an arbitrary point in E2 . Any orthonormal frame (e1 , e2 ) for the tangent space Tx E2 can be written as e1 =
cos(Īø)
,
sin(Īø)
e2 =
ā sin(Īø)
,
cos(Īø)
where Īø is the angle between e1 and the standard basis vector e1 = āxā 1 . Thus, Īø may be regarded as a local coordinate on the 1-dimensional ļ¬bers of the orthonormal frame bundle F (E2 ). Now, write cos(Īø) ā sin(Īø)
A = e1 e2 = . sin(Īø) cos(Īø) Then (Ļ 1 , Ļ 2 ) are given by
Ļ1 Ļ2
(3.5)
= Aā1 dx =
sin(Īø)
ā sin(Īø) cos(Īø)
=
cos(Īø)
dx1
dx2
cos(Īø) dx1 + sin(Īø) dx2
ā sin(Īø) dx1 + cos(Īø) dx2
,
3.3. Orthonormal frames on Euclidean space
79
and the 1-forms (Ļji ) are given by 1 Ļ1 Ļ21 = Aā1 dA Ļ12 Ļ22 cos(Īø) sin(Īø) ā sin(Īø) dĪø ā cos(Īø) dĪø = (3.6) ā sin(Īø) cos(Īø) cos(Īø) dĪø ā sin(Īø) dĪø
0
ādĪø
dĪø
0
=
.
The 1-forms (Ļ 1 , . . . , Ļ n ) are often called the dual forms on the orthonormal frame bundle F (En ) because they have the property that at any point f ā F (En ), 1, i = j, i i (3.7) Ļ (ej ) = Ī“j = 0, i = j. Exercise 3.18. Verify equations (3.7) by direct computation. The dual forms also have the property that Ļ i (v) = 0 for any vector v ā T F (En ) that is tangent to the ļ¬bers of the projection Ļ : F (En ) ā En . The technical way to say this is that the pullback of Ļ i to any ļ¬ber Ļ ā1 (x0 ) of Ļ via the inclusion map Ī¹ : Ļ ā1 (x0 ) ā F (En ) vanishes. Exercise 3.19. Prove this statement. (Hint: Let Ī± : I ā F (En ) be a curve tangent to v. Since v is tangent to some ļ¬ber Ļ ā1 (x0 ) ā F (En ), you can assume that Ī±(I) ā Ļ ā1 (x 0 ). So, what is the value of x at any point Ī±(t)? d Now compute dx(v) = dt t=0 x(Ī±(t)).) Forms with this property (i.e., their pullbacks to each ļ¬ber of Ļ vanish) are called semi-basic for the projection Ļ; for this reason, the dual forms are also sometimes called the semi-basic forms on F (En ). The pullbacks of (Ļji ), on the other hand, form a basis for the 1-forms on each ļ¬ber of Ļ. They are called the connection forms on F (En ). Now for the fun part: Start diļ¬erentiating! In order to compute the exterior derivatives of the dual forms and connection forms, we need to diļ¬erentiate equations (3.1). *Exercise 3.20. Diļ¬erentiate equations (3.1) (taking the second equation into account!) and do some careful index-juggling to obtain 0 = ei (Ļji ā§ Ļ j + dĻ i ), 0 = ei (Ļki ā§ Ļjk + dĻji ).
80
3. Homogeneous spaces
Conclude that these 1-forms satisfy the Cartan structure equations (3.8)
dĻ i = āĻji ā§ Ļ j , dĻji = āĻki ā§ Ļjk .
(Hint: Recall the Leibniz rule for diļ¬erentiating p-forms, and observe that the vector ļ¬eld ei is a vector-valued 0-form. Note that if we wrote the terms in (3.1) in the other orderāand some authors do!āthe Leibniz rule would lead to diļ¬erent signs in the structure equations.) Exercise 3.21. Verify by direct computation that the dual forms (3.5) and connection forms (3.6) on F (E2 ) satisfy the Cartan structure equations (3.8). Up to this point, we havenāt taken advantage of the fact that we have a Euclidean structure on En . Since (e1 , . . . , en ) are orthonormal vectors, we have 1, i = j, (3.9) ei , ej = Ī“ij = 0, i = j. *Exercise 3.22. Diļ¬erentiate equations (3.9) and conclude that the connection forms (Ļji ) are skew-symmetric in their indices; that is, they have the property that Ļij = āĻji . Exercise 3.23. Write out equations (3.1) and the structure equations (3.8) explicitly (i.e., without the summation convention) in the case n = 3. What is the dimension of F (E3 )? How many linearly independent connection forms are there? 3.3.3. The Maurer-Cartan form. This all ļ¬ts into a larger structure, which is easier to see if we go back to regarding F (En ) as the Lie group E(n). Exercise 3.24. Prove that the Lie algebra of E(n) is ( ) 0 t0 n e(n) = : B ā so(n), b ā E . b B (Recall that B ā so(n) if and only if B is a skew-symmetric n Ć n matrix.) We can deļ¬ne an e(n)-valued 1-form Ļ on E(n) as follows: Recall that for h ā E(n), the left multiplication map Lh : E(n) ā E(n) is deļ¬ned by Lh (g) = hg, and for any g ā E(n), we have the diļ¬erential (Lh )ā : Tg E(n) ā Thg E(n).
3.3. Orthonormal frames on Euclidean space
81
Now, for any g ā E(n), v ā Tg E(n), deļ¬ne Ļ(v) = (Lgā1 )ā (v). Since (Lgā1 )ā maps Tg E(n) to TI E(n) = e(n), we have Ļ(v) ā e(n) for all v ā T E(n). The form Ļ is called the Maurer-Cartan form on E(n). *Exercise 3.25. Prove that the Maurer-Cartan form Ļ on E(n) is leftinvariant; i.e., show that for any h ā E(n), Lāh Ļ = Ļ. (Hint: Let g ā E(n), v ā Tg E(n), and compute (Lāh Ļ)(v) = Ļ((Lh )ā (v)). Try to keep track of which tangent space each object lives in!) While this deļ¬nition is fairly simple in theory, it is not so easy to work with computationally. The Maurer-Cartan form is more commonly written as (3.10)
Ļ = g ā1 dg.
This notation requires some explanation. The variable g here essentially denotes the identity map g : E(n) ā E(n), but it really should be thought of as a coordinate function on E(n), which realizes any element of the abstract Lie group E(n) as its (n + 1) Ć (n + 1) matrix representation. Speciļ¬cally, for any f = (x; e1 , . . . , en ) ā E(n), 1 0 Ā·Ā·Ā· 0 (3.11) g(f ) = . x e 1 Ā· Ā· Ā· en Now, let f0 be any element of E(n), and let g0 = g(f0 ). (So f0 and g0 represent the same element of E(n); the diļ¬erence in notation is meant to suggest that f0 is an element of the abstract group and g0 is its matrix representation.) Diļ¬erentiating (3.11) shows that the mapping dg : Tf0 E(n) ā Tg0 E(n) can be written as the matrix-valued 1-form 0 0 Ā·Ā·Ā· (3.12) dg = dx de1 Ā· Ā· Ā·
0
.
den
Since g is essentially the identity map, the same is true for dg: It simply identiļ¬es any tangent vector v ā Tf0 E(n) with its matrix representation as a tangent vector to the matrix group E(n) at g0 . So dg(v) = v ā Tg0 E(n). The left multiplication by g ā1 , applied to a vector v ā Tg0 E(n), then means to multiply the matrix representation for v by the matrix representation for g0ā1 .
82
3. Homogeneous spaces
Remark 3.26. The realization of E(n) as a matrix group is crucial in order for this notation to make any sense at all: In the abstract setting, there is no way to multiply v by an element of E(n) because v is a tangent vector and not a group element. The only sensible interpretation of āg ā1 dg(v)ā is to act on the vector dg(v) = v ā Tg0 E(n) by the diļ¬erential of the leftmultiplication map Lgā1 . (Note that this is our original deļ¬nition for Ļ!) 0 But because everything here is matrix-valued, (Lgā1 )ā (v) is, in fact, given 0
by the product of the matrices g0ā1 and v.
As you might suspect, it is no accident that we have used the letter Ļ both for the scalar-valued dual and connection forms and for the matrix-valued Maurer-Cartan form. The following exercise shows how they are related: *Exercise 3.27. (a) Show that the Maurer-Cartan form on E(n) is given by ā” ā¤ 0 0 Ā·Ā·Ā· 0 ā¢Ļ1 Ļ1 Ā· Ā· Ā· Ļ1 ā„ nā„ 1 ā¢ Ļ=ā¢ . .. .. ā„ . ā£ .. . . ā¦ Ļ n Ļ1n Ā· Ā· Ā·
Ļnn
(Hint: Use equations (3.1), (3.11), (3.12) and the fact that Ļ = g ā1 dg is equivalent to dg = gĻ.) Write out this matrix explicitly in the case n = 3. (b) Use the skew-symmetry of the forms (Ļji ) to conļ¬rm that for any v ā T E(n), we have Ļ(v) ā e(n). (This is what it means to say that Ļ is āe(n)-valuedā.) Because the matrix-valued Maurer-Cartan form Ļ contains the scalar-valued dual forms and connection forms as its entries, the dual forms and connection forms are collectively referred to as āthe Maurer-Cartan formsā. The fact that Ļ is left-invariant implies that the Maurer-Cartan forms are leftinvariant as well; in fact, they form a basis for the left-invariant 1-forms on E(n). *Exercise 3.28. Show that the structure equations (3.8) are equivalent to the Maurer-Cartan equation dĻ = āĻ ā§ Ļ. (The wedge product of two matrices of 1-forms is computed just like the ordinary matrix product, substituting wedge product for ordinary multiplication of the appropriate entries.) Note that, despite the skew-symmetry of the wedge product for scalar-valued 1-forms, the wedge product of a matrixvalued 1-form with itself does not necessarily vanish!
3.3. Orthonormal frames on Euclidean space
83
*Exercise 3.29. Suppose that we choose a particular orthonormal frame (e1 (x), . . . , en (x)) for Tx En at each point x ā En . This amounts to choosing a section Ļ : En ā F (En ) of the orthonormal frame bundle Ļ : F (En ) ā En , also known as an orthonormal frame ļ¬eld on En . The pullbacks (Ļ ā (Ļ i ), Ļ ā (Ļji )) are then 1-forms on En . In order to reduce notational clutter, we will denote these pullbacks by (ĀÆ Ļi, Ļ ĀÆ ji ). Show that if we set
A(x) = e1 (x) . . . en (x) , then the dual forms and connection given by ā” 1ā¤ ā” 1ā¤ Ļ ĀÆ dx ā¢ .. ā„ ā1 ā¢ .. ā„ ā£ . ā¦ = A(x) ā£ . ā¦ , Ļ ĀÆn
dxn
forms associated to this frame ļ¬eld are ā”
Ļ ĀÆ 11 . . . ā¢ .. ā£ . Ļ ĀÆ 1n . . .
ā¤ Ļ ĀÆ n1 .. ā„ = A(x)ā1 dA(x). . ā¦ Ļ ĀÆ nn
(Compare with equations (3.3), (3.4).) Show that if we write ā¤ ā” 0 0 Ā· Ā· Ā· 0 t0 1 Ļ ā¢Ļ 1 ĀÆ 11 Ā· Ā· Ā· Ļ ĀÆ n1 ā„ ā„ ā¢ĀÆ g(x) = , Ļ ĀÆ =ā¢ . .. .. ā„ , . ā£ . . . ā¦ x A(x) n n Ļ ĀÆ Ļ ĀÆ1 Ā· Ā· Ā· Ļ ĀÆ nn then these equations imply that Ļ ĀÆ = g(x)ā1 dg(x). Remark 3.30. The dual forms (Ļ i ) and connection forms (Ļji ) are all linearly independent āupstairsā on the frame bundle F (En ), but this is no longer the case when these forms are pulled back ādownstairsā to En via a section, as in Exercise 3.29. The pulled-back dual forms (ĀÆ Ļ i ) are linearly 1 n independent, linear combinations of (dx , . . . , dx ), but the pulled-back connection forms (ĀÆ Ļji ) are no longer linearly independent from the dual forms; indeed, they may be expressed as linear combinations of (ĀÆ Ļ1, . . . , Ļ ĀÆ n ), which n form a basis for the 1-forms on E . Exercise 3.31. Consider E3 (minus the z-axis) with cylindrical coordinates (r, Īø, z), related to the usual Euclidean coordinates (x, y, z) by x = r cos(Īø),
y = r sin(Īø),
z = z.
Apply the result of Exercise 3.29 to the cylindrical frame ļ¬eld ā” ā¤ ā” ā¤ ā” ā¤ cos(Īø) ā sin(Īø) 0 ā¢ ā„ ā¢ ā„ ā¢ ā„ ā„ ā„ ā„ e1 = ā¢ e2 = ā¢ e3 = ā¢ ā£ sin(Īø) ā¦ , ā£ cos(Īø) ā¦ , ā£0ā¦ 0 0 1
84
3. Homogeneous spaces
to compute the dual forms (ĀÆ Ļ i ) and the connection forms (ĀÆ Ļji ) for this frame ļ¬eld. Show by direct computation that these forms satisfy the structure equations (3.8). Exercise 3.32. Repeat Exercise 3.31 for the spherical frame ļ¬eld ā” ā¤ ā” ā¤ ā” ā¤ cos(Ļ) cos(Īø) ā sin(Īø) ā sin(Ļ) cos(Īø) ā¢ ā„ ā¢ ā„ ā¢ ā„ ā„ ā„ ā„ e1 = ā¢ e2 = ā¢ e3 = ā¢ ā£ cos(Ļ) sin(Īø) ā¦ , ā£ cos(Īø) ā¦ , ā£ ā sin(Ļ) sin(Īø) ā¦ , sin(Ļ)
0
cos(Ļ)
where (Ļ, Ļ, Īø) are spherical coordinates on E3 (minus the z-axis), related to the usual Euclidean coordinates (x, y, z) by x = Ļ cos(Ļ) cos(Īø),
y = Ļ cos(Ļ) sin(Īø),
z = Ļ sin(Ļ).
3.4. Homogeneous spaces Letās take a moment to summarize our work so far: We started with Euclidean space En , i.e., the vector space Rn endowed with an inner product Ā·, Ā·. By considering the group of symmetries of this structure, we arrived at a description of En as the set of left cosets of the closed subgroup SO(n) of the Lie group E(n). These particular groups arose as the symmetry group of the structure (E(n)) and the isotropy group of a particular point (SO(n)). This description of En as the group quotient E(n)/SO(n) realizes En as a homogeneous space; it is a special case of the following general construction: Deļ¬nition 3.33. A homogeneous space is the set G/H of left cosets of a closed subgroup H of a Lie group G, endowed with the unique manifold structure with respect to which the natural action of G on G/H is smooth. (The existence and uniqueness of this manifold structure on G/H is a consequence of a theorem of Cartan [Car30] which states that every closed subgroup of a Lie group is a Lie subgroup.) The Lie group G is called the symmetry group of the homogeneous space G/H. In most commonly encountered examples, G is a subgroup of some matrix group GL(m). G acts on G/H by left multiplication in the obvious way, and we are generally interested in those properties (such as the inner product on Euclidean space) that are preserved under this action. For any point x = gH ā G/H, the isotropy group Hx of x is the group Hx = gHg ā1 , where g is any element of the coset gH. All the isotropy groups Hx are clearly conjugate to H in G. Moreover, the projection map Ļ : G ā G/H
3.5. Minkowski space
85
deļ¬ned by Ļ(g) = gH describes G as a principal bundle over G/H with ļ¬ber H: H
- G Ļ
?
G/H. Just as we identiļ¬ed the group E(n) with the set of orthonormal frames on Euclidean space, there is often some natural way to think of the group G as the set of āframesā on the space G/H. The Maurer-Cartan form on G is deļ¬ned by Ļ(v) = (Lgā1 )ā (v) for any v ā Tg G, and if G is a matrix group, then we can write Ļ = g ā1 dg. Either way, Ļ takes values in the Lie algebra g of G, and it satisļ¬es the Maurer-Cartan equation: (3.13)
dĻ = āĻ ā§ Ļ.
This equation will turn out to play a crucial role in the geometry of submanifolds of the space G/H. *Exercise 3.34. Suppose that G is a matrix Lie group, so that Ļ = g ā1 dg. Prove directly that Ļ satisļ¬es the Maurer-Cartan equation (3.13). (Hint: Show by diļ¬erentiating the equation gg ā1 = I that d(g ā1 ) = āg ā1 dg g ā1 .) In the remainder of this chapter, we will explore some other examples of homogeneous spaces: Minkowski space, equi-aļ¬ne space, and projective space.
3.5. Minkowski space 3.5.1. The Minkowski inner product. Minkowski space was introduced in 1907 by Hermann Minkowski [Min78] as a geometric setting for Einsteinās theory of special relativity. In Minkowskiās original formulation, the three dimensions of Euclidean space are combined with a single time dimension to create a 4-dimensional spacetime. More generally, we can consider Minkowski spaces consisting of n space dimensions and one time dimension.
86
3. Homogeneous spaces
What distinguishes Minkowski space from Euclidean space is its inner product structure: Instead of a positive deļ¬nite inner product, Minkowski space is equipped with an indeļ¬nite inner product. Given any symmetric bilinear form Q on the vector space Rn , there exists a basis (e1 , . . . , en ) for Rn and integers p, q ā„ 0 with the property that for any v = v i ei ā Rn , Q(v, v) = (v 1 )2 + Ā· Ā· Ā· + (v p )2 ā (v p+1 )2 ā Ā· Ā· Ā· ā (v p+q )2 . The integers p, q are invariants of Q, and the ordered pair (p, q) is called the signature of Q. The form Q is nondegenerate if and only if p + q = n. A positive deļ¬nite form Q has signature (n, 0), whereas Q is called indeļ¬nite if p and q are both greater than zero. Deļ¬nition 3.35. The (1 + n)-dimensional Minkowski space M1,n is the vector space Rn+1 endowed with a nondegenerate, symmetric bilinear form Ā·, Ā· with signature (1, n). This bilinear form is called the Minkowski inner product on M1,n . Equivalently, for any basis (e0 , . . . , en ) for Rn+1 , the symmetric matrix g = [gĪ±Ī² ] = [eĪ± , eĪ² ] that describes the Minkowski inner product Ā·, Ā· in terms of this basis has one positive eigenvalue and n negative eigenvalues. Remark 3.36. Even when n has a ļ¬xed value, say n = 3, the Minkowski space M1,3 is still often referred to as ā(1+3)-dimensional Minkowski spaceā to emphasize the distinction between the time and space dimensions. Remark 3.37. The Minkowski inner product is sometimes taken to have signature (n, 1) rather than (1, n). Our convention is the same as that used in [Cal00]; it is chosen so that tangent vectors v to curves corresponding to the world lines of particles in spacetime will have v, v > 0. Remark 3.38. For clarity, we will generally use Roman letters (i, j, etc.) for indices that range from 1 to n and Greek letters (Ī±, Ī², etc.) for indices that range from 0 to n. Just as in the Euclidean case, all Minkowski inner products on Rn+1 are equivalent, as the following exercise shows. (Cf. Exercise 3.6.) Exercise 3.39. Let Ā·, Ā· be any Minkowski inner product on Rn+1 . Construct a basis (e0 , . . . , en ) for Rn+1 as follows: Choose any nonzero vector v0 ā Rn+1 with v0 , v0 > 0 , and set v0 e0 =
. v0 , v0
3.5. Minkowski space
87
Let (e0 )ā„ denote the orthogonal complement of e0 with respect to Ā·, Ā·; i.e., (e0 )ā„ = {v ā Rn+1 | v, e0 = 0}. (a) Show that for any nonzero vector v ā (e0 )ā„ , v, v < 0. (Hint: Let (v1 , . . . , vn ) be any basis for (e0 )ā„ . Show that the matrix of g with respect to the basis (e0 , v1 , . . . , vn ) has the form t 1 0 g= , 0 A where A is a symmetric n Ć n matrix with all negative eigenvalues. Use this to prove the result.) (b) By part (a), the restriction of Ā·, Ā· to (e0 )ā„ is negative deļ¬nite. Construct an orthonormal basis (e1 , . . . , en ) for (e0 )ā„ as in Exercise 3.6. (For purposes of this construction, there is essentially no diļ¬erence between negative deļ¬nite and positive deļ¬nite inner products.) (c) Show that with respect to the basis ā§ āŖ āØ1, (3.14) gĪ±Ī² = eĪ± , eĪ² = ā1, āŖ ā© 0,
(e0 , . . . , en ), the (gĪ±Ī² ) are given by Ī± = Ī² = 0, Ī± = Ī² = 1, . . . , n, Ī± = Ī².
We say that the basis (e0 , . . . , en ) is orthonormal with respect to the inner product Ā·, Ā·, and (3.14) is the standard Minkowski inner product on M1,n . Henceforth, we will let (e0 , . . . , en ) denote the standard basis for M1,n . Nonzero vectors in M1,n are divided into three types: Deļ¬nition 3.40. A nonzero vector v ā M1,n is called ā¢ timelike if v, v > 0; ā¢ spacelike if v, v < 0; ā¢ lightlike or null if v, v = 0. A timelike vector v ā M1,n is called future-pointing if v, e0 > 0 and pastpointing if v, e0 < 0. The lightlike vectors form a cone, called the light cone or null cone, opening in both directions with axis parallel to e0 . The interior of this cone consists of timelike vectors, and the exterior consists of spacelike vectors.
88
3. Homogeneous spaces
Remark 3.41. This terminology comes from special relativity, where e0 is regarded as the time direction and (e1 , . . . , en ) as the spatial directions. The path of a particle traces out a curve in spacetime, called its world line. Since nothing can travel faster than the speed of light (which is normalized to c = 1 for the standard Minkowski inner product), the world line of any particle traveling at speed v < c must be a timelike curve (i.e., a curve whose tangent vectors are all timelike). The world line of a photon traveling at the speed of light is a lightlike curve (i.e., a curve whose tangent vectors are all lightlike). Any two points q1 , q2 ā M1,n with spacelike separation (i.e., for which the vector q2 ā q1 is spacelike) represent events that cannot have any causal impact on each other because they cannot be connected by a smooth timelike curve. Deļ¬nition 3.42. The Minkowski norm v of a nonzero vector v ā M1,n is deļ¬ned to be
(1) v = v, v if v is timelike;
(2) v = āv, v if v is spacelike; (3) v = 0 if v is lightlike. (Note that v ā„ 0 for all v ā M1,n .) Exercise 3.43. For n ā„ 2 and any Ī» ā R, show that the āsphereā SĪ» consisting of all vectors v ā M1,n satisfying v, v = Ī» is (1) a hyperboloid of two sheets when Ī» > 0; (2) a cone when Ī» = 0; (3) a hyperboloid of one sheet when Ī» < 0. (See Figure 3.1; note that the e0 -axis is drawn as the vertial axis.)
Figure 3.1. Minkowski āspheresā SĪ» with Ī» = 1, Ī» = 0, Ī» = ā1
3.5.2. Symmetries and isotropy groups. Let M1,n denote the Minkowski space of dimension 1 + n with the standard inner product (3.14). Just as we did for Euclidean space, we ask: What are the symmetries of M1,n ?
3.5. Minkowski space
89
*Exercise 3.44. Let A ā GL(n + 1). Show that Av, Av = v, v for every v ā M1,n if and only if A is an element of the Lie group O(1, n) = ā§ ā” 1 0 Ā·Ā·Ā· āŖ āŖ āŖ āØ ā¢0 ā1 Ā· Ā· Ā· ā¢ A ā GL(n + 1) : tA ā¢ . .. āŖ ā£ .. . āŖ āŖ ā© 0 0 Ā·Ā·Ā·
ā” 1 0 Ā·Ā·Ā· ā„ ā¢0 ā1 Ā· Ā· Ā· ā„ ā¢ ā„ A = ā¢ .. .. ā¦ ā£. . ā1 0 0 Ā·Ā·Ā· 0 0 .. .
ā¤
0 0 .. . ā1
ā¤ā« āŖ āŖ āŖ ā„ā¬ ā„ ā„ . ā¦āŖ āŖ āŖ ā
Furthermore, show that O(1, n) consists of all matrices of the form
A = e 0 Ā· Ā· Ā· en , where (e0 , . . . , en ) is an orthonormal basis for M1,n . O(1, n) is called the Lorentz group, and elements of O(1, n) are called Lorentz transformations. Remark 3.45. Unlike O(n), which has two connected components, O(1, n) has four connected components. The connected component that contains the identity matrix is denoted SO + (1, n), and it consists of those transformations in O(1, n) that have determinant equal to 1 and map the vector e0 to a future-pointing vector. These transformations are called proper (because they preserve the orientation of the spatial dimensions) and orthochronous (because they preserve the orientation of the time direction). SO+ (1, n) is therefore referred to as the āproper, orthochronous Lorentz groupā or, more succinctly, the ārestricted Lorentz groupā. We will restrict our attention to the proper, orthochronous symmetries of M1,n , and for simplicity we will refer to elements of this group as Lorentz transformations. *Exercise 3.46. (a) Show that 1 cosh(Īø) sinh(Īø) + SO (1, 1) = :ĪøāR . sinh(Īø) cosh(Īø) What are the other components of O(1, 1)? (b) Show that the light cone 2t 0 1 3 [x , x ] ā M1,1 | (x0 )2 ā (x1 )2 = 0 and the hyperbolas 2t 0 1 3 [x , x ] ā M1,1 | (x0 )2 ā (x1 )2 = Ā±r2 are each preserved under the action of SO+ (1, 1).
90
3. Homogeneous spaces
(c) Experiment with the action of SO + (1, 1) on various shapes in the plane. For instance, consider the unit circle, parametrized by x(t) = t[x0 (t), x1 (t)] = t[cos(t), sin(t)]. Use Maple to plot the curve Ax(t), where A ā SO+ (1, 1), for various values of the group parameter Īø. Exercise 3.47. Show that the Lie algebra so(1, n) of O(1, n) is given by so(1, n) = ā§ ā” 1 0 Ā·Ā·Ā· āŖ āŖ āŖ āØ ā¢0 ā1 Ā· Ā· Ā· ā¢ B ā M(n+1)Ć(n+1) : tB ā¢ . .. āŖ ā£ .. . āŖ āŖ ā© 0 0 Ā·Ā·Ā·
ā¤ ā”
1 0 Ā·Ā·Ā· ā„ ā¢0 ā1 Ā· Ā· Ā· ā„ ā¢ ā„ + ā¢ .. .. ā¦ ā£. . ā1 0 0 Ā·Ā·Ā· 0 0 .. .
In particular, this implies that the entries
(bĪ±Ī² )
0 0 .. . ā1
ā¤
ā« āŖ āŖ āŖ ā¬
ā„ ā„ ā„B = 0 . āŖ ā¦ āŖ āŖ ā
of B satisfy the relations
b00 = 0, bi0 = b0i , bji
=
ābij ,
i = 1, . . . , n, i, j = 1, . . . , n.
Since the Minkowski inner product is also preserved by translation, the full symmetry group of M1,n consists of all transformations of the form Ļ(x) = Ax + b, where A ā SO+ (1, n) and b ā M1,n . These transformations form a Lie group, called the PoincarĀ“e group M (1, n), which can be represented as ( ) 1 t0 M (1, n) = : A ā SO + (1, n), b ā M1,n . b A 1,n As in the Euclideancase, if we represent a vector x ā M by the (n + 2)1 dimensional vector , then elements of M (1, n) act by matrix multiplicax tion: 1 t0 1 1 = . b A x Ax + b
Once again, we ask: Given a vector x ā M1,n , what is the isotropy group of x? *Exercise 3.48. Show that: (a) The isotropy group H0 of 0 ā M1,n is ( t ) 1 0 + H0 = : A ā SO (1, n) ā¼ = SO + (1, n). 0 A
3.5. Minkowski space
91
(b) The isotropy group Hx of any other point x ā M1,n is
1 where tx = x
Hx = tx H0 tā1 x ,
t0
I
represents the translation tx (y) = y + x.
(c) There is a natural correspondence between M1,n and M (1, n)/SO+ (1, n), the set of left cosets of SO+ (1, n) in M (1, n). 3.5.3. Orthonormal frames and Maurer-Cartan forms. Orthonormal frames on M1,n are deļ¬ned as follows. Deļ¬nition 3.49. An orthonormal frame f on M1,n is a list of vectors f = (x; e0 , . . . , en ), where x ā M1,n and (e0 , . . . , en ) is an orthonormal basis for the tangent space Tx M1,n . (We may also say that (e0 , . . . , en ) is an orthonormal frame based at x.) The same reasoning as in the Euclidean case shows that the set of orthonormal frames may be regarded as the PoincarĀ“e group M (1, n) via the one-toone correspondence 1 0 Ā·Ā·Ā· 0 g(x; e0 , . . . , en ) = . x e0 Ā· Ā· Ā· en We can deļ¬ne a projection map Ļ : M (1, n) ā M1,n by Ļ(x; e0 , . . . , en ) = x; the ļ¬ber of this map is the set of all orthonormal frames based at x. SO + (1, n) acts freely and transitively on each ļ¬ber, and so this map gives an explicit description of M (1, n) as a principal bundle over M1,n with ļ¬ber group SO+ (1, n): SO + (1, n)
-M (1, n) Ļ
?
M1,n ā¼ = M (1, n)/SO + (1, n). In this context, M (1, n) is also called the orthonormal frame bundle of M1,n , and it is denoted F (M1,n ). The Maurer-Cartan forms on M (1, n) are deļ¬ned exactly as in the Euclidean case by equations (3.1). The structure equations (3.8) are also the same as in the Euclidean case; the only diļ¬erence is that with the indeļ¬nite inner product, the connection forms (ĻĪ²Ī± ) are no longer quite skew-symmetric.
92
3. Homogeneous spaces
*Exercise 3.50. Diļ¬erentiate equations (3.14) and conclude that (1) Ļ00 = 0; (2) Ļ0i = Ļi0 for i = 1, . . . , n; (3) Ļij = āĻji for i, j = 1, . . . , n. *Exercise 3.51. Prove that the Lie algebra of M (1, n) is ( ) 0 t0 1,n m(1, n) = . : B ā so(1, n), b ā M b B Exercise 3.52. Repeat Exercise 3.23 for M1,2 .
3.6. Equi-aļ¬ne space 3.6.1. Volume forms. Aļ¬ne geometry is the study of geometric properties that are preserved by the action of invertible linear transformations and translations on Rn . There is less structure in aļ¬ne geometry than in Euclidean or Minkowski geometry; there are, for instance, no invariant notions of lengths of vectors or angles between vectors. Aļ¬ne transformations do, however, preserve collinearity of points, and the notion of parallel lines is still well-deļ¬ned in aļ¬ne geometry. In this book, we will study equi-aļ¬ne geometry (also known as special aļ¬ne geometry), which has slightly more structure than general aļ¬ne geometry, in that we endow Rn with a volume form. Deļ¬nition 3.53. A volume form on Rn is a nonzero element dV of the 1-dimensional vector space Īn (Rn )ā . Equivalently, dV is a skew-symmetric, multilinear function n copies 4 56 7 dV : Rn Ć Ā· Ā· Ā· Ć Rn ā R. A volume form provides a way to measure the volume of the parallelepiped spanned by any n vectors (v1 , . . . , vn ). It does not provide a way to measure lengths of individual vectors or angles between vectors. The following exercise shows that all volume forms on Rn are equivalent. *Exercise 3.54. Let dV0 be the standard volume form on Rn , deļ¬ned by the property that dV0 (e1 , . . . , en ) = 1, where (e1 , . . . , en ) is the standard basis for Rn .
3.6. Equi-aļ¬ne space
93
(a) Show that for any n vectors (v1 , . . . , vn ) ā Rn ,
dV0 (v1 , . . . , vn ) = det v1 Ā· Ā· Ā· vn . (b) Let dV be any other volume form on Rn . Show that dV = Ī» dV0 for some nonzero real number Ī». (Hint: Let Ī» = dV (e1 , . . . , en ).) (c) Let (v1 , . . . , vn ), (w1 , . . . , wn ) be two bases for Rn , and let dV be a volume form on Rn . Show that dV (v1 , . . . , vn ) = dV (w1 , . . . , wn ) if and only if
w1 Ā· Ā· Ā· wn = A v1 Ā· Ā· Ā· vn for some matrix A ā SL(n). (Hint: Consider the matrix
ā1 w1 Ā· Ā· Ā· wn v1 Ā· Ā· Ā· vn .)
(d) Let dV = Ī» dV0 . Show that the isomorphism Ļ : Rn ā Rn deļ¬ned by Ļ(v) = Ī»(1/n) v identiļ¬es dV with the standard volume form dV0 , in the sense that dV (v1 , . . . , vn ) = dV0 (Ļ(v1 ), . . . , Ļ(vn )). (If Ī» < 0 and n is even, ļ¬rst perform a linear transformation that interchanges e1 and e2 ; this will have the eļ¬ect of reversing the sign of Ī».) Therefore, all volume forms on Rn are equivalent. Deļ¬nition 3.55. The vector space Rn endowed with a volume form dV is called equi-aļ¬ne space and is denoted An . Remark 3.56. The volume form is not really part of the deļ¬nition of equiaļ¬ne space; where it comes into play is when we decide what transformations to allow in our symmetry group. Aļ¬ne transformations are those that preserve the linear structure of An ; equi-aļ¬ne (or special aļ¬ne) transformations are those that preserve the volume form as well. It is most commonāalthough not universalāto study geometric properties that are preserved under the group of equi-aļ¬ne transformations, and this is the approach that we will take. It is fairly common to use the terms āaļ¬ne spaceā and āaļ¬ne transformationsā to refer to the equi-aļ¬ne versions, but we will attempt to resist the temptation to do so. By Exercise 3.54, we may assume without loss of generality that dV is the standard volume form on Rn , so that for any n vectors (v1 , . . . , vn ) ā Rn ,
dV (v1 , . . . , vn ) = det v1 Ā· Ā· Ā· vn .
94
3. Homogeneous spaces
3.6.2. Symmetries and isotropy groups. Now we ask: What are the symmetries of equi-aļ¬ne space? (It turns out that āorientation-preservingā is an unneccesary stipulation here, because any volume-preserving transformation necessarily preserves orientation.) Since equi-aļ¬ne space has less structure that must be preserved by a symmetry than Euclidean space does, we might expect that the symmetry group would be larger. *Exercise 3.57. Let A ā GL(n). Show that for any basis (e1 , . . . , en ) for An , we have dV (Ae1 , . . . , Aen ) = (det A) dV (e1 , . . . , en ). Therefore, dV (Ae1 , . . . , Aen ) = dV (e1 , . . . , en ) if and only if A ā SL(n). Since the volume form is also preserved by translation, the full symmetry group of An consists of all transformations Ļ : An ā An of the form Ļ(x) = Ax + b, where A ā SL(n) and b ā An . These transformations form a Lie group, called the equi-aļ¬ne group or special aļ¬ne group A(n), which can be represented as ( ) 1 t0 n A(n) = : A ā SL(n), b ā A . b A Now, given a vector x ā An , what is the isotropy group of x? *Exercise 3.58. Show that: (a) The isotropy group H0 of 0 ā An is ) ( t 1 0 H0 = : A ā SL(n) ā¼ = SL(n). 0 A (b) The isotropy group Hx of any other point x ā An is
1 where tx = x
t0
I
Hx = tx H0 tā1 x , represents the translation tx (y) = y + x.
(c) There is a natural correspondence between An and A(n)/SL(n), the set of left cosets of SL(n) in A(n). 3.6.3. Unimodular frames and Maurer-Cartan forms. Unimodular frames on An are deļ¬ned as follows: Deļ¬nition 3.59. A unimodular frame f on An is a list of vectors f = (x; e1 , . . . , en ), where x ā An and (e1 , . . . , en ) is a basis for the tangent
3.6. Equi-aļ¬ne space
95
space Tx An that spans a parallelepiped of volume 1. (We may also say that (e1 , . . . , en ) is a unimodular frame based at x.) The same reasoning as in the Euclidean case shows that the set of unimodular frames may be regarded as the equi-aļ¬ne group A(n) via the one-to-one correspondence 1 0 Ā·Ā·Ā· 0 g(x; e1 , . . . , en ) = . x e1 Ā· Ā· Ā· en We can deļ¬ne a projection map Ļ : A(n) ā An by Ļ(x; e1 , . . . , en ) = x; the ļ¬ber of this map is the set of all unimodular frames based at x. SL(n) acts freely and transitively on each ļ¬ber, and so this map gives an explicit description of A(n) as a principal bundle over An with ļ¬ber group SL(n): SL(n) - A(n) Ļ
?
An ā¼ = A(n)/SL(n). In this context, A(n) is also called the unimodular frame bundle of An , and it is denoted F (An ). Just as for Euclidean and Minkowski spaces, the Maurer-Cartan forms on A(n) are deļ¬ned by equations (3.1). The structure equations (3.8) are also the same as before, but without any sort of inner product structure, there is no symmetry or skew-symmetry condition on the connection forms (Ļji ). There is, however, one relation among the (Ļji ) that comes from the unimodular condition
det e1 Ā· Ā· Ā· en = 1. We can diļ¬erentiate this condition more easily if we express it as (3.15)
e1 ā§ Ā· Ā· Ā· ā§ en = e1 ā§ Ā· Ā· Ā· ā§ en ,
where (e1 , . . . , en ) denotes the standard basis for An . (Recall that wedge product is simply the skew-symmetrization of the tensor product.) Exercise 3.60. Show that equation (3.15) holds for any unimodular basis by showing that for any basis (e1 , . . . , en ) of An , unimodular or not, we have
e1 ā§ Ā· Ā· Ā· ā§ en = det e1 Ā· Ā· Ā· en e1 ā§ Ā· Ā· Ā· ā§ en . *Exercise 3.61. Diļ¬erentiate equation (3.15) and conclude that the connection forms (Ļji ) satisfy the trace condition Ļii = 0. (Hint: Donāt let the
96
3. Homogeneous spaces
wedge product confuse you: (e1 , . . . , en ) are vector-valued 0-forms, so there are no minus signs in the Leibniz rule. Also, note that the right-hand side of (3.15) is constant.) *Exercise 3.62. Prove that the Lie algebra of the A(n) is ( ) 0 t0 n a(n) = : B ā sl(n), b ā A . b B (Recall that B ā sl(n) if and only if tr(B) = 0.) Exercise 3.63. Repeat Exercise 3.23 for A3 .
3.7. Projective space 3.7.1. The structure of projective space. Projective space is related to the study of perspectiveāfor instance, how a 3-dimensional object appears when projected onto a 2-dimensional image. All points in 3-dimensional space that lie on a line intersecting a common focal point (such as the lens of a camera) are projected onto the same point in the 2-dimensional image. In this geometry, lines that are parallel in 3-dimensional space may appear to intersect āat inļ¬nityā, but this intersection point may appear at some ļ¬nite distance in the 2-dimensional projection. Projective transformations of 2-dimensional space are the result of changing the focal point in 3-dimensional space. Projective transformations are more general than aļ¬ne transformations; among other things, they can map points at a ļ¬nite distance from the origin to points āat inļ¬nityā, and vice versa. Because of this, n-dimensional projective space must be constructed so as to include not only the space Rn of ļ¬nite-distance points, but also a point āat inļ¬nityā for every direction in Rn . Recall the deļ¬nition from Example 1.9: Deļ¬nition 3.64. The n-dimensional (real) projective space Pn is the set of lines through the origin in the vector space Rn+1 . In order to see that this captures the idea of āRn plus points at inļ¬nityā, we will use homogeneous coordinates [x0 : Ā· Ā· Ā· : xn ] (cf. Example 1.9) to represent a point in Pn . Consider the dense, open subset V0 of Pn given by V0 = {[x0 : Ā· Ā· Ā· : xn ] ā Pn | x0 = 0}. Any point [x] ā V0 can be written as
x1 xn [x] = [x : Ā· Ā· Ā· : x ] = 1 : 0 : Ā· Ā· Ā· : 0 x x 0
n
3.7. Projective space
97
and so may be represented by the point
n x1 , . . . , xx0 x0
!
ā Rn . We will denote
ĀÆ = (ĀÆ this point by x x1 , . . . , x ĀÆn ); these are referred to as aļ¬ne coordinates on the open set V0 . We can visualize this by observing that any line in Rn+1 that passes through the origin and another point x with x0 = 0 intersects the plane x0 = 1 in exactly one point. (See Figure 3.2.)
Figure 3.2. V0 ā Pn
On the other hand, any point [x] ā Pn \ V0 has the form [x] = [0 : x ĀÆ1 : Ā· Ā· Ā· : x ĀÆn ]. This point occurs as a limit point of the line [x(t)] = [1 : tĀÆ x1 : Ā· Ā· Ā· : tĀÆ xn ] in V0 as t ā Ā±ā and so may be thought of as a āpoint at inļ¬nityā correĀÆ = t[ĀÆ sponding to the direction of the vector x x1 , . . . , x ĀÆn ] ā Rn . *Exercise 3.65. Show that the set Pn \ V0 may be naturally identiļ¬ed with Pnā1 . So inductively, we can think of Pn as Pn = Rn āŖ Rnā1 āŖ Ā· Ā· Ā· āŖ R āŖ {0}, where P0 = {0} represents a single point. 3.7.2. Symmetries and isotropy groups. The symmetries of projective space are called projective transformations. Based on Deļ¬nition 3.64 of Pn as the set of lines through the origin in Rn+1 , the symmetry group of Pn is simply the group of linear transformations of Rn+1 , i.e, GL(n+1). Note that every element of GL(n + 1) maps any line through the origin to another line through the origin; thus the action of GL(n+1) on Rn+1 gives a well-deļ¬ned action on Pn . However, there is a slight catch. While it is true that all elements of GL(n + 1) are symmetries of Pn , some of them act trivially on Pn . A matrix
98
3. Homogeneous spaces
A ā GL(n + 1) ļ¬xes every line in Rn+1 āand therefore acts as the identity transformation on Pn āif and only if A = Ī»I for some Ī» = 0. Moreover, any two matrices A, B ā GL(n + 1) induce the same transformation on Pn if and only if A = Ī»B for some Ī» = 0. Therefore, the most natural choice for the symmetry group of Pn is the quotient group GL(n + 1)/ ā¼, where A ā¼ B if and only if A = Ī»B for some Ī» = 0. *Exercise 3.66. Show that GL(m)/ ā¼ is isomorphic to (1) SL(m) if m is odd; (2) a semidirect product (SL(m)/{Ā±I}) {Ā±1} if m is even. (The quotient of SL(m) by {Ā±I} reļ¬ects the fact that āI is an element of SL(m), and it acts trivially on Pn . The semidirect product with {Ā±1} reļ¬ects the fact that the sign of the determinant is ļ¬xed under scaling, and so this group has two components: one whose elements have determinant 1 and one whose elements have determinant ā1.) Remark 3.67. The group GL(m)/ ā¼ is called the projective general linear group, denoted P GL(m); similarly, the group SL(m)/ ā¼ is called the projective special linear group, denoted P SL(m). When m is odd, these two groups are identical and isomorphic to SL(m); when m is even, P GL(m) is the group (SL(m)/{Ā±I}) {Ā±1} of Exercise 3.66, while P SL(m) is the proper subgroup SL(m)/{Ā±I} of P GL(m). In order to keep things as simple as possible, we will generally take the symmetry group of Pn to be SL(n + 1) regardless of whether n is odd or even. In the odd case, this means that we will restrict to the identity component of the symmetry group (as we did with orientation-preserving transformations in the Euclidean case) and that our choices of frames on Pn will be determined only up to sign. Now, it is not at all obvious that this symmetry group has much to do with the idea of projective transformations as representing a change in perspective. The following exercise will illustrate how this works in the case of P1 . *Exercise 3.68. Consider the plane R2 with coordinates (ĀÆ x, yĀÆ), and let the x ĀÆ-axis represent the open set V0 = {[x0 : x1 ] ā P1 | x0 = 0} 1
via the identiļ¬cation x ĀÆ = xx0 . Suppose that an object lies along the line L with equation yĀÆ = mĀÆ x +b and that two viewers located at points p = (p1 , p2 )
3.7. Projective space
99
and q = (q 1 , q 2 ) see the object as if it were projected from their respective viewpoints onto the x ĀÆ-axis. (See Figure 3.3.) Observer p sees the point r as sq
p s
Q
Q Q
``` Q ``Q
`s`
Q ```` `
r QQ L Q
Q
Q s
Qs
x ĀÆ
T (ĀÆ x)
Figure 3.3. Projective transformation of P1
lying at (ĀÆ x, 0), and observer q sees r as lying at (T (ĀÆ x), 0). Use the following steps to compute the transformation T (ĀÆ x): (a) Show that the point r of intersection between L and the line joining the point (ĀÆ x, 0) to p has coordinates " r=
(p2 ā b)ĀÆ x + bp1 ĀÆ + bp2 mp2 x , mĀÆ x + p2 ā mp1 mĀÆ x + p2 ā mp1
# .
(b) Show that the line joining q to r intersects the x ĀÆ-axis at the point (T (ĀÆ x), 0), where (3.16)
T (ĀÆ x) =
Ī±ĀÆ x+Ī² Ī³x ĀÆ+Ī“
for some constants Ī±, Ī², Ī³, Ī“ with Ī±Ī“ ā Ī²Ī³ = 0. (You can keep up with the precise constants if you want, but theyāre kind of a mess!) The map (3.16) is called a linear fractional transformation. (c) For what value(s) of x ĀÆ would it make sense to deļ¬ne T (ĀÆ x) = ā? What value would you assign to T (ā)? Can you see how to interpret these assignments in terms of the observers in Figure 3.3? 1
(d) Recall that x ĀÆ = xx0 , where [x0 : x1 ] are homogeneous coordinates on P1 . Show that T corresponds to the map T ([x0 : x1 ]) = [Ī“x0 + Ī³x1 : Ī²x0 + Ī±x1 ]
100
3. Homogeneous spaces
on P1 , which in turn corresponds to the linear transformation x0 Ī“ Ī³ x0 T8 = Ī² Ī± x1 x1 Ī“ Ī³ 2 on R , where ā GL(2). Ī² Ī± Ī“ Ī³ (e) Show that the matrix can be modiļ¬ed so as to have determiĪ² Ī± nant equal to Ā±1 without changing the transformation T . Thus, T can be regarded as an element of the group P GL(2). *Exercise 3.69. Let V0 ā P2 be the open set V0 = {[x0 : x1 : x2 ] ā P2 | x0 = 0}, with aļ¬ne coordinates
" (ĀÆ x1 , x ĀÆ2 ) =
x1 x2 , x0 x0
# .
(a) Let
ā” 0 0 0ā¤ a0 a1 a2 ā¢ 1 1 1ā„ A = ā£a0 a1 a2 ā¦ ā SL(3). a20 a21 a22 Show that the transformation T ([x]) = [Ax] corresponds to the map 1
2
T (ĀÆ x ,x ĀÆ )=
"
ĀÆ1 + a12 x ĀÆ2 a20 + a21 x ĀÆ1 + a22 x ĀÆ2 a10 + a11 x , a00 + a01 x ĀÆ1 + a02 x ĀÆ2 a00 + a01 x ĀÆ1 + a02 x ĀÆ2
#
on V0 . (b) Show that if A has the form
ā”
1
0
0
ā¤
ā¢ ā„ A = ā£b1 a11 a12 ā¦ b2 a21 a22 with a11 a22 ā a12 a21 = 1, then T : V0 ā V0 is the equi-aļ¬ne transformation ĀÆ ĀÆx + b, T (ĀÆ x) = AĀÆ a11 a12 b1 ĀÆ = where AĀÆ = 2 2 , b . Therefore, the equi-aļ¬ne group A(2) is a a1 a2 b2 proper subgroup of the group of projective transformations.
3.7. Projective space
101
(c) (Maple recommended) Consider the image of the unit circle {(ĀÆ x1 , x ĀÆ2 ) ā V0 | (ĀÆ x1 )2 + (ĀÆ x2 )2 = 1} in V0 under the transformation corresponding to the matrix ā” ā¤ cos(Ļ) ā sin(Ļ) 0 ā¢ ā„ A = ā£ sin(Ļ) cos(Ļ) 0ā¦ . 0 0 1 Show that: (1) If 0 < Ļ < Ļ4 , then the image of the circle is an ellipse in V0 . (2) If Ļ = Ļ4 , then the image of the circle is a parabola in V0 . (3) If
Ļ 4
< Ļ ā¤ Ļ2 , then the image of the circle is a hyperbola in V0 .
Thus, projective transformations can transform any nondegenerate conic section into any other nondegenerate conic section. (However, they do preserve the family of nondegenerate conic sections.) Now, given a point [x] ā Pn , what is the isotropy group of [x]? *Exercise 3.70. Let [x0 ] = [1 : 0 : Ā· Ā· Ā· : 0] ā Pn . Show that: (a) The isotropy group H[x0 ] of [x0 ] in SL(n + 1) is ā§ā” ā« ā¤ ā1 r Ā· Ā· Ā· r (det A) āŖ āŖ 1 n āŖ āŖ āŖ āŖ ā„ āØā¢ ā¬ 0 ā„ ā¢ (3.17) H[x0 ] = ā¢ : A ā GL(n), r , . . . , r ā R . ā„ .. 1 n A āŖ āŖ ā¦ ā£ . āŖ āŖ āŖ āŖ ā© ā 0 (b) The isotropy group H[x] of any other point [x] ā Pn is H[x] = t[x] H[x0 ] tā1 [x] , where t[x] is any matrix in SL(n + 1) whose ļ¬rst column is [x]. (t[x] will then have the property that t[x] ([x0 ]) = [x].) (c) There is a natural correspondence between Pn and SL(n + 1)/H[x0 ] , the set of left cosets of H[x0 ] in SL(n + 1). 3.7.3. Projective frames and Maurer-Cartan forms. Projective frames on Pn are deļ¬ned as follows: n Deļ¬nition 3.71. A projective frame
f on P is a list of vectors f = (e0 , n+1 . . ., en ), where eĪ± ā R and det e0 Ā· Ā· Ā· en = 1. We identify [e0 ] with n the position vector [x] ā P , and (e1 , . . . , en ) may be regarded as a basis for
102
3. Homogeneous spaces
the tangent space T[e0 ] Pn . (We may also say that (e0 , . . . , en ) is a projective frame based at [e0 ].) Note that e0 needs to be part of the frame since it is not uniquely determined by its equivalence class [e0 ] and choosing a diļ¬erent representative for [e0 ] would aļ¬ect the scaling of the vectors (e1 , . . . , en ). The same reasoning as in the Euclidean case shows that the set of projective frames may be regarded as the group SL(n + 1) via the one-to-one correspondence
g(e0 , . . . , en ) = e0 Ā· Ā· Ā· en . We can deļ¬ne a projection map Ļ : SL(n + 1) ā Pn by Ļ(e0 , . . . , en ) = [e0 ]; the ļ¬ber of this map is the set of all projective frames based at [e0 ]. H[x0 ] acts freely and transitively on each ļ¬ber, and so this map gives an explicit description of SL(n+1) as a principal bundle over Pn with ļ¬ber group H[x0 ] : H[x0 ] - SL(n + 1) Ļ
?
Pn ā¼ = SL(n + 1)/H[x0 ] . In this context, SL(n + 1) is also called the projective frame bundle of Pn , and it is denoted F (Pn ). The Maurer-Cartan forms on SL(n + 1) are deļ¬ned by the equations (3.18)
deĪ± = eĪ² ĻĪ±Ī² ,
where 0 ā¤ Ī±, Ī² ā¤ n. The structure equations are (3.19)
dĻĪ±Ī² = āĻĪ³Ī² ā§ ĻĪ±Ī³ ,
where the sum is over 0 ā¤ Ī³ ā¤ n. Since the Maurer-Cartan form Ļ = [ĻĪ±Ī² ] takes values in the Lie algebra sl(n+1), the only relation among the 1-forms (ĻĪ±Ī² ) is the trace condition ĻĪ±Ī± = 0. As for which forms are semi-basic for the projection Ļ : SL(n + 1) ā Pn and which should be regarded as connection forms, recall that the semi-basic forms are those that pull back to each ļ¬ber of Ļ to be zero, i.e., those with the property that ĻĪ±Ī² (v) = 0 whenever d[e0 ](v) = 0. But d[e0 ](v) = 0 if and
3.8. Maple computations
103
only if de0 (v) is a multiple of e0 . Therefore, d[e0 ](v) = 0 if and only if Ļ01 (v) = Ā· Ā· Ā· = Ļ0n (v) = 0, and so the semi-basic forms for Ļ are (Ļ01 , . . . , Ļ0n ). The remaining (ĻĪ±Ī² ) are the connection forms. Exercise 3.72. Repeat Exercise 3.23 for P3 .
3.8. Maple computations In this section, we will explore how the Cartan package may be used to facilitate some of the computations involved in the exercises in this chapter. While most of these computations are simple enough to be done by hand, seeing how to do them in Maple will help you to familiarize yourself with how the package works. You should begin by loading the Cartan package into Maple, and the LinearAlgebra package will be useful as well: > with(Cartan); > with(LinearAlgebra); Exercise 3.32: Deļ¬ne A to be the matrix whose columns are the frame ļ¬eld vectors: > A:= Matrix([ [cos(phi)*cos(theta), -sin(theta), -sin(phi)*cos(theta)], [cos(phi)*sin(theta), cos(theta), -sin(phi)*sin(theta)], [sin(phi),0,cos(phi)] ]); Let dx be the vector of the Cartesian coordinate 1-forms: > dx:= Vector([d(x), d(y), d(z)]); We know that the dual forms (ĀÆ Ļ1, Ļ ĀÆ 2, Ļ ĀÆ 3 ) are the components of the following vector: > dualforms:= simplify(MatrixInverse(A).dx); We can assign them as follows: > for i from 1 to 3 do omega[i]:= dualforms[i]; end do;
104
3. Homogeneous spaces
Similarly, the connection forms (ĀÆ Ļji ) are given by the following matrix: > connectionforms:= simplify(MatrixInverse(A).map(d, A)); Assign these as follows: > for i from 1 to 3 do for j from 1 to 3 do omega[i,j]:= connectionforms[i,j]; end do; end do; In order to verify the structure equations for the (dĀÆ Ļ i ), check that the following quantities are all equal to zero: > for i from 1 to 3 do Simf(d(omega[i]) + add(omega[i,j] &Ė omega[j], j=1..3)); end do; In order to verify the structure equations for the (dĀÆ Ļji ), check that the following quantities are all equal to zero. (The print command is to force Maple to print the output; normally output is suppressed for computations nested this deeply.) > for i from 1 to 3 do for j from 1 to 3 do print(Simf(d(omega[i,j]) + add(omega[i,k] &Ė omega[k,j], k=1..3))); end do; end do; The following two exercises donāt require the Cartan package, but they nicely illustrate some of Mapleās graphic capabilities, so we will include them here anyway. First, we need to load the plots package: > with(plots); Exercise 3.46, part (c): The general element of SO+ (1, 1) is represented by the following matrix: > A:= Matrix([[cosh(theta), sinh(theta)], [sinh(theta), cosh(theta)]]); Consider the unit circle, parametrized as follows: > unitcircle:= Vector([cos(t), sin(t)]);
3.8. Maple computations
105
The image of this curve under the action of the matrix A (for a ļ¬xed value of Īø) is given by > newcurve:= A.unitcircle; We can now deļ¬ne plots of this curve for various values of Īø; by using the display command, we can view them all on one graph: > g1:= plot([subs([theta=0], newcurve[1]), subs([theta=0], newcurve[2]), t=0..2*Pi], scaling=constrained): g2:= plot([subs([theta=0.5], newcurve[1]), subs([theta=0.5], newcurve[2]), t=0..2*Pi], scaling=constrained): g3:= plot([subs([theta=1], newcurve[1]), subs([theta=1], newcurve[2]), t=0..2*Pi], scaling=constrained): display(g1, g2, g3);
Exercise 3.69, part (c): Deļ¬ne A to be the transformation matrix: > A:= Matrix([[cos(phi), -sin(phi), 0], [sin(phi), cos(phi), 0], [0,0,1]]); Parametrize the unit circle in terms of homogeneous coordinates with x0 = 1: -> x homog:= Vector([1, cos(t), sin(t)]); Apply the transformation: > Tx homog:= A.x homog; If we write the transformed curve in terms of homogeneous coordinates with x0 = 1, then the other two coordinates are given by > Tx[1]:= Tx homog[2]/Tx homog[1]; Tx[2]:= Tx homog[3]/Tx homog[1];
106
3. Homogeneous spaces
We can animate the resulting curve, using Ļ as the time parameter, to see how the unit circle transforms through a family of quadrics and back again as Ļ varies from 0 to Ļ: > animate(plot, [ [Tx[1], Tx[2], t=0..2*Pi], view = [-5..5, -5..5], scaling=constrained], phi = 0..Pi, frames=100); In order to view the animation, click on the plot, and then from the āPlotā menu choose āAnimation ā Playā. (Alternatively, right-click on the plot and select āAnimation ā Playā from the pop-up menu.)
10.1090/gsm/178/04
Chapter 4
Curves and surfaces in Euclidean space
4.1. Introduction In this chapter, we get to the heart of the matter: Cartanās method of moving frames. This method is used to study the geometry of submanifolds of homogeneous spaces; in this chapter, we will see how it applies to curves and surfaces in E3 . The main idea goes something like this: By associating a frame to each point of a submanifold in some geometrically natural way and then studying how the frame varies along the submanifold, we can construct a complete set of invariants for a given class of submanifolds. Invariants are quantities associated to a submanifold (such as curvature and torsion for curves in E3 ) that remain unchanged when the submanifold is acted on by an element of the symmetry group of the homogeneous space. A complete set of invariants contains enough information to determine a submanifold uniquely up to the group action. This perspective naturally leads to two questions:
(1) How can we tell when we have found a complete set of invariants? This is a question about uniqueness: Given two submanifolds of a homogeneous space, when is it possible to transform one into the other via an element of the symmetry group? This is also known as the equivalence problem: When are two submanifolds equivalent under the action of the symmetry group?
107
108
4. Curves and surfaces in Euclidean space
(2) Once we know what a complete set of invariants should look like, can they be prescribed arbitrarily? This is a question about existence: Given prescribed values for the invariants, does there necessarily exist a submanifold whose invariants coincide with the given values? In Ā§4.2, we will address the theory underlying the ļ¬rst question, and in Ā§4.3 we will show how it applies to curves in E3 . Then in Ā§4.4 and Ā§4.5, we will take up the second question and show how the theory as a whole applies to surfaces in E3 .
4.2. Equivalence of submanifolds of a homogeneous space We will approach the equivalence problem for submanifolds of a homogeneous space G/H by considering the restriction of certain frames on the underlying space G/H to the submanifold in question. Remark 4.1. If M ā G/H and f : U ā G/H is an immersion with f (U ) = M (typically U is some open, connected, and simply connected region in Rn and f is a parametrization of M ), then ārestriction to M ā really means pullback to U . The pullback bundle (or induced bundle) f ā G of the principal bundle Ļ : G ā G/H is the bundle over U whose ļ¬ber over a point u ā U is just the ļ¬ber of G over the point f (u) ā G/H: f ā G = {(u, f ) ā U Ć G | f (u) = Ļ(f )}. The bundle f ā G is a principal bundle over U , with ļ¬ber group H. There is a natural map fĖ : f ā G ā G deļ¬ned by fĖ(u, f ) = f . When Ļ : G ā G/H is regarded as a frame bundle, the image fĖ(u, f ) of any element (u, f ) ā f ā G may be thought of as a frame based at the point f (u) ā G/H. These maps may be represented by the following commutative diagram: f āG
fĖ
-G Ļ
Ļ ĀÆ
?
U
f
?
- G/H.
We will generally be interested in choosing a frame at each point of M ā G/Hāi.e., a āframe ļ¬eldā on M āaccording to certain geometric considerations. Technically, this means choosing a section of the bundle f ā G over U ,
4.2. Equivalence of submanifolds of a homogeneous space
109
but we will usually regard it as choosing a lifting fĖ : U ā G, i.e., a function fĖ with the property that for any u ā U , (Ļ ā¦ fĖ)(u) = f (u) ā M ā G/H. In other words, we choose fĖ so that the following diagram commutes: > fĖ
U
f
G Ļ
?
- G/H.
When choosing a lifting fĖ : U ā G, we will want to choose frame ļ¬elds that are adapted to M . This means that, instead of just choosing arbitrary frame ļ¬elds, we will use the geometry of M to choose āniceā frame ļ¬elds. This is somewhat analogous to choosing āniceā coordinates on a neighborhood of a point on a surface to study the geometry at that point; the beauty of the method of moving frames is that we can do this at all points simultaneously. Once we have chosen a nice lifting (called an adapted frame ļ¬eld, or sometimes simply an adapted frame) fĖ : U ā G, we can consider the pullback fĖā Ļ of the Maurer-Cartan form Ļ of G and its structure equations to U . The pulled-back Maurer-Cartan form fĖā Ļ will generally contain quantities that are invariants of M : If we act on M by a symmetry of the ambient space G/H, then these quantities remain unchanged. (The invariance of fĖā Ļ under such an action follows from the left-invariance of Ļ under action by an element of G.) Typical examples of invariants are quantities such as arc length, curvature, etc. In order for the adapted frame ļ¬eld fĖ : U ā G to contain useful information about the invariants of M , the algorithm for choosing fĖ should be completely determined in some canonical way by the geometry of M . Moreover, the adapted frame ļ¬eld itself should be equivariant; this means that (g Ā· f ) = g Ā· fĖ for any g ā G. If such an equivariant adapted frame ļ¬eld exists, then the question of equivalence is completely answered by the following important lemma: Lemma 4.2. Let U ā Rn be a connected, open set, and let fĖ1 , fĖ2 : U ā G be two immersions. Then there exists an element g ā G such that fĖ1 (u) = g Ā· fĖ2 (u) for all u ā U if and only if fĖ1ā Ļ = fĖ2ā Ļ, where Ļ is the Maurer-Cartan form of G.
110
4. Curves and surfaces in Euclidean space
Proof. First, observe that for any map fĖ : U ā G, we have fĖā Ļ = fĖā1 dfĖ.
(4.1)
Remark 4.3. What does equation (4.1) really mean? Recall that for any g ā G, Ļ is a linear map from Tg G to Te G = g deļ¬ned by Ļ(w) = (Lgā1 )ā (w) for w ā Tg G. Now, if fĖ : U ā G is a diļ¬erentiable map, then fĖā Ļ is a linear map from Tu U to g deļ¬ned by ā1 Ė Ė fĖā Ļ(v) = Ļ(fĖā (v)) = (L(f(u)) Ā· dfĖ(v) ā1 )ā (fā (v)) = (f (u)) Ė
for v ā Tu U . Therefore,
fĖā Ļ = fĖā1 dfĖ.
Now, given fĖ1 , fĖ2 : U ā G, there exists a unique function g : U ā G satisfying the condition that fĖ2 (u) = g(u)fĖ1 (u)
(4.2)
for all u ā U āspeciļ¬cally, g(u) = fĖ2 (u)(fĖ1 (u))ā1 . Diļ¬erentiating (4.2) yields dfĖ2 = dg fĖ1 + g dfĖ1 ; therefore, fĖ2ā Ļ = fĖ2ā1 dfĖ2 = fĖā1 dg fĖ1 + fĖā1 g dfĖ1 = = = =
2 fĖ2ā1 dg fĖ1 fĖ2ā1 dg fĖ1 fĖ2ā1 dg fĖ1 fĖ2ā1 dg fĖ1
2
+ (g fĖ1 )ā1 g dfĖ1 + fĖā1 g ā1 g dfĖ1 1
+ fĖ1ā1 dfĖ1 + fĖā Ļ. 1
It follows that fĖ1ā Ļ = fĖ2ā Ļ if and only if dg = 0, i.e., if and only if g(u) is constant. This lemma is more powerful than it looks; it says that: (1) Whatever geometric information is contained in fĖā Ļ remains unchanged when M is transformed by a symmetry g of the ambient homogeneous space G/H. (2) Conversely, fĖā Ļ contains enough information about the geometry of M to completely determine it up to a symmetry of the ambient space.
4.3. Moving frames for curves in E3
111
So, our approach from here on will go something like this: Given an immersion f : U ā G/H, we will look for an equivariant method of constructing a canonical adapted frame ļ¬eld fĖ : U ā G. Then we will examine the pulled-back Maurer-Cartan form fĖā Ļ, which will contain a complete set of geometric invariants for the original immersion f . This is known as the method of moving frames, and we will start by demonstrating how to carry it out for curves in E3 .
4.3. Moving frames for curves in E3 Consider a smooth, parametrized curve Ī± : I ā E3 that maps some open interval I ā R into Euclidean space. E3 has the structure of the homogeneous space E(3)/SO(3), so an adapted frame ļ¬eld along Ī± should be a lifting Ī± Ė : I ā E(3). Any such lifting can be written as Ī± Ė (t) = (Ī±(t); e1 (t), e2 (t), e3 (t)), where for each t ā I, (e1 (t), e2 (t), e3 (t)) is an oriented, orthonormal basis for the tangent space TĪ±(t) E3 . Such an adapted frame ļ¬eld is usually called an orthonormal frame ļ¬eld along Ī±. If the curve is ānice enoughā (the precise meaning of this will become clear shortly), then we will be able to choose such a frame ļ¬eld in a canonical way, based on the geometry of the curve. Remark 4.4. While the orthonormal frame ļ¬eld is technically the image of the map Ī± Ė and so includes the position vector Ī±(t) at each point, it is common to refer to the triple of vector ļ¬elds (e1 (t), e2 (t), e3 (t)) as an āorthonormal frame ļ¬eld along Ī±ā. Hopefully this terminology will not cause any confusion. Recall that Ī± is regular if Ī± (t) = 0 for every t ā I. The ļ¬rst condition that we will require in order for Ī± to be ānice enoughā is that Ī± must be a regular curve. With this assumption, we can make our ļ¬rst frame adaptation by setting Ī± (t) e1 (t) = ; |Ī± (t)| i.e., we require that e1 (t) be the unit tangent vector to the curve at Ī±(t). Exercise 4.5. Show that this choice of e1 (t) is equivariant under the action of E(3): If we replace Ī± by g Ā· Ī± for some g ā E(3), then e1 (t) ā TĪ±(t) E3 will be replaced by (Lg )ā (e1 (t)) ā TgĀ·Ī±(t) E3 . The vector e1 (t) is now uniquely determined, but we still have the freedom to vary the pair (e2 (t), e3 (t)) by an arbitrary rotation in SO(2). We will
112
4. Curves and surfaces in Euclidean space
need to delve deeper into the geometry of the curve Ī± in order to determine how to choose the remainder of the adapted frame ļ¬eld. Here we make an observation that will simplify the remainder of our computations. Fix t0 ā I and deļ¬ne the arc length function along Ī± to be & t s(t) = |Ī± (u)| du. t0
Exercise 4.6. Show that s(t) is invariant under the action of E(3); that is, for any g ā E(3), the curves Ī± and g Ā· Ī± have the same arc length function. Since Ī± (t) = 0 for all t ā I, the inverse function theorem implies that s(t) has a diļ¬erentiable inverse function t(s). By setting Ī±(s) = Ī±(t(s)), we may assume that Ī± is parametrized by arc length, so that |Ī± (s)| = 1 and e1 (s) = Ī± (s). In order to make the next adaptation, we need to make another assumption about the curve. We will say that Ī± is nondegenerate if Ī± is regular and, in addition, e1 (s) = 0 for all s ā I. In this case, diļ¬erentiating the equation e1 (s), e1 (s) = 1 with respect to s yields
e1 (s), e1 (s) = 0.
Thus, e1 (s) is orthogonal to e1 (s), and we can make our second adaptation by setting e (s) e2 (s) = 1 . |e1 (s)| This vector is called the unit normal vector to the curve at Ī±(s). Exercise 4.7. Show that e2 (s) is equivariant under the action of E(3): If we replace Ī± by g Ā·Ī± for some g ā E(3), then e2 (s) ā TĪ±(s) E3 will be replaced by (Lg )ā (e2 (s)) ā TgĀ·Ī±(s) E3 . The adapted frame ļ¬eld is now uniquely determined: Because the frame must be oriented and orthonormal, e3 (s) is uniquely determined by the condition that e3 (s) = e1 (s) Ć e2 (s). The vector e3 (s) is called the binormal vector to the curve at Ī±(s). The adapted frame ļ¬eld (e1 (s), e2 (s), e3 (s)) is called the Frenet frame of the curve Ī±(s); it determines a canonical, left-invariant lifting Ī± Ė : I ā E(3) given by Ī± Ė (s) = (Ī±(s); e1 (s), e2 (s), e3 (s)) for any nondegenerate curve Ī± parametrized by arc length. (See Figure 4.1.)
4.3. Moving frames for curves in E3
113
Figure 4.1. Frenet frame at a point of a curve in E3
Now consider the pullbacks of equations (3.1) to I via Ī± Ė . We have Ī± Ė ā (x) = Ī±(s),
Ī± Ė ā (ei ) = ei (s).
Therefore, Ī± Ė ā (dx) = d(Ī± Ė ā (x)) = Ī± (s)ds, Ī± Ė ā (dei ) = d(Ī± Ė ā (ei )) = ei (s)ds. As in Chapter 3, write (ĀÆ Ļi, Ļ ĀÆ ji ) for the pulled-back forms (Ī± ĖāĻi, Ī± Ė ā Ļji ). Note that, since these are all 1-forms on I, they must all be multiples of ds. We can write the pullbacks of equations (3.1) as (4.3)
Ī± (s)ds = ei (s) Ļ ĀÆ i, ei (s)ds = ej (s) Ļ ĀÆ ij .
Now recall how we constructed our adapted frame ļ¬eld. First, we chose e1 (s) so that Ī± (s) = e1 (s); therefore, the ļ¬rst equation in (4.3) implies that Ļ ĀÆ 1 = ds,
Ļ ĀÆ2 = Ļ ĀÆ 3 = 0.
Then we chose e2 (s) so that e1 (s) is a multiple of e2 (s), say e1 (s) = Īŗ(s)e2 (s). The function Īŗ(s) is called the curvature of Ī± at s; note that Ī± is nondegenerate if and only if Īŗ(s) > 0 for all s ā I. So the equation for e1 (s) in (4.3) implies that Ļ ĀÆ 12 = Īŗ(s)ds,
Ļ ĀÆ 13 = 0.
(Recall that Ļ ĀÆ 11 = 0 by the skew-symmetry of the (Ļji ).) The only remaining Maurer-Cartan form is Ļ ĀÆ 32 ; it must be equal to some multiple of ds, so deļ¬ne a function Ļ (s) by the condition that Ļ ĀÆ 32 = āĻ (s)ds. The function Ļ (s) is called the torsion of Ī± at s.
114
4. Curves and surfaces in Euclidean space
Remark 4.8. The minus sign in the deļ¬nition of Ļ (s) is a convention preferred by some authors, but it is not universal. This choice of sign has the feature that it results in a positive value of Ļ for the standard right-handed helix Ī±(s) = t [a cos(s), a sin(s), b] , where a, b > 0 and a2 + b2 = 1. Using the skew-symmetry of the (ĀÆ Ļji ), the remaining two equations in (4.3) become e2 (s)ds = e1 (s) Ļ ĀÆ 21 + e3 (s) Ļ ĀÆ 23 = (āe1 (s)Īŗ(s) + e3 (s)Ļ (s))ds, e3 (s)ds = e1 (s) Ļ ĀÆ 31 + e2 (s) Ļ ĀÆ 32 = āe2 (s)Ļ (s)ds. Thus, we have the familiar Frenet equations:
ā” ā¤ 0 0 0 0 ā¢ ā„
ā¢1 0 āĪŗ(s) 0 ā„ ā¢ ā„. Ī± (s) e1 (s) e2 (s) e3 (s) = Ī±(s) e1 (s) e2 (s) e3 (s) ā¢ ā„ 0 Īŗ(s) 0 āĻ (s) ā£ ā¦ 0 0 Ļ (s) 0
Note that if we regard Ī±(s) Ė as the matrix
Ī± Ė (s) = Ī±(s) e1 (s) e2 (s) e3 (s) , then the matrix on the right multiplied by the 1-form ds is equal to Ī± Ė (s)ā1 d(Ī± Ė (s)), and so it is exactly the pullback of the Maurer-Cartan form Ļ = g ā1 dg on E(3) via Ī± Ė. Applying Lemma 4.2 yields the following theorem: Theorem 4.9. Two nondegenerate curves Ī±1 , Ī±2 : I ā E3 parametrized by arc length diļ¬er by a rigid motion if and only if they have the same curvature Īŗ(s) and torsion Ļ (s). This is the uniqueness portion of the fundamental theorem of space curves (cf. Theorem 3.1). We will address the existence portion in Ā§4.4. Exercise 4.10. Repeat the analysis of this section for curves in E4 . Here are some things to think about along the way: ā¢ Is there a natural choice of parametrization for the curve? ā¢ How should you choose the vectors (e1 (s), e2 (s), e3 (s), e4 (s)) of the frame ļ¬eld? (And how do you ensure that these vectors form an orthonormal frame ļ¬eld?) Prove that your choice is equivariant
4.4. Compatibility conditions
115
under the action of E(4). (Hint: The tricky part is how to choose e3 (s) so that it is orthogonal to both e1 (s) and e2 (s). For guidance, use the Frenet equations to convince yourself that for curves in E3 , e3 (s) =
e2 (s) ā e2 (s), e1 (s)e1 (s) . |e2 (s) ā e2 (s), e1 (s)e1 (s)|
In other words, e3 (s) is obtained by taking the orthogonal projection of e2 (s) onto the orthogonal complement of e1 (s) and e2 (s) and then normalizing it to have unit length.) ā¢ What is the right deļ¬nition of ānondegenerateā for curves in E4 ? ā¢ Where do invariants appear in the pullbacks of equations (3.1)? What can you conclude from the skew-symmetry of the connection forms? ā¢ What is the 4-dimensional analog of the Frenet equations? ā¢ How do you think the analysis would go for curves in En ?
4.4. Compatibility conditions and existence of submanifolds with prescribed invariants In Ā§4.3, we saw that a curve Ī± : I ā E3 parametrized by arc length s is completely determined up to rigid motions of E3 by its curvature Īŗ(s) and torsion Ļ (s). We may express this by saying that the curvature and torsion form a complete set of invariants for curves in E3 . In general, Lemma 4.2 tells us when we have found a complete set of invariants for a āniceā immersion f : U ā G/H: Assuming that we can ļ¬nd a canonical, equivariant way of choosing a lifting fĖ : U ā G (this is what āniceā means), a complete set of invariants is contained in fĖā Ļ, the pullback via fĖ of the Maurer-Cartan form Ļ of G. For curves in E3 , it is now natural to ask whether the functions Īŗ(s) and Ļ (s) may be prescribed arbitrarily. In other words, given arbitrary functions Īŗ(s), Ļ (s) with Īŗ(s) > 0, does there necessarily exist a curve Ī± : I ā E3 that is parametrized by arc length and has curvature Īŗ(s) and torsion Ļ (s)? Exercise 4.11. Why must we require Īŗ(s) > 0? The answer to this existence question is yes, but this result is particular to 1-dimensional submanifolds of homogeneous spaces G/H. It follows from
116
4. Curves and surfaces in Euclidean space
the following lemma: Lemma 4.12. Let G be a Lie group with Lie algebra g, and suppose that Ļ ĀÆ is a g-valued 1-form on a connected and simply connected manifold U . Then there exists a smooth map fĖ : U ā G with fĖā Ļ = Ļ ĀÆ if and only if Ļ ĀÆ satisļ¬es the Maurer-Cartan equation (4.4)
dĀÆ Ļ = āĀÆ Ļā§Ļ ĀÆ.
Outline of Proof. The full proof of this lemma requires the Frobenius theorem and is beyond the scope of this book. (If youāre curious, the proof may be found in [Gri74].) However, the main idea goes something like this: fĖā Ļ contains quantities involving derivatives of the unknown function fĖ : U ā G, and for any given g-valued 1-form Ļ ĀÆ on U , the equation fĖā Ļ = Ļ ĀÆ may be regarded as a system of partial diļ¬erential equations for fĖ. In general, this system is overdetermined and may have no solutions. However, equation (4.4) is precisely the compatibility condition that must be satisļ¬ed in order to guarantee that solutions exist, at least locally. (In this case, it turns out that a solution is uniquely determined by specifying an initial condition fĖ(u0 ) for any u0 ā U .) Once we know that local solutions exist, a patching argument can be used to construct a solution fĖ on the entire domain U . Remark 4.13. Even without the hypothesis that U is simply connected, the result of Lemma 4.12 holds in some neighborhood of any point u ā U ; simple connectivity is only necessary to ensure that these local solutions can be patched together to form a single solution that is globally deļ¬ned on U . For simplicity of exposition, we will not explicitly state topological hypotheses on the domain U every time we introduce an immersion f : U ā G/H. But keep in mind that if U is topologically nontrivial, then many of our constructions may be possible only locally and not globally on U . For example, because a frame bundle over a topologically nontrivial base space may have no global sections, it might not be possible to construct an adapted frame ļ¬eld globally on U . Because of these limitations, the method of moving frames is a tool best suited to the study of the local geometry of submanifolds of homogeneous spaces; it has very little to say about global properties. Assuming that the conditions of Lemma 4.12 are satisļ¬ed, composing the map fĖ with the natural projection Ļ : G ā G/H gives a smooth map f : U ā G/H that, in most cases of interest, realizes M = f (U ) as a submanifold of the homogeneous space G/H. According to Lemma 4.2, specifying a g-valued 1-form Ļ ĀÆ on U is equivalent to prescribing the values of a complete set of invariants for an unknown submanifold M ā G/H;
4.5. Moving frames for surfaces in E3
117
Lemma 4.12 then gives a necessary and suļ¬cient condition for the existence of a smooth map f : U ā G/H whose image has the prescribed invariants. Moreover, Lemma 4.2 implies that any such f is unique up to left action by an element g ā G. *Exercise 4.14. Let (Ļ 1 , . . . , Ļ n ) be a basis for the left-invariant 1-forms on G that are semi-basic for the projection Ļ : G ā G/H, and let (ĀÆ Ļ1, . . . , Ļ ĀÆ n) be the corresponding components of the g-valued 1-form Ļ ĀÆ on U . Show that the map f = Ļ ā¦ fĖ of Lemma 4.12 is an immersion if and only if (ĀÆ Ļ1, . . . , Ļ ĀÆ n) ā span the cotangent space Tu U at every point u ā U . (Note that typically the dimension of U is less than n, so the forms (ĀÆ Ļ1, . . . , Ļ ĀÆ n ) will generally not be linearly independent.) Corollary 4.15. Let I ā R be an open interval, and let Īŗ, Ļ : I ā R be any diļ¬erentiable functions satisfying Īŗ(s) > 0 for all s ā I. Then there exists a nondegenerate curve Ī± : I ā E3 , parametrized by arc length, with curvature Īŗ(s) and torsion Ļ (s). *Exercise 4.16. Show how Corollary 4.15 follows from Lemma 4.12. (Hint: Observe that both sides of equation (4.4) are 2-forms on I.) Corollary 4.15 applies more generally to curves in any homogeneous space G/H: Once we know how to construct equivariant frame ļ¬elds and ļ¬nd a complete set of invariants, Lemma 4.12 implies that these invariants may be prescribed arbitrarily. But for surfaces (and generally for submanifolds of any dimension greater than one), equation (4.4) will give compatibility conditions that a prescribed set of invariants must satisfy in order for an immersed submanifold with the given invariants to exist.
4.5. Moving frames for surfaces in E3 Let U be an open set in R2 (assumed here and throughout the remainder of the book to be connected and simply connected; cf. Remark 4.13), and let x : U ā E3 be an immersion whose image is a regular surface Ī£ = x(U ). Ė : Just as for curves, an adapted frame ļ¬eld along Ī£ should be a lifting x U ā E(3) of the form Ė (u) = (x(u); e1 (u), e2 (u), e3 (u)) , x where for each u ā U , (e1 (u), e2 (u), e3 (u)) is an oriented, orthonormal basis for the tangent space Tx(u) E3 . Since x is an immersion, there is a well-deļ¬ned tangent plane Tx(u) Ī£ for each point x(u) ā Ī£. Thus, we can make our ļ¬rst frame adaptation by requiring
118
4. Curves and surfaces in Euclidean space
that e3 (u) be orthogonal to Tx(u) Ī£. This determines e3 (u) uniquely up to sign, and an orthonormal frame ļ¬eld satisfying this condition will be called adapted. Exercise 4.17. Show that this choice of e3 (u) is equivariant (up to sign) under the action of E(3). Having chosen e3 (u) in this way, e1 (u) and e2 (u) must form a basis for Tx(u) Ī£ no matter how we choose them. We will explore how we might reļ¬ne our choices later, but for now, we allow (e1 (u), e2 (u)) to be an arbitrary orthonormal basis of Tx(u) Ī£. Ė . (As in Chapter Now consider the pullbacks of equations (3.1) to U via x Ė ā Ļji ) on U .) Our ļ¬rst 3, write (ĀÆ Ļi, Ļ ĀÆ ji ) for the pulled-back forms (Ė xā Ļ i , x observation about these forms is the following: Proposition 4.18. Let U ā R2 be an open set, and let x : U ā E3 be an immersion. For any adapted orthonormal frame ļ¬eld (e1 (u), e2 (u), e3 (u)) along Ī£ = x(U ), the associated dual and connection forms (ĀÆ Ļi, Ļ ĀÆ ji ) have the property that Ļ ĀÆ 3 = 0. Proof. The pullback of the ļ¬rst equation in (3.1) is dx = ei Ļ ĀÆ i. Let u ā U . Then dxu is a linear map from Tu U to Tx(u) Ī£, and so for any v ā Tu U , we must have dxu (v) = ei (u)ĀÆ Ļ i (v) ā Tx(u) Ī£. Since Tx(u) Ī£ is spanned by e1 (u) and e2 (u), the e3 (u) term in this sum must vanish; therefore, Ļ ĀÆ 3 (v) = 0. And since v ā Tu U is arbitrary, it follows that Ļ ĀÆ 3 = 0. *Exercise 4.19. Show that (ĀÆ Ļ1, Ļ ĀÆ 2 ) are linearly independent 1-forms on U . Therefore, they form a basis for the 1-forms on U . (Hint: Evaluate Ļ ĀÆ1 and Ļ ĀÆ 2 on vectors v1 , v2 ā Tu U with the property that dx(vi ) = ei (u) for i = 1, 2.) You may recall that the metric properties of a regular surface in E3 are encapsulated in the ļ¬rst fundamental form of the surface. Deļ¬nition 4.20. Let U ā R2 be an open set, and let x : U ā E3 be an immersion. The ļ¬rst fundamental form of Ī£ = x(U ) is the quadratic form I on T U deļ¬ned by I(v) = dx(v), dx(v) for v ā Tu U .
4.5. Moving frames for surfaces in E3
119
In other words, I is just the restriction of the Euclidean metric on E3 to vectors which are tangent to Ī£. Its primary function is to describe how to compute this metric in terms of the local coordinates u on Ī£ that are given by the parametrization x : U ā E3 . *Exercise 4.21. Show that for any v ā Tu U , 1 2 2 2 I(v) = Ļ ĀÆ (v) + Ļ ĀÆ (v) . This is often written more concisely as I = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 , and each term in the sum should be interpreted as the symmetric product Ļ ĀÆi ā¦ Ļ ĀÆ i. While the ļ¬rst fundamental form is deļ¬ned as a function of a single tangent vector, it can be used to deļ¬ne an inner product Ā·, Ā·u on each tangent space Tu U through a process called polarization. Deļ¬nition 4.22. The inner product Ā·, Ā·u is deļ¬ned by v, wu =
1 4
(I(v + w) ā I(v ā w))
for v, w ā Tu U . *Exercise 4.23. (a) Show that v, wu = Ļ ĀÆ 1 (v) Ļ ĀÆ 1 (w) + Ļ ĀÆ 2 (v) Ļ ĀÆ 2 (w). (b) Convince yourself that Ā·, Ā·u is a section of the symmetric tensor bundle S 2 (T ā U ). Any section of this bundle deļ¬nes a symmetric bilinear form B : T U Ć T U ā R, which in turn deļ¬nes a quadratic form Q : TU ā R by setting Q(v) = B(v, v). *Exercise 4.24. If youāve seen the ļ¬rst fundamental form before, you probably saw it written as I = E du2 + 2F du dv + G dv 2 , where (u, v) are local coordinates on U and E = xu , xu ,
F = xu , xv ,
G = xv , xv .
Suppose that x : U ā E3 is an immersion with F = 0. (Such a parametrization for a given surface Ī£ always exists, at least locally; the proof is beyond the scope of this book but can be found in [dC76]. This assumption isnāt necessary, but it keeps the calculations simpler.)
120
4. Curves and surfaces in Euclidean space
(a) Show that the frame ļ¬eld 1 1 e1 (u) = ā xu , e2 (u) = ā xv , e3 (u) = e1 (u) Ć e2 (u) E G is an oriented, orthonormal frame ļ¬eld along Ī£ = x(U ), with e3 (u) orthogonal to Tx(u) Ī£. (b) Show that the dual forms of this frame ļ¬eld are ā ā Ļ ĀÆ 1 = E du, Ļ ĀÆ 2 = G dv, Ļ ĀÆ3 = 0 and that I = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 = E du2 + G dv 2 . (c) Show that if (e1 (u), e2 (u), e3 (u)) is replaced by another adapted frame Ė2 (u), e Ė3 (u)) of the form ļ¬eld (Ė e1 (u), e Ė1 = cos(Īø) e1 + sin(Īø) e2 , e Ė2 = ā sin(Īø) e1 + cos(Īø) e2 , e Ė 3 = e3 , e ĖĀÆ 1 , Ļ ĖĀÆ 2 ) of the new adapted frame ļ¬eld are then the dual forms (Ļ ĖĀÆ 1 = cos(Īø) Ļ Ļ ĀÆ 1 + sin(Īø) Ļ ĀÆ 2, Ļ ĀÆĖ 2 = ā sin(Īø) Ļ ĀÆ 1 + cos(Īø) Ļ ĀÆ 2. Moreover, ĖĀÆ 1 )2 + (Ļ ĖĀÆ 2 )2 . I = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 = (Ļ Now letās see what we can learn by diļ¬erentiating! Since Ļ ĀÆ 3 = 0, we must have dĀÆ Ļ 3 = 0 as well. According to the Cartan structure equations (3.8), this implies that dĀÆ Ļ 3 = āĀÆ Ļ13 ā§ Ļ ĀÆ1 ā Ļ ĀÆ 23 ā§ Ļ ĀÆ 2 = 0. Since (ĀÆ Ļ1, Ļ ĀÆ 2 ) are linearly independent 1-forms, Cartanās lemma (cf. Lemma 2.49) implies that there exist real-valued functions h11 , h12 , h22 on U such that 3 1 Ļ ĀÆ1 h11 h12 Ļ ĀÆ = . Ļ ĀÆ 23 h12 h22 Ļ ĀÆ2 How should we interpret the functions (hij )? Recall that de3 = e1 Ļ31 + e2 Ļ32 = ā(e1 Ļ13 + e2 Ļ23 ). For any tangent vector w ā Tx Ī£, de3 (w) measures the directional derivative of the normal vector ļ¬eld e3 in the direction of w. So, up to sign, Ļ13 (w) measures the e1 component of this directional derivative, and Ļ23 (w) measures its e2 component. In other words, Ļi3 (w) measures how rapidly e3
4.5. Moving frames for surfaces in E3
121
rotates towards ei if we move in the direction w. When we pull everything back to U via the parametrization x and express Ļ ĀÆ i3 as a linear combination 1 2 of Ļ ĀÆ and Ļ ĀÆ , we see that hij measures how rapidly e3 rotates towards ei if we move in the direction ej . Recall that, in addition to the metric properties of a regular surface, there are various types of curvature that arise from the geometry of the Gauss map of the surface. This is the map from the surface to the unit sphere S2 ā E3 that sends any point of the surface to the unit normal vector of the surface at that point. In our context, it can be deļ¬ned as follows: Deļ¬nition 4.25. Let U ā R2 be an open set, and let x : U ā E3 be an immersion with image Ī£ = x(U ). The Gauss map of Ī£ = x(U ) is the map N : Ī£ ā S2 deļ¬ned by N (x(u)) = e3 (u), where (e1 (u), e2 (u), e3 (u)) is any adapted frame ļ¬eld on Ī£ = x(U ). (Note that N is well-deļ¬ned up to sign.) Notions of curvature typically associated with surfaces (Gauss curvature, mean curvature, etc.) arise as linear-algebraic properties of the diļ¬erential dN of the Gauss map, also known as the shape operator of the surface. The relevant information is contained in the second fundamental form of the surface. Deļ¬nition 4.26. Let U ā R2 be an open set, and let x : U ā E3 be an immersion. The second fundamental form of Ī£ = x(U ) is the quadratic form II on T U deļ¬ned by II(v) = āde3 (v), dx(v) for v ā Tu U , where (e1 (u), e2 (u), e3 (u)) is any adapted frame ļ¬eld on Ī£ = x(U ). Since curvature is related to how rapidly the normal vector varies as we move around the surface, we might expect the functions (hij ) to show up in the second fundamental form. *Exercise 4.27. (a) Show that for any v ā Tu U , II(v) = Ļ ĀÆ 13 (v) Ļ ĀÆ 1 (v) + Ļ ĀÆ 23 (v) Ļ ĀÆ 2 (v) = h11 (ĀÆ Ļ 1 (v))2 + 2h12 Ļ ĀÆ 1 (v) Ļ ĀÆ 2 (v) + h22 (ĀÆ Ļ 2 (v))2 . This is often written more concisely as II = Ļ ĀÆ 13 Ļ ĀÆ1 + Ļ ĀÆ 23 Ļ ĀÆ 2 = h11 (ĀÆ Ļ 1 )2 + 2h12 Ļ ĀÆ1 Ļ ĀÆ 2 + h22 (ĀÆ Ļ 2 )2 .
122
4. Curves and surfaces in Euclidean space
(b) Suppose that x : U ā E3 is a parametrization with F = 0, and let (e1 (u), e2 (u), e3 (u)) be the frame ļ¬eld in part (a) of Exercise 4.24. Show that ā II = Eh11 du2 + 2 EGh12 du dv + Gh22 dv 2 . (c) The second fundamental form is more commonly written as II = e du2 + 2f du dv + g dv 2 , where e = e3 , xuu = ā(e3 )u , xu = āde3 f = e3 , xuv = ā(e3 )v , xu = = ā(e3 )u , xv = g = e3 , xvv = ā(e3 )v , xv =
ā ā āu , dx āu , ā ā , dx āu āde3 āv ā ā āde3 āu , dx āv , ā ā , dx āv . āde3 āv
(Some authors use , m, n or L, M, N in place of e, f, g.) Show that this agrees with Deļ¬nition 4.26, and conclude that h11 =
e , E
f h12 = ā , EG
h22 =
g . G
Now, we still havenāt ļ¬gured out how we should choose the vectors (e1 (u), e2 (u)), except that they should form an orthonormal basis for Tx(u) Ī£ at each point. In order to reļ¬ne our adapted frame ļ¬eld further, we will examine how the matrix [hij ] changes if we vary the frame. So, let (e1 (u), e2 (u), e3 (u)) be any adapted frame ļ¬eld, with associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ). Ė2 (u), e Ė3 (u)) has the form (up to sign) Any other adapted frame ļ¬eld (Ė e1 (u), e ā” ā¤ cos(Īø) ā sin(Īø) 0
ā¢ ā„ Ė1 (u) e Ė2 (u) e Ė3 (u) = e1 (u) e2 (u) e3 (u) ā£ sin(Īø) cos(Īø) 0ā¦ e 0 0 1 ĖĀÆ i , Ļ ĖĀÆ ji ) be the Maurer-Cartan forms associfor some function Īø on U . Let (Ļ cos(Īø) ā sin(Īø) ated to the new frame ļ¬eld, and set B = . sin(Īø) cos(Īø) *Exercise 4.28. (a) Show that the result in part (c) of Exercise 4.24 can be expressed as 1 1 ĖĀÆ Ļ Ļ ĀÆ (4.5) = B ā1 . ĖĀÆ 2 Ļ Ļ ĀÆ2
4.5. Moving frames for surfaces in E3
(b) Show that (4.6)
ĖĀÆ 13 Ļ ĖĀÆ 23 Ļ
=B
ā1
Ļ ĀÆ 13
123
Ļ ĀÆ 23
t
= B
Ļ ĀÆ 13 Ļ ĀÆ 23
.
(Hint: Use the equation for de3 in (3.1).) Ė 11 , h Ė 12 , h Ė 22 on U such (c) Cartanās lemma implies that there exist functions h that 3 1 Ė 11 h Ė 12 Ļ ĖĀÆ 1 ĖĀÆ Ļ h = . Ė 12 h Ė 22 Ļ ĖĀÆ 23 ĖĀÆ 2 Ļ h Show that Ė Ė 12 h11 h h11 h12 h11 h12 ā1 t (4.7) =B B= B B. Ė 12 h Ė 22 h12 h22 h12 h22 h Recall from linear algebra that any symmetric matrix can be transformed to a diagonal matrix by just such an orthogonal change of basis. Therefore, for each u ā U there exists an adapted frame (e1 (u), e2 (u), e3 (u)) at the point x(u) ā Ī£ with the property that h11 (u) h12 (u) 0 Īŗ1 (u) (4.8) = h12 (u) h22 (u) 0 Īŗ2 (u) for some real numbers Īŗ1 (u), Īŗ2 (u). *Exercise 4.29. Let (e1 (u), e2 (u), e3 (u)) be an adapted frame satisfying (4.8), and let v1 , v2 ā Tu U be vectors with the property that dx(vi ) = ei (u) for i = 1, 2. Show that d(e3 )u (vi ) = dNx(u) (ei (u)) = āĪŗi (u)ei (u),
i = 1, 2.
This implies that e1 (u) and e2 (u) are eigenvectors for the linear transformation dNx(u) , the diļ¬erential of the Gauss map N : Ī£ ā S2 at the point x(u) ā Ī£, with eigenvalues āĪŗ1 (u), āĪŗ2 (u), respectively. You may recall the following deļ¬nition: Deļ¬nition 4.30. The eigenvectors for ādNx(u) are called principal vectors or principal directions at the point x(u) ā Ī£. The associated eigenvalues Īŗ1 (u), Īŗ2 (u) are called the principal curvatures of Ī£ at x(u). Therefore, there exists an adapted frame (e1 (u), e2 (u), e3 (u)) at each point x(u) ā Ī£ with the property that e1 (u) and e2 (u) are principal vectors at x(u). Such a frame will be called a principal adapted frame at the point x(u) ā Ī£, and an adapted frame ļ¬eld on Ī£ which has this property at every point x(u) ā Ī£ will be called a principal adapted frame ļ¬eld on Ī£.
124
4. Curves and surfaces in Euclidean space
Deļ¬nition 4.31. If Īŗ1 (u) = Īŗ2 (u) for some point u ā U , then the corresponding point x(u) of Ī£ is called an umbilic point of Ī£. If Ī£ has no umbilic points, then a principal adapted frame ļ¬eld can be determined uniquely (up to sign) by requiring that Īŗ1 > Īŗ2 ; moreover, Ė : U ā E(3). However, it can this frame ļ¬eld determines a smooth map x happen that a principal adapted frame ļ¬eld cannot be chosen smoothly in a neighborhood of an umbilic point; for this reason, umbilic points can be somewhat problematic. *Exercise 4.32. (a) Show that if Ī£ has no umbilic points, then the choice of a principal adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)) is equivariant (up to sign) under the action of E(3). (b) Show that for a principal adapted frame ļ¬eld, the second fundamental form is given by II = Īŗ1 (ĀÆ Ļ 1 )2 + Īŗ2 (ĀÆ Ļ 2 )2 . Remark 4.33. Exactly how much freedom does the phrase āup to signā represent? Given any principal adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)), we can (1) replace e3 (u) by āe3 (u); (2) depending on whether or not we changed the sign of e3 (u), replace one or both of e1 (u) and e2 (u) by their opposites so as to preserve the orientation of the basis (e1 (u), e2 (u), e3 (u)). (We might also exchange e1 (u) and e2 (u) with appropriately chosen signs, but for the most part, we will ignore this option.) So the other choices for a principal adapted frame ļ¬eld are (āe1 (u), e2 (u), āe3 (u)), (e1 (u), āe2 (u), āe3 (u)), (āe1 (u), āe2 (u), e3 (u)). *Exercise 4.34. For each of the principal adapted frame ļ¬elds in Remark 4.33, how do the sign changes to the frame vectors aļ¬ect the Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji )? Suppose that Ī£ = x(U ) has no umbilic points. Now that we (ļ¬nally!) have a way of deļ¬ning a canonical adapted frame ļ¬eld along Ī£, we can apply Lemma 4.2 to ļ¬nd a complete set of invariants for the surface. Theorem 4.35 (Bonnet). Let U ā R2 be an open set. Two immersions x1 , x2 : U ā E3 without umbilic points diļ¬er by a rigid motion if and only if they have the same ļ¬rst and second fundamental forms.
4.5. Moving frames for surfaces in E3
125
Remark 4.36. This theorem is true even for surfaces with umbilic points, but the proof is slightly more involved due to the issue of how to choose a canonical adapted frame ļ¬eld near umbilic points. Proof. One direction is clear: Since all our constructions are equivariant under the action of E(3), any two surfaces that diļ¬er by a rigid motion must have the same ļ¬rst and second fundamental forms. Conversely, suppose that x1 , x2 have the same ļ¬rst and second fundamental Ė1, x Ė 2 : U ā E(3) be principal adapted frame ļ¬elds for x1 , x2 , forms. Let x respectively; let (ĀÆ Ļi, Ļ ĀÆ ji ) denote the pulled-back dual and connection forms ĀÆ i, Ī© ĀÆ i ) denote those for x2 . By hypothesis, for x1 and let (Ī© j
IIx1
ĀÆ 1 )2 + (Ī© ĀÆ 2 )2 = Ix , Ix1 = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 = (Ī© 2 1 2 2 2 1 2 ĀÆ ) + (Īŗ2 )x (Ī© ĀÆ 2 )2 = IIx . = (Īŗ1 )x1 (ĀÆ Ļ ) + (Īŗ2 )x1 (ĀÆ Ļ ) = (Īŗ1 )x2 (Ī© 2 2
Equality of the ļ¬rst fundamental forms implies that 1 1 ĀÆ Ī© cos(Īø) ā sin(Īø) Ļ ĀÆ = ĀÆ2 Ā± sin(Īø) Ā± cos(Īø) Ļ Ī© ĀÆ2 for some function Īø on U , where the signs on the bottom row of the matrix are the same. Substituting this relation into the equation for the second fundamental forms yields 1 2 (Īŗ1 )x1 (ĀÆ Ļ 1 )2 + (Īŗ2 )x1 (ĀÆ Ļ 2 )2 = (Īŗ1 )x2 cos2 (Īø) + (Īŗ2 )x2 sin2 (Īø) (ĀÆ Ļ ) + 2 (((Īŗ2 )x2 ā (Īŗ1 )x2 ) sin(Īø) cos(Īø)) Ļ ĀÆ 1Ļ ĀÆ2 2 2 + (Īŗ1 )x2 sin2 (Īø) + (Īŗ2 )x2 cos2 (Īø) (ĀÆ Ļ ) . Since (Īŗ1 )x2 > (Īŗ2 )x2 , the vanishing of the middle term on the right-hand side implies that Īø is a multiple of Ļ2 . Then the remaining terms, together with the inequality Īŗ1 > Īŗ2 on both sides, imply that Īø is a multiple of Ļ. Therefore, (Īŗ1 )x2 = (Īŗ1 )x1 , (Īŗ2 )x2 = (Īŗ2 )x1 , and ĀÆ 1 = Ā±ĀÆ Ī© Ļ1,
ĀÆ 2 = Ā±ĀÆ Ī© Ļ2.
By making one of the permissible frame changes described in Remark 4.33 on one side or the other if necessary, we can arrange that both signs above are positive. ĀÆ1 = Ļ ĀÆ 1 = dĀÆ Since we now have Ī© ĀÆ 1 , we must have dĪ© Ļ 1 . According to the Cartan structure equations (3.8), this implies that ĀÆ1 ā Ļ (Ī© ĀÆ 21 ) ā§ Ļ ĀÆ 2 = 0. 2 ĀÆ1 ā Ļ By Cartanās lemma, Ī© ĀÆ 21 must be a multiple of Ļ ĀÆ 2 . But the same 2 ĀÆ 2 = dĀÆ ĀÆ1 āĻ reasoning applied to the equation dĪ© Ļ 2 implies that Ī© ĀÆ 21 must also 2
126
4. Curves and surfaces in Euclidean space
be a multiple of Ļ ĀÆ 1 . Since (ĀÆ Ļ1, Ļ ĀÆ 2 ) are linearly independent, it follows that ĀÆ1 = Ļ Ī© ĀÆ 1. 2
2
Ė 1 and x Ė 2 are both principal adapted frame ļ¬elds, we have Finally, since x ĀÆ 31 = (Īŗ1 )x2 Ī© ĀÆ 1 = (Īŗ1 )x1 Ļ Ī© ĀÆ1 = Ļ ĀÆ 13 , 2 ĀÆ 3 = (Īŗ2 )x Ī© ĀÆ 2 = (Īŗ2 )x Ļ Ī© ĀÆ 23 . 2 1ĀÆ = Ļ 2
The theorem now follows from Lemma 4.2.
Now we consider the question discussed in Ā§4.4; namely, can the ļ¬rst and second fundamental forms be prescribed arbitrarily? We must require that I be a positive deļ¬nite quadratic form (i.e., that I(v) > 0 for every v = 0 ā T U ) in order to deļ¬ne a metric on the surface. And in order to avoid the issue of umbilic points, we will assume that I and II are prescribed in such a way that IIu is not a scalar multiple of Iu at any point u ā U . Exercise 4.37. Why is this the right condition to impose on I and II in order to avoid umbilic points? We saw in the proof of Theorem 4.35 that prescribing these fundamental forms determines the 1-forms (ĀÆ Ļ1, Ļ ĀÆ 2, Ļ ĀÆ 13 , Ļ ĀÆ 23 ) associated to a principal adapted frame ļ¬eld up to sign and that these forms will have the properties that (ĀÆ Ļ1, Ļ ĀÆ 2 ) are linearly independent and that Ļ ĀÆ i3 is a multiple of Ļ ĀÆ i for 1 2 3 3 i = 1, 2. So, suppose that we are given 1-forms (ĀÆ Ļ ,Ļ ĀÆ ,Ļ ĀÆ1 , Ļ ĀÆ 2 ) on an open 2 set U ā R that satisfy these conditions. What additional conditions must these forms satisfy in order that there exist an embedding x : U ā E3 whose ļ¬rst and second fundamental forms are I = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 , II = Ļ ĀÆ 13 Ļ ĀÆ1 + Ļ ĀÆ 23 Ļ ĀÆ 2? Lemma 4.12 gives the answer: The forms Ļ ĀÆ 1, Ļ ĀÆ 2, Ļ ĀÆ 13 = āĀÆ Ļ31 , Ļ ĀÆ 23 = āĀÆ Ļ32 , 3 1 2 together with the form Ļ ĀÆ = 0 and some additional form Ļ ĀÆ 2 = āĀÆ Ļ1 , must satisfy the structure equations (3.8) for the Maurer-Cartan forms on E(3). Because Ļ ĀÆ 3 = 0, the ļ¬rst three of these equations may be written as dĀÆ Ļ 1 = āĀÆ Ļ21 ā§ Ļ ĀÆ 2, (4.9)
dĀÆ Ļ2 = Ļ ĀÆ 21 ā§ Ļ ĀÆ 1, dĀÆ Ļ 3 = 0 = āĀÆ Ļ13 ā§ Ļ ĀÆ1 ā Ļ ĀÆ 23 ā§ Ļ ĀÆ 2.
*Exercise 4.38. Show that the ļ¬rst two equations in (4.9) uniquely determine the 1-form Ļ ĀÆ 21 . (Hint: Ļ ĀÆ 21 must be equal to some linear combination 1 2 of (ĀÆ Ļ ,Ļ ĀÆ ). Show that each of the ļ¬rst two equations determines one of the unknown coeļ¬cients.)
4.5. Moving frames for surfaces in E3
127
The form Ļ ĀÆ 21 determined by the ļ¬rst two equations in (4.9) is called the Levi-Civita connection form of the metric deļ¬ned by the ļ¬rst fundamental form I = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 . The third equation just says that (ĀÆ Ļ13 , Ļ ĀÆ 23 ) must be 1 2 symmetric linear combinations of (ĀÆ Ļ ,Ļ ĀÆ ), which will automatically be true under our assumptions. The remaining structure equations may be written as dĀÆ Ļ21 = Ļ ĀÆ 13 ā§ Ļ ĀÆ 23 , (4.10)
dĀÆ Ļ13 = Ļ ĀÆ 23 ā§ Ļ ĀÆ 21 , dĀÆ Ļ23 = āĀÆ Ļ13 ā§ Ļ ĀÆ 21 .
The ļ¬rst of these equations is called the Gauss equation, and the last two are called the Codazzi-Mainardi equations, or simply the Codazzi equations. By Lemma 4.12, we have the following theorem: Theorem 4.39 (Bonnet). Let (ĀÆ Ļ1, Ļ ĀÆ 2, Ļ ĀÆ 13 , Ļ ĀÆ 23 ) be 1-forms on a connected and simply connected open set U ā R2 satisfying the conditions that (ĀÆ Ļ1, Ļ ĀÆ 2 ) are 3 linearly independent at each point of U and that Ļ ĀÆ i is a scalar multiple of Ļ ĀÆ i for i = 1, 2. Suppose that, together with the Levi-Civita connection form Ļ ĀÆ 21 determined by Ļ ĀÆ 1 and Ļ ĀÆ 2 , these forms satisfy the Gauss and Codazzi equations (4.10). Then there exists an immersed surface x : U ā E3 , unique up to rigid motion, whose ļ¬rst and second fundamental forms are I = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 , II = Ļ ĀÆ 13 Ļ ĀÆ1 + Ļ ĀÆ 23 Ļ ĀÆ 2. Because of this result, the Gauss and Codazzi equations are also referred to as the compatibility equations of the theory of surfaces in E3 . Exercise 4.40. Let (u, v) be local coordinates on R2 . Use the following steps to determine whether there exists an immersion x : R2 ā E3 with ļ¬rst and second fundamental forms I = cosh2 (v) (du2 + dv 2 ), II = du2 ā dv 2 . (a) Show that the 1-forms (ĀÆ Ļ1, Ļ ĀÆ 2, Ļ ĀÆ 13 , Ļ ĀÆ 23 ) determined by I and II according to the conditions of Theorem 4.39 are Ļ ĀÆ 1 = cosh(v) du, 1 Ļ ĀÆ 13 = du, cosh(v)
Ļ ĀÆ 2 = cosh(v) dv, 1 Ļ ĀÆ 23 = ā dv. cosh(v)
128
4. Curves and surfaces in Euclidean space
(b) Show that the Levi-Civita connection form is Ļ ĀÆ 21 = tanh(v) du. (Hint: Set Ļ ĀÆ 21 = a du + b dv for some unknown functions a, b on R2 . Use the structure equations for dĀÆ Ļ1 2 and dĀÆ Ļ to determine a and b.) (c) Check that these forms satisfy the Gauss and Codazzi equations. Therefore, Theorem 4.39 implies that the desired surface exists. (In fact, it is a catenoid.) *Exercise 4.41. This exercise is a continuation of Exercise 4.24. Suppose that x : U ā E3 is an immersion whose coordinate curves are all principal curves. (This means that xu , xv are both principal vectors at each point of Ī£ = x(U ).) (a) Show that the frame ļ¬eld in part (a) of Exercise 4.24 is a principal adapted frame ļ¬eld along Ī£. (b) Show that the condition that all coordinate curves of x are principal curves is equivalent to the condition that the ļ¬rst and second fundamental forms I = E du2 + 2F du dv + G dv 2 , II = e du2 + 2f du dv + g dv 2 have the property that F = f = 0. (c) Use the structure equations for the dual forms in part (b) of Exercise 4.24 to show that 1 Ļ ĀÆ 21 = ā (Ev du ā Gu dv). 2 EG (d) Show that e 1 Ļ ĀÆ = E g 2 Ļ ĀÆ 23 = Ļ ĀÆ = G
Ļ ĀÆ 13 =
e ā du, E g ā dv. G
(e) Show that the Gauss equation is equivalent to " " # # eg Gu 1 Ev ā (4.11) . + ā =ā ā EG 2 EG EG v EG u
4.5. Moving frames for surfaces in E3
129
(f) Show that the Codazzi equations are equivalent to e 1 g! ev = Ev + , 2 E G (4.12) e 1 g! g u = Gu + . 2 E G While isolated umbilic points on a surface can be problematic, it is natural to ask whether we can categorize those surfaces that are totally umbilic, i.e., surfaces with the property that every point is an umbilic point. Exercise 4.42. Suppose that the surface x : U ā E3 is totally umbilic. (a) Show that any adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)) is a principal adapted frame ļ¬eld. (b) Let (ĀÆ Ļi, Ļ ĀÆ ji ) be the Maurer-Cartan forms for an adapted frame ļ¬eld on Ī£ = x(U ). Show that there exists a smooth function Ī» : U ā R such that (4.13)
Ļ ĀÆ 13 = Ī»ĀÆ Ļ1,
Ļ ĀÆ 23 = Ī»ĀÆ Ļ2.
Conclude that the second fundamental form of Ī£ is a scalar multiple of the ļ¬rst fundamental form, i.e., that II = Ī»I. (c) Prove that Ī» is constant. (Hint: Use the structure equations to diļ¬erentiate equations (4.13), taking into account the fact that we must have dĪ» = Ī»1 Ļ ĀÆ 1 + Ī»2 Ļ ĀÆ2 for some functions Ī»1 , Ī»2 on U . Then use Cartanās lemma.) (d) Show that if Ī» = 0, then de3 = 0. Conclude that the normal vector ļ¬eld of Ī£ is constant and that Ī£ is therefore contained in a plane. (e) Show that if Ī» = 0, then d(x+ Ī»1 e3 ) = 0. Conclude that the vector-valued function x + Ī»1 e3 : U ā E3 is equal to some constant point q ā E3 and that 1 Ī£ is therefore contained in the sphere of radius |Ī»| centered at q. Thus, the only totally umbilic surfaces in E3 are (open subsets of) planes and spheres. One of the conclusions of Exercise 4.42 is that if the principal curvatures Īŗ1 , Īŗ2 of a regular surface Ī£ are equal at every point of Ī£, then they must in fact be constant. This suggests a related question: Can we categorize those surfaces for which Īŗ1 , Īŗ2 are constants, but not necessarily equal?
130
4. Curves and surfaces in Euclidean space
Exercise 4.43. Suppose that the surface x : U ā E3 has the property that both principal curvatures Īŗ1 , Īŗ2 are constants. We know from Exercise 4.42 that if Īŗ1 = Īŗ2 , then Ī£ = x(U ) is contained in either a plane or a sphere, so assume that Īŗ1 = Īŗ2 . Let (e1 (u), e2 (u), e3 (u)) be a principal adapted Ė : U ā E(3) denote the corresponding lifting of frame ļ¬eld on Ī£, and let x x : U ā E3 , with associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ). Then we have (4.14)
Ļ ĀÆ 13 = Īŗ1 Ļ ĀÆ 1,
Ļ ĀÆ 23 = Īŗ2 Ļ ĀÆ 2.
(a) Diļ¬erentiate equations (4.14) to obtain (Īŗ1 ā Īŗ2 )ĀÆ Ļ21 ā§ Ļ ĀÆ 1 = (Īŗ1 ā Īŗ2 )ĀÆ Ļ21 ā§ Ļ ĀÆ 2 = 0. Use Cartanās lemma to conclude that Ļ ĀÆ 21 = 0. (b) Diļ¬erentiate the equation Ļ ĀÆ 21 = 0 and show that Īŗ1 Īŗ2 = 0. Without loss of generality, we may assume that Īŗ1 = 0, Īŗ2 = 0. In the remainder of this exercise, we will see how the structure equations can be integrated in order to determine the surface Ī£. (c) Show that dĀÆ Ļ 1 = dĀÆ Ļ 2 = 0. Apply the PoincarĀ“e lemma (cf. Theorem 2.31) to conclude that there exist functions u, v on U such that Ļ ĀÆ 1 = du,
Ļ ĀÆ 2 = dv.
Ė can be written Thus the pullbacks of the structure equations (3.1) to U via x as dx = e1 du + e2 dv, de1 = 0,
(4.15)
de2 = e3 (Īŗ2 dv), de3 = āe2 (Īŗ2 dv).
(d) Integrate equations (4.15) (beginning with the equations for de1 , de2 , de3 ĀÆ1 , e ĀÆ2 , e ĀÆ3 , and working backwards) to show that there exist constant vectors e 3 ĀÆ ā E such that x ĀÆ1 , e1 (u, v) = e ĀÆ2 + sin(Īŗ2 v) e ĀÆ3 , e2 (u, v) = cos(Īŗ2 v) e (4.16)
ĀÆ2 + cos(Īŗ2 v) e ĀÆ3 , e3 (u, v) = ā sin(Īŗ2 v) e 1 1 ĀÆ + ue ĀÆ1 + ĀÆ2 ā ĀÆ3 . x(u, v) = x sin(Īŗ2 v) e cos(Īŗ2 v) e Īŗ2 Īŗ2
(e) Use equations (4.16) and the fact that (e1 (u), e2 (u), e3 (u)) is an orĀÆ2 , e ĀÆ3 ) is an orthonormal frame. thonormal frame ļ¬eld to show that (ĀÆ e1 , e
4.5. Moving frames for surfaces in E3
131
ĀÆ=0 Conclude that via a Euclidean transformation, we can arrange that x and ā” ā¤ ā” ā¤ ā” ā¤ 1 0 0 ĀÆ1 = ā£0ā¦ , ĀÆ2 = ā£1ā¦ , ĀÆ3 = ā£0ā¦ , e e e 0 0 1 and hence that ā” ā¤ u ā¢ ā„ (4.17) x(u, v) = ā£ Īŗ12 sin(Īŗ2 v) ā¦ . ā Īŗ12 cos(Īŗ2 v) Equation (4.17) describes a parametrization for the cylinder of radius |Īŗ12 | centered along the x1 -axis; therefore, Ī£ = x(U ) is contained in a cylinder of radius |Īŗ12 | . Together, Exercises 4.42 and 4.43 prove the following classiļ¬cation theorem: Theorem 4.44. Let Ī£ be a connected, regular surface in E3 whose principal curvatures are constant. Then Ī£ is contained in either a plane, sphere, or cylinder. Any invariant of an immersed surface x : U ā E3 that can be expressed purely in terms of the ļ¬rst fundamental form I = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 is called an intrinsic invariant of the surface. For instance, arc length and area are intrinsic quantities on Ī£ = x(U ). The principal curvatures Īŗ1 , Īŗ2 , however, are not intrinsic; they depend not only on the metric, but also on how the surface is immersed. Two important notions of curvature for surfaces are given in the following deļ¬nition: Deļ¬nition 4.45. The function K = Īŗ1 Īŗ2 on Ī£ is called the Gauss curvature of Ī£. The function H = 12 (Īŗ1 + Īŗ2 ) on Ī£ is called the mean curvature of Ī£. Remark 4.46. It is not necessary that an adapted frame ļ¬eld be principal in order to compute the Gauss and mean curvatures. For any adapted frame ļ¬eld on Ī£ with associated matrix [hij ], we have K = det[hij ],
H = 12 tr[hij ].
Even though Īŗ1 , Īŗ2 are not intrinsic quantities, Gaussās āTheorema Egregiumā states that their product K is, in fact, intrinsic. (The mean curvature H, however, is not intrinsic.) In the following exercise, we will prove this in
132
4. Curves and surfaces in Euclidean space
several steps. (For simplicity, we will assume that the surface is oriented, meaning that a choice of e3 has been speciļ¬ed.) Exercise 4.47. Let x : U ā E3 be an immersed surface. The 1-forms (ĀÆ Ļ1, Ļ ĀÆ 2 ) are determined by the ļ¬rst fundamental form of x up to a transformation of the form 1 1 ĖĀÆ Ļ cos(Īø) sin(Īø) Ļ ĀÆ = ĖĀÆ 2 ā sin(Īø) cos(Īø) Ļ Ļ ĀÆ2 for some function Īø on U . (a) Show that the area form dA = Ļ ĀÆ1 ā§ Ļ ĀÆ2 is an intrinsic quantity, i.e., that ĖĀÆ 1 ā§ Ļ ĖĀÆ 2 = dA. dAĖ = Ļ (Note: The notation dA for the area form is traditional, but the d does not signify that dA is the exterior derivative of some 1-form.) (b) Show that if Ļ ĀÆ 21 is the Levi-Civita connection form corresponding to 1 2 (ĀÆ Ļ ,Ļ ĀÆ ), then Ļ ĀÆĖ 21 = Ļ ĀÆ 21 ā dĪø. Conclude that dĀÆ Ļ21 is an intrinsic quantity. (c) Show that dĀÆ Ļ21 = Ļ ĀÆ 13 ā§ Ļ ĀÆ 23 = K Ļ ĀÆ1 ā§ Ļ ĀÆ 2 = K dA. Conclude that K must be an intrinsic quantity. (Note that this is simply another version of equation (4.11), which expresses the Gauss curvature eg K = EG as a function of E, G, and their derivatives.) Surfaces for which the mean curvature H is identically zero are called minimal surfaces; these surfaces are of considerable interest and will be treated in detail in Chapter 8. Surfaces for which the Gauss curvature K is identically zero are called ļ¬at, and we will conclude this chapter with a brief exploration of their local theory. Because the Gauss curvature of a regular surface Ī£ is an intrinsic quantity, it is not changed by any deformation of Ī£ that preserves the ļ¬rst fundamental form of Ī£. Intuitively, this means that the surface may be smoothly bent and/or twisted, but not stretched or contracted. So for instance, any surface that can be obtained by smoothly bending a sheet of paper must be ļ¬at.
4.5. Moving frames for surfaces in E3
133
Since any ļ¬at surface Ī£ must have K = Īŗ1 Īŗ2 = 0, one of the principal curvatures Īŗ1 , Īŗ2 must be identically zero on Ī£. As we saw in Exercise 4.42 that any surface with Īŗ1 = Īŗ2 = 0 must be contained in a plane, we will disregard this case and, to keep things simple, we will assume that Ī£ has no umbilic points. (In practice, this simply means that we restrict our attention to the open subset of Ī£ consisting of the non-umbilic points.) Exercise 4.48. Suppose that the surface x : U ā E3 is ļ¬at and has no umbilic points. Without loss of generality, we may assume that the principal curvatures of Ī£ = x(U ) satisfy Īŗ1 = 0, Īŗ2 = 0. Let (e1 (u), e2 (u), e3 (u)) Ė : U ā E(3) denote the be a principal adapted frame ļ¬eld on Ī£, and let x corresponding lifting of x : U ā E3 , with associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ). Then since Īŗ1 = 0, we have (4.18)
Ļ ĀÆ 13 = 0,
Ļ ĀÆ 23 = Īŗ2 Ļ ĀÆ 2.
(a) Diļ¬erentiate the equation Ļ ĀÆ 13 = 0 and use Cartanās lemma to conclude that Ļ ĀÆ 21 = Ī¼ Ļ ĀÆ2
(4.19) for some function Ī¼ : U ā R.
(b) Show that dĀÆ Ļ 1 = 0, and apply the PoincarĀ“e lemma (cf. Theorem 2.31) to conclude that there exists a function u on U such that Ļ ĀÆ 1 = du. (c) Use the structure equation for dĀÆ Ļ 2 and the Frobenius theorem (cf. Theorem 2.33) to conclude that for any point u ā U , there exist a neighborhood V ā U of u and diļ¬erentiable functions Ī», v : V ā R (with Ī» = 0) such that the restriction of Ļ ĀÆ 2 to V is given by Ļ ĀÆ 2 = Ī» dv. (For simplicity, we will shrink U if necessary and assume that these functions are deļ¬ned on the entire open set U .) (d) Since Ļ ĀÆ1 ā§ Ļ ĀÆ 2 = 0, the functions (u, v) form a local coordinate system on U , and we may regard Ī», Ī¼ as functions of u and v. Show that the structure equation for dĀÆ Ļ 2 implies that Ī¼ = ā Ī»Ī»u , and therefore (4.20)
Ļ ĀÆ 21 = āĪ»u dv.
(e) Use the structure equation for dĀÆ Ļ21 to conclude that Ī»uu = 0, and therefore (4.21)
Ī»(u, v) = uf1 (v) + f0 (v)
for some smooth functions f0 (v), f1 (v).
134
4. Curves and surfaces in Euclidean space
(f) Use the structure equation for dĀÆ Ļ23 to conclude that (Īŗ2 Ī»)u = 0. Integrate and use equation (4.21) to conclude that (4.22)
Īŗ2 (u, v) =
f2 (v) uf1 (v) + f0 (v)
for some smooth functions f0 (v), f1 (v), f2 (v). Ė can now be (g) The pullbacks of the structure equations (3.1) to U via x written as dx = e1 du + e2 (uf1 (v) + f0 (v))dv, (4.23)
de1 = e2 f1 (v) dv, de2 = āe1 f1 (v) dv + e3 f2 (v) dv, de3 = āe2 f2 (v) dv.
Conclude that the u-parameter curves are straight line segments (and that u is an arc-length parameter along these curves) and hence that Ī£ is a ruled surface. We have now proved the following theorem, keeping in mind that with the notation of Exercise 4.48, the mean curvature H of Ī£ is given by H = 12 Īŗ2 with Īŗ2 as in equation (4.22): Theorem 4.49. Let Ī£ be a ļ¬at surface whose mean curvature H is nonzero everywhere. Then for each point x ā Ī£, there exists a unique straight line x in E3 such that x ā x and x ā© Ī£ is an open neighborhood of x in x . Moreover, the restriction of the function H1 to the open interval x ā© Ī£ is an aļ¬ne linear function of the arc length parameter along this interval. A more traditional proof of this result is given in [MR05].
4.6. Maple computations In order to get set up to use Maple for some of the exercises in this chapter, begin by loading the Cartan and LinearAlgebra packages into Maple: > with(Cartan); > with(LinearAlgebra); Next, introduce the Maurer-Cartan forms on the frame bundle F (E3 ); these need to be declared so that Maple will recognize them as 1-forms. It suļ¬ces to declare a linearly independent subset; weāll deļ¬ne the others in terms of these shortly.
4.6. Maple computations
135
> Form(omega[1], omega[2], omega[3]); Form(omega[1,2], omega[3,1], omega[3,2]); Next, tell Maple about the symmetries in the connection forms: > omega[1,1]:= omega[2,2]:= omega[3,3]:= omega[2,1]:= omega[1,3]:= omega[2,3]:=
0; 0; 0; -omega[1,2]; -omega[3,1]; -omega[3,2];
Tell Maple how to diļ¬erentiate these forms according to the Cartan structure equations (3.8): > for i from 1 to 3 do d(omega[i]):= -add(āomega[i,j] &Ė omega[j]ā, j=1..3); end do; d(omega[1,2]):= -add(āomega[1,k] &Ė omega[k,2]ā, k=1..3); d(omega[3,1]):= -add(āomega[3,k] &Ė omega[k,1]ā, k=1..3); d(omega[3,2]):= -add(āomega[3,k] &Ė omega[k,2]ā, k=1..3); Now consider the pullbacks of the Maurer-Cartan forms to the surface via an adapted frame ļ¬eld. The ļ¬rst condition that these forms must satisfy is Ļ ĀÆ 3 = 0. In Maple, itās often useful to impose such conditions via a substitution rather than by simply setting Ļ ĀÆ 3 equal to zero. The reason for this is that if we make the assignment Ļ ĀÆ 3 = 0, we lose the ability to use the structure 3 equation for dĀÆ Ļ because Maple will just evaluate dĀÆ Ļ 3 as d(0) = 0. Using a substitution allows us to choose when we want Maple to be aware that Ļ ĀÆ 3 = 0 and when we donāt. So, introduce the following substitution for the Maurer-Cartan forms associated to an adapted frame ļ¬eld. (Weāll add more information to this substitution as we learn more about the Maurer-Cartan forms.) > adaptedsub1:= [omega[3]=0]; Now, since we have Ļ ĀÆ 3 = 0, we must have dĀÆ Ļ 3 = 0 as well. So the following quantity must be zero: > zero1:= Simf(subs(adaptedsub1, Simf(d(omega[3]))));
zero1 := (Ļ1 ) &Ė (Ļ31 ) + (Ļ2 ) &Ė (Ļ32 )
136
4. Curves and surfaces in Euclidean space
Note that we ļ¬rst computed dĀÆ Ļ 3 and then applied the substitution to tell Maple that Ļ ĀÆ 3 = 0. In this case the knowledge that Ļ ĀÆ 3 = 0 didnāt aļ¬ect 3 the computation of dĀÆ Ļ , but itās a good idea to get in the habit of applying such substitutions when you intend for them to be in eļ¬ect. Applying Cartanās lemma tells us that (ĀÆ Ļ13 , Ļ ĀÆ 23 ) must be symmetric linear combinations of (ĀÆ Ļ1, Ļ ĀÆ 2 ), so we add this information to our substitution: > adaptedsub1:= [op(adaptedsub1), omega[3,1] = h[1,1]*omega[1] + h[1,2]*omega[2], omega[3,2] = h[1,2]*omega[1] + h[2,2]*omega[2]]; Exercise 4.28: In order to keep up with both the original Maurer-Cartan ĖĀÆ i , Ļ ĖĀÆ ji ), introduce new 1-forms to forms (ĀÆ Ļi, Ļ ĀÆ ji ) and the transformed forms (Ļ represent the transformed forms, with the same symmetry conditions as the original forms: > Form(Omega[1], Omega[2], Omega[3]); Form(Omega[1,2], Omega[3,1], Omega[3,2]); Omega[1,1]:= 0; Omega[2,2]:= 0; Omega[3,3]:= 0; Omega[2,1]:= -Omega[1,2]; Omega[1,3]:= -Omega[3,1]; Omega[2,3]:= -Omega[3,2]; (It wonāt be necessary to assign their exterior derivatives because these will be computed in terms of the exterior derivatives of the original forms when needed.) We can introduce the relations (4.5), (4.6) via the following substitution: > framechangesub:= [ Omega[1] = cos(theta)*omega[1] + sin(theta)*omega[2], Omega[2] = -sin(theta)*omega[1] + cos(theta)*omega[2], Omega[3,1] = cos(theta)*omega[3,1] + sin(theta)*omega[3,2], Omega[3,2] = -sin(theta)*omega[3,1] + cos(theta)*omega[3,2]]; Weāll also need the reverse substitution so that we can go back and forth between the two sets of Maurer-Cartan forms: > framechangebacksub:= makebacksub(framechangesub); Ė ij ) associated to the transformed forms In order to compare the functions (h to the functions (hij ) associated to the original forms, introduce another
4.6. Maple computations
137
substitution describing the adaptations of the transformed frame: > adaptedsub2:= [Omega[3]=0, Omega[3,1] = H[1,1]*Omega[1] + H[1,2]*Omega[2], Omega[3,2] = H[1,2]*Omega[1] + H[2,2]*Omega[2]]; Ė ij ) are expressed in Now combine all these substitutions to see how the (h 3 3 3 Ė terms of the (hij ): First, write Ļ ĀÆ 1 in terms of (ĀÆ Ļ1 , Ļ ĀÆ 2 ): > Simf(subs(framechangesub, Omega[3,1])); cos(Īø) Ļ3,1 + sin(Īø) Ļ3,2 Next, convert this to an expression in terms of the (hij ) and (ĀÆ Ļ1, Ļ ĀÆ 2 ): > Simf(subs(adaptedsub1, %)); (cos(Īø) h1,1 + sin(Īø) h1,2 ) Ļ1 + (cos(Īø) h1,2 + sin(Īø) h2,2 ) Ļ2 ĖĀÆ 1 , Ļ ĖĀÆ 2 ): Finally, convert this to an expression in terms of (Ļ > Simf(subs(framechangebacksub, %)); (cos(Īø)2 h1,1 + 2 cos(Īø) sin(Īø) h1,2 + h2,2 ā h2,2 cos(Īø)2 ) Ī©1 + (ā cos(Īø) sin(Īø) h1,1 + cos(Īø) sin(Īø) h2,2 + 2 cos(Īø)2 h1,2 ā h1,2 )Ī©2 Of course, this sequence of operations can be combined into a single command: > Simf(subs(framechangebacksub, Simf(subs(adaptedsub1, Simf(subs(framechangesub, Omega[3,1])))))); Ė 11 , h Ė 12 , ĖĀÆ 1 , Ļ ĖĀÆ 2 ) in the output are, of course, equal to h Now, the coeļ¬cients of (Ļ respectively. But in order to illustrate how we might handle a slightly more complicated situation, we will let Maple do the work of comparing this ĖĀÆ 13 : expression to our original expression for Ļ > zero2:= Simf(subs(adaptedsub2, Omega[3,1]) - %); The coeļ¬cients of this expression must both be zero, which gives us two Ė 11 and h Ė 12 . These equations can be exequations that can be solved for h tracted as follows: > eqns:= {op(ScalarForm(zero2))}; Before solving these equations, we might as well compute the analogous ĖĀÆ 23 . We can add these to our equations that result from consideration of Ļ
138
4. Curves and surfaces in Euclidean space
system of equations as follows: > zero3:= Simf(subs(adaptedsub2, Omega[3,2]) - Simf(subs(framechangebacksub, Simf(subs(adaptedsub1, Simf(subs(framechangesub, Omega[3,2]))))))); > eqns:= eqns union {op(ScalarForm(zero3))}; Ė ij ): Now solve these equations for the functions (h > solve(eqns, {H[1,1], H[1,2], H[2,2]}); Ė ij ): Then, we might as well actually assign these values to the (h > assign(%); Now, itās not entirely obvious how to recognize these expressions as those of equation (4.7), but we can at least check that our computations are consisĖ ij ], B as follows: tent with these expressions. First, deļ¬ne matrices [hij ], [h > hmatrix:= Matrix([[h[1,1], h[1,2]], [h[1,2], h[2,2]]]); Hmatrix:= Matrix([[H[1,1], H[1,2]], [H[1,2], H[2,2]]]); B:= Matrix([[cos(theta), -sin(theta)], [sin(theta), cos(theta)]]); If everything has gone according to plan, the following matrix should be zero: > Hmatrix - simplify(Transpose(B).hmatrix.B); Exercise 4.40: Set up a substitution for the forms that we know from part (a), together with an expression for Ļ ĀÆ 21 with coeļ¬cients to be determined later: > examplesub:= [omega[1] = cosh(v)*d(u), omega[2] = cosh(v)*d(v), omega[3]=0, omega[3,1] = d(u)/cosh(v), omega[3,2] = -d(v)/cosh(v), omega[1,2] = a*d(u) + b*d(v)]; Now, compute dĀÆ Ļ 1 in two ways: by ļ¬rst making the substitution into Ļ ĀÆ1 and then diļ¬erentiating, and by applying the structure equations and then making the substitution. Then the diļ¬erence of the resulting expressions must be equal to zero: > Simf(d(Simf(subs(examplesub, omega[1]))) - subs(examplesub, Simf(d(omega[1])))); (sinh(v) ā cosh(v) a) d(v) &Ė d(u) > a:= solve(%, a);
4.6. Maple computations
139
a :=
sinh(v) cosh(v)
An analogous computation for dĀÆ Ļ 2 yields b = 0; once we have made this 1 assignment, we will have Ļ ĀÆ 2 = tanh(v) du, as expected. Finally, verifying the Gauss and Codazzi equations simply involves checking that both ways of computing the structure equations for each of the (dĀÆ Ļji ) yield the same result: > Simf(d(Simf(subs(examplesub, omega[1,2]))) - subs(examplesub, Simf(d(omega[1,2])))); 0 > Simf(d(Simf(subs(examplesub, omega[3,1]))) - subs(examplesub, Simf(d(omega[3,1])))); 0 > Simf(d(Simf(subs(examplesub, omega[3,2]))) - subs(examplesub, Simf(d(omega[3,2])))); 0 Exercise 4.41: For a principal parametrization as in Exercise 4.24, we have ā ā Ļ ĀÆ 1 = E du, Ļ ĀÆ 2 = G dv. Then, in order for the second fundamental form to have the desired form, we must have e g Ļ ĀÆ 13 = ā du, Ļ ĀÆ 23 = ā dv. E G 1 Moreover, Ļ ĀÆ 2 must be equal to some linear combination of du and dv. Start by unassigning the variables a, b so that we can use them again and declaring that E, G, e, g are functions of u and v. (This declaration isnāt strictly necessary, but it will make the output of some computations look nicer.) > unassign(āaā, ābā); > PDETools[declare](E(u,v), G(u,v), e(u,v), g(u,v)); Introduce a substitution for the Maurer-Cartan forms in terms of the coordinate 1-forms: > coordsub:= [omega[3]=0, omega[1] = sqrt(E(u,v))*d(u), omega[2] = sqrt(G(u,v))*d(v),
140
4. Curves and surfaces in Euclidean space
omega[3,1] = (e(u,v)/sqrt(E(u,v)))*d(u), omega[3,2] = (g(u,v)/sqrt(G(u,v)))*d(v), omega[1,2] = a*d(u) + b*d(v)]; Compute the coeļ¬cients in Ļ ĀÆ 21 as we did in the previous exercise: > Simf(d(Simf(subs(coordsub, omega[1]))) - subs(coordsub, Simf(d(omega[1])))); > a:= solve(%, a); > Simf(d(Simf(subs(coordsub, omega[2]))) - subs(coordsub, Simf(d(omega[2])))); > b:= solve(%, b); The Gauss equation comes from comparing the two expressions for dĀÆ Ļ21 : > Simf(d(Simf(subs(coordsub, omega[1,2]))) - subs(coordsub, Simf(d(omega[1,2])))); > Gausseq1:= pick(%, d(u), d(v)); Now, youāll probably notice that this expression doesnāt look quite like the one in part (e) of the exercise. But we can ask Maple to compare the two expressions to conļ¬rm that they are, in fact, equivalent. First, give a name to the expression that results from moving all the terms in equation (4.11) to the left-hand side: > Gausseq2:= (e(u,v)*g(u,v))/(E(u,v)*G(u,v)) + (1/(2*sqrt(E(u,v)*G(u,v))))* (diff(diff(E(u,v), v)/sqrt(E(u,v)*G(u,v)), v) + diff(diff(G(u,v), u)/sqrt(E(u,v)*G(u,v)), u)); Solve this equation for one of the variables (say, g), and then substitute this expression into the ļ¬rst version of the Gauss equation. If the two equations are equivalent, then the result should be zero. > solve(Gausseq2, {g(u,v)}); > Simf(subs(%, Gausseq1)); 0
Similar manipulations involving dĀÆ Ļ13 and dĀÆ Ļ23 will conļ¬rm that their structure equations are equivalent to the Codazzi equations (4.12). Now, in fact, thereās nothing special about assuming that F = f = 0, except that it makes the computations simpler. For a challenge, you might try redoing this exercise without this assumption. Youāll need to start by
4.6. Maple computations
141
applying the Gram-Schmidt algorithm to the basis (xu , xv ) in order to obtain an orthonormal frame ļ¬eld and then compute the dual forms (ĀÆ Ļ1, Ļ ĀÆ 2 ) for this frame ļ¬eld. (This part isnāt too bad to do by hand.) Details are given in the Maple worksheet for this chapter on the AMS webpage.
10.1090/gsm/178/05
Chapter 5
Curves and surfaces in Minkowski space
5.1. Introduction In physics, the study of relativity generally begins with special relativity, which is primarily the study of timelike curves in the Minkowski space M1,3 . As discussed in Ā§3.5, such curves represent the world lines of particles in spacetime, and special relativity describes how particles behave in the absence of a gravitational ļ¬eld. The eļ¬ects of gravity are considered in general relativity, where the Minkowski space M1,3 is replaced by a more general 4-dimensional manifold with a pseudo-Riemannian metric of signature (1, 3) and the strength of the gravitational ļ¬eld is reļ¬ected in the curvature of the metric. An excellent introduction to these topics from a geometric point of view is given in [Cal00]. In keeping with our general treatment of 3-dimensional homogeneous spaces, we will conļ¬ne our attention to the study of curves and surfaces in the Minkowski space M1,2 . In Ā§5.2, we will apply the method of moving frames to study the geometry of timelike curves in M1,2 , and many of the features of special relativity will already be apparent here. In Ā§5.3, we will study the geometry of timelike surfaces in M1,2 , i.e., surfaces for which the restriction of the Minkowski metric on M1,2 to each tangent plane has signature (1, 1). (By contrast, spacelike surfaces are those for which the restriction of the Minkowski metric to each tangent plane has signature (0, 2).) Such a surface may be regarded as a 2-dimensional model for the 4-dimensional pseudoRiemannian manifolds that are studied in general relativity.
143
144
5. Curves and surfaces in Minkowski space
5.2. Moving frames for timelike curves in M1,2 Consider a smooth, parametrized timelike curve Ī± : I ā M1,2 , where I ā R is some open interval. M1,2 has the structure of the homogeneous space M (1, 2)/SO+ (1, 2), so an adapted frame ļ¬eld along Ī± should be a lifting Ī± Ė : I ā M (1, 2). Any such lifting can be written as Ī± Ė (t) = (Ī±(t); e0 (t), e1 (t), e2 (t)), where for each t ā I, (e0 (t), e1 (t), e2 (t)) is an oriented, orthonormal basis for the tangent space TĪ±(t) M1,2 . (Recall that, by convention, we require that e0 (t) be timelike and that e1 (t) and e2 (t) be spacelike; cf. Exercise 3.39.) As in the Euclidean case, such an adapted frame ļ¬eld is usually called an orthonormal frame ļ¬eld along Ī±. As in the Euclidean case, we say that Ī± is regular if Ī± (t) = 0 for every t ā I; henceforth, we will only consider regular curves. The construction of an orthonormal frame ļ¬eld along Ī± is very similar to that for a curve in E3 ; we begin by setting Ī± (t) e0 (t) = , |Ī± (t)|
where |Ī± (t)| = Ī± (t), Ī± (t) is computed using the Minkowski inner product; i.e., we take e0 (t) to be the unit tangent vector to the curve at Ī±(t). (Note that, since Ī± is a timelike curve, e0 (t) is a timelike vector, as desired.) The Minkowski analog of arc length is the following: Deļ¬nition 5.1. Given a ļ¬xed point t0 ā I, the proper time function along Ī± is & t Ļ (t) = |Ī± (u)| du. t0
This terminology arises from the key fact that in relativity, time is not an absolute quantity; diļ¬erent observers may measure the passage of time diļ¬erently, and the proper time along a world line Ī±(t) represents how time passes for an observer traveling along Ī±. Remark 5.2. The use of the letter Ļ to denote proper time is traditional in relativity; however, it should not be confused with the torsion of a nondegenerate curve in E3 , which is also denoted by Ļ ! Hopefully it will be clear from the context which quantity is intended. This convention will also necessitate a departure from the traditional geometric notation for curves in E3 ; thus, we will denote the analogs of curvature and torsion for timelike curves in M1,2 by Īŗ1 , Īŗ2 .
5.2. Moving frames for timelike curves in M1,2
145
Exercise 5.3. Show that Ļ (t) is invariant under the action of M (1, 2); that is, for any g ā M (1, 2), the curves Ī± and g Ā· Ī± have the same proper time function. *Exercise 5.4. Consider two particles with world lines Ī±, Ī² : I ā M1,2 given by Ī±(t) = t[t, 0, 0],
Ī²(t) = t[t, vt, 0],
so that Ī± represents a particle that remains stationary at the origin in R2 and Ī² represents a particle traveling in the positive direction along the x1 axis with speed v < 1. (Recall that the inner product on M1,2 is normalized so that the speed of light is c = 1; so this just means that v is less than the speed of light.) (a) Show that the proper time function forāĪ± (with t0 = 0) is ĻĪ± (t) = t and the proper time function for Ī² is ĻĪ² (t) = t 1 ā v 2 . (b) Reparametrize Ī² according to its proper time function: t Ļ vĻ Ī²(Ļ ) = Ī²(t(Ļ )) = ā , ā ,0 . 1 ā v2 1 ā v2 Let u = tanhā1 (v), so that v = tanh(u), and consider the Lorentz transformation T : M1,2 ā M1,2 deļ¬ned by ā” 0ā¤ āā” 0 ā¤ā ā” ā¤ ā” 0ā¤ x Ė x cosh(u) ā sinh(u) 0 x ā¢ 1ā„ āā¢ 1 ā„ā ā¢ ā„ ā¢ 1ā„ Ė ā¦ = T āā£x ā¦ā = ā£ā sinh(u) cosh(u) 0ā¦ ā£x ā¦ . ā£x x Ė2
x2
0
0
1
x2
Show that with respect to the (Ė x0 , x Ė1 , x Ė2 )-coordinates, t t āvt Ī±(t) = ā , ā ,0 , Ī²(Ļ ) = t[Ļ, 0, 0]. 1 ā v2 1 ā v2 So in this new coordinate system, Ī² represents a stationary particle and Ī± represents a particle traveling in the negative direction along the x1 -axis with speed v. (Hint: You will need to make use of the hyperbolic trig identity sech2 (u) = 1 ā tanh2 (u).) (c) Let t1 = Ļ1 = T . Show that in either coordinate system, the points Ī±(t1 ) and Ī²(Ļ1 ) both lie on the hyperboloid {t[x0 , x1 , x2 ] ā M1,2 | (x0 )2 ā (x1 )2 ā (x2 )2 = T 2 }. This hyperboloid represents the points that can be reached from the origin in proper time Ļ = T by particles traveling with any constant spatial velocity v = v 1 āxā 1 + v 2 āxā 2 with (v 1 )2 + (v 2 )2 = v 2 < 1.
146
5. Curves and surfaces in Minkowski space
Henceforth, we will assume that Ī± is parametrized by proper time Ļ , and so e0 (Ļ ) = Ī± (Ļ ). As in the Euclidean case, we will say that a timelike curve Ī± is nondegenerate if Ī± is regular and, in addition, e0 (Ļ ) = 0 for all Ļ ā I. In this case, diļ¬erentiating the equation e0 (Ļ ), e0 (Ļ ) = 1 yields e0 (Ļ ), e0 (Ļ ) = 0. Thus, e0 (Ļ ) is orthogonal to e0 (Ļ ), and we deļ¬ne e1 (Ļ ) =
e0 (Ļ ) . |e0 (Ļ )|
This vector will be called the unit normal vector to the curve at Ī±(Ļ ); note that e1 (Ļ ) is a spacelike vector (cf. Exercise 3.39). By analogy with the Euclidean case, at this point we would like to deļ¬ne e2 (Ļ ) = e0 (Ļ ) Ć e1 (Ļ ). But ļ¬rst, we have to deļ¬ne an appropriate notion of the cross product in M1,2 and check that it has the desired properties. Recall that the cross product for two vectors v = t[v 1 , v 2 , v 3 ], w = t[w1 , w2 , w3 ] in E3 is v Ć w = t[v 2 w3 ā v 3 w2 , v 3 w1 ā v 1 w3 , v 1 w2 ā v 2 w1 ], often written schematically as
e1 e2 e3 v Ć w = v1 v2 v3 . w 1 w 2 w 3
The analogous notion for Minkowski space is Deļ¬nition 5.5. Let v = t[v 0 , v 1 , v 2 ], w = t[w0 , w1 , w2 ] ā M1,2 . The Minkowski cross product v Ć w is v Ć w = t[v 1 w2 ā v 2 w1 , v 0 w2 ā v 2 w0 , v 1 w0 ā v 0 w1 ], written schematically as
e0 e1 e2 v Ć w = v 0 āv 1 āv 2 . w0 āw1 āw2
The following exercise shows that the Minkowski cross product has properties analogous to those of the Euclidean cross product.
5.2. Moving frames for timelike curves in M1,2
147
Exercise 5.6. (a) Show that with respect to the Minkowski inner product, v Ć w is orthogonal to both v and w. (b) Show that v Ć w, v Ć w = v, vw, w ā v, w2 . Conclude that if v, w are orthogonal vectors of Minkowski norm 1, then v Ć w also has Minkowski norm equal to 1. (c) Show that for any three vectors u, v, w ā M1,2 ,
u, v Ć w = det u v w . (This is an orientation condition that dictates how the sign of v Ć w should be chosen.) As a consequence of Exercise 5.6, we can complete our adapted frame ļ¬eld along Ī± by deļ¬ning e2 (Ļ ) = e0 (Ļ ) Ć e1 (Ļ ). Exercise 5.7. Prove that this choice of frame ļ¬eld (e0 (Ļ ), e1 (Ļ ), e2 (Ļ )) is equivariant under the action of M (1, 2): If we replace Ī± by g Ā· Ī± for some g ā M (1, 2), then eĪ± (Ļ ) ā TĪ±(Ļ ) M1,2 will be replaced by (Lg )ā (eĪ± (Ļ )) ā TgĀ·Ī±(Ļ ) M1,2 . We now have a canonical adapted frame ļ¬eld (e0 (Ļ ), e1 (Ļ ), e2 (Ļ )) deļ¬ned at each point of Ī±(Ļ ), which in turn deļ¬nes a canonical, left-invariant lifting Ī± Ė : I ā M (1, 2) for any nondegenerate timelike curve Ī±, given by Ī± Ė (Ļ ) = (Ī±(Ļ ); e0 (Ļ ), e1 (Ļ ), e2 (Ļ )) . Now consider the pullbacks of equations (3.1) to I via Ī± Ė ; these can be written as Ī± (Ļ )dĻ = eĪ± (Ļ ) Ļ ĀÆ Ī±, (5.1) eĪ± (Ļ )dĻ = eĪ² (Ļ ) Ļ ĀÆ Ī±Ī² . Recall that we constructed our adapted frame ļ¬eld so that Ī± (Ļ ) = e0 (Ļ ); therefore, the ļ¬rst equation in (5.1) implies that Ļ ĀÆ 0 = dĻ,
Ļ ĀÆ1 = Ļ ĀÆ 2 = 0.
Then we chose e1 (Ļ ) so that e0 (Ļ ) is a multiple of e1 (Ļ ), say e0 (Ļ ) = Īŗ1 (Ļ )e1 (Ļ ). The function Īŗ1 (Ļ ) is the analog of the curvature Īŗ(s) for curves in E3 ; note that Ī± is nondegenerate if and only if Īŗ1 (Ļ ) > 0 for all Ļ ā I. We will call Īŗ1 (Ļ ) the Minkowski curvature of Ī±. So the equation for e0 (Ļ ) in (5.1) implies that Ļ ĀÆ 01 = Īŗ1 (Ļ )dĻ,
Ļ ĀÆ 02 = 0.
148
5. Curves and surfaces in Minkowski space
(Recall from Exercise 3.50 that Ļ ĀÆ 00 = 0.) The only remaining Maurer-Cartan form is Ļ ĀÆ 21 ; it must be equal to some multiple of dĻ , so deļ¬ne a function Īŗ2 (Ļ ) by the condition that Ļ ĀÆ 21 = Īŗ2 (Ļ )dĻ. The function Īŗ2 (Ļ ) is the analog of the torsion Ļ (s) for curves in E3 . We will call Īŗ2 (Ļ ) the Minkowski torsion of Ī±. Using the symmetry relations between the (ĀÆ ĻĪ²Ī± ) from Exercise 3.50, we have the Minkowski analog of the Frenet equations for timelike curves: ā” ā¤ 0 0 0 0 ā¢ ā„
ā¢1 0 Īŗ1 (Ļ ) 0 ā„ ā„ Ī± (Ļ ) e0 (Ļ ) e1 (Ļ ) e2 (Ļ ) = Ī±(Ļ ) e0 (Ļ ) e1 (Ļ ) e2 (Ļ ) ā¢ ā¢0 Īŗ (Ļ ) 0 Īŗ (Ļ )ā„ . 2 ā£ 1 ā¦ 0 0 āĪŗ2 (Ļ ) 0 Applying Lemma 4.2 yields the following theorem: Theorem 5.8. Two nondegenerate timelike curves Ī±1 , Ī±2 : I ā M1,2 parametrized by proper time diļ¬er by a Lorentz transformation if and only if they have the same Minkowski curvature Īŗ1 (Ļ ) and Minkowski torsion Īŗ2 (Ļ ). *Exercise 5.9. Consider the world line Ī± of a particle moving along the x1 -axis, with acceleration proportional to its distance from the origin and directed away from the origin and with initial conditions 1 , (x1 ) (0) = 0, a where a > 0 is the constant of proportionality for which x1 (0) =
(x1 ) (Ļ ) = a2 x1 (Ļ ). (Note that Ī± and all its derivatives will be contained in the plane spanned by (e0 , e1 ).) (a) Show that
t
Ī±(Ļ ) =
1 1 sinh(aĻ ), cosh(aĻ ), 0 . a a
(Hint: Let Ī±(Ļ ) = t[x0 (Ļ ), x1 (Ļ ), 0]. Solve the given initial value problem for x1 (Ļ ), and then use the condition that Ī± is timelike and Ī± (Ļ ) = 1 to solve for x0 (Ļ ).) (b) Show that the world line of Ī± lies on a hyperbola in the (x0 , x1 )-plane. (c) Show that the Minkowski curvature and torsion of Ī± are Īŗ1 (Ļ ) = a,
Īŗ2 (Ļ ) = 0.
5.3. Moving frames for timelike surfaces in M1,2
149
Thus, Ī± is the Minkowski analog of a circle in E3 , i.e., a nondegenerate curve with constant nonzero curvature and zero torsion. *Exercise 5.10. Consider the world line Ī± of a particle moving along a circle of radius r in the (x1 , x2 )-plane with what a stationary observer believes to be constant angular velocity k > 0. This means that Ī± has a parametrization of the form Ī±(t) = t[t, r cos(kt), r sin(kt)]. (a) Show that the proper time function for Ī± is
Ļ (t) = t 1 ā k 2 r2 ; therefore, the proper time parametrization for Ī± is # " # " t Ļ kĻ kĻ Ī±(Ļ ) = ā , r sin ā . , r cos ā 1 ā k2 r2 1 ā k2 r2 1 ā k2 r2 (Note that this means that an observer traveling along Ī± believes its angular k velocity to be kĖ = ā1āk .) 2 r2 (b) Compute the Frenet frame (e0 (Ļ ), e1 (Ļ ), e2 (Ļ )) for Ī± and show that the Minkowski curvature and torsion of Ī± are k2r k Īŗ1 (Ļ ) = , Īŗ2 (Ļ ) = . 2 2 1āk r 1 ā k2 r2 Thus, Ī± is the Minkowski analog of a helix in E3 , i.e., a nondegenerate curve with constant nonzero curvature and torsion. (This shouldnāt be too surprising since the world line of Ī± is a helix in M1,2 !)
5.3. Moving frames for timelike surfaces in M1,2 Now, let U be an open set in R2 , and let x : U ā M1,2 be an immersion whose image is a timelike surface Ī£ = x(U ). Just as for curves, an adapted Ė : U ā M (1, 2) of the form frame ļ¬eld along Ī£ is a lifting x Ė (u) = (x(u); e0 (u), e1 (u), e2 (u)) , x where for each u ā U , (e0 (u), e1 (u), e2 (u)) is an oriented, orthonormal basis for the tangent space Tx(u) M1,2 . As in the Euclidean case, we begin by choosing e2 (u) to be orthogonal to the tangent plane Tx(u) Ī£. (Note that, since Ī£ is assumed to be timelike, this implies that e2 (u) is a spacelike vector of Minkowski norm 1.) Exercise 5.11. Show that this choice of e2 (u) is equivariant (up to sign) under the action of M (1, 2).
150
5. Curves and surfaces in Minkowski space
The vectors (e0 (u), e1 (u)) must now form a basis for the tangent space Tx(u) Ī£. We will assume that e0 (u) is timelike and e1 (u) is spacelike, but otherwise (e0 (u), e1 (u)) is allowed to be an arbitrary orthonormal basis of Tx(u) Ī£. An orthonormal frame ļ¬eld satisfying this condition will be called adapted. Ė ā ĻĪ²Ī± ) on U . Precisely Let (ĀÆ ĻĪ±, Ļ ĀÆ Ī²Ī± ) represent the pulled-back forms (Ė xā Ļ Ī± , x the same reasoning as in the Euclidean case can be used to prove the following: Proposition 5.12. Let U ā R2 be an open set, and let x : U ā M1,2 be a timelike immersion. For any adapted orthonormal frame ļ¬eld (e0 (u), e1 (u), e2 (u)) along Ī£ = x(U ), the associated dual and connection forms (ĀÆ ĻĪ±, Ļ ĀÆ Ī²Ī± ) have the property that Ļ ĀÆ 2 = 0. Moreover, (ĀÆ Ļ0, Ļ ĀÆ 1 ) form a basis for the 1-forms on U . The metric properties of the surface Ī£ are once again contained in the ļ¬rst fundamental form of the surface: Deļ¬nition 5.13. Let U ā R2 be an open set, and let x : U ā M1,2 be a timelike immersion. The ļ¬rst fundamental form of Ī£ = x(U ) is the quadratic form I on T U deļ¬ned by I(v) = dx(v), dx(v) for v ā Tu U , where Ā·, Ā· is the Minkowski inner product. *Exercise 5.14. Show that for any v ā Tu U , 0 2 1 2 I(v) = Ļ ĀÆ (v) ā Ļ ĀÆ (v) . This is often written more concisely as I = (ĀÆ Ļ 0 )2 ā (ĀÆ Ļ 1 )2 . Note that I is a quadratic form of signature (1, 1) rather than a positive deļ¬nite quadratic form. This reļ¬ects the fact that Ī£ is a timelike surface; if Ī£ were spacelike, then its ļ¬rst fundamental form would have signature (0, 2). As in the Euclidean case, diļ¬erentiating the equation Ļ ĀÆ 2 = 0 yields dĀÆ Ļ 2 = āĀÆ Ļ02 ā§ Ļ ĀÆ0 ā Ļ ĀÆ 12 ā§ Ļ ĀÆ 1 = 0, and Cartanās lemma (cf. Lemma 2.49) implies that there exist functions h00 , h01 , h11 on U such that 2 0 Ļ ĀÆ0 h00 h01 Ļ ĀÆ =ā . Ļ ĀÆ 12 h01 h11 Ļ ĀÆ1
5.3. Moving frames for timelike surfaces in M1,2
151
(The minus sign is included here in order to minimize minus signs in the second fundamental form below.) Once again, the functions (hĪ±Ī² ) are related to the diļ¬erential of the Gauss map. The Gauss map and the second fundamental form are deļ¬ned essentially as before: Deļ¬nition 5.15. Let U ā R2 be an open set, and let x : U ā M1,2 be a timelike immersion with image Ī£ = x(U ). The Gauss map of Ī£ = x(U ) is the map N : Ī£ ā M1,2 deļ¬ned by N (x(u)) = e2 (u), where (e0 (u), e1 (u), e2 (u)) is any adapted frame ļ¬eld on Ī£ = x(U ). (Note that N takes values in the āsphereā Sā1 .) Deļ¬nition 5.16. Let U ā R2 be an open set, and let x : U ā M1,2 be a timelike immersion. The second fundamental form of Ī£ = x(U ) is the quadratic form II on T U deļ¬ned by II(v) = āde2 (v), dx(v) for v ā Tu U , where (e0 (u), e1 (u), e2 (u)) is any adapted frame ļ¬eld on Ī£ = x(U ). *Exercise 5.17. Show that for any v ā Tu U , 2 II(v) = ā Ļ ĀÆ 0 (v) Ļ ĀÆ 0 (v) + Ļ ĀÆ 12 (v) Ļ ĀÆ 1 (v) = h00 (ĀÆ Ļ 0 (v))2 + 2h01 Ļ ĀÆ 0 (v) Ļ ĀÆ 1 (v) + h11 (ĀÆ Ļ 1 (v))2 . This is often written more concisely as II = ā(ĀÆ Ļ02 Ļ ĀÆ0 + Ļ ĀÆ 12 Ļ ĀÆ 1 ) = h00 (ĀÆ Ļ 0 )2 + 2h01 Ļ ĀÆ0 Ļ ĀÆ 1 + h11 (ĀÆ Ļ 1 )2 . (Hint: Although the result looks the same as in the Euclidean case, there are some sign diļ¬erences in the details of the computation. Note that, because of the slightly diļ¬erent symmetries in the (ĀÆ ĻĪ²Ī± ), we have de2 = e0 Ļ20 + e1 Ļ21 = e0 Ļ02 ā e1 Ļ12 . And donāt forget to use the Minkowski inner product in the deļ¬nition of II!) Now we will examine how the matrix [hĪ±Ī² ] changes if we vary the frame. Let (e0 (u), e1 (u), e2 (u)) be any adapted frame ļ¬eld along Ī£, with associated Ė1 (u), Maurer-Cartan forms (ĀÆ ĻĪ±, Ļ ĀÆ Ī²Ī± ). Any other adapted frame ļ¬eld (Ė e0 (u), e Ė e2 (u)) has the form (up to sign) (5.2) ā” ā¤ cosh(Īø) sinh(Īø) 0
ā¢ ā„ Ė0 (u) e Ė1 (u) e Ė2 (u) = e0 (u) e1 (u) e2 (u) ā£ sinh(Īø) cosh(Īø) 0ā¦ e 0 0 1
152
5. Curves and surfaces in Minkowski space
ĖĀÆ Ī± , Ļ ĖĀÆ Ī²Ī± ) be the Maurer-Cartan forms associfor some function Īø on U . Let (Ļ cosh(Īø) sinh(Īø) ated to the new frame ļ¬eld, and set B = . sinh(Īø) cosh(Īø) *Exercise 5.18 (Cf. Exercise 4.28). (a) Show that 0 0 ĖĀÆ Ļ Ļ ĀÆ ā1 =B . ĖĀÆ 1 Ļ Ļ ĀÆ1 (b) Show that
ĖĀÆ 20 Ļ
ĖĀÆ 21 Ļ
Ļ ĀÆ 20
= B ā1
Ļ ĀÆ 21
.
(Hint: Use the equation for de2 in (3.1).) Ė 00 , h Ė 01 , h Ė 11 on U such (c) Cartanās lemma implies that there exist functions h that 2 Ė Ė 01 Ļ ĖĀÆ 0 ĖĀÆ 0 Ļ h00 h =ā . Ė 01 h Ė 11 Ļ ĖĀÆ 12 ĖĀÆ 1 Ļ h Show that
Ė Ė 01 āh00 āh Ė 01 h
Ė 11 h
=B
ā1
āh00 āh01 h01
B.
h11
(Hint: Use the fact that Ļ ĀÆ 20 = Ļ ĀÆ 02 , Ļ ĀÆ 21 = āĀÆ Ļ12 .) (d) Use part (c) to show that Ė Ė 01 h00 h h00 h01 t (5.3) = B B. Ė 01 h Ė 11 h01 h11 h (Hint: Note that B ā1 is not equal to tB.) Recall that at this point in the Euclidean case, we used the fact that any symmetric matrix has an orthogonal basis of eigenvectors to conclude that we could choose B so as to diagonalize the matrix [hij ]. The key fact is not so much that the matrix [hij ] is symmetric (which depends on the fact that it is expressed relative to an orthonormal basis (e1 , e2 )), but rather that the linear operator de3 : Tx(u) Ī£ ā Tx(u) Ī£ is self-adjoint; i.e., for any two vectors v, w ā Tx(u) Ī£, we have de3 (v), w = v, de3 (w). The formal statement of the linear algebra theorem is that a self-adjoint operator on En has all real eigenvalues and an orthogonal basis of eigenvectors.
5.3. Moving frames for timelike surfaces in M1,2
153
Unfortunately, this result is not true in Minkowski space! The following exercise shows how this property can fail in the Minkowski setting. *Exercise 5.19. (a) Show that de2 = e0 Ļ ĀÆ 20 + e1 Ļ ĀÆ 21 = e0 (āh00 Ļ ĀÆ 0 ā h01 Ļ ĀÆ 1 ) + e1 (h01 Ļ ĀÆ 0 + h11 Ļ ĀÆ 1 ). (You probably did this as part of Exercise 5.18.) Conclude that the matrix of the linear transformation de2 : Tx(u) Ī£ ā Tx(u) Ī£ with respect to the basis (e0 (u), e1 (u)) for Tx(u) Ī£ is āh00 āh01 . h01 h11 (b) Show that the linear transformation de2 : Tx(u) Ī£ ā Tx(u) Ī£ is selfadjoint with respect to the Minkowski metric; that is, for any two vectors v, w ā Tx(u) Ī£, we have de2 (v), w = v, de2 (w). Conclude that the matrix S that expresses a self-adjoint operator on a (1+1)-dimensional Minkowski vector space V relative to an orthonormal basis (e0 , e1 ) for V has the property that the matrix S + tS is diagonal. (Contrast this with the Euclidean case, where any self-adjoint operator is expressed by a symmetric matrix S relative to an orthonormal basis.) (c) Show that: ā¢ If |h00 + h11 | > 2|h01 |, then de2 has distinct, real eigenvalues and an orthogonal basis of eigenvectors, one timelike and one spacelike. ā¢ If |h00 + h11 | < 2|h01 |, then de2 has complex eigenvalues and no real eigenvectors. ā¢ If |h00 + h11 | = 2|h01 | = 0, then de2 is a multiple of the identity transformation. ā¢ If |h00 + h11 | = 2|h01 | = 0, then de2 has a repeated real eigenvalue and a 1-dimensional, lightlike eigenspace. In order to make sense of the result of Exercise 5.19, we introduce the following analogs of the Euclidean notions of Gauss and mean curvature. Deļ¬nition 5.20. The function K = det(de2 ) on Ī£ is called the Gauss curvature of Ī£. The function H = ā 12 tr(de2 ) on Ī£ is called the mean curvature of Ī£. *Exercise 5.21. Show that K = h201 ā h00 h11 and H = 12 (h00 ā h11 ).
154
5. Curves and surfaces in Minkowski space
ā Deļ¬nition 5.22. The function H = H 2 ā K on Ī£ is called the skew curvature of Ī£. (By convention, if H 2 ā K < 0, then H is chosen so that iH < 0.) *Exercise 5.23. (a) Show that for a regular surface Ī£ in Euclidean space E3 , the skew curvature H is always real and H = 0 precisely at umbilic points. (b) Show that for a timelike, regular surface in Minkowski space M1,2 , 9 H = 12 (h00 + h11 )2 ā 4h201 . (c) Conclude from the result of Exercise 5.19 that at any point x(u) ā Ī£: ā¢ If H (u) is real and H (u) > 0, then d(e2 )x(u) has distinct, real eigenvalues and an orthogonal basis of eigenvectors, one timelike and one spacelike. ā¢ If H (u) is imaginary, then d(e2 )x(u) has complex eigenvalues and no real eigenvectors. ā¢ If H (u) = 0 and h01 = 0, then d(e2 )x(u) is a multiple of the identity transformation. ā¢ If H (u) = 0 and h01 = 0, then d(e2 )x(u) has a repeated real eigenvalue and a 1-dimensional, lightlike eigenspace. In light of this result, we will need to divide into cases based on the skew curvature in order to make further reļ¬nements to our adapted frame ļ¬eld. 5.3.1. Case 1: H (u) is real and d(e2 )x(u) has an orthogonal basis of eigenvectors for all u ā U . This assumption covers the ļ¬rst and third cases in Exercise 5.23, and the frame adaptation proceeds much as it did for surfaces in E3 . In this case, there exists an adapted frame (e0 (u), e1 (u), e2 (u)) at each point x(u) ā Ī£ with the property that āh00 (u) āh01 (u) 0 āĪŗ0 (u) = h01 (u) 0 āĪŗ1 (u) h11 (u) for some real numbers Īŗ0 (u), Īŗ1 (u). As in the Euclidean case, e0 (u) and e1 (u) are eigenvectors for d(e2 )u , with (5.4)
d(e2 )u (e0 (u)) = āĪŗ0 e0 (u),
d(e2 )u (e1 (u)) = āĪŗ1 e1 (u).
5.3. Moving frames for timelike surfaces in M1,2
155
We have the following analog of Deļ¬nition 4.30: Deļ¬nition 5.24. The vectors e0 (u) and e1 (u) in equations (5.4) are called principal vectors or principal directions at x(u) ā Ī£, and Īŗ0 (u), Īŗ1 (u) are called the principal curvatures of Ī£ at x(u). Such a frame (e0 (u), e1 (u), e2 (u)) is called a principal adapted frame at the point x(u) ā Ī£, and an adapted frame ļ¬eld on Ī£ which has this property at every point x(u) ā Ī£ is called a principal adapted frame ļ¬eld on Ī£. Exercise 5.25. Show that K = Īŗ0 Īŗ1 and H = 12 (Īŗ0 + Īŗ1 ). Deļ¬nition 5.26. If Īŗ0 (u) = Īŗ1 (u) for some point u ā U , then the corresponding point x(u) of Ī£ is called an umbilic point of Ī£. If Ī£ has no umbilic points, then a principal adapted frame ļ¬eld along Ī£ is determined uniquely (up to sign) since e0 must be timelike and e1 must be spacelike. *Exercise 5.27. Show that the umbilic points of Ī£ are precisely those points where H = 0. *Exercise 5.28. (a) Show that if Ī£ has no umbilic points, then a principal adapted frame ļ¬eld (e0 (u), e1 (u), e2 (u)) is equivariant (up to sign) under the action of M (1, 2). (b) Show that for a principal adapted frame ļ¬eld, the second fundamental form is II = Īŗ0 (ĀÆ Ļ 0 )2 ā Īŗ1 (ĀÆ Ļ 1 )2 . The analog of Bonnetās theorem (cf. Theorem 4.35) is the following: Theorem 5.29. Let U ā R2 be an open set. Two timelike immersions x1 , x2 : U ā M1,2 for which the diļ¬erential of the Gauss map has two linearly independent real eigenvectors diļ¬er by a Lorentz transformation if and only if they have the same ļ¬rst and second fundamental forms. The proof of this theorem (at least for surfaces with no umbilic points) is very similar to that in the Euclidean case. *Exercise 5.30. Prove Theorem 5.29 for surfaces without umbilic points as follows: Let x1 , x2 : U ā M1,2 be immersions as in Theorem 5.29, with the same ļ¬rst and second fundamental forms and Īŗ0 (u) = Īŗ1 (u) for every u ā U. Ė1, x Ė 2 : U ā M (1, 2) be principal adapted frame ļ¬elds for x1 , x2 . (a) Let x Let (ĀÆ ĻĪ±, Ļ ĀÆ Ī²Ī± ) denote the pulled-back dual and connection forms for x1 and
156
5. Curves and surfaces in Minkowski space
ĀÆ Ī±, Ī© ĀÆ Ī± ) denote those for x2 . Show that up to sign, let (Ī© Ī² ĀÆĪ± = Ļ Ī© ĀÆ Ī±,
ĀÆĪ± = Ļ Ī© ĀÆ Ī²Ī± . Ī²
Moreover, by making appropriate sign changes in the principal adapted frame ļ¬elds, we can arrange that these equations hold exactly. (b) Use Lemma 4.2 to conclude that x1 and x2 diļ¬er by a Lorentz transformation. *Exercise 5.31. Let x : U ā M1,2 be a timelike immersion as in Theorem 5.29, and let (u, v) be local coordinates on U . We can write the ļ¬rst and second fundamental forms as I = E du2 + 2F du dv + G dv 2 , II = e du2 + 2f du dv + g dv 2 for some functions E, F, G, e, f, g on U . The fact that Ī£ = x(U ) is timelike implies that EG ā F 2 < 0. Given a surface Ī£ of this type with no umbilic points, there exists a parametrization x : U ā M1,2 of Ī£ (at least locally) with the property that the coordinate curves of x are all principal curves, and such a parametrization has the property that F = f = 0, as in the Euclidean case. Without loss of generality, we may assume that E > 0 and G < 0. (Show that this condition implies that xu is timelike and xv is spacelike.) Follow the outline of Exercises 4.24 and 4.27 to derive the analogs of the Gauss and Codazzi equations for such surfaces. *Exercise 5.32. In this exercise, we will classify the totally umbilic timelike surfaces in M1,2 . So, suppose that x : U ā M1,2 is a timelike immersion as in Theorem 5.29, with the property that every point of Ī£ = x(U ) is umbilic. (a) Show that any adapted frame ļ¬eld (e0 (u), e1 (u), e2 (u)) is a principal adapted frame ļ¬eld. (b) Let (ĀÆ ĻĪ±, Ļ ĀÆ Ī²Ī± ) be the Maurer-Cartan forms for an adapted frame ļ¬eld on Ī£ = x(U ). Show that there exists a smooth function Ī» : U ā R such that Ļ ĀÆ 02 = āĪ»ĀÆ Ļ0,
Ļ ĀÆ 12 = Ī»ĀÆ Ļ1.
Conclude that the second fundamental form of Ī£ is a scalar multiple of the ļ¬rst fundamental form, i.e., that II = Ī»I. (c) Prove that Ī» is constant. (Hint: Use the structure equations to differentiate the equations above, taking into account the fact that we must
5.3. Moving frames for timelike surfaces in M1,2
157
have dĪ» = Ī»0 Ļ ĀÆ 0 + Ī»1 Ļ ĀÆ1 for some functions Ī»0 , Ī»1 on U . Then use Cartanās lemma.) (d) Show that if Ī» = 0, then de2 = 0. Conclude that the normal vector ļ¬eld of Ī£ is constant and that Ī£ is therefore contained in a plane. (e) Show that if Ī» = 0, then d(x+ Ī»1 e2 ) = 0. Conclude that the vector-valued function x + Ī»1 e2 : U ā M1,2 is equal to some constant point q ā M1,2 and that Ī£ is therefore contained in the hyperboloid of one sheet deļ¬ned by the equation 1 x ā q, x ā q = ā 2 . |Ī»| (Note that the minus sign on the right-hand side is due to the fact that e2 is a spacelike vector and that this hyperboloid is actually the āsphereā Sā 1 .) Ī»2
Thus, the only totally umbilic timelike surfaces are (open subsets of) planes and hyperboloids of one sheet. Example 5.33 (De Sitter spacetime). In relativity, the de Sitter spacetime dS4 is perhaps the simplest non-ļ¬at model for general relativity. It can be deļ¬ned as the set of spacelike unit vectors in the Minkowski space M1,4 : dS4 = {x ā M1,4 | x, x = ā1}. De Sitter space is the maximally symmetric vacuum solution of Einsteinās ļ¬eld equations with a positive cosmological constant. It represents a universe that is spatially homogeneous, diļ¬eomorphic to a 3-dimensional sphere (and hence compact), and expanding in size for t > 0. A 2-dimensional analog is the de Sitter spacetime dS2 , which is the āsphereā Sā1 of spacelike unit vectors in M1,2 and is a totally umbilic surface as in Exercise 5.32. It represents a 1-dimensional universe that is diļ¬eomorphic to a circle. *Exercise 5.34. We can parametrize dS2 by the map x : R2 ā M1,2 given by (5.5)
x(u, v) = t[sinh(u), cosh(u) cos(v), cosh(u) sin(v)] ,
where u should be thought of as a time parameter and v as a spatial parameter. (a) Show that the metric on dS2 corresponding to the parametrization (5.5) has E = 1, F = 0, G = ā cosh2 (u). (Cf. Exercise 5.31.)
158
5. Curves and surfaces in Minkowski space
(b) Show that the circumference of the space at time u = Ļ is & Ļ xv (Ļ, v) dv = 2Ļ cosh(Ļ ). āĻ
Thus, the āradiusā of the space at time Ļ is cosh(Ļ ), which grows exponentially as Ļ increases for Ļ > 0. Now consider the possible world lines for particles traveling in the de Sitter space dS2 . Since dS2 is spatially ļ¬nite, we might think that a moving particle should be able to ācircumnavigateā the space and return to its original position. But in fact this is not possible, as the following exercise shows. *Exercise 5.35. Consider the path Ī² of a photon emitted at the point x(0, 0) and traveling in the positive v direction. We can parametrize the photonās world line as Ī²(t) = x(t, v(t)) for some function v(t) which is determined by the conditions that v(0) = 0, v (t) > 0, and Ī² (t) is a lightlike vector. (a) Show that the lightlike tangent directions to dS2 at the point x(u0 , v0 ) are spanned by the vectors vĀ± = cosh(u0 ) xu Ā± xv . (b) Show that the function v(t) above must satisfy the diļ¬erential equation v (t) = sech(t) and that the solution to this equation with v(0) = 0 and v (t) > 0 is Ļ v(t) = 2 tanā1 et ā . 2 (c) Show that
Ļ Ļ , lim v(t) = ā . tā+ā tāāā 2 2 Therefore, no photon ever travels more than halfway around the circle. (And, of course, no other particle can travel farther than a photon!) lim v(t) =
5.3.2. Case 2: H (u) is imaginary for all u ā U . This assumption covers the second case in Exercise 5.23. *Exercise 5.36. (a) Show that under a transformation of the form (5.3), Ė 00 + h Ė 11 = cosh(2Īø)(h00 + h11 ) + 2 sinh(2Īø)h01 . h Conclude that there exists an adapted frame ļ¬eld (e0 (u), e1 (u), e2 (u)) with the property that h00 + h11 = 0,
5.3. Moving frames for timelike surfaces in M1,2
and therefore,
h00 h01
=
h01 h11
Ī»0
Ī»1
159
Ī»1 āĪ»0
for some functions Ī»0 , Ī»1 on U with |Ī»1 | > 0 (cf. Exercise 5.18). (b) Show that the Maurer-Cartan forms associated to such a frame ļ¬eld satisfy Ļ ĀÆ 02 = āĪ»0 Ļ ĀÆ 0 ā Ī»1 Ļ ĀÆ 1, Ļ ĀÆ 12 = āĪ»1 Ļ ĀÆ 0 + Ī»0 Ļ ĀÆ 1. (c) Show that for such a frame ļ¬eld, the second fundamental form is 0 2 II = Ī»0 (ĀÆ Ļ ) ā (ĀÆ Ļ 1 )2 + 2Ī»1 Ļ ĀÆ 0Ļ ĀÆ 1 = Ī»0 I + 2Ī»1 Ļ ĀÆ 0Ļ ĀÆ 1. (d) What goes wrong with part (a) when H = 0 and h01 = 0? In order to get a feel for what such a surface Ī£ might look like, consider the case where Ī£ is a graph of the form x2 = f (x0 , x1 ). (Any timelike surface can locally be described in this way, possibly after a rotation in the (x1 , x2 )-plane.) Consider a parametrization x : U ā M1,2 of the form x(u, v) = t[u, v, f (u, v)]. Exercise 5.37. Show that Ī£ is timelike if and only if |fu | < 1. Now choose a point (u0 , v0 ) ā U , and ārotateā the surface (via a Lorentzian transformation) if necessary so that fu (u0 , v0 ) = fv (u0 , v0 ) = 0, so that the tangent plane to Ī£ at the point x0 = x(u0 , v0 ) is parallel to the (x0 , x1 )-plane. Exercise 5.38. (a) Show that the normal vector ļ¬eld along Ī£ is e2 (u, v) =
1
[āfu , fv , ā1].
t
1 + fv2 ā fu2
In particular, e2 (u0 , v0 ) = t[0, 0, ā1]. (b) Choose an adapted frame ļ¬eld along Ī£ so that e0 (u0 , v0 ) = t[1, 0, 0],
e1 (u0 , v0 ) = t[0, 1, 0],
160
5. Curves and surfaces in Minkowski space
and let (ĀÆ ĻĪ±, Ļ ĀÆ Ī²Ī± ) be the associated Maurer-Cartan forms. Show that at the point x0 , Ļ ĀÆ 0 = du, Ļ ĀÆ 02 = āfuu du ā fuv dv,
Ļ ĀÆ 1 = dv,
Therefore,
h00 h01 h01 h11
Ļ ĀÆ 12 = āfuv du ā fvv dv.
= (u0 ,v0 )
fuu (u0 , v0 ) fuv (u0 , v0 )
fuv (u0 , v0 ) fvv (u0 , v0 )
.
(c) Conclude that the graph of the function f (u, v) = a(u2 ā v 2 ) + 2buv, where a, b are constants with |b| > 0, is one example of a surface of this type. (See Figure 5.1; note that the e0 -axis is drawn as the vertical axis.)
Figure 5.1. A timelike surface with H imaginary
5.3.3. Case 3: H (u) = 0 and h01 (u) = 0 for all u ā U . Note that this is a degenerate condition, and generically it will only hold along a closed subset of Ī£, similar to the set of umbilic points for a surface in Case 1. Deļ¬nition 5.39. A point x(u) in a timelike surface Ī£ = x(U ) in M1,2 will be called a quasi-umbilic point if H (u) = 0 and h01 (u) = 0. If every point of Ī£ is quasi-umbilic, then Ī£ is called totally quasi-umbilic. Exercise 5.40. Suppose that x : U ā M1,2 is a totally quasi-umbilic timelike surface. (a) Show that |h01 | = 12 |h00 + h11 | = 0.
5.4. An alternate construction for timelike surfaces
161
(b) Show that under a transformation of the form (5.3), Ė 01 = e2Īø h01 . h Conclude that there exists an adapted frame ļ¬eld (e0 (u), e1 (u), e2 (u)) with the property that |h01 | = 12 |h00 + h11 | = 1, and therefore, h00 h01 Īµ1 + Ī» Īµ2 = , h01 h11 Īµ2 Īµ1 ā Ī» where Ī» is a function on U and Īµi = Ā±1, i = 1, 2. (In fact, Ī» is equal to the mean curvature H of Ī£.) (c) Show that the Maurer-Cartan forms associated to such a frame ļ¬eld satisfy Ļ ĀÆ 02 = ā(Īµ1 + Ī»)ĀÆ Ļ 0 ā Īµ2 Ļ ĀÆ 1, Ļ ĀÆ 12 = āĪµ2 Ļ ĀÆ 0 ā (Īµ1 ā Ī»)ĀÆ Ļ1. (d) Show that for such a frame ļ¬eld, the second fundamental form is 0 2 0 2 II = Ī» (ĀÆ Ļ ) ā (ĀÆ Ļ ) + (ĀÆ Ļ 1 )2 + Īµ1 (ĀÆ Ļ 1 )2 + 2Īµ2 Ļ ĀÆ 0Ļ ĀÆ1 0 2 Ļ ) + (ĀÆ = Ī»I + Īµ1 (ĀÆ Ļ 1 )2 + 2Īµ2 Ļ ĀÆ 0Ļ ĀÆ 1. By analogy with the totally umbilic surfaces, one might expect that the totally quasi-umbilic surfaces would form a fairly limited set. But in fact, the analysis of totally quasi-umbilic surfaces is considerably more involved than that for totally umbilic surfaces, and it turns out that there is an inļ¬nite-dimensional family of such surfaces! We will explore some examples in Exercise 5.49.
5.4. An alternate construction for timelike surfaces In Ā§5.3, we used orthonormal frame ļ¬elds to study timelike surfaces in M1,2 . This is in keeping with our general perspective that a choice of frame ļ¬eld along a surface x : U ā G/H in a homogeneous space G/H should deļ¬ne Ė : U ā G to a surface in the Lie group G. However, this is not a lifting x always the most geometrically natural way to choose frames, and in this section, we will explore an alternate method for constructing moving frames for timelike surfaces in M1,2 . Let Ī£ = x(U ) be a timelike surface in M1,2 . Since the ļ¬rst fundamental form of a timelike surface has signature (1, 1), there are two linearly independent null (i.e., lightlike) directions in each tangent space Tx(u) Ī£. In this section,
162
5. Curves and surfaces in Minkowski space
we will construct frame ļ¬elds (f1 (u), f2 (u), f3 (u)) along Ī£ = x(U ) with the property that f1 (u) and f2 (u) are null vectors. Exercise 5.41. Suppose that (e0 (u), e1 (u), e2 (u)) is an orthonormal frame ļ¬eld along Ī£ = x(U ). Show that the vectors f1 (u) =
ā1 (e0 (u) 2
+ e1 (u)),
f2 (u) =
ā1 (e0 (u) 2
ā e1 (u))
are linearly independent null vectors and that f1 (u), f2 (u) = 1. Deļ¬nition 5.42. A null adapted frame ļ¬eld (f1 (u), f2 (u), f3 (u)) along a timelike surface Ī£ = x(U ) is a basis for the tangent space Tx(u) M1,2 with the properties that (1) (f1 (u), f2 (u)) are null vectors that span the tangent space Tx(u) Ī£ at each point x(u) ā Ī£; (2) f1 (u), f2 (u) = 1; (3) f3 (u) is the normal vector f3 (u) = f1 (u) Ć f2 (u) to Tx(u) Ī£ at each point x(u) ā Ī£. Now, let (ĀÆ Ī· i , Ī·ĀÆji ) be the Maurer-Cartan forms associated to a null adapted frame ļ¬eld along Ī£. These forms are deļ¬ned just as in (3.1): dx = fi Ī·ĀÆi , dfi = fj Ī·ĀÆij , where 1 ā¤ i, j ā¤ 3, and they still satisfy the structure equations dĀÆ Ī· i = āĀÆ Ī·ji ā§ Ī·ĀÆj , dĀÆ Ī·ji = āĀÆ Ī·ki ā§ Ī·ĀÆjk . The main diļ¬erence is that the vectors (f1 , f2 , f3 ) satisfy the somewhat unusual inner product relations (5.6)
f1 , f1 = 0,
f2 , f2 = 0,
f3 , f3 = ā1,
f1 , f2 = 1,
f1 , f3 = 0,
f2 , f3 = 0.
*Exercise 5.43. Diļ¬erentiate the inner product relations (5.6) to obtain the following relations among the (ĀÆ Ī·ji ): (5.7)
Ī·ĀÆ21 = Ī·ĀÆ12 = Ī·ĀÆ33 = 0, Ī·ĀÆ22 = āĀÆ Ī·11 ,
Ī·ĀÆ31 = Ī·ĀÆ23 ,
Ī·ĀÆ32 = Ī·ĀÆ13 .
5.4. An alternate construction for timelike surfaces
163
The same reasoning as before can be used to prove the following: Proposition 5.44. Let U ā R2 be an open set, and let x : U ā M1,2 be a timelike immersion. For any null adapted frame ļ¬eld (f1 (u), f2 (u), f3 (u)) along Ī£ = x(U ), the associated dual and connection forms (ĀÆ Ī· i , Ī·ĀÆji ) have the property that Ī·ĀÆ3 = 0. Moreover, (ĀÆ Ī· 1 , Ī·ĀÆ2 ) form a basis for the 1-forms on U . *Exercise 5.45. Show that the ļ¬rst fundamental form of Ī£ = x(U ) (cf. Deļ¬nition 5.13) is I = 2ĀÆ Ī· 1 Ī·ĀÆ2 . *Exercise 5.46. (a) Diļ¬erentiate the equation Ī·ĀÆ3 = 0 and use Cartanās lemma to conclude that there exist functions k11 , k12 , k22 on U such that 3 Ī·ĀÆ1 k11 k12 Ī·ĀÆ1 (5.8) =ā . Ī·ĀÆ23 k12 k22 Ī·ĀÆ2 (b) The Gauss map (cf. Deļ¬nition 5.15) of Ī£ = x(U ) is the map N : Ī£ ā M1,2 deļ¬ned by N (x(u)) = f3 (u). Show that df3 = f1 Ī·ĀÆ31 + f2 Ī·ĀÆ32 = āf1 (k12 Ī·ĀÆ1 + k22 Ī·ĀÆ2 ) ā f2 (k11 Ī·ĀÆ1 + k12 Ī·ĀÆ2 ). Conclude that with respect to the basis (f1 (u), f2 (u)) for Tx(u) Ī£, the matrix of the linear transformation df3 : Tx(u) Ī£ ā Tx(u) Ī£ is k12 k22 ā . k11 k12 (c) The second fundamental form (cf. Deļ¬nition 5.16) of Ī£ = x(U ) is deļ¬ned by II(v) = ādf3 (v), dx(v) for v ā Tu U . Show that II = k11 (ĀÆ Ī· 1 )2 + 2k12 Ī·ĀÆ1 Ī·ĀÆ2 + k22 (ĀÆ Ī· 2 )2 . (d) Show that the Gauss and mean curvatures (cf. Deļ¬nition 5.20) of Ī£ are 2 K = k12 ā k11 k22 ,
H = k12
and that the skew curvature (cf. Deļ¬nition 5.22) is
H = k11 k22 .
164
5. Curves and surfaces in Minkowski space
Conclude that the analogous conditions to those in Exercise 5.19(c) are ā¢ k11 k22 > 0 (H real and H > 0); ā¢ k11 k22 < 0 (H imaginary); ā¢ k11 = k22 = 0 (umbilic points); ā¢ k11 k22 = 0, and exactly one of k11 , k22 vanishes (quasi-umbilic points). Now we will examine how the matrix [kij ] changes if we vary the frame. *Exercise 5.47. Let (f1 (u), f2 (u), f3 (u)) be any null adapted frame ļ¬eld along Ī£, with associated Maurer-Cartan forms (ĀÆ Ī· i , Ī·ĀÆji ). (a) Show that any other null adapted frame ļ¬eld (Ėf1 (u), Ėf2 (u), Ėf3 (u)) has the form (up to sign) ā” Īø ā¤ 0 0 e
ā„ Ėf1 (u) Ėf2 (u) Ėf3 (u) = f1 (u) f2 (u) f3 (u) ā¢ (5.9) ā£ 0 eāĪø 0ā¦ 0
0
1
for some function Īø on U . (Hint: Consider the transformation (5.2) for the adapted orthonormal frame ļ¬eld e0 (u) =
ā1 (f1 (u) 2
+ f2 (u)),
e1 (u) =
ā1 (f1 (u) 2
ā f2 (u)),
e2 (u) = f3 (u).)
(b) Let (Ī·ĖĀÆi , Ī·ĖĀÆji ) be the Maurer-Cartan forms associated to the new frame eĪø 0 ļ¬eld, and set B = . Show that 0 eāĪø 1 1 āĪø 1 1 1 āĪø 1 Ī·ĖĀÆ Ī·ĀÆ e Ī·ĀÆ Ī·ĖĀÆ3 Ī·ĀÆ3 e Ī·ĀÆ3 ā1 ā1 (5.10) =B = , =B = . Ī·ĖĀÆ2 Ī·ĖĀÆ32 Ī·ĀÆ2 Ī·ĀÆ32 eĪø Ī·ĀÆ2 eĪø Ī·ĀÆ32 (c) Cartanās lemma implies that there exist functions kĖ11 , kĖ12 , kĖ22 on U such that 3 Ī·ĖĀÆ1 kĖ11 kĖ12 Ī·ĖĀÆ1 =ā . Ī·ĖĀÆ23 kĖ12 kĖ22 Ī·ĖĀÆ2 Show that
Ė k12 kĖ22 kĖ11 kĖ12
= B ā1
k12 k22 k11 k12
B
5.4. An alternate construction for timelike surfaces
and that (5.11)
Ė k11 kĖ12 kĖ12 kĖ22
= tB
k11 k12
B=
k12 k22
165
e2Īø k11
k12
k12
eā2Īø k22
.
The result of Exercise 5.47 provides a guide as to how to choose a canonical null adapted frame ļ¬eld in most cases: (1) If H = 0, then there exists a null adapted frame ļ¬eld (f1 (u), f2 (u), f3 (u)) (unique up to sign) along Ī£ with the property that k11 = Ā±k22 . The sign is positive if H is real and negative if H is imaginary. (2) It may not be possible to choose such a frame ļ¬eld continuously near umbilic or quasi-umbilic points. But in a totally quasi-umbilic neighborhood where, without loss of generality, k22 = 0 and k11 = 0, there exists a null adapted frame ļ¬eld (f1 (u), f2 (u), f3 (u)) (unique up to sign) along Ī£ with the property that k11 = Ā±1. *Exercise 5.48 (Cf. Exercise 5.31). A timelike surface Ī£ always has a local parametrization x : U ā M1,2 in terms of null coordinates. Such a parametrization has the property that I = 2F du dv for some function F > 0 on U . We can write the second fundamental form as II = e du2 + 2f du dv + g dv 2 for some functions e, f, g on U . Follow the outline of Exercises 4.24 and 4.27 to derive the analogs of the Gauss and Codazzi equations for timelike surfaces in null coordinates. Exercise 5.49. Let Ī± : I ā M1,2 be a nondegenerate null curve, i.e., a curve with the property that Ī± (u) is a null vector for all u ā I. (The condition that Ī± is nondegenerate means that Ī± (u) and Ī± (u) are linearly independent for all u ā I.) Let f 0 be any null vector that is linearly independent from Ī± (u) for all u ā I, and consider the cylindrical surface Ī£ with parametrization x : I Ć R ā M1,2 given by x(u, v) = Ī±(u) + vf 0 . (a) Show that Ī± can be reparametrized so as to satisfy the condition Ī± (u), f 0 = 1.
166
5. Curves and surfaces in Minkowski space
(b) With Ī± as in part (a), show that the vector ļ¬elds f1 (u, v) = Ī± (u),
f3 (u, v) = f1 (u, v) Ć f2 (u, v)
f2 (u, v) = f 0 ,
form a null adapted frame ļ¬eld along Ī£. (c) Show by direct computation of dx, df2 , and df1 (preferably in that order!) that the Maurer-Cartan forms associated to this frame ļ¬eld satisfy Ī·ĀÆ1 = du, Ī·ĀÆ31 = Ī·ĀÆ23 = 0,
Ī·ĀÆ2 = dv, Ī·ĀÆ32 = Ī·ĀÆ13 = Ī» du
for some nonvanishing function Ī» : I ā R. (d) Conclude that k11 = āĪ», k12 = 0, k22 = 0 and that Ī£ is therefore totally quasi-umbilic. (e) Show that Ī£ has Gauss and mean curvatures K ā” 0,
H ā” 0.
This exercise shows that, unlike in the Euclidean case, there exist non-planar timelike surfaces in M1,2 whose Gauss and mean curvatures are identically zero! Conversely, it can be shown that if Ī£ ā M1,2 is a timelike surface with K ā” H ā” 0, then every point of Ī£ is either umbilic or quasi-umbilic; see [Cle12] for details.
5.5. Maple computations The Maple setup for the orthonormal frame approach in this chapter is much the same as in Chapter 4, except that indices now range from 0 to 2 and the symmetries of the connection forms are slightly diļ¬erent. Likewise, Exercise 5.18 is very similar to Exercise 4.28; aside from the adjustments in the index ranges, the only change is in the matrix B. We leave these modiļ¬cations as an exercise for the reader. (More details are given in the Maple worksheet for this chapter on the AMS webpage. Here, we will explore how to do analogous computations for a null adapted frame ļ¬eld along a timelike surface in M1,2 . After loading the Cartan and LinearAlgebra packages into Maple, begin by declaring the necessary 1forms: > Form(eta[1], eta[2], eta[3]); Form(eta[1,1], eta[3,1], eta[3,2]);
5.5. Maple computations
167
Tell Maple about the symmetries in the connection forms: > eta[1,2]:= eta[2,1]:= eta[3,3]:= eta[2,2]:= eta[1,3]:= eta[2,3]:=
0; 0; 0; -eta[1,1]; eta[3,2]; eta[3,1];
Tell Maple how to diļ¬erentiate these forms according to the Cartan structure equations (3.8): > for i from 1 to 3 do d(eta[i]):= -add(āeta[i,j] &Ė eta[j]ā, j=1..3); end do; d(eta[1,1]):= -add(āeta[1,k] &Ė eta[k,1]ā, k=1..3); d(eta[3,1]):= -add(āeta[3,k] &Ė eta[k,1]ā, k=1..3); d(eta[3,2]):= -add(āeta[3,k] &Ė eta[k,2]ā, k=1..3); Set up a substitution for the Maurer-Cartan forms of a null adapted frame ļ¬eld, taking into account the relations (5.8) that result from computing dĀÆ Ī· 3 = 0: > adaptedsub1:= [eta[3]=0, eta[3,1] = -(k[1,1]*eta[1] + k[1,2]*eta[2]), eta[3,2] = -(k[1,2]*eta[1] + k[2,2]*eta[2])]; Exercise 5.47: Introduce new 1-forms to represent the transformed forms, with the same symmetry conditions as the original forms: > Form(Eta[1], Eta[2], Eta[3]); Form(Eta[1,1], Eta[3,1], Eta[3,2]); Eta[1,2]:= 0; Eta[2,1]:= 0; Eta[3,3]:= 0; Eta[2,2]:= -Eta[1,1]; Eta[1,3]:= Eta[3,2]; Eta[2,3]:= Eta[3,1]; Introduce the relations (5.10) via the following substitution, along with its reverse substitution: > framechangesub:= [Eta[1] = exp(-theta)*eta[1], Eta[2] = exp(theta)*eta[2], Eta[3,1] = exp(theta)*eta[3,1], Eta[3,2] = exp(-theta)*eta[3,2]]; > framechangebacksub:= makebacksub(framechangesub);
168
5. Curves and surfaces in Minkowski space
In order to compare the functions (kĖij ) to the functions (kij ), introduce another substitution describing the adaptations of the transformed frame: > adaptedsub2:= [Eta[3]=0, Eta[3,1] = -(K[1,1]*Eta[1] + K[1,2]*Eta[2]), Eta[3,2] = -(K[1,2]*Eta[1] + K[2,2]*Eta[2])]; Now combine all these substitutions to see how the (kĖij ) are expressed in terms of the (kij ): > zero2:= Simf(subs(adaptedsub2, Eta[3,1]) - Simf(subs(framechangebacksub, Simf(subs(adaptedsub1, Simf(subs(framechangesub, Eta[3,1]))))))); > eqns:= {op(ScalarForm(zero2))}; > zero3:= Simf(subs(adaptedsub2, Eta[3,2]) - Simf(subs(framechangebacksub, Simf(subs(adaptedsub1, Simf(subs(framechangesub, Eta[3,2]))))))); > eqns:= eqns union {op(ScalarForm(zero3))}; > solve(eqns, {K[1,1], K[1,2], K[2,2]}); Exercise 5.48: Start by declaring that F, e, f, g are functions of u and v: > PDETools[declare](F(u,v), e(u,v), f(u,v), g(u,v)); For a null parametrization x : U ā M1,2 of a timelike surface with F > 0, we can deļ¬ne a null adapted frame ļ¬eld by 1 f1 (u) = ā xu , F
1 f2 (u) = ā xv , F
The associated dual forms are ā Ī·ĀÆ1 = F du,
Ī·ĀÆ2 =
ā
F dv,
f3 (u) = f1 (u) Ć f2 (u).
Ī·ĀÆ3 = 0,
and in order for the second fundamental form to have the desired form, we must have " " # # e f f g 3 3 Ī·ĀÆ1 = ā ā du + ā dv , Ī·ĀÆ2 = ā ā du + ā dv . F F F F Introduce a substitution for the Maurer-Cartan forms in terms of the coordinate 1-forms: > coordsub:= [eta[3]=0, eta[1] = sqrt(F(u,v))*d(u), eta[2] = sqrt(F(u,v))*d(v), eta[3,1] = -((e(u,v)/sqrt(F(u,v)))*d(u) + (f(u,v)/sqrt(F(u,v)))*d(v)), eta[3,2] = -((f(u,v)/sqrt(F(u,v)))*d(u)
5.5. Maple computations
169
+ (g(u,v)/sqrt(F(u,v)))*d(v)), eta[1,1] = a*d(u) + b*d(v)]; Use the structure equations for dĀÆ Ī· 1 and dĀÆ Ī· 2 to compute the coeļ¬cients in 1 Ī·1 : > Simf(d(Simf(subs(coordsub, eta[1]))) - subs(coordsub, Simf(d(eta[1])))); > b:= solve(%, b); > Simf(d(Simf(subs(coordsub, eta[2]))) - subs(coordsub, Simf(d(eta[2])))); > a:= solve(%, a); The Gauss equation comes from comparing the two expressions for dĀÆ Ī·11 : > Simf(d(Simf(subs(coordsub, eta[1,1]))) - subs(coordsub, Simf(d(eta[1,1])))); > Gausseq:= pick(%, d(u), d(v)); The Gauss curvature is deļ¬ned to be K = f Fāeg 2 , and from the output of the last computation we see that the Gauss equation takes the form 1 (ln F )uv K = 3 (F Fuv ā Fu Fv ) = . F F 2
Similar manipulations involving dĀÆ Ī·13 and dĀÆ Ī·23 show that the Codazzi equations take the form " # " # F fu ā f Fu F fv ā f Fv f f ev = , gu = . =F =F F F u F F v
10.1090/gsm/178/06
Chapter 6
Curves and surfaces in equi-aļ¬ne space
6.1. Introduction Now we will apply the method of moving frames to study the geometry of curves and surfaces in equi-aļ¬ne space. The lack of a metric structure will lead to results that may seem less intuitive than those of the last two chapters, but the general procedure for constructing invariants remains the same. And rather than rediscovering the familiar invariants for curves and surfaces in Euclidean space, we will see how this method naturally leads to an alternative set of invariants that are preserved under the action of the entire equi-aļ¬ne group. Exercise 6.1. Show by example that an equi-aļ¬ne transformation T : R3 ā R3 does not necessarily preserve the curvature Īŗ(s) and torsion Ļ (s) of a curve Ī± : I ā E3 . (Hint: Let Ī± be something simple, like a helix, and consider a diagonal transformation such as
T (x1 , x2 , x3 ) = t 2x1 , 12 x2 , x3 .) A remark about the notation: A curve in R3 may be regarded as a curve in either E3 or A3 , depending on what symmetry group we allow to act on R3 . In general, an equi-aļ¬ne transformation acting on a curve in E3 will transform the curve to another curve that is not equivalent to the original curve under the action of the Euclidean group. 171
172
6. Curves and surfaces in equi-aļ¬ne space
6.2. Moving frames for curves in A3 Consider a smooth, parametrized curve Ī± : I ā A3 that maps some open interval I ā R into equi-aļ¬ne space. A3 has the structure of the homogeneous space A(3)/SL(3), so an adapted frame ļ¬eld along Ī± should be a lifting Ī± Ė : I ā A(3). Any such lifting can be written as Ī± Ė (t) = (Ī±(t); e1 (t), e2 (t), e3 (t)), where for each t ā I, (e1 (t), e2 (t), e3 (t)) is a unimodular basis for the tangent space TĪ±(t) A3 . Such an adapted frame ļ¬eld is called a unimodular frame ļ¬eld along Ī±. The situation is quite diļ¬erent from that of either Euclidean or Minkowski space; for instance, there is no obvious notion of arc length for a curve that is invariant under the action of A(3). Moreover, we have much greater freedom in choosing our frame; the only requirement is that det[e1 (t) e2 (t) e3 (t)] = 1. So, how should we proceed? In the Euclidean case, we used the ļ¬rst derivative of Ī± to choose e1 (t) and the second derivative of Ī± to choose e2 (t), pausing along the way to normalize according to arc length so that the frame would be orthonormal. These choices determined e3 (t) uniquely, but it is clear from the structure equations that e3 (t) is related to the third derivative of Ī±. In order for this procedure to work, we had to assume that Ī± was ānondegenerateā, i.e., that the vectors (Ī± (t), Ī± (t)) were linearly independent for each t ā I. So, how might we use similar reasoning to construct a canonical adapted frame ļ¬eld in the equi-aļ¬ne case, and what would be the right notion of ānondegenerateā for equi-aļ¬ne curves? Since orthonormality is no longer required, our ļ¬rst guess towards constructing an adapted frame ļ¬eld might be to take e1 (t) = Ī± (t), e2 (t) = Ī± (t), e3 (t) = Ī± (t). In order for this to work, we must assume that the vectors (Ī± (t), Ī± (t), Ī± (t)) are linearly independent for each t ā I; such a curve will be called nondegenerate. (Note that this word means something diļ¬erent for curves in equi-aļ¬ne space than for curves in Euclidean space!) For nondegenerate curves, the only problem with this choice of frame ļ¬eld is that it is not
6.2. Moving frames for curves in A3
173
necessarily unimodular. But we could ļ¬x this by deļ¬ning the adapted frame ļ¬eld to be e1 (t) =
3 (6.1)
e2 (t) =
3 e3 (t) =
3
Ī± (t) det[Ī± (t) Ī± (t) Ī± (t)] Ī± (t) det[Ī± (t) Ī± (t) Ī± (t)] Ī± (t) det[Ī± (t) Ī± (t) Ī± (t)]
, , .
Exercise 6.2. Prove that this choice of frame ļ¬eld (e1 (t), e2 (t), e3 (t)) is equivariant under the action of A(3): If we replace Ī± by g Ā· Ī± for some g ā A(3), then ei (t) ā TĪ±(t) A3 will be replaced by (Lg )ā (ei (t)) ā TgĀ·Ī±(t) . Now, wouldnāt it be nice to get rid of that ugly denominator? In the Euclidean case, we were able to get rid of the denominator for e1 (t) by reparametrizing according to arc length. So letās see if we can ļ¬nd a suitable reparametrization to do the trick here; speciļ¬cally, we would like to ļ¬nd a reparametrization Ī±(s) of Ī± for which det[Ī± (s) Ī± (s) Ī± (s)] = 1. This will be a bit more complicated than in the Euclidean case since the denominators in (6.1) involve the ļ¬rst three derivatives of Ī± instead of just the ļ¬rst derivative. Suppose that we reparametrize the curve by setting Ī±(s) = Ī±(t(s)) for some invertible function t(s). Then dĪ± dĪ± = t (s) , ds dt (6.2)
2 d2 Ī± 2 d Ī± ā” t (s) ds2 dt2
mod
dĪ± , ds
d3 Ī± d3 Ī± ā” t (s)3 3 3 ds dt
mod
dĪ± d2 Ī± . , ds ds2 2
d Ī± Note that it doesnāt matter what multiple of dĪ± ds we are ignoring in ds2 ā 2 3 d Ī± d Ī± and similarly for the dĪ± ds , ds terms in ds2 ābecause they wonāt aļ¬ect the determinant det[Ī± (s) Ī± (s) Ī± (s)]. Therefore,
det[Ī± (s) Ī± (s) Ī± (s)] = t (s)6 det[Ī± (t(s)) Ī± (t(s)) Ī± (t(s))]. Note that the sign of det[Ī± (s) Ī± (s) Ī± (s)] is ļ¬xed, so the best that we can hope for is to arrange that det[Ī± (s) Ī± (s) Ī± (s)] = Ā±1. This suggests that
174
6. Curves and surfaces in equi-aļ¬ne space
we make the following deļ¬nition: Deļ¬nition 6.3. Let Ī± : I ā A3 be a nondegenerate curve, and ļ¬x t0 ā I. The function & t9 6 s(t) = det[Ī± (u) Ī± (u) Ī± (u)] du t0
is called the equi-aļ¬ne arc length function along Ī±. *Exercise 6.4. (a) Show that any nondegenerate curve Ī± : I ā A3 can be smoothly reparametrized by its equi-aļ¬ne arc length. (b) Show that if Ī± is parametrized by equi-aļ¬ne arc length s, then det[Ī± (s) Ī± (s) Ī± (s)] = Ā±1. Remark 6.5. As for curves in Euclidean space, it may or may not be possible to ļ¬nd an explicit parametrization by equi-aļ¬ne arc length. The existence of such a parametrization is of more importance for developing the geometric theory than for working with explicit examples. Exercise 6.6. Show that for a nondegenerate curve Ī± : I ā A3 , the equiaļ¬ne arc length s(t) is invariant under the action of A(3); that is, for any g ā A(3), the curves Ī± and gĀ·Ī± have the same equi-aļ¬ne arc length function. Equi-aļ¬ne arc length is a very diļ¬erent notion from Euclidean arc length. Some of the most notable diļ¬erences are: (1) Unlike Euclidean arc length, which depends only on the ļ¬rst derivative of Ī±, the equi-aļ¬ne arc length depends on the ļ¬rst three derivatives of Ī±. In fact, this number is dependent on the dimension of the ambient equi-aļ¬ne space: The equi-aļ¬ne arc length of a curve Ī± : I ā An depends on the ļ¬rst n derivatives of Ī±. (2) The equi-aļ¬ne arc length is only nonzero for nondegenerate curves; so, for instance, any curve contained in a plane in A3 has equi-aļ¬ne arc length zero according to this deļ¬nition. It could, however, have nonzero equi-aļ¬ne arc length when regarded as a curve in A2 . (3) There is no ambient metricāEuclidean or otherwiseāon A3 whose restriction to a curve gives its equi-aļ¬ne arc length, as there is for curves in Euclidean space. This makes it diļ¬cult to develop any real intuition for what the equi-aļ¬ne arc length represents, but we will at least make an attempt in the following exercise.
6.2. Moving frames for curves in A3
175
*Exercise 6.7. Let Ī± : I ā A3 be a nondegenerate curve, parametrized by its Euclidean arc length sĀÆ. Show that the equi-aļ¬ne arc length of Ī± is given by & sĀÆ 9 6 (6.3) s(ĀÆ s) = Īŗ(u)2 Ļ (u) du, sĀÆ0
where Īŗ and Ļ represent the Euclidean curvature and torsion, respectively, of Ī±. So, the greater a curveās Euclidean curvature and torsion, the greater its equi-aļ¬ne arc length. (Hint: Use the Frenet equations for Ī± to compute Ī± (ĀÆ s), Ī± (ĀÆ s), Ī± (ĀÆ s).) Remark 6.8. This exercise implies that, while the curvature Īŗ and torsion Ļ of a unit-speed curve in E3 are not preserved by the action of the equi
6 aļ¬ne group, the 1-form |Īŗ2 Ļ | dĀÆ s (where sĀÆ is the Euclidean arc length of the curve) is invariant under the action of this group! By Exercise 6.4, we can assume that Ī± is parametrized by equi-aļ¬ne arc length, so that det[Ī± (s) Ī± (s) Ī± (s)] = Ā±1. Then the frame ļ¬eld (6.4)
e1 (s) = Ā±Ī± (s),
e2 (s) = Ā±Ī± (s),
e3 (s) = Ā±Ī± (s)
(with signs chosen to agree with the sign of det[Ī± (s) Ī± (s) Ī± (s)]) is unimodular and equivariant under the action of A(3). Moreover, this frame ļ¬eld is canonical, in that it is uniquely determined by the geometry of Ī±. Deļ¬nition 6.9. Let Ī± : I ā A3 be a nondegenerate curve, parametrized by equi-aļ¬ne arc length s. The unimodular frame ļ¬eld (6.4) is called the equi-aļ¬ne Frenet frame of Ī±. Remark 6.10. The vectors (e1 (s), e2 (s), e3 (s)) of the equi-aļ¬ne Frenet frame (6.4) may not be exactly equal to those of the frame ļ¬eld (6.1); they will diļ¬er by precisely the terms that we ignored in the chain rule computations (6.2). In order to avoid a proliferation of sign ambiguities in what follows, for the rest of this section we will assume that det[Ī± (s) Ī± (s) Ī± (s)] = 1, so that the signs in equation (6.4) are all positive. Inserting the appropriate minus signs when det[Ī± (s) Ī± (s) Ī± (s)] = ā1 is left as an exercise for the reader. Now that we have a canonical adapted frame ļ¬eld along Ī±, we can compute invariants for nondegenerate equi-aļ¬ne curves in A3 . As in the Euclidean
176
6. Curves and surfaces in equi-aļ¬ne space
case, the pullbacks of equations (3.1) to I via Ī± can be written as Ī± (s)ds = ei (s) Ļ ĀÆ i,
(6.5)
ei (s)ds = ej (s) Ļ ĀÆ ij .
But we constructed our adapted frame ļ¬eld so that Ī± (s) = e1 (s),
e1 (s) = e2 (s),
e2 (s) = e3 (s).
So, the ļ¬rst equation in (6.5) implies that Ļ ĀÆ 1 = ds,
Ļ ĀÆ2 = Ļ ĀÆ 3 = 0,
and the equations for e1 (s), e2 (s) in (6.5) imply that Ļ ĀÆ 12 = Ļ ĀÆ 23 = ds,
Ļ ĀÆ 11 = Ļ ĀÆ 13 = Ļ ĀÆ 21 = Ļ ĀÆ 22 = 0.
Finally, e3 (s) must satisfy e3 (s) = Īŗ1 (s) e1 (s) + Īŗ2 (s) e2 (s) for some functions Īŗ1 (s), Īŗ2 (s). These functions are called the equi-aļ¬ne curvatures of Ī± at s. *Exercise 6.11. Why is there no e3 (s) term in the equation for e3 (s)? (Hint: What does the equation for e3 (s) in (6.5) tell you about Ļ ĀÆ 31 , Ļ ĀÆ 32 , and Ļ ĀÆ 33 ? What relations must the (ĀÆ Ļji ) satisfy?) Thus, the equi-aļ¬ne analog of the Frenet equations is ā” 0 ā¢
ā¢1 Ī± (s) e1 (s) e2 (s) e3 (s) = Ī±(s) e1 (s) e2 (s) e3 (s) ā¢ ā¢0 ā£ 0
0 0
0
ā¤
ā„ 0 0 Īŗ1 (s)ā„ ā„. 1 0 Īŗ2 (s)ā„ ā¦ 0 1 0
Applying Lemma 4.2 yields the following theorem: Theorem 6.12. Two nondegenerate equi-aļ¬ne curves Ī±1 , Ī±2 : I ā A3 parametrized by equi-aļ¬ne arc length diļ¬er by an equi-aļ¬ne transformation if and only if they have the same equi-aļ¬ne curvatures Īŗ1 (s), Īŗ2 (s). *Exercise 6.13. Let Ī± : I ā A2 be a curve in the equi-aļ¬ne plane A2 . (a) How would you deļ¬ne an adapted frame ļ¬eld (e1 (t), e2 (t)) at each point of the curve? (b) When should a curve Ī± : I ā A2 be called ānondegenerateā? (c) How would you deļ¬ne equi-aļ¬ne arc length for a nondegenerate curve?
6.2. Moving frames for curves in A3
177
(d) Use the pullbacks of equations (3.1) to ļ¬nd a complete set of invariants for nondegenerate equi-aļ¬ne curves Ī± : I ā A2 parametrized by equi-aļ¬ne arc length. *Exercise 6.14. Hopefully you discovered a single invariant Īŗ(s), called the equi-aļ¬ne curvature of Ī±, in Exercise 6.13. Suppose that Ī± : I ā A2 is nondegenerate and that its equi-aļ¬ne curvature Īŗ(s) is equal to a constant Īŗ. Show that: (a) If Īŗ = 0, then Ī± is a parabola. (b) If Īŗ > 0, then Ī± is a hyperbola. (c) If Īŗ < 0, then Ī± is a circle or an ellipse. (Hint: In each case, you should be able to show that Ī±(s) satisļ¬es a differential equation whose general solution is not diļ¬cult to ļ¬nd. Since each of these conditions is preserved under equi-aļ¬ne transformations, you can perform an equi-aļ¬ne transformation to eliminate most of the arbitrary constants in the general solution. Finally, eliminate the parameter s to show that Ī± lies on the appropriate conic section.) *Exercise 6.15. Let Ī± : I ā A2 be a curve, parametrized by its Euclidean arc length sĀÆ, and let Īŗ ĀÆ (ĀÆ s) denote the Euclidean curvature of Ī±. (a) Show that the equi-aļ¬ne arc length of Ī± is given by &
3 (6.6) s(ĀÆ s) = Īŗ ĀÆ (u)2 du. sĀÆ0
ĀÆ2 (ĀÆ (b) Let (ĀÆ e1 (ĀÆ s), e s)) denote the Euclidean Frenet frame ļ¬eld along Ī±. Show that the equi-aļ¬ne Frenet frame ļ¬eld along Ī± is given by ĀÆ1 (ĀÆ e1 (s) = Īŗ ĀÆ (ĀÆ s(s))ā1/3 e s(s)), (6.7)
e2 (s) = ā
s(s)) Īŗ ĀÆ (ĀÆ ĀÆ1 (ĀÆ ĀÆ2 (ĀÆ e s(s)) + Īŗ ĀÆ (ĀÆ s(s))1/3 e s(s)), 3ĀÆ Īŗ(ĀÆ s(s))5/3
where sĀÆ(s) is the inverse function of the equi-aļ¬ne arc length function (6.6) and primes denote derivatives of Īŗ ĀÆ (ĀÆ s) with respect to sĀÆ. (Hint: Use the chain rule very carefully!) (c) Diļ¬erentiate the expression (6.7) for e2 (s) (again, being very careful with the chain rule) and conclude that the equi-aļ¬ne curvature of Ī± is given by 1 (6.8) Īŗ(s) = ā 3ĀÆ Īŗ(ĀÆ s(s))ĀÆ Īŗ (ĀÆ s(s)) ā 5(ĀÆ Īŗ (ĀÆ s(s)))2 + 9ĀÆ Īŗ(ĀÆ s(s))4 . 8/3 9ĀÆ Īŗ(ĀÆ s(s)) This shows that, while Euclidean curvature is a second-order invariant of Ī± (i.e., it may be expressed in terms of Ī± and its derivatives up to order 2), the
178
6. Curves and surfaces in equi-aļ¬ne space
equi-aļ¬ne curvature is a fourth-order invariant of Ī±. (And, like the equiaļ¬ne arc length, the order of this invariant is dependent on the dimension of the ambient space.) This expression for the equi-aļ¬ne curvature in terms of the Euclidean curvature is originally due to Blaschke [Bla23]; for an alternative approach, see [Kog03] or [Olv11b]. Exercise 6.16. Now that you know how things work for curves in A2 and A3 , what sorts of invariants would you expect to appear for curves in An ? *Exercise 6.17. Let Ī± : I ā A3 be a nondegenerate curve parametrized by equi-aļ¬ne arc length s, and suppose that its equi-aļ¬ne curvatures Īŗ1 (s), Īŗ2 (s) are both identically equal to zero. (a) Show that there exist vectors v0 , v1 , v2 , v3 ā A3 , with det[v1 v2 v3 ] = 1, such that Ī±(s) = v0 + sv1 + 12 s2 v2 + 16 s3 v3 . (b) Show that there exists an equi-aļ¬ne transformation g ā A(3) such that
g Ā· Ī±(s) = t s, 12 s2 , 16 s3 . This curve is called the rational normal curve of degree 3; we will encounter it again in a slightly diļ¬erent form in Chapter 7 (cf. Exercise 7.31).
6.3. Moving frames for surfaces in A3 Now, let U be an open set in R2 , and let x : U ā A3 be an immersion whose image is a surface Ī£ = x(U ). Just as for curves, an adapted frame ļ¬eld Ė : U ā A(3) of the form along Ī£ is a lifting x Ė (u) = (x(u); e1 (u), e2 (u), e3 (u)) , x where for each u ā U , (e1 (u), e2 (u), e3 (u)) is a unimodular basis for the tangent space Tx(u) A3 . In the Euclidean case, we began by choosing e3 (u) to be orthogonal to the tangent plane Tx(u) Ī£; having done so, it followed that (e1 (u), e2 (u)) must be a basis for Tx(u) Ī£. But in equi-aļ¬ne space, there is no notion of orthogonality, so we donāt have any obvious way to normalize e3 (u) immediately. However, we can still make our ļ¬rst adaptation by requiring that (e1 (u), e2 (u)) span Tx(u) Ī£. A frame ļ¬eld satisfying this condition will be called 0-adapted. (The 0 reļ¬ects the fact that we will be making further reļ¬nements to the frame later on; these frame ļ¬elds will be called 1-adapted, 2-adapted, etc.) Exercise 6.18. Show that the condition that (e1 (u), e2 (u)) span Tx(u) Ī£ is equivariant under the action of A(3).
6.3. Moving frames for surfaces in A3
179
Ė ā Ļji ) on U . Precisely the Let (ĀÆ Ļi, Ļ ĀÆ ji ) represent the pulled-back forms (Ė xā Ļ i , x same reasoning as in the Euclidean case can be used to prove the following: Proposition 6.19. Let U ā R2 be an open set, and let x : U ā A3 be an immersion. For any 0-adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)) along Ī£ = x(U ), the associated dual and connection forms (ĀÆ Ļi, Ļ ĀÆ ji ) have the property that Ļ ĀÆ 3 = 0. Moreover, (ĀÆ Ļ1, Ļ ĀÆ 2 ) form a basis for the 1-forms on U . As in the Euclidean case, diļ¬erentiating the equation Ļ ĀÆ 3 = 0 yields dĀÆ Ļ 3 = āĀÆ Ļ13 ā§ Ļ ĀÆ1 ā Ļ ĀÆ 23 ā§ Ļ ĀÆ 2 = 0, and Cartanās lemma (cf. Lemma 2.49) implies that there exist functions h11 , h12 , h22 on U such that 3 1 Ļ ĀÆ1 h11 h12 Ļ ĀÆ (6.9) = . Ļ ĀÆ 23 h12 h22 Ļ ĀÆ2 Once again, with an eye towards making further adaptations to our frame ļ¬eld, we will investigate how the matrix [hij ] changes if we vary the frame. At this point things begin to look diļ¬erent from the Euclidean case because we have a larger symmetry group to make use of. Let (e1 (u), e2 (u), e3 (u)) be any 0-adapted frame ļ¬eld, with associated MauĖ2 (u), rer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ). Any other 0-adapted frame ļ¬eld (Ė e1 (u), e Ė3 (u)) must have the properties that e Ė2 (u)) = span(e1 (u), e2 (u)) span(Ė e1 (u), e and
Ė1 (u) e Ė2 (u) e Ė3 (u) = det e1 (u) e2 (u) e3 (u) . det e
This will be true if and only if ā” (6.10)
ā¢ B Ė1 (u) e Ė2 (u) e Ė3 (u) = e1 (u) e2 (u) e3 (u) ā£ e 0
r1 r2
ā¤ ā„ ā¦
0 (det B)ā1
for some GL(2)-valued function B and real-valued functions r1 , r2 on U . Let ĖĀÆ i , Ļ ĖĀÆ ji ) be the Maurer-Cartan forms associated to the new frame ļ¬eld. (Ļ *Exercise 6.20. (a) Show that 1 1 ĖĀÆ Ļ Ļ ĀÆ (6.11) = B ā1 . ĖĀÆ 2 Ļ Ļ ĀÆ2
180
6. Curves and surfaces in equi-aļ¬ne space
(b) Show that (6.12)
ĖĀÆ 13 Ļ
Ļ ĀÆĖ 23
t
= (det B) B
Ļ ĀÆ 13 Ļ ĀÆ 23
.
(Hint: It is no longer true that Ļ ĀÆ i3 = āĀÆ Ļ3i , so you must use the equations 3 for de1 and de2 in (3.1) to compute Ļ ĀÆ 1 and Ļ ĀÆ 23 . It will be simplest to keep track of everything if you write them in the form ā” 1 1ā¤ ĀÆ2 Ļ ĀÆ1 Ļ
ā¢ 2 2ā„ ĀÆ1 Ļ ĀÆ2 ā„ d e1 e2 = e1 e2 e3 ā¢ ā¦ .) ā£Ļ Ļ ĀÆ 13 Ļ ĀÆ 23 Ė 11 , h Ė 12 , h Ė 22 on U such (c) Cartanās lemma implies that there exist functions h that 3 Ė Ė 1 ĖĀÆ 1 ĖĀÆ Ļ h11 h12 Ļ = . Ė 12 h Ė 22 Ļ ĖĀÆ 23 ĖĀÆ 2 Ļ h Show that Ė Ė h11 h12 h11 h12 (6.13) = (det B) tB B. Ė Ė h12 h22 h12 h22 The transformation (6.13) has the property that Ė ij ] = (det B)4 det[hij ], det[h so the sign of det[hij ] is ļ¬xed. Deļ¬nition 6.21. Assume that the matrix [hij ] is nonsingular at every point of U . The surface Ī£ = x(U ) is called (1) elliptic if det[hij ] > 0 at every point of U ; (2) hyperbolic if det[hij ] < 0 at every point of U . For the remainder of this section, we will assume that Ī£ is elliptic; the hyperbolic case will be treated in Exercise 6.42. The transformation (6.13) acts transitively on the set of 2 Ć 2 matrices with positive determinant; therefore, there exists a choice of 0-adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)) for which h11 h12 1 0 = . 0 1 h12 h22 Such a frame ļ¬eld will be called 1-adapted.
6.3. Moving frames for surfaces in A3
181
*Exercise 6.22. Let (e1 (u), e2 (u), e3 (u)) be any 1-adapted frame ļ¬eld for an elliptic equi-aļ¬ne surface Ī£ = x(U ) ā A3 . Ė2 (u), e Ė3 (u)) for Ī£ (a) Show that any other 1-adapted frame ļ¬eld (Ė e1 (u), e must have the form ā¤ ā” r1
ā¢ B ā„ Ė1 (u) e Ė2 (u) e Ė3 (u) = e1 (u) e2 (u) e3 (u) ā£ e (6.14) ā¦ r2 ā1 0 0 (det B) for some SO(2)-valued function B and real-valued functions r1 , r2 on U . (b) Let (ĀÆ Ļi, Ļ ĀÆ ji ) be the Maurer-Cartan forms associated to a 1-adapted frame ĖĀÆ i , Ļ ĖĀÆ ji ) be the Maurer-Cartan forms asļ¬eld (e1 (u), e2 (u), e3 (u)), and let (Ļ sociated to the 1-adapted frame ļ¬eld (6.14). Show that ĖĀÆ 1 + Ļ ĖĀÆ 2 = (Ļ ĖĀÆ 23 Ļ ĖĀÆ 1 )2 + (Ļ ĖĀÆ 2 )2 = (ĀÆ Ļ ĀÆĖ 13 Ļ Ļ 1 )2 + (ĀÆ Ļ 2 )2 = Ļ ĀÆ 13 Ļ ĀÆ1 + Ļ ĀÆ 23 Ļ ĀÆ2 (cf. Exercise 6.20). Deļ¬nition 6.23. Let U ā R2 be an open set, and let x : U ā A3 be an immersion. Let (e1 (u), e2 (u), e3 (u)) be any 1-adapted frame ļ¬eld along Ī£ = x(U ), with associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ). The quadratic form I=Ļ ĀÆ 13 Ļ ĀÆ1 + Ļ ĀÆ 23 Ļ ĀÆ 2 = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 is called the equi-aļ¬ne ļ¬rst fundamental form of Ī£. The result of Exercise 6.22 implies that the equi-aļ¬ne ļ¬rst fundamental form is well-deļ¬ned, independent of the choice of a particular 1-adapted frame ļ¬eld on Ī£. Exercise 6.24. Show that the equi-aļ¬ne ļ¬rst fundamental form is invariant under the action of A(3). Because the equi-aļ¬ne ļ¬rst fundamental form is a positive deļ¬nite quadratic form at each point u ā U , it deļ¬nes a metric on Ī£. But unlike in the Euclidean case, this metric is not the restriction of any ambient metric on A3 to Ī£. Moreover, this metric depends on ļ¬rst and second derivatives of x, whereas in the Euclidean case, the ļ¬rst fundamental form depends only on ļ¬rst derivatives of x. Remark 6.25. Suppose that Ī£ is an elliptic surface in A3 and that Ī± : I ā A3 is a nondegenerate curve such that Ī±(I) ā Ī£. We now have two ways of deļ¬ning an arc length function on Ī±: We can use the equi-aļ¬ne arc length function from Ā§6.2 or we can use the restriction of the metric given by the equi-aļ¬ne ļ¬rst fundamental form on Ī£ to Ī±. For curves in either Euclidean or Minkowski space, these two notions of arc length would agree since both
182
6. Curves and surfaces in equi-aļ¬ne space
come from the restriction of the same ambient metric on the entire space. But there is no reason why they should be the same for curves in equi-aļ¬ne space, especially considering the fact that the equi-aļ¬ne arc length of a curve Ī± : I ā A3 depends on the ļ¬rst three derivatives of Ī±, whereas the equi-aļ¬ne ļ¬rst fundamental form of a surface x : U ā A3 depends only on the ļ¬rst two derivatives of x. In fact, as the following exercise will show, these two notions of arc length do not necessarily agree! Exercise 6.26. Let U = (0, 2Ļ) Ć (ā Ļ2 , Ļ2 ) ā R2 , and let x : U ā A3 be the standard parametrization of the unit sphere S2 : x(u, v) = t[cos(u) cos(v), sin(u) cos(v), sin(v)] . Let Ī± : (ā Ļ2 , Ļ2 ) ā A3 be the curve
Ī±(t) = x(t, t) = t cos2 (t), sin(t) cos(t), sin(t) . (a) Show that the equi-aļ¬ne arc length function for Ī± is & t
6 s(t) = 6 cos(u) du. 0
(b) Verify that, because S2 (regarded as a surface in E3 ) is totally umbilic with all principal curvatures equal to 1, the Maurer-Cartan forms associated to any adapted orthonormal frame ļ¬eld (e1 (u), e2 (u), e3 (u)) on S2 satisfy Ļ ĀÆ 13 = Ļ ĀÆ 1,
Ļ ĀÆ 23 = Ļ ĀÆ 2.
This means that any such frame ļ¬eld is also a 1-adapted frame ļ¬eld for S2 regarded as a surface in A3 . Therefore, the equi-aļ¬ne ļ¬rst fundamental form of S2 is the same as its Euclidean ļ¬rst fundamental form. (Moreover, the unit sphere is the only surface in A3 with this property!) (c) By part (b), the restriction of the equi-aļ¬ne ļ¬rst fundamental form on S2 to Ī± is the same as the restriction of the Euclidean ļ¬rst fundamental form on S2 to Ī±, which in turn is the same as the Euclidean arc length of Ī±. Show that this arc length function for Ī± is given by & t
s(t) = cos2 (u) + 1 du. 0
(d) (Maple recommended) Plot both arc length functions for Ī±, and compute the arc length of Ī± with respect to both arc length functions. What happens if you change the pitch of Ī±, i.e., set Ī±(t) = x(ct, t) for various choices of c? This raises the question: Given an elliptic surface Ī£ ā A3 , are there any curves in Ī£ for which these two notions of arc length do agree? This question is explored in [CEM+ 14].
6.3. Moving frames for surfaces in A3
183
Now, we still donāt have a well-deļ¬ned normal vector ļ¬eld e3 (u) on Ī£ because we still allow transformations between 1-adapted frame ļ¬elds with Ė3 (u) = e3 (u) + r1 (u)e1 (u) + r2 (u)e2 (u) e for arbitrary functions r1 , r2 . In order to address this issue, consider the connection form Ļ ĀÆ 33 . *Exercise 6.27. (a) Show that under a transformation of the form (6.14), we have ĖĀÆ 33 = Ļ Ļ ĀÆ 33 + r1 Ļ ĀÆ 1 + r2 Ļ ĀÆ 2. Since (ĀÆ Ļ1, Ļ ĀÆ 2 ) form a basis for the 1-forms on U , there is a unique choice of r1 , r2 for which Ļ ĀÆ 33 = 0. A 1-adapted frame ļ¬eld satisfying the condition Ļ ĀÆ 33 = 0 will be called 2-adapted. (b) Give a geometric interpretation of the condition Ļ ĀÆ 33 = 0. *Exercise 6.28. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame ļ¬eld for an elliptic equi-aļ¬ne surface Ī£ = x(U ) ā A3 . Ė2 (u), e Ė3 (u)) for Ī£ (a) Show that any other 2-adapted frame ļ¬eld (Ė e1 (u), e must have the form ā” ā¤ 0
ā¢ B ā„ Ė1 (u) e Ė2 (u) e Ė3 (u) = e1 (u) e2 (u) e3 (u) ā£ e (6.15) 0ā¦ 0 0 1 for some SO(2)-valued function B on U . Ė3 (u) = e3 (u) and that the vector ļ¬eld e3 (u) on Ī£ is (b) Conclude that e therefore now well-deļ¬ned. (c) Show that this choice of e3 (u) is equivariant under the action of A(3). Deļ¬nition 6.29. Let U ā R2 be an open set, and let x : U ā A3 be an immersion. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame ļ¬eld along Ī£ = x(U ). The vector ļ¬eld e3 (u) along Ī£ is called the equi-aļ¬ne normal vector ļ¬eld of Ī£. Remark 6.30. The equi-aļ¬ne normal direction at a point x0 ā Ī£ has the following geometric interpretation: Consider the family of planes in A3 that are parallel to the tangent plane Tx0 Ī£. Because Ī£ is elliptic, planes in this family suļ¬ciently close to Tx0 Ī£ each intersect Ī£ in a convex plane curve. Each of these plane curves bounds a region that has a center of mass. These centers of mass trace out a smooth curve in A3 , and the limiting tangent line to this curve as the curve approaches the original point x0 is parallel to the equi-aļ¬ne normal vector to Ī£ at x0 . And since the notions of parallelism
184
6. Curves and surfaces in equi-aļ¬ne space
and center of mass are equivariant under the action of A(3), this construction is equivariant as well. There is still more to be learned by diļ¬erentiating. Since we now have Ļ ĀÆ 33 = 0, we must have dĀÆ Ļ33 = 0 as well. According to the Cartan structure equations (3.8) and the normalizations that we have made thus far, this implies that dĀÆ Ļ33 = āĀÆ Ļ13 ā§ Ļ ĀÆ 31 ā Ļ ĀÆ 23 ā§ Ļ ĀÆ 32 = Ļ ĀÆ 31 ā§ Ļ ĀÆ1 + Ļ ĀÆ 32 ā§ Ļ ĀÆ 2 = 0. Cartanās lemma implies that there exist functions 11 , 12 , 22 on U such that 1 1 Ļ ĀÆ3 11 12 Ļ ĀÆ (6.16) = . Ļ ĀÆ 32 12 22 Ļ ĀÆ2 Deļ¬nition 6.31. Let U ā R2 be an open set, and let x : U ā A3 be an immersion. The equi-aļ¬ne second fundamental form of Ī£ = x(U ) is the quadratic form II on T U deļ¬ned by II = Ļ ĀÆ 31 Ļ ĀÆ 13 + Ļ ĀÆ 32 Ļ ĀÆ 23 =Ļ ĀÆ 31 Ļ ĀÆ1 + Ļ ĀÆ 32 Ļ ĀÆ2 = 11 (ĀÆ Ļ 1 )2 + 212 Ļ ĀÆ1 Ļ ĀÆ 2 + 22 (ĀÆ Ļ 2 )2 , where (ĀÆ Ļi, Ļ ĀÆ ji ) are the Maurer-Cartan forms associated to any 2-adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)) on Ī£ = x(U ). *Exercise 6.32. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame ļ¬eld Ė2 (u), along Ī£ with associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ), and let (Ė e1 (u), e Ė3 (u)) be any other 2-adapted frame ļ¬eld of the form (6.15), with associated e ĖĀÆ i , Ļ ĖĀÆ ji ). Maurer-Cartan forms (Ļ (a) Show that (6.17)
Ļ ĀÆĖ 31
Ļ ĀÆĖ 32
= B ā1
Ļ ĀÆ 31 Ļ ĀÆ 32
.
(b) Show that (6.18)
Ė11 Ė12 Ė12 Ė22
t
=B
11 12
B.
12 22
(c) Show that the equi-aļ¬ne second fundamental form of x is well-deļ¬ned, independent of the choice of 2-adapted frame ļ¬eld and associated MaurerCartan forms.
6.3. Moving frames for surfaces in A3
185
Remark 6.33. The quantity L = 12 (11 + 22 ) is called the equi-aļ¬ne mean curvature of Ī£. It has the property (similar to that of the mean curvature H in the Euclidean case) that it vanishes identically if and only if Ī£ is a critical point of the equi-aļ¬ne area functional.We will explore this further in Chapter 8. We still have a bit more diļ¬erentiating to do: *Exercise 6.34. (a) Diļ¬erentiate the equations Ļ ĀÆ 13 = Ļ ĀÆ 1,
Ļ ĀÆ 23 = Ļ ĀÆ2
and use Cartanās lemma to conclude that there exist functions h111 , h112 , h122 , h222 on U such that ā” ā¤ ā” ā¤ 2ĀÆ Ļ11 h111 h112 ĀÆ1 ā¢ 1 ā„ ā¢ ā„ Ļ 2 ā¢ ā„ ā¢ ā„ ĀÆ2 + Ļ ĀÆ 1 ā¦ = ā£h112 h122 ā¦ (6.19) . ā£Ļ Ļ ĀÆ2 2ĀÆ Ļ22 h122 h222 (b) Use the fact that Ļ ĀÆ 11 + Ļ ĀÆ 22 = 0 (Why is this true for a 2-adapted frame ļ¬eld? Hint: It is not true for a 1-adapted frame ļ¬eld.) to show that (6.20)
h122 = āh111 ,
h112 = āh222 .
Remark 6.35. If we were to deļ¬ne functions (hijk ) with i, j, k = 1, 2 by the equations 2ĀÆ Ļ11 = h111 Ļ ĀÆ 1 + h112 Ļ ĀÆ 2, Ļ ĀÆ 21 + Ļ ĀÆ 12 = h121 Ļ ĀÆ 1 + h122 Ļ ĀÆ 2 = h211 Ļ ĀÆ 1 + h212 Ļ ĀÆ 2, 2ĀÆ Ļ22 = h221 Ļ ĀÆ 1 + h222 Ļ ĀÆ2 (where the second line makes sense because the expression on the left is symmetric in the ļ¬rst two indices), then the result of Exercise 6.34 says that the (hijk ) are symmetric in all their indices. Deļ¬nition 6.36. Let U ā R2 be an open set, and let x : U ā A3 be an immersion. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame ļ¬eld along Ī£ = x(U ), with associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ). The cubic form P = hijk Ļ ĀÆ iĻ ĀÆjĻ ĀÆ k = h111 (ĀÆ Ļ 1 )3 + 3h112 (ĀÆ Ļ 1 )2 Ļ ĀÆ 2 + 3h122 Ļ ĀÆ 1 (ĀÆ Ļ 2 )2 + h222 (ĀÆ Ļ 2 )3 1 3 2 3 Ļ ) ā 3ĀÆ Ļ ) ā 3(ĀÆ = h111 (ĀÆ Ļ 1 (ĀÆ Ļ 2 )2 + h222 (ĀÆ Ļ 1 )2 Ļ ĀÆ2 is called the Fubini-Pick form of Ī£. P is also known as the cubic form of Ī£. *Exercise 6.37. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame ļ¬eld Ė2 (u), along Ī£ with associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ), and let (Ė e1 (u), e
186
6. Curves and surfaces in equi-aļ¬ne space
Ė3 (u)) be any other 2-adapted frame ļ¬eld of the form (6.15), with associated e ĖĀÆ i , Ļ ĖĀÆ ji ). We can write Maurer-Cartan forms (Ļ cos(Īø) ā sin(Īø) B= sin(Īø) cos(Īø) for some function Īø on U . (a) Show that
ĖĀÆ 11 Ļ ĖĀÆ 21 Ļ
(6.21)
ĖĀÆ 22 Ļ ĀÆĖ 12 Ļ
=B
ā1
Ļ ĀÆ 11 Ļ ĀÆ 21
B+
Ļ ĀÆ 12 Ļ ĀÆ 22
0 ādĪø dĪø
.
0
More explicitly: ĖĀÆ 11 = cos2 (Īø) Ļ Ļ ĀÆ 11 + sin(Īø) cos(Īø) (ĀÆ Ļ21 + Ļ ĀÆ 12 ) + sin2 (Īø) Ļ ĀÆ 22 , (6.22)
ĖĀÆ 21 = cos2 (Īø) Ļ Ļ ĀÆ 21 + sin(Īø) cos(Īø) (ĀÆ Ļ22 ā Ļ ĀÆ 11 ) ā sin2 (Īø) Ļ ĀÆ 12 ā dĪø, Ļ ĀÆĖ 12 = cos2 (Īø) Ļ ĀÆ 12 + sin(Īø) cos(Īø) (ĀÆ Ļ22 ā Ļ ĀÆ 11 ) ā sin2 (Īø) Ļ ĀÆ 21 + dĪø, ĖĀÆ 22 = cos2 (Īø) Ļ Ļ ĀÆ 22 ā sin(Īø) cos(Īø) (ĀÆ Ļ21 + Ļ ĀÆ 12 ) + (sin2 Īø)ĀÆ Ļ11 .
(b) Show that
Ė 111 h
Ė 222 h
=B
3
h111
.
h222
Conclude that the quantity h2111 + h2222 is well-deļ¬ned, independent of the choice of 2-adapted frame ļ¬eld and associated Maurer-Cartan forms. The quantity J = 16 (h2111 + h2222 ) is called the Pick invariant of Ī£. (c) Show that the Fubini-Pick form of x is well-deļ¬ned, independent of the choice of 2-adapted frame ļ¬eld and associated Maurer-Cartan forms. *Exercise 6.38. Suppose that the equi-aļ¬ne second fundamental form is not a multiple of the equi-aļ¬ne ļ¬rst fundamental form at any point of U . (This is equivalent to assuming that the matrix [ij ] is not a multiple of the identity at any point of U , similar to the assumption that a surface in E3 has no umbilic points.) (a) Show that there exists a 2-adapted frame ļ¬eld on Ī£ for which [ij ] is a diagonal matrix and that this choice of frame ļ¬eld is unique up to signs (and possibly exchanging e1 (u) and e2 (u)). We will call such a frame ļ¬eld an equi-aļ¬ne principal adapted frame ļ¬eld on Ī£. (Cf. Exercise 4.32.) (b) Show that prescribing the ļ¬rst and second equi-aļ¬ne fundamental forms and the Fubini-Pick form for an equi-aļ¬ne principal adapted frame ļ¬eld uniquely determines all the connection forms (ĀÆ Ļji ) up to sign. (Cf. Exercise 4.38.)
6.3. Moving frames for surfaces in A3
187
(c) Conclude that any two elliptic surfaces Ī£1 , Ī£2 ā A3 with the same ļ¬rst and second equi-aļ¬ne fundamental forms and Fubini-Pick form diļ¬er by an equi-aļ¬ne transformation. In other words, the ļ¬rst and second equi-aļ¬ne fundamental forms, together with the Fubini-Pick form, form a complete set of local invariants for elliptic equi-aļ¬ne surfaces with no umbilic points. Exercise 6.39 (Maple recommended). Let Ī£ be an elliptic equi-aļ¬ne surface as in Exercise 6.38, and suppose that x : U ā A3 is a parametrization of Ī£ with the property that the coordinate curves are all equi-aļ¬ne principal curves. (This means that xu , xv are multiples of the vectors e1 (u), e2 (u) in an equi-aļ¬ne principal adapted frame ļ¬eld for Ī£.) The equi-aļ¬ne ļ¬rst and second fundamental forms and the Fubini-Pick form of Ī£ can be written in terms of such a parametrization as I = M du2 + N dv 2 , (6.23)
II = m du2 + n dv 2 , P = p111 du3 + 3p112 du2 dv + 3p122 du dv 2 + p222 dv 3
for some functions M, N, m, n, pijk on U . (a) Show that the equi-aļ¬ne principal vectors tangent to Ī£ are 1 e1 (u) = ā xu , M
1 e2 (u) = ā xv . N
(b) Show that the Maurer-Cartan forms associated to this equi-aļ¬ne principal adapted frame ļ¬eld are given by ā ā Ļ ĀÆ1 = Ļ ĀÆ 13 = M du, Ļ ĀÆ2 = Ļ ĀÆ 23 = N dv, m n Ļ ĀÆ 31 = ā du, Ļ ĀÆ 32 = ā dv, M N p111 p122 p112 p222 1 2 Ļ ĀÆ1 = du + dv, Ļ ĀÆ2 = du + dv, 2M 2M # 2N " 2N # " p p ā112 + s1 du + ā122 + s2 dv, Ļ ĀÆ 21 = 2 MN 2 MN # " # " p122 p112 2 ā ā Ļ ĀÆ1 = ā s1 du + ā s2 dv 2 MN 2 MN for some functions s1 , s2 on U and that (6.24)
M p122 + N p111 = M p222 + N p112 = 0.
(Hint: Use the result of Exercise 6.34.)
188
6. Curves and surfaces in equi-aļ¬ne space
(c) Use the Cartan structure equations for dĀÆ Ļ 1 and dĀÆ Ļ 2 to ļ¬nd the functions s1 , s2 . (d) Suppose that the functions M, N, m, n, pijk are prescribed arbitrarily, subject to the conditions that M, N > 0 and that equation (6.24) is satisļ¬ed. Use the Cartan structure equations (3.8) to determine the PDE system that must be satisļ¬ed by these functions in order to guarantee the local existence of an elliptic equi-aļ¬ne surface with equi-aļ¬ne ļ¬rst and second fundamental forms and Fubini-Pick form given by (6.23). This PDE system is the analog of the Gauss-Codazzi system for elliptic surfaces in A3 . Exercise 6.40 (Maple recommended). Let x : U ā A3 be an elliptic equiaļ¬ne surface, and suppose that x has the property that, when regarded as a surface in Euclidean space, the coordinate curves are principal curves (cf. Exercise 4.41). Let Īŗ1 =
e , E
Īŗ2 =
g G
denote the principal curvatures of the Euclidean surface, where E, G, e, g are as in Exercise 4.41. (a) Let (e1 (u), e2 (u), e3 (u)) be the Euclidean adapted frame ļ¬eld of Exercise 4.41. Show that the frame ļ¬eld # Īŗ2 1/8 Ė1 (u) = Ė2 (u) = e e1 (u), e e2 (u), Īŗ31 (Īŗ1 Īŗ2 )u (Īŗ1 Īŗ2 )v 1/4 Ė3 (u) = (Īŗ1 Īŗ2 ) e3 (u) ā e ā 7/4 3/4 e1 (u) ā ā 3/4 7/4 e2 (u) 4 EĪŗ1 Īŗ2 4 GĪŗ1 Īŗ2 "
Īŗ1 Īŗ32
#1/8
"
is a 2-adapted equi-aļ¬ne frame ļ¬eld along Ī£. (b) Use the Maurer-Cartan equation (3.1) to compute the Maurer-Cartan ĖĀÆ i , Ļ Ė forms (Ļ ĀÆ ji ) associated to this frame ļ¬eld. (Hint: The ļ¬rst equation in (3.1) implies that ĖĀÆ 1 + e ĖĀÆ 2 , Ė1 Ļ Ė2 Ļ dx = e1 Ļ ĀÆ 1 + e2 Ļ ĀÆ2 = e ĖĀÆ 1 , Ļ ĖĀÆ 2 ) and (ĀÆ so the relationship between (Ļ Ļ1, Ļ ĀÆ 2 ) is closely related to the Ė2 ) and (e1 , e2 ).) relationship between (Ė e1 , e (c) Use the result of part (b) to compute the ļ¬rst and second equi-aļ¬ne fundamental forms and Fubini-Pick form of Ī£ in terms of the Euclidean
6.3. Moving frames for surfaces in A3
189
invariants of Ī£. In particular, note that Iaļ¬ = K ā1/4 IIEuc , where K is the Euclidean Gauss curvature of Ī£. Exercise 6.41. Let x : U ā A3 be an elliptic equi-aļ¬ne surface, and suppose that the equi-aļ¬ne second fundamental form is a multiple of the equi-aļ¬ne ļ¬rst fundamental form, so that the Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ) associated to a 2-adapted frame ļ¬eld satisfy the conditions Ļ ĀÆ 31 = Ī» Ļ ĀÆ 1,
Ļ ĀÆ 32 = Ī» Ļ ĀÆ2
for some function Ī» on U . (This is the analog of assuming that a surface in E3 is totally umbilic.) (a) Prove that Ī» is constant. (Hint: Use the structure equations to differentiate the equations above, taking into account the fact that we must have dĪ» = Ī»1 Ļ ĀÆ 1 + Ī»2 Ļ ĀÆ2 for some functions Ī»1 , Ī»2 on U . Then use Cartanās lemma.) (b) Show that if Ī» = 0, then de3 = 0, and therefore the equi-aļ¬ne normals of Ī£ are all parallel. Such surfaces are called improper equi-aļ¬ne spheres. (c) Show that if Ī» = 0, then d(x(u) ā Ī»1 e3 (u)) = 0. Therefore, all the equiaļ¬ne normals of Ī£ intersect at the point q = x(u) ā Ī»1 e3 (u). Such surfaces are called proper equi-aļ¬ne spheres. Equi-aļ¬ne spheres, both proper and improper, are much more plentiful than spheres in Euclidean space; in fact, there is an inļ¬nite-dimensional family of such surfaces. Exercise 6.42. In this exercise, we will explore the frame adaptation process for hyperbolic equi-aļ¬ne surfaces. Suppose that x : U ā A3 is a parametrization for a hyperbolic equi-aļ¬ne surface Ī£ = x(U ) ā A3 . Then the matrix [hij ] in (6.9) has det[hij ] < 0. (a) Show that there exists a 0-adapted frame ļ¬eld (f1 (u), f2 (u), f3 (u)) for which h11 h12 0 1 = . 1 0 h12 h22
190
6. Curves and surfaces in equi-aļ¬ne space
Such a frame ļ¬eld will be called a 1-adapted null frame ļ¬eld on Ī£ = x(U ). The associated Maurer-Cartan forms (ĀÆ Ī· i , Ī·ĀÆji ) satisfy Ī·ĀÆ13 = Ī·ĀÆ2 ,
Ī·ĀÆ23 = Ī·ĀÆ1 ,
and the equi-aļ¬ne ļ¬rst fundamental form becomes I = 2 Ī·ĀÆ1 Ī·ĀÆ2 . Thus, it deļ¬nes an indeļ¬nite metric on Ī£. (Based on this observation, we might expect that the geometry of hyperbolic equi-aļ¬ne surfaces will be somewhat reminiscent of the geometry of timelike surfaces in Minkowski space!) (b) Let (f1 (u), f2 (u), f3 (u)) be any 1-adapted null frame ļ¬eld for a hyperbolic equi-aļ¬ne surface Ī£ = x(U ) ā A3 . Show that any other 1-adapted null frame ļ¬eld (Ėf1 (u), Ėf2 (u), Ėf3 (u)) for Ī£ must have the form ā” Īø ā¤ e 0 r1
ā„ Ėf1 (u) Ėf2 (u) Ėf3 (u) = f1 (u) f2 (u) f3 (u) ā¢ (6.25) ā£ 0 eāĪø r2 ā¦ 0 0 1 for some functions Īø, r1 , r2 on U . (c) Show that under a transformation of the form (6.25), we have Ī·ĀÆĖ33 = Ī·ĀÆ33 + r2 Ī·ĀÆ1 + r1 Ī·ĀÆ2 . (No, thatās not a typo in the indices!) Conclude that there is a unique choice of r1 , r2 for which Ī·ĀÆ33 = 0. A 1-adapted null frame ļ¬eld satisfying the condition Ī·ĀÆ33 = 0 will be called a 2-adapted null frame ļ¬eld. For the remainder of this exercise, assume that (f1 (u), f2 (u), f3 (u)) is a 2adapted null frame ļ¬eld along Ī£. (d) Diļ¬erentiate the equation Ī·ĀÆ33 = 0 and use Cartanās lemma to conclude that there exist functions , 12 , 21 on U such that 1 Ī·ĀÆ3 12 Ī·ĀÆ1 = . Ī·ĀÆ32 21 Ī·ĀÆ2 The equi-aļ¬ne second fundamental form of Ī£ is given by II = Ī·ĀÆ31 Ī·ĀÆ13 + Ī·ĀÆ32 Ī·ĀÆ23 = Ī·ĀÆ31 Ī·ĀÆ2 + Ī·ĀÆ32 Ī·ĀÆ2 = 21 (ĀÆ Ī· 1 )2 + 2ĀÆ Ī· 1 Ī·ĀÆ2 + 12 (ĀÆ Ī· 2 )2 .
6.4. Maple computations
191
(e) Diļ¬erentiate the equations Ī·ĀÆ13 = Ī·ĀÆ2 ,
Ī·ĀÆ23 = Ī·ĀÆ1
and use Cartanās lemma to conclude that there exist functions h111 , h112 , h122 , h222 on U such that ā” ā¤ ā” ā¤ 2ĀÆ Ī·12 h111 h112 ā¢ 1 ā„ ā¢ ā„ Ī·ĀÆ1 ā¢Ī·ĀÆ + Ī·ĀÆ2 ā„ = ā¢h112 h122 ā„ 2ā¦ ā£ 1 ā£ ā¦ 2 . Ī·ĀÆ 2ĀÆ Ī·21 h122 h222 (f) Show that Ī·ĀÆ11 + Ī·ĀÆ22 = 0 (Hint: The reasoning is the same as for a 2-adapted frame ļ¬eld for an elliptic surface.), and conclude that h112 = h122 = 0. The Fubini-Pick form of Ī£ is (6.26)
P = h111 (ĀÆ Ī· 1 )3 + h222 (ĀÆ Ī· 2 )3 .
Further adaptations could be made to the frame in order to further normalize the equi-aļ¬ne second fundamental form; this would require considering several cases similar to those that arose when we considered the second fundamental form for timelike surfaces in Minkowski space.
6.4. Maple computations Once again, the Maple setup is similar to that of Chapter 4, but since there are fewer relations among the connection forms, we need to declare more of them. (In fact, weāll go ahead and declare them all; it turns out to be convenient to wait until after we deļ¬ne the structure equations to tell Maple about the single relation among the connection forms because it allows us to use a for loop to deļ¬ne the structure equations without creating redundant assignments.) So, after loading the Cartan and LinearAlgebra packages into Maple, begin by declaring the necessary 1-forms: > Form(omega[1], omega[2], omega[3]); Form(omega[1,1], omega[1,2], omega[1,3], omega[2,1], omega[2,2], omega[2,3], omega[3,1], omega[3,2], omega[3,3]);
192
6. Curves and surfaces in equi-aļ¬ne space
Tell Maple how to diļ¬erentiate these forms according to the Cartan structure equations (3.8): > for i from 1 to 3 do d(omega[i]):= -add(āomega[i,j] &Ė omega[j]ā, j=1..3); end do; for i from 1 to 3 do for j from 1 to 3 do d(omega[i,j]):= -add(āomega[i,k] &Ė omega[k,j]ā, k=1..3); end do; end do; Now tell Maple about the relation between the connection forms: > omega[3,3]:= -(omega[1,1] + omega[2,2]); Set up a substitution for the Maurer-Cartan forms associated to a 0-adapted frame ļ¬eld, taking into account the relations (6.9) that result from computing dĀÆ Ļ 3 = 0: > adaptedsub1:= [omega[3]=0, omega[3,1] = h[1,1]*omega[1] + h[1,2]*omega[2], omega[3,2] = h[1,2]*omega[1] + h[2,2]*omega[2]]; Exercise 6.20: Introduce new 1-forms to represent the transformed forms, with the same relation as the original forms: > Form(Omega[1], Omega[2], Omega[3]); Form(Omega[1,1], Omega[1,2], Omega[1,3], Omega[2,1], Omega[2,2], Omega[2,3], Omega[3,1], Omega[3,2], Omega[3,3]); Omega[3,3]:= -(Omega[1,1] + Omega[2,2]); Under a transformation of the form (6.10), we have the relations (6.11), (6.12). Since B is now an arbitrary matrix in GL(2), the corresponding substitution requires a bit more typing than in previous chapters: > framechangesub:= [ Omega[1] = (1/(b[1,1]*b[2,2] - b[1,2]*b[2,1]))* (b[2,2]*omega[1] - b[1,2]*omega[2]), Omega[2] = (1/(b[1,1]*b[2,2] - b[1,2]*b[2,1]))* (-b[2,1]*omega[1] + b[1,1]*omega[2]), Omega[3,1] = (b[1,1]*b[2,2] - b[1,2]*b[2,1])* (b[1,1]*omega[3,1] + b[2,1]*omega[3,2]), Omega[3,2] = (b[1,1]*b[2,2] - b[1,2]*b[2,1])*
6.4. Maple computations
193
(b[1,2]*omega[3,1] + b[2,2]*omega[3,2])]; > framechangebacksub:= makebacksub(framechangesub); Ė ij ) to the functions (hij ), introduce In order to compare the functions (h another substitution describing the adaptations for the transformed frame: > adaptedsub2:= [Omega[3]=0, Omega[3,1] = H[1,1]*Omega[1] + H[1,2]*Omega[2], Omega[3,2] = H[1,2]*Omega[1] + H[2,2]*Omega[2]]; Ė ij ) are expressed in Now combine all these substitutions to see how the (h terms of the (hij ): > zero2:= Simf(subs(adaptedsub2, Omega[3,1]) - Simf(subs(framechangebacksub, Simf(subs(adaptedsub1, Simf(subs(framechangesub, Omega[3,1]))))))); > eqns:= {op(ScalarForm(zero2))}; > zero3:= Simf(subs(adaptedsub2, Omega[3,2]) - Simf(subs(framechangebacksub, Simf(subs(adaptedsub1, Simf(subs(framechangesub, Omega[3,2]))))))); > eqns:= eqns union {op(ScalarForm(zero3))}; > solve(eqns, {H[1,1], H[1,2], H[2,2]}); > assign(%); Ė ij ) agree with those in equation (6.13): Check that these expressions for the (h > hmatrix:= Matrix([[h[1,1], h[1,2]], [h[1,2], h[2,2]]]); Hmatrix:= Matrix([[H[1,1], H[1,2]], [H[1,2], H[2,2]]]); B:= Matrix([[b[1,1], b[1,2]], [b[2,1], b[2,2]]]); > Hmatrix - simplify(Determinant(B)*Transpose(B).hmatrix.B); 0 Exercise 6.32: Now suppose that Ī£ is an elliptic surface and that we have chosen a 2-adapted frame ļ¬eld, so that [hij ] is the identity matrix, B ā SO(2), and the equi-aļ¬ne normal vector ļ¬eld e3 (u) is well-deļ¬ned. Since we now wish to explore transformations among 2-adapted frame ļ¬elds, Ė ij ): assign these conditions for both (hij ) and (h > h[1,1]:= 1; h[1,2]:= 0; h[2,2]:= 1; H[1,1]:= 1; H[1,2]:= 0; H[2,2]:= 1;
194
6. Curves and surfaces in equi-aļ¬ne space
b[1,1]:= b[1,2]:= b[2,1]:= b[2,2]:=
cos(theta); -sin(theta); sin(theta); cos(theta);
Note that our adapted frame substitution is now as it should be: > Simf(adaptedsub1); [Ļ3 = 0, Ļ3,1 = Ļ1 , Ļ3,2 = Ļ2 ] Similarly, our frame change substitution has simpliļ¬ed considerably: > Simf(framechangesub); [Ī©1 = cos(Īø) Ļ1 + sin(Īø) Ļ2 , Ī©2 = ā sin(Īø) Ļ1 + cos(Īø) Ļ2 , Ī©3,1 = cos(Īø) Ļ3,1 + sin(Īø) Ļ3,2 , Ī©3,2 = ā sin(Īø) Ļ3,1 + cos(Īø) Ļ3,2 ] At this point, we know that Ļ ĀÆ 33 = ā(ĀÆ Ļ11 + Ļ ĀÆ 22 ) = 0, and from equation (6.16) 1 2 we have expressions for Ļ ĀÆ 3 and Ļ ĀÆ 3 . Add these relations to our substitution (we will use ell for the letter in order to avoid confusing it with the number 1): > adaptedsub1:= [op(adaptedsub1), omega[2,2] = -omega[1,1], omega[1,3] = ell[1,1]*omega[1] + ell[1,2]*omega[2], omega[2,3] = ell[1,2]*omega[1] + ell[2,2]*omega[2]]; Add similar relations to the substitution for the transformed forms (we canāt use L because it is a command in the Cartan package, so use LL instead): > adaptedsub2:= [op(adaptedsub2), Omega[2,2] = -Omega[1,1], Omega[1,3] = LL[1,1]*Omega[1] + LL[1,2]*Omega[2], Omega[2,3] = LL[1,2]*Omega[1] + LL[2,2]*Omega[2]]; Next, add the relations (6.17) to our frame change substitution (keeping in mind that we now know that B ā SO(2)): > framechangesub:= Simf([op(framechangesub), Omega[1,3] = cos(theta)*omega[1,3] + sin(theta)*omega[2,3], Omega[2,3] = -sin(theta)*omega[1,3] + cos(theta)*omega[2,3]]); > framechangebacksub:= makebacksub(framechangesub); Now combine all these substitutions to see how the (Ėij ) are expressed in terms of the (ij ): > zero4:= Simf(subs(adaptedsub2, Omega[1,3]) - Simf(subs(framechangebacksub, Simf(subs(adaptedsub1,
6.4. Maple computations
> >
> > >
195
Simf(subs(framechangesub, Omega[1,3]))))))); eqns:= {op(ScalarForm(zero4))}; zero5:= Simf(subs(adaptedsub2, Omega[2,3]) - Simf(subs(framechangebacksub, Simf(subs(adaptedsub1, Simf(subs(framechangesub, Omega[2,3]))))))); eqns:= eqns union {op(ScalarForm(zero5))}; solve(eqns, {LL[1,1], LL[1,2], LL[2,2]}); assign(%);
Check that these expressions for the (Ėij ) agree with those in equation (6.18): > ellmatrix:= Matrix([[ell[1,1], ell[1,2]], [ell[1,2], ell[2,2]]]); LLmatrix:= Matrix([[LL[1,1], LL[1,2]], [LL[1,2], LL[2,2]]]); > LLmatrix - simplify(Transpose(B).ellmatrix.B); 0 Exercise 6.34: Continuing on, we now need to compute the quantities d(ĀÆ Ļ13 ā Ļ ĀÆ 1 ),
d(ĀÆ Ļ23 ā Ļ ĀÆ 2 ),
both of which should be equal to zero, and then substitute in what we already know: > zero6:= Simf(subs(adaptedsub1, Simf(d(omega[3,1] - omega[1])))); ā2 (Ļ1 ) &Ė (Ļ1,1 ) ā (Ļ2 ) &Ė (Ļ1,2 ) ā (Ļ2 ) &Ė (Ļ2,1 ) (See the Maple worksheet for this chapter on the AMS webpage for an illustration of why the inner Simf command is a good ideaāor experiment for yourself and see what happens without it!) We can recognize this as an expression of the form Ļ1 ā§ Ļ ĀÆ 1 + Ļ2 ā§ Ļ ĀÆ 2, to which we should apply Cartanās lemma. If you want Maple to help you identify Ļ1 and Ļ2 , you can use the pick command: > pick(zero6, omega[1]); 2 Ļ1,1 > pick(zero6, omega[2]); Ļ1,2 + Ļ2,1
196
6. Curves and surfaces in equi-aļ¬ne space
Similarly for d(ĀÆ Ļ23 ā Ļ ĀÆ 2 ): > zero7:= Simf(subs(adaptedsub1, Simf(d(omega[3,2] - omega[2])))); ā(Ļ1 ) &Ė (Ļ2,1 ) + 2 (Ļ2 ) &Ė (Ļ1,1 ) ā (Ļ1 ) &Ė (Ļ1,2 ) > pick(zero7, omega[1]); Ļ1,2 + Ļ2,1 > pick(zero7, omega[2]); ā2 Ļ1,1 Applying Cartanās lemma to both of these expressions yields equations (6.19), (6.20). We need to add these expressions to adaptedsub1 (say, by solving for Ļ ĀÆ 11 and Ļ ĀÆ 12 ), but we must do so carefully. Because these forms already occur in some of the right-hand sides of equations in adaptedsub1, we need to add them in two steps: First, substitute the new expressions into adaptedsub1 as it currently is, and then add them as new list items in adaptedsub1: > adaptedsub1:= Simf(subs([ omega[1,1] = (1/2)*(h[1,1,1]*omega[1] - h[2,2,2]*omega[2]), omega[2,1] = -omega[1,2] - h[2,2,2]*omega[1] - h[1,1,1]*omega[2]], adaptedsub1)); > adaptedsub1:= [op(adaptedsub1), omega[1,1] = (1/2)*(h[1,1,1]*omega[1] - h[2,2,2]*omega[2]), omega[2,1] = -omega[1,2] - h[2,2,2]*omega[1] - h[1,1,1]*omega[2]]; Exercise 6.37: First, we need to make the additions to adaptedsub2 corresponding to those that we just made to adaptedsub1: > adaptedsub2:= Simf(subs([ Omega[1,1] = (1/2)*(H[1,1,1]*Omega[1] - H[2,2,2]*Omega[2]), Omega[2,1] = -Omega[1,2] - H[2,2,2]*Omega[1] - H[1,1,1]*Omega[2]], adaptedsub2)); > adaptedsub2:= [op(adaptedsub2), Omega[1,1] = (1/2)*(H[1,1,1]*Omega[1] - H[2,2,2]*Omega[2]), Omega[2,1] = -Omega[1,2] - H[2,2,2]*Omega[1] - H[1,1,1]*Omega[2]]; We can get Maple to help us expand equation (6.21) into the expressions (6.22): > omegamatrix:= Matrix([[omega[1,1], omega[1,2]], [omega[2,1], omega[2,2]]]);
6.4. Maple computations
197
> Omegamatrix:= simplify(MatrixInverse(B).omegamatrix.B + Matrix([[0, -d(theta)], [d(theta), 0]])); Then we need to add these expressions to our frame change substitution: > framechangesub:= Simf([op(framechangesub), Omega[1,1] = Omegamatrix[1,1], Omega[1,2] = Omegamatrix[1,2], Omega[2,1] = Omegamatrix[2,1], Omega[2,2] = Omegamatrix[2,2]]); > framechangebacksub:= makebacksub(framechangesub); Ė ijk ) are expressed in Now combine all these substitutions to see how the (h terms of the (hijk ): > zero8:= Simf(subs(adaptedsub2, Omega[1,1]) - Simf(subs(framechangebacksub, Simf(subs(adaptedsub1, Simf(subs(framechangesub, Omega[1,1]))))))); > eqns:= {op(ScalarForm(zero8))}; > solve(eqns, {H[1,1,1], H[2,2,2]}); > assign(%); ĖĀÆ 21 and Ļ ĖĀÆ 12 and If you like, you can perform the analogous computations for Ļ conļ¬rm that they are now identically zero. But be warned that sometimes Maple can be clumsy about substitutions and require that they be applied ĖĀÆ 21 repeatedly. For instance, performing the same computation as above for Ļ now yields an apparently nonzero expression, but applying adaptedsub2 to this expression yet again does, in fact, yield zero. Ė ijk ) agree with those in equaFinally, check that these expressions for the (h tion (6.22): > hvector:= Vector([h[1,1,1], h[2,2,2]]); > Hvector:= Vector([H[1,1,1], H[2,2,2]]); > simplify(Hvector - B.B.B.hvector); 0 0 Exercise 6.40: First declare the functions associated to the Euclidean ļ¬rst and second fundamental forms of Ī£. Since we want to express everything in terms of Īŗ1 , Īŗ2 , declare these functions and then express e, g in terms of them: > PDETools[declare](E(u,v), G(u,v), kappa1(u,v), kappa2(u,v)); > e(u,v):= E(u,v)*kappa1(u,v); g(u,v):= G(u,v)*kappa2(u,v);
198
6. Curves and surfaces in equi-aļ¬ne space
For Maple purposes, use (ĀÆ Ī· i , Ī·ĀÆji ) to denote the Maurer-Cartan forms associated to the Euclidean adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)) and use (ĀÆ Ļi, Ļ ĀÆ ji ) to denote those associated to the 2-adapted equi-aļ¬ne frame ļ¬eld Ė2 (u), e Ė3 (u)). We have computed (ĀÆ (Ė e1 (u), e Ī· i , Ī·ĀÆji ) previously, and we can simply assign them: > eta[1]:= E(u,v)Ė(1/2)*d(u); eta[2]:= G(u,v)Ė(1/2)*d(v); eta[1,2]:= (diff(E(u,v), v)*d(u) - diff(G(u,v), u)*d(v))/ (2*E(u,v)Ė(1/2)*G(u,v)Ė(1/2)); eta[3,1]:= (e(u,v)/E(u,v)Ė(1/2))*d(u); eta[3,2]:= (g(u,v)/G(u,v)Ė(1/2))*d(v); eta[2,1]:= -eta[1,2]; eta[1,3]:= -eta[3,1]; eta[2,3]:= -eta[3,2]; Next, we need to introduce variables for the vector ļ¬elds (ei (u)) and (Ė ei (u)) so that we can use their structure equations to compute the associated Maurer-Cartan forms. Thereās no good way to tell Maple that these functions are vector-valued, but it wonāt really matter. First we tell Maple the structure equations for the exterior derivatives of the Euclidean frame ļ¬eld, the components of which we call (e01, e02, e03): > d(e01):= e02*eta[2,1] + e03*eta[3,1]; d(e02):= e01*eta[1,2] + e03*eta[3,2]; d(e03):= e01*eta[1,3] + e02*eta[2,3]; Weāll need the Gauss and Codazzi equations later, and these can be computed directly from the equations d(dei ) = 0. Start with d(de3 ): zero9:= Simf(d(d(e03))); Pick oļ¬ the scalar coeļ¬cient of du ā§ dv, and collect terms in (e1 (u), e2 (u)): > zero9a:= collect(pick(zero9, d(u), d(v)), {e01, e02}); Since (e1 (u), e2 (u)) are linearly independent, both coeļ¬cients must be zero; indeed, these are the Codazzi equations. We can collect these into a convenient substitution: > Codazzisub:= [ op(solve({coeff(zero9a, e01), coeff(zero9a, e02)}, {diff(kappa1(u,v), v), diff(kappa2(u,v), u)}))];
6.4. Maple computations
199
Now compute d(de1 ), taking the Codazzi equations into account: > zero10:= Simf(subs(Codazzisub, Simf(d(d(e01))))); The coeļ¬cient of du ā§ dv in the resulting expression is a multiple of e2 (u), and the scalar multiple is precisely the Gauss equation. We can add this equation to our substitution by solving for any term we like, but since we want to have things expressed in terms of Īŗ1 , Īŗ2 later, itās best to solve for something else. The other choices are the functions E, G, and their assorted derivatives. As a general rule, the safest thing to do is to solve for one of the highest-order derivatives in the expression; in this case, Guu will work. > GaussCodazzisub:= [op(Codazzisub), op(solve(coeff(pick(zero10, d(u), d(v)), e02), {diff(G(u,v), u,u)}))]; Check that there are no additional conditions lurking in the equations d(dei ) = 0: > Simf(subs(GaussCodazzisub, Simf(d(d(e01))))); Simf(subs(GaussCodazzisub, Simf(d(d(e02))))); Simf(subs(GaussCodazzisub, Simf(d(d(e03))))); 0 0 0 Rather than just computing the Maurer-Cartan forms for the frame ļ¬eld given in Exercise 6.40(a), letās explore where these expressions came from. First, the fact that the coordinate curves of Ī£ are principal curves means that the Maurer-Cartan forms associated to this Euclidean frame ļ¬eld have the property that Ī·ĀÆ13 is a multiple of Ī·ĀÆ1 and Ī·ĀÆ23 is a multiple of Ī·ĀÆ2 . This means that this Euclidean frame ļ¬eld is actually a 0-adapted equi-aļ¬ne frame ļ¬eld for which the matrix [hij ] is diagonal. Based on the transformation rule (6.13), this suggests that we should be able to create a 2-adapted equi-aļ¬ne Ė2 (u)) are scalar multiples of (e1 (u), e2 (u)), frame ļ¬eld for which (Ė e1 (u), e respectively. So, set Ė1 (u) = Ī»1 e1 (u), e
Ė2 (u) = Ī»2 e2 (u) e
for some (unknown) functions Ī»1 , Ī»2 . Then in order to keep the frame unimodular, we must have Ė3 (u) = r1 e1 (u) + r2 e2 (u) + e
1 e3 Ī»1 Ī»2
200
6. Curves and surfaces in equi-aļ¬ne space
for some functions r1 , r2 . For Maple purposes, we will call this frame ļ¬eld (e1, e2, e3): > e1:= lambda1(u,v)*e01; e2:= lambda2(u,v)*e02; e3:= r1(u,v)*e01 + r2(u,v)*e02 + e03/(lambda1(u,v)*lambda2(u,v)); In order for this frame ļ¬eld to be 2-adapted, the corresponding MaurerCartan forms must satisfy the conditions Ļ ĀÆ 13 = Ļ ĀÆ 1,
Ļ ĀÆ 23 = Ļ ĀÆ 2,
Ļ ĀÆ 11 + Ļ ĀÆ 22 = 0.
Moreover, we must have ā E 1 1 Ļ ĀÆ = Ī·ĀÆ = du, Ī»1 Ī»1 1
ā G 1 2 Ļ ĀÆ = Ī·ĀÆ = dv. Ī»2 Ī»2 2
So, set up a substitution with these conditions: > affineframesub:= [omega[3] = 0, omega[1] = eta[1]/lambda1(u,v), omega[2] = eta[2]/lambda2(u,v), omega[3,1] = eta[1]/lambda1(u,v), omega[3,2] = eta[2]/lambda2(u,v), omega[2,2] = -omega[1,1]]; We can now determine the functions Ī»1 , Ī»2 , r1 , r2 by using the structure equations for dĖ e1 , dĖ e2 , dĖ e3 : > zero11:= collect(Simf(subs(affineframesub, Simf(d(e1) - (e1*omega[1,1] + e2*omega[2,1] + e3*omega[3,1])))), {e01, e02, e03}); zero12:= collect(Simf(subs(affineframesub, Simf(d(e2) - (e1*omega[1,2] + e2*omega[2,2] + e3*omega[3,2])))), {e01, e02, e03}); zero13:= collect(Simf(subs(affineframesub, Simf(d(e3) - (e1*omega[1,3] + e2*omega[2,3] + e3*omega[3,3])))), {e01, e02, e03}); If you look at the output from these computations, youāll see that the e1 (u) and e2 (u) terms are rather a mess because they involve Maurer-Cartan forms that havenāt been computed yet. But the e3 (u) terms are manageable, and we can use them to solve for the unknown functions. It turns out to be best to do this in two steps, mainly because Maple is very clumsy at handling algebraic expressions. So, start by using the e3 (u) terms in dĖ e1 and dĖ e2 to
6.4. Maple computations
201
solve for Ī»1 , Ī»2 : > solve({pick(coeff(zero11, e03), d(u)), pick(coeff(zero12, e03), d(v))}, {lambda1(u,v), lambda2(u,v)}); At this point, Maple returns the rather obtuse expression ( 1 Ī»1 (u, v) = , RootOf (āĪŗ1 + Z 8 Īŗ23 )3 Īŗ2 Ī»2 (u, v) = RootOf (āĪŗ1 + Z 8 Īŗ23 )
3
and we will need to intervene by hand. The expression RootOf (āĪŗ1 + Z 8 Īŗ23 ) means any one of the values obtained by setting the expression in parentheses equal to zero and solving for Z. Since we know that we want everything to be positive-valued, we can easily see by inspection that the desired solution is (1/8) Īŗ Z = 1(3/8) , Īŗ2 which leads to the solution (1/8)
Ī»1 =
Īŗ2
, (3/8)
Īŗ1
(1/8)
Ī»2 =
Īŗ1
(3/8)
.
Īŗ2
So make these assignments: > lambda1(u,v):= kappa2(u,v)Ė(1/8)/kappa1(u,v)Ė(3/8); lambda2(u,v):= kappa1(u,v)Ė(1/8)/kappa2(u,v)Ė(3/8); Now we can use the e3 (u) term in dĖ e3 to solve for r1 , r2 : > solve({op(ScalarForm(Simf(coeff(zero13, e03))))}, {r1(u,v), r2(u,v)}); > assign(%); You can now inspect the following expressions to see that they agree with those in Exercise 6.40(a): > Simf(e1); Simf(e2); Simf(e3); Moreover, the remaining Maurer-Cartan forms can now be computed from the e1 (u) and e2 (u) terms in zero11, zero12, zero13. This is left as an exercise for the reader; details may be found in the Maple worksheet for this chapter on the AMS webpage.
10.1090/gsm/178/07
Chapter 7
Curves and surfaces in projective space
7.1. Introduction Applying the method of moving frames to curves and surfaces in projective space leads to invariants that are a bit more complicated than those that we have encountered thus far. For nondegenerate curves in the projective space Pn , there is no projectively invariant notion of arc lengthānot even an unconventional one such as the equi-aļ¬ne arc length that we deļ¬ned for curves in An . (It may, however, be possible to deļ¬ne an arc length function with some additional restrictions on the curve; cf. Remark 7.13.) Instead, a nondegenerate curve in Pn carries a canonical projective structure. Moreover, additional invariants associated to the curve (i.e., the projective analogs of curvature, torsion, etc.) are no longer real-valued functions on the curve; rather, they are more general geometric objects. In order to keep the complexity to a minimum while introducing these concepts, we will start one dimension lower than in previous chapters and begin by investigating curves in P2 . This is already a topic with important applications, notably in areas such as computer graphics. From there, we will move on to curves and surfaces in P3 .
203
204
7. Curves and surfaces in projective space
7.2. Moving frames for curves in P2 Consider a smooth, parametrized curve [Ī±] : I ā P2 that maps some open interval I ā R into the projective plane. P2 has the structure of the homogeneous space SL(3)/H[x0 ] , where H[x0 ] is the isotropy group from equation (3.17), so an adapted frame ļ¬eld along [Ī±] should be a lifting Ī± Ė : I ā SL(3). Any such lifting can be written as Ī± Ė (t) = (e0 (t), e1 (t), e2 (t)), where for each t ā I, [e0 (t)] = [Ī±(t)] ā P2 and
det e0 (t) e1 (t) e2 (t) = 1. Such an adapted frame ļ¬eld is called a projective frame ļ¬eld along [Ī±]. How should we choose such a lifting in some geometrically natural way? As in the equi-aļ¬ne case, it seems natural to choose the vectors (e0 (t), e1 (t), e2 (t)) so that e1 (t) = e0 (t),
e2 (t) = e1 (t).
This suggests that we look for a lifting Ī± : I ā R3 of [Ī±] with the property that (7.1)
det Ī±(t) Ī± (t) Ī± (t) = 1.
In particular, the vectors (Ī±(t), Ī± (t), Ī± (t)) should be linearly independent for all t ā I. Exercise 7.1. Given a curve [Ī±] : I ā P2 , let Ī±1 , Ī±2 : I ā R3 be two liftings of [Ī±], so that [Ī±1 (t)] = [Ī±2 (t)] = [Ī±(t)] ā P2 for all t ā I. Show that the vectors (Ī±1 (t), Ī±1 (t), Ī±1 (t)) are linearly independent if and only if the vectors (Ī±2 (t), Ī±2 (t), Ī±2 (t)) are linearly independent. (Hint: Ī±2 (t) = Ī»(t)Ī±1 (t) for some nonvanishing, real-valued function Ī»(t).) Exercise 7.1 implies that the linear dependence or independence of the vectors (Ī±(t), Ī± (t), Ī± (t)) does not depend on the choice of lifting Ī± : I ā R3 .
7.2. Moving frames for curves in P2
205
Therefore, the following deļ¬nition makes sense: Deļ¬nition 7.2. A regular curve [Ī±] : I ā P2 will be called nondegenerate if for any lifting Ī± : I ā R3 of [Ī±], the vectors (Ī±(t), Ī± (t), Ī± (t)) are linearly independent for all t ā I. *Exercise 7.3. Let [Ī±] : I ā P2 be a nondegenerate curve. Show that [Ī±] has a unique lifting Ī± : I ā R3 satisfying the determinant condition (7.1). (Hint: Let Ī±0 : I ā R3 be any lifting of [Ī±], and consider an arbitrary lifting Ī±(t) = Ī»(t)Ī±0 (t). Show that (7.1) uniquely determines the function Ī»(t).) Deļ¬nition 7.4. Let [Ī±] : I ā P2 be a nondegenerate curve. The lifting Ī± : I ā R3 of Exercise 7.3 will be called the canonical lifting of [Ī±]. The projective frame ļ¬eld e0 (t) = Ī±(t),
e1 (t) = Ī± (t),
e2 (t) = Ī± (t)
will be called the canonical projective frame ļ¬eld associated to [Ī±]. Exercise 7.5. Prove that the canonical projective frame ļ¬eld (e0 (t), e1 (t), e2 (t)) associated to [Ī±] is equivariant under the action of SL(3) on P2 : If we replace [Ī±] by g Ā· [Ī±] for some g ā SL(3), then eĪ³ (t) ā TĪ±(t) R3 will be replaced by g Ā· eĪ³ (t) ā TgĀ·Ī±(t) R3 for Ī³ = 0, 1, 2. Now things start to get interesting. The canonical projective frame ļ¬eld depends on the parametrization of [Ī±]. How will it change if we reparametrize [Ī±]? Is there some particular parametrization of [Ī±]āsomething akin to the arc length parametrization for curves in Euclidean spaceāthat is somehow geometrically natural? In order to answer this question, we will ļ¬rst compute invariants for parametrized nondegenerate curves. Then we will investigate how these invariants transform under a change of parametrization for [Ī±] and look for special parametrizations that normalize the invariants in some natural way. The pullbacks of equations (3.1) to I via Ī± can be written as (7.2)
eĪ³ (t)dt = eĪ“ (t) Ļ ĀÆ Ī³Ī“ ,
where the indices Ī³, Ī“ range from 0 to 2. But we constructed the canonical projective frame ļ¬eld so that e0 (t) = e1 (t),
e1 (t) = e2 (t).
206
7. Curves and surfaces in projective space
The ļ¬rst of these equations implies that Ļ ĀÆ 01 = dt,
Ļ ĀÆ 00 = Ļ ĀÆ 02 = 0,
and the second equation implies that Ļ ĀÆ 12 = dt,
Ļ ĀÆ 10 = Ļ ĀÆ 11 = 0.
Finally, e2 (t) must satisfy e2 (t) = Īŗ0 (t) e0 (t) + Īŗ1 (t) e1 (t) for some functions Īŗ0 (t), Īŗ1 (t). These functions are called the Wilczynski invariants of the parametrized curve [Ī±] : I ā P2 . They were ļ¬rst introduced by Wilczynski in the context of studying invariants for linear diļ¬erential operators; see [Wil62] for details. *Exercise 7.6. Why is there no e2 (t) term in the equation for e2 (t)? (Cf. Exercise 6.11.) Next, we investigate how the Wilczynski invariants transform under a reparametrization of Ī±. *Exercise 7.7 (Maple recommended). Let J ā R, and let [Ī²] : J ā P2 be a reparametrization of [Ī±] given by [Ī²(s)] = [Ī±(t(s))], with t (s) = 0. (a) Show that the canonical lifting of [Ī²] is given by Ī²(s) =
1 t (s)
Ī±(t(s)),
where Ī±(t) is the canonical lifting of [Ī±]. Ė1 (s), e Ė2 (s)) asso(b) Show that the canonical projective frame ļ¬eld (Ė e0 (s), e ciated to [Ī²] is given by Ė0 (s) = e
1
e0 (t(s)), t (s)
t (s) e0 (t(s)) + e1 (t(s)), (t (s))2 # " t (s) (t (s))2 t (s) Ė2 (s) = ā e e ā 2 (t(s)) ā e1 (t(s)) + t (s)e2 (t(s)), 0 (t (s))2 (t (s))3 t (s)
Ė1 (s) = ā e
where (e0 (t), e1 (t), e2 (t)) is the canonical projective frame ļ¬eld associated to [Ī±].
7.2. Moving frames for curves in P2
207
(c) Show that the Wilczynski invariants Īŗ Ė 0 (s), Īŗ Ė 1 (s) associated to [Ī²] are given by (7.3)
t(4) (s) t (s)t (s) Īŗ Ė 0 (s) = ā ā3 +4 t (s) (t (s))2
"
t (s) t (s)
#3
+ (t (s))3 Īŗ0 (t(s)) + t (s)t (s)Īŗ1 (t(s)), (7.4)
t (s) Īŗ Ė 1 (s) = ā2 +3 t (s)
"
t (s) t (s)
#2
+ (t (s))2 Īŗ1 (t(s)),
where Īŗ0 (t), Īŗ1 (t) are the Wilczynski invariants associated to [Ī±]. The transformation rule (7.4) for Īŗ Ė 1 is simpler than the transformation rule (7.3) for Īŗ Ė 0 , in that (7.4) involves only Īŗ1 , whereas (7.3) involves both Īŗ0 and Īŗ1 . So, ļ¬rst consider equation (7.4). Local existence theory for ordinary diļ¬erential equations guarantees that, given any smooth function Īŗ1 (t) on an interval I ā R, any point t0 ā I, and any point s0 ā R, there exists an interval J ā R containing the point s0 and a function t : J ā I with the properties that (1) t(s0 ) = t0 ; (2) t (s) = 0 for all s ā J (indeed, t (s0 ) = 0 can be prescribed arbitrarily, and then J can be chosen to be small enough so that t (s) = 0 on the entire interval); (3) t(s) satisļ¬es the diļ¬erential equation (7.5)
ā2
t (s) +3 t (s)
"
t (s) t (s)
#2
+ (t (s))2 Īŗ1 (t(s)) = 0.
According to (7.4), the corresponding reparametrization [Ī²(s)] will have Īŗ Ė 1 (s) ā” 0. Deļ¬nition 7.8. Let I ā R, and let [Ī±] : I ā P2 be a nondegenerate curve. [Ī±] is called a projective parametrization if the Wilczynski invariants Īŗ0 (t), Īŗ1 (t) associated to [Ī±] have the property that Īŗ1 (t) ā” 0. In this case, the parameter t ā I is called a projective parameter for [Ī±]. A projective parametrization for a nondegenerate curve in P2 is the natural analog of an arc length parametrization for a nondegenerate curve in En , M1,n , or An . This is certainly a less intuitive notion than that of an arc length parametrization, and as such it merits further investigation.
208
7. Curves and surfaces in projective space
For instance, does a nondegenerate curve in P2 have a unique projective parametrization? If not, how much ļ¬exibility does this notion admit? In order to address this question, suppose that [Ī±] : I ā P2 is a nondegenerate, projectively parametrized curve, and suppose that [Ī²(s)] = [Ī±(t(s))] is a reparametrization of [Ī±] that is also a projective parametrization. This means that the Wilczynski invariants Īŗ1 (t), Īŗ Ė 1 (s) associated to [Ī±], [Ī²], respectively, are both identically zero. According to (7.4), it follows that the function t(s) must satisfy the diļ¬erential equation (7.6)
t (s) ā2 +3 t (s)
"
t (s) t (s)
#2 = 0.
The expression on the left-hand side of (7.6) is equal to ā2 times the Schwarzian derivative S(t) of the function t(s), deļ¬ned by (7.7)
t 3 S(t) = ā t 2
"
t t
#2 .
The Schwarzian derivative is the fundamental invariant diļ¬erential operator of projective diļ¬erential geometry; it has the property that for any diļ¬erentiable function f : P1 ā P1 and any projective transformation g : P1 ā P1 , S(g ā¦ f ) = S(f ). Moreover, S(f ) = 0 if and only if f is a projective transformation. *Exercise 7.9. Recall that a projective transformation g : P1 ā P1 has the form g([x0 : x1 ]) = [dx0 + cx1 : bx0 + ax1 ], where
a b det = 1. c d
In terms of the aļ¬ne coordinate x ĀÆ= transformation
x1 x0
g(ĀÆ x) =
on P1 , g is given by a linear fractional
aĀÆ x+b . cĀÆ x+d
(a) Show by direct computation that if g is a linear fractional transformation, then S(g) = 0. (b) Suppose that a diļ¬erentiable function f : P1 ā P1 satisļ¬es S(f ) = 0. Show that f is a linear fractional transformation. (Hint: First show that
7.2. Moving frames for curves in P2
209
S(f ) can be rewritten as " S(f ) =
f f
#
1 ā 2
"
f f
#2 .
Then solve the diļ¬erential equation z ā 12 z 2 = 0 for the function z =
f f .)
As a consequence of Exercise 7.9, any projective reparametrization of a projectively parametrized nondegenerate curve [Ī±(t)] must have the form [Ī²(s)] = [Ī±(t(s))], where as + b t(s) = cs + d with ad ā bc = 1. Thus, a given nondegenerate curve in P2 does not have a unique projective parametrization, but rather a 3-parameter family of projective parametrizations related to each other by linear fractional transformations. Remark 7.10. Lest this seem too bizarre, note that a curve in Euclidean space does not actually have a unique arc length parameter s. Given an arc length parameter s along a curve Ī± : I ā En , any reparametrization of the form (7.8)
sĖ = s + c,
where c ā R is constant, yields an equally valid arc length parameter for Ī±. It is even possibleāfor instance, when Ī± is a closed curveāthat there is no single, continuous arc length parameter deļ¬ned on the entire curve. Rather, the curve may be covered with open intervals, each with its own arc length parameter, and transition functions of the form (7.8) deļ¬ned on the regions where intervals overlap. (Recall the discussion of transition functions between systems of local coordinates on manifolds in Ā§1.1.) What we really have is not so much an arc length parameter as a metric structure on the curve, deļ¬ned by the condition that the transition functions between systems of local coordinates are isometries. Similarly, the projective parameter along a nondegenerate curve in P2 given by the solution of (7.5) may not be well-deļ¬ned along the entire curve, as only local existence of a solution is guaranteed. However, a local projective parameter may be found in a neighborhood of any point on the curve, and the transition functions between projective parameters on overlapping inter-
210
7. Curves and surfaces in projective space
vals must be linear fractional transformations. Thus, what we really have is a well-deļ¬ned projective structure on the curve, analogous to the metric structure on a curve in En . Now, assume that [Ī±] : I ā P2 is projectively parametrized, and consider the remaining invariant function Īŗ0 (t). How does it transform under a projective reparametrization? *Exercise 7.11. Let [Ī±] : I ā P2 be a nondegenerate, projectively parametrized curve, and let [Ī²(s)] = [Ī±(t(s))] be a projective reparametrization of [Ī±] given by a linear fractional transformation (7.9)
t(s) =
as + b cs + d
with ad ā bc = 1. Let Īŗ0 (t), Īŗ Ė 0 (s) be the invariants associated to [Ī±], [Ī²], respectively. Use equation (7.3) to show that Īŗ0 (t(s)) Īŗ Ė 0 (s) = = Īŗ0 (t(s)) (cs + d)6
"
dt ds
#3 .
Conclude that under the transformation (7.9), we have Īŗ Ė 0 (s)(ds)3 = Īŗ0 (t)(dt)3 . Thus, the curvature āfunctionā Īŗ0 is not really a well-deļ¬ned function on the curve at all, but rather a ācubic formā that can be represented in terms of any projective parameter t as (7.10)
Īŗ Ė 0 = Īŗ0 (t)(dt)3 .
The cubic form (7.10) is called the projective curvature form of [Ī±]. In the following exercise, we show that the projective curvature form is a welldeļ¬ned rank 3 symmetric tensor along [Ī±]. *Exercise 7.12. Let [Ī±] : I ā P2 be a nondegenerate curve (not necessarily projectively parametrized), and let [Ī²(s)] = [Ī±(t(s))] be an arbitrary reparametrization of [Ī±]. Let Īŗ0 (t), Īŗ1 (t) be the invariants associated to [Ī±] and let Īŗ Ė 0 (s), Īŗ Ė 1 (s) be the invariants associated to [Ī²]. Use the transformation rules (7.3), (7.4) to show that Īŗ Ė 0 (s) ā 12 Īŗ Ė 1 (s) (ds)3 = Īŗ0 (t) ā 12 Īŗ1 (t) (dt)3 .
7.2. Moving frames for curves in P2
211
Conclude that the cubic form (7.11)
Īŗ Ė 0 = Īŗ0 (t) ā 12 Īŗ1 (t) (dt)3
is a well-deļ¬ned rank 3 tensor along [Ī±] that has the form (7.10) for any projective parametrization of [Ī±]. Remark 7.13. Our terminology here is somewhat nonstandard. In the case where Īŗ Ė 0 is nonvanishing, the 1-form
1 3 Īŗ Ė 0 = Īŗ0 (t) ā 12 Īŗ1 (t) 3 dt is sometimes referred to as the projective arc length element and the function & t
3 s(t) = Īŗ Ė0 t0
is sometimes referred to as the projective arc length of [Ī±]. Parametrizing [Ī±] according to this projective arc length function then leads to a notion of projective curvature that is somewhat diļ¬erent from ours. This approach has the advantage that the projective curvature so obtained is a well-deļ¬ned function on [Ī±] rather than a tensor; the disadvantage is that, while any nondegenerate curve can be projectively parametrized, not all nondegenerate curves in P2 can be parametrized by projective arc length, so this approach is more restrictive. (For instance, conic curves have projective arc length identically equal to zero.) Likewise, the canonical projective frame ļ¬eld associated to a projectively parametrized nondegenerate curve is not quite invariant under a projective reparametrization. *Exercise 7.14. Let [Ī±] : I ā P2 be a nondegenerate, projectively parametrized curve, and let [Ī²(s)] = [Ī±(t(s))] be a projective reparametrization of [Ī±] given by a linear fractional transformation as in equation (7.9). Let (e0 (t), e1 (t), e2 (t)) be the canonical projective frame ļ¬eld associated to [Ī±]. Use the result of Exercise 7.7(b) to show that the canonical projective frame Ė1 (s), e Ė2 (s)) associated to [Ī²] is given by ļ¬eld (Ė e0 (s), e Ė0 (s) = (cs + d)2 e0 (t(s)), e Ė1 (s) = 2c(cs + d)e0 (t(s)) + e1 (t(s)), e 2c 1 Ė2 (s) = 2c2 e0 (t(s)) + e e2 (t(s)). e1 (t(s)) + cs + d (cs + d)2 Thus, the projective structure gives rise to a 2-parameter family of canonical projective frame ļ¬elds along a nondegenerate curve [Ī±] : I ā P2 . Note that this is not the same thing as a rank 2 principal bundle of projective frames along the curve: In a rank 2 principal bundle, the choice of a frame at each
212
7. Curves and surfaces in projective space
point amounts to choosing two arbitrary functions along the curve. Here we have only two constant parametersā worth of choices for a projective frame ļ¬eld along the curve, and choosing a speciļ¬c projective frame at any point of the curve uniquely determines the canonical projective frame ļ¬eld along the entire curve. For any canonical projective frame ļ¬eld (e0 (t), e1 (t), e2 (t)) associated to a nondegenerate, projectively parametrized curve [Ī±] : I ā P2 , we have the following analog of the Frenet equations: ā¤ ā” 0 0 Īŗ0 (t)
ā¢ ā„ e0 (t) e1 (t) e2 (t) = e0 (t) e1 (t) e2 (t) ā£1 0 0 ā¦ . 01 0 Applying Lemma 4.2 yields the following theorem: Theorem 7.15. Two nondegenerate, projectively parametrized curves [Ī±1 ], [Ī±2 ] : I ā P2 diļ¬er by a projective transformation if and only if they have the same projective curvature form Īŗ Ė 0 = Īŗ0 (t)(dt)3 . *Exercise 7.16. Suppose that a nondegenerate, projectively parametrized curve [Ī±] : I ā P2 has projective curvature Īŗ0 (t) = 0. (Note that this condition is invariant under projective reparametrization.) (a) Show that the canonical lifting Ī± : I ā R3 of [Ī±] is part of a parabola contained in a plane in R3 . (b) Conclude that all conic sections in P2 have projective curvature identically equal to zero. Exercise 7.17. This exercise demonstrates the result of Exercise 7.16 explicitly in the case where [Ī±] is a circle. Let [Ī±] : R ā P2 be the circle parametrized in homogeneous coordinates as [Ī±(t)] = [1 : cos(t) : sin(t)]. (a) Show that the canonical lifting Ī± : R ā R3 of [Ī±] is Ī±(t) = t[1, cos(t), sin(t)]. (b) Show that the invariants Īŗ0 (t), Īŗ1 (t) associated to [Ī±] are Īŗ0 (t) = 0,
Īŗ1 (t) = ā1.
Conclude that [Ī±] is not a projective parametrization.
7.2. Moving frames for curves in P2
213
(c) Show that the function t(s) = 2 tanā1 (s) satisļ¬es equation (7.5); thus, the curve [Ī²(s)] = [Ī±(2 tanā1 (s))] is a projective reparametrization of [Ī±]. (d) Show that the canonical lifting of [Ī²] is ā” ā” ā¤ ā” ā¤ ā” ā¤ ā¤ 1 + s2 1 0 1 ā¢ ā„ 1ā¢ ā„ ā„ 1 2ā¢ ā„ 1ā¢ 2 Ī²(s) = 2 ā£1 ā s ā¦ = 2 s ā£ā1ā¦ + s ā£0ā¦ + 2 ā£1ā¦ . 0 0 1 2s Therefore, Ī² is a parabola in the plane spanned by the vectors t[1, ā1, 0], 0, 1] and passing through the point t[ 12 , 12 , 0].
t[0,
Exercise 7.18. We like to think of P2 as the plane R2 , represented as the open set 3 2 V0 = 1 : x ĀÆ1 : x ĀÆ2 ā P2 , plus points āat inļ¬nityā. Thus, we like to think of a curve [Ī±] : I ā P2 as having a parametrization of the form (7.12)
[Ī±(t)] = [1 : x ĀÆ1 (t) : x ĀÆ2 (t)]
that is valid at all but ļ¬nitely many points on the curve. (a) Show that any nondegenerate curve [Ī±] : I ā P2 of the form (7.12) has a reparametrization [Ī²(ĀÆ s)] = [Ī±(t(ĀÆ s))] whose canonical lifting is Ī²(ĀÆ s) = t[1, x ĀÆ1 (t(ĀÆ s)), x ĀÆ2 (t(ĀÆ s))]. (b) Show that the invariants Īŗ0 (ĀÆ s), Īŗ1 (ĀÆ s) associated to [Ī²] have the property that Īŗ0 (ĀÆ s) ā” 0. (c) Conclude that [Ī²(ĀÆ s)] is not a projective parametrization unless [Ī±] is a conic section. It follows that for non-conic curves, the canonical lifting of a projective parametrization is never contained in a plane in R3 . (d) Show that, while [Ī²(ĀÆ s)] is not a projective parametrization, it has the ĀÆ property that the curve Ī² : I ā A2 deļ¬ned by ĀÆ s) = t[ĀÆ Ī²(ĀÆ x1 (t(ĀÆ s)), x ĀÆ2 (t(ĀÆ s))] is parametrized by its equi-aļ¬ne arc length, and the Wilczynski invariant ĀÆ This is related to Īŗ1 (ĀÆ s) of [Ī²] is equal to the equi-aļ¬ne curvature Īŗ(ĀÆ s) of Ī².
214
7. Curves and surfaces in projective space
the fact that the equi-aļ¬ne transformation group A(2) can be realized as the subgroup of the projective transformation group SL(3) that preserves the plane {t[x0 , x1 , x2 ] ā R3 | x0 = 1} in R3 . (e) Let [Ī³(s)] = [Ī²(ĀÆ s(s))] be a projective reparametrization of [Ī²(ĀÆ s)], so that the Wilczynski invariants Īŗ Ė 0 (s), Īŗ Ė 1 (s) associated to [Ī³] satisfy Īŗ Ė 1 (s) = 0. Use the formulas (7.3), (7.4) and the fact that Īŗ0 (ĀÆ s) = 0 to show that Īŗ Ė 0 (s) = ā 12 (ĀÆ s (s))3 Īŗ (ĀÆ s(s)), where Īŗ(ĀÆ s) is the equi-aļ¬ne curvature of Ī²ĀÆ and Īŗ (ĀÆ s(s)) denotes the derivative of Īŗ(ĀÆ s) with respect to sĀÆ, evaluated at sĀÆ = sĀÆ(s). (f) Conclude that the projective curvature form of the original curve [Ī±] may be expressed as Īŗ Ė0 = Īŗ Ė 0 (s)(ds)3 = ā 12 Īŗ (ĀÆ s)(dĀÆ s)3 , where sĀÆ is the equi-aļ¬ne arc length of the curve Ī²ĀÆ : I ā A2 and Īŗ(ĀÆ s) is the ĀÆ equi-aļ¬ne curvature of Ī². Comparing with the result of Exercise 6.15, we see that the projective curvature form is a ļ¬fth-order invariant of [Ī±].
7.3. Moving frames for curves in P3 Now we will extend the ideas developed in Section 7.2 to curves in P3 . Consider a smooth parametrized curve [Ī±] : I ā P3 that maps some open interval I ā R into the projective space P3 . The construction of the canonical projective frame ļ¬eld associated to [Ī±] is analogous to that for curves in P2 : A projective frame ļ¬eld along [Ī±] is a lifting Ī± Ė : I ā SL(4), written as Ī± Ė (t) = (e0 (t), e1 (t), e2 (t), e3 (t)), with the properties that [e0 (t)] = [Ī±(t)] ā P3 and (7.13)
det e0 (t) e1 (t) e2 (t) e3 (t) = 1.
As for curves in P2 , this suggests that we should look for a lifting Ī± : I ā R4 with the property that
(7.14) det Ī±(t) Ī± (t) Ī± (t) Ī± (t) = 1. In particular, the vectors (Ī±(t), Ī± (t), Ī± (t), Ī± (t)) should be linearly independent for all t ā I.
7.3. Moving frames for curves in P3
215
Deļ¬nition 7.19. A regular curve [Ī±] : I ā P3 will be called nondegenerate if for any lifting Ī± : I ā R4 of [Ī±], the vectors (Ī±(t), Ī± (t), Ī± (t), Ī± (t)) are linearly independent for all t ā I. Exercise 7.20 (Cf. Exercise 7.1). Show that Deļ¬nition 7.19 makes sense, as follows. Given a curve [Ī±] : I ā P3 , let Ī±1 , Ī±2 : I ā R4 be two liftings of [Ī±], so that [Ī±1 (t)] = [Ī±2 (t)] = [Ī±(t)] ā P3 for all t ā I. Show that the vectors (Ī±1 (t), Ī±1 (t), Ī±1 (t), Ī±1 (t)) are linearly independent if and only if the vectors (Ī±2 (t), Ī±2 (t), Ī±2 (t), Ī±2 (t)) are linearly independent. *Exercise 7.21 (Cf. Exercise 7.3). (a) Let [Ī±] : I ā P3 be a nondegenerate curve. Show that [Ī±] has a lifting Ī± : I ā R4 satisfying the determinant condition
det Ī±(t) Ī± (t) Ī± (t) Ī± (t) = Ā±1 and that this lifting is uniquely determined up to sign (i.e., Ī± may be replaced by āĪ±). (b) Explain the ambiguities of sign in this result, which did not appear in Exercise 7.3. (Hint: Cf. Exercise 3.66.) In order to simplify the exposition in the remainder of this section, we will assume that
det Ī±(t) Ī± (t) Ī± (t) Ī± (t) = 1. (This can be achieved by replacing [Ī±] by its reļ¬ection through a plane in P3 if necessary.) Deļ¬nition 7.22. Let [Ī±] : I ā P3 be a nondegenerate curve. The lifting Ī± : I ā R4 of Exercise 7.21 will be called the canonical lifting of [Ī±]. The projective frame ļ¬eld e0 (t) = Ī±(t),
e1 (t) = Ī± (t),
e2 (t) = Ī± (t),
e3 (t) = Ī± (t)
will be called the canonical projective frame ļ¬eld associated to [Ī±]. Exercise 7.23. Prove that the canonical projective frame ļ¬eld (e0 (t), e1 (t), e2 (t), e3 (t)) associated to [Ī±] is equivariant under the action of SL(4) on P3 : If we replace [Ī±] by g Ā· [Ī±] for some g ā SL(4), then eĪ³ (t) ā TĪ±(t) R4 will be replaced by g Ā· eĪ³ (t) ā TgĀ·Ī±(t) R4 for Ī³ = 0, 1, 2, 3. As in Section 7.2, we will proceed by computing invariants for parametrized nondegenerate curves and then investigating how these invariants transform under reparametrizations.
216
7. Curves and surfaces in projective space
*Exercise 7.24. Show that the canonical projective frame ļ¬eld associated to a parametrized nondegenerate curve [Ī±] : I ā P3 satisļ¬es the structure equations e0 (t) = e1 (t),
e1 (t) = e2 (t),
e2 (t) = e3 (t),
e3 (t) = Īŗ0 (t) e0 (t) + Īŗ1 (t) e1 (t) + Īŗ2 (t) e2 (t) for some functions Īŗ0 (t), Īŗ1 (t), Īŗ2 (t). As for curves in P2 , the functions Īŗ0 (t), Īŗ1 (t), Īŗ2 (t) are called the Wilczynski invariants of the parametrized curve [Ī±] : I ā P3 . Next, we investigate how the Wilczynski invariants transform under a reparametrization of [Ī±]. *Exercise 7.25 (Maple recommended; cf. Exercise 7.7). Let J ā R, and let [Ī²] : J ā P3 be a reparametrization of [Ī±] given by [Ī²(s)] = [Ī±(t(s))], with t (s) > 0. (The sign of t (s) is not particularly important, but choosing it to be positive will simplify the following computations.) (a) Show that the canonical lifting of [Ī²] is given by Ī²(s) =
1 Ī±(t(s)), t (s)
where Ī±(t) is the canonical lifting of [Ī±]. Ė1 (s), e Ė2 (s), e Ė3 (s)) (b) Show that the canonical projective frame ļ¬eld (Ė e0 (s), e associated to [Ī²] is given by Ė0 (s) = e
1
e0 (t(s)), (t (s))3/2
3t (s) 1 e (t(s)) + e1 (t(s)), 0 (2t (s))5/2 (t (s))1/2 # " 3t (s) 15(t (s))2 t (s) Ė2 (s) = ā e e ā (t(s)) ā 2 e1 (t(s)) 0 2(t (s))5/2 4(t (s))7/2 (t (s))3/2
Ė1 (s) = ā e
+ (t (s))1/2 e2 (t(s)), Ė3 (s) = ā e " ā
3t(4) (s) 45t (s)t (s) 105(t (s))3 ā + 2(t (s))5/2 4(t (s))7/2 8(t (s))9/2 7t (s) 27(t (s))2 ā 2(t (s))3/2 4(t (s))5/2
+ (t (s))3/2 e3 (t(s)),
# e1 (t(s)) ā
e0 (t(s))
3t (s) e2 (t(s)) 2(t (s))1/2
7.3. Moving frames for curves in P3
217
where (e0 (t), e1 (t), e2 (t), e3 (t)) is the canonical projective frame ļ¬eld associated to [Ī±]. (c) Show that the Wilczynski invariants Īŗ Ė 0 (s), Īŗ Ė 1 (s), Īŗ Ė 2 (s) associated to [Ī²] are given by " #2 1 t (s) t(5) (s) t(4) (s)t (s) Īŗ Ė 0 (s) = + 60 ā24 + 120 2 16 t (s) (t (s)) t (s) t (s)(t (s))2 ā300 + 135 (t (s))3
(7.15)
(7.16)
"
t (s) t (s)
#4
3 + (t (s))4 Īŗ0 (t(s)) + t (s)(t (s))2 Īŗ1 (t(s)) 2 " # 3 3 2 + t (s)t (s) ā (t (s)) Īŗ2 (t(s)), 2 4 " #3 t(4) (s) t (s) t (s)t (s) Īŗ Ė 1 (s) = 5 ā ā3 +4 2 t (s) (t (s)) t (s) + (t (s))3 Īŗ1 (t(s)) + 2t (s)t (s)Īŗ2 (t(s)),
(7.17)
5 Īŗ Ė 2 (s) = 2
t (s) ā2 +3 t (s)
"
t (s) t (s)
#2
+ (t (s))2 Īŗ2 (t(s)),
where Īŗ0 (t), Īŗ1 (t), Īŗ2 (t) are the Wilczynski invariants associated to [Ī±]. Observe that equation (7.17) is very similar to equation (7.4). Precisely the same argument that we used in Section 7.2 can be applied to equation (7.17) to show that there exists a reparametrization [Ī²(s)] for which Īŗ Ė 2 (s) ā” 0. Thus, we have the analog of Deļ¬nition 7.8: Deļ¬nition 7.26. Let I ā R, and let [Ī±] : I ā P3 be a nondegenerate curve. [Ī±] is called a projective parametrization if the Wilczynski invariants Īŗ0 (t), Īŗ1 (t), Īŗ2 (t) associated to [Ī±] have the property that Īŗ2 (t) ā” 0. In this case, the parameter t ā I is called a projective parameter for [Ī±]. As for curves in P2 , a projective parametrization is unique up to reparametrizations of the form [Ī²(s)] = [Ī±(t(s))], where t(s) is a linear fractional transformation. Therefore, any nondegenerate curve in P3 has a well-deļ¬ned projective structure. Now, assume that [Ī±] : I ā P3 is projectively parametrized, and consider the remaining invariant functions Īŗ0 (t), Īŗ1 (t).
218
7. Curves and surfaces in projective space
*Exercise 7.27 (Cf. Exercise 7.11). Let [Ī±] : I ā P3 be a nondegenerate, projectively parametrized curve, and let [Ī²(s)] = [Ī±(t(s))] be a projective reparametrization of [Ī±] given by a linear fractional transformation as in equation (7.9). Let Īŗ0 (t), Īŗ1 (t) and Īŗ Ė 0 (s), Īŗ Ė 1 (s) be the invariants associated to [Ī±], [Ī²], respectively. (a) Use equations (7.15) and (7.16) to show that Īŗ Ė 1 (s) = Īŗ Ė 0 (s) =
Īŗ1 (t(s)) , (cs + d)6
Īŗ0 (t(s)) 3c Īŗ1 (t(s)) ā . 8 (cs + d) (cs + d)7
(b) Show that under the transformation (7.9), we have Īŗ Ė 1 (s)(ds)3 = Īŗ1 (t)(dt)3 , Īŗ Ė 0 (s) ā 12 Īŗ Ė 1 (s) (ds)4 = Īŗ0 (t) ā 12 Īŗ1 (t) (dt)4 . Therefore, the cubic form Īŗ Ė 1 = Īŗ1 (t)(dt)3
(7.18) and the quartic form (7.19)
Īŗ Ė 0 = Īŗ0 (t) ā 12 Īŗ1 (t) (dt)4
are invariant under projective transformations. We will call these forms the projective curvature forms of [Ī±]. *Exercise 7.28. Let [Ī±] : I ā P3 be a nondegenerate curve (not necessarily projectively parametrized), and let [Ī²(s)] = [Ī±(t(s))] be an arbitrary reparametrization of [Ī±]. Let Īŗ0 (t), Īŗ1 (t), Īŗ2 (t) be the invariants associated to [Ī±], and let Īŗ Ė 0 (s), Īŗ Ė 1 (s), Īŗ Ė 2 (s) be the invariants associated to [Ī²]. (a) Use the transformation rules (7.15), (7.16), (7.17) to show that Īŗ Ė 1 (s) ā Īŗ Ė 2 (s) (ds)3 = Īŗ1 (t) ā Īŗ2 (t) (dt)3 . Conclude that the cubic form (7.20)
Īŗ Ė 1 = Īŗ1 (t) ā Īŗ2 (t) (dt)3
is a well-deļ¬ned rank 3 tensor along [Ī±] that has the form (7.18) for any projective parametrization of [Ī±]. (b) Use the transformation rules (7.15), (7.16), (7.17) to show that 9 Īŗ Ė 0 (s) ā 12 Īŗ Ė 1 (s) + 15 Īŗ Ė 2 (s) + 100 (Ė Īŗ2 (s))2 (ds)4 = Īŗ0 (t) ā 12 Īŗ1 (t) + 15 Īŗ2 (t) +
2 9 100 (Īŗ2 (t))
(dt)4 .
7.3. Moving frames for curves in P3
Conclude that the quartic form Īŗ Ė 0 = Īŗ0 (t) ā 12 Īŗ1 (t) + 15 Īŗ2 (t) +
219
2 9 100 (Īŗ2 (t))
(dt)4
is a well-deļ¬ned rank 4 tensor along [Ī±] that has the form (7.19) for any projective parametrization of [Ī±]. As for curves in P2 , the canonical projective frame ļ¬eld associated to a projectively parametrized nondegenerate curve in P3 is not quite invariant under a projective reparametrization: *Exercise 7.29 (Cf. Exercise 7.14). Let [Ī±] : I ā P3 be a nondegenerate, projectively parametrized curve, and let [Ī²(s)] = [Ī±(t(s))] be a projective reparametrization of [Ī±] given by a linear fractional transformation as in equation (7.9). Let (e0 (t), e1 (t), e2 (t), e3 (t)) be the canonical projective frame ļ¬eld associated to [Ī±]. Use the result of Exercise 7.25(b) to show that Ė1 (s), e Ė2 (s), e Ė3 (s)) associated to the canonical projective frame ļ¬eld (Ė e0 (s), e [Ī²] is given by Ė0 (s) = (cs + d)3 e0 (t(s)), e Ė1 (s) = 3c(cs + d)2 e0 (t(s)) + (cs + d)e1 (t(s)), e 1 e2 (t(s)), (cs + d) 6c2 3c Ė3 (s) = 6c3 e0 (t(s)) + e e2 (t(s)) e1 (t(s)) + (cs + d) (cs + d)2 1 + e3 (t(s)). (cs + d)3
Ė2 (s) = 6c2 (cs + d)e0 (t(s)) + 4ce1 (t(s)) + e
Thus, the projective structure gives rise to a 2-parameter family of canonical projective frame ļ¬elds along a nondegenerate curve [Ī±] : I ā P3 , just as it did for curves in P2 . For any canonical projective frame ļ¬eld (e0 (t), e1 (t), e2 (t), e3 (t)) associated to such a curve, we have the following analog of the Frenet equations: ā¤ ā” 0 0 0 Īŗ0 (t) ā„ ā¢
ā¢1 0 0 Īŗ1 (t)ā„ ā„. ā¢ e0 (t) e1 (t) e2 (t) e3 (t) = e0 (t) e1 (t) e2 (t) e3 (t) ā¢ ā„ 0 1 0 0 ā¦ ā£ 001 0 Applying Lemma 4.2 yields the following theorem: Theorem 7.30. Two nondegenerate, projectively parametrized curves [Ī±1 ], [Ī±2 ] : I ā P3 diļ¬er by a projective transformation if and only if they have
220
7. Curves and surfaces in projective space
the same projective curvature forms Īŗ Ė 1 = Īŗ1 (t)(dt)3 ,
Īŗ Ė 0 = Īŗ0 (t) ā 12 Īŗ1 (t) (dt)4 .
*Exercise 7.31. Suppose that a nondegenerate, projectively parametrized curve [Ī±] : I ā P3 has projective curvatures Īŗ0 (t) = Īŗ1 (t) = 0. (Note that this condition is invariant under projective reparametrization.) (a) Show that the canonical lifting Ī± : I ā R4 of [Ī±] has the form Ī±(t) = v0 + v1 t + 12 v2 t2 + 16 v3 t3 , where (v0 , v1 , v2 , v3 ) are linearly independent vectors in R4 with
det v0 v1 v2 v3 = 1. In particular, note that Ī± is contained in a 3-dimensional aļ¬ne hyperplane in R4 ānamely, the plane passing through v0 and spanned by the vectors (v1 , v2 , v3 ). (b) Show that [Ī±] is projectively equivalent to the curve [Ī²] with canonical lifting Ī²(t) = t[1, t, 12 t2 , 16 t3 ]. [Ī²] is called the rational normal curve of degree 3 (cf. Exercise 6.17). In general, the rational normal curve [Ī²] of degree n in Pn is the curve parametrized in homogeneous coordinates as [Ī²(t)] = [1 : t : 12 t2 : Ā· Ā· Ā· :
1 n n! t ].
7.4. Moving frames for surfaces in P3 Now, let U be an open set in R2 , and let [x] : U ā P3 be an immersion whose image is a surface [Ī£] = [x(U )]. Just as for curves in P3 , an adapted Ė : U ā SL(4) of the form frame ļ¬eld along [Ī£] is a lifting x Ė (u) = (e0 (u), e1 (u), e2 (u), e3 (u)) , x where for each u ā U , we have [e0 (u)] = [x(u)] ā P3 and
det e0 (u) e1 (u) e2 (u) e3 (u) = 1.
Any choice for the function e0 (u) deļ¬nes a lifting x : U ā R4 of [x], given by x(u) = e0 (u), whose image is a surface Ī£ ā R4 .
7.4. Moving frames for surfaces in P3
221
We might like to proceed as in the equi-aļ¬ne case, by choosing (e1 (u), e2 (u)) so that they span the tangent space Tx(u) Ī£. However, this choice is complicated by the fact that this space is not well-deļ¬ned: If we replace the lifting x by an alternate lifting Ė (u) = Ī»(u)x(u) x for some nonvanishing function Ī» : U ā R, then the tangent vectors xu , xv , which span Tx(u) Ī£, are replaced by the tangent vectors Ė u = Ī»xu + Ī»u x, x
Ė v = Ī»xv + Ī»v x. x
Since the function Ī» and its partial derivatives at any point are arbitrary (aside from the requirement that Ī» = 0), the tangent space Tx(u) Ī£ is only well-deļ¬ned modulo the vector x(u) = e0 (u). Consequently, we can only require that (e1 (u), e2 (u)) span Tx(u) Ī£ modulo e0 (u). More concretely, this means that we will require that the three vectors (e0 (u), e1 (u), e2 (u)) span the 3-dimensional subspace of R4 determined by the position vector x(u) and the tangent plane Tx(u) Ī£. In particular, we will require that this space be 3dimensional, which means that the position vector of Ī£ cannot be contained in its tangent plane at any point. (In fact, this condition is precisely what it means for the smooth map [x] : U ā P3 to be an immersion.) A projective frame ļ¬eld (e0 (u), e1 (u), e2 (u), e3 (u)) along [Ī£] satisfying these conditions will be called 0-adapted. Exercise 7.32. Show that, given a smooth map [x] : U ā P3 , the condition that span(x, xu , xv ) be 3-dimensional is independent of the choice of lifting x : U ā R4 of [x]. Ė : U ā SL(4) is a 0-adapted frame ļ¬eld along [Ī£] = Now, suppose that x [x(U )], and let (ĀÆ ĻĪ²Ī± ) represent the pulled-back forms (Ė xā ĻĪ²Ī± ) on U . Recall from Ā§3.7 that the 1-forms (Ļ01 , Ļ02 , Ļ03 ) are the semi-basic forms for the projection Ļ : SL(4) ā P3 . Thus, these should be regarded as the dual forms, and the remaining (ĻĪ²Ī± ) as the connection forms. Precisely the same reasoning as in the Euclidean and equi-aļ¬ne cases can be used to prove the following: Proposition 7.33. Let U ā R2 be an open set, and let [x] : U ā P3 be an immersion. For any 0-adapted frame ļ¬eld (e0 (u), e1 (u), e2 (u), e3 (u)) along [Ī£] = [x(U )], the associated dual and connection forms (ĀÆ ĻĪ²Ī± ) have the 3 1 2 property that Ļ ĀÆ 0 = 0. Moreover, (ĀÆ Ļ0 , Ļ ĀÆ 0 ) form a basis for the 1-forms on U . Diļ¬erentiating the equation Ļ ĀÆ 03 = 0 yields dĀÆ Ļ03 = āĀÆ Ļ13 ā§ Ļ ĀÆ 01 ā Ļ ĀÆ 23 ā§ Ļ ĀÆ 02 = 0,
222
7. Curves and surfaces in projective space
and Cartanās lemma (cf. Lemma 2.49) implies that there exist functions h11 , h12 , h22 on U such that 3 1 Ļ ĀÆ1 h11 h12 Ļ ĀÆ0 (7.21) = . Ļ ĀÆ 23 h12 h22 Ļ ĀÆ 02 The next step is to investigate how the matrix [hij ] changes if we vary the frame. First, we must identify the group of transformations between 0-adapted frames. Let (e0 (u), e1 (u), e2 (u), e3 (u)) be any 0-adapted frame ļ¬eld, with associated Ė1 (u), Maurer-Cartan forms (ĀÆ ĻĪ²Ī± ). Any other 0-adapted frame ļ¬eld (Ė e0 (u), e Ė2 (u), e Ė3 (u)) must have the properties that e span(Ė e0 (u)) = span(e0 (u)), Ė1 (u), e Ė2 (u)) = span(e0 (u), e1 (u), e2 (u)), span(Ė e0 (u), e and
Ė0 (u) e Ė1 (u) e Ė2 (u) e Ė3 (u) = det e0 (u) e1 (u) e2 (u) e3 (u) = 1. det e
This will be true if and only if
Ė0 (u) e Ė1 (u) e Ė2 (u) e Ė3 (u) e ā¤ s0 Ī» r 1 r2 ā„
ā¢ s1 ā„ ā¢0 = e0 (u) e1 (u) e2 (u) e3 (u) ā¢ ā„ B ā¦ ā£0 s2 0 0 0 (Ī» det B)ā1 ā”
(7.22)
for some GL(2)-valued function B and real-valued functions Ī», r1 , r2 , s0 , s1 , ĖĀÆ Ī²Ī± ) be the Maurer-Cartan forms associated to s2 on U , with Ī» = 0. Let (Ļ the new frame ļ¬eld. The computations in the following exercise should feel familiar from the equi-aļ¬ne case: *Exercise 7.34 (Cf. Exercise 6.20). (a) Show that 1 1 ĖĀÆ 0 Ļ Ļ ĀÆ0 = Ī»B ā1 . ĖĀÆ 02 Ļ Ļ ĀÆ 02 (b) Show that
ĖĀÆ 13 Ļ ĖĀÆ 23 Ļ
= Ī»(det B) tB
Ļ ĀÆ 13 Ļ ĀÆ 23
.
7.4. Moving frames for surfaces in P3
223
Ė 11 , h Ė 12 , h Ė 22 on U such (c) Cartanās lemma implies that there exist functions h that 3 Ė Ė 1 ĖĀÆ 1 ĖĀÆ 0 Ļ h11 h12 Ļ = . Ė 12 h Ė 22 Ļ ĖĀÆ 23 ĖĀÆ 02 Ļ h Show that Ė Ė h11 h12 h11 h12 (7.23) = (det B) tB B. Ė Ė h12 h22 h12 h22 The transformation (7.23) is precisely the same as in the equi-aļ¬ne case, so we use the same terminology here: Deļ¬nition 7.35. Assume that the matrix [hij ] is nonsingular at every point of U . The surface [Ī£] = [x(U )] is called (1) elliptic if det[hij ] > 0 at every point of U ; (2) hyperbolic if det[hij ] < 0 at every point of U . For the remainder of this section, we will assume that [Ī£] is elliptic. The hyperbolic case was treated in detail by Cartan in [Car20]; we will explore this case in Exercise 7.50. The transformation (7.23) acts transitively on the set of 2 Ć 2 matrices with positive determinant; therefore, there exists a choice of 0-adapted frame ļ¬eld (e0 (u), e1 (u), e2 (u), e3 (u)) for which h11 h12 1 0 = . 0 1 h12 h22 Such a frame ļ¬eld will be called 1-adapted. *Exercise 7.36. Let (e0 (u), e1 (u), e2 (u), e3 (u)) be any 1-adapted frame ļ¬eld for an elliptic projective surface [Ī£] = [x(U )] ā P3 . Ė1 (u), e Ė2 (u), e Ė3 (u)) (a) Show that any other 1-adapted frame ļ¬eld (Ė e0 (u), e for [Ī£] must have the form
Ė0 (u) e Ė1 (u) e Ė2 (u) e Ė3 (u) e ā¤ s0 Ī» r 1 r2 ā„
ā¢ s1 ā„ ā¢0 = e0 (u) e1 (u) e2 (u) e3 (u) ā¢ ā„ B ā¦ ā£0 s2 ā1 0 0 0 (Ī» det B) ā”
(7.24)
for some SO(2)-valued function B and real-valued functions Ī», r1 , r2 , s0 , s1 , s2 on U , with Ī» = 0.
224
7. Curves and surfaces in projective space
(b) Let (ĀÆ ĻĪ²Ī± ) be the Maurer-Cartan forms associated to a 1-adapted frame ĖĀÆ Ī²Ī± ) be the Maurer-Cartan forms ļ¬eld (e0 (u), e1 (u), e2 (u), e3 (u)), and let (Ļ associated to the 1-adapted frame ļ¬eld (7.24). Show that 1 2 3 1 ĖĀÆ 13 Ļ ĖĀÆ 01 + Ļ ĖĀÆ 02 = (Ļ ĖĀÆ 23 Ļ ĖĀÆ 01 )2 + (Ļ ĖĀÆ 02 )2 = Ī»2 (ĀÆ Ļ Ļ0 ) + (ĀÆ ĀÆ1 Ļ Ļ02 )2 = Ī»2 Ļ ĀÆ0 + Ļ ĀÆ 23 Ļ ĀÆ 02 . (Cf. Exercise 7.34.) Note that the result of Exercise 7.36(b) is slightly diļ¬erent from the equiaļ¬ne case. Now the quadratic form (7.25)
I=Ļ ĀÆ 13 Ļ ĀÆ 01 + Ļ ĀÆ 23 Ļ ĀÆ 02 = (ĀÆ Ļ01 )2 + (ĀÆ Ļ02 )2
is well-deļ¬ned only up to a scalar multiple. Rather than deļ¬ning a metric, it deļ¬nes a conformal structure on [Ī£], in which angles between tangent vectors are well-deļ¬ned, but lengths of tangent vectors are not. Exercise 7.37. Show that the conformal structure I is invariant (up to a scalar multiple) under the action of SL(4). In order to continue making further adaptations, our next step is to diļ¬erentiate the equations (7.26)
Ļ ĀÆ 13 = Ļ ĀÆ 01 ,
Ļ ĀÆ 23 = Ļ ĀÆ 02 ,
which hold for the Maurer-Cartan forms associated to any 1-adapted frame ļ¬eld. *Exercise 7.38. Diļ¬erentiate the equations (7.26) and use Cartanās lemma to conclude that there exist functions h111 , h112 , h122 , h222 on U such that ā” 1 ā¤ ā¤ ā” 2ĀÆ Ļ1 ā (ĀÆ Ļ00 + Ļ ĀÆ 33 ) h111 h112 ĀÆ 01 ā¢ ā„ Ļ ā„ ā¢ 1 2 ā¢ ā„ ā„ ā¢ Ļ ĀÆ2 + Ļ ĀÆ1 (7.27) ā£ ā¦ = ā£h112 h122 ā¦ 2 . Ļ ĀÆ0 2ĀÆ Ļ22 ā (ĀÆ Ļ00 + Ļ ĀÆ 33 ) h122 h222 Next, we need to compute how the functions (hijk ) vary under a transformation of the form (7.24). This computation gets rather complicated, but we can make it simpler by breaking it down into two steps. *Exercise 7.39. Consider transformations of the form (7.24) with B equal to the identity matrix and Ī» = 1, so that ā” ā¤ 1 r1 r2 s0
ā¢0 1 0 s1 ā„ ā„ Ė0 (u) e Ė1 (u) e Ė2 (u) e Ė3 (u) = e0 (u) e1 (u) e2 (u) e3 (u) ā¢ e ā£0 0 1 s2 ā¦ . 0 0 0 1
7.4. Moving frames for surfaces in P3
225
Show that under such a transformation, we have Ė 111 = h111 + 3(r1 ā s1 ), h Ė 112 = h112 + (r2 ā s2 ), h Ė 122 = h122 + (r1 ā s1 ), h Ė 222 = h222 + 3(r2 ā s2 ). h Therefore, Ė 111 + h Ė 122 = h111 + h122 + 4(r1 ā s1 ), h Ė 112 + h Ė 222 = h112 + h222 + 4(r2 ā s2 ), h so there exists a choice of 1-adapted frame ļ¬eld (e0 (u), e1 (u), e2 (u), e3 (u)) for which (7.28)
h111 + h122 = h112 + h222 = 0.
Any 1-adapted frame ļ¬eld satisfying the condition (7.28) will be called 2adapted. Remark 7.40. This is certainly not the only way that one might choose to normalize these functions! But our experience with the equi-aļ¬ne caseā where a similar relation held for any 2-adapted frame ļ¬eldāsuggests that this might turn out to be a nice choice. *Exercise 7.41. Let (e0 (u), e1 (u), e2 (u), e3 (u)) be any 2-adapted frame ļ¬eld for an elliptic projective surface [Ī£] = [x(U )] ā P3 . Ė1 (u), e Ė2 (u), e Ė3 (u)) (a) Show that any other 2-adapted frame ļ¬eld (Ė e0 (u), e for [Ī£] must have the form (7.24), where r1 s1 (7.29) = Ī» tB . r2 s2 (b) Show that under a transformation of the form (7.24) satisfying equation (7.29), we have Ė h111 h111 ā1 3 (7.30) =Ī» B . Ė 222 h222 h (c) Let (ĀÆ ĻĪ²Ī± ) be the Maurer-Cartan forms associated to the 2-adapted frame ĖĀÆ Ī²Ī± ) be the Maurer-Cartan forms ļ¬eld (e0 (u), e1 (u), e2 (u), e3 (u)), and let (Ļ Ė1 (u), e Ė2 (u), e Ė3 (u)). Show associated to the 2-adapted frame ļ¬eld (Ė e0 (u), e
226
7. Curves and surfaces in projective space
that 1 3 2 3 Ė 111 (Ļ Ė 222 (Ļ ĖĀÆ 0 ) ā 3Ļ ĖĀÆ 0 ) ā 3(Ļ ĖĀÆ 02 ĖĀÆ 01 (Ļ ĖĀÆ 02 )2 + h ĖĀÆ 01 )2 Ļ h
1 3 2 3 = Ī»2 h111 (ĀÆ Ļ0 ) ā 3ĀÆ Ļ0 ) ā 3(ĀÆ Ļ01 (ĀÆ Ļ02 )2 + h222 (ĀÆ Ļ01 )2 Ļ ĀÆ 02 . Conclude that the cubic form P = hijk Ļ ĀÆ 0i Ļ ĀÆ 0j Ļ ĀÆ 0k (7.31)
= h111 (ĀÆ Ļ01 )3 + 3h112 (ĀÆ Ļ01 )2 Ļ ĀÆ 02 + 3h122 Ļ ĀÆ 01 (ĀÆ Ļ02 )2 + h222 (ĀÆ Ļ02 )3 1 3 2 3 Ļ0 ) ā 3 Ļ Ļ0 ) ā 3(ĀÆ = h111 (ĀÆ ĀÆ 01 (ĀÆ Ļ02 )2 + h222 (ĀÆ Ļ01 )2 Ļ ĀÆ 02
is well-deļ¬ned up to a scalar multiple, independent of the choice of a particular 2-adapted frame ļ¬eld on [Ī£]. (P is the projective analog of the Fubini-Pick form for surfaces in equi-aļ¬ne space.) According to (7.30), under a transformation from one 2-adapted frame to another, the vector t h111 h222 transforms by a rotation and a scaling. This vector is an example of a relative invariant: If it vanishes for any 2-adapted frame based at a point u ā U , then it vanishes for every 2-adapted frame based at u. On the other hand, if this vector is nonzero for some 2-adapted frame based at u, then we can ļ¬nd a 2-adapted frame based at u for which it is equal to any nonzero vector we choose. Any point [x(u)] ā [Ī£] at which h111 = h222 = 0 will be called an umbilic point of Ī£.
For the remainder of this section, we will assume that the vector t h111 h222 is nonzero for every 2-adapted frame based at each point u ā U , i.e., that Ī£ contains no umbilic points. (We will treat the totally umbilic case, where h111 ā” h222 ā” 0, in Exercises 7.48 and 7.49.) With this assumption, equation (7.30) implies that there exists a choice of 2-adapted frame ļ¬eld for which h111 = 2,
h222 = 0.
Such a frame ļ¬eld will be called 3-adapted. *Exercise 7.42. (a) Let (e0 (u), e1 (u), e2 (u), e3 (u)) be any 3-adapted frame ļ¬eld for an elliptic projective surface [Ī£] = [x(U )] ā P3 with no umbilic Ė1 (u), e Ė2 (u), points. Show that any other 3-adapted frame ļ¬eld (Ė e0 (u), e Ė3 (u)) for [Ī£] must have the form described in Exercise 7.41, where either e Ī» = 1 and B is one of the three matrices ā ā 1 0 3 1 ā1 ā 3 1 ā1 , , ā ā 2 2 ā 3 ā1 0 1 3 ā1 or Ī» = ā1 and B is one of the three matrices ā ā1 0 1 1 ā 3 , , ā 2 0 ā1 3 1
ā 3 1 1 . ā 2 ā 3 1
7.4. Moving frames for surfaces in P3
227
(b) Let (ĀÆ ĻĪ²Ī± ) be the Maurer-Cartan forms associated to a 3-adapted frame ĖĀÆ Ī²Ī± ) be the Maurer-Cartan forms ļ¬eld (e0 (u), e1 (u), e2 (u), e3 (u)), and let (Ļ Ė1 (u), e Ė2 (u), e Ė3 (u)). associated to any other 3-adapted frame ļ¬eld (Ė e0 (u), e Show that ĖĀÆ 01 )2 + (Ļ ĖĀÆ 02 )2 = (ĀÆ (Ļ Ļ01 )2 + (ĀÆ Ļ02 )2 . Therefore, the conformal structure (7.25) associated to a 3-adapted frame ļ¬eld provides a well-deļ¬ned metric for an elliptic surface in P3 with no umbilic points. Deļ¬nition 7.43. Let U ā R2 be an open set, and let [x] : U ā P3 be an elliptic immersion with no umbilic points. Let (e0 (u), e1 (u), e2 (u), e3 (u)) be any 3-adapted frame ļ¬eld along [Ī£] = [x(U )], with associated MaurerCartan forms (ĀÆ ĻĪ²Ī± ). The quadratic form I = (ĀÆ Ļ01 )2 + (ĀÆ Ļ02 )2 is called the projective ļ¬rst fundamental form of [Ī£]. Remark 7.44. For a 3-adapted frame ļ¬eld, the cubic form (7.31) becomes 1 3 P = 2 (ĀÆ Ļ0 ) ā 3 Ļ ĀÆ 01 (ĀÆ Ļ02 )2 . The 3-adapted frames have the property that the tangent vector e2 (u) is one of the three null directions for P at the point [x(u)] ā [Ī£]. These directions are called the Darboux tangents at the point [x(u)] ā [Ī£]; for more details, see [Su83]. For simplicity, we will restrict our attention to transformations between 3adapted frames with B equal to the identity matrix; thus, we will assume that any two 3-adapted frame ļ¬elds vary by a transformation of the form (7.32) ā” ā¤ 1 s1 s2 s0
ā¢0 1 0 s1 ā„ ā„ Ė0 (u) e Ė1 (u) e Ė2 (u) e Ė3 (u) = e0 (u) e1 (u) e2 (u) e3 (u) ā¢ e ā£0 0 1 s2 ā¦ . 0 0 0 1 *Exercise 7.45. (a) Show that the Maurer-Cartan forms associated to any 3-adapted frame ļ¬eld satisfy the equations Ļ ĀÆ 00 + Ļ ĀÆ 33 = Ļ ĀÆ 11 + Ļ ĀÆ 22 = 0. (Hint: Apply the result of Exercise 7.38, keeping in mind condition (7.28), and recall the deļ¬ning condition for the Lie algebra sl(4).)
228
7. Curves and surfaces in projective space
(b) Diļ¬erentiate the equation Ļ ĀÆ 00 + Ļ ĀÆ 33 = 0 and use Cartanās lemma to conclude that there exist functions 11 , 12 , 22 on U such that
Ļ ĀÆ 10 ā Ļ ĀÆ 31
(7.33)
Ļ ĀÆ 20 ā Ļ ĀÆ 32
=
11 12
12 22
Ļ ĀÆ 01 Ļ ĀÆ 02
.
*Exercise 7.46. (a) Show that under a transformation of the form (7.32), we have Ė11 = 11 + s21 + s22 ā 2s0 ā 2s1 , Ė12 = 12 + 2s2 , Ė22 = 22 + s2 + s2 ā 2s0 + 2s1 . 1
2
(b) Conclude that there exists a choice of 3-adapted frame ļ¬eld for which 11 = 12 = 22 = 0. Any frame ļ¬eld satisfying this condition will be called 4-adapted. (c) Show that any two 4-adapted frame ļ¬elds vary by a transformation of the form ā¤ Ī» 0 0 0
ā¢ 0 ā„ ā„ ā¢0 Ė0 (u) e Ė1 (u) e Ė2 (u) e Ė3 (u) = e0 (u) e1 (u) e2 (u) e3 (u) ā¢ e ā„, B ā£0 0 ā¦ 0 0 0 Ī»ā1 ā”
where (Ī», B) are one of the pairs from Exercise 7.42. Therefore, the 4adapted frames at each point form a discrete set, and a continuous 4-adapted frame ļ¬eld along [Ī£] is uniquely determined by its value at any point u ā U . *Exercise 7.47. (a) Show that the Maurer-Cartan forms associated to a 4-adapted frame ļ¬eld satisfy the equations (7.34)
Ļ ĀÆ 11 = Ļ ĀÆ 01 ,
Ļ ĀÆ 22 = āĀÆ Ļ01 , Ļ ĀÆ 10 = Ļ ĀÆ 31 ,
Ļ ĀÆ 21 + Ļ ĀÆ 12 = ā2ĀÆ Ļ02 , Ļ ĀÆ 20 = Ļ ĀÆ 32 .
7.4. Moving frames for surfaces in P3
229
(b) Diļ¬erentiate the equations in part (a) and use Cartanās lemma to conclude that there exist functions p1 , p2 , q11 , q12 , q21 , q22 such that Ļ ĀÆ 21 = p1 Ļ ĀÆ 01 + (p2 ā 1)ĀÆ Ļ02 , Ļ ĀÆ 12 = āp1 Ļ ĀÆ 01 ā (p2 + 1)ĀÆ Ļ02 , (7.35)
Ļ ĀÆ 31 = q11 Ļ ĀÆ 01 + q12 Ļ ĀÆ 02 , Ļ ĀÆ 32 = q21 Ļ ĀÆ 01 + q22 Ļ ĀÆ 02 , Ļ ĀÆ 00 = āĀÆ Ļ33 = 3p2 Ļ ĀÆ 01 ā 3p1 Ļ ĀÆ 02 , Ļ ĀÆ 30 = (q11 ā q22 )ĀÆ Ļ01 ā (q12 + q21 )ĀÆ Ļ02 .
The functions p1 , p2 , q11 , q12 , q21 , q22 are the fundamental invariants for elliptic surfaces in P3 with no umbilic points. In the next two exercises, we consider the totally umbilic case. Exercise 7.48 (Projective spheres, Part 1). Let (e0 (u), e1 (u), e2 (u), e3 (u)) be any 2-adapted frame ļ¬eld for an elliptic projective surface [Ī£] = [x(U )] ā P3 , and suppose that [Ī£] is totally umbilic, i.e., that h111 = h222 = 0. (a) Use the result of Exercise 7.38 to show that the Maurer-Cartan forms associated to this frame ļ¬eld satisfy (7.36)
Ļ ĀÆ 11 = Ļ ĀÆ 22 = Ļ ĀÆ 21 + Ļ ĀÆ 12 = Ļ ĀÆ 00 + Ļ ĀÆ 33 = 0.
(b) Diļ¬erentiate equations (7.36) to obtain (ĀÆ Ļ10 ā Ļ ĀÆ 31 ) ā§ Ļ ĀÆ 01 = 0, (ĀÆ Ļ20 ā Ļ ĀÆ 32 ) ā§ Ļ ĀÆ 02 = 0, (ĀÆ Ļ20 ā Ļ ĀÆ 32 ) ā§ Ļ ĀÆ 01 + (ĀÆ Ļ10 ā Ļ ĀÆ 31 ) ā§ Ļ ĀÆ 02 = 0. Use Cartanās lemma to conclude that there exists a function Ļ such that Ļ ĀÆ 10 ā Ļ ĀÆ 31 = Ļ Ļ ĀÆ 01 , Ļ ĀÆ 20 ā Ļ ĀÆ 32 = Ļ Ļ ĀÆ 02 . (c) Show that under a transformation of the form
ā” 1
ā¢0 Ė0 (u) e Ė1 (u) e Ė2 (u) e Ė3 (u) = e0 (u) e1 (u) e2 (u) e3 (u) ā¢ e ā£0 0
we have Ļ Ė = Ļ ā 2s0 .
0 1 0 0
0 0 1 0
ā¤ s0 0ā„ ā„, 0ā¦ 1
230
7. Curves and surfaces in projective space
Conclude that there exists a 2-adapted frame ļ¬eld with Ļ = 0 and that for such a frame ļ¬eld we have Ļ ĀÆ 31 = Ļ ĀÆ 10 ,
(7.37)
Ļ ĀÆ 32 = Ļ ĀÆ 20 .
We will call such a frame ļ¬eld fully adapted. (d) Show that any two fully adapted frame ļ¬elds diļ¬er by a of the form
Ė0 (u) e Ė1 (u) e Ė2 (u) e Ė3 (u) e ā” Ī» r 1 r2 ā¢ (7.38)
ā¢ 0 = e0 (u) e1 (u) e2 (u) e3 (u) ā¢ B ā£0 0 0 0
transformation
ā¤ s0 s1 ā„ ā„ ā„, s2 ā¦ Ī»ā1
where B ā SO(2),
[r1 r2 ] = Ī»[s1 s2 ]B,
1 s0 = Ī»(s21 + s22 ). 2
(e) Diļ¬erentiate equations (7.37) to obtain 2ĀÆ Ļ30 ā§ Ļ ĀÆ 01 = 0, 2ĀÆ Ļ30 ā§ Ļ ĀÆ 02 = 0. Use Cartanās lemma to conclude that Ļ ĀÆ 30 = 0. (f) Show that diļ¬erentiating the equation Ļ ĀÆ 30 = 0 yields an identity. At this point, the Maurer-Cartan form for the bundle of fully adapted frames over [Ī£] is ā” 0 0 0 ā¤ Ļ ĀÆ0 Ļ ĀÆ1 Ļ ĀÆ2 0 ā¢ 1 ā„ ā¢Ļ ĀÆ 21 Ļ ĀÆ 10 ā„ ā¢ ĀÆ0 0 Ļ ā„ Ļ ĀÆ=ā¢ ā„. 2 1 0 ā¢Ļ Ļ2 0 Ļ ĀÆ2 ā„ ā£ ĀÆ 0 āĀÆ ā¦ 0
Ļ ĀÆ 01 Ļ ĀÆ 02 āĀÆ Ļ00
We have not yet found a unique frame over each point of [Ī£], but since diļ¬erentiating the structure equations yields no further relations, this is as far as the frame bundle can be reduced. This means that the bundle of fully adapted frames over [Ī£] is itself a Lie group G whose Lie algebra g is the space of matrices with the symmetries of the Maurer-Cartan form above and that [Ī£] is a homogeneous space of the form G/H, where H is the transformation group in equation (7.38). All that remains is to identify the group G and the quotient space G/H.
7.4. Moving frames for surfaces in P3
231
Exercise 7.49 (Projective spheres, Part 2). Let Q be the matrix ā” ā¤ 0 0 0 ā1 ā¢0 10 0ā„ ā„ Q=ā¢ ā£ 0 0 1 0 ā¦. ā1 0 0 0 Q represents the quadratic form Q(x) = (x1 )2 + (x2 )2 ā 2x0 x3 , which is an indeļ¬nite quadratic form of signature (3, 1) on R4 . The Lie group SO(Q) is deļ¬ned to be the group of matrices of determinant 1 that preserve this quadratic form; i.e., SO(Q) = {A ā SL(4) | tAQA = Q}. The group SO(Q) is isomorphic to the Lie group SO(3, 1). (a) Show that the Lie algebra so(Q) is deļ¬ned by so(Q) = {B ā sl(4) | tBQ + QB = 0}. (Hint: Let A(t) be a curve in SO(Q) with A(0) = I. Diļ¬erentiate the equation t
A(t)QA(t) = Q
and evaluate at t = 0.) (b) Show that so(Q) consists of all matrices of the form ā” 0 0 0 ā¤ a0 a1 a2 0 ā¢ 1 ā„ ā¢a0 0 a12 a01 ā„ ā„ B=ā¢ ā¢a2 āa1 0 a0 ā„ . ā£ 0 2 2 ā¦ 0 a10 a20 āa00 Conclude that the reduced Maurer-Cartan form on the bundle of fully adapted frames in Exercise 7.48 takes values in so(Q). (c) Recall that for any fully adapted frame ļ¬eld Ė (u) = (e0 (u), e1 (u), e2 (u), e3 (u)) x Ė as in Exercise 7.48, the pullback of the Maurer-Cartan form Ļ ĀÆ to U via x ā ā1 Ė Ė is given by x (ĀÆ Ļ ) = x dĖ x. Use this fact to show that any fully adapted frame ļ¬eld has the form Ė (u) = A0 A(u), x where A0 ā SL(4) is a constant matrix and A(u) ā SO(Q) for all u ā U .
232
7. Curves and surfaces in projective space
Ė (u) is projectively equivalent to the Since A0 ā SL(4), the frame ļ¬eld x ā1 Ė (u). (This equivalence replaces the surface frame ļ¬eld A0 x [Ī£] = [x(U )] = [e0 (U )] ā P3 with the projectively equivalent surface ā1 ā1 3 [Aā1 0 Ī£] = [A0 x(U )] = [A0 e0 (U )] ā P .)
Ė (u) ā SO(Q) Therefore, up to projective equivalence, we can assume that x for all u ā U .
(d) Show that for any matrix e0 e1 e2 e3 ā SO(Q), the vector e0 is a null vector for Q; i.e., Q(e0 ) = 0. Therefore, the surface x(U ) = e0 (U ) ā R4 \{0} must be contained in the 3-dimensional hypersurface S ā R4 \ {0} deļ¬ned by the equation Q(x) = 0. (e) Show that S consists of all lines through the origin passing through all points of the form x = t[1, x1 , x2 , x3 ] with (x1 )2 + (x2 )2 + (x3 )2 = 1. Thus, the image of S under the quotient map R4 \ {0} ā P3 may be regarded as a sphere in P3 , and [Ī£] = [x(U )] must be an open subset of this sphere. Exercise 7.50. In this exercise, we explore the frame adaptation process for hyperbolic projective surfaces. Let [x] : U ā P3 be a smooth immersion whose image is a hyperbolic projective surface [Ī£]. Then the matrix [hij ] in (7.21) has det[hij ] < 0. (a) Show that there exists a 0-adapted frame ļ¬eld (f0 (u), f1 (u), f2 (u), f3 (u)) along Ī£ for which h11 h12 0 1 = . 1 0 h12 h22 Such a frame ļ¬eld will be called a 1-adapted null frame ļ¬eld along [Ī£]. The associated Maurer-Cartan forms (ĀÆ Ī·Ī²Ī± ) satisfy (7.39)
Ī·ĀÆ13 = Ī·ĀÆ02 ,
Ī·ĀÆ23 = Ī·ĀÆ01 ,
and the conformal structure on [Ī£] is the indeļ¬nite quadratic form I = Ī·ĀÆ13 Ī·ĀÆ01 + Ī·ĀÆ23 Ī·ĀÆ02 = 2ĀÆ Ī·01 Ī·ĀÆ02 .
7.4. Moving frames for surfaces in P3
233
(b) Let (f0 (u), f1 (u), f2 (u), f3 (u)) be any 1-adapted null frame ļ¬eld for a hyperbolic projective surface [Ī£] = [x(U )] ā P3 . Show that any other 1adapted null frame ļ¬eld (Ėf0 (u), Ėf1 (u), Ėf2 (u), Ėf3 (u)) for Ī£ must have the form (7.40) ā¤ ā” Ī» r1 r2 s0 ā„ ā¢
ā¢ 0 eĪø 0 s1 ā„ ā„ Ėf0 (u) Ėf1 (u) Ėf2 (u) Ėf3 (u) = f0 (u) f1 (u) f2 (u) f3 (u) ā¢ ā¢ 0 0 eāĪø s ā„ 2 ā¦ ā£ 0 0 0 Ī»ā1 for some real-valued functions Ī», Īø, r1 , r2 , s0 , s1 , s2 on U , with Ī» = 0. (c) Diļ¬erentiate equations (7.39) and use Cartanās lemma to conclude that there exist functions h111 , h112 , h122 , h222 on U such that ā” ā¤ ā” ā¤ 2ĀÆ Ī·12 h111 h112 ā¢ 1 ā„ ā¢ ā„ Ī·ĀÆ01 2 0 3 ā¢(ĀÆ ā„ ā¢ Ī·0 + Ī·ĀÆ3 )ā¦ = ā£h112 h122 ā„ ā£ Ī·1 + Ī·ĀÆ2 ) ā (ĀÆ ā¦ 2 . Ī·ĀÆ0 2ĀÆ Ī·21 h122 h222 (d) Show that under a transformation of the form (7.40), we have
(7.41)
Ė 111 = Ī»ā1 e3Īø h111 , h Ė 112 = Ī»ā1 eĪø h112 + 2(Ī»ā1 r1 ā eĪø s2 ), h Ė 122 = Ī»ā1 eāĪø h122 + 2(Ī»ā1 r2 ā eāĪø s1 ), h Ė 222 = Ī»ā1 eā3Īø h222 . h
Therefore, there exists a choice of 1-adapted null frame ļ¬eld (f0 (u), f1 (u), f2 (u), f3 (u)) for which h112 = h122 = 0. Any 1-adapted null frame ļ¬eld satisfying this condition will be called a 2adapted null frame ļ¬eld. (e) Show that any other 2-adapted null frame ļ¬eld (Ėf0 (u), Ėf1 (u), Ėf2 (u), Ėf3 (u)) for [Ī£] must have the form (7.40), where (7.42)
r1 = Ī»eĪø s2 ,
r2 = Ī»eāĪø s1 .
Equations (7.41) imply that the functions h111 , h222 are relative invariants; for the remainder of this exercise, we will assume that both h111 , h222 are nonzero. (Cartan showed in [Car20] that if either of these functions is identically zero, then [Ī£] is a ruled surface in P3 .) Equations (7.41) then imply that there exists a choice of 2-adapted null frame ļ¬eld for which h111 = h222 = 2.
234
7. Curves and surfaces in projective space
(This may require a slight extension of the transformation (7.40) to include transformations of the form fĪ± (u) ā āfĪ± (u) for an even number of indices Ī±.) Any 2-adapted null frame ļ¬eld satisfying this condition will be called a 3-adapted null frame ļ¬eld. (f) Show that, up to signs, any two 3-adapted null frame ļ¬elds vary by a transformation of the form ā” ā¤ 1 s2 s1 s0
ā¢0 1 0 s1 ā„ ā„ (7.43) Ėf0 (u) Ėf1 (u) Ėf2 (u) Ėf3 (u) = f0 (u) f1 (u) f2 (u) f3 (u) ā¢ ā£0 0 1 s2 ā¦ . 0 0 0 1 (g) Show that the Maurer-Cartan forms associated to any 3-adapted null frame ļ¬eld satisfy the equations Ī·ĀÆ00 + Ī·ĀÆ33 = Ī·ĀÆ11 + Ī·ĀÆ22 = 0. Diļ¬erentiate the equation Ī·ĀÆ00 + Ī·ĀÆ33 = 0 and use Cartanās lemma to conclude that there exist functions 11 , 12 , 22 on U such that 0 Ī·ĀÆ1 ā Ī·ĀÆ32 11 12 Ī·ĀÆ01 = . Ī·ĀÆ20 ā Ī·ĀÆ31 12 22 Ī·ĀÆ02 (h) Show that under a transformation of the form (7.43), we have Ė11 = 11 ā 2s1 , Ė12 = 12 + 2s1 s2 ā 2s0 , Ė22 = 22 ā 2s2 . Conclude that there exists a choice of 3-adapted null frame ļ¬eld for which 11 = 12 = 22 = 0. Any frame ļ¬eld satisfying this condition will be called a 4-adapted null frame ļ¬eld. The 4-adapted null frames at each point form a discrete set (they are unique up to signs), and a continuous 4-adapted null frame ļ¬eld along [Ī£] is uniquely determined by its value at any point u ā U . (i) Show that the Maurer-Cartan forms associated to a 4-adapted null frame ļ¬eld satisfy the equations Ī·ĀÆ12 = Ī·ĀÆ01 ,
Ī·ĀÆ21 = Ī·ĀÆ02 ,
Ī·ĀÆ10 = Ī·ĀÆ32 ,
Ī·ĀÆ20 = Ī·ĀÆ31 .
7.5. Maple computations
235
Diļ¬erentiate these equations and use Cartanās lemma to conclude that there exist functions p1 , p2 , q11 , q12 , q21 , q22 such that Ī·ĀÆ11 = āĀÆ Ī·22 = p1 Ī·ĀÆ01 + p2 Ī·ĀÆ02 , Ī·ĀÆ00 = āĀÆ Ī·33 = ā3p1 Ī·ĀÆ01 + 3p2 Ī·ĀÆ02 , Ī·ĀÆ31 = q11 Ī·ĀÆ01 + q12 Ī·ĀÆ02 , Ī·ĀÆ32 = q21 Ī·ĀÆ01 + q22 Ī·ĀÆ02 , Ī·ĀÆ30 = q12 Ī·ĀÆ01 + q21 Ī·ĀÆ02 . The functions p1 , p2 , q11 , q12 , q21 , q22 are the fundamental invariants for nonruled, hyperbolic surfaces in P3 .
7.5. Maple computations In this chapter, the computations for curves in P2 and P3 are already suļ¬ciently complicated that some assistance from Maple may be appreciated. Here we will explore some of the exercises for curves in P3 ; the case of curves in P2 is left as an exercise for the reader. As usual, begin by loading the Cartan and LinearAlgebra packages into Maple. Ė0 (s) of [Ī²(s)] must be equal to a Exercise 7.25: The canonical lifting e scalar multiple of e0 (t(s)); say Ė0 (s) = Ī»(s) e0 (t(s)). e For Maple purposes, we will denote the canonical projective frame ļ¬eld for [Ī²(s)] by (f0(s), f1(s), f2(s), f3(s)). > f0(s):= lambda(s)*e0(t(s)); Ė0 (s): Compute the ļ¬rst three derivatives of e > f1(s):= diff(f0(s), s); f2(s):= diff(f1(s), s); f3(s):= diff(f2(s), s); Maple expresses the output from these computations in terms of the derivatives of e0 (t(s)); for instance, it returns f 1(s) := Ī»s e0(t(s)) + Ī»(s) D(e0)(t(s)) ts (The expression D(e0)(t(s)) refers to the derivative t = t(s).)
d dt
(e0 (t)) evaluated at
We donāt have any good way of telling Maple that these derivatives are all linearly independent, but we can still collect terms and extract the coeļ¬cients needed to compute the necessary determinant.
236
7. Curves and surfaces in projective space
> collect(f1(s), {e0(t(s)), D(e0)(t(s))}); collect(f2(s), {e0(t(s)), D(e0)(t(s)), D(D(e0))(t(s))}); collect(f3(s), {e0(t(s)), D(e0)(t(s)), D(D(e0))(t(s)), D(D(D(e0)))(t(s))});
Ė0 (s) e Ė1 (s) e Ė2 (s) e Ė3 (s) can be Upon inspection, it becomes clear that det e computed as follows: > detf:= coeff(f0(s), e0(t(s)))*coeff(f1(s), D(e0)(t(s)))* coeff(f2(s), D(D(e0))(t(s)))*coeff(f3(s), D(D(D(e0)))(t(s))); detf := Ī»(s)4 t6s Since this determinant must be equal to 1, we must have Ī»(s) =
1 (t (s))(3/2)
.
Now computing the rest of the canonical projective frame ļ¬eld for [Ī²] is simply a matter of diļ¬erentiating the expression Ė0 (s) = e
1
e0 (t(s)). (t (s))(3/2)
But keep in mind that we still need to compute the Wilczynski invariants, Ė3 (s) as a linear combination of (Ė Ė1 (s), which requires expressing e e0 (s), e Ė2 (s)). This can get a bit messy, but we can keep things better organized if, e instead of explicitly expressing everything as functions of t and s, we instead work with appropriately chosen diļ¬erential forms. In this context, we assign the derivatives (and the Wilczynski invariants) of the original frame ļ¬eld as follows: > d(e0):= d(e1):= d(e2):= d(e3):=
e1*d(t); e2*d(t); e3*d(t); e0*kappa0*d(t) + e1*kappa1*d(t) + e2*kappa2*d(t);
Next, observe that the function t(s) never appears explicitly; all that matters is its derivative t (s). So, let r(s) denote t (s), and set up substitutions that Ė0 to e0 : relate dt to ds and e > PDETools[declare](r(s)); tsub:= [d(t) = r(s)*d(s)]; fsub:= [f0 = e0/r(s)Ė(3/2)];
7.5. Maple computations
237
Ė0 with respect to s by computing dĖ Compute the derivative of e e0 and then substituting for dt in terms of ds: > Simf(subs(tsub, Simf(d(Simf(subs(fsub, f0)))))); Ė1 (s), so add it to fsub: The coeļ¬cient of ds in this expression should be e > fsub:= [op(fsub), f1 = pick(%, d(s))]; Ė2 (s) and e Ė3 (s): Similarly for e > Simf(subs(tsub, Simf(d(Simf(subs(fsub, f1)))))); fsub:= [op(fsub), f2 = pick(%, d(s))]; Simf(subs(tsub, Simf(d(Simf(subs(fsub, f2)))))); fsub:= [op(fsub), f3 = pick(%, d(s))]; In order to compute the Wilczinksi invariants for the reparametrized curve, Ė1 , e Ė2 ). The makebacksub command we need to express dĖ e3 in terms of (Ė e0 , e wonāt work here because (e0 , e1 , e2 , e3 ) are not 1-forms, but we can create the reverse substitution as follows: > esub:= [op(solve({op(fsub)}, {e0, e1, e2, e3}))]; Now we can compute dĖ e3 and express the result in terms of the canonical projective frame ļ¬eld for [Ī²]: > Simf(subs(esub, Simf(subs(tsub, Simf(d(Simf(subs(fsub, f3)))))))); df3:= collect(pick(%, d(s)), {f0, f1, f2, f3}); Ė3 term, and the invariants As expected, we see that the output contains no e Īŗ Ė 0 (s), Īŗ Ė 1 (s), Īŗ Ė 2 (s) associated to [Ī²] are given by > Kappa0:= coeff(df3, f0); Kappa1:= coeff(df3, f1); Kappa2:= coeff(df3, f2); Exercise 7.28: First, it will be helpful to introduce variables to represent the derivatives of Īŗ0 (t), Īŗ1 (t), Īŗ2 (t): > d(kappa0):= kappa0 t*d(t); d(kappa1):= kappa1 t*d(t); d(kappa2):= kappa2 t*d(t); d(kappa0 t):= kappa0 tt*d(t); d(kappa1 t):= kappa1 tt*d(t); d(kappa2 t):= kappa2 tt*d(t);
238
7. Curves and surfaces in projective space
Next, we will need to express some of the derivatives of Īŗ Ė 0 (s), Īŗ Ė 1 (s), Īŗ Ė 2 (s) (with respect to s) in terms of Īŗ0 (t), Īŗ1 (t), Īŗ2 (t) and their derivatives (with respect to t). As it turns out, these are the ones that we will need: > Kappa1 s:= Simf(pick(Simf(subs(tsub, d(Kappa1))), d(s))); Kappa2 s:= Simf(pick(Simf(subs(tsub, d(Kappa2))), d(s))); Kappa2 ss:= Simf(pick(Simf(subs(tsub, d(Kappa2 s))), d(s))); In order to identify invariant forms, the trick is to look for combinations of Īŗ Ė 0 (s), Īŗ Ė 1 (s), Īŗ Ė 2 (s) and their derivatives that donāt contain any derivatives of r(s), so that they transform tensorially under the change of variables t ā t(s). The ļ¬rst one is relatively easy to spot: The computations above show that Īŗ Ė 1 (s) = r(s)3 Īŗ1 (t) + 2 r(s)r (s)Īŗ2 (t) 1 2 3 + 20 r(s)r (s)r (s) ā 5 r(s) r (s) ā 15 r (s) r(s)3 (compare with equation (7.16)), while Īŗ Ė 2 (s) = r(s)3 Īŗ2 (t) + 2 r(s)r (s)Īŗ2 (t) 1 + 20 r(s)r (s)r (s) ā 5 r(s)2 r (s) ā 15 r (s)3 . 3 r(s) Therefore, we have Īŗ Ė 1 (s) ā Īŗ Ė 2 (s) = r(s)3 Īŗ1 (t) ā Īŗ2 (t) , and hence Īŗ Ė 1 (s) ā Īŗ Ė 2 (s) ds3 = Īŗ1 (t) ā Īŗ2 (t) (r(s) ds)3 = Īŗ1 (t) ā Īŗ2 (t) dt3 . The quartic form requires a bit more ļ¬nesse to identify; for details, see the Maple worksheet for this chapter on the AMS webpage. But once we know the answer, verifying it is straightforward: > Simf(Kappa0 - (1/2)*Kappa1 s + (1/5)*Kappa2 ss + (9/100)*Kappa2Ė2); 1 4 r 100 Īŗ0 + 9 Īŗ22 + 20 kappa2 tt ā 50 kappa1 t 100 Now we move on to surfaces in P3 . The initial setup is similar to that in Chapter 6, with a slightly diļ¬erent collection of indices: > Form(omega[0,0], omega[0,1], omega[0,2], omega[0,3], omega[1,0], omega[1,1], omega[1,2], omega[1,3],
7.5. Maple computations
239
omega[2,0], omega[2,1], omega[2,2], omega[2,3], omega[3,0], omega[3,1], omega[3,2], omega[3,3]); for i from 0 to 3 do for j from 0 to 3 do d(omega[i,j]):= -add(āomega[i,k] &Ė omega[k,j]ā, k=0..3); end do; end do; Itās a little bit clumsy to introduce the relation Ļ ĀÆ 00 + Ļ ĀÆ 11 + Ļ ĀÆ 22 + Ļ ĀÆ 33 = 0 at this point because doing so would require solving for one of the forms in terms of the others, and itās not yet clear which form would be best to solve for. So for now, weāll simply keep the relation in mind, and weāll tell Maple about it when it becomes convenient to do so. Next, set up the initial substitution for the Maurer-Cartan forms of a 0adapted frame ļ¬eld: > adaptedsub1:= [omega[3,0]=0, omega[3,1] = h[1,1]*omega[1,0] + h[1,2]*omega[2,0], omega[3,2] = h[1,2]*omega[1,0] + h[2,2]*omega[2,0]]; Exercise 7.34: Introduce new 1-forms to represent the transformed forms: > Form(Omega[0,0], Omega[0,1], Omega[0,2], Omega[0,3], Omega[1,0], Omega[1,1], Omega[1,2], Omega[1,3], Omega[2,0], Omega[2,1], Omega[2,2], Omega[2,3], Omega[3,0], Omega[3,1], Omega[3,2], Omega[3,3]); Ultimately, weāre going to need to know how all of the Maurer-Cartan forms transform as we gradually reduce the transformation group. And while it would be quite a mess to write everything out explicitly, we know that under a transformation of the form (7.22), we have (7.44)
Ī±
Ī± ĖĀÆ Ī² = Aā1 dA + Aā1 Ļ Ļ ĀÆ Ī² A,
where A is the matrix in equation (7.22). We can go ahead and set up this substitution for all the Maurer-Cartan forms at onceāalthough you will probably want to suppress the output so as not to have Maple spew out pages and pages of nasty expressions that we donāt really need to see!
240
7. Curves and surfaces in projective space
We can set up the matrix computation (7.44) as follows: First, introduce the necessary matrices: > omegamatrix:= Matrix([ [omega[0,0], omega[0,1], omega[0,2], omega[0,3]], [omega[1,0], omega[1,1], omega[1,2], omega[1,3]], [omega[2,0], omega[2,1], omega[2,2], omega[2,3]], [omega[3,0], omega[3,1], omega[3,2], omega[3,3]]]); > Omegamatrix:= Matrix([ [Omega[0,0], Omega[0,1], Omega[0,2], Omega[0,3]], [Omega[1,0], Omega[1,1], Omega[1,2], Omega[1,3]], [Omega[2,0], Omega[2,1], Omega[2,2], Omega[2,3]], [Omega[3,0], Omega[3,1], Omega[3,2], Omega[3,3]]]); > groupmatrix:= Matrix([ [lambda, r[1], r[2], s[0]], [0, b[1,1], b[1,2], s[1]], [0, b[2,1], b[2,2], s[2]], [0,0,0,1/(lambda*(b[1,1]*b[2,2] - b[1,2]*b[2,1]))]]); Next, compute the right-hand side of equation (7.44) (and youāll deļ¬nitely want to suppress the output): > RHSmatrix:= map(Simf, MatrixInverse(groupmatrix).map(d, groupmatrix) + MatrixInverse(groupmatrix).omegamatrix.groupmatrix): Now, hereās the desired substitution: > framechangesub:= [ Omega[0,0] = RHSmatrix[1,1], Omega[0,2] = RHSmatrix[1,3], Omega[1,0] = RHSmatrix[2,1], Omega[1,2] = RHSmatrix[2,3], Omega[2,0] = RHSmatrix[3,1], Omega[2,2] = RHSmatrix[3,3], Omega[3,0] = RHSmatrix[4,1], Omega[3,2] = RHSmatrix[4,3],
Omega[0,1] Omega[0,3] Omega[1,1] Omega[1,3] Omega[2,1] Omega[2,3] Omega[3,1] Omega[3,3]
= = = = = = = =
RHSmatrix[1,2], RHSmatrix[1,4], RHSmatrix[2,2], RHSmatrix[2,4], RHSmatrix[3,2], RHSmatrix[3,4], RHSmatrix[4,2], RHSmatrix[4,4]]:
We need to be careful when constructing the reverse substitution; the expressions on the right-hand sides in framechangesub involve exterior derivatives of the group parameters as well as the forms (ĀÆ ĻĪ²Ī± ), and this may prevent the makebacksub command from working as desired. Itās safer to do the following instead:
7.5. Maple computations
241
> framechangebacksub:= [op(solve({op(framechangesub)}, {seq(seq(omega[i,j], j=0..3), i=0..3)}))]: We now proceed much as we did in Exercise 6.20: First, introduce another substitution describing the adaptations for the transformed frame: > adaptedsub2:= [Omega[3,0]=0, Omega[3,1] = H[1,1]*Omega[1,0] + H[1,2]*Omega[2,0], Omega[3,2] = H[1,2]*Omega[1,0] + H[2,2]*Omega[2,0]]; Ė ij ) may be expressed in terms of the (hij ). (Note Next, determine how the (h that we need one extra substitution here that wasnāt needed in Exercise 6.20; if you wonder why itās there, see what happens when you leave it out!) > zero2:= Simf(subs(adaptedsub2, Omega[3,1]) - Simf(subs(adaptedsub2, Simf(subs(framechangebacksub, Simf(subs(adaptedsub1, Simf(subs(framechangesub, Omega[3,1]))))))))); > eqns:= {op(ScalarForm(zero2))}; > zero3:= Simf(subs(adaptedsub2, Omega[3,2]) - Simf(subs(adaptedsub2, Simf(subs(framechangebacksub, Simf(subs(adaptedsub1, Simf(subs(framechangesub, Omega[3,2]))))))))); > eqns:= eqns union {op(ScalarForm(zero3))}; > solve(eqns, {H[1,1], H[1,2], H[2,2]}); > assign(%); You can now check that the result is precisely the same as that in Exercise 6.20. Exercise 7.38: Now suppose that [Ī£] is an elliptic surface and that we have chosen a 1-adapted frame ļ¬eld, so that [hij ] is the identity matrix and B ā SO(2). Since we now wish to explore transformations among 1-adapted Ė ij ): frame ļ¬elds, assign these conditions for both (hij ) and (h > h[1,1]:= 1; h[1,2]:= 0; h[2,2]:= 1; H[1,1]:= 1; H[1,2]:= 0; H[2,2]:= 1; b[1,1]:= cos(theta); b[1,2]:= -sin(theta);
242
7. Curves and surfaces in projective space
b[2,1]:= sin(theta); b[2,2]:= cos(theta); Diļ¬erentiate the ļ¬rst equation in (7.26): > zero4:= Simf(subs(adaptedsub1, Simf(d(omega[3,1] - omega[1,0])))); Use the pick command to write this expression in the form Ļ1 ā§ Ļ ĀÆ 01 + Ļ2 ā§ Ļ ĀÆ 02 so that we can apply Cartanās lemma: pick(zero4, omega[1,0]); āĻ0,0 + 2 Ļ1,1 ā Ļ3,3 pick(zero4, omega[2,0]); Ļ1,2 + Ļ2,1 Similarly for the second equation in (7.26): > zero5:= Simf(subs(adaptedsub1, Simf(d(omega[3,2] - omega[2,0])))); > pick(zero5, omega[1,0]); Ļ1,2 + Ļ2,1 > pick(zero5, omega[2,0]); āĻ0,0 + 2 Ļ2,2 ā Ļ3,3 Equation (7.27) then follows from Cartanās lemma. Now add these conditions to adaptedsub1: > adaptedsub1:= [op(adaptedsub1), omega[1,1] = (1/2)*(omega[0,0] + omega[3,3] + h[1,1,1]*omega[1,0] + h[1,1,2]*omega[2,0]), omega[2,1] = -omega[1,2] + h[1,1,2]*omega[1,0] + h[1,2,2]*omega[2,0], omega[2,2] = (1/2)*(omega[0,0] + omega[3,3] + h[1,2,2]*omega[1,0] + h[2,2,2]*omega[2,0])]; Exercise 7.39: We may as well go ahead and determine how the (hijk ) transform under the full 1-adapted group action. We can compute the funcĖ ijk ) as follows: tions (h > newform1:= Simf(subs(adaptedsub2, Simf(subs(framechangebacksub, Simf(subs(adaptedsub1,
7.5. Maple computations
243
Simf(subs(framechangesub, 2*Omega[1,1] - Omega[0,0] - Omega[3,3])))))))); > H[1,1,1]:= pick(newform1, Omega[1,0]); H[1,1,2]:= pick(newform1, Omega[2,0]); > newform2:= Simf(subs(adaptedsub2, Simf(subs(framechangebacksub, Simf(subs(adaptedsub1, Simf(subs(framechangesub, 2*Omega[2,2] - Omega[0,0] - Omega[3,3])))))))); > H[1,2,2]:= pick(newform2, Omega[1,0]); H[2,2,2]:= pick(newform2, Omega[2,0]); The resulting expressions look fairly complicated, but they simplify considerably if we take Īø = 0, Ī» = 1: > Simf(subs([theta=0, lambda=1], H[1,1,1])); h1,1,1 ā 3 s1 + 3 r1 > Simf(subs([theta=0, lambda=1], H[1,1,2])); h1,1,2 ā s2 + r2 > Simf(subs([theta=0, lambda=1], H[1,2,2])); h1,2,2 ā s1 + r1 > Simf(subs([theta=0, lambda=1], H[2,2,2])); h2,2,2 ā 3 s2 + 3 r2 Exercise 7.41: Now we assume that both frame ļ¬elds are 2-adapted, so that the condition (7.28) holds for both sets of Maurer-Cartan forms. So, we tell Maple that it holds for the (hijk ) and then determine what conditions must Ė ijk ) as be satisļ¬ed by the group parameters in order for it to hold for the (h well: > h[1,2,2]:= -h[1,1,1]; h[1,1,2]:= -h[2,2,2]; > zero6:= Simf(H[1,1,1] + H[1,2,2]); zero7:= Simf(H[1,1,2] + H[2,2,2]); > solve({zero6, zero7}, {r[1], r[2]}); {r1 = sin(Īø)Ī» s2 + Ī» s1 cos(Īø), r2 = ā sin(Īø)Ī» s1 + Ī» cos(Īø)s2 } > assign(%);
244
7. Curves and surfaces in projective space
Now compute how the functions h111 , h222 transform under this restricted group action: > collect(Simf(H[1,1,1]), {h[1,1,1], h[2,2,2]}); > collect(Simf(H[2,2,2]), {h[1,1,1], h[2,2,2]}); If the output looks too complicated to recognize immediately, we can check that this does, in fact, yield the result in equation (7.30): > Simf(Vector([H[1,1,1], H[2,2,2]]) - (1/lambda)*B.B.B.Vector([h[1,1,1], h[2,2,2]])); 0 0 The remainder of the
adaptation process divides into cases based on whether or not the vector t h111 h222 vanishes. For Maple purposes, it becomes prudent to use substitutions rather than assignments from this point on, so that whatever information we learn about each case can be restricted to that case rather than applied globally. In the non-umbilic case, we now restrict to 3-adapted frame ļ¬elds, so we assume that h111 = 2, h222 = 0. In order to keep computations as simple as possible, we will restrict to those transformations in Exercise 7.42 for which Ī» = 1 and B is the identity matrix. We can collect this information into the following substitution: > threeadaptedsub:= [h[1,1,1] = 2, h[2,2,2] = 0, theta = 0, lambda = 1]; Exercise 7.45: At this point, we have the following relation: > Simf(subs(adaptedsub1, omega[1,1] + omega[2,2])); Ļ0,0 + Ļ3,3 Therefore, since Ļ ĀÆ 00 + Ļ ĀÆ 11 + Ļ ĀÆ 22 + Ļ ĀÆ 33 = 0, we must have Ļ ĀÆ 00 + Ļ ĀÆ 33 = Ļ ĀÆ 11 + Ļ ĀÆ 22 = 0. We need to add these relations to adaptedsub1 (and give it a new name since weāre working with a subcase) via the same two-step process that we used in Chapter 6. Since the substitution already contains expressions for Ļ ĀÆ 11 and Ļ ĀÆ 22 , we only need to add the relation Ļ ĀÆ 00 + Ļ ĀÆ 33 = 0: > adaptedsub13:= Simf(subs([omega[3,3] = -omega[0,0]], Simf(adaptedsub1))); > adaptedsub13:= [op(adaptedsub13), omega[3,3] = -omega[0,0]];
7.5. Maple computations
245
Now diļ¬erentiate this relation: > zero8:= Simf(subs(adaptedsub13, Simf(d(omega[0,0] + omega[3,3])))); Identify the appropriate 1-forms so that we can apply Cartanās lemma: > pick(zero8, omega[1,0]); āĻ0,1 + Ļ1,3 > pick(zero8, omega[2,0]); āĻ0,2 + Ļ2,3 It follows from Cartanās lemma that equations (7.33) must hold, and so we add them to our substitution: > adaptedsub13:= [op(adaptedsub13), omega[1,3] = omega[0,1] + ell[1,1]*omega[1,0] + ell[1,2]*omega[2,0], omega[2,3] = omega[0,2] + ell[1,2]*omega[1,0] + ell[2,2]*omega[2,0]]; Exercise 7.46: Now we compute how the (ij ) transform under the transformation (7.32) between 3-adapted frame ļ¬elds. > newform4:= Simf(subs(threeadaptedsub, Simf(subs(adaptedsub2, Simf(subs(framechangebacksub, Simf(subs(adaptedsub13, Simf(subs(framechangesub, Omega[1,3] - Omega[0,1])))))))))); > newform5:= Simf(subs(threeadaptedsub, Simf(subs(adaptedsub2, Simf(subs(framechangebacksub, Simf(subs(adaptedsub13, Simf(subs(framechangesub, Omega[2,3] - Omega[0,2])))))))))); > LL[1,1]:= pick(newform4, Omega[1,0]); LL[1,2]:= pick(newform4, Omega[2,0]); LL[2,2]:= pick(newform5, Omega[2,0]); In order to show that we can always ļ¬nd a 3-adapted frame for which the (ij ) are all equal to zero, it suļ¬ces to show that we can solve the equations Ė11 = Ė12 = Ė22 = 0 for s0 , s1 , s2 : > solve({LL[1,1], LL[1,2], LL[2,2]}, {s[0], s[1], s[2]}); 1 1 1 2 1 2 1 1 2 {s0 = ā ell1,1 ā ell2,2 + ell1,2 + ell2,2 ā ell2,2 ell1,1 + ell1,1 , 4 4 8 32 16 32 1 1 1 s1 = ell2,2 ā ell1,1 , s2 = ell1,2 } 4 4 2
246
7. Curves and surfaces in projective space
Exercise 7.47: For a 4-adapted frame ļ¬eld, we have the following conditions: > fouradaptedsub:= [h[1,1,1] = 2, h[2,2,2] = 0, ell[1,1] = 0, ell[1,2] = 0, ell[2,2] = 0]; We can reļ¬ne our 3-adapted frame adaptation by incorporating these conditions as follows: > adaptedsub14:= Simf(subs(fouradaptedsub, adaptedsub13)); We can now read oļ¬ the relations (7.34) directly from adaptedsub14. At this point, all the remaining Maurer-Cartan forms must be linear combinations of (ĀÆ Ļ01 , Ļ ĀÆ 02 ). It follows from equations (7.34) that the ļ¬rst four equations in (7.35) must hold for some functions (pi , qij ). Before we continue, add these expressions to adaptedsub14 in two steps: > adaptedsub14:= Simf(subs([ omega[1,2] = p[1]*omega[1,0] + (p[2] - 1)*omega[2,0], omega[0,1] = q[1,1]*omega[1,0] + q[1,2]*omega[2,0], omega[0,2] = q[2,1]*omega[1,0] + q[2,2]*omega[2,0]], adaptedsub14)); > adaptedsub14:= [op(adaptedsub14), omega[1,2] = p[1]*omega[1,0] + (p[2] - 1)*omega[2,0], omega[0,1] = q[1,1]*omega[1,0] + q[1,2]*omega[2,0], omega[0,2] = q[2,1]*omega[1,0] + q[2,2]*omega[2,0]]; Now diļ¬erentiate the ļ¬rst two equations in (7.34): > Simf(subs(adaptedsub14, Simf(d(omega[1,1] - omega[1,0])))); > pick(%, omega[1,0]); āĻ0,0 ā 3 p1 Ļ2,0 By Cartanās lemma, (ĀÆ Ļ00 + 3p1 Ļ ĀÆ 02 ) must be equal to a multiple of Ļ ĀÆ 01 . > Simf(subs(adaptedsub14, Simf(d(omega[1,2] + omega[2,1] + 2*omega[2,0])))); > pick(%, omega[2,0]); 2 Ļ0,0 ā 6 p2 Ļ1,0 By Cartanās lemma, (2ĀÆ Ļ00 ā 6p2 Ļ ĀÆ 01 ) must be equal to a multiple of Ļ ĀÆ 02 . Together, these conditions imply that Ļ ĀÆ 00 = 3p2 Ļ ĀÆ 01 ā 3p1 Ļ ĀÆ 02 .
7.5. Maple computations
247
Add this condition to adaptedsub14: > adaptedsub14:= Simf(subs([ omega[0,0] = 3*p[2]*omega[1,0] - 3*p[1]*omega[2,0]], adaptedsub14)); > adaptedsub14:= [op(adaptedsub14), omega[0,0] = 3*p[2]*omega[1,0] - 3*p[1]*omega[2,0]]; Finally, diļ¬erentiating the last two equations in (7.34) and applying Cartanās lemma in this same fashion yields Ļ ĀÆ 30 = (q11 ā q22 )ĀÆ Ļ01 ā (q12 + q21 )ĀÆ Ļ02 , as desired. Add this last condition to adaptedsub14: > adaptedsub14:= [op(adaptedsub14), omega[0,3] = (q[1,1] - q[2,2])*omega[1,0] - (q[1,2] + q[2,1])*omega[2,0]]; The last step in the process is to check the remaining structure equations for any additional relations among the functions (pi , qij ). We can check them all at once with the following sequence of commands, which will produce any and all such relations: > for i from 0 to 3 do for j from 0 to 3 do print(Simf(subs(adaptedsub14, Simf(d(omega[i,j]) - d(Simf(subs(adaptedsub14, omega[i,j]))))))); end do; end do; The resulting equations all involve the exterior derivatives of the functions (pi , qij ); they are equivalent to a system of PDEs that these functions must satisfy. This system is analogous to the Gauss-Codazzi system for surfaces in E3 ; it determines the compatibility conditions for the invariants of elliptic projective surfaces without umbilic points. The totally umbilic case is treated in the the Maple worksheet for this chapter on the AMS webpage.
Part 3
Applications of moving frames
10.1090/gsm/178/08
Chapter 8
Minimal surfaces in E3 and A3
8.1. Introduction The study of minimal surfaces goes back to 1760, when Lagrange posed the following problem: Given a closed curve in E3 , can we ļ¬nd a surface of minimum area among all surfaces that have the given curve as their boundary? This question is called the Plateau problem; it is named after the physicist Joseph Plateau, who in the nineteenth century performed experiments with wire frames and soap ļ¬lms in order to study the properties of such surfaces. Despite the long history, rigorous existence results for area-minimizing surfaces with general boundary curves were only proved beginning in the 1930s [Dou31].
8.2. Minimal surfaces in E3 A regular surface in E3 is called minimal if it is ālocally area minimizingā. More precisely, Ī£ ā E3 is minimal if for any suļ¬ciently small open set V ā Ī£, the closure V has the minimum area of all surfaces in E3 with the same boundary as V . Classical examples of minimal surfaces include the plane, the catenoid, and the helicoid. 8.2.1. Minimal surfaces and the calculus of variations. In order to study properties of minimal surfaces, we need to introduce some ideas from the calculus of variations. (For a more comprehensive introduction to the 251
8. Minimal surfaces in E3 and A3
252
calculus of variations, see, e.g., [Dac08] or [Olv00].) Intuitively, we consider the set S of closed and bounded surfaces Ī£ ā E3 to be an inļ¬nite-dimensional space. (Making this notion precise would require that we deļ¬ne a topology on this space, but we wonāt need this level of technicality.) We then deļ¬ne the area functional A on this space to be & A(Ī£) = dA; Ī£
i.e., A(Ī£) is simply the area of Ī£. The key idea is that minimal surfaces should be critical points of this functional. Unfortunately, since the space S is inļ¬nite-dimensional (with a rather complicated topology!), it isnāt entirely obvious how to go about ļ¬nding critical points of the functional A. The solution to this problem lies in the idea of a variation. We start with the observation that if f :M āR is a smooth function on a ļ¬nite-dimensional manifold M and q0 ā M is a critical point of f , then for any smooth curve Ī± : I ā M with Ī±(0) = q0 , we must have d (f ā¦ Ī±) = 0. dt t=0
Conversely, if q0 ā M is not a critical point of f , then there exists a smooth d curve Ī± : I ā M with Ī±(0) = q0 and dt t=0 (f ā¦ Ī±) = 0. If we could come up with a good deļ¬nition for a āsmooth curveā in the space S, then we could apply this same idea: In order for a surface Ī£ to be a critical point of A, it should have the property that for any āsmooth curveā Ī± : I ā S with Ī±(0) = Ī£, d (8.1) (A ā¦ Ī±) = 0. dt t=0 Conversely, if Ī£ is not a critical point of A, then there should exist a āsmooth d curveā Ī± : I ā S with Ī±(0) = Ī£ and dt (A ā¦ Ī±) = 0. t=0 Remark 8.1. It is customary to denote the surface Ī±(t) ā S by Ī£t , so that Ī£0 = Ī£ and the condition (8.1) becomes d A(Ī£t ) = 0. dt t=0 The intuitive idea of a āsmooth curveā in S can be deļ¬ned rigorously as follows.
8.2. Minimal surfaces in E3
253
Deļ¬nition 8.2. Let U ā R2 be an open set, and let x : U ā E3 be a smooth immersion whose image is a surface Ī£ = x(U ). A variation of x is a smooth map X : U Ć (āĪµ, Īµ) ā E3 for some Īµ > 0, with the properties that: (1) For each t ā (āĪµ, Īµ), the set Ī£t = X(U, t) is a regular surface in E3 . (2) For each u ā U , X(u, 0) = x(u). (In particular, Ī£0 = Ī£.) So intuitively, a variation deļ¬nes a 1-parameter family of surfaces Ī£t ā E3 that āvary smoothlyā with t. Deļ¬nition 8.3. A variation X of an immersion x : U ā E3 is called (1) compactly supported if there exists a compact set D ā U such that for all t ā (āĪµ, Īµ) and all u ā U \ D, X(u, t) = X(u, 0) = x(u); (2) normal if the vector āX āt (u, t) is parallel to the unit normal vector to the surface Ī£t for all t ā (āĪµ, Īµ) and all u ā U . We will only consider compactly supported, normal variations. For surfaces with a given boundary, the restriction to compactly supported variations corresponds to considering only variations that leave the boundary ļ¬xed, which is a natural assumption in the context of the Plateau problem. (Moreover, it guarantees that the integrals that arise in the computations below remain ļ¬nite.) The restriction to normal variations will simplify the analysis of minimal surfaces, and the following exercise shows that any compactly supported variation can be made normal via a reparametrization; hence, this assumption is not really a signiļ¬cant restriction. *Exercise 8.4. Let x : U ā E3 be an immersion, and let X : U Ć (āĪµ, Īµ) ā E3 be a compactly supported variation of x. For each t ā (āĪµ, Īµ), the function xt : U ā E3 deļ¬ned by xt (u) = X(u, t) deļ¬nes a parametrization of the surface Ī£t . ĀÆ : U Ć (āĪµ, Īµ) ā E3 of X of the form Consider a reparametrization X ĀÆ u, t) = X(u(ĀÆ X(ĀÆ u, t), t), ĀÆ for all u ĀÆ ā U . If we write u = (u1 , u2 ), then we can write with u(ĀÆ u, 0) = u this reparametrization as 1 2 ĀÆ u X ĀÆ ,u ĀÆ , t) = X(u1 (ĀÆ u1 , u ĀÆ2 , t), u2 (ĀÆ u1 , u ĀÆ2 , t), t .
8. Minimal surfaces in E3 and A3
254
(a) Show that for each t ā (āĪµ, Īµ), the function ĀÆ u, t) ĀÆ t (ĀÆ x u) = X(ĀÆ ĀÆ t is the same as is a reparametrization of the surface Ī£t ; i.e., the image of x ĀÆ 0 = x0 = x. the image of xt . Moreover, x (b) Show that ĀÆ āX āu2 āX āX āu1 āX + + = . āt āt āu1 āt āu2 āt (c) Let nt (u) denote the unit normal vector to the surface Ī£t at the point xt (u). Then we can decompose āX āt as āX āX āX = a(u, t) 1 + b(u, t) 2 + c(u, t)nt āt āu āu for some functions a(u, t), b(u, t), c(u, t). Show that the condition that is parallel to nt (u) is equivalent to the system of diļ¬erential equations (8.2)
āu1 = āa(u1 , u2 , t), āt
ĀÆ āX āt
āu2 = āb(u1 , u2 , t) āt
for the functions u1 (ĀÆ u, t), u2 (ĀÆ u, t). (These equations should be regarded as ĀÆ= ordinary diļ¬erential equations, with t as the independent variable and u ĀÆ is equivalent (ĀÆ u1 , u ĀÆ2 ) as parameters.) Moreover, the condition u(ĀÆ u, 0) = u to the initial conditions (8.3)
u1 (ĀÆ u1 , u ĀÆ2 , 0) = u ĀÆ1 ,
u2 (ĀÆ u1 , u ĀÆ2 , 0) = u ĀÆ2 .
(d) Conclude from the existence/uniqueness theorem for ordinary diļ¬erential equations that for each choice of (ĀÆ u1 , u ĀÆ2 ), the system (8.2), (8.3) has a unique solution for t in some interval (āĪµ, Īµ). Therefore, there exists a ĀÆ of X (possibly deļ¬ned for a smaller value of Īµ than reparametrization X ĀÆ is the original variation) with the property that the variation deļ¬ned by X normal. Remark 8.5. You might worry that, because the value of Īµ in the previous ĀÆ , it could happen that there is no single Īµ > 0 that exercise depends on u ĀÆ ā U . Fortunately, the hypothesis that the variation is comsuļ¬ces for all u pactly supported saves the day: Any positive-valued function on a compact set must have a positive lower bound; therefore there exists a real number ĀÆ is well-deļ¬ned on U Ć (āĪµ, Īµ). Īµ > 0 such that the desired variation X
8.2. Minimal surfaces in E3
255
Now we can give a rigorous deļ¬nition that corresponds to our intuitive idea of a critical point for the area functional A on S: Deļ¬nition 8.6. Let U ā R2 be an open set, and let x : U ā E3 be an immersion whose image is a surface Ī£ = x(U ). Ī£ is called a minimal surface if for every compactly supported variation X : U Ć (āĪµ, Īµ) of x, we have d (8.4) A(Ī£t ) = 0. dt t=0 Remark 8.7. Since the area functional is invariant under reparametrizations, minimality is a property of the surface Ī£, independent of the choice of parametrization x : U ā E3 of Ī£. Moreover, in order to show that a given immersion x is minimal, it suļ¬ces to consider normal variations of x. Remark 8.8. Despite the name, minimal surfaces are not necessarily global minimizers for the area functional; for instance, there may be multiple minimal surfaces with diļ¬erent areas spanning a given boundary curve or curves. However, it can be shown that any minimal surface Ī£ is locally area-minimizing in the sense that, given any point q ā Ī£, there exists a neighborhood V ā Ī£ of q that is area-minimizing among all surfaces with the same boundary curve as V . Now, it still isnāt entirely obvious how to use this deļ¬nition to ļ¬nd minimal surfaces. In order to ļ¬nd critical points q0 for a function f : M ā R on a ļ¬nite-dimensional manifold M , it suļ¬ces to identify those points q0 ā M where all the partial derivatives of f vanish; however, for the functional A on the inļ¬nite-dimensional space S, there is no ļ¬nite set of compactly supported variations that plays a role analogous to the partial derivatives of f . Instead, we will need to ļ¬nd a way to work directly with Deļ¬nition 8.6. Let x : U ā E3 be an immersion whose image is a surface Ī£ ā E3 . We begin by choosing an adapted orthonormal frame ļ¬eld (e1 (u), e2 (u), e3 (u)) along Ī£, so that (e1 (u), e2 (u)) span the tangent plane Tx(u) Ī£ and e3 (u) is orthogonal to Tx(u) Ī£. Recall that choosing an orthonormal frame ļ¬eld along Ė : U ā E(3) deļ¬ned by Ī£ is equivalent to choosing a lifting x Ė (u) = (x(u); e1 (u), e2 (u), e3 (u)) . x The pullbacks (ĀÆ Ļi, Ļ ĀÆ ji ) to U of the Maurer-Cartan forms (Ļ i , Ļji ) on E(3) via Ė satisfy the conditions that Ļ the lifting x ĀÆ 3 = 0 and 3 1 Ļ ĀÆ1 h11 h12 Ļ ĀÆ = Ļ ĀÆ 23 h12 h22 Ļ ĀÆ2
8. Minimal surfaces in E3 and A3
256
for some functions h11 , h12 , h22 on U , and the Gauss and mean curvature functions of Ī£ (cf. Deļ¬nition 4.45) are given by K = h11 h22 ā h212 ,
H = 12 (h11 + h22 ).
Now, consider a compactly supported, normal variation X : U Ć (āĪµ, Īµ) ā E3 d of x. In order to compute dt A(Ī£t ), we will choose an orthonormal frame t=0 8 : U Ć (āĪµ, Īµ) ā E(3)āand consider ļ¬eld on the variation X, i.e., a lifting X i i the pullbacks (ĀÆ Ļ ,Ļ ĀÆ j ) of the Maurer-Cartan forms on E(3) to U Ć (āĪµ, Īµ) 8 via X. We can deļ¬ne such a frame ļ¬eld as follows: For each (u, t) ā U Ć (āĪµ, Īµ), let (e1 (u, t), e2 (u, t), e3 (u, t)) be an orthonormal frame for the surface Ī£t at the point xt (u), with e3 (u, t) normal to the tangent plane Txt (u) Ī£t . Then 8 must satisfy the equation the pullbacks (ĀÆ Ļ i ) of the (Ļ i ) to U via X dX = ei Ļ ĀÆ i. *Exercise 8.9. (a) For each t ā (āĪµ, Īµ), let ıt : U ā U Ć (āĪµ, Īµ) denote the inclusion map deļ¬ned by ıt (u) = (u, t). Show that the pullbacks of the 1-forms Ļ ĀÆ 1 = dX, e1 , Ļ ĀÆ 2 = dX, e2 to U via ıt are the usual dual forms on the surface Ī£t = xt (U ). (Hint: This isnāt as complicated as it sounds. It is an immediate consequence of the fact that the immersion xt can be written as the composition xt = X ā¦ ıt and the fact that pullbacks behave well with respect to compositionāspeciļ¬cally, (xt )ā = (ıt )ā ā¦ (X)ā .) (b) Show that
āX dt. Ļ ĀÆ = dX, e3 = Ā± āt (The sign is dependent on the choice of orientation for Ī£, as determined by the choice of the vector ļ¬eld e3 (u) on Ī£.) 3
In particular, note that Ļ ĀÆ 3 is no longer necessarily equal to zero as a 1-form on the variation X; instead, it is a multiple of dt. Moreover, the 1-forms (ĀÆ Ļ1, Ļ ĀÆ 2 , dt) are linearly independent and form a basis for the 1-forms on the 3-dimensional manifold U Ć (āĪµ, Īµ).
8.2. Minimal surfaces in E3
257
For convenience, let Ļ(u, t) = Ā± āX āt (u, t) , with the sign chosen so that 3 Ļ ĀÆ = Ļ(u, t) dt. *Exercise 8.10. (a) Diļ¬erentiate the equation Ļ ĀÆ 3 = Ļ dt to obtain Ļ ĀÆ 13 ā§ Ļ ĀÆ1 + Ļ ĀÆ 23 ā§ Ļ ĀÆ 2 + dĻ ā§ dt = 0. (b) Apply Cartanās lemma to conclude that ā” 3ā¤ ā” ā¤ ā” 1ā¤ Ļ ĀÆ1 Ļ ĀÆ h11 h12 Ļ1 ā¢ 3ā„ ā¢ ā„ ā¢ 2ā„ (8.5) ĀÆ 2 ā¦ = ā£h12 h22 Ļ2 ā¦ ā£Ļ ĀÆ ā¦ ā£Ļ dĻ Ļ1 Ļ2 Ļ3 dt for some functions h11 , h12 , h22 , Ļ1 , Ļ2 , Ļ3 on U Ć (āĪµ, Īµ). (c) Show that the functions (hij ) are precisely the coeļ¬cients of the second fundamental form of the surface Ī£t , while the functions (Ļi ) are the directional derivatives of Ļ in the directions of the vectors ei (u, t). Ė : U ā E(3) on Now, recall that for any adapted orthonormal frame ļ¬eld x the surface Ī£ = x(U ), the area of Ī£ is given by & & 1 2 Ė ā (Ļ 1 ā§ Ļ 2 ) A(Ī£) = x Ļ ĀÆ ā§Ļ ĀÆ = U
U
(cf. Exercise 4.47). Applying this formula to each surface Ī£t in the variation yields & Ė āt (Ļ 1 ā§ Ļ 2 ). x
A(Ī£t ) = U
The surface Ī£ = Ī£0 is minimal if and only if for every compactly supported normal variation, we have d 0 = A(Ī£t ) dt t=0 & d Ė ā (Ļ 1 ā§ Ļ 2 ) = x (8.6) dt t=0 U t & d ā 1 2 Ė = x (Ļ ā§ Ļ ) . t U dt t=0 *Exercise 8.11. Show that Ė āt (Ļ 1 ā§ Ļ 2 ) = ıāt (ĀÆ x Ļ1 ā§ Ļ ĀÆ 2 ), 8 (cf. Exercise where (ĀÆ Ļ1, Ļ ĀÆ 2 ) are the pullbacks of (Ļ 1 , Ļ 2 ) to U Ć(āĪµ, Īµ) via X 8.9). Therefore, the condition (8.6) can be written as & d ā 1 (8.7) ıt (ĀÆ Ļ ā§Ļ ĀÆ 2 ) = 0. dt U
t=0
8. Minimal surfaces in E3 and A3
258
In order to evaluate this integral, recall from Ā§2.11 that the Lie derivative of a p-form Ī¦ along a vector ļ¬eld v is the p-form Ļāt Ī¦ ā Ī¦ , tā0 t where Ļt is the ļ¬ow of the vector ļ¬eld v. In other words, d Lv Ī¦ = Ļāt Ī¦. dt Lv Ī¦ = lim
t=0
*Exercise 8.12. (a) Show that the inclusion map ıt : U ā U Ć (āĪµ, Īµ) may ā be regarded as the ļ¬ow of the vector ļ¬eld āt on U Ć (āĪµ, Īµ), restricted to the set U Ć {0} ā U Ć (āĪµ, Īµ). (b) Use part (a) to show that d ā 1 ıt (ĀÆ Ļ ā§Ļ ĀÆ 2 ) = Lā/āt (ĀÆ Ļ1 ā§ Ļ ĀÆ 2 ). dt t=0 (See Ā§2.11 for the relevant deļ¬nitions.) (c) Use equation (8.7) to conclude that if the immersion x : U ā E3 is minimal, then for any compactly supported, normal variation of x, we must have & (8.8) Lā/āt (ĀÆ Ļ1 ā§ Ļ ĀÆ 2 ) = 0. U
Now weāre ready to put all the pieces together! *Exercise 8.13. Let x : U ā E3 be an immersion, and let X : U Ć(āĪµ, Īµ) ā E3 be a compactly supported, normal variation of x. (a) Use Cartanās formula for the Lie derivative (cf. Theorem 2.55) to show that ā Lā/āt (ĀÆ d(ĀÆ Ļ1 ā§ Ļ Ļ1 ā§ Ļ ĀÆ 2) = ĀÆ 2) . āt (Hint: Since X is a normal variation of x, the vector " # āX ā = Xā āt āt is a multiple of the frame vector e3 (u, t). Therefore, " " # # āX āX Ļ1 = Ļ2 = 0, āt āt and pulling these equations back via Xā yields " # " # ā ā 1 2 Ļ ĀÆ =Ļ ĀÆ = 0. āt āt
8.2. Minimal surfaces in E3
259
(b) Use the Cartan structure equations, the equation Ļ ĀÆ 3 = Ļ dt, and equation (8.5) to show that d(ĀÆ Ļ1 ā§ Ļ ĀÆ 2 ) = ā(h11 + h22 )Ļ Ļ ĀÆ1 ā§ Ļ ĀÆ 2 ā§ dt = ā2HĻ Ļ ĀÆ1 ā§ Ļ ĀÆ 2 ā§ dt. (c) Use parts (a) and (b) to show that & & Lā/āt (ĀÆ Ļ1 ā§ Ļ ĀÆ 2) = ā2HĻ Ļ ĀÆ1 ā§ Ļ ĀÆ 2. U
U
(d) Conclude that if the mean curvature H of Ī£ is identically equal to zero, then & Lā/āt (ĀÆ Ļ1 ā§ Ļ ĀÆ 2) = 0 U
for every compactly supported, normal variation of x, and hence Ī£ is minimal. Thus, we have proved the following proposition: Proposition 8.14. If a regular surface Ī£ ā E3 has mean curvature H identically equal to zero, then Ī£ is minimal. In the following exercise, we will prove the converse of Propsition 8.14: If Ī£ = x(U ) is a surface whose mean curvature is not identically zero, then Ī£ is not a critical point for the area functional A, and hence Ī£ is not minimal. *Exercise 8.15. Let Ī£ = x(U ) be a surface whose mean curvature is not identically zero. (a) Let u0 ā U be a point where H(u0 ) = 0. Show that there exists a ĀÆ is contained in U and such neighborhood U ā U of u0 whose closure U that H is nonzero and does not change sign on U . ĀÆ ; i.e., (b) Let X : U Ć (āĪµ, Īµ) be a normal variation of x supported on U X(u, t) = x(u) ĀÆ , U
for all u ā U \ chosen so that Ļ ĀÆ 3 = Ļ dt for some function Ļ that is nonzero in a neighborhood of {u0 } Ć (āĪµ, Īµ) and has the same sign as H wherever it is nonzero. Show that for this variation, d A(Ī£t ) < 0. dt t=0
Conclude that Ī£ = Ī£0 is not a critical point for A. Together with Proposition 8.14, this proves the following theorem: Theorem 8.16. A regular surface Ī£ ā E3 is minimal if and only if its mean curvature H is identically equal to zero.
8. Minimal surfaces in E3 and A3
260
Remark 8.17. Theorem 8.16 is often taken as a deļ¬nition for minimal surfaces because it is much easier to work with than Deļ¬nition 8.6. In the following two exercises, we explore two classical minimal surfaces. Exercise 8.18. The catenoid is the surface Ī£ ā E3 obtained by rotating the curve x = cosh(z) about the z-axis. It can be parametrized by x(u, v) = t[cos(u) cosh(v), sin(u) cosh(v), v]. (a) Show that the frame ļ¬eld e1 (u, v) =
xu = t[ā sin(u), cos(u), 0], |xu |
e2 (u, v) =
xv 1 t [cos(u) sinh(v), sin(u) sinh(v), 1], = |xv | cosh(v)
e3 (u, v) = e1 (u, v) Ć e2 (u, v) =
1 t [cos(u), sin(u), ā sinh(v)] cosh(v)
is orthonormal and that (e1 (u, v), e2 (u, v)) span the tangent space to Ī£ at each point x(u, v) ā Ī£. (b) Show that the dual forms of this frame ļ¬eld are Ļ ĀÆ 1 = cosh(v) du,
Ļ ĀÆ 2 = cosh(v) dv.
(c) Compute de3 , and show that Ļ ĀÆ 13 = ā
1 du, cosh(v)
Ļ ĀÆ 23 =
1 dv. cosh(v)
(d) Use the results of parts (b) and (c) to compute the matrix [hij ], and show that the mean curvature of Ī£ is H = 0. Therefore, the catenoid is a minimal surface. (e) (Maple recommended) Repeat the computations of parts (a)ā(d) for an arbitrary non-planar surface of revolution, parametrized by x(u, v) = t[Ļ(v) cos(u), Ļ(v) sin(u), v]. Show that the surface is minimal if and only if the function Ļ(v) satisļ¬es the diļ¬erential equation (8.9)
ĻĻ = (Ļ )2 + 1.
8.2. Minimal surfaces in E3
261
(f) Show that the only solutions of equation (8.9) are 1 cosh(av + b), a where a, b are constants. Conclude that catenoids are the only non-planar minimal surfaces of revolution. Ļ(v) =
Exercise 8.19. The helicoid is the ruled surface Ī£ ā E3 parametrized by x(u, v) = t[v cos(u), v sin(u), u]. (a) Show that the frame ļ¬eld e1 (u, v) =
xu 1 t [āv sin(u), v cos(u), 1], =ā |xu | v2 + 1
e2 (u, v) =
xv = t[cos(u), sin(u), 0], |xv |
1 t e3 (u, v) = e1 (u, v) Ć e2 (u, v) = ā [ā sin(u), cos(u), āv] 2 v +1 is orthonormal and that (e1 (u, v), e2 (u, v)) span the tangent space to Ī£ at each point x(u, v) ā Ī£. (b) Show that the dual forms of this frame ļ¬eld are
Ļ ĀÆ 1 = v 2 + 1 du, Ļ ĀÆ 2 = dv. (c) Compute de3 , and show that Ļ ĀÆ 13 =
v2
1 dv, +1
1 Ļ ĀÆ 23 = ā du. 2 v +1
(d) Use the results of parts (b) and (c) to compute the matrix [hij ], and show that the mean curvature of Ī£ is H = 0. Therefore, the helicoid is a minimal surface. 8.2.2. The Weierstrass-Enneper representation for minimal surfaces. There is a beautiful connection between minimal surfaces in E3 and the theory of holomorphic functions of a complex variable, which was ļ¬rst described by Weierstrass and Enneper in the late nineteenth century [Wei66]. In this section, we will see how moving frames can be used to explore this relationship. Let Ī£ ā E3 be a regular surface with parametrization x : U ā E3 , and let (e1 (u), e2 (u), e3 (u)) be an orthonormal frame ļ¬eld on Ī£, with e3 (u) normal to the tangent plane Tx(u) Ī£ at each point x(u) ā Ī£. As we saw in the
8. Minimal surfaces in E3 and A3
262
previous subsection, the associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ) have the property that 3 1 Ļ ĀÆ1 h11 h12 Ļ ĀÆ = Ļ ĀÆ 23 h12 h22 Ļ ĀÆ2 for some functions (hij ) on U , and Ī£ is minimal if and only if h11 + h22 = 0. Consider the complex, vector-valued 1-form Ī¾ = (e1 ā ie2 )(ĀÆ Ļ 1 + iĀÆ Ļ2) = (e1 Ļ ĀÆ 1 + e2 Ļ ĀÆ 2 ) + i(e1 Ļ ĀÆ 2 ā e2 Ļ ĀÆ 1) on U , where i =
ā
ā1.
Remark 8.20. In keeping with the notation used throughout this book, we will continue to use (ĀÆ Ļi, Ļ ĀÆ ji ) to denote the (real-valued!) Maurer-Cartan forms associated to the orthonormal frame ļ¬eld (e1 (u), e2 (u), e3 (u)). In order to avoid confusion, we will use the notation z ā rather than zĀÆ to denote complex conjugation. *Exercise 8.21. (a) Show that Ī¾ is a well-deļ¬ned 1-form on U , independent of the choice of orthonormal frame ļ¬eld (e1 (u), e2 (u), e3 (u)). (Hint: This is the same sort of computation that you used to show that the ļ¬rst fundamental form was well-deļ¬ned in Exercise 4.24.) (b) Show that dĪ¾ = i(h11 + h22 ) e3 Ļ ĀÆ1 ā§ Ļ ĀÆ 2. Therefore, dĪ¾ = 0 if and only if Ī£ is minimal. Now suppose that Ī£ is minimal. Let (e1 (u), e2 (u), e3 (u)) be any adapted orthonormal frame ļ¬eld along Ī£, with associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ), 1 2 and consider the complex, scalar-valued 1-form Ļ ĀÆ + iĀÆ Ļ on U . *Exercise 8.22. (a) Use the Frobenius theorem (cf. Theorem 2.33) to show that every point u ā U has a neighborhood V ā U on which there exist complex-valued functions z, Ļ : V ā C such that (8.10)
Ļ ĀÆ 1 + iĀÆ Ļ 2 = Ļ dz.
(Hint: Show that the hypothesis of Theorem 2.33 is automatically satisļ¬ed for any 1-form on a 2-dimensional surface. Moreover, this theorem holds for complex-valued 1-forms Ī¦, in which case the functions f, g of Theorem 2.33 are complex-valued as well.)
8.2. Minimal surfaces in E3
263
(b) Show that the function Ļ must be nonzero at every point of V . (Hint: The 2-form (ĀÆ Ļ 1 + iĀÆ Ļ 2 ) ā§ (ĀÆ Ļ 1 + iĀÆ Ļ 2 )ā = (ĀÆ Ļ 1 + iĀÆ Ļ 2 ) ā§ (ĀÆ Ļ 1 ā iĀÆ Ļ 2 ) = ā2iĀÆ Ļ1 ā§ Ļ ĀÆ2 is nonvanishing on Ī£.) (c) Write the function z as z = u + iv, where u, v : V ā R are real-valued functions on V , and let z ā = u ā iv denote the complex conjugate of z. Use your calculation from part (b) to show that dz ā§ dz ā = ā2i du ā§ dv = 0 at every point of V ; therefore, the functions (u, v) can be used as a system of local (real) coordinates on V . (In fact, by a reparametrization of V , we can safely assume that u = u1 , v = u2 .) The next exercise introduces some basic results from complex analysis that will be needed for the remainder of this section. *Exercise 8.23. Let V ā R2 be a simply connected, open set with local coordinates (u, v). If we introduce the complex coordinate z = u + iv on V , then z and its complex conjugate z ā = u ā iv form a complex-valued local coordinate system (z, z ā ) on V . (a) Show that we can write the original coordinates (u, v) as 1 u = (z + z ā ), 2
v=
1 (z ā z ā ). 2i
Now, let w : V ā C be a complex-valued, diļ¬erentiable function on V . The function w may be considered as a function of either the coordinates (u, v) or the coordinates (z, z ā ). The function w is called holomorphic or complex analytic if, when considered as a function w(z, z ā ), it satisļ¬es the condition āw = 0, āz ā
(8.11) i.e., if w is a function of z alone.
(b) Show that equation (8.11) is equivalent to (8.12)
wu + iwv = 0.
8. Minimal surfaces in E3 and A3
264
(c) Write w(u, v) = x(u, v) + iy(u, v), where x, y : V ā R are real-valued, diļ¬erentiable functions on V . Show that equation (8.12) is equivalent to the pair of equations (8.13)
xu = yv ,
xv = āyu .
Equations (8.13) are called the Cauchy-Riemann equations. (d) Show that any diļ¬erentiable functions x, y : V ā R satisfying equations (8.13) must also satisfy the equations (8.14)
xuu + xvv = yuu + yvv = 0.
In other words, both x and y must be harmonic functions. Now we return to the complex-valued function z = u + iv of Exercise 8.22. Since this function satisļ¬es dz ā§ dz ā = 0, it can be used as a local complex coordinate on V . This choice of complex coordinate z allows us to write the restriction of the parametrization x to the subset V ā U in the form x(z), thereby deļ¬ning a complex structure on Ī£; in other words, it gives Ī£ the structure of a 1-dimensional complex manifold. Next, with Ļ as in equation (8.10), let f : V ā C3 be the complex, vectorvalued function f (z, z ā ) = (e1 (z, z ā ) ā ie2 (z, z ā ))Ļ(z, z ā ). Then we can write Ī¾ as Ī¾ = f (z, z ā ) dz. *Exercise 8.24. (a) Show that the assumption that Ī£ is minimal, and hence dĪ¾ = 0, is equivalent to āf = 0. āz ā Therefore, if Ī£ is minimal, then f (z, z ā ) = f (z) is a holomorphic function on V . For the remainder of this exercise, assume that this condition holds. (b) Show that f (z), f (z) = 0. (c) Apply the PoincarĀ“e lemma (cf. Theorem 2.31) to the closed 1-form Ī¾ = f (z) dz to conclude that there exists a vector-valued, holomorphic function
8.2. Minimal surfaces in E3
265
z : V ā C3 such that Ī¾ = dz, and therefore, f (z) =
z (z).
(d) Show that Re(dz) = Re(Ī¾) = e1 Ļ ĀÆ 1 + e2 Ļ ĀÆ 2 = dx, where x : U ā Ī£ is the original parametrization of Ī£. (Here Re denotes the āreal partā; i.e., Re(a + ib) = a.) Conclude that, up to a translation, we must have x(z) = Re (z(z)) . Therefore, we can write z(z) as z(z) = x(z) + iy(z) for some function y : V ā R3 . Moreover, from part (b), we have z (z), z (z) = 0. (Note, however, that the requirement that x be an immersion implies that the vector z (z) is never zero for any z ā V .) Exercise 8.24 shows that any minimal surface x : U ā E3 can locally be written as the real part of a holomorphic function z : U ā C3 with z (z), z (z) = 0. The next exercise shows that the converse is true as well. *Exercise 8.25. Let V ā C be an open set with local coordinates (u, v) and complex coordinate z = u + iv. Let z : V ā C3 be a complex vector-valued, holomorphic function with the properties that z (z) is never zero on V and z (z), z (z) = 0. Write z(u, v) = x(u, v) + iy(u, v), where x, y : V ā
R3
are real vector-valued functions on V .
(a) Use the Cauchy-Riemann equations (8.13) to show that dz = (xu ā ixv )(du + i dv) = (xu ā ixv ) dz. Therefore, z (z) = xu ā ixv . (b) Use the condition z (z), z (z) = 0 to show that the parametrization x : V ā R3 is conformal; i.e., xu , xu = xv , xv ,
xu , xv = 0.
In classical notation (cf. Exercise 4.24), this means that E = G and F = 0. (c) Use the fact that xuu + xvv = 0 (cf. Exercise 8.23) to show that the surface Ī£ = x(V ) has mean curvature H = 0 and therefore Ī£ is minimal. (Hint: Exercise 4.27 may be helpful.)
8. Minimal surfaces in E3 and A3
266
Together, Exercises 8.24 and 8.25 imply the following proposition: Proposition 8.26. A regular surface Ī£ ā E3 is minimal if and only if every point of Ī£ has a neighborhood that can be parametrized by a smooth immersion x : V ā E3 that is the real part of a holomorphic function z : V ā C3 with z (z), z (z) = 0. This brings us to the Weierstrass-Enneper representation for minimal surfaces, which is deļ¬ned as follows. Let U ā C be open, and suppose that we are given (1) a meromorphic function g : U ā C, i.e., a holomorphic function that may have isolated singularities z0 ā U called poles, where g(z) = (zāz1 0 )k h(z) for some holomorphic function h(z) and some integer k ā„ 1, called the order of the pole at z0 ; (2) a holomorphic function f : U ā C with the properties that whenever g has a pole of order k at z0 ā U , f has a zero of order exactly 2k at z0 , and f does not have a zero at any point that is not a pole of g. Choose a base point z0 ā U , and deļ¬ne z : U ā C3 by (8.15) & z & t& z 2 2 1 i z(z) = 2 f (Ī¶)(1 ā g(Ī¶) ) dĪ¶, 2 f (Ī¶)(1 + g(Ī¶) ) dĪ¶, z0
z0
z
f (Ī¶)g(Ī¶) dĪ¶ .
z0
The following exercise shows that for any choice of functions f, g : U ā C as above, the real part of z(z) deļ¬nes a parametrization for a minimal surface: *Exercise 8.27. (a) Show that under the given assumptions on f and g, the integrands in (8.15) are holomorphic and therefore the function z(z) is holomorphic. (b) Show that z (z) = 0 for all z ā U and that z (z), z (z) = 0. Conclude that x = Re(z) : U ā R3 is a parametrization for an immersed minimal surface in E3 . Conversely, the following exercise shows that every minimal surface has a local parametrization of this form. *Exercise 8.28. Let U ā R2 , and let x : U ā E3 be a parametrization for an immersed minimal surface Ī£ ā R3 . Choose a complex coordinate z on U as in Exercise 8.22 and a holomorphic function z : U ā C3 such that z (z), z (z) = 0 and x = Re(z).
8.2. Minimal surfaces in E3
267
(a) Write z (z) = t[Ī¾1 , Ī¾2 , Ī¾3 ], where Ī¾1 , Ī¾2 , Ī¾3 : U ā C3 are holomorphic functions. Show that the condition z (z), z (z) = 0 is equivalent to Ī¾12 + Ī¾22 + Ī¾32 = 0. (b) Deļ¬ne f (z) = Ī¾1 (z) ā iĪ¾2 (z),
g(z) =
Ī¾3 (z) . Ī¾1 (z) ā iĪ¾2 (z)
Show that f is holomorphic and that if g has a pole of order k at z0 ā U , then f has a zero of order 2k at z0 . (Hint: Clearly, zeros of f and poles of g both occur at points z0 ā U where Ī¾1 (z0 ) ā iĪ¾2 (z0 ) = 0. Use the relation in part (a) to relate the order of the pole of g at z0 to the order of the zero of f at z0 .) (c) Show that with f, g as in part (b), z has the form (8.15). *Exercise 8.29 (The associated family of a minimal surface). Let x : U ā E3 be a parametrization for a minimal surface Ī£ ā E3 with WeierstrassEnneper representation x = Re(z), with z as in (8.15). (a) Show that the ļ¬rst fundamental form of Ī£ can be written as I = 12 dz, dzā . (Hint: Recall that dz = Ļ ĀÆ 1 + iĀÆ Ļ 2 .) (b) For each t ā [0, 2Ļ], let zt (z) = eit z(z). Show that the function zt is holomorphic, with zt (z), zt (z) = 0, and therefore, the function xt = Re(zt ) deļ¬nes a parametrization for a minimal surface Ī£t ā E3 . The 1-parameter family of minimal surfaces Ī£t is called the associated family of Ī£. (c) Use part (a) to show that the ļ¬rst fundamental form of Ī£t is independent of t; therefore, the minimal surfaces Ī£t are all isometric. In particular, the surface Ī£āĻ/2 with parametrization y = Im(z) is isometric to Ī£; this surface is called the conjugate surface of Ī£. Exercise 8.30 (Enneperās surface). (a) In the Weierstrass-Enneper representation (8.15), let f (z) = 2,
g(z) = z.
8. Minimal surfaces in E3 and A3
268
Show that the resulting minimal surface Ī£ is parametrized by
x(u, v) = t u ā 13 u3 + uv 2 , āv + 13 v 3 ā vu2 , u2 ā v 2 , where z = u + iv. This surface is called Enneperās surface. (b) Use Maple to plot Enneperās surface over various ranges in u and v. Is it an embedded surface in E3 ? (c) Compute the parametrizations xt for the associated family of Enneperās surface. (Hint: The Weierstrass-Enneper representation for Ī£t can be obtained by taking f (z) = 2eit , g(z) = z.) (d) Use Maple to create an animation of plots of the family of surfaces Ī£t , with t ā [0, 2Ļ] as the time parameter. Exercise 8.31. Let Ī£ ā E3 be the catenoid, parametrized as in Exercise 8.18. (a) Show that the Weierstrass-Enneper representation for Ī£ is obtained (up to a translation) by taking f (z) = āieāiz ,
g(z) = eiz
and that the conjugate surface of the catenoid is the helicoid. (Hint: The formula that you ļ¬nd for the conjugate surface will require a slight reparametrization before it looks like the parametrization for the helicoid from Exercise 8.19.) (b) Use Maple to create an animation of plots of the family of surfaces Ī£t in the associated family, with t ā [0, 2Ļ] as the time parameter.
8.3. Minimal surfaces in A3 Recall that in the process of constructing adapted frame ļ¬elds for elliptic surfaces Ī£ ā A3 , we found two invariant quadratic forms that we referred to as the ļ¬rst and second equi-aļ¬ne fundamental forms. These were deļ¬ned in terms of the Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ) associated to any 2-adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)) along Ī£ by I = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 , II = Ļ ĀÆ 31 Ļ ĀÆ1 + Ļ ĀÆ 32 Ļ ĀÆ 2. These quadratic forms are equi-aļ¬ne analogs of the Euclidean ļ¬rst and second fundamental forms for an elliptic surface Ī£ in E3 , but despite the apparent similarities, they are quite diļ¬erent from their Euclidean counterparts. For instance, if x : U ā A3 is a parametrization of Ī£, then the coeļ¬cients of
8.3. Minimal surfaces in A3
269
the equi-aļ¬ne ļ¬rst fundamental form are deļ¬ned in terms of second derivatives of x, whereas the Euclidean ļ¬rst fundamental form is deļ¬ned in terms of ļ¬rst derivatives of x. (In fact, the equi-aļ¬ne ļ¬rst fundamental form of a surface Ī£ is a scalar multiple of the Euclidean second fundamental form of Ī£; cf. Exercise 6.40(c).) Nevertheless, we will see that an equi-aļ¬ne analog of Theorem 8.16 is true: If we deļ¬ne an area functional for equi-aļ¬ne surfaces based on the equi-aļ¬ne ļ¬rst fundamental form I, then the critical points of this functional are precisely those surfaces for which the trace of II with respect to I is identically zero. 8.3.1. Variational calculations. Let x : U ā A3 be a smooth immersion whose image is an elliptic surface Ī£ = x(U ). Let (e1 (u), e2 (u), e3 (u)) be a 2-adapted frame ļ¬eld along Ī£, and let (ĀÆ Ļi, Ļ ĀÆ ji ) be the associated MaurerCartan forms. *Exercise 8.32. Show that the 2-form dA = Ļ ĀÆ1 ā§ Ļ ĀÆ2 is well-deļ¬ned, independent of the choice of 2-adapted frame ļ¬eld and associated Maurer-Cartan forms. (Hint: This computation makes use of Exercises 6.20 and 6.22.) This 2-form is called the equi-aļ¬ne area form of Ī£. By analogy with the Euclidean case, we deļ¬ne the equi-aļ¬ne area functional A on the set S of closed and bounded elliptic surfaces Ī£ ā A3 to be & A(Ī£) = dA. Ī£
Ė : U ā A(3) is a 2-adapted If x : U ā A3 is a parametrization of Ī£ and x frame ļ¬eld along the surface Ī£ = x(U ), then the equi-aļ¬ne area of Ī£ is given by & & 1 2 Ė ā (Ļ 1 ā§ Ļ 2 ). A(Ī£) = x Ļ ĀÆ ā§Ļ ĀÆ = U
U
In order to look for critical points of the equi-aļ¬ne area functional, we apply the same ideas that we developed in the Euclidean case. The notion of a compactly supported normal variation is precisely the same for surfaces in A3 as for surfaces in E3 ; the only diļ¬erence is that ānormalā refers to the equi-aļ¬ne normal vector, which is well-deļ¬ned for any 2-adapted frame ļ¬eld on Ī£. Thus, the equi-aļ¬ne analog of Deļ¬nition 8.6 is as follows: Deļ¬nition 8.33. Let U ā R2 be an open set, and let x : U ā A3 be an immersion whose image is an elliptic surface Ī£ = x(U ). Ī£ is called
270
8. Minimal surfaces in E3 and A3
an equi-aļ¬ne minimal surface if for every compactly supported variation X : U Ć (āĪµ, Īµ) of x, we have d (8.16) A(Ī£t ) = 0. dt t=0 *Exercise 8.34. Convince yourself that the results of Exercise 8.4 are equally valid in the equi-aļ¬ne case; the only change is that the unit normal vector nt (u) to the surface Ī£t should be replaced with the equi-aļ¬ne normal vector to Ī£t at the point xt (u). Therefore, as in the Euclidean case, it suļ¬ces to consider normal variations of x. Now, let U ā R2 be an open set; let x : U ā A3 be an immersion whose image is an elliptic surface Ī£ = x(U ); and let X : U Ć (āĪµ, Īµ) ā A3 be a compactly supported, normal variation of x. By analogy with the Euclidean case, deļ¬ne a 2-adapted frame ļ¬eld on the variation X as follows: For each (u, t) ā U Ć (āĪµ, Īµ), let (e1 (u, t), e2 (u, t), e3 (u, t)) be a 2-adapted frame for the surface Ī£t at the point xt (u), so that e3 (u, t) is the equi-aļ¬ne normal to the surface Ī£t at the point xt (u). Let (ĀÆ Ļi, Ļ ĀÆ ji ) denote the pullbacks of the 8 Maurer-Cartan forms on A(3) to U Ć (āĪµ, Īµ) via X. *Exercise 8.35 (Cf. Exercise 8.9). (a) For each t ā (āĪµ, Īµ), let ıt : U ā U Ć (āĪµ, Īµ) denote the inclusion map deļ¬ned by ıt (u) = (u, t). Show that the pullbacks of the 1-forms (ĀÆ Ļ1, Ļ ĀÆ 2 ) to U via ıt are the usual dual forms on the surface Ī£t = xt (U ). (b) Show that Ļ ĀÆ 3 = Ļ dt, where the function Ļ(u, t) is determined by the condition that āX e3 (u, t) Ļ ĀÆ3 = dt. āt (c) Show that the pullbacks of the connection forms (ĀÆ Ļji ) to U via ıt are the usual connection forms on the surface Ī£t = xt (U ). In particular, the relations 1 1 Ļ ĀÆ3 11 12 Ļ ĀÆ (8.17) = , Ļ ĀÆ 32 12 22 Ļ ĀÆ2 which hold for a 2-adapted frame ļ¬eld on an elliptic equi-aļ¬ne surface, still hold modulo dt for the corresponding forms on U Ć (āĪµ, Īµ). (This means, e.g., that the 1-form Ļ ĀÆ 31 ā (11 Ļ ĀÆ 1 + 12 Ļ ĀÆ 2) on U Ć (āĪµ, Īµ) is equal to a multiple of dt.)
8.3. Minimal surfaces in A3
271
As in the Euclidean case, the equi-aļ¬ne area functional of the surface Ī£t is given by & Ė āt (Ļ 1 ā§ Ļ 2 ), A(Ī£t ) = x U
and the surface Ī£ = Ī£0 is minimal if and only if for every compactly supported normal variation of x, we have & d d ā 1 Ė t (Ļ ā§ Ļ 2 ) . 0 = A(Ī£t ) = x dt t=0 U dt t=0 *Exercise 8.36. Convince yourself that the results of Exercises 8.11 and 8.12 hold in the equi-aļ¬ne case. *Exercise 8.37 (Cf. Exercise 8.13). Let x : U ā A3 be an immersion, and let X : U Ć (āĪµ, Īµ) ā A3 be a compactly supported, normal variation of x. (a) Use Cartanās formula for the Lie derivative to show that Lā/āt (ĀÆ Ļ1 ā§ Ļ ĀÆ 2) =
ā āt
d(ĀÆ Ļ1 ā§ Ļ ĀÆ 2) .
(b) Use the Cartan structure equations for a 2-adapted coframing, the equation Ļ ĀÆ 3 = Ļ dt, and equation (8.17) to show that d(ĀÆ Ļ1 ā§ Ļ ĀÆ 2 ) = 2LĻĀÆ Ļ1 ā§ Ļ ĀÆ 2 ā§ dt, where L = 12 (11 + 22 ) is the equi-aļ¬ne mean curvature of Ī£. (Note that you only need to know that equation (8.17) holds modulo dt for this computation.) (c) Use parts (a) and (b) to show that & & 1 2 Lā/āt (ĀÆ Ļ ā§Ļ ĀÆ )= 2LĻ Ļ ĀÆ1 ā§ Ļ ĀÆ 2. U
U
(d) Conclude that if the equi-aļ¬ne mean curvature L of Ī£ is identically equal to zero, then & Lā/āt (ĀÆ Ļ1 ā§ Ļ ĀÆ 2) = 0 U
for every compactly supported, normal variation of x and hence Ī£ is equiaļ¬ne minimal. Thus, we have proved the following proposition: Proposition 8.38. If a regular elliptic surface Ī£ ā A3 has mean curvature L identically equal to zero, then Ī£ is equi-aļ¬ne minimal.
8. Minimal surfaces in E3 and A3
272
*Exercise 8.39. Adapt the argument of Exercise 8.15 to the equi-aļ¬ne case to prove the converse of Proposition 8.38: If Ī£ ā A3 is an elliptic surface whose equi-aļ¬ne mean curvature is not identically zero, then Ī£ is not a critical point for the equi-aļ¬ne area functional A, and hence Ī£ is not equi-aļ¬ne minimal. Together with Proposition 8.38, this proves the following equi-aļ¬ne analog of Theorem 8.16: Theorem 8.40. A regular elliptic surface Ī£ ā A3 is equi-aļ¬ne minimal if and only if its equi-aļ¬ne mean curvature L is identically equal to zero. Exercise 8.41. (a) Show that for any values of a, b, c with ac ā b2 > 0, the elliptic paraboloid z = ax2 + bxy + cy 2 is equi-aļ¬nely equivalent to the paraboloid Ī£ ā A3 deļ¬ned by the equation z = 12 (x2 + y 2 ).
(8.18)
(b) Consider the parametrization x : R2 ā A3 of Ī£ given by
x(u, v) = t u, v, 12 u2 + v 2 . Show that the equi-aļ¬ne frame ļ¬eld e1 (u, v) = xu = t[1, 0, u] , (8.19)
e2 (u, v) = xv = t[0, 1, v] , e3 (u, v) = t[0, 0, 1]
is a 2-adapted frame ļ¬eld on Ī£ by computing its dual and connection forms and showing that they satisfy the deļ¬ning conditions Ļ ĀÆ 13 = Ļ ĀÆ 1,
Ļ ĀÆ 23 = Ļ ĀÆ 2,
Ļ ĀÆ 33 = 0
for a 2-adapted frame ļ¬eld. (c) Show that the equi-aļ¬ne mean curvature of Ī£ is identically zero. Conclude that any elliptic paraboloid is equi-aļ¬ne minimal. Exercise 8.42 (Maple recommended). In this exercise, we will derive the conditions that a function f (x, y) must satisfy in order for the graph z = f (x, y) to be an elliptic equi-aļ¬ne minimal surface Ī£ ā A3 . Consider the parametrization x : R2 ā A3 of Ī£ given by x(u, v) = t[u, v, f (u, v)] .
8.3. Minimal surfaces in A3
273
(a) Let (e1 (u), e2 (u), e3 (u)) be the 0-adapted equi-aļ¬ne frame ļ¬eld on Ī£ given by e1 (u, v) = xu = t [1, 0, fu ] , e2 (u, v) = xv = t[0, 1, fv ] , e3 (u, v) = t [0, 0, 1] . Show that the dual forms associated to this frame ļ¬eld are Ļ ĀÆ 1 = du,
Ļ ĀÆ 2 = dv
and that the only nonzero connection forms are Ļ ĀÆ 13 = fuu du + fuv dv, Ļ ĀÆ 23 = fuv du + fvv dv. Thus, we have
h11 h12
h12 h22
=
fuu fuv fuv fvv
.
2 > 0, so that Ī£ is elliptic, and for simplicity assume Assume that fuu fvv āfuv that fuu , fvv > 0. In order to compute the equi-aļ¬ne mean curvature of Ī£, we need to construct a 2-adapted frame ļ¬eld on Ī£ and compute the associated Maurer-Cartan forms. Recall from Chapter 6 that any other Ė2 (u), e Ė3 (u)) on Ī£ has the form 0-adapted frame ļ¬eld (Ė e1 (u), e ā¤ ā” r1
ā¢ B ā„ r2 Ė1 (u) e Ė2 (u) e Ė3 (u) = e1 (u) e2 (u) e3 (u) ā£ e (8.20) ā¦
0
0 (det B)ā1
for some GL(2)-valued function B and real-valued functions r1 , r2 on U and ĖĀÆ i , Ļ ĖĀÆ ji ) associated to this frame ļ¬eld satisfy that the Maurer-Cartan forms (Ļ the relations 1 1 3 3 ĖĀÆ ĖĀÆ 1 Ļ Ļ ĀÆ Ļ Ļ ĀÆ1 (8.21) = B ā1 , = (det B) tB . ĖĀÆ 2 ĖĀÆ 23 Ļ Ļ Ļ ĀÆ2 Ļ ĀÆ 23 (b) Show that if we take ā” ā¤ 2 )1/8 (fuu fvv ā fuv āfuv ā ā ā¢ 2 )3/8 ā„ fuu fuu (fuu fvv ā fuv ā¢ ā„ ā¢ ā„, (8.22) B=ā¢ ā„ ā ā£ ā¦ fuu 0 2 )3/8 (fuu fvv ā fuv
8. Minimal surfaces in E3 and A3
274
then
Ė 11 h Ė 12 h
Ė 12 10 h = , Ė 01 h22
ĖĀÆ 13 = Ļ ĖĀÆ 1 , Ļ ĖĀÆ 23 = Ļ ĖĀÆ 2 . and therefore Ļ (c) Show that for this frame ļ¬eld (with r1 , r2 still arbitrary), we have " # (fuu r1 + fuv r2 ) (fuu fuvv ā 2fuv fuuv + fvv fuuu ) ĖĀÆ 33 = Ļ + du 2 ) 2 )1/4 4(fuu fvv ā fuv (fuu fvv ā fuv " +
(fuv r1 + fvv r2 ) (fuu fvvv ā 2fuv fuvv + fvv fuuv ) + 2 ) 2 1/4 4(fuu fvv ā fuv (fuu fvv ā fuv )
Conclude that by choosing
(8.23)
r1 r2
=ā
fuu fuv
# dv.
ā”
ā¤ (fuu fuvv ā 2fuv fuuv + fvv fuuu ) ā1 ā¢ 2 )3/4 ā„ 4(fuu fvv ā fuv fuv ā¢ ā„ ā¢ ā„, ā¢ ā„ fvv ā£ (fuu fvvv ā 2fuv fuvv + fvv fuuv ) ā¦ 2 )3/4 4(fuu fvv ā fuv
ĖĀÆ 33 = 0 and hence (together with the result of part (b)) we can arrange that Ļ Ė2 (u), e Ė3 (u)) is 2-adapted. that the resulting frame ļ¬eld (Ė e1 (u), e (d) Show that the dual forms associated to this frame ļ¬eld are ā fuu fuv 1 Ė Ļ ĀÆ = du + ā dv, 2 1/8 2 )1/8 (fuu fvv ā fuv ) fuu (fuu fvv ā fuv ĖĀÆ 2 = Ļ
2 )3/8 (fuu fvv ā fuv ā dv. fuu
ĖĀÆ 32 , and ļ¬nd the functions (ij ) such that Then compute Ļ ĀÆĖ 31 and Ļ 1 1 ĖĀÆ 3 ĖĀÆ Ļ 11 12 Ļ = . ĖĀÆ 32 ĖĀÆ 2 Ļ 12 22 Ļ Finally, write the equi-aļ¬ne mean curvature equation L=
1 2
(11 + 22 ) = 0
as a (rather nasty!) fourth-order diļ¬erential equation for f . 8.3.2. A Weierstrass-Enneper-type representation for elliptic equiaļ¬ne minimal surfaces in A3 . In this section, we will derive a Weierstrass-Enneper-type representation for elliptic equi-aļ¬ne minimal surfaces. This formula is originally due to Blaschke and may be found in [Bla85].
8.3. Minimal surfaces in A3
275
Let Ī£ ā A3 be an elliptic surface with parametrization x : U ā A3 , and let (e1 (u), e2 (u), e3 (u)) be a 2-adapted frame ļ¬eld along Ī£, so that the associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ) satisfy Ļ ĀÆ 13 = Ļ ĀÆ 1,
Ļ ĀÆ 23 = Ļ ĀÆ 2,
Ļ ĀÆ 33 = 0.
Recall that for such a frame ļ¬eld, we have 1 1 Ļ ĀÆ3 11 12 Ļ ĀÆ = , Ļ ĀÆ 32 12 22 Ļ ĀÆ2 ā”
ā¤ h1 āh2 ĀÆ1 ā¢ 1 ā„ ā¢ ā„ Ļ 2 ā¢Ļ ā„ ā¢ ā„ , ĀÆ 1 ā¦ = ā£āh2 āh1 ā¦ ā£ ĀÆ2 + Ļ Ļ ĀÆ2 2ĀÆ Ļ22 āh1 h2 2ĀÆ Ļ11
ā¤
ā”
where we have set h1 = h111 = āh122 ,
h2 = h222 = āh112 ,
and that Ī£ is equi-aļ¬ne minimal if and only if 11 + 22 = 0. Now, let A3C denote the complexiļ¬ed equi-aļ¬ne space A3 ā C. (This is simply the vector space C3 , but with an equi-aļ¬ne structure rather than a Euclidean structure.) Consider the Ī2 A3C -valued 1-form (cf. Deļ¬nition 2.11) Ī¾ = 12 e3 ā§ (e1 ā ie2 )(ĀÆ Ļ 1 + iĀÆ Ļ2) = 12 e3 ā§ [(e1 Ļ ĀÆ 1 + e2 Ļ ĀÆ 2 ) + i(e1 Ļ ĀÆ 2 ā e2 Ļ ĀÆ 1 )] on U . *Exercise 8.43. (a) Show that Ī¾ is a well-deļ¬ned 1-form on U , independent of the choice of 2-adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)). (b) Show that dĪ¾ = 12 (11 + 22 )(e1 ā§ e2 ) Ļ ĀÆ1 ā§ Ļ ĀÆ 2. Therefore, dĪ¾ = 0 if and only if Ī£ is equi-aļ¬ne minimal. Now, suppose that Ī£ is equi-aļ¬ne minimal. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame ļ¬eld on Ī£, with associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ), and consider the scalar-valued, complex 1-form Ļ ĀÆ 1 + iĀÆ Ļ 2 on U . *Exercise 8.44. Convince yourself that the results of Exercise 8.22 hold in the equi-aļ¬ne case. Thus, every point u ā U has a neighborhood V ā U on which there exist complex-valued functions z, Ļ : V ā C such that Ļ ĀÆ 1 + iĀÆ Ļ 2 = Ļ dz.
8. Minimal surfaces in E3 and A3
276
Without loss of generality, we can assume that z = u + iv, where (u, v) are local coordinates on V , and we can write the restriction of the parametrization x : U ā A3 of Ī£ to V ā U in the form x(z), thereby deļ¬ning a complex structure on Ī£. Let F : V ā Ī2 A3C be the function F(z, z ā ) = 12 e3 ā§ (e1 ā ie2 )Ļ, so that Ī¾ = F(z, z ā ) dz. *Exercise 8.45. (a) Show that the assumption that Ī£ is equi-aļ¬ne minimal, and hence dĪ¾ = 0, is equivalent to āF = 0. āz ā Therefore, F(z, z ā ) = F(z) is a holomorphic, Ī2 A3C -valued function on V . For the remainder of this exercise, assume that this condition holds. (b) Apply the PoincarĀ“e lemma (cf. Theorem 2.31) to the closed 1-form Ī¾ = F(z) dz to conclude that there exists a Ī2 A3C -valued holomorphic function Z : V ā Ī2 A3C such that Ī¾ = dZ, and therefore, F(z) = Z (z). For ease of notation, let e = 12 (e1 ā ie2 ),
Ļ ĀÆ=Ļ ĀÆ 1 + iĀÆ Ļ2,
so that Ī¾ = dZ = e3 ā§ e Ļ ĀÆ. Then the complex conjugate of Ī¾ is Ī¾ ā = dZā = e3 ā§ eā Ļ ĀÆ ā. (c) Show that d(e ā§ eā ) = 12 (Ī¾ ā ā Ī¾) = 12 (dZā ā dZ). Conclude that Zā ā Z = 2e ā§ eā + 2iC for some real-valued constant C ā Ī2 A3 . (Why must C be real-valued?) (d) Show that by adding an imaginary constant to Z, we can arrange that C = 0. Moreover, this will have no eļ¬ect on the condition that Ī¾ = dZ.
8.3. Minimal surfaces in A3
277
At this point, we need to introduce an operation called the special linear cross product. This is somewhat diļ¬erent from the usual cross product on R3 , in that it operates on a pair of elements of Ī2 A3 and produces an element of A3 . Deļ¬nition 8.46. The special linear cross product is the unique skew-symmetric, bilinear map Ćsl : Ī2 A3 Ć Ī2 A3 ā A3 that is SL(3)-equivariant and satisļ¬es (8.24)
(e1 ā§ e2 ) Ćsl (e1 ā§ e3 ) = e1
for any unimodular basis (e1 , e2 , e3 ) of A3 . A few comments on this deļ¬nition are in order: (1) āSL(3)-equivariantā means that for any matrix A ā SL(3), we have (Ae1 ā§ Ae2 ) Ćsl (Ae1 ā§ Ae3 ) = Ae1 . (2) Let (e1 , e2 , e3 ) denote the standard basis for A3 . Since any other unimodular basis (e1 , e2 , e3 ) can be expressed as (e1 , e2 , e3 ) = (Ae1 , Ae2 , Ae3 ) for some matrix A ā SL(3), requiring that (8.24) hold for every unimodular basis is equivalent to requiring that it hold only for the standard basis and that it be SL(3)-equivariant. (3) This cross product extends via bilinearity in the usual way to elements of the complexiļ¬ed space Ī2 A3C . For a geometric interpretation of this cross product, think of an element v1 ā§ v2 ā Ī2 A3 as representing the oriented plane spanned by the vectors (v1 , v2 ) in A3 . The cross product of two such planes (v1 ā§ v2 , w1 ā§ w2 ) is a vector that spans the line of intersection of the two planes. Exercise 8.47. Show that the special linear cross product can be expressed in terms of the usual cross product on R3 (or C3 ) as (v1 ā§ v2 ) Ćsl (w1 ā§ w2 ) = (v1 Ć v2 ) Ć (w1 Ć w2 ). (Hint: Since both sides are skew-symmetric and bilinear, it suļ¬ces to show that the equation holds in the case where v1 = w1 = e1 ,
v2 = e2 ,
w2 = e 3
and that the right-hand side is SL(3)-equivariant.)
8. Minimal surfaces in E3 and A3
278
We are now ready to derive Blaschkeās formula. *Exercise 8.48. (a) Use the results of Exercise 8.45 to show that (Zā ā Z) Ćsl d(Zā + Z) = āi(e Ļ ĀÆ + eā Ļ ĀÆ ā ) = āi dx, where x : V ā A3 is the given parametrization of Ī£. (b) Conclude that dx = i[(Zā ā Z) Ćsl d(Zā + Z)] = i[Zā Ćsl dZā ā Z Ćsl dZ + d(Zā Ćsl Z)]. Therefore, up to translation, the parametrization x : V ā A3 of the Ī£ is given in terms of the complex coordinate z on V by (8.25) " # & z ā ā ā x(z) = i Z(z) Ćsl Z(z) + Z(Ī¶) Ćsl Z (Ī¶) ā Z(Ī¶) Ćsl Z (Ī¶) dĪ¶ z0
for any choice of base point z0 ā V . This formula is called the Blaschke representation for Ī£. Therefore, for any equi-aļ¬ne minimal surface x : U ā A3 and any point u ā U , there exists a neighborhood V ā U of u for which the restriction of x to V can be written in the form (8.25) for some holomorphic function Z : V ā Ī2 A3C . *Exercise 8.49. Write Z = 12 (X + iY) for some smooth functions X, Y : V ā Ī2 A3R . Recall that the functions X, Y must satisfy the Cauchy-Riemann equations (8.26)
Xu = Yv ,
Xv = āYu ,
which in turn imply that X and Y are harmonic; i.e., (8.27)
Xuu + Xvv = Yuu + Yvv = 0
(cf. Exercise 8.23). (a) Show that dx = (Y Ćsl Yv ) du ā (Y Ćsl Yu ) dv. (b) Conclude that in order for x to be an immersion, the vectors (Y(z), Yu (z), Yv (z)) must be linearly independent elements of Ī2 A3R for each z ā V.
8.3. Minimal surfaces in A3
279
Conversely, let V ā C be an open set, and let Z = 12 (X + iY) : V ā Ī2 A3C be a holomorphic function such that the vectors (Y(z), Yu (z), Yv (z)) are linearly independent for each z ā V . Deļ¬ne an immersion x : V ā A3 by (8.25). *Exercise 8.50. Let Ī» : V ā R be a smooth function, and consider the frame ļ¬eld along Ī£ = x(V ) deļ¬ned by e1 (u) = Ī»xu = Ī»(Y Ćsl Yv ), (8.28)
e2 (u) = Ā±Ī»xv = āĪ»(Y Ćsl Yu ), e3 (u) = Ī»2 (Yu Ćsl Yv ),
with the sign for e2 (u) chosen so that the matrix e1 (u) e2 (u) e3 (u) has positive determinant. (a) Show that there exists a unique choice for the function Ī» (up to sign) for which (e1 (u), e2 (u), e3 (u)) is a unimodular frame ļ¬eld. For the remainder of this exercise, assume that Ī» has been chosen accordingly. (b) Show that the dual forms associated to the frame ļ¬eld (8.28) are Ļ ĀÆ 1 = Ī»ā1 du,
Ļ ĀÆ 2 = Ā±Ī»ā1 dv,
with the sign of Ļ ĀÆ 2 chosen according to the sign of e2 (u). In order to compute the connection forms associated to the frame ļ¬eld (8.28), observe that, since (Y, Yu , Yv ) are linearly independent, the second derivatives of Y can be written as Yuu = h0 Y + h1 Yu + h2 Yv , Yuv = k0 Y + k1 Yu + k2 Yv , Yvv = āh0 Y ā h1 Yu ā h2 Yv for some functions hi , ki : V ā R. (c) Diļ¬erentiate equations (8.28) and show that the connection forms associated to the frame ļ¬eld (8.28) satisfy the conditions Ļ ĀÆ 13 = Ļ ĀÆ 1,
Ļ ĀÆ 23 = Ļ ĀÆ 2,
Ļ ĀÆ 33 = 0.
Conclude that the frame ļ¬eld (8.28) is 2-adapted. (Hint: For the last condition, ļ¬rst show that Ļ ĀÆ 11 + Ļ ĀÆ 22 = Ļ ĀÆ 33 = 2Ī»ā1 dĪ» + Ī»(h1 + k2 )ĀÆ Ļ 1 ā Ī»(h2 + k1 )ĀÆ Ļ2. Then use the assumption that the frame ļ¬eld is unimodular to conclude that Ļ ĀÆ 33 = 0.)
8. Minimal surfaces in E3 and A3
280
(d) Show that
1 Ļ ĀÆ3
2
h0 k 0
=Ī» Ļ ĀÆ 32 k0 āh0 Conclude that Ī£ is equi-aļ¬ne minimal.
Ļ ĀÆ1
Ļ ĀÆ2
.
Together, Exercises 8.48 and 8.50 imply the following proposition: Proposition 8.51. A regular elliptic surface Ī£ ā A3 is equi-aļ¬ne minimal if and only if every point of Ī£ has a neighborhood that can be parametrized by a smooth immersion x : V ā A3 of the form (8.25) for some holomorphic function Z : V ā Ī2 A3C . *Exercise 8.52. Let Ī£ be the elliptic paraboloid (8.18) of Exercise 8.41, with the 2-adapted frame ļ¬eld (8.19). Show that the Blaschke representation for Ī£ may be obtained by taking Z(z) =
1 2
(āie1 ā§ e2 + iz e2 ā§ e3 + z e3 ā§ e1 ) ,
where (e1 , e2 , e3 ) represents the standard basis of A3 . (Hint: Write the frame vectors (e1 (u, v), e2 (u, v), e3 (u, v)) as e1 (u, v) = e1 + ue3 , e2 (u, v) = e2 + ve3 , e3 (u, v) = e3 , and note that, since (e1 , e2 , e3 ) is a unimodular basis, we have (e1 ā§ e2 ) Ć (e1 ā§ e3 ) = e1 , (e2 ā§ e3 ) Ć (e2 ā§ e1 ) = e2 , (e3 ā§ e1 ) Ć (e3 ā§ e2 ) = e3 .)
8.4. Maple computations As usual, begin by loading the Cartan and LinearAlgebra packages into Maple. Exercise 8.18 (e): Deļ¬ne the parametrization for Ī£ and the orthonormal frame ļ¬eld (e1 (u), e2 (u), e3 (u)): > PDETools[declare](rho(v)); X:= Vector([rho(v)*cos(u), rho(v)*sin(u), v]); > Xu:= map(diff, X, u); Xv:= map(diff, X, v); > e1:= Xu/simplify(Norm(Xu, Euclidean, conjugate=false),
8.4. Maple computations
281
symbolic); e2:= Xv/simplify(Norm(Xv, Euclidean, conjugate=false), symbolic); e3:= simplify(CrossProduct(e1, e2)); We can use the following substitution to go back and forth between the (du, dv) basis and the (ĀÆ Ļ1, Ļ ĀÆ 2 ) basis as necessary (but ļ¬rst we need to declare 1 2 the 1-forms (ĀÆ Ļ ,Ļ ĀÆ )): > Form(omega[1], omega[2]); > framesub:= [ omega[1] = simplify(Norm(Xu, Euclidean, conjugate=false), symbolic)*d(u), omega[2] = simplify(Norm(Xv, Euclidean, conjugate=false), symbolic)*d(v)]; > framebacksub:= makebacksub(framesub); In order to compute the (hij ), we need to express de3 as a linear combination of (e1 (u), e2 (u)). We know from the Cartan structure equations (3.1) that the coeļ¬cients of (e1 (u), e2 (u)) will be the 1-forms Ļ ĀÆ 31 = āĀÆ Ļ13 , Ļ ĀÆ 32 = āĀÆ Ļ23 , respectively. > de3:= map(Simf, subs(framebacksub, map(d, e3))); > zero1:= Simf(de3 + (omega[3,1]*e1 + omega[3,2]*e2)); > Simf(solve({zero1[1], zero1[2], zero1[3]}, {omega[3,1], omega[3,2]})); 1
Ļv,v Ļ2 1 + Ļ2v Ļ1 Ļ3,1 = ā
, Ļ3,2 = 1 + 2 Ļ2v + Ļ4v 1 + Ļ2v Ļ > assign(%); We can read oļ¬ the (hij ) directly from Ļ ĀÆ 13 and Ļ ĀÆ 23 : > h:= Matrix([ [pick(omega[3,1], omega[1]), pick(omega[3,1], omega[2])], [pick(omega[3,2], omega[1]), pick(omega[3,2], omega[2])]]); Finally, Ī£ is minimal if and only if H = 12 (h11 + h22 ) = 0: > minsurfeq:= numer(simplify(Trace(h))); minsurf eq := āĻ2v ā 1 + Ļv,v Ļ Now, Maple can solve this equation with the dsolve command. Unfortunately, Maple can be rather clumsy about simplifying exponentials and
8. Minimal surfaces in E3 and A3
282
trigonometric functions, so it requires a bit of manipulation to get the solution into a nice form. > soln:= dsolve(minsurfeq, rho(v)); ā ā soln := Ļ =
1 ā C1 ā 2
e
v C1
1 !2 e
C2 C1
ā !2 + 1ā e
v C1
e
C2 C1
,
" 1 f= 2
C1
e
v C1
e
!2 e v C1
e
C2 C1
#
!2 +1
C2 C1
Maple apparently doesnāt even realize that these two solutions are, in fact, the same! But we can verify that our eyes are not deceiving us on this point: > Simf(subs(soln[1], rho(v)) - subs(soln[2], rho(v))); 0 So, we can choose either solution and perform some gymnastics to force Maple to put it into a nicer form. (Try unpacking this command and applying these operations one at a time to see the intermediate steps.) > Simf(convert(combine(expand(Simf(subs(soln[1], rho(v)))), exp), trig)); " # v + C2 C1 cosh C1 Exercise 8.30: Deļ¬ne the Weierstrass parametrization associated to functions f (z), g(z): > PDETools[declare](f(z), g(z)); > dZ:= Vector([(1/2)*f(z)*(1 - g(z)Ė2), (I/2)*f(z)*(1 + g(z)Ė2), f(z)*g(z)]); > Z:= map(int, dZ, z); Now consider the case f (z) = 2, g(z) = z: > Z0:= Simf(subs([f(z) = 2, g(z) = z], Z)); In order to compute the parametrization x(u, v), we need to introduce the real and imaginary parts of z and tell Maple that the components are real: > assume(u, real); assume(v, real); > X0:= map(Re, Simf(subs([z = u + I*v], Z0)));
8.4. Maple computations
283
The plot3d command can be used to plot the surface over various parameter ranges; e.g., > plot3d(X0, u=-2..2, v=-2..2, axes = normal, scaling=constrained);
Next, compute the associated family of surfaces: > assume(t, real); > Zt:= Simf(subs([f(z) = 2*exp(I*t), g(z) = z], Z)); > Xt:= map(Re, Simf(subs([z = u + I*v], Zt))); In order to animate the associated family, we need to load the plots package, and then we can use the animate3d command: > with(plots); > animate3d(Xt, u=-2..2, v=-2..2, t=0..2*Pi, axes=normal, scaling=constrained, frames=50); To view the animation, click on the plot. Then, from the Plot menu, select Animation ā Play. Exactly the same procedure can be used for Exercise 8.31 to animate the associated family that interpolates between the helicoid and the catenoid. Exericse 8.41: First of all, letās remove the assumptions on (u, v), just because the trailing tildes in the output are distracting: > unassign(āuā, āvā); Deļ¬ne the parametrization for Ī£ and the unimodular frame ļ¬eld (e1 (u), e2 (u), e3 (u)): > X:= Vector([u, v, (1/2)*(uĖ2 + vĖ2)]); > e1:= map(diff, X, u); e2:= map(diff, X, v); e3:= Vector([0,0,1]);
8. Minimal surfaces in E3 and A3
284
Since we have e1 (u) = xu (u), e2 (u) = xv (u), it follows that the associated Maurer-Cartan forms Ļ ĀÆ 1, Ļ ĀÆ 2 are equal to Ļ ĀÆ 1 = du,
Ļ ĀÆ 2 = dv.
Set this up as a substitution: > framesub:= [omega[1] = d(u), omega[2] = d(v)]; framebacksub:= makebacksub(framesub); In order to compute the connection forms (ĀÆ Ļji ), diļ¬erentiate the frame ļ¬elds: > de1:= map(Simf, subs(framebacksub, map(d, e1))); de2:= map(Simf, subs(framebacksub, map(d, e2))); de3:= map(Simf, subs(framebacksub, map(d, e3))); From the output, it is easy to read oļ¬ that de1 = e3 Ļ ĀÆ 1,
de2 = e3 Ļ ĀÆ 2,
de3 = 0.
Therefore, we have Ļ ĀÆ 13 = Ļ ĀÆ 1,
Ļ ĀÆ 23 = Ļ ĀÆ 2,
and the remaining connection forms (ĀÆ Ļji ) are equal to zero. It follows that this frame ļ¬eld is 2-adapted. Moreover, since Ļ ĀÆ 31 = Ļ ĀÆ 32 = 0, Ī£ is an equiaļ¬ne minimal surface. Exercise 8.42: Deļ¬ne the parametrization for Ī£ and the 0-adapted frame ļ¬eld of part (a) (weāll call this frame ļ¬eld (e10, e20, e30) because it will be reļ¬ned later): > PDETools[declare](f(u,v)); > X:= Vector([u, v, f(u,v)]); > e10:= map(diff, X, u); e20:= map(diff, X, v); e30:= Vector([0,0,1]); As in the previous exercise, for this frame ļ¬eld we have Ļ ĀÆ 1 = du, Ļ ĀÆ 2 = dv: > framesub0:= [omega[1] = d(u), omega[2] = d(v)]; > framebacksub0:= makebacksub(framesub0); In order to compute the connection forms (ĀÆ Ļji ), diļ¬erentiate the frame ļ¬elds: > de10:= map(Simf, subs(framebacksub0, map(d, e10))); de20:= map(Simf, subs(framebacksub0, map(d, e20))); de30:= map(Simf, subs(framebacksub0, map(d, e30)));
8.4. Maple computations
285
From the output, we see that Ļ ĀÆ 13 = fuu Ļ ĀÆ 1 + fuv Ļ ĀÆ 2,
Ļ ĀÆ 23 = fuv Ļ ĀÆ 1 + fvv Ļ ĀÆ 2,
and the remaining connection forms (ĀÆ Ļji ) are equal to zero. Now we need to ļ¬nd a matrix-valued function B and real-valued functions r1 , r2 on U that will make the frame ļ¬eld (8.20) 2-adapted. Rather than simply verifying that the expressions given in equations (8.22) and (8.23) will do the trick, letās see if we can ļ¬gure out where they came from. First, look for a matrix-valued function B such that the transformed MaurerĖĀÆ 13 = Ļ ĖĀÆ 1 , Cartan forms in equation (8.21) will satisfy the 1-adapted condition Ļ ĖĀÆ 23 = Ļ ĖĀÆ 2 . We can make this problem slightly simpler by observing that, since Ļ B is only determined up to multiplication by a rotation matrix, we should be able to ļ¬nd a suitable matrix B whose lower left-hand entry is zero. With this assumption, deļ¬ne the new frame ļ¬eld and the new Maurer-Cartan forms as follows: > B:= Matrix([[b[1,1], b[1,2]], [0, b[2,2]]]); > e1:= Simf(B[1,1]*e10 + B[2,1]*e20); e2:= Simf(B[1,2]*e10 + B[2,2]*e20); e3:= Simf(r1*e10 + r2*e20 + (1/Determinant(B))*e30); > dualformsvec:= Simf(MatrixInverse(B). Vector([Simf(subs(framesub0, omega[1])), Simf(subs(framesub0, omega[2]))])); > connformsvec1:= Simf(Determinant(B)*Transpose(B). Vector([Simf(subs(framesub0, omega[3,1])), Simf(subs(framesub0, omega[3,2]))])); The 1-adapted condition is simply the condition that these last two vectors of 1-forms are equal. Thus, collecting all the scalar coeļ¬cients of the 1forms in their diļ¬erence gives a system of equations that can be solved for the entries of B: > zero2:= map(Simf,connformsvec1 - dualformsvec); > eqns:= {op(ScalarForm(zero2[1])), op(ScalarForm(zero2[2]))}; > solve(eqns, {b[1,1], b[1,2], b[2,2]}); The resulting output gives complicated expressions involving RootOf, but it is straightforward to check that the solution yields the matrix B in equation (8.22). Make these assignments so that we can go on to the next step: > b[1,1]:= (diff(f(u,v),u,u)*diff(f(u,v),v,v) - diff(f(u,v),u,v)Ė2)Ė(1/8)/sqrt(diff(f(u,v),u,u)); b[1,2]:= -diff(f(u,v),u,v)/(sqrt(diff(f(u,v),u,u))* (diff(f(u,v),u,u)*diff(f(u,v),v,v)
8. Minimal surfaces in E3 and A3
286
- diff(f(u,v),u,v)Ė2)Ė(3/8)); b[2,2]:= sqrt(diff(f(u,v),u,u))/ (diff(f(u,v),u,u)*diff(f(u,v),v,v) - diff(f(u,v),u,v)Ė2)Ė(3/8); Set up a substitution for the Maurer-Cartan forms for this new frame ļ¬eld: > framesub:= [omega[1] = Simf(dualformsvec[1]), omega[2] = Simf(dualformsvec[2])]; > framebacksub:= makebacksub(framesub); We still need to solve for r1 , r2 . This requires computing dĖ e3 so that we can ĖĀÆ 33 and set it equal to zero. We can do this fairly compactly as follows: ļ¬nd Ļ If we let A be the matrix
Ė1 (u) e Ė2 (u) e Ė3 (u) , A= e then we have
ā¤ Ļ ĀÆĖ 31 ā¢ Ė 2ā„ ĀÆ3 ā¦ . dĖ e3 = A ā£ Ļ Ļ ĀÆĖ 33 ā”
ĖĀÆ 33 is the last entry of the vector Aā1 dĖ Therefore, Ļ e3 , and we can solve for ĖĀÆ 33 equal to zero. r1 , r2 by setting the scalar coeļ¬cients of Ļ > > > > > >
A:= Matrix([e1, e2, e3]); de3:= map(Simf, subs(framebacksub, map(d, e3))); connformsvec2:= map(Simf, MatrixInverse(A).de3); zero3:= connformsvec2[3]; solve({op(ScalarForm(zero3))}, {r1, r2}); assign(%);
You should check that the resulting expressions for r1 , r2 agree with equation (8.23). ĖĀÆ 31 and Finally, the equi-aļ¬ne mean curvature L of Ī£ can be computed from Ļ 2 ĖĀÆ 3 , which are the ļ¬rst two entries of connformsvec2. First, we need to use Ļ ĖĀÆ 1 , Ļ ĖĀÆ 2 ) the substitution to express these forms as linear combinations of (Ļ and then compute the trace of the associated coeļ¬cient matrix: > omega[1,3]:= Simf(subs(framebacksub, connformsvec2[1])); omega[2,3]:= Simf(subs(framebacksub, connformsvec2[2])); > LL:= (1/2)*(Simf(pick(omega[1,3], omega[1]) + pick(omega[2,3], omega[2]))); Now, arenāt you glad that you didnāt have to compute L by hand?
10.1090/gsm/178/09
Chapter 9
Pseudospherical surfaces and BĀØ acklundās theorem
9.1. Introduction In this chapter, we will show how moving frames may be used to prove BĀØ acklundās theorem. BĀØacklundās theorem concerns surfaces of constant negative Gauss curvature, also known as pseudospherical surfaces. The bestknown pseudospherical surface is, of course, the pseudosphere, which is the surface of revolution obtained by revolving the tractrix Ī±(t) = t[t ā tanh t, sech t, 0] about the x-axis. But there are inļ¬nitely many other pseudospherical surfaces as well. BĀØacklundās theorem is based on a beautiful geometric construction that starts with a given pseudospherical surface and produces from it a 2-parameter family of new pseudospherical surfaces. (See [RS82] for a discussion of BĀØacklundās original construction.) The construction can be iterated, thereby producing an arbitrary number of increasingly complicated families of pseudospherical surfaces from a single starting surface. For example, we might take the pseudosphere as our initial surface and use BĀØacklundās construction to generate new families of surfaces. 287
288
9. Pseudospherical surfaces and BĀØacklundās theorem
In Ā§9.4, we will see how pseudospherical surfaces are intimately connected with solutions Ļ(x, y) of the partial diļ¬erential equation known as the sineGordon equation: (9.1)
Ļxy = sin(Ļ).
This is a nonlinear partial diļ¬erential equation, and it is one of a number of nonlinear PDEs known as āintegrable systemsā. PDEs in this class share a number of important features, including special families of solutions known as āsolitonā solutions. As we will see, BĀØacklundās geometric construction for pseudospherical surfaces gives rise to a corresponding transformation between solutions of equation (9.1). This transformation is called a BĀØ acklund transformation, and, just as BĀØacklundās construction generates new pseudospherical surfaces from a known pseudospherical surface, the BĀØacklund transformation for equation (9.1) can be used to generate new solutions from any known solution (cf. Exercise 9.17). Applying this transformation, even starting from the trivial solution Ļ(x, y) = 0, produces nontrivial new solutions; in fact, the soliton solutions of (9.1) can be constructed in precisely this way.
9.2. Line congruences BĀØacklundās construction begins with the notion of a line congruence. Roughly, a line congruence in E3 is simply a 2-parameter family of lines in E3 . A more formal deļ¬nition requires that the set of lines in E3 be given a topological structure so that it becomes a manifold, called the aļ¬ne Grassmannian G1 (E3 ). This can be accomplished as follows: Any line in E3 is determined by a pair (x, e), where x is a point on and e is a unit vector parallel to . The set of all such pairs is equal to the product E3 Ć S2 . Now deļ¬ne an equivalence relation ā¼ on E3 Ć S2 by the condition that (x, e) ā¼ (y, f ) if and only if the pairs (x, e) and (y, f ) determine the same line. Exercise 9.1. Show that (x, e) ā¼ (y, f ) if and only if f = Ā±e and y = x+te for some t ā R. The aļ¬ne Grassmannian G1 (E3 ) is then deļ¬ned to be the set of equivalence classes G1 (E3 ) = E3 Ć S2 / ā¼, with the line ā G1 (E3 ) determined by the pair (x, e) denoted by [(x, e)]. The set G1 (E3 ) can be given the structure of a smooth manifold of dimension 4 in a natural way. With this deļ¬nition in hand, the formal deļ¬nition of a line congruence is as follows: Deļ¬nition 9.2. A line congruence in E3 is an immersed surface in G1 (E3 ).
9.3. BĀØacklundās theorem
289
Line congruences were the object of much study in the nineteenth century; for a thorough treatment, see [Eis60]. Example 9.3. Let U be an open set in R2 , and let x : U ā E3 be an immersion whose image is a regular surface Ī£ ā E3 . For each u ā U , let (u) denote the line in E3 passing through the point x(u) and parallel to the normal vector e3 (u). Then the collection {(u) | u ā U } is a line congruence in E3 . A line congruence of this type is called a normal congruence. Given an open set U ā R2 and a line congruence : U ā G1 (E3 ), we can express the congruenceāin inļ¬nitely many diļ¬erent waysāas (u) = [(x(u), e(u)], where x(u) is a point on the line (u) and e(u) is a unit vector parallel to (u). If x(u) is chosen to be a smooth function of u, then the surface Ī£ = x(U ) is called a surface of reference for the line congruence. Any other 8 =x Ė (U ) for the congruence can then be parametrized surface of reference Ī£ as Ė (u) = x(u) + Ī»(u)e(u) x for some smooth, real-valued function Ī» on U . For a generic line congruence : U ā G1 (E3 ), there are two distinguished 8 = x Ė (U ), called focal surfaces of the surfaces of reference Ī£ = x(U ), Ī£ congruence. The deļ¬nition and construction of the focal surfaces are rather involved and will be omitted here (for details, see [Eis60]); for our purposes, the key property of these surfaces is that each line in the congruence is tangent to both focal surfaces. We will assume that both focal surfaces are Ė (u) are parametrized as above, so that for each u ā U , the points x(u), x 8 contained in the line (u) and the line (u) is tangent to the surfaces Ī£, Ī£ Ė (u), respectively. at the points x(u), x
9.3. BĀØ acklundās theorem BĀØacklundās theorem concerns a special category of line congruences known as pseudospherical congruences. Deļ¬nition 9.4. Let U ā R2 , and let : U ā G1 (E3 ) be a line congruence 8 parametrized as above. The congruence is in E3 with focal surfaces Ī£, Ī£, called pseudospherical if the following two conditions hold: (1) The distance r = |Ė x(u) ā x(u)| is a constant, independent of u.
290
9. Pseudospherical surfaces and BĀØacklundās theorem
(2) The angle Ī± (assumed to be nonzero) between the surface normal 8 at the points x(u), x Ė3 (u) to the surfaces Ī£, Ī£ Ė (u), vectors e3 (u), e respectively, is a constant, independent of u. Theorem 9.5 (BĀØacklund). (1) Let U ā R2 , and suppose that : U ā G1 (E3 ) is a pseudospherical 8 Then both Ī£ and Ī£ 8 line congruence in E3 with focal surfaces Ī£, Ī£. sin2 (Ī±) have constant negative Gauss curvature K = ā r2 , where r, Ī± are as in Deļ¬nition 9.4. (2) If Ī£ ā E3 is any surface of constant negative Gauss curvature 2 K = ā sinr2(Ī±) , then given any point x0 ā Ī£ and any unit tangent vector e0 ā Tx0 Ī£ that is not a principal direction at x0 , there exists 8 ā E3 and a pseudospherical line congruence with a unique surface Ī£ 8 such that if x 8 corresponding Ė 0 is the point in Ī£ focal surfaces Ī£, Ī£ Ė 0 ā x0 = re0 and the angle between the surface normal to x0 , then x 8 at the points x0 , x Ė 0 , respectively, is Ī±. vectors to Ī£, Ī£ The construction in part (2) of Theorem 9.5 is due to Bianchi and BĀØ acklund ([BĀØ 83], [Bia79]) and is called a BĀØ acklund transformation; the terminology 8 is a ātransformationā of the original refers to the idea that the new surface Ī£ surface Ī£. BĀØacklundās theorem can be proved using local coordinates on the surfaces 8 but the computations are rather ugly. The proof can be greatly simĪ£, Ī£, pliļ¬ed by using the method of moving frames because frame ļ¬elds can be adapted to the geometry of the problem in a way that local coordinates cannot. Whereas in previous chapters we have adapted our frames according to the geometry of a single surface, here we have to consider two surfaces and the geometric conditions relating them. We will use these considerations to 8 (This guide our choices of orthonormal frame ļ¬elds on the surfaces Ī£, Ī£. proof is taken from [CT80].) 8 respectively, Ė : U ā E3 be the parametrizations of Ī£, Ī£, Proof. (1) Let x, x 3 induced from the line congruence : U ā G1 (E ) as above. We can choose Ė2 (u), orthonormal frame ļ¬elds (e1 (u), e2 (u), e3 (u)) along Ī£ and (Ė e1 (u), e 8 such that: Ė3 (u)) along Ī£ e Ė3 (u) is the unit (1) e3 (u) is the unit normal vector to Ī£ at x(u) and e 8 Ė (u); therefore, (e1 (u), e2 (u)) span the tannormal vector to Ī£ at x 8 Ė2 (u)) span the tangent space TxĖ (u) Ī£. gent space Tx(u) Ī£ and (Ė e1 (u), e
9.3. BĀØacklundās theorem
291
Ė1 (u) is the common unit tangent vector to both surfaces (2) e1 (u) = e Ė (u) ā x(u). in the direction of x Remark 9.6. The notation ei (u) is intended to distinguish these orthonormal frame vectors from the principal orthonormal frame vectors ei (u) that will be introduced in Ā§9.4. This notational distinction will become important in Ā§9.5, when we need to consider both of these orthonormal frame ļ¬elds simultaneously! Similarly, we will denote the Maurer-Cartan forms associated to the frame ļ¬eld (e1 (u), e2 (u), e3 (u)) by (ĀÆ Ļi , Ļ ĀÆ ij ), and those i i ĖĀÆ , Ļ ĖĀÆ j ). Ė2 (u), e Ė3 (u)) by (Ļ associated to the frame ļ¬eld (Ė e1 (u), e *Exercise 9.7. Show that it follows from these conditions (and the deļ¬niĖ3 (u)) that tion of Ī± as the angle between e3 (u), e Ė1 (u) = e1 (u), e (9.2)
Ė2 (u) = cos(Ī±)e2 (u) + sin(Ī±)e3 (u), e Ė3 (u) = ā sin(Ī±)e2 (u) + cos(Ī±)e3 (u) e
and that (9.3)
Ė (u) = x(u) + re1 (u). x
Let (ĀÆ Ļi , Ļ ĀÆ ij ) denote the pullbacks of the Maurer-Cartan forms (Ļ i , Ļji ) on ĖĀÆ i , Ļ ĖĀÆ ji ) E(3) to U via the frame ļ¬eld (x(u); e1 (u), e2 (u), e3 (u)) for Ī£, and let (Ļ i i Ė1 (u), e Ė2 (u), denote the pullbacks of (Ļ , Ļj ) to U via the frame ļ¬eld (Ė x(u); e 8 Ė3 (u)) for Ī£. e *Exercise 9.8. (a) Show that taking the exterior derivative of equation (9.3) and applying the Cartan structure equations (3.1) yields (9.4)
ĖĀÆ 1 + e ĖĀÆ 2 = e1 Ļ Ė1 Ļ Ė2 Ļ e ĀÆ 1 + e2 Ļ ĀÆ 2 + r(e2 Ļ ĀÆ 21 + e3 Ļ ĀÆ 31 ).
Ė2 (u)) into equation (9.4) Then substitute the expressions (9.2) for (Ė e1 (u), e to obtain ĖĀÆ 1 + e2 cos(Ī±) Ļ ĖĀÆ 2 + e3 sin(Ī±) Ļ ĖĀÆ 2 = e1 Ļ e1 Ļ ĀÆ 1 + e2 (ĀÆ Ļ2 + r Ļ ĀÆ 21 ) + e3 (r Ļ ĀÆ 31 ). Conclude that we have the following relationships between the Maurer8 Cartan forms on Ī£ and those on Ī£: ĖĀÆ 1 = Ļ Ļ ĀÆ 1, (9.5)
ĖĀÆ 2 = Ļ cos(Ī±) Ļ ĀÆ2 + r Ļ ĀÆ 21 , ĖĀÆ 2 = r Ļ sin(Ī±) Ļ ĀÆ 31 .
(b) Show that the last two equations in (9.5) imply that (9.6)
Ļ ĀÆ2 + r Ļ ĀÆ 21 = r cot(Ī±) Ļ ĀÆ 31 .
292
9. Pseudospherical surfaces and BĀØacklundās theorem
(c) Use the fact that Ļ ĀÆ ij = dej , ei (and similarly for Ļ ĀÆĖ ji ), the expressions (9.2), and equation (9.6) to show that (9.7)
ĖĀÆ 13 = Ļ
sin(Ī±) 2 Ļ ĀÆ , r
ĖĀÆ 23 = Ļ Ļ ĀÆ 32 .
Next, recall that the coeļ¬cients (hij ) of the second fundamental form of Ī£ are deļ¬ned by the equations (9.8)
Ļ ĀÆ 31 = h11 Ļ ĀÆ 1 + h12 Ļ ĀÆ 2, Ļ ĀÆ 32 = h12 Ļ ĀÆ 1 + h22 Ļ ĀÆ 2.
Because we have not made any attempt to arrange for e1 (u) and e2 (u) to be principal vector ļ¬elds, we should not expect to have h12 = 0; in fact, the following exercise shows that h12 cannot be equal to zero. *Exercise 9.9. Use the ļ¬rst and third equations in (9.5), the ļ¬rst equation ĖĀÆ 1 , Ļ ĖĀÆ 2 ) are linearly independent 1-forms on U to in (9.8), and the fact that (Ļ conclude that h12 = 0. Ė of Ī£. 8 Recall that Finally, consider the Gauss curvature K ĖĻ ĖĀÆ 3 ā§ Ļ ĖĀÆ 3 = K ĖĀÆ 1 ā§ Ļ ĖĀÆ 2 (9.9) Ļ 1
2
(cf. Exericse 4.47). *Exercise 9.10. (a) Use equations (9.7) and (9.8) to show that ĖĀÆ 13 ā§ Ļ ĖĀÆ 23 = ā Ļ
sin(Ī±) ĀÆ1 ā§ Ļ ĀÆ 2. h12 Ļ r
(b) Use equation (9.9), the ļ¬rst and third equations of (9.5), and (9.8) to show that Ė r h12 Ļ ĖĀÆ 13 ā§ Ļ ĖĀÆ 23 = K Ļ ĀÆ1 ā§ Ļ ĀÆ 2. sin(Ī±) (c) Conclude from parts (a) and (b) and the fact that h12 = 0 that 2
Ė = ā sin (Ī±) . K r2 2
An analogous argument shows that K = ā sinr2(Ī±) as well. The proof of part (2) of Theorem 9.5 involves concepts from the theory of exterior diļ¬erential systems similar to those needed for the proof of Lemma 4.12. The key step involves constructing an adapted orthonormal frame ļ¬eld (e1 (u), e2 (u), e3 (u)) along Ī£ with the property that a parametrization 8 will be given by Ė : U ā E3 for the desired surface Ī£ x (9.10)
Ė (u) = x(u) + re1 (u). x
9.4. Pseudospherical surfaces and the sine-Gordon equation
293
The Maurer-Cartan forms for such a frame ļ¬eld must satisfy equation (9.6); in fact (as we will see in Ā§9.5), equation (9.6) is equivalent to an overdetermined system of partial diļ¬erential equations for the frame ļ¬eld (e1 (u), e2 (u), e3 (u)), and the compatibility condition for this system is precisely the 2 condition that Ī£ has Gauss curvature K = ā sinr2(Ī±) . The initial condition e1 (u0 ) = e0 ā Tx0 Ī£ then determines the desired frame ļ¬eld uniquely, and 8 deļ¬ned by the remainder of the proof consists of showing that the surface Ī£ (9.10) satisļ¬es all the desired conditions. The proof of part (2) of Theorem 9.5 shows how, given a surface of constant negative Gauss curvature K, one can construct new pseudospherical surfaces 8 The 2-parameter family of such surfaces alluded to earlier arises as Ī£. follows: One parameter comes from the choice of constants r, Ī± such that 2 K = ā sinr2(Ī±) , and one comes from the choice of a non-principal unit vector e0 ā Tx0 Ī£. The choice of r, Ī± determines the coeļ¬cients of the PDE system (9.6), while the choice of e0 determines the initial conditions that give rise to a particular solution of this PDE system.
9.4. Pseudospherical surfaces and the sine-Gordon equation Let Ī£ be a pseudospherical surface, and for simplicity assume that its Gauss curvature is K = ā1. Since the Gauss curvature of Ī£ is negative, Ī£ cannot have any umbilic points; consequently, it can be shown (for a proof, see [dC76]) that every point x ā Ī£ has a neighborhood for which there exists a local parametrization x : U ā E3 of Ī£ whose coordinate curves are principal curves in Ī£ (cf. Exercises 4.24 and 4.27). As in Exercise 4.24, we can then choose the adapted orthonormal frame ļ¬eld 1 1 e1 (u) = ā xu , e2 (u) = ā xv , e3 (u) = e1 (u) Ć e2 (u) E G along Ī£. (Note that, unlike in Ā§9.3, we are only considering a single pseudospherical surface, so there is no line congruence to take into consideration when choosing an adapted frame ļ¬eld.) Then, as we saw in Exercises 4.24 and 4.27, the coeļ¬cients of the ļ¬rst and second fundamental forms satisfy F = f = 0, and the associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ) are given by ā ā Ļ ĀÆ 1 = E du, Ļ ĀÆ 2 = G dv, ā ā e g Ļ ĀÆ 13 = ā du = Īŗ1 E du, Ļ ĀÆ 23 = ā dv = Īŗ2 G dv, (9.11) E G 1 Ļ ĀÆ 21 = ā (Ev du ā Gu dv), 2 EG where Īŗ1 = Ee , Īŗ2 = Gg are the principal curvatures of Ī£.
294
9. Pseudospherical surfaces and BĀØacklundās theorem
The following exercises will show how the surface Ī£ = x(U ) gives rise to a solution of the sine-Gordon equation (9.1). First, we investigate the Gauss and Codazzi equations for Ī£ and show that the parametrization x : U ā E3 can be ļ¬ne-tuned to arrange that the ļ¬rst and second fundamental forms of Ī£ can be expressed nicely in terms of a single function Ļ : U ā R. *Exercise 9.11. (a) Show that the Codazzi equations of Exercise 4.41(f) can be written in the form ā ! ā ! āĪŗ1 ā āĪŗ2 ā (9.12) = (Īŗ2 ā Īŗ1 ) ln( E) , = (Īŗ1 ā Īŗ2 ) ln( G) . āv āv āu āu (b) Divide equations (9.12) by (Īŗ1 ā Īŗ2 ), multiply the left-hand sides by ĪŗĪŗ11 and ĪŗĪŗ22 , respectively, and use the Gauss equation Īŗ1 Īŗ2 = ā1 to show that ā ! ā ā ln(Īŗ21 + 1) = ā2 ln( E) , āv āv (9.13) ā ! ā ā ln(Īŗ22 + 1) = ā2 ln( G) . āu āu (c) Integrate equations (9.13) and conclude that c1 (u) , E for some functions c1 (u), c2 (v) > 0.
(9.14)
Īŗ21 + 1 =
Īŗ22 + 1 =
c2 (v) G
(d) Show that under a change of coordinates of the form (9.15)
u Ė = h(u),
vĖ = k(v),
the coeļ¬cients of the ļ¬rst fundamental form with respect to the coordinates (Ė u, vĖ) become 1 1 Ė= Ė= E E, G G. 2 (h (u)) (k (v))2 (The functions Īŗ1 , Īŗ2 , however, are invariants and are unchanged by the coordinate transformation.) Conclude that the functions h(u), k(v) can be chosen so as to arrange that cĖ1 (Ė u) = cĖ2 (Ė v ) = 1. (e) Now, assume that the coordinate functions (u, v) (without the tildes) have been chosen so that c1 (u) = c2 (v) = 1. Show that there exists a function Ļ(u, v) such that Īŗ1 = tan(Ļ), 2
E = cos (Ļ),
Īŗ2 = ā cot(Ļ), G = sin2 (Ļ).
9.4. Pseudospherical surfaces and the sine-Gordon equation
295
Thus, the ļ¬rst and second fundamental forms of Ī£ are I = cos2 (Ļ) du2 + sin2 (Ļ) dv 2 , II = sin(Ļ) cos(Ļ) (du2 ā dv 2 ). (Hint: Recall that Īŗ1 Īŗ2 = ā1 and that the tangent and cotangent functions are surjective onto R.) Next, we give a geometric interpretation of the function Ļ and show that the function Ļ = 2Ļ is a solution of the sine-Gordon equation (9.1). *Exercise 9.12. (a) Show that the angle between the asymptotic directions at any point x(u, v) is equal to 2Ļ(u, v). (Hint: The asymptotic directions are the null directions for the second fundamental form; in this case, they are represented by the tangent vectors xu Ā± xv .) (b) Show that Ļ ĀÆ 21 (cf. equation (9.11)) is given by Ļ ĀÆ 21 = āĻv du ā Ļu dv. (c) Use the Gauss equation dĀÆ Ļ21 = K Ļ ĀÆ1 ā§ Ļ ĀÆ2 to show that the function Ļ satisfes the PDE (9.16)
Ļuu ā Ļvv = sin(Ļ) cos(Ļ).
(d) Let Ļ = 2Ļ, so that Ļ(u, v) is the angle between the asymptotic directions at the point x(u, v). Show that equation (9.16) is equivalent to the PDE (9.17)
Ļuu ā Ļvv = sin(Ļ)
for the function Ļ. (e) Consider the change of coordinates 1 1 x = (u + v), y = (u ā v). 2 2 Show that in terms of the (x, y)-coordinates, the ļ¬rst and second fundamental forms of Ī£ are given by (9.18)
I = dx2 + 2 cos(2Ļ) dx dy + dy 2 , II = 2 sin(2Ļ) dx dy.
Note that the x- and y-coordinate directions are now the asymptotic directions at each point of Ī£; for this reason, (x, y) are called asymptotic coordinates on Ī£, and the corresponding parametrization is called an asymptotic parametrization of Ī£. (f) Show that equation (9.17) is equivalent to equation (9.1).
296
9. Pseudospherical surfaces and BĀØacklundās theorem
Remark 9.13. Equations (9.17) and (9.1) are both referred to as the sineGordon equation. The local coordinates (u, v) of equation (9.17) are called space-time coordinates because the left-hand side has the same form as the wave equation Ļuu ā Ļvv = 0, where u is often thought of as a time coordinate and v as a spatial coordinate. The local coordinates (x, y) of equation (9.1), on the other hand, are called null or characteristic coordinates. The term ācharacteristicā comes from the fact that the x- and y-coordinate curves are the characteristic curves for the PDE (9.1), while the term ānullā arises from thinking of the (u, v)-plane as the Minkowski space M1,1 with its standard metric, for which the x- and y-coordinate curves are the null lines. We have shown that, at least locally, any pseudospherical surface Ī£ determines a solution Ļ of the sine-Gordon equation (9.1). The following exercise shows that the converse is true as well. *Exercise 9.14. Let Ļ : U ā R be any solution of the sine-Gordon equation (9.1), and let Ļ(x, y) = 12 Ļ(x, y). Deļ¬ne 1-forms (ĀÆ Ļ1, Ļ ĀÆ 2, Ļ ĀÆ 13 , Ļ ĀÆ 23 , Ļ ĀÆ 21 ) by Ļ ĀÆ 1 = cos(Ļ)(dx + dy), (9.19)
Ļ ĀÆ 13 = sin(Ļ)(dx + dy),
Ļ ĀÆ 2 = sin(Ļ)(dx ā dy), Ļ ĀÆ 23 = ā cos(Ļ)(dx ā dy),
Ļ ĀÆ 21 = āĻx dx + Ļy dy. (a) Check that the ļ¬rst and second fundamental forms I = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 , II = Ļ ĀÆ 13 Ļ ĀÆ1 + Ļ ĀÆ 23 Ļ ĀÆ2 agree with those in equations (9.18). (b) Check that these 1-forms (together with Ļ ĀÆ 3 = 0) satisfy the Cartan structure equations (3.8). (Hint: This result depends on the fact that Ļ is a solution of equation (9.1).) (c) Conclude from Bonnetās theorem (cf. Theorem 4.39) that there exists an immersed surface x : U ā E3 whose ļ¬rst and second fundamental forms are given by (9.18). In particular, the Gauss curvature of the surface Ī£ = x(U ) is K = ā1, and the angle between the asymptotic directions at the point x(x, y) is equal to Ļ(x, y). Remark 9.15. In recent years, many other integrable systems have been shown to be connected with pseudospherical geometry, and this connection
9.5. BĀØacklund transformation for the sine-Gordon equation
297
provides an important tool for studying the space of solutions to these equations. This connection was ļ¬rst considered by Chern and Tenenblat [CT86] and further explored by Reyes [Rey98] and others.
9.5. The BĀØ acklund transformation for the sine-Gordon equation In this section, we will see how the geometric BĀØ acklund transformation between pseudospherical surfaces gives rise to a corresponding analytic transformation between solutions of the sine-Gordon equation (9.1). Suppose that we have a BĀØacklund transformation between two pseudospher8 of Gauss curvature K = ā1. (Note that the condition ical surfaces Ī£, Ī£ K = ā1 implies that r = sin(Ī±).) Let (e1 (u), e2 (u), e3 (u)) be the orthonormal frame ļ¬eld on Ī£ adapted to the BĀØ acklund transformation as in Ā§9.3, and let (e1 (u), e2 (u), e3 (u)) be the principal adapted frame ļ¬eld on Ī£ as in Ā§9.4. Let Ī·(u) denote the angle between e1 (u) and e1 (u). The following exercise shows how the function Ī· is related to the function Ļ of Ā§9.4. *Exercise 9.16. (a) Show that
(9.20)
e1 (u) e2 (u) = e1 (u) e2 (u)
cos(Ī·(u)) āsin(Ī·(u))
.
sin(Ī·(u)) cos(Ī·(u))
(b) Show that 1 1 Ļ ĀÆ cos(Ī·) sin(Ī·) Ļ ĀÆ cos(Ļ ā Ī·) cos(Ļ + Ī·) dx = = , āsin(Ī·) cos(Ī·) Ļ sin(Ļ ā Ī·) āsin(Ļ + Ī·) dy Ļ ĀÆ2 ĀÆ2 (9.21)
Ļ ĀÆ 31
Ļ ĀÆ 32
=
cos(Ī·) sin(Ī·)
āsin(Ī·) cos(Ī·)
Ļ ĀÆ 13 Ļ ĀÆ 23
=
sin(Ļ ā Ī·) sin(Ļ + Ī·)
ācos(Ļ ā Ī·) cos(Ļ + Ī·)
dx , dy
Ļ ĀÆ 12 = Ļ ĀÆ 21 ā dĪ· = ā(Ļx + Ī·x ) dx + (Ļy ā Ī·y ) dy. (Hint: Cf. Exercises 4.28 and 4.47 and equations (9.19).) (c) Show that the BĀØacklund equation (9.6) is equivalent to the ļ¬rst-order system of partial diļ¬erential equations (9.22)
Ļx + Ī·x = Ī» sin(Ļ ā Ī·), 1 Ļy ā Ī·y = sin(Ļ + Ī·), Ī»
298
9. Pseudospherical surfaces and BĀØacklundās theorem
where Ī» = cot(Ī±) ā csc(Ī±). (Hint: Recall that, since K = ā1, we have r = sin(Ī±), and you will need the trigonometric identity ā(cot(Ī±) + csc(Ī±)) =
1 .) (cot(Ī±) ā csc(Ī±))
The system (9.22) is a coupled system of partial diļ¬erential equations for the pair of functions (Ļ(x, y), Ī·(x, y)). The following exercise explores some properties of solutions of this system. *Exercise 9.17. (a) Suppose that the pair of functions (Ļ(x, y), Ī·(x, y)) satisļ¬es the PDE system (9.22), where Ī» is any nonzero constant. Show that the functions 2Ļ, 2Ī· must each be solutions of the sine-Gordon equation (9.1). (Hint: Diļ¬erentiate the ļ¬rst equation in (9.22) with respect to y and the second equation with respect to x; then add and subtract the resulting equations and apply trigonometric identities to simplify the right-hand sides.) (b) Now, suppose that 2Ļ is any known solution of (9.1). Then the system (9.22) can be regarded as the overdetermined PDE system Ī·x = āĻx + Ī» sin(Ļ ā Ī·), 1 Ī·y = Ļy ā sin(Ļ + Ī·) Ī» for the unknown function Ī·(x, y). Show that this system is compatible, i.e., that (Ī·x )y = (Ī·y )x āprecisely because 2Ļ satisļ¬es (9.1). It follows that this system has a 1-parameter family of solutions Ī·(x, y) and that these solutions can be constructed using only techniques of ordinary diļ¬erential equations. The PDE system (9.22) is called a BĀØ acklund transformation for the sineGordon equation (9.1); the construction in Exercise 9.17 is the analog of part (2) of Theorem 9.5. It shows how, given one solution 2Ļ of the sineGordon equation (9.1), the PDE system (9.22) can be used to construct a 2-parameter family of new solutions 2Ī·: One parameter comes from the choice of the constant Ī» = 0, and one comes from the 1-parameter family of solutions Ī· to the system (9.22). At this point, we have established several relationships involving pseudospherical surfaces and solutions of the sine-Gordon equation: (1) Every pseudospherical surface Ī£ of Gauss curvature K = ā1 is associated with a solution 2Ļ to the sine-Gordon equation (9.1) that describes the angle between its asymptotic directions at each point (and vice versa).
9.5. BĀØacklund transformation for the sine-Gordon equation
299
(2) Every pseudospherical surface Ī£ of Gauss curvature K = ā1 has a 8 gener2-parameter family of associated pseudospherical surfaces Ī£ ated by the construction in part (2) of Theorem 9.5. (3) Every solution 2Ļ to the sine-Gordon equation (9.1) has a 2-parameter family of associated solutions 2Ī· to (9.1) generated by the PDE system (9.22). It seems natural to ask: Given a pseudospherical surface Ī£ with associated 8 solution 2Ļ to (9.1), what is the relationship between the new surfaces Ī£ associated to Ī£ and the new solutions 2Ī· to (9.1) associated to 2Ļ? The following exercise answers this question. *Exercise 9.18. Let K = ā1 (so that r = sin(Ī±)), and let (ĀÆ Ļi , Ļ ĀÆ ij ) be as in Exercise 9.16. ĖĀÆ 1 , (a) Use equations (9.5), (9.7) to show that the Maurer-Cartan forms (Ļ 8 are given by ĖĀÆ 2 , Ļ ĖĀÆ 13 , Ļ ĖĀÆ 23 ) on Ī£ Ļ ĖĀÆ 1 = cos(Ļ ā Ī·) dx + cos(Ļ + Ī·) dy, Ļ (9.23)
ĖĀÆ 2 = sin(Ļ ā Ī·) dx + sin(Ļ + Ī·) dy, Ļ ĖĀÆ 13 = sin(Ļ ā Ī·) dx ā sin(Ļ + Ī·) dy, Ļ ĖĀÆ 23 = ā cos(Ļ ā Ī·) dx + cos(Ļ + Ī·) dy. Ļ
8 are (b) Show that the ļ¬rst and second fundamental forms of Ī£ (9.24)
ĖĀÆ 1 )2 + (Ļ ĖĀÆ 2 )2 = dx2 + 2 cos(2Ī·) dx dy + dy 2 , I = (Ļ ĖĀÆ 13 Ļ ĖĀÆ 1 + Ļ ĖĀÆ 2 = ā2 sin(2Ī·) dx dy. ĖĀÆ 23 Ļ II = Ļ
8 =x Ė (U ) (c) Conclude that the angle between the asymptotic directions of Ī£ Ė (x, y) is equal to 2Ī·(x, y). (The sign of the second fundamental at the point x form is unimportant, as it can be reversed by reversing the orientation of 8 Ī£.) The result of Exercise 9.18 completes the picture: The analytic BĀØ acklund transformation (9.22) between solutions of the sine-Gordon equation (9.1) is the precise analog of the geometric BĀØ acklund transformation between pseudospherical surfaces of Gauss curvature K = ā1. Exercise 9.19. Let 2Ļ(x, y) ā” 0 be the trivial solution of the sine-Gordon equation (9.1). (This solution corresponds to the degenerate āsurfaceā consisting of a straight line in E3 .) Show that the corresponding solutions 2Ī·(x, y) generated by the BĀØacklund transformation (9.22) are given by Ī·(x, y) = 2 tanā1 (Ceā(Ī»x+ Ī» y) ), 1
300
9. Pseudospherical surfaces and BĀØacklundās theorem
where C = 0 is constant. (Hint: You may ļ¬nd the trig identity csc(Ī·) + cot(Ī·) = cot( 12 Ī·) useful.) The functions 2Ī· = 4 tanā1 (Ceā(Ī»x+ Ī» y) ) 1
(9.25)
are called the 1-soliton solutions of the sine-Gordon equation. Iterating this procedure gives the 2-solitons, etc. Exercise 9.20. Let Ī£ be the pseudosphere, with parametrization ā”
x + y ā tanh(x + y)
ā¤
ā¢ ā„ ā„ x(x, y) = ā¢ ā£sech(x + y) cos(x ā y)ā¦ . sech(x + y) sin(x ā y) (a) Show that the ļ¬rst and second fundamental forms of Ī£ are given by I = dx2 +
2(cosh2 (x + y) ā 2) dx dy + dy 2 , 2 cosh (x + y)
II =
4 sinh(x + y) dx dy. cosh2 (x + y)
(b) Conclude that: (1) (x, y) are asymptotic coordinates for Ī£. (2) The Gauss curvature of Ī£ is K = ā1. (3) The angle 2Ļ(x, y) between the asymptotic directions of Ī£ at the point x(x, y) satisļ¬es the conditions (9.26)
cos(2Ļ(x, y)) =
cosh2 (x + y) ā 2 2 sinh(x + y) , sin(2Ļ(x, y)) = . 2 cosh (x + y) cosh2 (x + y)
(In particular, check that "
cosh2 (x + y) ā 2 cosh2 (x + y)
"
#2 +
2 sinh(x + y) cosh2 (x + y)
#2 = 1.)
(c) Show that the function Ļ(x, y) deļ¬ned by equation (9.26) is Ļ(x, y) = 2 tanā1 (ex+y ).
9.5. BĀØacklund transformation for the sine-Gordon equation
301
Thus, the pseudosphere is one of the family of pseudospherical surfaces that arise from the 1-soliton solutions (9.25) of the sine-Gordon equation (9.1). Surfaces in this family are sometimes referred to as ā1-solitonā pseudospherical surfaces. Exercise 9.21. In this exercise, we will give a proof of Hilbertās theorem, which states that there is no isometric immersion of the complete hyperbolic plane H2 into the Euclidean space E3 . (For purposes of this exercise, all you need to know about H2 is that it is a complete surface of inļ¬nite area, diļ¬eomorphic to R2 , with constant Gauss curvature K = ā1. We will explore hyperbolic spaces in more detail in Chapter 11.) Suppose that x : H2 ā E3 is an isometric immersion. Recall that H2 has Gauss curvature K = ā1; therefore the image Ī£ = x(H2 ) is an immersed pseudospherical surface. (āImmersedā allows for the possibility of self-intersection; an immersion x : H2 ā E3 must be a regular mapping at every point of the domain H2 , but it need not necessarily be one-to-one.) Moreover, since H2 is a complete surface of inļ¬nite area, the same must be true of Ī£. (a) Show that, because x is an immersion, there exist global coordinates (x, y) on H2 such that x(x, y) is an asymptotic parametrization of Ī£ and for which the ļ¬rst and second fundamental forms of Ī£ are as in equations (9.18). (Hint: We already know that local coordinates (x, y) satisfying these conditions exist in a neighborhood of each point in H2 . Moreover, these local coordinates are determined only up to additive constants. Show that for any two such overlapping coordinate patches with local coordinates (x, y) and (Ė x, yĖ), respectively, we must have x Ė = x + x0 ,
yĖ = y + y0
for some constants x0 , y0 . Since each pair of coordinates is only determined up to additive constants, we can set x Ė = x, yĖ = y. A topological argument shows that this patching construction works globally because H2 is simply connected.) (b) Let 2Ļ : H2 ā R be the solution of the sine-Gordon equation (9.1) associated to Ī£. Show that, because x is an immersion, the function Ļ must satisfy Ļ 0 < Ļ(x, y) < 2 2 for all (x, y) ā H . (Hint: An immersed pseudospherical surface must have linearly independent asymptotic directions at each point.) (c) Let R ā H2 be a rectangle of the form a ā¤ x ā¤ b,
c ā¤ y ā¤ d.
302
9. Pseudospherical surfaces and BĀØacklundās theorem
Use the sine-Gordon equation (9.1) and the Fundamental Theorem of Calculus to show that the area of x(R) ā Ī£ is given by & A(x(R)) = Ļ ĀÆ1 ā§ Ļ ĀÆ2 R
= 2 (Ļ(b, d) ā Ļ(b, c) ā Ļ(a, d) + Ļ(a, c)) . Conclude from part (b) that A(x(R)) < 2Ļ. (d) Observe that, because H2 is complete and has inļ¬nite area, the area of R can be made arbitrarily large by choosing a, b, c, d appropriately. Therefore, since x is an isometric immersion, the area of x(R) can be made arbitrarily large as well. This contradicts the result of part (c); thus, no such isometric immersion can exist. Exercise 9.22. While the BĀØ acklund transformation (9.22) relates two different solutions of the same PDE (9.1), it is also possible for a BĀØ acklund transformation to relate solutions of two diļ¬erent PDEs. For example, consider the ļ¬rst-order system of partial diļ¬erential equations (ĻāĪ·)
(9.27)
Ļx + Ī·x = 2Ī»e 2 , 1 (Ļ+Ī·) Ļy ā Ī·y = e 2 . Ī»
(a) Suppose that the pair of functions (Ļ(x, y), Ī·(x, y)) satisļ¬es the PDE system (9.27), where Ī» is any nonzero constant. Show that Ī·(x, y) must be a solution of the wave equation (in characteristic coordinates) (9.28)
Ī·xy = 0,
while Ļ(x, y) must be a solution of Liouvilleās equation (9.29)
Ļxy = eĻ .
(b) Show by integration that the general solution of the wave equation (9.28) is (9.30)
Ī·(x, y) = Ļ(x) + Ļ(y),
where Ļ(x), Ļ(y) are arbitrary functions of a single variable. (c) Substitute equation (9.30) into the system (9.27) to obtain a system of two ļ¬rst-order PDEs for the function Ļ(x, y). By treating each of these equations as a separable ODE in the appropriate variable, show that " & # & 1 1 ā 21 Ļ (Ļ(x)āĻ(y)) āĻ(x) Ļ(y) (9.31) e Ī» e = āe 2 dx + dy . e 2Ī»
9.6. Maple computations
303
(Hint: Write the right-hand sides of equations (9.27) as 1 1 1 (ĻāĻ(y)+Ļ(x)+2Ļ(y)) 2Ī»e 2 (Ļ+Ļ(x)ā2Ļ(x)āĻ(y)) and .) e2 Ī» (d) Set
&
& 1 X(x) = āĪ» e dx, Y (y) = ā eĻ(y) dy. 2Ī» Solve equation (9.31) for Ļ to obtain the general solution of Liouvilleās equation: " # 2X (x)Y (y) (9.32) Ļ(x, y) = ln , (X(x) + Y (y))2 āĻ(x)
where X(x) and Y (y) are arbitrary functions of a single variable, with the property that X (x) and Y (y) are nonzero and have the same sign.
9.6. Maple computations As usual, begin by loading the Cartan and LinearAlgebra packages into Maple. Since we have three diļ¬erent adapted frame ļ¬elds in play in 8 adapted to the this chapter (one frame ļ¬eld on each of the surfaces Ī£, Ī£ BĀØacklund transformation and one principal adapted frame ļ¬eld on Ī£), weāll need to keep track of three diļ¬erent sets of Maurer-Cartan forms. In order to distinguish them a bit, weāll use (theta1[i], theta1[i,j]) and 8 respec(theta2[i], theta2[i,j]) for the Maurer-Cartan forms on Ī£, Ī£, tively, corresponding to the BĀØ acklund-adapted frame ļ¬elds, and (omega[i], omega[i,j]) for the Maurer-Cartan forms corresponding to the principal frame ļ¬eld on Ī£. Start by declaring the necessary 1-forms and telling Maple about their symmetries and structure equations: > Form(omega[1], omega[2], omega[3]); Form(omega[1,2], omega[3,1], omega[3,2]); Form(theta1[1], theta1[2], theta1[3]); Form(theta1[1,2], theta1[3,1], theta1[3,2]); Form(theta2[1], theta2[2], theta2[3]); Form(theta2[1,2], theta2[3,1], theta2[3,2]); > omega[1,1]:= omega[2,2]:= omega[3,3]:= omega[2,1]:= omega[1,3]:= omega[2,3]:=
0; 0; 0; -omega[1,2]; -omega[3,1]; -omega[3,2];
304
9. Pseudospherical surfaces and BĀØacklundās theorem
theta1[1,1]:= theta1[2,2]:= theta1[3,3]:= theta1[2,1]:= theta1[1,3]:= theta1[2,3]:=
0; 0; 0; -theta1[1,2]; -theta1[3,1]; -theta1[3,2];
theta2[1,1]:= theta2[2,2]:= theta2[3,3]:= theta2[2,1]:= theta2[1,3]:= theta2[2,3]:=
0; 0; 0; -theta2[1,2]; -theta2[3,1]; -theta2[3,2];
> for i from 1 to d(omega[i]):= end do; d(omega[1,2]):= d(omega[3,1]):= d(omega[3,2]):=
3 do -add(āomega[i,j] &Ė omega[j]ā, j=1..3); -add(āomega[1,k] &Ė omega[k,2]ā, k=1..3); -add(āomega[3,k] &Ė omega[k,1]ā, k=1..3); -add(āomega[3,k] &Ė omega[k,2]ā, k=1..3);
for i from 1 to 3 do d(theta1[i]):= -add(ātheta1[i,j] end do; d(theta1[1,2]):= -add(ātheta1[1,k] d(theta1[3,1]):= -add(ātheta1[3,k] d(theta1[3,2]):= -add(ātheta1[3,k] for i from 1 to 3 do d(theta2[i]):= -add(ātheta2[i,j] end do; d(theta2[1,2]):= -add(ātheta2[1,k] d(theta2[3,1]):= -add(ātheta2[3,k] d(theta2[3,2]):= -add(ātheta2[3,k]
&Ė theta1[j]ā, j=1..3); &Ė theta1[k,2]ā, k=1..3); &Ė theta1[k,1]ā, k=1..3); &Ė theta1[k,2]ā, k=1..3); &Ė theta2[j]ā, j=1..3); &Ė theta2[k,2]ā, k=1..3); &Ė theta2[k,1]ā, k=1..3); &Ė theta2[k,2]ā, k=1..3);
Weāll also need the vector form of the structure equations; weāll use x1 and 8 respectively, and (e11, e12, e13), x2 for the parametrizations of Ī£ and Ī£, (e21, e22, e23) for the frame ļ¬elds on the two surfaces that are adapted to the BĀØ acklund transformation. > d(x1):= e11*theta1[1] + e12*theta1[2]; d(e11):= e12*theta1[2,1] + e13*theta1[3,1]; d(e12):= e11*theta1[1,2] + e13*theta1[3,2]; d(e13):= e11*theta1[1,3] + e12*theta1[2,3];
9.6. Maple computations
305
d(x2):= e21*theta2[1] + e22*theta2[2]; d(e21):= e22*theta2[2,1] + e23*theta2[3,1]; d(e22):= e21*theta2[1,2] + e23*theta2[3,2]; d(e23):= e21*theta2[1,3] + e22*theta2[2,3]; Exercise 9.8: Now, assume that we have chosen adapted frame ļ¬elds 8 as in equations Ė2 (u), e Ė3 (u)) on Ī£ (e1 (u), e2 (u), e3 (u)) on Ī£ and (Ė e1 (u), e (9.2) and that equation (9.3) holds. First, declare Ī± and r to be constants, and set up a substitution to move from one frame ļ¬eld to the other: > Form(r=-1, alpha=-1); > frame2to1sub:= [e21 = e11, e22 = cos(alpha)*e12 + sin(alpha)*e13, e23 = -sin(alpha)*e12 + cos(alpha)*e13]; Now diļ¬erentiate equation (9.3), and write the results in terms of the ļ¬rst adapted frame ļ¬eld: > zero1:= Simf(subs(frame2to1sub, d(x2 - x1 - r*e11))); Collect terms, and then use the fact that (e1 (u), e2 (u), e3 (u)) are linearly independent to conclude that each of their coeļ¬cients must be zero: > collect(zero1, {e11, e12, e13}); > zero1a:= coeff(zero1, e11); zero1b:= coeff(zero1, e12); zero1c:= coeff(zero1, e13); This computation yields equations (9.5), and we can combine the last two equations to obtain a relation that only involves the Maurer-Cartan forms on Ī£: > zero1d:= Simf(zero1b - cot(alpha)*zero1c); The result yields equation (9.6). Next, we need to compute the 1-forms ĖĀÆ 13 = dĖ Ė3 , Ļ e1 , e
ĖĀÆ 23 = dĖ Ė3 . Ļ e2 , e
Since we havenāt told Maple that (e11, e12, e13) are vectors, weāll have to deļ¬ne our own procedure to compute these inner products. We can do this as follows: > innprod1:= proc(exp1, exp2) RETURN(coeff(exp1, e11)*coeff(exp2, e11) + coeff(exp1, e12)*coeff(exp2, e12)
306
9. Pseudospherical surfaces and BĀØacklundās theorem
+ coeff(exp1, e13)*coeff(exp2, e13)); end proc; ĖĀÆ 13 and Ļ ĖĀÆ 23 via the formulas above. First, compute Ļ ĖĀÆ 13 : Now we can compute Ļ > Simf(innprod1(d(Simf(subs(frame2to1sub, e21))), Simf(subs(frame2to1sub, e23)))); Īø11,2 sin(Ī±) + Īø13,1 cos(Ī±) Apply the relation (9.6): > Simf(subs(solve({zero1d}, {theta1[3,1]}), %)); This yields the expression in the ļ¬rst equation of (9.7). Similarly, the computation > Simf(innprod1(d(Simf(subs(frame2to1sub, e22))), Simf(subs(frame2to1sub, e23)))); yields the expression in the second equation of (9.7). Exercise 9.14: Set up two substitutions: one to tell Maple that the function Ļ(x, y) = 2Ļ(x, y) satisļ¬es the sine-Gordon equation (9.1) and one to deļ¬ne the desired Maurer-Cartan forms associated to the principal adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)): > PDETools[declare](psi(x,y)); > SGEsub:= [diff(psi(x,y), x, y) = (1/2)*sin(2*psi(x,y))]; > SGEformsub:= [omega[1] = cos(psi(x,y))*(d(x) + d(y)), omega[2] = sin(psi(x,y))*(d(x) - d(y)), omega[3] = 0, omega[3,1] = sin(psi(x,y))*(d(x) + d(y)), omega[3,2] = -cos(psi(x,y))*(d(x) - d(y)), omega[1,2] = -diff(psi(x,y), x)*d(x) + diff(psi(x,y), y)*d(y)]; Maple doesnāt really know how to compute symmetric products of diļ¬erential forms, but the following commands will work for computing the ļ¬rst and second fundamental forms: > collect(simplify(Simf(subs(SGEformsub, omega[1]))Ė2 + Simf(subs(SGEformsub, omega[2]))Ė2), {d(x), d(y)}); > collect(simplify(Simf(subs(SGEformsub, omega[3,1]))* Simf(subs(SGEformsub, omega[1])) + Simf(subs(SGEformsub, omega[3,2]))* Simf(subs(SGEformsub, omega[2]))), {d(x), d(y)});
9.6. Maple computations
307
In order to verify that the structure equations are satisļ¬ed, check that computations such as the following all yield zero: > Simf(d(Simf(subs(SGEformsub, omega[1])))) - Simf(subs(SGEformsub, Simf(d(omega[1])))); The only one that doesnāt immediately reduce to zero is the one for Ļ ĀÆ 21 , which requires an application of SGEsub in order to see that it vanishes. Exercise 9.16: Equation (9.20) is immediate from the deļ¬nition of Ī·(u), and the results of Chapter 4 yield the expressions for (ĀÆ Ļ1 , Ļ ĀÆ 2, Ļ ĀÆ 31 , Ļ ĀÆ 32 ) in equations (9.21). Set up a substitution to go back and forth between the two sets of Maurer-Cartan forms: > PDETools[declare](eta(x,y)); > rotationsub:= [ theta1[1] = cos(eta(x,y))*omega[1] + sin(eta(x,y))*omega[2], theta1[2] = -sin(eta(x,y))*omega[1] + cos(eta(x,y))*omega[2], theta1[3,1] = cos(eta(x,y))*omega[3,1] + sin(eta(x,y))*omega[3,2], theta1[3,2] = -sin(eta(x,y))*omega[3,1] + cos(eta(x,y))*omega[3,2]]; > rotationbacksub:= makebacksub(rotationsub); We can then compute Ļ ĀÆ 12 in terms of Ļ ĀÆ 21 as follows: Compute dĀÆ Ļ 1 and dĀÆ Ļ2 by diļ¬erentiating the expressions for Ļ ĀÆ 1 and Ļ ĀÆ 2 in equation (9.21), and then use the reverse substitution to express the result in terms of (ĀÆ Ļ1 , Ļ ĀÆ 2 ): > dtheta1[1]:= Simf(subs([omega[3]=0], Simf(subs(rotationbacksub, Simf(d(Simf(subs(rotationsub, theta1[1])))))))); > pick(dtheta1[1], theta1[2]); āĻ1,2 + Ī·x d(x) + Ī·y d(y) > dtheta1[2]:= Simf(subs([omega[3]=0], Simf(subs(rotationbacksub, Simf(d(Simf(subs(rotationsub, theta1[2])))))))); > pick(dtheta1[2], theta1[1]); Ļ1,2 ā Ī·x d(x) ā Ī·y d(y) It follows from the structure equations for dĀÆ Ļ 1 and dĀÆ Ļ 2 that Ļ ĀÆ 12 = Ļ ĀÆ 21 ā dĪ·. Add this to our substitution: > rotationsub:= [op(rotationsub), theta1[1,2] = omega[1,2] - d(eta(x,y))];
308
9. Pseudospherical surfaces and BĀØacklundās theorem
We can combine this with the expressions for (ĀÆ Ļi, Ļ ĀÆ ji ) in SGEformsub in order to write (ĀÆ Ļi , Ļ ĀÆ ij ) in terms of (dx, dy): > SGErotationsub:= Simf(subs(SGEformsub, rotationsub)); Since K = ā1, we have r = sin(Ī±): > r:= sin(alpha); Now consider the BĀØacklund equation (9.6), which says that the following expression is zero: > Backlundeq:= theta1[2] + r*theta1[2,1] - r*cot(alpha)*theta1[3,1]; We can express this equation in terms of (dx, dy) via SGErotationsub: > Backlundzero:= Simf(subs(SGErotationsub, Backlundeq)); The coeļ¬cients of dx and dy in this expression are PDEs involving the functions Ļ and Ī·: > PDE1:= pick(Backlundzero, d(x)); PDE2:= pick(Backlundzero, d(y)); These PDEs contain some coeļ¬cients involving trigonometric functions of Ī±, and they will look nicer if we rename them. In the ļ¬rst equation, divide by sin(Ī±) and set Ī» = cot(Ī±) ā csc(Ī±) via the substitution: > lambdasub:= [cos(alpha) = lambda*sin(alpha) + 1]; > Simf(subs(lambdasub, PDE1/sin(alpha))); āĪ» sin(Ļ) cos(Ī·) + Ī» sin(Ī·) cos(Ļ) + Ļx + Ī·x > combine(%, trig); Ī» sin(Ī· ā Ļ) + Ļx + Ī·x Similarly, for the second equation divide by sin(Ī±) and set Ī¼ = ā(cot(Ī±) + csc(Ī±)): > musub:= [cos(alpha) = -mu*sin(alpha) - 1]; > Simf(subs(musub, PDE2/sin(alpha))); Ī¼ sin(Ļ) cos(Ī·) + Ī¼ sin(Ī·) cos(Ļ) ā Ļy + Ī·y > combine(%, trig); Ī¼ sin(Ī· + Ļ) ā Ļy + Ī·y
9.6. Maple computations
309
All that remains is to show that Ī»Ī¼ = 1, so that Ī¼ = Ī»1 : > Simf(subs(solve({op(lambdasub), op(musub)}, {lambda, mu}), lambda*mu)); 1 Exercise 9.18: This exercise is now a simple matter of combining things ĖĀÆ i , Ļ ĖĀÆ ij ) in terms that we have already computed in order to write the forms (Ļ of (dx, dy). Equations (9.7) and the ļ¬rst and third equations in (9.5) allow ĖĀÆ 1 , Ļ ĖĀÆ 2 , Ļ ĖĀÆ 31 , Ļ ĖĀÆ 32 ) in terms of (ĀÆ us to write (Ļ Ļ1 , Ļ ĀÆ 2, Ļ ĀÆ 31 , Ļ ĀÆ 32 ) as follows: > theta2sub:= [op(solve({zero1a, zero1c}, {theta2[1], theta2[2]})), theta2[3,1] = theta1[2], theta2[3,2] = theta1[3,2]]; Then we apply SGErotationsub to express these forms in terms of (dx, dy): > theta2coordsub:= Simf(subs(SGErotationsub, theta2sub)); We can compactify these expressions a bit as follows: > map(combine, theta2coordsub, trig); This should yield the expressions in equation (9.23). We can compute the ļ¬rst and second fundamental forms in equation (9.24) as follows: > collect(simplify(Simf(subs(theta2coordsub, theta2[1]))Ė2 + Simf(subs(theta2coordsub, theta2[2]))Ė2), {d(x), d(y)}); > collect(simplify(Simf(subs(theta2coordsub, theta2[3,1]))* Simf(subs(theta2coordsub, theta2[1])) + Simf(subs(theta2coordsub, theta2[3,2]))* Simf(subs(theta2coordsub, theta2[2]))), {d(x), d(y)});
10.1090/gsm/178/10
Chapter 10
Two classical theorems
In this chapter, we will see how moving frames may be used to prove two classical results in diļ¬erential geometry: the classiļ¬cation of doubly ruled surfaces in R3 and the Cauchy-Crofton formula for the length of a curve in the Euclidean plane E2 .
10.1. Doubly ruled surfaces in R3 A regular surface Ī£ ā R3 is called ruled if Ī£ contains a straight line segment passing through each point x ā Ī£. Ruled surfaces have been the subject of much study in classical diļ¬erential geometry; see, e.g., [Eis60] or [Wil62]. A surface Ī£ is called doubly ruled if it can be realized as a ruled surface in two distinct ways, i.e., if Ī£ contains two linearly independent line segments passing through each point x ā Ī£. There are two well-known examples of non-planar doubly ruled surfaces in R3 : the hyperboloid of one sheet 2 3 (10.1) Ī£ = t[x, y, z] ā R3 | x2 + y 2 ā z 2 = r2 , where r > 0 is a positive constant, and the hyperbolic paraboloid 2 3 (10.2) Ī£ = t[x, y, z] ā R3 | z = xy . *Exercise 10.1. Recall that a regular surface Ī£ ā R3 is ruled if and only if every point x ā Ī£ has a neighborhood that can be given a parametrization of the form x(u, v) = Ī±(u) + vĪ²(u), where Ī± is a regular curve in R3 and Ī²(u) = 0. 311
312
10. Two classical theorems
(a) Show that the following maps x1 , x2 : R2 ā R3 are parametrizations of the hyperboloid (10.1): x1 (u, v) = t [r cos(u), r sin(u), 0] + v t[ā sin(u), cos(u), 1] , x2 (u, v) = t [r cos(u), r sin(u), 0] + v t[ā sin(u), cos(u), ā1] . Conclude that the hyperboloid (10.1) is doubly ruled. (The rulings described by these parametrizations are shown in Figure 10.1.)
Figure 10.1. Two families of rulings on the hyperboloid of one sheet
(b) Show that the following map x : R2 ā R3 is a parametrization of the hyperbolic paraboloid (10.2) that simultaneously realizes both families of rulings: x(u, v) = t [u, v, uv] . (Hint: Find curves Ī±(u), Ī²(u), Ī³(v), Ī“(v) in R3 such that x(u, v) = Ī±(u) + vĪ²(u) = Ī³(v) + uĪ“(v).) Conclude that the hyperbolic paraboloid (10.2) is doubly ruled. (The rulings described by this parametrization are shown in Figure 10.2.)
Figure 10.2. Two families of rulings on the hyperbolic paraboloid
You may have noticed that so far in this chapter, we have been describing surfaces as subsets of R3 without specifying any particular homogeneous
10.1. Doubly ruled surfaces in R3
313
space structure on R3 . The following exercise suggests such a structure that might be well-suited to our purposes: *Exercise 10.2. Show that the set of doubly ruled surfaces is invariant under the action of the equi-aļ¬ne group A(3). Speciļ¬cally, if Ī£ ā R3 is a doubly ruled surface and g ā A(3) is an equi-aļ¬ne transformation, then the surface g Ā· Ī£ ā R3 is also doubly ruled. Based on the result of Exercise 10.2, for the remainder of this section, we will regard R3 as the equi-aļ¬ne space A3 . The goal of this section is to prove the following theorem [HCV52]: Theorem 10.3. Any non-planar doubly ruled regular surface Ī£ ā A3 is equivalent via an equi-aļ¬ne transformation to (an open subset of ) either a hyperboloid of one sheet as in (10.1) or a hyperbolic paraboloid as in (10.2). Remark 10.4. We will use the theory developed in Chapter 6 to prove Theorem 10.3 as stated, but in fact more is true: (1) The set of doubly ruled surfaces is invariant under the action of the full aļ¬ne group without the equi-aļ¬ne restriction. Under this group action, the hyperboloids of one sheet (10.1) for all positive values of r are all equivalent, whereas diļ¬erent values of r give surfaces that are inequivalent under the action of the equi-aļ¬ne group. (2) If we regard R3 as an open subset of P3 via the identiļ¬cation
(x1 , x2 , x3 ) ā 1 : x1 : x2 : x3 , the set of doubly ruled surfaces is also invariant under the action of the group of projective transformations on P3 . Under this group action, the surfaces (10.1) and (10.2) are all equivalent to each other. To begin the proof, let U be an open set in R2 , and let x : U ā A3 be an immersion whose image is a regular surface Ī£ ā A3 that is doubly ruled. In order to apply the method of moving frames, it seems natural to choose a frame ļ¬eld (e1 (u), e2 (u), e3 (u)) along Ī£ with the property that for each u ā U , the vectors e1 (u) and e2 (u) are each tangent to one of the rulings passing through the point x(u). Here we see another reason to regard R3 as A3 (as opposed to, say, E3 ): Having the freedom to choose a unimodular frame ļ¬eld on Ī£ (as opposed to, say, an orthonormal frame ļ¬eld) allows us suļ¬cient ļ¬exibility to adapt our choice of frame ļ¬eld to the surface in precisely this way. So, in this context, a unimodular frame ļ¬eld
314
10. Two classical theorems
(e1 (u), e2 (u), e3 (u)) along Ī£ will be called 0-adapted if for each u ā U , the vectors e1 (u) and e2 (u) are each tangent to one of the rulings passing through the point x(u). (Note that this notion of ā0-adaptedā is more restrictive than that in Chapter 6 because here we have additional geometric information to guide our initial choice for an adapted frame ļ¬eld.) Exercise 10.5. Explain why orthonormal frame ļ¬elds would generally be too restrictive to allow for an adapted frame ļ¬eld satisfying this tangency condition. Recall that choosing such a unimodular frame ļ¬eld along Ī£ is equivalent to Ė : U ā A(3) deļ¬ned by choosing a lifting x Ė (u) = (x(u); e1 (u), e2 (u), e3 (u)) x and that the pullbacks (ĀÆ Ļi, Ļ ĀÆ ji ) to U of the Maurer-Cartan forms (Ļ i , Ļji ) on Ė satisfy the conditions that Ļ A(3) via the lifting x ĀÆ 3 = 0 and (10.3)
dx = e1 Ļ ĀÆ 1 + e2 Ļ ĀÆ 2, dei = e1 Ļ ĀÆ i1 + e2 Ļ ĀÆ i2 + e3 Ļ ĀÆ i3 ,
i = 1, 2, 3.
*Exercise 10.6. (a) Show that the condition that the vector ļ¬elds (e1 (u), e2 (u)) are tangent to the rulings implies that (10.4)
de1 (e1 ) ā” 0
mod e1 ,
de2 (e2 ) ā” 0
mod e2 .
(Hint: Recall that de1 (e1 ) computes the directional derivative of the vector ļ¬eld e1 (u) in the direction of e1 (u). What limitations are imposed on this derivative by the condition that the vector ļ¬eld e1 (u) is tangent to a straight line along this direction?) (b) Show that the condition (10.4) implies that (10.5)
de1 ā” 0 mod (e1 , Ļ ĀÆ 2 ),
de2 ā” 0
mod (e2 , Ļ ĀÆ 1 ).
Conclude from equations (10.3) and (10.5) that (10.6)
Ļ ĀÆ 12 = a212 Ļ ĀÆ 2,
Ļ ĀÆ 13 = a312 Ļ ĀÆ 2,
Ļ ĀÆ 21 = a121 Ļ ĀÆ 1,
Ļ ĀÆ 23 = a321 Ļ ĀÆ1
for some functions a121 , a212 , a312 , a321 on U . (c) Use the Cartan structure equation 0 = dĀÆ Ļ 3 = āĀÆ Ļ13 ā§ Ļ ĀÆ1 ā Ļ ĀÆ 23 ā§ Ļ ĀÆ2 to show that a321 = a312 .
10.1. Doubly ruled surfaces in R3
315
In order to minimize notational clutter, we will rename the (akij ) and write (10.7)
Ļ ĀÆ 12 = a Ļ ĀÆ 2,
Ļ ĀÆ 13 = c Ļ ĀÆ 2,
Ļ ĀÆ 21 = b Ļ ĀÆ 1,
Ļ ĀÆ 23 = c Ļ ĀÆ 1.
Now we will investigate how the functions a, b, c transform if we vary our choice of 0-adapted frame ļ¬eld on Ī£. Ė2 (u), e Ė3 (u)) be any two *Exercise 10.7. Let (e1 (u), e2 (u), e3 (u)), (Ė e1 (u), e 0-adapted frame ļ¬elds on Ī£, with associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ) i i ĖĀÆ , Ļ ĖĀÆ j ), respectively. For simplicity, assume that the vector ļ¬eld e Ė1 (u) and (Ļ is tangent to the same family of rulings as e1 (u), and similarly for the vector Ė2 (u). ļ¬elds e2 (u) and e (a) Show that
(10.8)
ā¤ ā” Ī» 1 0 r1 ā„
ā¢ ā„ Ė1 (u) e Ė2 (u) e Ė3 (u) = e1 (u) e2 (u) e3 (u) ā¢ e ā£ 0 Ī» 2 r2 ā¦ 0 0 Ī»11Ī»2
for some functions Ī»1 , Ī»2 , r1 , r2 on U , with Ī»1 , Ī»2 = 0. (b) Show that under a transformation 1 1 1 ĖĀÆ ĀÆ Ļ Ī»1 Ļ (10.9) = , 1 2 ĖĀÆ 2 ĀÆ Ļ Ī»2 Ļ
of the form (10.8), we have 3 2 ĖĀÆ 1 Ļ Ī» 1 Ī»2 Ļ ĀÆ 13 = . ĖĀÆ 23 Ļ Ī»1 Ī»22 Ļ ĀÆ 23
Conclude that the transformed function cĖ deļ¬ned by the conditions ĖĀÆ 13 = cĖ Ļ ĖĀÆ 2 , Ļ
ĖĀÆ 23 = cĖ Ļ ĖĀÆ 1 Ļ
is given by (10.10)
cĖ = Ī»21 Ī»22 c.
Thus, the function c is a relative invariant: If it vanishes for any 0-adapted frame based at a point u ā U , then it vanishes for every 0-adapted frame based at u. At this point, we need to consider separately the cases where c is zero or nonzero. First, we examine the case where c vanishes identically on U . *Exercise 10.8. Suppose that for any 0-adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)) along Ī£, the function c is identically equal to zero on U . Show that this assumption implies that de1 ā” de2 ā” 0
mod (e1 , e2 ).
Conclude that the plane spanned by the vectors (e1 (u), e2 (u)) is constant on U and hence that the surface Ī£ is contained in this plane.
316
10. Two classical theorems
As a consequence of Exercise 10.8, the function c associated to any 0-adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)) on a non-planar doubly ruled surface cannot vanish identically. By shrinking the domain U of our parametrization x : U ā A3 of Ī£ if necessary, we can assume that c = 0 at every point of U . Remark 10.9. The assumption that the function c is either identically zero or never zero on U is an example of a constant type assumption; it may be thought of as somewhat analogous to the assumption that a regular surface in E3 either contains no umbilic points or is totally umbilic. In principle, there could exist doubly ruled surfaces for which this assumption does not hold, i.e., for which c = 0 only on a proper, nonempty subset of U . But once we prove Theorem 10.3 for the open subset of U on which c = 0, a patching argument based on the assumption that Ī£ is smooth can be used to show that, in fact, this cannot happen. According to equation (10.10), there exists a choice of 0-adapted frame ļ¬eld (e1 (u), e2 (u), e3 (u)) for which c = Ā±1. If c = ā1, we can exchange e1 (u) and e2 (u) and replace e3 (u) by āe3 (u) to arrive at a new 0-adapted frame ļ¬eld for which c = 1; thus, without loss of generality, we will assume that we can choose a 0-adapted frame ļ¬eld for which c = 1. Such a frame ļ¬eld will be called 1-adapted. *Exercise 10.10. Let (e1 (u), e2 (u), e3 (u)) be any 1-adapted frame ļ¬eld along Ī£. Ė2 (u), e Ė3 (u)) along Ī£ (a) Show that any other 1-adapted frame ļ¬eld (Ė e1 (u), e must have the form ā” ā¤ Ī» 0 r1 ā„
ā¢ 1 ā„ Ė1 (u) e Ė2 (u) e Ė3 (u) = e1 (u) e2 (u) e3 (u) ā¢ e (10.11) ā£ 0 Ā± Ī» r2 ā¦ 0 0 Ā±1 for some functions Ī», r1 , r2 on U , with Ī» = 0. (b) Note that the condition c = 1 is equivalent to the relations (10.12)
Ļ ĀÆ 13 = Ļ ĀÆ 2,
Ļ ĀÆ 23 = Ļ ĀÆ1
among the associated Maurer-Cartan forms. Compare with Exercise 6.42(a), and conclude that any 1-adapted frame ļ¬eld along Ī£ satisļ¬es the deļ¬ning conditions for what we called a ā1-adapted null frameā there. In particular, Ī£ must be a hyperbolic equi-aļ¬ne surface; otherwise no such frame ļ¬eld could exist along Ī£. (c) Apply the result of Exercise 6.42(c) to show that there exists a choice of 1-adapted frame ļ¬eld along Ī£ for which Ļ ĀÆ 33 = 0. (This condition corresponds
10.1. Doubly ruled surfaces in R3
317
to choosing e3 (u) to be parallel to the equi-aļ¬ne normal direction.) Such a frame ļ¬eld will be called 2-adapted. (d) Show that any two 2-adapted frame ļ¬elds (e1 (u), e2 (u), e3 (u)), (Ė e1 (u), Ė2 (u), e Ė3 (u)) along Ī£ must be related by a transformation of the form e ā” ā¤ Ī» 0 0 ā„
ā¢ 1 ā„ Ė1 (u) e Ė2 (u) e Ė3 (u) = e1 (u) e2 (u) e3 (u) ā¢ e (10.13) ā£0 Ā± Ī» 0 ā¦ 0 0 Ā±1 for some function Ī» = 0 on U . *Exercise 10.11. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame ļ¬eld along Ī£. (a) Apply the results of Exercises 6.42(e) and 6.42(f) and equations (10.7) to show that a = b = 0. (Alternatively, diļ¬erentiate equations (10.12) to arrive at the same result.) Consequently, the Maurer-Cartan forms associated to any 2-adapted frame ļ¬eld on Ī£ must satisfy the conditions (10.14)
Ļ ĀÆ 21 = Ļ ĀÆ 12 = 0.
Conclude from equation (10.14) that the Fubini-Pick form (6.26) of Ī£ is identically zero. (b) Conversely, show that if Ī£ is a hyperbolic equi-aļ¬ne surface whose Fubini-Pick form is identically zero, then Ī£ is doubly ruled. (Hint: It suļ¬ces to show that the condition (10.5) holds.) (c) Diļ¬erentiate equations (10.14) and apply Cartanās lemma to conclude that there exist functions f, g on U such that (10.15)
Ļ ĀÆ 31 = f Ļ ĀÆ 1,
Ļ ĀÆ 32 = g Ļ ĀÆ 2.
(d) Diļ¬erentiate the equation Ļ ĀÆ 33 = 0 and conclude that g = f . (e) Substitute equations (10.15) (with g = f ) into the Cartan structure equations (3.8) for dĀÆ Ļ31 and dĀÆ Ļ32 to obtain df ā§ Ļ ĀÆ 1 = df ā§ Ļ ĀÆ 2 = 0. Conclude that df = 0, and hence f is equal to a constant C ā R on U .
318
10. Two classical theorems
Once again, we need to divide into cases based on whether or not C is zero. Case 1: C = 0. *Exercise 10.12. Suppose that C = 0. (a) Show that de3 = 0, and conclude that the vector ļ¬eld e3 (u) is constant on U . (Note that this implies that Ī£ is an improper equi-aļ¬ne sphere; cf. Exercise 6.41.) (b) Recall that the Maurer-Cartan forms associated to a 2-adapted frame ļ¬eld on Ī£ satisfy Ļ ĀÆ 11 + Ļ ĀÆ 22 = 0. Show that dĀÆ Ļ11 = dĀÆ Ļ22 = 0, and use the PoincarĀ“e lemma (cf. Theorem 2.31) to conclude that there exists a function Ī¼ on U such that (10.16)
Ļ ĀÆ 11 = dĪ¼ = āĀÆ Ļ22 .
Remark 10.13. Technically, this result only holds if U is homeomorphic to an open disk in R2 . If this is not the case, then U can be covered by such open sets, and the result of Theorem 10.3 can be applied to the restriction of the parametrization x : U ā A3 to each of these open subsets of U . The theorem can then be obtained for the entire surface Ī£ = x(U ) via a patching argument. (c) Use equations (10.14) and (10.16) to show that (10.17) (d) Show that
dĀÆ Ļ 1 = ādĪ¼ ā§ Ļ ĀÆ 1,
dĀÆ Ļ 2 = dĪ¼ ā§ Ļ ĀÆ 2.
d eĪ¼ Ļ ĀÆ 1 = d eāĪ¼ Ļ ĀÆ 2 = 0.
Conclude that by a transformation of the form (10.13) with Ī» = eāĪ¼ , we can arrange that (10.18)
dĀÆ Ļ 1 = dĀÆ Ļ 2 = 0.
Then use equation (10.17) to conclude that dĪ¼ = 0, and hence (10.19)
Ļ ĀÆ 11 = Ļ ĀÆ 22 = 0.
To summarize, we have shown that when C = 0, there exists a 2-adapted frame ļ¬eld along Ī£ for which the associated Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ) satisfy the following conditions: dĀÆ Ļ 1 = dĀÆ Ļ 2 = 0, (10.20)
Ļ ĀÆ 11 = Ļ ĀÆ 21 = Ļ ĀÆ 12 = Ļ ĀÆ 22 = Ļ ĀÆ 31 = Ļ ĀÆ 32 = Ļ ĀÆ 33 = 0, Ļ ĀÆ 13 = Ļ ĀÆ 2,
Ļ ĀÆ 23 = Ļ ĀÆ 1.
10.1. Doubly ruled surfaces in R3
319
The following exercise shows how these equations can be integrated in order to determine the surface Ī£. *Exercise 10.14. (a) Apply the PoincarĀ“e lemma to the ļ¬rst equations in (10.20) to show that there exist functions u, v on U such that Ļ ĀÆ 1 = du,
(10.21)
Ļ ĀÆ 2 = dv
(cf. Remark 10.13). Note that the condition Ļ ĀÆ1 ā§ Ļ ĀÆ 2 = 0 implies that the functions (u, v) form a local coordinate system on U . (b) From equations (10.21) and the remainder of equations (10.20), conclude that the structure equations (10.3) now take the form dx = e1 du + e2 dv, de1 = e3 dv,
(10.22)
de2 = e3 du, de3 = 0.
(c) Integrate equations (10.22) (beginning with the equation for de3 and ĀÆ1 , e ĀÆ2 , e ĀÆ3 ) working backwards) to show that there exist constant vectors (ĀÆ x, e such that ĀÆ3 , e3 (u, v) = e (10.23)
ĀÆ1 + vĀÆ e1 (u, v) = e e3 , ĀÆ2 + uĀÆ e2 (u, v) = e e3 , ĀÆ + uĀÆ x(u, v) = x e1 + vĀÆ e2 + uvĀÆ e3 .
(d) Use equations (10.23) and the fact that (e1 (u), e2 (u), e3 (u)) is a unimodular frame ļ¬eld to show that
ĀÆ1 e ĀÆ2 e ĀÆ3 = 1. det e ĀÆ = Conclude that via an equi-aļ¬ne transformation, we can arrange that x 0, 0] and
t [0,
ĀÆ1 = t [1, 0, 0] , e
ĀÆ2 = t [0, 1, 0] , e
ĀÆ3 = t[0, 0, 1] , e
and hence x(u, v) = t [u, v, uv] . Therefore, up to equi-aļ¬ne equivalence, Ī£ must be an open subset of the hyperbolic paraboloid (10.2).
320
10. Two classical theorems
Case 2: C = 0. Suppose that C = 0. Without loss of generality, we may suppose that C > 0: If instead we have C < 0, an equi-aļ¬ne transformation of the form
x = t x1 , x2 , x3 ā t āx1 , x2 , āx3 will reverse the sign of C. *Exercise 10.15. (a) Suppose that C > 0. Show that (10.24)
dĀÆ Ļ 1 = āĀÆ Ļ11 ā§ Ļ ĀÆ 1,
dĀÆ Ļ 2 = āĀÆ Ļ22 ā§ Ļ ĀÆ 2,
and use the Frobenius theorem (cf. Theorem 2.33) to conclude that every point u ā U has a neighborhood V ā U on which there exist functions u, v, g1 , g2 such that (10.25)
Ļ ĀÆ 1 = eg1 du,
Ļ ĀÆ 2 = eg2 dv.
It suļ¬ces to prove Theorem 10.3 on each of these restricted neighborhoods V (cf. Remark 10.13); for simplicity, we will assume that V = U . Note that the condition Ļ ĀÆ1 ā§ Ļ ĀÆ 2 = 0 implies that the functions (u, v) form a local coordinate system on U . (b) Show that by a transformation of the form (10.13) with Ī» = e 2 (g1 āg2 ) , we can arrange that 1
(10.26)
Ļ ĀÆ 1 = eh du,
Ļ ĀÆ 2 = eh dv,
where h = 12 (g1 + g2 ). (c) Use equations (10.24) (together with the fact that Ļ ĀÆ 11 + Ļ ĀÆ 22 = 0) to show that (10.27)
Ļ ĀÆ 11 = āĀÆ Ļ22 = hu du ā hv dv.
Then substitute this expression into the structure equation for dĀÆ Ļ11 to show that the function h must satisfy the PDE (10.28)
2huv = Ce2h .
Conclude that the function z(u, v) = 2h(u, v) + ln C must be a solution of Liouvilleās equation (10.29)
zuv = ez .
*Exercise 10.16. (a) The general solution to Liouvilleās equation (10.29) is " # 2U (u)V (v) (10.30) z(u, v) = ln , (U (u) + V (v))2
10.1. Doubly ruled surfaces in R3
321
where U (u) and V (v) are arbitrary functions of a single variable with the property that U (u) and V (v) are nonzero and have the same sign (cf. Exercise 9.22). Show that by making the change of coordinates u Ė = U (u), we can arrange that
vĖ = V (v), "
z(Ė u, vĖ) = ln
2 (Ė u + vĖ)2
# .
Dropping the tildes, this yields (10.31)
1 h(u, v) = (z(u, v) ā ln C) = ln 2
ā
2 ā C(u + v)
.
(b) Use equations (10.12), (10.14), (10.15), (10.26), (10.27), and (10.31) to show that ā ā 2 2 1 2 Ļ ĀÆ =ā du, Ļ ĀÆ =ā dv, C(u + v) C(u + v)
(10.32)
Ļ ĀÆ 21 = Ļ ĀÆ 12 = Ļ ĀÆ 33 = 0, 1 Ļ ĀÆ 11 = āĀÆ Ļ22 = (ādu + dv), (u + v) ā ā 2 2 3 3 Ļ ĀÆ1 = ā dv, Ļ ĀÆ2 = ā du, C(u + v) C(u + v) ā ā 2C 2C 1 2 Ļ ĀÆ3 = du, Ļ ĀÆ3 = dv. (u + v) (u + v)
(c) From equations (10.32), conclude that the structure equations (10.3) now take the form ā ā 2 2 dx = e1 ā du + e2 ā dv , C(u + v) C(u + v) ā " # 1 2 de1 = e1 (ādu + dv) + e3 ā dv , (u + v) C(u + v) (10.33) ā " # 1 2 de2 = e2 (du ā dv) + e3 ā du , (u + v) C(u + v) ā ā 2C 2C de3 = e1 du + e2 dv . (u + v) (u + v)
322
10. Two classical theorems
*Exercise 10.17. (a) Show that equations (10.33) are equivalent to the PDE system (10.34) ā ā 2 2 xu = ā xv = ā e1 , e2 , C(u + v) C(u + v) ā 2 1 1 (e1 )u = ā (e1 )v = e1 , e1 + ā e3 , (u + v) (u + v) C(u + v) ā 2 1 1 (e2 )u = e2 + ā e2 , e3 , (e2 )v = ā (u + v) (u + v) C(u + v) ā ā 2C 2C (e3 )u = (e3 )v = e1 , e2 . (u + v) (u + v) (b) Integrate the equations for (e1 )u and (e2 )v to show that there exist A3 -valued functions f (v), g(u) such that (10.35)
e1 (u, v) =
1 f (v), (u + v)
e2 (u, v) =
1 g(u). (u + v)
(c) Use the equations for (e1 )v and (e2 )u to write (10.36)
ā ā C C e3 (u, v) = ā ((u + v)(e2 )u ā e2 ) = ā ((u + v)(e1 )v ā e1 ) . 2 2
Substitute equations (10.35) into this equation to show that (10.37)
(u + v)f (v) ā 2f (v) = (u + v)g (u) ā 2g(u).
(d) Diļ¬erentiate equation (10.37) with respect to u and v, and use the result to show that there exist constant vectors c10 , c20 , c11 , c21 , c2 such that (10.38)
f (v) = c2 v 2 + c11 v + c10 ,
g(u) = c2 u2 + c21 u + c20 .
Substitute these expressions into equation (10.37), and by comparing like powers of u and v, conclude that c21 = āc11 , c20 = c10 . Set c1 = c11 = āc21 ,
c0 = c10 = c20 ,
so that (10.38) becomes (10.39)
f (v) = v 2 c2 + vc1 + c0 ,
g(u) = u2 c2 ā uc1 + c0 .
10.1. Doubly ruled surfaces in R3
323
(e) Substitute (10.39) into equations (10.35), (10.36) to obtain 1 (v 2 c2 + vc1 + c0 ), (u + v) 1 e2 (u, v) = (u2 c2 ā uc1 + c0 ), (u + v) ā C e3 (u, v) = ā (2uvc2 + (u ā v)c1 ā 2c0 ). 2(u + v)
e1 (u, v) = (10.40)
Observe from equations (10.33) that d(x ā C1 e3 ) = 0; this implies that x ā C1 e3 is constant along Ī£ (in particular, Ī£ is a proper equi-aļ¬ne sphere; cf. Exercise 6.41), and by a translation we can arrange that x ā C1 e3 = 0. Then we have 1 1 (10.41) x(u, v) = e3 (u, v) = ā (2uvc2 + (u ā v)c1 ā 2c0 ). C 2C(u + v) (f) Verify that the functions in equations (10.40), (10.41) satisfy the PDE system (10.34). (g) Use equations (10.40) to show that
ā
C det e1 (u, v) e2 (u, v) e3 (u, v) = ā det c2 c1 c0 . 2
Since (e1 , e2 , e3 ) must be a unimodular frame, conclude that ā
2 (10.42) det c2 c1 c0 = ā . C (h) By an equi-aļ¬ne transformation, we can arrange that ā 2t 1 t 1 t c2 = ā ā [1, 0, 1] , c1 = ā [0, 1, 0] , c0 = ā ā [ā1, 0, 1] . 6 6 2 C C 26C Check that these vectors satisfy the determinant condition (10.42) and that with this choice, we have x(u, v) =
1 2 3
C (u + v)
t
[uv + 1, u ā v, uv ā 1] .
Finally, check that the coordinates of x(u, v) satisfy the deļ¬ning equation 2 (10.1) for the hyperboloid, with r = C ā 3 . Therefore, up to equi-aļ¬ne equivalence, Ī£ must be an open subset of a hyperboloid of one sheet. This completes the proof of Theorem 10.3.
324
10. Two classical theorems
10.2. The Cauchy-Crofton formula The Cauchy-Crofton formula comes from the subject of integral geometry. It relates the length of a curve in the Euclidean plane to the āsizeā of the set of lines in the plane that intersect the curve (counted with multiplicity). A classical approach to this formula may be found in [dC76]; the moving frames approach described here is outlined in [Che42]. In order to make sense of the idea of the āsizeā of a set of lines, we need to introduce the notion of a measure on the set of lines in E2 , a.k.a. the aļ¬ne Grassmannian G1 (E2 ) (cf. Ā§9.2). In general, a measure on a manifold M is a diļ¬erential form dĪ¼ on M (which, despite the notation, is not necessarily an exact form) that is used to deļ¬ne the size of any subset of M via integration. Speciļ¬cally, the measure of a subset Ī© ā M is given by & Ī¼(Ī©) = dĪ¼. Ī©
One familiar example is the area measure on the Euclidean plane E2 : The area measure is the 2-form dA = dx ā§ dy, and the area of any (measurable) subset Ī© ā E2 is given by & A(Ī©) = dA. Ī©
An important feature of the area measure is that it is invariant under the action of the Euclidean group E(2): For any element g ā E(2) and any measurable subset Ī© ā E2 , we have A(g Ā· Ī©) = A(Ī©). We can see how this invariance comes about as follows: Recall that we can regard E2 as the homogeneous space E2 ā¼ = E(2)/SO(2); i.e., E2 is the set of left cosets of the subgroup SO(2) ā E(2): SO(2) - E(2) Ļ
?
E2 ā¼ = E(2)/SO(2). The Maurer-Cartan forms (Ļ 1 , Ļ 2 , Ļ21 ) form a basis for the left-invariant 1forms on the Lie group E(2), and of these, Ļ 1 and Ļ 2 are semi-basic for the projection Ļ : E(2) ā E2 (cf. Exercise 3.19). The area measure on the quotient space E2 is given by the wedge product of these semi-basic forms: dA = Ļ 1 ā§ Ļ 2 .
10.2. The Cauchy-Crofton formula
325
(While the 1-forms (Ļ 1 , Ļ 2 ) are not individually well-deļ¬ned on E2 , it turns out that their wedge product is; cf. Exercise 4.47.) So the invariance of the area measure follows immediately from the left-invariance of the MaurerCartan forms: Given any measurable subset Ī© ā E2 and any element g ā E(2), we have & A(g Ā· Ī©) = Ļ1 ā§ Ļ2 gĀ·Ī© & = Lāg (Ļ 1 ā§ Ļ 2 ) &Ī© = Ļ1 ā§ Ļ2 Ī©
= A(Ī©). Now consider the space of lines in E2 . As we saw in Ā§9.2 when we constructed the aļ¬ne Grassmannian G1 (E3 ), we can specify a line in the plane E2 by choosing a point x on and a unit vector e1 parallel to . Two pairs (x, e1 ) and (y, f1 ) determine the same line if and only if f1 = Ā±e1 and y = x + te1 for some t ā R (cf. Exercise 9.1). In this case, we write (x, e1 ) ā¼ (y, f1 ), and the aļ¬ne Grassmannian is deļ¬ned to be the quotient space G1 (E2 ) = E2 Ć S1 / ā¼ . *Exercise 10.18. (a) Show that the product E2 ĆS1 is diļ¬eomorphic to the oriented, orthonormal frame bundle F (E2 ), which in turn is diļ¬eomorphic to the Lie group E(2). (Hint: Choosing one unit vector e1 ā Tx E2 uniquely determines a second unit vector e2 ā Tx E2 such that (e1 , e2 ) is an oriented, orthonormal frame for Tx E2 . This is a feature that is particular to E2 and not true for En when n ā„ 3.) Consequently, we can write G1 (E2 ) = E(2)/ ā¼ . (b) Show that the equivalence relation (x, e1 ) ā¼ (y, f1 ) on E2 Ć S1 corresponds to the following equivalence relation on E(2): If 1 0 0 1 0 0 Ė= Ė= x , y ā E(2), x e1 e2 y f1 f2 Ėā¼y Ė if and only if then x
ā” ā¤ 1 0 0 Ė=x Ė ā£ t Ā±1 0 ā¦ y 0 0 Ā±1
326
10. Two classical theorems
for some t ā R, where the choice of sign is the same in both diagonal entries. Therefore, G1 (E2 ) may be regarded as the set of left cosets of the subgroup ā§ā” ā« ā¤ āØ 1 0 0 ā¬ H = ā£ t Ā±1 0 ā¦ : t ā R ā E(2), ā© ā 0 0 Ā±1 and E(2) may be regarded as a principal bundle over G1 (E2 ) with ļ¬ber group H: H
- E(2) Ļ
?
G1 (E2 ) ā¼ = E(2)/H. (c) Show that the Maurer-Cartan forms (Ļ 2 , Ļ21 ) on E(2) are semi-basic for the projection Ļ : E(2) ā G1 (E2 ). This suggests that a reasonable way to deļ¬ne a measure dĪ¼ on G1 (E2 ) might be to set dĪ¼ = |Ļ 2 ā§ Ļ21 |. Like the area measure dA = Ļ 1 ā§ Ļ 2 , the 2-form dĪ¼ is well-deļ¬ned on the quotient G1 (E2 ), even though the individual 1-forms (Ļ 2 , Ļ21 ) are not. (The absolute value sign is necessary because we have deļ¬ned G1 (E2 ) to be the space of unoriented lines, but reversing the orientation changes the sign of the 2-form Ļ 2 ā§ Ļ21 ; therefore, this 2-form is only well-deļ¬ned up to sign on G1 (E2 ).) Then, given any (measurable) subset Ī© ā G1 (E2 ), we deļ¬ne the measure Ī¼(Ī©) to be & & Ī¼(Ī©) = dĪ¼ = |Ļ 2 ā§ Ļ21 |. Ī©
Ī©
Exercise 10.19. Another way to determine a line in E2 is to specify two parameters: the angle Īø with 0 ā¤ Īø < Ļ such that the vector e1 = t[cos(Īø), sin(Īø)] is parallel to and the shortest distance Ļ from the origin to . This identiļ¬cation gives rise to a diļ¬eomorphism G1 (E2 ) ā¼ = R Ć S1 (where S1 represents R/ĻZ rather than the more typical R/2ĻZ), with coordinates {(Ļ, Īø) | Ļ ā R, 0 ā¤ Īø < Ļ} .
10.2. The Cauchy-Crofton formula
327
Show that the measure dĪ¼ on G1 (E2 ) is then given by the 2-form dĪ¼ = |dĻ ā§ dĪø| on R Ć S1 . (Hint: How are the Maurer-Cartan forms (Ļ 2 , Ļ21 ) related to (dĻ, dĪø)?) Before we state the Cauchy-Crofton formula, we need to introduce the notion of incidence: Deļ¬nition 10.20. A point x ā E2 and a line ā G1 (E2 ) are called incident if the point x lies on the line . A curve Ī± in E2 and a line ā G1 (E2 ) are called incident if any point of Ī± is incident with . *Exercise 10.21. Let Ļ1 : E(2) ā E2 denote the projection from E(2) to E2 and let Ļ2 : E(2) ā G1 (E2 ) denote the projection from E(2) to G1 (E2 ), as in the following diagram: SO(2)
- E(2)
Ļ1 /
E2
H
S Ļ2 SS w
G1 (E2 ).
Show that a point x ā E2 and a line ā G1 (E2 ) are incident if and only if the left cosets Ļ1ā1 (x), Ļ2ā1 () ā E(2) have nonempty intersection. We are now ready to state the Cauchy-Crofton formula. Theorem 10.22. Let Ī± : [a, b] ā E2 be a curve in the Euclidean plane, and let Ī© ā G1 (E2 ) be the set of lines incident with Ī±, counted with multiplicity (e.g., if a line intersects Ī± at two distinct points, then it counts twice). Then Ī¼(Ī©) = 2L, where L is the length of Ī±. To begin the proof, let 8 = Ļ ā1 (Ī±([a, b])) ā E(2). Ī© 1 8 consists of all oriented, orthonormal frames (x; e1 , e2 ) based at all The set Ī© points x ā Ī±, each of which corresponds to a line passing through the point x and parallel to the frame vector e1 . Moreover, any line that intersects 8 Each line Ī± in k distinct points is represented by 2k distinct points in Ī©: 8 for each point x where it intersects Ī±, represented by appears twice in Ī© frames of the form (x; e1 , e2 ) and (x; āe1 , āe2 ).
328
10. Two classical theorems
8 is a 2-dimensional submanifold *Exercise 10.23. Convince yourself that Ī© of E(2). Now, let 8 ā G1 (E2 ), Ī© = Ļ2 (Ī©) and observe that Ī© consists of all lines incident with Ī±. Moreover, for any 8 is precisely twice the number of ā Ī©, the cardinality of the set Ļ2ā1 () ā© Ī© 8 represents all the lines distinct points in which intersects Ī±. So the set Ī© 2 in E that are incident with Ī±, counted with double multiplicity. 8 is an The restriction of the projection Ļ2 : E(2) ā G1 (E2 ) to the set Ī© 8 8 immersion Ļ2 : Ī© ā Ī©; in fact, this map realizes Ī© as a double cover of Ī©. Because the measure dĪ¼ = |Ļ 2 ā§ Ļ21 | is semi-basic for the projection Ļ2 , it follows that & & & ā 1 1 8 Ī¼(Ī©) = dĪ¼ = 2 Ļ2 (dĪ¼) = 2 dĪ¼ = 12 Ī¼(Ī©), Ī©
Ī©
Ī©
where the integral on the left must be counted with multiplicity. (We are abusing notation slightly here by writing dĪ¼ for both the 2-form |Ļ 2 ā§ Ļ21 | on E(2) and the well-deļ¬ned 2-form dĪ¼ on G1 (E2 ) whose pullback to E(2) via Ļ2 is equal to |Ļ 2 ā§ Ļ21 |.) *Exercise 10.24. Let Ī± : [a, b] ā E2 be an arc-length parametrization of Ī±, given by Ī±(s) = t[x(s), y(s)]. (a) Show that the vectors
ĀÆ1 (s) = t x (s), y (s) , e
ĀÆ2 (s) = t āy (s), x (s) e
form an oriented, orthonormal frame at the point Ī±(s). 8 can be parametrized by the map (b) Show that the set Ī© Ī± Ė : [a, b] Ć [0, 2Ļ] ā E(2) deļ¬ned by Ī± Ė (s, Īø) = (x(s, Īø); e1 (s, Īø), e2 (s, Īø)) ĀÆ1 (s) + sin(Īø) e ĀÆ2 (s), ā sin (Īø)ĀÆ ĀÆ2 (s)) = (Ī±(s); cos(Īø) e e1 (s) + cos(Īø) e x(s) x (s) cos(Īø)āy (s) sin(Īø) āx (s) sin(Īø)āy (s) cos(Īø) = ; , . y(s) y (s) cos(Īø)+x (s) sin(Īø) āy (s) sin(Īø)+x (s) cos(Īø)
10.3. Maple computations
329
(c) Pull back the Cartan structure equation dx = e1 Ļ 1 + e2 Ļ 2 on E(2) via Ī± Ė to show that the 1-forms Ļ ĀÆ1 = Ī± Ė ā Ļ 1 and Ļ ĀÆ2 = Ī± Ė ā Ļ 2 are given by (10.43)
Ļ ĀÆ 2 = ā sin(Īø) ds.
Ļ ĀÆ 1 = cos(Īø) ds,
Then use the structure equations for dĀÆ Ļ 1 and dĀÆ Ļ 2 to show that Ļ ĀÆ 21 = ādĪø + Ī» ds
(10.44)
for some function Ī»(s, Īø). (In fact, the structure equation for dĀÆ Ļ21 implies that Ī» is a function of s alone.) (d) Use equations (10.43) and (10.44), together with the fact that & & & 2 1 2 1 |Ļ ā§ Ļ2 | = |Ļ ā§ Ļ2 | = Ī± Ė ā (|Ļ 2 ā§ Ļ21 |), Ī©
Ī±([a,b]Ć[0,2Ļ]) Ė
to show that 8 = Ī¼(Ī©)
&
2Ļ
[a,b]Ć[0,2Ļ]
&
0
b
| sin(Īø)| ds dĪø = 4L,
a
where L = (b ā a) is the length of Ī±. Conclude that Ī¼(Ī©) = 2L. This completes the proof of Theorem 10.22.
10.3. Maple computations Begin by setting up exactly as we did for Chapter 6: Load the Cartan and LinearAlgebra packages into Maple, declare the Maurer-Cartan forms (ĀÆ Ļi, Ļ ĀÆ ji ) associated to a unimodular frame ļ¬eld on Ī£, and tell Maple about the structure equations and the relation Ļ ĀÆ 11 + Ļ ĀÆ 22 + Ļ ĀÆ 33 = 0. Now, suppose that Ī£ is doubly ruled and that we have chosen an adapted frame ļ¬eld with (e1 (u), e2 (u)) tangent to the rulings. Set up a substitution that encodes the conditions (10.7) for the associated Maurer-Cartan forms: > ruledsub:= [omega[3]=0, omega[2,1] = a*omega[2], omega[1,2] = b*omega[1], omega[3,1] = c*omega[2], omega[3,2] = c*omega[1]]; ĖĀÆ i , Ļ ĖĀÆ ji ) to represent the Exercise 10.7: Begin by introducing new 1-forms (Ļ transformed forms: > Form(Omega[1], Omega[2], Omega[3]); Form(Omega[1,1], Omega[1,2], Omega[1,3], Omega[2,1],
330
10. Two classical theorems
Omega[2,2], Omega[2,3], Omega[3,1], Omega[3,2], Omega[3,3]); Omega[3,3]:= -(Omega[1,1] + Omega[2,2]); Under a transformation of the form (10.8), we have the relations (10.9). Set this up as a substitution: > framechangesub:= [Omega[1] = (1/lambda1)*omega[1], Omega[2] = (1/lambda2)*omega[2], Omega[3,1] = lambda1Ė2*lambda2*omega[3,1], Omega[3,2] = lambda1*lambda2Ė2*omega[3,2]]; > framechangebacksub:= makebacksub(framechangesub); In order to ļ¬nd how the function c transforms, combine the substitutions ĖĀÆ 13 and Ļ ĖĀÆ 23 are related to Ļ ĖĀÆ 2 and framechangesub and ruledsub to see how Ļ ĖĀÆ 1 , respectively: Ļ > Simf(subs(framechangebacksub, Simf(subs(ruledsub, Simf(subs(framechangesub, Omega[3,1])))))); Ī»12 Ī»22 c Ī©2 > Simf(subs(framechangebacksub, Simf(subs(ruledsub, Simf(subs(framechangesub, Omega[3,2])))))); Ī»12 Ī»22 c Ī©1 So we have cĖ = Ī»21 Ī»22 c, as in equation (10.10). Exercise 10.11: Since (e1 (u), e2 (u), e3 (u)) is a 2-adapted frame ļ¬eld, we have c = 1, and since we have already set Ļ ĀÆ 33 = ā(ĀÆ Ļ11 + Ļ ĀÆ 22 ), the condition 3 1 2 Ļ ĀÆ 3 = 0 is equivalent to Ļ ĀÆ 1 +ĀÆ Ļ2 = 0. Add these conditions to our substitution: > c:= 1; > ruledsub:= [op(ruledsub), omega[2,2] = -omega[1,1]]; Now diļ¬erentiate equations (10.12): > Simf(subs(ruledsub, Simf(d(omega[3,1] - omega[2])))); ā2 a (Ļ1 ) &Ė (Ļ2 ) > Simf(subs(ruledsub, Simf(d(omega[3,2] - omega[1])))); 2 b (Ļ1 ) &Ė (Ļ2 ) Since Ļ ĀÆ1 ā§ Ļ ĀÆ 2 = 0, it follows that a = b = 0. > a:= 0; b:= 0;
10.3. Maple computations
331
Now diļ¬erentiate equations (10.14): > Simf(subs(ruledsub, Simf(d(omega[1,2])))); (Ļ1 ) &Ė (Ļ1,3 ) > Simf(subs(ruledsub, Simf(d(omega[2,1])))); (Ļ2 ) &Ė (Ļ2,3 ) By Cartanās lemma, equations (10.15) must hold. Add these conditions to our substitution: > ruledsub:= [op(ruledsub), omega[1,3] = f*omega[1], omega[2,3] = g*omega[2]]; Now diļ¬erentiate the equation Ļ ĀÆ 11 + Ļ ĀÆ 22 = 0: > Simf(subs(ruledsub, Simf(d(omega[1,1] + omega[2,2])))); (āf + g) (Ļ1 ) &Ė (Ļ2 ) Therefore, g = f . > g:= f; Finally, diļ¬erentiate equations (10.15): > Simf(subs(ruledsub, Simf(d(omega[1,3])) - d(Simf(subs(ruledsub, omega[1,3]))))); (Ļ1 ) &Ė (d(f )) > Simf(subs(ruledsub, Simf(d(omega[2,3])) - d(Simf(subs(ruledsub, omega[2,3]))))); (Ļ2 ) &Ė (d(f )) It follows from Cartanās lemma that df = 0, and hence f is equal to a constant C ā R. Details for the case C = 0 may be found in the Maple worksheet for this chapter on the AMS webpage; here we will explore the case C > 0. Exercise 10.15: First, declare C to be constant and set f = C. (It is useful to set up separate substitutions for the cases C = 0, C > 0, so we give our substitution a new name when we divide into cases.) > Form(C=-1); > ruledsubcase2:= Simf(subs([f=C], ruledsub));
332
10. Two classical theorems
Compute dĀÆ Ļ 1 and dĀÆ Ļ2: > Simf(subs(ruledsubcase2, Simf(d(omega[1])))); > Simf(subs(ruledsubcase2, Simf(d(omega[2])))); This should yield equations (10.24). By an application of the Frobenius theorem followed by a transformation of the form (10.13), we can arrange that (ĀÆ Ļ1, Ļ ĀÆ 2 ) have the form (10.26). Add these conditions to our substitution, and note that since Ļ ĀÆ 1 and Ļ ĀÆ 2 already appear on the right-hand sides of some of the equations in ruledsubcase2, we need to do this via the two-step process that we introduced in Chapter 6: > PDETools[declare](h(u,v)); > ruledsubcase2:= Simf(subs([omega[1] = exp(h(u,v))*d(u), omega[2] = exp(h(u,v))*d(v)], ruledsubcase2)); > ruledsubcase2:= [op(ruledsubcase2), omega[1] = exp(h(u,v))*d(u), omega[2] = exp(h(u,v))*d(v)]; Now we can use the equations (10.24) to determine the 1-form Ļ ĀÆ 11 = āĀÆ Ļ22 : > zero1:= Simf(subs(ruledsubcase2, Simf(d(omega[1])) - d(Simf(subs(ruledsubcase2, omega[1]))))); > factor(pick(zero1, d(u))); āeh (Ļ11 + hv d(v)) > zero2:= Simf(subs(ruledsubcase2, Simf(d(omega[2])) - d(Simf(subs(ruledsubcase2, omega[2]))))); > factor(pick(zero1, d(v))); āeh (āĻ11 + hu d(u)) Equation (10.27) then follows from Cartanās lemma. Add this condition to our substitution: > ruledsubcase2:= Simf(subs([omega[1,1] = diff(h(u,v), u)*d(u) - diff(h(u,v), v)*d(v)], ruledsubcase2)); > ruledsubcase2:= [op(ruledsubcase2), omega[1,1] = diff(h(u,v), u)*d(u) - diff(h(u,v), v)*d(v)]; Finally, consider the structure equation for dĀÆ Ļ11 : > Simf(subs(ruledsubcase2, Simf(d(omega[1,1])) - d(Simf(subs(ruledsubcase2, omega[1,1]))))); (C e2 h ā 2 hu,v ) (d(v)) &Ė (d(u)) Therefore, h must be a solution of equation (10.28).
10.3. Maple computations
333
Exercise 10.16: Set h equal to the function given in equation (10.31): > ruledsubcase2:= Simf(subs([ h(u,v) = ln(sqrt(2)/(sqrt(C)*(u+v)))], ruledsubcase2)); Equations (10.32) can then be read oļ¬ directly from the equations in ruledsubcase2. Exercise 10.17: In order to construct the PDE system (10.34), declare the variables (X, e1, e2, e3) to be functions of u and v; then apply our substitution to the structure equations (10.3): > PDETools[declare](X(u,v), e1(u,v), e2(u,v), e3(u,v)); > zero1:= Simf(subs(ruledsubcase2, Simf(d(X(u,v)) - (e1(u,v)*omega[1] + e2(u,v)*omega[2])))); zero2:= Simf(subs(ruledsubcase2, Simf(d(e1(u,v)) - (e1(u,v)*omega[1,1] + e2(u,v)*omega[2,1] + e3(u,v)*omega[3,1])))); zero3:= Simf(subs(ruledsubcase2, Simf(d(e2(u,v)) - (e1(u,v)*omega[1,2] + e2(u,v)*omega[2,2] + e3(u,v)*omega[3,2])))); zero4:= Simf(subs(ruledsubcase2, Simf(d(e3(u,v)) - (e1(u,v)*omega[1,3] + e2(u,v)*omega[2,3] + e3(u,v)*omega[3,3])))); The PDE system (10.34) now consists of the scalar coeļ¬cients of du and dv in each of these expressions: > pde1a:= pde1b:= pde2a:= pde2b:= pde3a:= pde3b:= pde4a:= pde4b:=
pick(zero1, pick(zero1, pick(zero2, pick(zero2, pick(zero3, pick(zero3, pick(zero4, pick(zero4,
d(u)); d(v)); d(u)); d(v)); d(u)); d(v)); d(u)); d(v));
In this case, it turns out that Maple can solve the entire system in one step. (This often isnāt the case, especially with overdetermined systems, so the pdsolve command should generally be used with caution!) > pdsolve({pde1a, pde1b, pde2a, pde2b, pde3a, pde3b, pde4a, pde4b}, {X(u,v), e1(u,v), e2(u,v), e3(u,v)});
334
10. Two classical theorems
(The constants in the resulting solution should be interpreted as being vector-valued.) Up to renaming the constants and shifting x by a translation to eliminate one of them, this gives the expressions in equations (10.40), (10.41). Exercise 10.24: Since this exercise requires diļ¬erent Maurer-Cartan forms with diļ¬erent structure equations from the previous exercises, itās probably best to restart Maple and reload the Cartan and LinearAlgebra packages. Declare the Maurer-Cartan forms on F (E2 ) and tell Maple about their symmetries and structure equations: > Form(omega[1], omega[2], omega[1,2]); > omega[2,1]:= -omega[1,2]; > d(omega[1]):= -omega[1,2] &Ė omega[2]; d(omega[2]):= -omega[2,1] &Ė omega[1]; d(omega[1,2]):= 0; Declare the functions (x(s), y(s)), and deļ¬ne the vectors (x(s, Īø), e1 (s, Īø), e2 (s, Īø)): > PDETools[declare](x(s), y(s)); > X:= Vector([x(s), y(s)]); e10:= Vector([diff(x(s), s), diff(y(s), s)]); e20:= Vector([-diff(y(s), s), diff(x(s), s)]); e1:= cos(theta)*e10 + sin(theta)*e20; e2:= -sin(theta)*e10 + cos(theta)*e20; Now, express dx as a linear combination of (e1 (s, Īø), e2 (s, Īø)) in order to determine (ĀÆ Ļ1, Ļ ĀÆ 2 ) via the structure equation dx = e1 Ļ ĀÆ 1 + e2 Ļ ĀÆ2 : > zero1:= map(d, X) - e1*omega[1] - e2*omega[2]; > coordsub:= [op(Simf(solve({zero1[1], zero1[2]}, {omega[1], omega[2]})))]; This should yield the expressions (10.43) for (ĀÆ Ļ1, Ļ ĀÆ 2 ). Now use the structure 1 2 1 equations for dĀÆ Ļ and dĀÆ Ļ to determine Ļ ĀÆ2 : > zero2:= Simf(subs(coordsub, Simf(d(omega[1])) - d(Simf(subs(coordsub, omega[1]))))); > factor(pick(zero2, d(s))); sin(Īø) (Ļ1,2 + d(Īø)) > zero3:= Simf(subs(coordsub, Simf(d(omega[2])) - d(Simf(subs(coordsub, omega[2])))));
10.3. Maple computations
335
> factor(pick(zero3, d(s))); cos(Īø) (Ļ1,2 + d(Īø)) Since sin(Īø) and cos(Īø) cannot vanish simultaneously, Cartanās lemma implies that Ļ ĀÆ 21 has the form (10.44). Add this expression to coordsub: > coordsub:= [op(coordsub), omega[1,2] = -d(theta) + lambda*d(s)]; Now we can compute Ī± Ė ā (Ļ 2 ā§ Ļ21 ) as follows: > Simf(subs(coordsub, omega[2] &Ė omega[1,2])); ā sin(Īø) (d(Īø) &Ė d(s))
Part 4
Beyond the ļ¬at case: Moving frames on Riemannian manifolds
10.1090/gsm/178/11
Chapter 11
Curves and surfaces in elliptic and hyperbolic spaces
11.1. Introduction Until now, all the homogeneous spaces that we have encountered have been modeled on the vector space Rn , and we have relied extensively on the fact that all the various structures that we have deļ¬ned (Euclidean, Minkowski, equi-aļ¬ne, projective) are ļ¬at. This property is encoded in the Cartan structure equations (3.8): According to the second equation in (3.8), the matrix of connection forms Ļc = [Ļji ] satisļ¬es the structure equation (11.1)
dĻc + Ļc ā§ Ļc = 0.
(Note that this is not quite the same thing as the Maurer-Cartan equation (3.13) because Ļc is a submatrix of the matrix-valued Maurer-Cartan form Ļ.) This might not seem like such a big deal, but in fact ļ¬atness is a necessary condition for the existence of canonical isomorphisms Tx Rn ā¼ = Rn (cf. Remark 3.13 and the discussion in Ā§3.3.2). It is these canonical isomorphisms that allow us to think of the components (e1 (x), . . . , en (x)) of a frame ļ¬eld on Rn as functions from Rn to Rn rather than as sections of the tangent bundle T Rn , which in turn allows us to deļ¬ne their exterior derivatives in a straightforward way. 339
340
11. Curves and surfaces in elliptic and hyperbolic spaces
In this chapter, we will explore moving frames on two homogeneous spaces that are ācurvedā versions of En : elliptic space Sn (which is diļ¬eomorphic to an n-dimensional sphere) and hyperbolic space Hn (which is diļ¬eomorphic to Rn , but with a diļ¬erent metric structure than that of En ). We will see that for these spaces, the ļ¬atness condition (11.1) no longer holds; rather, the matrix of 2-forms (11.2)
Ī© = dĻc + Ļc ā§ Ļc ,
called the curvature of the connection matrix Ļc , is nonzero and is an important feature of the associated geometry.
11.2. The homogeneous spaces Sn and Hn In this section, we will introduce Riemannian homogeneous space structures on Sn and Hn , much as we did for ļ¬at homogeneous spaces in Chapter 3. Fortunately, Sn and Hn may naturally be regarded as submanifolds of the ļ¬at spaces En+1 and M1,n , respectively, so we can still use the tools developed in Chapter 3 to get started. 11.2.1. Elliptic space Sn . Deļ¬nition 11.1. The n-dimensional elliptic space Sn is the unit sphere in En+1 ; i.e., Sn = {x ā En+1 | x, x = 1}. The symmetry group of Sn is deļ¬ned to be the subgroup of E(n + 1) that preserves the set Sn . *Exercise 11.2. Show that the symmetry group of Sn consists of all matrices AĖ ā GL(n + 2) of the form t 1 0 Ė (11.3) A= , 0A where A ā SO(n+1). Therefore, the symmetry group of Sn is isomorphic to SO + (n + 1), and we will generally identify the element AĖ in equation (11.3) with the corresponding element A ā SO(n + 1). In order to describe Sn as a homogeneous space of the Lie group SO(n + 1), we need to compute the isotropy group of a point x ā Sn . *Exercise 11.3. Let x0 = t[1, 0, . . . , 0] ā Sn . Show that: (a) The isotropy group Hx0 of x0 in SO(n + 1) is ) ( t 1 0 ĀÆ (11.4) Hx0 = : A ā SO(n) . 0 AĀÆ
11.2. The homogeneous spaces Sn and Hn
341
(b) The isotropy group Hx of any other point x ā Sn is Hx = tx Hx0 tā1 x , where tx is any matrix in SO(n + 1) whose ļ¬rst column is x. (The transformation tx will then have the property that tx (x0 ) = x.) Orthonormal frames on Sn are deļ¬ned as follows: Deļ¬nition 11.4. An orthonormal frame f on Sn is a list of vectors f = (e0 , . . . , en ), where e0 ā Sn ā En+1 , e1 , . . . , en ā En+1 , and e0 Ā· Ā· Ā· en ā SO(n+1). We identify e0 with the position vector x ā Sn , and the condition
that e0 Ā· Ā· Ā· en ā SO(n + 1) implies that the vectors (e1 , . . . , en ) may be regarded as an orthonormal basis for the tangent space Te0 Sn ā Te0 En+1 ā¼ = En+1 . (We may also say that (e1 , . . . , en ) is an orthonormal frame based at e0 .) The collection of all orthonormal frames on Sn is called the orthonormal frame bundle of Sn , denoted F (Sn ). The same reasoning as in prior cases shows that the orthonormal frame bundle F (Sn ) may be regarded as the group SO(n + 1) via the one-to-one correspondence
g(e0 , . . . , en ) = e0 Ā· Ā· Ā· en . The projection map Ļ : SO(n + 1) ā Sn deļ¬ned by Ļ([e0 . . . en ]) = e0 describes SO(n + 1) as a principal bundle over Sn with ļ¬ber group SO(n); therefore, we have a natural correspondence Sn ā¼ = SO(n + 1)/SO(n). Because we have deļ¬ned the components (e0 , . . . , en ) of a frame as elements of the vector space En+1 , we can regard them as functions eĪ± : F (Sn ) ā En+1 and deļ¬ne the Maurer-Cartan forms (ĻĪ²Ī± ) on SO(n + 1) as usual by the equations (11.5)
deĪ± = eĪ² ĻĪ±Ī² ,
where 0 ā¤ Ī±, Ī² ā¤ n and ĻĪ±Ī² = āĻĪ²Ī± . *Exercise 11.5. (a) Show that the forms (Ļ0i ) for 1 ā¤ i ā¤ n are semi-basic for the projection Ļ : SO(n + 1) ā Sn and so may be regarded as the dual forms of any orthonormal frame (e1 , . . . , en ) based at e0 ā Sn . The forms (Ļji ) for 1 ā¤ i, j ā¤ n may then be regarded as the connection forms. (b) Set Ļ i = Ļ0i . Show that the dual forms (Ļ i ) and the connection forms (Ļji ) satisfy the structure equations (11.6)
dĻ i = āĻji ā§ Ļ j , dĻji = āĻki ā§ Ļjk + Ļ i ā§ Ļ j .
342
11. Curves and surfaces in elliptic and hyperbolic spaces
The second equation in (11.6) implies that the curvature (11.2) of the connection matrix Ļc = [Ļji ] is given by Ī© = [Ī©ij ] = [Ļ i ā§ Ļ j ]. The nonzero entries of Ī© reļ¬ect the fact that all sectional curvatures of Sn are identically equal to 1 (see, e.g., [dC92]). (c) Show that the structure equations (11.6) are equivalent to the MaurerCartan equation dĻ = āĻ ā§ Ļ, where
ā”
0 āĻ 1 Ā· Ā· Ā· āĻ n
ā¤
ā¢Ļ1 Ļ1 Ā· Ā· Ā· Ļ1 ā„ ā¢ n ā„ 1 Ļ=ā¢ . .. .. ā„ ā£ .. . . ā¦ Ļ n Ļ1n Ā· Ā· Ā· Ļnn is the so(n + 1)-valued Maurer-Cartan form on SO(n + 1). 11.2.2. Hyperbolic space Hn . Deļ¬nition 11.6. The n-dimensional hyperbolic space Hn is the upper sheet of the two-sheeted hyperboloid (x0 )2 ā (x1 )2 ā Ā· Ā· Ā· ā (xn )2 = 1 in M1,n ; i.e., Hn = {x ā M1,n | x, x = 1,
x0 > 0}.
Hn is a spacelike hypersurface in M1,n , so the restriction of the Minkowski metric on M1,n deļ¬nes a Riemannian metric on Hn . (Technically, this requires reversing signs in the deļ¬nition of the Minkowski inner product so that its restriction to each tangent space Tx Hn yields a quadratic form that is positive deļ¬nite rather than negative deļ¬nite, but for the sake of consistency we will retain the sign conventions of Chapter 5.) The symmetry group of Hn is deļ¬ned to be the subgroup of M (1, n) that preserves the set Hn . *Exercise 11.7. Show that the symmetry group of Hn consists of all matrices AĖ ā GL(n + 2) of the form t 1 0 (11.7) AĖ = , 0A where A ā SO+ (1, n). Therefore, the symmetry group of Hn is isomorphic to SO+ (1, n), and we will generally identify the element AĖ in equation (11.7) with the corresponding element A ā SO + (1, n).
11.2. The homogeneous spaces Sn and Hn
343
In order to describe Hn as a homogeneous space of the Lie group SO + (1, n), we need to compute the isotropy group of a point x ā Hn . *Exercise 11.8. Let x0 = t[1, 0, . . . , 0] ā Hn . Show that: (a) The isotropy group Hx0 of x0 in SO + (1, n) is ) ( t 1 0 ĀÆ (11.8) Hx0 = : A ā SO(n) . 0 AĀÆ (b) The isotropy group Hx of any other point x ā Hn is Hx = tx Hx0 tā1 x , where tx is any matrix in SO+ (1, n) whose ļ¬rst column is x. (The transformation tx will then have the property that tx (x0 ) = x.) Orthonormal frames on Hn are deļ¬ned as follows: Deļ¬nition 11.9. An orthonormal frame f on Hn is a list of vectors f = (e0 , . . . , en ), where e0 ā Hn ā M1,n , e1 , . . . , en ā M1,n , and e0 Ā· Ā· Ā· en ā SO+ (1, n). We identify the position vector x ā Hn , and the con e0 with + dition that e0 Ā· Ā· Ā· en ā SO (1, n) implies that the vectors (e1 , . . . , en ) may be regarded as an orthonormal basis for the tangent space Te0 Hn . (We may also say that (e1 , . . . , en ) is an orthonormal frame based at e0 .) The collection of all orthonormal frames on Hn is called the orthonormal frame bundle of Hn , denoted F (Hn ). Remark 11.10. For an orthonormal frame f = (e0 , . . . , en ) on Sn , each of the vectors eĪ± ā En+1 satisļ¬es eĪ± , eĪ± = 1, and so could, if desired, be identiļ¬ed with a point of Sn . But for an orthonormal frame on Hn , only the vector e0 satisļ¬es the deļ¬ning condition e0 , e0 = 1 for points in Hn , while the vectors (e1 , . . . , en ) each satisfy ei , ei = ā1. This illustrates the fact that for non-ļ¬at homogeneous spaces (and for Riemannian manifolds in general), we will no longer be able to regard the frame vectors (e1 , . . . , en ) as taking values in the same space as the position vector e0 , as we did for frames on ļ¬at homogeneous spaces. Instead, we must regard each of the vectors (e1 , . . . , en ) as taking values in the tangent bundle of the underlying manifold. The same reasoning as in prior cases shows that the orthonormal frame bundle F (Hn ) may be regarded as the group SO+ (1, n) via the one-to-one correspondence
g(e0 , . . . , en ) = e0 Ā· Ā· Ā· en . The projection map Ļ : SO + (1, n) ā Hn deļ¬ned by Ļ([e0 . . . en ]) = e0
344
11. Curves and surfaces in elliptic and hyperbolic spaces
describes SO+ (1, n) as a principal bundle over Hn with ļ¬ber group SO(n); therefore, we have a natural correspondence Hn ā¼ = SO + (1, n)/SO(n). Because we have deļ¬ned the components (e0 , . . . , en ) of a frame as elements of the vector space M1,n , we can regard them as functions eĪ± : F (Hn ) ā M1,n and deļ¬ne the Maurer-Cartan forms (ĻĪ²Ī± ) on SO+ (1, n) as usual by the equations (11.9)
deĪ± = eĪ² ĻĪ±Ī² ,
where 0 ā¤ Ī±, Ī² ā¤ n. Recall from Exercise 3.50 that these forms satisfy the relations ā§ āŖ Ī± = Ī², āØ0, Ī² Ī± ĻĪ± = ĻĪ² , Ī± = 0 or Ī² = 0, āŖ ā© Ī± āĻĪ² , Ī±, Ī² ā„ 1. *Exercise 11.11. (a) Show that the forms (Ļ0i ) for 1 ā¤ i ā¤ n are semi-basic for the projection Ļ : SO + (1, n) ā Hn and so may be regarded as the dual forms of any orthonormal frame (e1 , . . . , en ) based at e0 ā Hn . The forms (Ļji ) for 1 ā¤ i, j ā¤ n may then be regarded as the connection forms. (b) Set Ļ i = Ļ0i . Show that the dual forms (Ļ i ) and the connection forms (Ļji ) satisfy the structure equations (11.10)
dĻ i = āĻji ā§ Ļ j , dĻji = āĻki ā§ Ļjk ā Ļ i ā§ Ļ j .
(Compare with equations (11.6).) The second equation in (11.10) implies that the curvature (11.2) of the connection matrix Ļc = [Ļji ] is given by Ī© = [Ī©ij ] = [āĻ i ā§ Ļ j ]. The nonzero entries of Ī© reļ¬ect the fact that all sectional curvatures of Hn are identically equal to ā1. (c) Show that the structure equations (11.10) are equivalent to the MaurerCartan equation dĻ = āĻ ā§ Ļ, where
ā”
0 Ļ1 Ā· Ā· Ā· Ļn
ā¢Ļ1 ā¢ Ļ=ā¢ . ā£ ..
Ļ11 Ā· Ā· Ā· .. .
ā¤
Ļn1 ā„ ā„ .. ā„ . ā¦
Ļ n Ļ1n Ā· Ā· Ā· Ļnn is the so+ (1, n)-valued Maurer-Cartan form on SO+ (1, n).
11.3. A more intrinsic view of Sn and Hn
345
11.3. A more intrinsic view of Sn and Hn While it is certainly useful to regard Sn and Hn as submanifolds of the ļ¬at metric spaces En+1 and M1,n , it is also somewhat artiļ¬cial. These manifolds are perfectly well-deļ¬ned as intrinsic n-dimensional homogeneous spaces of the Lie groups SO(n + 1) and SO+ (1, n), respectively, and they shouldnāt need to be extrinsically embedded as submanifolds of larger spaces in order to study their geometry. For ease of exposition, let Xn denote either Sn or Hn ; let Vn+1 denote the ļ¬at homogeneous space in which Xn is deļ¬ned as an embedded hypersurface (En+1 or M1,n ), and let G denote the symmetry group of Xn (SO(n + 1) or SO+ (1, n)). Each tangent space Tx Xn ā Tx Vn+1 ā¼ = Vn+1 inherits an inner product Ā·, Ā· from the inner product (either Euclidean or Minkowski) on Vn+1 , thereby giving Xn the structure of a Riemannian manifold. We can give an intrinsic deļ¬nition for orthonormal frames on Xn that is equivalent to Deļ¬nitions 11.4 and 11.9, as follows. (Note that the condition
det e0 Ā· Ā· Ā· en = 1 implied by these deļ¬nitions induces an orientation on each tangent space Te0 Xn .) Deļ¬nition 11.12. An (oriented) orthonormal frame f on Xn is a list f = (e0 , . . . , en ), where e0 ā Xn and (e1 , . . ., en ) is an oriented, orthonormal basis for the tangent space Te0 Xn . Alternatively, we may say that (e1 , . . . , en ) is an orthonormal frame based at e0 . The collection of all orthonormal frames on Xn is called the orthonormal frame bundle of Xn and is denoted F (Xn ); it may be identiļ¬ed with the Lie group G. The Maurer-Cartan form Ļ and its components (Ļ i , Ļji ) are deļ¬ned as usual on G, and the structure equations (11.6) and (11.10) remain valid in the intrinsic setting as consequences of the Maurer-Cartan equation on G. The primary issue that must be addressed is how to make sense of the idea of diļ¬erentiating vector ļ¬elds on Xn . We can no longer think of the components (e0 , . . . , en ) of an orthonormal frame as functions from F (Xn ) to Vn+1 ; rather, e0 is a function from F (Xn ) to Xn , while (e1 , . . . , en ) are functions from F (Xn ) to the tangent bundle T Xn , with the property that for any frame f = (e0 , . . . , en ) ā F (Xn ), we have ei (f ) ā Te0 (f ) Xn for 1 ā¤ i ā¤ n. Moreover, the derivatives of the functions ei : F (Xn ) ā T Xn at a point f ā F (Xn ) must also take values in the tangent space Te0 (f ) Xn and hence must be linear combinations of the vectors (e1 (f ), . . . , en (f )). This is manifestly not the case for the extrinsically deļ¬ned exterior derivatives of
346
11. Curves and surfaces in elliptic and hyperbolic spaces
equations (11.5) and (11.9), as each of the exterior derivatives de1 , . . . , den contains a nonzero e0 term. It turns out that the way to solve this problem is simply to take the orthogonal projection of these extrinsically deļ¬ned exterior derivatives onto the tangent plane Te0 Xn . This idea leads to the notion of the covariant derivative for vector ļ¬elds on Xn : ā¼ Deļ¬nition 11.13. Let e0 ā Xn ā Vn+1 , and let w ā Te0 Xn ā Te0 Vn+1 = n+1 n V . Let v be a (tangent) vector ļ¬eld on X , regarded as a function v : Xn ā Vn+1 with the property that for every x ā Xn , we have v(x) ā Tx Xn ā Vn+1 . Let Ļe0 : Vn+1 ā Te0 Xn denote orthogonal projection with respect to the (Euclidean or Minkowski) metric on Vn+1 . The covariant derivative of v with respect to w is the vector āw v ā Te0 Xn deļ¬ned by (11.11)
āw v = Ļe0 (dv(w)).
In particular, if we take v(x) = ei (x) (1 ā¤ i ā¤ n) for some orthonormal frame ļ¬eld (e1 (x), . . . , en (x)) on Xn , then we have āw ei = Ļe0 (dei (w)) = ej Ļ ĀÆ ij (w), where the repeated index j is summed from 1 to n and Ļ ĀÆ ji represents the pullback of the Maurer-Cartan form Ļji on F (Xn ) to Xn via the frame ļ¬eld (e1 (x), . . . , en (x)). Moreover, the Leibniz rule for exterior derivatives implies that for any vector ļ¬eld v(x) = v i (x)ei (x) on Xn , we have āw v = w(v i )ei (x) + v i āw ei (x) (11.12)
= w(v i )ei (x) + v i ej (x)ĀÆ Ļij (w) = ei (x) w(v i ) + v j Ļ ĀÆ ji (w) .
The remarkable fact about equation (11.12) is that, even though we used extrinsic objects to deļ¬ne the covariant derivative, the result is described entirely in intrinsic terms: The tangent vector ļ¬elds (ei (x)) and the pulledback Maurer-Cartan forms (ĀÆ Ļji ) are well-deļ¬ned on Xn as a homogeneous space of the Lie group G, without regard to its embedding as a submanifold of Vn+1 . So, unlike the exterior derivative of equations (11.5) and (11.9), the covariant derivative is intrinsically deļ¬ned on Xn . The following exercise shows how the covariant derivative may be thought of as an analog to the exterior derivative for vector ļ¬elds on Xn . *Exercise 11.14. Given a vector ļ¬eld v on Xn , consider the map āv : T Xn ā T Xn
11.3. A more intrinsic view of Sn and Hn
347
deļ¬ned by (11.13)
āv(w) = āw v.
Use the explicit formula (11.12) to show that āv is a T Xn -valued 1-form on Xn (cf. Deļ¬nition 2.18). Speciļ¬cally, if (e1 (x), . . . , en (x)) is an orthonormal frame ļ¬eld on Xn with associated connection forms (ĀÆ Ļji ), then āv = ei dv i + v j Ļ ĀÆ ji . In particular, if v(x) = ei (x), then we have (11.14)
āei (x) = ej (x) Ļ ĀÆ ij .
Thus, we can think of ā as a generalization of the exterior derivative d that appears in the structure equations (3.1). An important property of the covariant derivative is that it is compatible with the metric on Xn . This means that for any vector ļ¬elds v1 , v2 on Xn and any vector w ā Tx Xn , we have (11.15)
w(v1 , v2 ) = āw v1 , v2 + v1 , āw v2 .
(You can think of this as a Leibniz rule for computing the directional derivative of the real-valued function v1 , v2 in the direction of w.) *Exercise 11.15. Use equation (11.12) to show that equation (11.15) holds. (Hint: You will need to use the fact that Ļ ĀÆ ij = āĀÆ Ļji for 1 ā¤ i, j ā¤ n, which n is true for the connection forms on both S and Hn .) Deļ¬nition 11.16. Let Ī(T Xn ) denote the space of smooth local sections of T Xn (i.e., smooth vector ļ¬elds on open sets in Xn ). The map ā : Ī(T Xn ) Ć Ī(T Xn ) ā Ī(T Xn ) deļ¬ned by ā(v, w)(x) = āw(x) v ā Tx Xn is called the Levi-Civita connection on Xn . The 1-forms (ĀÆ Ļji ) determined by the frame ļ¬eld (e1 (x), . . . , en (x)) and equation (11.14) are called the connection forms associated to ā and the frame ļ¬eld (e1 (x), . . . , en (x)). (Note that this terminology is consistent with their deļ¬nition as the pullbacks of the connection forms on the frame bundle G ā Xn via the frame ļ¬eld.) Remark 11.17. Although the Levi-Civita connection is deļ¬ned as an operator on vector ļ¬elds on Xn , equation (11.14) suggestsācorrectly!āthat there should be a related operator (also denoted ā) on an appropriate class of smooth maps from F (Xn ) to T Xn , determined by the condition (11.16)
āei = ej Ļij
348
11. Curves and surfaces in elliptic and hyperbolic spaces
Ė : F (Xn ) ā T Xn via together with an extension to appropriate maps v linearity and a Leibniz rule akin to equation (11.12). Once all the details are worked out, equation (11.16) is the natural analog of the structure equations (3.1). We will develop this idea more fully in Chapter 12. One of the most important diļ¬erences between the Levi-Civita operator ā and the exterior derivative d is that there is no analog of the identity dā¦d = 0 for ā, and so it is not immediately clear how to diļ¬erentiate equation (11.16) in order to obtain structure equations for the derivatives (dĻji ). Fortunately, we have already derived these for the cases at hand via extrinsic techniques (cf. equations (11.6) and (11.10)). In Chapter 12, we will see how to derive these equations more intrinsically and in more general scenarios. For the remainder of this chapter, we will restrict our consideration to the case n = 3. We will continue to use the notation (X3 , V4 , G) to denote either (S3 , E4 , SO(4)) or (H3 , M1,3 , SO + (1, 3)).
11.4. Moving frames for curves in S3 and H3 Consider a smooth, parametrized curve Ī± : I ā X3 that maps some open interval I ā R into X3 . Since X3 has the structure of the homogeneous space G/SO(3), an adapted frame ļ¬eld along Ī± should be a lifting Ī± Ė : I ā G. Any such lifting can be written as Ī± Ė (t) = (e0 (t), e1 (t), e2 (t), e3 (t)), where for each t ā I, e0 (t) = Ī±(t) and (e1 (t), e2 (t), e3 (t)) is an oriented, orthonormal basis for the tangent space TĪ±(t) X3 . Such an adapted frame ļ¬eld is usually called an orthonormal frame ļ¬eld along Ī±. As in the Euclidean case, we say that Ī± is regular if Ī± (t) = 0 for every t ā I; as usual, we will only consider regular curves. We begin our construction of an adapted orthonormal frame ļ¬eld along Ī± by setting e1 (t) =
Ī± (t) ; |Ī± (t)|
i.e., we require that e1 (t) be the unit tangent vector to the curve at Ī±(t). The same argument as in the Euclidean case shows that Ī± can be smoothly reparametrized by its arc length function s(t), so henceforth we will assume that Ī± = Ī±(s) is parametrized by arc length and that e1 (s) = Ī± (s). Note that this makes sense even in the intrinsic setting: Since Ī±(s) ā X3 , the derivative e1 (s) = Ī± (s) is an element of TĪ±(s) X3 , which is where we expect the frame vectors to live.
11.4. Moving frames for curves in S3 and H3
349
The next step is where things start to look a bit diļ¬erent from the Euclidean case: Since e1 (s) is a vector ļ¬eld along Ī±, we have to use the covariant derivative to diļ¬erentiate it. In particular, diļ¬erentiating any vector ļ¬eld along the curve Ī± means taking its covariant derivative with respect to the unit tangent vector ļ¬eld along Ī±. So the natural analog for the Euclidean derivative e1 (s) is the covariant derivative āe1 (s) e1 (s). We will say that Ī± is nondegenerate if Ī± is regular and, in addition, āe1 (s) e1 (s) = 0 for all s ā I. *Exercise 11.18. In the Euclidean case, any regular curve Ī± : I ā E3 with e1 (s) = Ī± (s) = 0 for all s ā I is contained in a straight line in E3 . Consider the analogous condition for curves in X3 : Let Ī± : I ā X3 be a regular curve parametrized by arc length; let e0 (s) = Ī±(s), e1 (s) = Ī± (s), and suppose that āe1 (s) e1 (s) = 0 for all s ā I. (a) Use the extrinsic deļ¬nition (11.11) together with the structure equations (11.5) and (11.9) to show that, when regarded as functions e0 , e1 : I ā V4 , we have (11.17)
e0 (s) = e1 (s),
e1 (s) = k(s)e0 (s),
where k(s) = Ļ10 (e1 (s)). (b) Use the fact that Ļ10 = Ā±Ļ01 = Ā±Ļ 1 (with the sign depending on whether X3 = S3 or H3 ) to show that ā1, X3 = S3 , k(s) = 1, X3 = H3 . (c) Solve the diļ¬erential equations (11.17) and show that: ĀÆ0 , e ĀÆ1 ā E4 (1) If X3 = S3 , then there exist orthogonal unit vectors e such that (11.18)
ĀÆ0 + sin(s) e ĀÆ1 . Ī±(s) = e0 (s) = cos(s) e In particular, Ī± is contained in the great circle determined by the ĀÆ1 ). intersection of S3 with the plane spanned by (ĀÆ e0 , e
ĀÆ0 , e ĀÆ1 ā M1,3 , (2) If X3 = H3 , then there exist orthogonal unit vectors e ĀÆ0 timelike and e ĀÆ1 spacelike, such that with e (11.19)
ĀÆ0 + sinh(s) e ĀÆ1 . Ī±(s) = e0 (s) = cosh(s) e In particular, Ī± is contained in the āgreat hyperbolaā determined by ĀÆ1 ). the intersection of H3 with the timelike plane spanned by (ĀÆ e0 , e
The diļ¬erential equation āĪ± (s) Ī± (s) = 0 is called the geodesic equation, and the curves Ī± in equations (11.18) and (11.19) are the geodesics in S3 and H3 , respectively.
350
11. Curves and surfaces in elliptic and hyperbolic spaces
Now, suppose that Ī± : I ā X3 is a nondegenerate curve parametrized by arc length. According to equation (11.15), diļ¬erentiating the equation e1 (s), e1 (s) = 1 along Ī± yields āe1 (s) e1 (s), e1 (s) = 0. Thus, āe1 (s) e1 (s) is orthogonal to e1 (s), and we deļ¬ne e2 (s) =
āe1 (s) e1 (s) . |āe1 (s) e1 (s)|
This vector will be called the unit normal vector to the curve at Ī±(s). The adapted frame ļ¬eld is now uniquely determined: Because the frame must be oriented and orthonormal, e3 (s) is uniquely determined by the condition that e3 (s) = e1 (s) Ć e2 (s). This makes sense because e1 (s) and e2 (s) are elements of the oriented, 3-dimensional Euclidean vector space TĪ±(s) X3 , where the cross product is welldeļ¬ned. The vector e3 (s) is called the binormal vector to the curve at Ī±(s). The adapted frame ļ¬eld (e1 (s), e2 (s), e3 (s)) is called the Frenet frame of the curve Ī±(s); it determines a canonical, left-invariant lifting Ī± Ė : I ā G given by Ī±(s) Ė = (e0 (s), e1 (s), e2 (s), e3 (s)), where e0 (s) = Ī±(s), for any nondegenerate curve Ī± in X3 parametrized by arc length. *Exercise 11.19. Show that we have the following analog of the Frenet equations for nondegenerate curves Ī± : I ā X3 parametrized by arc length:
(11.20) Ī± (s) āe1 (s) e1 (s) āe1 (s) e2 (s) āe1 (s) e3 (s) ā” ā¤ 0 0 0 0 ā¢ ā„
ā¢1 0 āĪŗ(s) 0 ā„ ā¢ ā„, = Ī±(s) e1 (s) e2 (s) e3 (s) ā¢ ā„ 0 Īŗ(s) 0 āĻ (s) ā£ ā¦ 0 0 Ļ (s) 0 where Īŗ, Ļ : I ā R are smooth functions along Ī± with Īŗ(s) = |āe1 (s) e1 (s)| > 0. (Hint: Equation (11.14) might be helpful.) As in the Euclidean case, the functions Īŗ(s), Ļ (s) are called the curvature and torsion, respectively, of Ī±.
11.5. Moving frames for surfaces in S3 and H3
351
Exercise 11.20. Let Ī± : I ā X3 ā V4 be a nondegenerate curve parametrized by arc length. (a) Show that, when regarded as functions e0 , e1 , e2 , e3 ā” 0 ā¢
ā¢1 e0 (s) e1 (s) e2 (s) e3 (s) = e0 (s) e1 (s) e2 (s) e3 (s) ā¢ ā¢0 ā£ 0
: I ā V4 , we have ā¤ Ā±1 0 0 ā„ 0 āĪŗ(s) 0 ā„ ā„, Īŗ(s) 0 āĻ (s)ā„ ā¦ 0 Ļ (s) 0
with the sign depending on whether X3 = S3 or H3 . ĀÆ0 , e ĀÆ1 , e ĀÆ2 ā V4 such that Ī± (b) Show that there exist orthogonal unit vectors e 3 ĀÆ1 , e ĀÆ2 ) if is contained in the intersection of X with the plane spanned by (ĀÆ e0 , e and only if the torsion Ļ (s) is identically zero. (This intersection is a āgreat sphereā when X3 = S3 and a āgreat hyperboloidā when X3 = H3 .) This is the analog of the fact that a nondegenerate curve in E3 is contained in a plane if and only if its torsion Ļ (s) is identically zero.
11.5. Moving frames for surfaces in S3 and H3 Now, let U be an open set in R2 , and let x : U ā X3 be an immersion whose image is a surface Ī£ = x(U ). Just as for curves, an adapted frame ļ¬eld Ė : U ā G of the form along Ī£ is a lifting x Ė (u) = (e0 (u), e1 (u), e2 (u), e3 (u)) , x where for each u ā U , e0 (u) = x(u) and (e1 (u), e2 (u), e3 (u)) is an oriented, orthonormal basis for the tangent space Tx(u) X3 . *Exercise 11.21. Convince yourself that the following statements, which we proved for surfaces in E3 in Chapter 4, remain true for a surface Ī£ = x(U ) ā X3 . (In particular, this means that none of the following constructions relied on the ļ¬atness of E3 or involved diļ¬erentiating vector ļ¬elds on E3 .) (a) We can choose an orthonormal frame (e1 (u), e2 (u), e3 (u)) for each tangent space Tx(u) X3 so that that e3 (u) is orthogonal to Tx(u) Ī£ and (e1 (u), Ė : U ā G by e2 (u)) span Tx(u) Ī£. Any such choice deļ¬nes a lifting x Ė (u) = (e0 (u), e1 (u), e2 (u), e3 (u)) , x where e0 (u) = x(u). (b) Let (ĀÆ Ļi, Ļ ĀÆ ji ) denote the pullbacks of the Maurer-Cartan forms (Ļ i , Ļji ) Ė . Then we have Ļ on G to U via x ĀÆ 3 = 0, and the 1-forms (ĀÆ Ļ1, Ļ ĀÆ 2 ) form a basis for the 1-forms on U .
352
11. Curves and surfaces in elliptic and hyperbolic spaces
(c) The metric on X3 naturally induces a metric on Ī£ = x(U ) ā X3 , given by the ļ¬rst fundamental form I = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 . (d) Diļ¬erentiating the equation Ļ ĀÆ 3 = 0 and applying Cartanās lemma implies that there exist functions h11 , h12 , h22 on U such that 3 1 Ļ ĀÆ1 h11 h12 Ļ ĀÆ = . Ļ ĀÆ 23 h12 h22 Ļ ĀÆ2 Once again, the functions (hij ) are related to the derivative of the Gauss map on Ī£. However, there are two important diļ¬erences: (1) Since there are no canonical isomorphisms between the individual tangent spaces Tx(u) X3 , we cannot view the Gauss map as a map from Ī£ to the unit sphere in E3 ; rather, it is a map from Ī£ to the tangent bundle T X3 . (2) In order to deļ¬ne the second fundamental form, we must use the covariant derivative to diļ¬erentiate the Gauss map. Deļ¬nition 11.22. Let U ā R2 be an open set, and let x : U ā X3 be an immersion with image Ī£ = x(U ). The Gauss map of Ī£ = x(U ) is the map N : Ī£ ā T X3 deļ¬ned by N (x(u)) = e3 (u) ā Tx(u) X3 , where (e1 (u), e2 (u), e3 (u)) is any adapted frame ļ¬eld on Ī£ = x(U ). Deļ¬nition 11.23. Let U ā R2 be an open set, and let x : U ā X3 be an immersion. The second fundamental form of Ī£ = x(U ) is the quadratic form II on T U deļ¬ned by II(v) = āāv e3 , dx(v) for v ā Tu U , where (e1 (u), e2 (u), e3 (u)) is any adapted frame ļ¬eld on Ī£ = x(U ). *Exercise 11.24. Show that II = Ļ ĀÆ 13 Ļ ĀÆ1 + Ļ ĀÆ 23 Ļ ĀÆ 2 = h11 (ĀÆ Ļ 1 )2 + 2h12 Ļ ĀÆ1 Ļ ĀÆ 2 + h22 (ĀÆ Ļ 2 )2 . As in the Euclidean case, the eigenvalues Īŗ1 (u), Īŗ2 (u) of the matrix [hij (u)] are called the principal curvatures of Ī£ at the point x(u), and the eigenvectors of the self-adjoint map ādNx(u) : Tx(u) Ī£ ā Tx(u) Ī£ are called the principal vectors or principal directions of Ī£ at the point x(u).
11.5. Moving frames for surfaces in S3 and H3
Deļ¬nition 11.25. The functions ĀÆ = Īŗ1 Īŗ2 , (11.21) K
353
H = 12 (Īŗ1 + Īŗ2 )
on U are called the extrinsic curvature and the mean curvature, respectively, of Ī£. The Gauss curvature K of Ī£ (also called the intrinsic curvature of Ī£) is the function on U deļ¬ned by the condition that dĀÆ Ļ21 = K Ļ ĀÆ1 ā§ Ļ ĀÆ 2, where (ĀÆ Ļi, Ļ ĀÆ ji ) are the Maurer-Cartan forms associated to any adapted frame ļ¬eld on Ī£. ĀÆ and H are all well*Exercise 11.26. (a) Show that the functions K, K, deļ¬ned on U , independent of the choice of adapted frame ļ¬eld. (b) Use the structure equations (11.6) and (11.10) to show that: ĀÆ + 1. (1) If X3 = S3 , then K = K ĀÆ ā 1. (2) If X3 = H3 , then K = K So, unlike in the Euclidean case where the Gauss equation implies that ĀÆ here these two notions of curvature diļ¬er by the sectional curvature K = K, of the underlying homogeneous space. As in the Euclidean case, the pullbacks to U of the structure equations for the derivatives (dĀÆ Ļji ) in equations (11.6) and (11.10) are called the Gauss and Codazzi equations. *Exercise 11.27. Show that the Gauss and Codazzi equations take the form dĀÆ Ļ21 = Ļ ĀÆ 13 ā§ Ļ ĀÆ 23 Ā± Ļ ĀÆ1 ā§ Ļ ĀÆ 2, (11.22)
dĀÆ Ļ13 = Ļ ĀÆ 23 ā§ Ļ ĀÆ 21 , dĀÆ Ļ23 = āĀÆ Ļ13 ā§ Ļ ĀÆ 21 ,
with the sign in the ļ¬rst equation depending on whether X3 = S3 or H3 . *Exercise 11.28. Suppose that x : U ā X3 is an immersion whose coordinate curves are all principal curves. Then the ļ¬rst and second fundamental forms may be written as I = E du2 + G dv 2 , II = e du2 + g dv 2 . (a) Show that the principal adapted frame ļ¬eld 1 1 e1 (u) = ā xu , e2 (u) = ā xv , e3 (u) = e1 (u) Ć e2 (u) E G
354
11. Curves and surfaces in elliptic and hyperbolic spaces
has associated Maurer-Cartan forms ā ā Ļ ĀÆ 1 = E du, Ļ ĀÆ 2 = G dv, Ļ ĀÆ 3 = 0, 1 Ļ ĀÆ 21 = ā (Ev du ā Gu dv), 2 EG e 1 e Ļ ĀÆ 13 = Ļ ĀÆ = ā du, E E g g Ļ ĀÆ 23 = Ļ ĀÆ 2 = ā dv. G G (b) Show that the Gauss equation is equivalent to " " # # eg Gu 1 Ev ā (11.23) , + ā Ā±1=ā ā EG 2 EG EG v EG u with the sign on the left-hand side depending on whether X3 is equal to S3 or H3 . (c) Show that the Codazzi equations are equivalent to e 1 g! ev = Ev + , 2 E G (11.24) e 1 g! g u = Gu + 2 E G (cf. Exercises 4.24 and 4.41). As for surfaces in E3 , the ļ¬rst and second fundamental forms are invariants of the surface. Consequently, if two surfaces have diļ¬erent ļ¬rst and second fundamental forms, then they cannot be equivalent via an isometry of M . Lemmas 4.2 and 4.12 imply that the converse is true as well, and we have the following analog of Bonnetās theorem: Theorem 11.29. Let (ĀÆ Ļ1, Ļ ĀÆ 2, Ļ ĀÆ 13 , Ļ ĀÆ 23 ) be 1-forms on a connected and simply 2 connected open set U ā R satisfying the conditions that (ĀÆ Ļ1, Ļ ĀÆ 2 ) are linearly independent at each point of U and that Ļ ĀÆ i3 is a scalar multiple of Ļ ĀÆi for i = 1, 2. Suppose that, together with the Levi-Civita connection form Ļ ĀÆ 21 1 2 determined by Ļ ĀÆ and Ļ ĀÆ , these forms satisfy the Gauss and Codazzi equations (11.22). Then there exists an immersed surface x : U ā X3 , unique up to transformation by an element of G, whose ļ¬rst and second fundamental forms are I = (ĀÆ Ļ 1 )2 + (ĀÆ Ļ 2 )2 , II = Ļ ĀÆ 13 Ļ ĀÆ1 + Ļ ĀÆ 23 Ļ ĀÆ 2. We will conclude this chapter by exploring some special families of surfaces in S3 and H3 .
11.5. Moving frames for surfaces in S3 and H3
355
Deļ¬nition 11.30. A regular surface Ī£ = x(U ) ā X3 is called totally geodesic if every geodesic in Ī£ is also a geodesic in X3 . It turns out that Ī£ is totally geodesic if and only if (11.25)
āw e3 = 0 for all w ā T Ī£,
where e3 is a unit normal vector ļ¬eld along Ī£. Remark 11.31. By way of comparison, for a surface Ī£ ā E3 , the totally geodesic condition says that every geodesic in Ī£ is a straight line in E3 , which implies that Ī£ is contained in a plane in E3 . Meanwhile, the condition (11.25) says that the normal vector ļ¬eld e3 to Ī£ is constant along Ī£, which also implies that Ī£ is contained in a plane in E3 . Exercise 11.32. Suppose that Ī£ = x(U ) ā X3 is a totally geodesic surface in X3 . (a) Show that the condition (11.25) is equivalent to the condition that the Maurer-Cartan forms associated to any adapted frame ļ¬eld along Ī£ satisfy (11.26)
Ļ ĀÆ 13 = Ļ ĀÆ 23 = 0.
(b) Use the extrinsic deļ¬nition (11.11) together with the structure equations (11.5) and (11.9) to show that, when regarded as functions e0 , e1 , e2 , e3 : U ā V4 , equation (11.26) implies that de3 = 0. ĀÆ0 , e ĀÆ1 , e ĀÆ2 ā V4 such that Conclude that there exist orthogonal unit vectors e Ī£ = x(U ) is contained in the āgreat sphereā of S3 or the āgreat hyperboloidā of H3 determined by the intersection of X3 with the plane spanned ĀÆ1 , e ĀÆ2 ) (cf. Exercise 11.20). This is the analog of the fact that any by (ĀÆ e0 , e totally geodesic surface in E3 is contained in a plane. Deļ¬nition 11.33. A surface Ī£ = x(U ) ā X3 is called ļ¬at if its Gauss curvature K is identically zero. Exercise 11.34. Let Ī£ be a ļ¬at surface in S3 . (a) Show that every point x0 ā Ī£ has a neighborhood for which there exists an asymptotic parametrization x : U ā S3 of Ī£ such that the ļ¬rst and second fundamental forms of Ī£ = x(U ) are given by (11.27)
I = dx2 + 2 cos(2Ļ) dx dy + dy 2 , II = 2 sin(2Ļ) dx dy,
where the function Ļ : U ā R, called the angle function, satisļ¬es the wave equation Ļxy = 0. (Hint: Observe that Īŗ1 Īŗ2 = ā1, and adapt the construction of Ā§9.4.) Conversely, Theorem 11.29 implies that any solution Ļ(x, y)
356
11. Curves and surfaces in elliptic and hyperbolic spaces
of the wave equation gives rise to a ļ¬at surface in S3 whose ļ¬rst and second fundamental forms are given by (11.27). (b) Let a, b > 0 with a2 +b2 = 1, and consider the map x : S1 ĆS1 ā S3 ā E4 deļ¬ned by x(Īø, Ļ) = t [a cos(Īø), a sin(Īø), āb cos(Ļ), b sin(Ļ)] , where Īø and Ļ denote angle coordinates on the two copies of S1 . Show that Ī£ = x(S1 Ć S1 ) is a ļ¬at torus in S3 , with angle function equal to a constant Ļ0 such that cos(Ļ0 ) = a, sin(Ļ0 ) = b. (Hint: Begin by considering the principal adapted frame ļ¬eld e0 (Īø, Ļ) = x(Īø, Ļ) = t[a cos(Īø), a sin(Īø), āb cos(Ļ), b sin(Ļ)], e1 (Īø, Ļ) =
xĪø (Īø, Ļ) = t[ā sin(Īø), cos(Īø), 0, 0], |xĪø (Īø, Ļ)|
e2 (Īø, Ļ) =
xĻ (Īø, Ļ) = t[0, 0, sin(Ļ), cos(Ļ)], |xĻ (Īø, Ļ)|
e3 (Īø, Ļ) = t[āb cos(Īø), āb sin(Īø), āa cos(Ļ), a sin(Ļ)] and its associated Maurer-Cartan forms.) Exercise 11.35. Let Ī£ be a ļ¬at surface in H3 , and assume that Ī£ has no umbilic points (i.e., points where Īŗ1 = Īŗ2 ). (a) Show that every point x ā Ī£ has a neighborhood for which there exists a principal parametrization x : U ā H3 of Ī£ such that the ļ¬rst and second fundamental forms of Ī£ = x(U ) are given by (11.28)
I = cosh2 (Ļ) du2 + sinh2 (Ļ) dv 2 , II = sinh(Ļ) cosh(Ļ)(du2 + dv 2 ),
where the angle function Ļ : U ā R satisļ¬es Laplaceās equation Ļuu + Ļvv = 0. (Hint: Observe that Īŗ1 Īŗ2 = 1, and adapt the construction of Ā§9.4.) Conversely, Theorem 11.29 implies that any solution Ļ(u, v) of Laplaceās equation gives rise to a ļ¬at surface in H3 whose ļ¬rst and second fundamental forms are given by (11.28). (b) Let a, b > 0 with a2 ā b2 = 1, and consider the map x : R Ć S1 ā H3 ā M1,3 deļ¬ned by x(t, Ļ) = t[a cosh(t), a sinh(t), b cos(Ļ), b sin(Ļ)] , where t denotes a standard coordinate on R and Ļ denotes an angle coordinate on S1 . Show that Ī£ = x(R Ć S1 ) is a ļ¬at cylinder in H3 , with angle
11.6. Maple computations
357
function equal to a constant Ļ0 such that cosh(Ļ0 ) = a,
sinh(Ļ0 ) = b.
(Hint: Begin by considering the principal adapted frame ļ¬eld e0 (t, Ļ) = x(t, Ļ) = t[a cosh(t), a sinh(t), b cos(Ļ), b sin(Ļ)], e1 (t, Ļ) =
xt (t, Ļ) = t[sinh(t), cosh(t), 0, 0], |xt (t, Ļ)|
e2 (t, Ļ) =
xĻ (t, Ļ) = t[0, 0, ā sin(Ļ), cos(Ļ)], |xĻ (t, Ļ)|
e3 (t, Ļ) = t[āb cosh(t), āb sinh(t), āa cos(Ļ), āa sin(Ļ)] and its associated Maurer-Cartan forms.)
11.6. Maple computations We will need to set up separately for computations in S3 and H3 because the structure equations for their Maurer-Cartan forms are diļ¬erent due to the curvature terms. Here we will work through Exercise 11.34 regarding ļ¬at surfaces in S3 ; only minor modiļ¬cations are required for the computations for Exercise 11.35 regarding ļ¬at surfaces in H3 . (See the Maple worksheet for this chapter on the AMS webpage for details.) Exercise 11.34: After loading the Cartan and LinearAlgebra packages into Maple, declare the Maurer-Cartan forms on S3 , and tell Maple about their symmetries and structure equations: > Form(omega[1], omega[2], omega[3]); Form(omega[1,2], omega[3,1], omega[3,2]); > omega[1,1]:= 0; omega[2,2]:= 0; omega[3,3]:= 0; omega[2,1]:= -omega[1,2]; omega[1,3]:= -omega[3,1]; omega[2,3]:= -omega[3,2]; > for i from 1 to 3 do d(omega[i]):= -add(āomega[i,j] &Ė omega[j]ā, j=1..3); end do;
358
11. Curves and surfaces in elliptic and hyperbolic spaces
> d(omega[1,2]):= -omega[1,3] &Ė omega[3,2] + omega[1] &Ė omega[2]; d(omega[3,1]):= -omega[3,2] &Ė omega[2,1] + omega[3] &Ė omega[1]; d(omega[3,2]):= -omega[3,1] &Ė omega[1,2] + omega[3] &Ė omega[2]; Now, suppose that x : U ā S3 is a principal parametrization of a ļ¬at surface Ī£ ā S3 and that (e1 (u), e2 (u), e3 (u)) is a principal adapted frame ļ¬eld along Ī£. Since Ī£ is ļ¬at, the principal curvatures Īŗ1 (u), Īŗ2 (u) satisfy the condition Īŗ1 (u)Īŗ2 (u) = ā1. The same argument as that in Ā§9.4 shows that the coordinates (u, v) can be chosen in such a way that the associated dual forms are given by the following expressions for some function Ļ0 (u, v): > PDETools[declare](psi0(u,v)); > adaptedsub:= [ omega[1] = cos(psi0(u,v))*d(u), omega[2] = sin(psi0(u,v))*d(v), omega[3] = 0, omega[3,1] = sin(psi0(u,v))*d(u), omega[3,2] = -cos(psi0(u,v))*d(v), omega[1,2] = -diff(psi0(u,v), v)*d(u) - diff(psi0(u,v), u)*d(v)]; These forms satisfy the structure equations for the dual forms, as well as the Codazzi equations for dĀÆ Ļ13 and dĀÆ Ļ23 . Now check the Gauss equation: > Simf(d(Simf(subs(adaptedsub, omega[1,2]))) + Simf(subs(adaptedsub, omega[1,3] &Ė omega[3,2] - omega[1] &Ė omega[2]))); (āĻ0v,v + Ļ0u,u ) (d(v)) &Ė (d(u)) So the Gauss equation is satisļ¬ed if and only if the function Ļ0 (u, v) satisļ¬es the wave equation (Ļ0 )uu ā (Ļ0 )vv = 0. Now make the change to asymptotic coordinates x=
(u + v) , 2
y=
(u ā v) , 2
or, equivalently, u = x + y,
v = x ā y.
Weāll need to remove Ļ ĀÆ 21 from the list in adaptedsub because Maple wonāt like this substitution in the expressions diff(psi0(u,v), u) and
11.6. Maple computations
359
diff(psi0(u,v), v). Weāll also introduce a new name for the function Ļ(x, y) = Ļ0 (x + y, x ā y). > PDETools[declare](psi(x,y)); > adaptedsub asymp:= Simf(subs([u = x + y, v = x - y], Simf(subs([psi0(u,v) = psi(x,y)], [seq(adaptedsub[i], i=1..5)])))); Maple doesnāt really know how to compute symmetric products of diļ¬erential forms, but the following commands will work for computing the ļ¬rst and second fundamental forms: > collect(simplify(Simf(subs(adaptedsub asymp, omega[1]))Ė2 + Simf(subs(adaptedsub asymp, omega[2]))Ė2), {d(x), d(y)}); > collect(simplify(Simf(subs(adaptedsub asymp, omega[3,1]))* Simf(subs(adaptedsub asymp, omega[1])) + Simf(subs(adaptedsub asymp, omega[3,2]))* Simf(subs(adaptedsub asymp, omega[2]))), {d(x), d(y)}); For Exercise 11.34(b), we can compute the Maurer-Cartan forms associated to the given frame ļ¬eld, as follows. First, declare a, b to be constants and deļ¬ne the frame vectors (e0 , e1 , e2 , e3 ): > Form(a=-1, b=-1); > e0:= Vector([a*cos(theta), a*sin(theta), -b*cos(phi), b*sin(phi)]); e1:= Vector([-sin(theta), cos(theta), 0,0]); e2:= Vector([0,0, sin(phi), cos(phi)]); e3:= Vector([-b*cos(theta), -b*sin(theta), -a*cos(phi), a*sin(phi)]); The fastest way to compute the Maurer-Cartan forms is to deļ¬ne the corresponding group element
g = e0 e1 e2 e3 ā SO(4) and compute the matrix-valued Maurer-Cartan form Ļ ĀÆ = g ā1 dg, as follows: > g:= Matrix([e0, e1, e2, e3]); connection matrix:= simplify(MatrixInverse(g).map(d, g));
360
11. Curves and surfaces in elliptic and hyperbolic spaces
ā”
ā¤ d(Ļ) d(Īø)a āb 0 0 ā ā¢ ā„ a2 + b 2 a2 + b 2 ā¢ ā„ ā¢ d(Īø)a 0 0 āb d(Īø)ā„ ā¢ ā„ ā„ connection matrix = ā¢ ā¢b d(Ļ) 0 0 a d(Ļ) ā„ ā¢ ā„ ā¢ ā„ ā£ ā¦ d(Īø) d(Ļ)a 0 b 2 ā 2 0 2 2 a +b a +b Since a2 + b2 = 1, we see that we have the following Maurer-Cartan forms: > examplesub:= [omega[1] = a*d(theta), omega[2] = b*d(phi), omega[3] = 0, omega[1,2] = 0, omega[3,1] = b*d(theta), omega[3,2] = -a*d(phi)]; These forms agree with the forms in adaptedsub, with cos(Ļ0 ) = a, sin(Ļ0 ) = b, as desired.
10.1090/gsm/178/12
Chapter 12
The nonhomogeneous case: Moving frames on Riemannian manifolds
12.1. Introduction So far, we have been using moving frames to study the geometry of curves and surfaces as submanifolds Ī£ of homogeneous spaces G/H. In this context, the geometry of Ī£ is determined by the geometry of the ambient homogeneous space G/H and the particular way that Ī£ is embedded in G/H as a submanifold. But there are many interesting geometric problems for which this scenario is too restrictive. For instance, we may be interested in the geometric structure of a nonhomogeneous manifold that is deļ¬ned intrinsically and not as a submanifold of some larger ambient homogeneous space. Or, even in the study of submanifolds, we might be interested in submanifolds Ī£ of some manifold M that is not homogeneous. Recall that for a homogeneous space G/H, the natural projection map Ļ : G ā G/H leads to a description of G as the frame bundle F (G/H) of the space G/H and that the set of frames over a given point x ā G/H is in one-to-one correspondence with the subgroup H of G. The fundamental property of homogeneous spaces is that for any two points x, y ā G/H and any frames 361
362
12. Moving frames on Riemannian manifolds
fx , fy based at the points x and y, respectively, there is a symmetry of the manifold G/H that takes x to y and fx to fy . In particular, the entire frame bundle F (G/H) is diļ¬eomorphic to G, and G acts freely and transitively on F (G/H). When we replace the homogeneous space G/H with a more general ndimensional manifold M , the situation is a bit diļ¬erent. We can still associate to each point x ā M a collection of frames (x; e1 , . . . , en ) for the tangent space Tx M , and the speciļ¬c collection will depend on what sort of geometric structure we want to consider on M . For instance, if M is a Riemannian manifold, we might consider the collection of frames that are orthonormal with respect to the Riemannian metric on M . Moreover, we still have a group action on the set of frames based at each point x ā M , simply because any two frames for the tangent space Tx M are related by an element of GL(n). In the case of a Riemannian manifold M , any two orthonormal frames based at a point x ā M are related by an element of the orthogonal group O(n); in fact, the set of orthonormal frames over each point x is in one-to-one correspondence with the Lie group H = O(n). What is diļ¬erent is that if x, y are distinct points of M , then there is no obvious relationship between the orthonormal frames based at x and those based at y and no obvious group action that can be used to transform a frame based at x to one based at y. The collection of all orthonormal frames based at all points of M forms a principal bundle F (M ) with ļ¬ber group H = O(n) (cf. Ā§1.5), but there may be no larger Lie group G that acts transitively on the entire frame bundle F (M ). In this chapter, we will illustrate this more general scenario by exploring how moving frames can be applied to study the geometry of a Riemannian manifold M and submanifolds Ī£ of M . Similar constructions can be applied to manifolds with other types of geometryāe.g., Lorentzian, equi-aļ¬ne, or projective manifolds.
12.2. Orthonormal frames and connections on Riemannian manifolds Deļ¬nition 12.1. A Riemannian manifold of dimension n is a smooth manifold of dimension n, together with a smoothly varying metric g on M (cf. Exercise 1.15). The metric g determines an inner product Ā·, Ā· on each tangent space Tx M deļ¬ned by v, w = g(v, w) for v, w ā Tx M .
12.2. Orthonormal frames and connections
363
Exercise 12.2. The phrase āsmoothly varyingā in Deļ¬nition 12.1 means 1 n that for any local coordinate system 2 3 x = (x , . . . , x ) on M , the functions ā ā gij (x) = g āxi , āxj | 1 ā¤ i, j, ā¤ n are smooth functions of x. Show that this condition is independent of the choice of local coordinates on M : If ĀÆ = (ĀÆ x x1 , . . . , x ĀÆn ) is another local coordinate system on M with smooth local coordinate transformation functions xi = xi (ĀÆ x1 , . . . , x ĀÆn ), then the functions # " ā ā gĀÆk (ĀÆ x) = g , āx ĀÆk ā x ĀÆ are smooth if and only if the original functions (gij ) are smooth. (Hint: The results of Exercise 1.22 should be helpful.) For simplicity, we will assume that the manifold M is oriented. Orthonormal frames on M are deļ¬ned almost exactly as they were on En (cf. Deļ¬nition 3.12); the only diļ¬erence is that now x is simply a point of M and, in general, not an element of a vector space. Deļ¬nition 12.3. An (oriented) orthonormal frame f on an oriented Riemannian manifold M is a list f = (x; e1 , . . . , en ) where x ā M and (e1 , . . ., en ) is an oriented, orthonormal basis for the tangent space Tx M . Alternatively, we may say that (e1 , . . . , en ) is an orthonormal frame based at x. The set of orthonormal frames at each point is in one-to-one correspondence with the Lie group SO(n), and the set of orthonormal frames on M forms a principal bundle over M with ļ¬ber SO(n), called the (oriented) orthonormal frame bundle of M and denoted F (M ): SO(n) - F (M ) Ļ
?
M. There are several important diļ¬erences between the orthonormal frame bundle of the homogeneous space En and that of a general Riemannian manifold M . First, while each ļ¬ber of the frame bundle is acted on freely and transitively by SO(n), there is no larger group that acts transitively on the entire frame bundle F (M ). But the most signiļ¬cant change is that, given an orthonormal frame ļ¬eld (e1 (x), . . . , en (x)) on an open set in M , there is no natural way of thinking of the frame vector ļ¬elds (ei (x)) as functions from M to a ļ¬xed vector spaceānot even by regarding M as a submanifold of some larger Euclidean space, as we did for Sn and Hn in Chapter 11. Rather, they are sections of the tangent bundle T M , which means that for
364
12. Moving frames on Riemannian manifolds
each point x ā M , the vectors (e1 (x), . . . , en (x)) take values in the vector space Tx M . In the case of En , we were able to diļ¬erentiate vector ļ¬elds by using the fact that each tangent space Tx En is canonically isomorphic to En and regarding the vector ļ¬elds (e1 (x), . . . , en (x)) as functions into this ļ¬xed vector space (cf. Remark 3.13). And even for Sn and Hn , we were able to make use of these canonical isomorphisms for the ambient spaces En+1 and M1,n in order to deļ¬ne the covariant derivatives of vector ļ¬elds. But for a more general Riemannian manifold M , there is no such canonical isomorphism between each tangent space Tx M and a ļ¬xed vector space En , or even a canonical embedding of Tx M into a larger ļ¬xed vector space; indeed, there are inļ¬nitely many ways of identifying each tangent space Tx M with a ļ¬xed vector space En , all of which are equally valid. The following exercise illustrates some of the complications that may arise as a result of this ambiguity. *Exercise 12.4. To any local orthonormal frame ļ¬eld (e1 (x), . . . , en (x)) on an open set U ā M , we can associate a local trivialization (cf. Ā§1.5) Ļ : T U ā U Ć En of the tangent bundle T M by deļ¬ning (12.1) Ļ x, ai ei (x) = x, t[a1 , . . . , an ] . (a) Show that the map Ļ deļ¬nes an isometry between each tangent space Tx M and the vector space En . (b) We may regard a local trivialization Ļ as a local choice of basis vector ļ¬elds (e1 (x), . . ., en (x)) for sections of T M , and the identiļ¬cation (12.1) makes it tempting to think that we might be able to regard these vector ļ¬elds as āconstantā for purposes of diļ¬erentiation. But what happens when Ėn (x)) be any other orthonormal we choose a diļ¬erent basis? Let (Ė e1 (x), . . . , e frame ļ¬eld on U , related to the original frame ļ¬eld by
ĀÆ Ė1 (x) . . . e Ėn (x) = e1 (x) . . . en (x) A(x), e (12.2) ĀÆ where A(x) is an SO(n)-valued function on U . Show that under the analogous local trivialization ĻĖ associated to the orthonormal frame ļ¬eld (Ė e1 (x), Ėn (x)), the vector ļ¬elds (e1 (x), . . . , en (x)) are identiļ¬ed with the col. . ., e ā1 ĀÆ umns of the matrix A(x) . In particular, vector ļ¬elds that appear āconstantā with respect to one trivialization do not necessarily remain āconstantā with respect to a diļ¬erent trivialization. This exercises raises a crucial question: If there is no canonical trivialization of T M , and hence no consistent notion of a āconstantā vector ļ¬eld on M , then how can we possibly diļ¬erentiate vector ļ¬elds on M ? (This is a special case of the more general question of how to diļ¬erentiate sections of
12.2. Orthonormal frames and connections
365
a vector bundle.) It turns out that, in order to make sense of the notion of diļ¬erentiation for vector ļ¬elds on M , we need to introduce an additional structure, called a connection, on the tangent bundle Tx M . Deļ¬nition 12.5. Let Ī(T M ) denote the space of smooth local sections of T M (i.e., smooth vector ļ¬elds on open sets in M ). An aļ¬ne connection (or, more succinctly, a connection) ā on T M is a map ā : Ī(T M ) Ć Ī(T M ) ā Ī(T M ), with ā(w, v) denoted by āw v, such that for any vector ļ¬elds v, v1 , v2 , w, w1 , w2 ā Ī(T M ), any smooth, real-valued functions f, g on M , and any real numbers a, b ā R, we have (1) āf w1 +gw2 v = f āw1 v + gāw2 v; (2) āw (av1 + bv2 ) = aāw v1 + bāw v2 ; (3) āw (f v) = w(f )v + f āw v. Remark 12.6. The Levi-Civita connection of Deļ¬nition 11.16 is, of course, a connection according to this deļ¬nition. It also has certain additional properties, which we will discuss shortly. Conditions (1) and (2) are linearity properties: They say that the map ā is linear over smooth functions in its ļ¬rst input and linear over real numbers in its second input. Condition (3) is an analog of the Leibniz rule that describes how ā behaves when its second input is multiplied by a smooth function. The vector ļ¬eld āw v should be regarded as deļ¬ning a sort of ādirectional derivativeā of the vector ļ¬eld v in the direction of w, and conditions (1)ā(3) are precisely the conditions that such a directional derivative must satisfy. Remark 12.7. This deļ¬nition applies more generally to any smooth vector bundle B over a manifold M : If Ļ : B ā M is a vector bundle with ļ¬bers isomorphic to a ļ¬xed k-dimensional vector space V (cf. Ā§1.5), then an aļ¬ne connection ā on B is a map ā : Ī(T M ) Ć Ī(B) ā Ī(B) satisfying the properties of Deļ¬nition 12.5. This more general setting illustrates the fact that the two copies of Ī(T M ) used for the inputs of ā in Deļ¬nition 12.5 play signiļ¬cantly diļ¬erent roles: The ļ¬rst input is the direction along which diļ¬erentiation should occur (which must be a tangent vector to M ), and the second input is the object to be diļ¬erentiated. The output is the resulting diļ¬erentiated object, and it should live in the same space as the second input.
366
12. Moving frames on Riemannian manifolds
*Exercise 12.8. Let M = En , and let v, w be smooth vector ļ¬elds on En . By taking advantage of the usual canonical identiļ¬cation of each tangent space Tx En with the vector space En , we may regard these vector ļ¬elds as smooth functions v, w : En ā En . Having done so, deļ¬ne āw v = dv(w). (a) Show that ā is a connection on T En . (Hint: It might be helpful to write everything out in terms of the canonical local coordinates on T En ; i.e.,
v(x) = t v 1 (x), . . . , v n (x) , etc. This should make it more obvious that āw v is a vector ļ¬eld on En ā in fact, the components of āw v are obtained by computing the directional derivatives of the components of v in the direction w.) (b) The canonical isomorphism Tx En ā¼ = En allows us to write the tangent n bundle T E as the product manifold (12.3) T En ā¼ = En Ć En , where the ļ¬rst factor represents the base manifold En and the second factor represents the ļ¬bers Tx En . (In other words, we have a canonical global trivialization of the tangent bundle T En .) This, in turn, allows us to write the tangent bundle T (T En ) as the product manifold (12.4) T (T En ) ā¼ = T En Ć T En in the obvious way. The vector ļ¬eld v is a section of T En , and via the identiļ¬cation (12.3), we can write it as a function Ļ : En ā En Ć En deļ¬ned by Ļ(x) = (x, v(x)) . Show that āw v is given by the composition (12.5)
āw v = Ļ2 ā¦ dĻ(w),
where dĻ : T En ā T (T En ) is the diļ¬erential of the map Ļ (cf. Ā§1.3) and Ļ2 : T En Ć T En ā T En is the projection onto the second factor. This connection is called the ļ¬at connection on En . Remark 12.9. For a general Riemannian manifold M , there is generally no global trivialization analogous to equation (12.3) for T M . However, for each point (x, v) ā T M , the 2n-dimensional tangent space T(x,v) (T M ) can be decomposed in a manner analogous to equation (12.4) in many diļ¬erent ways. The n-dimensional subspace V(x,v) = T(x,v) (Tx M ) ā T(x,v) (T M ) (corresponding to the second factor in (12.4)) is canonically deļ¬ned: It is the tangent space to the ļ¬ber Tx M ā T M and is called the vertical tangent
12.2. Orthonormal frames and connections
367
space; moreover, V(x,v) is canonically isomorphic to Tx M . But there is no single canonical choice for the complementary n-dimensional subspace corresponding to the ļ¬rst factor in (12.4) (called the horizontal tangent space); in fact, any n-dimensional subspace H(x,v) ā T(x,v) (T M ) for which H(x,v) ā© V(x,v) = {0} is, a priori, an equally valid choice for the horizontal tangent space at (x, v). But the choice of a connection on T M can resolve this issue: A connection ā on T M determines a projection map Ļ2 : T(x,v) (T M ) ā V(x,v) as in equation (12.5), and this in turn determines the n-dimensional horizontal subspace H(x,v) = ker(Ļ2 ) ā T(x,v) (T M ) and the corresponding decomposition T(x,v) (T M ) = H(x,v) ā V(x,v) . Conversely, a choice of such a decomposition for each (x, v) ā T M (subject to certain consistency conditions implied by Deļ¬nition 12.5) determines a connection on T M via a projection formula analogous to (12.5). In the case of M = En , the canonical decomposition (12.4) for T (T En ) is precisely equivalent to the ļ¬at connection ā on T En . As a consequence of properties (1)ā(3) of Deļ¬nition 12.5, a connection ā is completely determined by its action on any given orthonormal frame ļ¬eld on M . Given an orthonormal frame ļ¬eld (e1 (x), . . . , en (x)) on M , the connection ā determines scalar-valued 1-forms (ĀÆ Ļji ), with 1 ā¤ i, j ā¤ n, on M , called the connection forms associated to this frame ļ¬eld, deļ¬ned by the condition that for any w ā Tx M , (12.6)
(āw ei ) (x) = ej (x) Ļ ĀÆ ij (w).
The following exercise shows how a connection on T M may be thought of as an analog to the exterior derivative for vector ļ¬elds on M (cf. Exercise 11.14). *Exercise 12.10. Given a connection ā on T M and a vector ļ¬eld v on M , consider the map āv : T M ā T M deļ¬ned by (12.7)
āv(w) = āw v.
(a) Use the linearity properties of Deļ¬nition 12.5 to show that āv is a T M -valued 1-form on M (cf. Deļ¬nition 2.18).
368
12. Moving frames on Riemannian manifolds
(b) Show that if (e1 (x), . . . , en (x)) is an orthonormal frame ļ¬eld on an open set U ā M , then āei is the T M -valued 1-form (12.8)
āei = ej Ļ ĀÆ ij ,
where (ĀÆ Ļij ) are the scalar-valued 1-forms on M deļ¬ned by equation (12.6). Thus, we can think of ā as a generalization of the exterior derivative d that appears in the structure equations (3.1). *Exercise 12.11. Let ā be a connection on T M , and let (e1 (x), . . . , en (x)) be an orthonormal frame ļ¬eld on an open set U ā M , with associated connection forms (ĀÆ Ļji ) on U . (Think of the frame ļ¬eld as determining a local trivialization of T M , as in Exercise 12.4.) : i (a) Show that for any vector ļ¬eld v(x) = v (x)ei (x) on U , (12.9)
(āv) (x) = ei (x)(dv i + v j Ļ ĀÆ ji ).
(Note that, while the exterior derivative of a vector ļ¬eld on M is not welldeļ¬ned, it still makes perfect sense to compute the exterior derivative of real-valued functions, such as v i , on M .) For this reason, a connection is sometimes expressed with respect to a given trivialization of T M as ā=d+Ļ ĀÆ, where Ļ ĀÆ is the matrix of 1-forms Ļ ĀÆ = [ĀÆ Ļji ]. The notation means that if : i the vector ļ¬eld v(x) = v (x)ei (x) is expressed as the column vector ĀÆ (x) = t[v 1 (x), . . . , v n (x)], then the vector ļ¬eld āv should be expressed as v ĀÆ , as indicated by equation (12.9). the column vector dĀÆ v+Ļ ĀÆv Ėn (x)) be any other orthonormal frame ļ¬eld on U , related (b) Let (Ė e1 (x), . . . , e to the original frame ļ¬eld as in equation (12.2). Use equation (12.8) to show ĖĀÆ = [Ļ ĖĀÆ ji ] of connection forms associated to the new frame that the matrix Ļ ļ¬eld is given by (12.10)
ĀÆ ĖĀÆ = AĀÆā1 dAĀÆ + AĀÆā1 Ļ Ļ ĀÆ A.
Remark 12.12. Connections on ļ¬ber bundles play an important role in theoretical physics, particularly in ļ¬eld theory. In the physics literature, a local section of the orthonormal frame bundle (which amounts to choosing a local trivialization for T M ) is called a gauge, and a frame transformation of the form (12.2) is called a gauge transformation. A connection is called a gauge ļ¬eld, and equation (12.10) describes how the gauge ļ¬eld transforms under a gauge transformation.
12.2. Orthonormal frames and connections
369
In addition to the connection forms (ĀÆ Ļji ), we can associate to an orthonormal frame ļ¬eld (e1 (x), . . . , en (x)) on M the dual forms (ĀÆ Ļ1, . . . , Ļ ĀÆ n ). These are deļ¬ned in essentially the same way as for orthonormal frame ļ¬elds on En : Deļ¬nition 12.13. Let M be a Riemannian manifold, and let (e1 (x), . . ., en (x)) be an orthonormal frame ļ¬eld on an open set U ā M . The dual forms (ĀÆ Ļ1, . . . , Ļ ĀÆ n ) associated to this frame ļ¬eld are the unique scalar-valued 1-forms on U deļ¬ned by the property that 1, i = j, i i Ļ ĀÆ (ej ) = Ī“j = 0, i = j (cf. (3.7)). *Exercise 12.14. (a) Show that the dual forms (ĀÆ Ļ1, . . . , Ļ ĀÆ n ) associated to an orthonormal frame ļ¬eld (e1 (x), . . . , en (x)) form a basis for the 1-forms on U . Ėn (x)) be any other orthonormal frame ļ¬eld on U , related (b) Let (Ė e1 (x), . . . , e to the original frame ļ¬eld as in equation (12.2). Show that the dual forms ĖĀÆ 1 , . . . , Ļ ĖĀÆ n ) associated to the new frame ļ¬eld are given by (Ļ ā” 1ā¤ ā” 1ā¤ ĖĀÆ Ļ Ļ ĀÆ ā¢ .. ā„ ā1 ā¢ .. ā„ ĀÆ (12.11) ā£ . ā¦ = A ā£ . ā¦. ĖĀÆ n Ļ Ļ ĀÆn
As you might have guessed from the notation, the dual forms and connection forms (ĀÆ Ļi, Ļ ĀÆ ji ) on M associated to an orthonormal frame ļ¬eld (e1 (x), . . ., en (x)) on M are the pullbacks to M of certain 1-forms (Ļ i , Ļji ) on the frame bundle F (M ) via the section f (x) = (x; e1 (x), . . . , en (x)) of F (M ). Fortunately, as the following exercise will show, equations (12.10) and (12.11) tell us precisely how these 1-forms should be deļ¬ned. (It might be helpful to review Ā§3.3.2 at this point, particularly the derivations of equations (3.3) and (3.4).) *Exercise 12.15. Let (e1 (x), . . ., en (x)) be a local orthonormal frame ļ¬eld on an open set U ā M , with associated dual and connection forms (ĀÆ Ļi, Ļ ĀÆ ji ). Just as this orthonormal frame ļ¬eld determines a local trivialization of T M (cf. Exercise 12.4), it can also be used to deļ¬ne a local trivialization of the principal bundle F (M ): For any x ā U and any orthonormal frame f = (x; e1 , . . . , en ) based at x, we can write
e1 Ā· Ā· Ā· e n = e 1 Ā· Ā· Ā· e n A (12.12)
370
12. Moving frames on Riemannian manifolds
for some unique matrix A = [aij ] ā SO(n). Deļ¬ne a map Ļ : F (U ) ā U Ć SO(n) by Ļ (x; e1 , . . . , en ) = (x, A) , where A is the matrix deļ¬ned by equation (12.12). (a) Show that the map Ļ deļ¬nes a diļ¬eomorphism between F (U ) = Ļ ā1 (U ) ā F (M ) and U Ć SO(n). Ėn (x)) be any other orthonormal frame ļ¬eld on U , related (b) Let (Ė e1 (x), . . . , e to the original frame ļ¬eld as in equation (12.2). Show that the analogous local trivialization ĻĖ associated to the orthonormal frame ļ¬eld (Ė e1 (x), . . ., Ėn (x)) is given by e ! ā1 ! ĀÆ (12.13) ĻĖ (x; e1 , . . . , en ) = x, AĖ = x, A(x) A . (c) Deļ¬ne 1-forms (Ļ i , Ļji ) on F (U ) by ā” 1ā¤ ā” 1ā¤ Ļ Ļ ĀÆ ā¢ .. ā„ ā1 ā¢ .. ā„ ā£ . ā¦ = A ā£ . ā¦, (12.14) Ļn Ļ ĀÆn Ļ = [Ļji ] = Aā1 dA + Aā1 Ļ ĀÆ A. Note the distinction between these 1-forms and those in equations (12.10) and (12.11): In those equations, AĀÆ is an SO(n)-valued function on the open set U ā M , whereas in equations (12.14), the entries of A represent local coordinates on SO(n), independent of the local coordinates on M . So, while the 1-forms (ĀÆ Ļi, Ļ ĀÆ ji ) are 1-forms on M and the (ĀÆ Ļji ) are linear combinations of the (ĀÆ Ļ i ), the 1-forms (Ļ i , Ļji ) are linearly independent 1-forms on the orthonormal frame bundle F (M ). Use equations (12.10), (12.11), and (12.13) to show that the 1-forms (12.14) are well-deļ¬ned, independent of the choice of orthonormal frame ļ¬eld (e1 (x), . . . , en (x)) used to deļ¬ne the 1-forms (ĀÆ Ļi, Ļ ĀÆ ji ) and the local trivialization Ļ of F (M ).
12.3. The Levi-Civita connection There are many ways of choosing a connection on T M ; indeed, given an orthonormal frame ļ¬eld on an open set U ā M , any n Ć n matrix Ļ ĀÆ of 1-forms on M can be used to deļ¬ne a connection on T U via the equation (12.8). But some connections have nicer properties than others; the following deļ¬nition describes two properties that are often considered desirable.
12.3. The Levi-Civita connection
371
Deļ¬nition 12.16. An aļ¬ne connection ā on T M is called (1) torsion-free or symmetric if for any vector ļ¬elds v, w on M , we have āv w ā āw v = [v, w], where [v, w] denotes the usual Lie bracket of vector ļ¬elds (cf. Ā§1.4); (2) compatible with the metric g on M if for any vector ļ¬elds v, w on M , we have dv, w = āv, w + v, āw. (This condition is often written as āg = 0.) A given connection ā on T M may have one, both, or neither of these properties. But it turns out that there is exactly one connection on T M that has both: Theorem 12.17 (Levi-Civita). Given a Riemannian manifold M , there exists a unique connection ā on T M that is both torsion-free and compatible with the metric. This connection is called the Levi-Civita connection on T M . Remark 12.18. The existence of a canonical āniceā connection is a very important feature of Riemannian geometry, and unless otherwise stated, the Levi-Civita connection is almost always the connection of choice for the tangent bundle of a Riemannian manifold. But in other types of geometry (e.g., equi-aļ¬ne, projective), there is often no single ābestā connection, and then the choice of connection must be stated explicitly. In some contexts, it is even desirable to consider a whole family of connections! In the following exercises, we will show how to prove Theorem 12.17. The strategy of the proof is to show that if a torsion-free, metric-compatible connection exists, then it must be unique. In the process, an explicit formula for this unique connection is derived, which serves to prove the existence result as well. To begin the proof, let (e1 (x), . . . , en (x)) be an orthonormal frame ļ¬eld on an open set U ā M , with associated dual forms (ĀÆ Ļ1, . . . , Ļ ĀÆ n ). (At this point, we will drop the underscore notation on the vector ļ¬elds ei (x) since we no longer need to think of them as ābasisā vector ļ¬elds.) Let ā be an aļ¬ne connection on M , and let (ĀÆ Ļji ) be the corresponding connection forms associated to the given frame ļ¬eld. Since the dual forms are a basis for the
372
12. Moving frames on Riemannian manifolds
1-forms on M , we can write the connection forms as Ļ ĀÆ ji = aijk Ļ ĀÆk
(12.15) for some functions (aijk ) on U .
*Exercise 12.19. Suppose that ā is compatible with the metric. (a) Show that the connection forms must satisfy the condition Ļ ĀÆ ij = āĀÆ Ļji ; i.e., the connection forms are skew-symmetric in their indices. (Hint: Differentiate the equations ei , ej = Ī“ij .) (b) Conclude that if ā is compatible with the metric, then the functions (aijk ) must satisfy aijk = āajik .
(12.16)
*Exercise 12.20. Suppose that ā is torsion-free, and let ckij (x) = āckji (x) be the functions on U deļ¬ned by the Lie bracket relations [ei , ej ] = ckij ek . (a) Show that for each i, j, k, Ļ ĀÆ jk (ei ) ā Ļ ĀÆ ik (ej ) = ckij . (Hint: Apply the torsion-free condition with v = ei , w = ej .) (b) Conclude that if ā is torsion-free, then the functions (aijk ) must satisfy (12.17)
akji ā akij = ckij .
*Exercise 12.21. Now, suppose that ā is both compatible with the metric and torsion-free. (a) Use equations (12.16) and (12.17) to show that ! (12.18) akij = 12 cjki ā cijk ā ckij . (Hint: This is an exercise in index juggling. Start with akij and apply equations (12.16) and (12.17) alternately until you come back to the index arrangement that you started with. And keep in mind that ckij = āckji !) (b) Conclude that Theorem 12.17 is true and that the Levi-Civita connection is deļ¬ned by equation (12.15), with (akij ) as in equation (12.18). From now on, we will assume that ā is the Levi-Civita connection on T M .
12.4. The structure equations
373
12.4. The structure equations Let (Ļ i , Ļji ) be the dual forms and connection forms on F (M ) associated to the Levi-Civita connection on T M . In order to compute geometric invariants associated to M , we ļ¬rst need to compute the structure equations of these 1-forms. The process is analogous to that of Ā§3.3.2, but it is complicated by the fact that we must use the connection ā rather than the exterior derivative d to diļ¬erentiate vector-valued quantities. As in the case of En , we can deļ¬ne projection maps x : F (M ) ā M and ei : F (M ) ā T M in the obvious way: If f = (x; e1 , . . . , en ) ā F (M ), then x(f ) = x ā M, ei (f ) = (x, ei ) ā Tx M. The diļ¬erentials of (x, ei ) are maps dx : Tf F (M ) ā Tx M, dei : Tf F (M ) ā T(x,ei ) (T M ). The map dx is exactly analogous to the Euclidean case: The vectors (e1 , . . ., en ) form a basis for Tx M at each point, and the dual forms are deļ¬ned precisely so that dx = ei Ļ i .
(12.19)
The maps (dei ) are a bit more complicated. For each f ā F (M ), the image of (dei )f takes values in the 2n-dimensional tangent space T(x,ei ) (T M ). The Levi-Civita connection ā determines a linear projection operator Ļ2 : T(x,ei ) (T M ) ā T(x,ei ) (Tx M ) ā¼ = Tx M, deļ¬ned by the condition that (12.20)
(Ļ2 ā¦ d)(ei ) = āei = ej Ļij
(cf. Remark 12.9). This equation is the analog of the equation for dei in equations (3.1). In order to compute the structure equations for the forms (Ļ i , Ļji ), we will ļ¬rst need to diļ¬erentiate equation (12.19). This requires some care: Since both sides of the equation are 1-forms that take values in T M , we must use the connection ā to diļ¬erentiate them. And in general, it is not necessarily true that ā ā¦ d = 0, so we cannot directly apply any obvious analog of the identity d ā¦ d = 0. Fortunately, we can get around this problem by considering two diļ¬erent expressions for the T M -valued 1-form dx, as the following two exercises show.
374
12. Moving frames on Riemannian manifolds
*Exercise 12.22. Recall that with respect to any local coordinate system x = (x1 , . . . , xn ) on an open set U ā M , we can write ā (12.21) dx = dxi , āxi where āxā 1 , . . . , āxān are the coordinate vector ļ¬elds on U . (a) Show that the coordinate vector ļ¬elds āxā 1 , . . . , āxān have pairwise Lie
ā ā brackets equal to zero; i.e., āxi , āxj = 0 (cf. Exercise 1.34). (b) Apply ā to equation (12.21) and use an argument similar to that of Exercise 12.11, part (a), to show that " # ā ā ā (dx) = ā ā§ dxi + i d(dxi ). i āx āx Conclude that " # ā (12.22) ā (dx) = ā ā§ dxi . āxi (c) Use the deļ¬nition (12.7) for the T M -valued 1-form ā " # " # ā ā (12.23) ā =ā ā dxj . āxi āxi āxj
ā āxi
to show that
(d) Use equations (12.22) and (12.23) to show that " # " ## " ā ā ā (dx) = ā ā āā ā dxi ā§ dxj . āxj āxi āxi āxj i b > c > 0, and let M = R3 , with metric 2 1 g= dx + dy 2 + dz 2 . 2 2 2 2 (ax + by + cz + 1) (a) Show that the frame ļ¬eld ā , āx ā e2 (x, y, z) = (ax2 + by 2 + cz 2 + 1) , (12.49) āy ā e3 (x, y, z) = (ax2 + by 2 + cz 2 + 1) āz on M is an orthonormal frame ļ¬eld for g and that its dual forms are dx Ļ ĀÆ1 = , (ax2 + by 2 + cz 2 + 1) dy Ļ ĀÆ2 = , (12.50) 2 (ax + by 2 + cz 2 + 1) dz Ļ ĀÆ3 = . 2 2 (ax + by + cz 2 + 1) e1 (x, y, z) = (ax2 + by 2 + cz 2 + 1)
12.6. Moving frames for surfaces in Riemannian manifolds
387
(b) Show that the Levi-Civita connection forms associated to the dual forms (12.50) are 2 Ļ ĀÆ 32 = (by dz ā cz dy), (ax2 + by 2 + cz 2 + 1) 2 Ļ ĀÆ 13 = (cz dx ā ax dz), (12.51) (ax2 + by 2 + cz 2 + 1) 2 Ļ ĀÆ 21 = (ax dy ā by dx). 2 2 (ax + by + cz 2 + 1) ĀÆ has the form (c) Show that the curvature matrix R ā” ĀÆ2 ā¤ R323 0 0 ā¢ ā„ ĀÆ3 ĀÆ=ā¢ 0 R ā„, 0 R 131 ā£ ā¦ 1 ĀÆ 0 0 R212 where (12.52)
ĀÆ 2 = 2 (b + c ā 2a)ax2 + (b ā c)(cz 2 ā by 2 ) + (b + c) , R 323 ĀÆ 3 = 2 (c + a ā 2b)by 2 + (c ā a)(ax2 ā cz 2 ) + (c + a) , R 131 ĀÆ 1 = 2 (a + b ā 2c)cz 2 + (a ā b)(by 2 ā ax2 ) + (a + b) . R 212
ĀÆ2 , R ĀÆ3 , R ĀÆ 1 ) of R ĀÆ are distinct for all x = Verify that the eigenvalues (R 323 131 212 ĀÆ in each (x, y, z) ā M , and conclude that the corresponding eigenvectors of R tangent space Tx M are precisely the coordinate directions (e1 , e2 , e3 ). (d) Conclude from the result of Exercise 12.41(c) that the only potential tangent planes P ā Tx M to totally geodesic surfaces in M are the coordinate planes P1 = span(e2 , e3 ),
P2 = span(e3 , e1 ),
P3 = span(e1 , e2 ).
It follows that if there exists a totally geodesic surface Ī£ ā M , it must be a level surface for one of the coordinate functions (x, y, z) on M . (e) Let U ā R2 , and consider a candidate surface Ī£ ā M with parametrization x : U ā M given by x(u, v) = (u, v, z0 ) for some z0 ā R. The frame ļ¬eld (12.49) is an adapted frame ļ¬eld along Ī£. Compute the pullbacks of the connection forms (12.51) to U ; in particular, show that 2cz0 du 2cz0 dv Ļ ĀÆ 13 = , Ļ ĀÆ 23 = . 2 2 2 2 (au + bv + cz0 + 1) (au + bv 2 + cz02 + 1) Conclude from the result of Exercise 12.41(a) that Ī£ is a totally geodesic surface if and only if z0 = 0.
388
12. Moving frames on Riemannian manifolds
A similar argument shows that the only totally geodesic surfaces in M are contained in the coordinate āplanesā Ī£1 = {(0, y, z) | y, z ā R} ā M, Ī£2 = {(x, 0, z) | x, z ā R} ā M, Ī£3 = {(x, y, 0) | x, y ā R} ā M. Any open submanifold M0 ā M that does not intersect these surfacesāfor instance, the ļ¬rst octant M0 = {(x, y, z) ā M | x, y, z > 0}ācontains no totally geodesic surfaces whatsoever.
12.7. Maple computations Exercise 12.30: Itās not so simple to tell Maple how to do computations on a manifold of arbitrary dimension, so weāll go through this exercise for the case n = m = 3, which represents a 3-dimensional Riemannian manifold M embedded in E6 . (This is the embedding dimension for which the isometric embedding problem is a determined system of PDEs, rather than an overdetermined or underdetermined system, when n = 3.) After loading the Cartan and LinearAlgebra packages into Maple, we ļ¬rst need to declare the Maurer-Cartan forms on F (E6 ). This is rather a lot of forms, and itās helpful to use the seq command to generate the necessary lists of forms: > Form(seq(omega[i], i=1..6)); Form(seq(seq(omega[i,j], j=1..6), i=1..6)); Tell Maple about the symmetries in the connection forms; for convenience, do this in such a way that the resulting basis includes forms indexed as (ĀÆ Ļia ) with 1 ā¤ i ā¤ 3, 4 ā¤ a ā¤ 6 rather than the other way around: > for i from 1 to 6 do omega[i,i]:= 0; end do; omega[2,1]:= -omega[1,2]; omega[3,2]:= -omega[2,3]; omega[1,3]:= -omega[3,1]; omega[5,4]:= -omega[4,5]; omega[6,5]:= -omega[5,6]; omega[4,6]:= -omega[6,4]; for a from 4 to 6 do for j from 1 to 3 do
12.7. Maple computations
389
omega[j,a]:= -omega[a,j]; end do; end do; Tell Maple how to diļ¬erentiate these forms according to the Cartan structure equations (3.8), and be sure to deļ¬ne structure equations only for those connection forms that have not been deļ¬ned in terms of other forms: > for i from 1 to 6 do d(omega[i]):= sum(ā-omega[i,j] &Ė omega[j]ā, ājā=1..6); end do; d(omega[1,2]):= sum(ā-omega[1,k] &Ė omega[k,2]ā, ākā=1..6); d(omega[2,3]):= sum(ā-omega[2,k] &Ė omega[k,3]ā, ākā=1..6); d(omega[3,1]):= sum(ā-omega[3,k] &Ė omega[k,1]ā, ākā=1..6); d(omega[4,5]):= sum(ā-omega[4,k] &Ė omega[k,5]ā, ākā=1..6); d(omega[5,6]):= sum(ā-omega[5,k] &Ė omega[k,6]ā, ākā=1..6); d(omega[6,4]):= sum(ā-omega[6,k] &Ė omega[k,4]ā, ākā=1..6); for a from 4 to 6 do for j from 1 to 3 do d(omega[a,j]):= sum(ā-omega[a,k] &Ė omega[k,j]ā, ākā=1..6); end do; end do; Now, consider an adapted orthonormal frame ļ¬eld along M for which the frame vectors (e1 (u), e2 (u), e3 (u)) are tangent to M at each point. According to equation (12.37), for such a frame ļ¬eld, we must have Ļ ĀÆ4 = Ļ ĀÆ5 = Ļ ĀÆ 6 = 0. Set up a substitution for the Maurer-Cartan forms associated to an adapted frame ļ¬eld: > adaptedsub:= [omega[4]=0, omega[5]=0, omega[6]=0]; Now diļ¬erentiate these equations: > Simf(subs(adaptedsub, Simf(d(omega[4])))); Simf(subs(adaptedsub, Simf(d(omega[5])))); Simf(subs(adaptedsub, Simf(d(omega[6])))); Applying Cartanās lemma tells us that equations (12.38) hold, and we can add these equations to our substitution as follows: > for a from 4 to 6 do h[a,2,1]:= h[a,1,2]; h[a,3,2]:= h[a,2,3];
390
12. Moving frames on Riemannian manifolds
h[a,1,3]:= h[a,3,1]; end do; > unassign(āiā, ājā, āaā); > adaptedsub:= [op(adaptedsub), seq(seq(omega[a,i] = sum(h[a,i,j]*omega[j], j=1..3), i=1..3), a=4..6)]; (The unassign command is needed because one of the previous for loops assigned values to these indices that need to be removed before using them in a seq command.) Now look at the structure equations for dĀÆ Ļ 1 , dĀÆ Ļ 2 , dĀÆ Ļ3: > Simf(subs(adaptedsub, Simf(d(omega[1])))); (Ļ2 ) &Ė (Ļ1,2 ) ā (Ļ3 ) &Ė (Ļ3,1 ) > Simf(subs(adaptedsub, Simf(d(omega[2])))); ā(Ļ1 ) &Ė (Ļ1,2 ) + (Ļ3 ) &Ė (Ļ2,3 ) > Simf(subs(adaptedsub, Simf(d(omega[3])))); (Ļ1 ) &Ė (Ļ3,1 ) ā (Ļ2 ) &Ė (Ļ2,3 ) It follows from these equations that the 1-forms (ĀÆ Ļji ) with 1 ā¤ i, j ā¤ 3 are the Levi-Civita connection forms associated to the metric 1 2 2 2 3 2 I= Ļ ĀÆ + Ļ ĀÆ + Ļ ĀÆ i ) on M . Therefore, they satisfy the structure equations (12.35), where (Rjk are the components of the Riemann curvature tensor of M . Now compare equations (12.35) to the structure equations for the (dĀÆ Ļji ) as Maurer-Cartan forms on E6 :
> zero1:= Simf(subs(adaptedsub, Simf(d(omega[1,2]) + omega[1,3] &Ė omega[3,2] - R[1,2,1,2]*omega[1] &Ė omega[2] - R[1,2,2,3]*omega[2] &Ė omega[3] - R[1,2,3,1]*omega[3] &Ė omega[1]))); > zero2:= Simf(subs(adaptedsub, Simf(d(omega[2,3]) + omega[2,1] &Ė omega[1,3] - R[2,3,1,2]*omega[1] &Ė omega[2] - R[2,3,2,3]*omega[2] &Ė omega[3] - R[2,3,3,1]*omega[3] &Ė omega[1]))); > zero3:= Simf(subs(adaptedsub, Simf(d(omega[3,1]) + omega[3,2] &Ė omega[2,1] - R[3,1,1,2]*omega[1] &Ė omega[2] - R[3,1,2,3]*omega[2] &Ė omega[3] - R[3,1,3,1]*omega[3] &Ė omega[1])));
12.7. Maple computations
391
The scalar coeļ¬cients in these quantities are equivalent to the Gauss equations (12.39). Next, look at the structure equations for the connection forms (ĀÆ Ļia ) with 1 ā¤ i ā¤ 3, 4 ā¤ a ā¤ 6: > zero41:= Simf(subs(adaptedsub, Simf(d(omega[4,1])) - d(Simf(subs(adaptedsub, omega[4,1]))))); The resulting expression has the form Ļ1 ā§ Ļ ĀÆ 1 + Ļ2 ā§ Ļ ĀÆ 2 + Ļ3 ā§ Ļ ĀÆ 3, where Ļ1 , Ļ2 , Ļ3 are the following 1-forms: > pick(zero41, omega[1]); ā2 h4,1,2 Ļ1,2 + 2 h4,3,1 Ļ3,1 ā h5,1,1 Ļ4,5 + h6,1,1 Ļ6,4 ā d(h4,1,1 ) > pick(zero41, omega[2]); ā (h4,2,2 ā h4,1,1 ) Ļ1,2 ā h4,3,1 Ļ2,3 + h4,2,3 Ļ3,1 ā h5,1,2 Ļ4,5 + h6,1,2 Ļ6,4 ā d(h4,1,2 ) > pick(zero41, omega[3]); ā h4,2,3 Ļ1,2 + h4,1,2 Ļ2,3 ā (āh4,3,3 + h4,1,1 ) Ļ3,1 ā h5,3,1 Ļ4,5 + h6,3,1 Ļ6,4 ā d(h4,3,1 ) Applying Cartanās lemma to all 9 structure equations for the (dĀÆ Ļia ) shows that there exist functions (haijk ), symmetric in their lower indices, such that these equations take the form dhaij = (terms involving the functions (hbk ) and the connection forms (ĀÆ Ļji , Ļ ĀÆ ba )) + haijk Ļ ĀÆk. These are the Codazzi equations; there are 18 of them, one for each of the components (haij ) of the second fundamental form. Remark 12.43. From the PDE perspective, the symmetry of the (haijk ) is equivalent to a system of PDEs of the form āhaij āhaik ā = (lower order terms). āxk āxj From this point of view, there are 27 Codazzi equations, 24 of which are independent.
392
12. Moving frames on Riemannian manifolds
Finally, look at the structure equations for the connection forms (ĀÆ Ļba ) with 4 ā¤ a, b ā¤ 6: > Simf(subs(adaptedsub, Simf(d(omega[4,5])))); > Simf(subs(adaptedsub, Simf(d(omega[5,6])))); > Simf(subs(adaptedsub, Simf(d(omega[6,4])))); These all have the form a dĀÆ Ļba = āĀÆ Ļca ā§ Ļ ĀÆ bc + Sbij Ļ ĀÆi ā§ Ļ ĀÆj, a ) are quadratic expressions in the functions (ha ), where the functions (Sbij ij similar to those found in the Gauss equations. These are the Ricci equations; a ) represent the components of the curvature tensor for the coeļ¬cients (Sbij the connection on the normal bundle of M induced from the Euclidean connection on E6 .
Exercise 12.42: Since this exercise requires diļ¬erent dual and connection forms with diļ¬erent structure equations from the previous exercise, itās probably best to restart Maple and reload the Cartan and LinearAlgebra packages. Declare the dual and connection forms on F (M ), and tell Maple about their symmetries: > Form(omega[1], omega[2], omega[3]); Form(omega[1,2], omega[3,1], omega[3,2]); > omega[1,1]:= 0; omega[2,2]:= 0; omega[3,3]:= 0; omega[2,1]:= -omega[1,2]; omega[1,3]:= -omega[3,1]; omega[2,3]:= -omega[3,2]; Declare constants a, b, c: > Form(a=-1, b=-1, c=-1); Set up a substitution for the dual forms of the given orthonormal frame ļ¬eld on M , as well as the reverse substitution: > dualformssub:= [ omega[1] = d(x)/(a*xĖ2 + b*yĖ2 + c*zĖ2 + 1), omega[2] = d(y)/(a*xĖ2 + b*yĖ2 + c*zĖ2 + 1), omega[3] = d(z)/(a*xĖ2 + b*yĖ2 + c*zĖ2 + 1)]; > dualformsbacksub:= makebacksub(dualformssub);
12.7. Maple computations
393
In order to compute the connection forms, diļ¬erentiate the dual forms and compare the result to the structure equations. (Itās helpful to go ahead and write the (ĀÆ Ļji ) as linear combinations of the (ĀÆ Ļ i ) with undetermined coeļ¬cients and then solve for these coeļ¬cients.) > connectionformssub:= [ omega[1,2] = h[1,2,1]*omega[1] + h[1,2,2]*omega[2] + h[1,2,3]*omega[3], omega[3,1] = h[3,1,1]*omega[1] + h[3,1,2]*omega[2] + h[3,1,3]*omega[3], omega[3,2] = h[3,2,1]*omega[1] + h[3,2,2]*omega[2] + h[3,2,3]*omega[3]]; According to the structure equations, the following expressions should be zero: > zero1:= Simf(d(Simf(subs(dualformssub, omega[1]))) + Simf(subs(dualformssub, Simf(subs(connectionformssub, omega[1,2] &Ė omega[2] + omega[1,3] &Ė omega[3]))))); zero2:= Simf(d(Simf(subs(dualformssub, omega[2]))) + Simf(subs(dualformssub, Simf(subs(connectionformssub, omega[2,1] &Ė omega[1] + omega[2,3] &Ė omega[3]))))); zero3:= Simf(d(Simf(subs(dualformssub, omega[3]))) + Simf(subs(dualformssub, Simf(subs(connectionformssub, omega[3,1] &Ė omega[1] + omega[3,2] &Ė omega[2])))));
(2cz ā h3,1,1 ) (d(x)) &Ė(d(z)) (2by + h1,2,1 ) (d(y)) &Ė(d(x)) ā (ax2 + by 2 + cz 2 + 1)2 (ax2 + by 2 + cz 2 + 1)2 (h1,2,3 + h3,1,2 ) (d(y)) &Ė(d(z)) ā (ax2 + by 2 + cz 2 + 1)2
zero1 :=
(h1,2,3 ā h3,2,1 ) (d(x)) &Ė(d(z)) (2ax ā h1,2,2 ) (d(y)) &Ė(d(x)) + (ax2 + by 2 + cz 2 + 1)2 (ax2 + by 2 + cz 2 + 1)2 (2cz ā h3,2,2 ) (d(y)) &Ė(d(z)) + (ax2 + by 2 + cz 2 + 1)2
zero2 :=
(2ax + h3,1,3 ) (d(x)) &Ė(d(z)) (ax2 + by 2 + cz 2 + 1)2 (h3,1,2 ā h3,2,1 ) (d(y)) &Ė(d(x)) (2by + h3,2,3 ) (d(y)) &Ė(d(z)) + ā (ax2 + by 2 + cz 2 + 1)2 (ax2 + by 2 + cz 2 + 1)2
zero3 := ā
394
12. Moving frames on Riemannian manifolds
We can set the scalar coeļ¬cients of these forms equal to zero and solve for the (hijk ) to determine the connection forms: > solve({op(ScalarForm(zero1)), op(ScalarForm(zero2)), op(ScalarForm(zero3))}, {h[1,2,1], h[1,2,2], h[1,2,3], h[3,1,1], h[3,1,2], h[3,1,3], h[3,2,1], h[3,2,2], h[3,2,3]}); {h1,2,1 = ā2by, h1,2,2 = 2ax, h1,2,3 = 0, h3,1,1 = 2cz, h3,1,2 = 0, h3,1,3 = ā2ax, h3,2,1 = 0, h3,2,2 = 2cz, h3,2,3 = ā2by} > assign(%); So here are the connection forms, expressed in terms of the coordinate 1forms: > connectionformssub:= Simf(subs(dualformssub, Simf(connectionformssub))); connectionf ormssub := Ļ1,2 = ā
2by d(x) 2ay d(y) + 2 , 2 2 + by + cz + 1 ax + by 2 + cz 2 + 1
2cz d(x) 2ax d(z) ā 2 , 2 2 + by + cz + 1 ax + by 2 + cz 2 + 1 2cz d(y) 2by d(z) = 2 ā ax + by 2 + cz 2 + 1 ax2 + by 2 + cz 2 + 1
Ļ3,1 = Ļ3,2
ax2
ax2
Now compute the curvature 2-forms, expressed in terms of the dual forms: > Omega[2,3]:= Simf(subs(dualformsbacksub, Simf(d(Simf(subs(connectionformssub, omega[2,3]))) + Simf(subs(connectionformssub, omega[2,1] &Ė omega[1,3]))))); Omega[3,1]:= Simf(subs(dualformsbacksub, Simf(d(Simf(subs(connectionformssub, omega[3,1]))) + Simf(subs(connectionformssub, omega[3,2] &Ė omega[2,1]))))); Omega[1,2]:= Simf(subs(dualformsbacksub, Simf(d(Simf(subs(connectionformssub, omega[1,2]))) + Simf(subs(connectionformssub,
12.7. Maple computations
395
omega[1,3] &Ė omega[3,2]))))); Ī©2,3 := 2cax2 + 2cby 2 ā 2c2 z 2 + 2c + 2bax2 ā 2b2 y 2 + 2bcz 2 + 2b ā 4a2 x2 (Ļ2 ) &Ė (Ļ3 ) Ī©3,1 := ā 2cax2 ā 2cby 2 + 2c2 z 2 ā 2c + 2a2 x2 ā 2aby 2 ā 2acz 2 ā 2a + 4b2 y 2 (Ļ1 ) &Ė (Ļ3 ) Ī©1,2 := 2bax2 ā 2b2 y 2 + 2bcz 2 + 2b ā 2a2 x2 + 2aby 2 + 2acz 2 + 2a ā 4c2 z 2 (Ļ1 ) &Ė (Ļ2 ) ĀÆ is diagonal, with the following diagonal entries: So the matrix R > R[2,3,2,3]:= pick(Omega[2,3], omega[2], omega[3]); R[3,1,3,1]:= pick(Omega[3,1], omega[3], omega[1]); R[1,2,1,2]:= pick(Omega[1,2], omega[1], omega[2]); ĀÆ are distinct: Check that the diagonal entries of R > factor(R[2,3,2,3] - R[3,1,3,1]); ā2 (āb + a) (ax2 + by 2 + cz 2 + 1) > factor(R[3,1,3,1] - R[1,2,1,2]); ā2 (āc + b) (ax2 + by 2 + cz 2 + 1) > factor(R[1,2,1,2] - R[2,3,2,3]); 2 (a ā c) (ax2 + by 2 + cz 2 + 1) It follows that any totally geodesic surface must be a level surface for one of the coordinate functions (x, y, z). Suppose that Ī£ is contained in the level set {(x, y, z) ā M | z = z0 } for some constant z0 and compute the pullbacks of the connection forms to Ī£: > Form(z0=-1); > Simf(subs([x=u, y=v, z=z0], connectionformssub)); 2bv d(u) 2au d(v) Ļ1,2 = ā 2 + , au + bv 2 + cz02 + 1 au2 + bv 2 + cz02 + 1 Ļ3,1
2cz0 d(u) 2cz0 d(v) = 2 , Ļ3,2 = 2 2 2 au + bv + cz0 + 1 au + bv 2 + cz02 + 1
Since a totally geodesic surface must have Ļ ĀÆ 13 = Ļ ĀÆ 23 = 0, Ī£ is not a totally geodesic surface unless z0 = 0.
Bibliography
[AEG06] Juan A. Aledo, JosĀ“ e M. Espinar, and JosĀ“ e A. GĀ“ alvez, Timelike surfaces in the LorentzMinkowski space with prescribed Gaussian curvature and Gauss map, J. Geom. Phys. 56 (2006), no. 8, 1357ā1369. [AI79] Robert L. Anderson and Nail H. Ibragimov, Lie-BĀØ acklund Transformations in Applications, SIAM Studies in Applied Mathematics, vol. 1, Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 1979. Ā“ [AR93] M. A. Akivis and B. A. Rosenfeld, Elie Cartan (1869ā1951), Translations of Mathematical Monographs, vol. 123, American Mathematical Society, Providence, RI, 1993, translated from the Russian manuscript by V. V. Goldberg. [BĀØ 83] A. V. BĀØ acklund, Om ytor med konstant negativ krĀØ okning, Lunds Universitets Ė Arsskrift 19 (1883), 1ā48. [BCG+ 91] R. L. Bryant, S. S. Chern, R. B. Gardner, H. L. Goldschmidt, and P. A. Griļ¬ths, Exterior Diļ¬erential Systems, Mathematical Sciences Research Institute Publications, vol. 18, Springer-Verlag, New York, 1991. [BG80] Richard L. Bishop and Samuel I. Goldberg, Tensor Analysis on Manifolds, Dover Publications Inc., New York, 1980, corrected reprint of the 1968 original. [Bia79] Luigi Bianchi, Ricerche sulle superļ¬cie elicoidali e sulle superļ¬cie a curvatura costante, Ann. Scuola Norm. Sup. Pisa Cl. Sci. 2 (1879), 285ā341. [Bla23] Wilhelm Blaschke, Vorlesungen u ĀØber Diļ¬erentialgeometrie, vol. II, Springer, Berlin, 1923. [Bla85]
, Gesammelte Werke. Band 4, Thales-Verlag, Essen, 1985, Aļ¬ne Diļ¬erentialgeometrie. Diļ¬erentialgeometrie der Kreis- und Kugelgruppen. [Aļ¬ne diļ¬erential geometry. Diļ¬erential geometry of circle and ball groups], with commentaries by Werner Burau and Udo Simon, edited by Burau, S. S. Chern, K. LeichtweiĆ, H. R. MĀØ uller, L. A. SantalĀ“ o, Simon and K. Strubecker.
[Cal00] James J. Callahan, The Geometry of Spacetime: An Introduction to Special and General Relativity, Undergraduate Texts in Mathematics, Springer-Verlag, New York, 2000.
397
398
Bibliography
Ā“ Ā“ [Car04] Elie Cartan, Sur la structure des groupes inļ¬nis de transformation, Ann. Sci. Ecole Norm. Sup. (3) 21 (1904), 153ā206. [Car20]
Ā“ , Sur la dĀ“ eformation projective des surfaces, Ann. Sci. Ecole Norm. Sup. (3) 37 (1920), 259ā356.
[Car27]
, Sur la possibilitĀ“ e de plonger un espace riemannien donnĀ“ e dans un espace euclidĀ“ een, Ann. Soc. Polon. Math. 6 (1927), 1ā7.
[Car30]
, La thĀ“ eorie des groupes ļ¬nis et continus et lāanalysis situs, MĀ“ emorial des sciences mathĀ“ ematiques, no. 42, Gauthier-Villars et Cie, 1930.
[Car35]
, La MĀ“ ethode du Rep` ere Mobile, la ThĀ“ eorie des Groupes Continus, et les Espaces GĀ“ enĀ“ eralisĀ“ es, ExposĀ“ es de GĀ“ eomĀ“ etrie, no. 5, Hermann, Paris, 1935.
[Car46]
, LeĀøcons sur la GĀ“ eomĀ“ etrie des Espaces de Riemann, 2nd ed., Gauthier-Villars, Paris, 1946.
[Car92]
, LeĀøcons sur la GĀ“ eomĀ“ etrie Projective Complexe. La ThĀ“ eorie des Groupes Finis et Continus et la GĀ“ eomĀ“ etrie Diļ¬Ā“ erentielle TraitĀ“ ees Par la MĀ“ ethode du Rep` ere Mobile. LeĀøcons sur la ThĀ“ eorie des Espaces a ` Connexion Projective, Les Grands Classiques Ā“ Gauthier-Villars. [Gauthier-Villars Great Classics], Editions Jacques Gabay, Sceaux, 1992, reprint of the 1931, 1937, and 1937 editions.
[CEM+ 14] Jeanne Clelland, Edward Estrada, Molly May, Jonah Miller, and Sean Peneyra, A Tale of Two Arc Lengths: Metric Notions for Curves in Surfaces in Equiaļ¬ne Space, Proc. Amer. Math. Soc. 142 (2014), 2543ā2558. [Che42] Shiing-shen Chern, On integral geometry in Klein spaces, Ann. of Math. (2) 43 (1942), 178ā189. [Cle99] Jeanne N. Clelland, Lie groups and the method of moving frames, 1999, lecture videos from Mathematical Sciences Research Institute Summer Graduate Workshop, http://www.msri.org/publications/video/index2.html. [Cle12]
, Totally quasi-umbilic timelike surfaces in R1,2 , Asian J. Math. 16 (2012), 189ā208.
[CT80] Shiing Shen Chern and Chuu Lian Terng, An analogue of BĀØ acklundās theorem in aļ¬ne geometry, Rocky Mountain J. Math. 10 (1980), no. 1, 105ā124. [CT86] Shiing-shen Chern and Keti Tenenblat, Pseudo-spherical surfaces and evolution equations, Stud. Appl. Math. 74 (1986), 55ā83. [Dac08] Bernard Dacorogna, Introduction to the Calculus of Variations, 2nd ed., Imperial College Press, London, 2008. [Dar72a] Gaston Darboux, LeĀøcons sur la ThĀ“ eorie GĀ“ enĀ“ erale des Surfaces et les Applications GĀ“ eomĀ“ etriques du Calcul Inļ¬nitĀ“ esimal. Deuxi` eme partie, Chelsea Publishing Co., Ā“ Bronx, N.Y., 1972, Les Congruences et les Equations LinĀ“eaires aux DĀ“ erivĀ“ees Partielles. Les Lignes TracĀ“ ees sur les Surfaces, rĀ“ eimpression de la deuxi` eme Ā“ edition de 1915. [Dar72b]
, LeĀøcons sur la ThĀ“ eorie GĀ“ enĀ“ erale des Surfaces et les Applications GĀ“ eomĀ“ etriques du Calcul Inļ¬nitĀ“ esimal. Premi` ere partie, Chelsea Publishing Co., Bronx, N.Y., 1972, GĀ“ enĀ“ eralitĀ“es. CoordonnĀ“ ees curvilignes. Surfaces minima, rĀ“ eimpression de la deuxi` eme Ā“ edition de 1914.
[Dar72c]
, LeĀøcons sur la ThĀ“ eorie GĀ“ enĀ“ erale des Surfaces et les Applications GĀ“ eomĀ“ etriques du Calcul Inļ¬nitĀ“ esimal. Quatri` eme partie, Chelsea Publishing Co., Bronx, N.Y.,
Bibliography
399
1972, DĀ“ eformation Inļ¬niment Petite et Representation SphĀ“ erique, rĀ“eimpression de la premi` ere Ā“ edition de 1896. [Dar72d]
, LeĀøcons sur la ThĀ“ eorie GĀ“ enĀ“ erale des Surfaces et les Applications GĀ“ eomĀ“ etriques du Calcul Inļ¬nitĀ“ esimal. Troisi` eme partie, Chelsea Publishing Co., Bronx, N.Y., 1972, Lignes GĀ“ eodĀ“ esiques et Courbure GĀ“eodĀ“ esique. Param`etres Diļ¬Ā“erentiels. DĀ“ eformation des Surfaces, rĀ“ eimpression de la premi` ere Ā“ edition de 1894.
[dC76] Manfredo P. do Carmo, Diļ¬erential Geometry of Curves and Surfaces, Prentice-Hall Inc., Englewood Cliļ¬s, N.J., 1976, translated from the Portuguese. [dC92]
, Riemannian Geometry, Mathematics: Theory & Applications, BirkhĀØ auser Boston Inc., Boston, MA, 1992, translated from the second Portuguese edition by Francis Flaherty.
[DFN92] B. A. Dubrovin, A. T. Fomenko, and S. P. Novikov, Modern GeometryāMethods and Applications. Part I: The Geometry of Surfaces, Transformation Groups, and Fields, 2nd ed., Graduate Texts in Mathematics, vol. 93, Springer-Verlag, New York, 1992, translated from the Russian by Robert G. Burns. [Dou31] Jesse Douglas, Solution of the problem of Plateau, Trans. Amer. Math. Soc. 33 (1931), no. 1, 263ā321. [Eis60] Luther Pfahler Eisenhart, A Treatise on the Diļ¬erential Geometry of Curves and Surfaces, Dover Publications Inc., New York, 1960. [Fav57] J. Favard, Cours de gĀ“ eomĀ“ etrie diļ¬Ā“ erentielle locale, Gauthier-Villars, Paris, 1957. [FO98] Mark Fels and Peter J. Olver, Moving coframes. I. A practical algorithm, Acta Appl. Math. 51 (1998), no. 2, 161ā213. [FO99]
, Moving coframes. II. Regularization and theoretical foundations, Acta Appl. Math. 55 (1999), no. 2, 127ā208.
[Fom90] A. T. Fomenko, The Plateau Problem. Part I, Studies in the Development of Modern Mathematics, vol. 1, Gordon and Breach Science Publishers, New York, 1990, Historical survey, translated from the Russian. [Fre47] J. F. Frenet, Sur les Courbes a ` Double Courbure, Ph.D. thesis, Toulouse, 1847. [GĀ“ al09] JosĀ“ e A. GĀ“ alvez, Surfaces of constant curvature in 3-dimensional space forms, Mat. Contemp. 37 (2009), 1ā42. [Gar89] Robert B. Gardner, The Method of Equivalence and Its Applications, CBMS-NSF Regional Conference Series in Applied Mathematics, vol. 58, Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA, 1989. [Gre78] Mark L. Green, The moving frame, diļ¬erential invariants and rigidity theorems for curves in homogeneous spaces, Duke Math. J. 45 (1978), no. 4, 735ā779. [Gri74] P. Griļ¬ths, On Cartanās method of Lie groups and moving frames as applied to uniqueness and existence questions in diļ¬erential geometry, Duke Math. J. 41 (1974), 775ā814. [Gug77] Heinrich W. Guggenheimer, Diļ¬erential Geometry, Dover Publications, Inc., New York, 1977, corrected reprint of the 1963 edition, Dover Books on Advanced Mathematics. [HCV52] D. Hilbert and S. Cohn-Vossen, Geometry and the Imagination, Chelsea Publishing Company, New York, N.Y., 1952, translated by P. NemĀ“ enyi.
400
Bibliography
[HH06] Qing Han and Jia-Xing Hong, Isometric Embedding of Riemannian Manifolds in Euclidean Spaces, Mathematical Surveys and Monographs, vol. 130, American Mathematical Society, Providence, RI, 2006. [IL03] Thomas A. Ivey and J. M. Landsberg, Cartan for Beginners: Diļ¬erential Geometry via Moving Frames and Exterior Diļ¬erential Systems, Graduate Studies in Mathematics, vol. 61, American Mathematical Society, Providence, RI, 2003. [Jan26] M. Janet, Sur la possibilitĀ“ e de plonger un espace Riemannien donnĀ“ e dans un espace euclidien, Ann. Soc. Polon. Math. 5 (1926), 38ā43. [Jen77] Gary R. Jensen, Higher Order Contact of Submanifolds of Homogeneous Spaces, Springer-Verlag, Berlin, 1977, Lecture Notes in Mathematics, Vol. 610. [Kle93a] Felix Klein, A comparative review of recent researches in geometry, Bull. Amer. Math. Soc. 2 (1893), no. 10, 215ā249. [Kle93b]
, Vergleichende Betrachtungen u ĀØber neuere geometrische Forschungen, Math. Ann. 43 (1893), no. 1, 63ā100.
[Kob95] Shoshichi Kobayashi, Transformation Groups in Diļ¬erential Geometry, Classics in Mathematics, Springer-Verlag, Berlin, 1995, reprint of the 1972 edition. [Kog03] Irina Kogan, Two algorithms for a moving frame construction, Canad. J. Math. 55 (2003), 266ā291. [Lee13] John M. Lee, Introduction to Smooth Manifolds, 2nd ed., Graduate Texts in Mathematics, vol. 218, Springer, New York, 2013. [Man10] Elizabeth Louise Mansļ¬eld, A Practical Guide to the Invariant Calculus, Cambridge Monographs on Applied and Computational Mathematics, vol. 26, Cambridge University Press, Cambridge, 2010. [Mil83] Tilla Klotz Milnor, Harmonic maps and classical surface theory in Minkowski 3-space, Trans. Amer. Math. Soc. 280 (1983), no. 1, 161ā185. [Min78] Hermann Minkowski, Die Grundgleichungen fĀØ ur die elektromagnetischen VorgĀØ ange in bewegten KĀØ orpern, Nachrichten von der Gesellschaft der Wissenschaften zu GĀØ ottingen, Mathematisch-Physikalische Klasse (1907/8), 53ā111. [Min89]
, Raum und Zeit, Jahresbericht der Deutschen Mathematiker-Vereinigung (1908/9), 75ā88.
[MR05] SebastiĀ“ an Montiel and Antonio Ros, Curves and surfaces, Graduate Studies in Mathematics, vol. 69, American Mathematical Society, Providence, RI; Real Sociedad MatemĀ“ atica EspaĖ nola, Madrid, 2005, translated and updated from the 1998 Spanish edition by the authors. [Nas56] John Nash, The imbedding problem for Riemannian manifolds, Ann. of Math. (2) 63 (1956), 20ā63. [NMS08] Mehdi Nadjaļ¬khah and Ali Mahdipour Sh., Aļ¬ne classiļ¬cation of n-curves, Balkan J. Geom. Appl. 13 (2008), no. 2, 66ā73. [NS94] Katsumi Nomizu and Takeshi Sasaki, Aļ¬ne Diļ¬erential Geometry, Geometry of aļ¬ne immersions, Cambridge Tracts in Mathematics, vol. 111, Cambridge University Press, Cambridge, 1994. [Olv00] Peter Olver, Applications of Lie Groups to Diļ¬erential Equations, 2nd ed., Graduate Texts in Mathematics, Springer-Verlag, New York, 2000.
Bibliography
401
[Olv10]
, Recent advances in the theory and application of Lie pseudo-groups, XVIII International Fall Workshop on Geometry and Physics, AIP Conf. Proc., vol. 1260, Amer. Inst. Phys., Melville, NY, 2010, pp. 35ā63.
[Olv11a]
, Lectures on Moving Frames, Symmetries and Integrability of Diļ¬erence Equations (Decio Levi, Peter Olver, Zora Thomova, and Pavel Winternitz, eds.), London Mathematical Society Lecture Note Series, vol. 381, Cambridge University Press, Cambridge, 2011, Lectures from the Summer School (SĀ“ eminaire de MĀ“ athematiques SupĀ“ erieures) held at the UniversitĀ“ e de MontrĀ“ eal, MontrĀ“eal, QC, June 8ā21, 2008.
[Olv11b]
, Recursive moving frames, Results Math. 60 (2011), 423ā452.
[OāN83] Barrett OāNeill, Semi-Riemannian Geometry, with applications to relativity, Pure and Applied Mathematics, vol. 103, Academic Press Inc. [Harcourt Brace Jovanovich Publishers], New York, 1983. [OāN06]
, Elementary Diļ¬erential Geometry, 2nd ed., Elsevier/Academic Press, Amsterdam, 2006.
[Opr07] John Oprea, Diļ¬erential Geometry and Its Applications, 2nd ed., Classroom Resource Materials Series, Mathematical Association of America, Washington, DC, 2007. [OT05] V. Ovsienko and S. Tabachnikov, Projective Diļ¬erential Geometry Old and New, Cambridge Tracts in Mathematics, vol. 165, Cambridge University Press, Cambridge, 2005. [Rad30] Tibor RadĀ“ o, On Plateauās problem, Ann. of Math. (2) 31 (1930), no. 3, 457ā469. [Rey98] E. G. Reyes, Pseudo-spherical surfaces and integrability of evolution equations, J. Diļ¬. Eq. 147 (1998), 195ā230. [RS82] C. Rogers and W. F. Shadwick, BĀØ acklund Transformations and Their Applications, Mathematics in Science and Engineering, vol. 161, Academic Press Inc. [Harcourt Brace Jovanovich Publishers], New York, 1982. [RS02] C. Rogers and W. K. Schief, BĀØ acklund and Darboux Transformations, Geometry and modern applications in soliton theory, Cambridge Texts in Applied Mathematics, Cambridge University Press, Cambridge, 2002. [Sen31] C. E. Senļ¬, Theoremata Princioalia e Theoria Curvarum et Su Perļ¬cierum, Dorpat Univ., 1831. [Ser51] J. Serret, MĀ“ emoire sur quelques formules relatives ` a la thĀ“ eorie des courbes a ` double courbure, J. Math. Pures Appl. 16 (1851), 193ā207. [She99] Mary D. Shepherd, Line congruences as surfaces in the space of lines, Diļ¬erential Geom. Appl. 10 (1999), no. 1, 1ā26. [Spi79] Michael Spivak, A Comprehensive Introduction to Diļ¬erential Geometry. 5 Vols., 2nd ed., Publish or Perish Inc., Wilmington, DE, 1979. [Ste07] James Stewart, Calculus: Early Transcendentals, 6th ed., Stewartās Calculus Series, Brooks Cole, Belmont, CA, 2007. [Su83] Bu Chin Su, Aļ¬ne Diļ¬erential Geometry, Science Press, Beijing, 1983. [Tsu96] Kazumi Tsukada, Totally geodesic submanifolds of Riemannian manifolds and curvature-invariant subspaces, Kodai Math. J. 19 (1996), no. 3, 395ā437. ĀØ [Wei66] K. Weierstrass, Uber die FlĀØ achen deren mittlere KrĀØ ummung u ĀØberall gleich null ist., Monatsber. Berliner Akad. (1866), 612ā625, 855ā856.
402
Bibliography
[Wil61] T. J. Willmore, The deļ¬nition of Lie derivative, Proc. Edinburgh Math. Soc. (2) 12 (1960/1961), 27ā29. [Wil62] E. J. Wilczynski, Projective Diļ¬erential Geometry of Curves and Ruled Surfaces, Chelsea Publishing Co., New York, 1962.
Index
0-form, 42 1-form, 35 on Rn , 40ā41 on a manifold, 47 An , see Equi-aļ¬ne space Adapted frame ļ¬eld, 109 on a surface in E3 , 118 on a surface in A3 , 178 on a surface in P3 , 221 on a timelike surface in M1,2 , 150 equi-aļ¬ne principal adapted frame ļ¬eld on an elliptic surface in A3 , 186 null adapted frame ļ¬eld on a hyperbolic surface in A3 , 190 on a hyperbolic surface in P3 , 232 on a timelike surface in M1,2 , 162 principal adapted frame ļ¬eld on a surface in E3 , 123 on a timelike surface in M1,2 , 155 Aļ¬ne connection, see Connection Aļ¬ne geometry, 92 Aļ¬ne Grassmannian, 288, 324 Aļ¬ne transformation, 93 Arc length, see Curve, arc length Area functional on surfaces in E3 , 252 equi-aļ¬ne, on surfaces in A3 , 269 Area measure, 324 Associated family of a minimal surface in E3 , 267
BĀØ acklund transformation for Liouvilleās equation, 302ā303 for pseudospherical surfaces, 290 for the sine-Gordon equation, 288, 298 BĀØ acklundās theorem, 287, 290 BĀØ acklund, Albert, 290 Baker-Jarvis, Duļ¬, xvi Bartels, Martin, xi Bianchi, Luigi, 290 Blaschke representation for an elliptic equi-aļ¬ne minimal surface in A3 , 278 Blaschke, Wilhelm, 178, 274 Bonnetās theorem for a surface in E3 existence, 127 uniqueness, 124 for a surface in S3 or H3 , 354 for a timelike surface in M1,2 , 155 Bryant, Robert, xv Bushek, Nathaniel, xv Canonical isomorphism for dual spaces, 36 for tangent spaces, 16, 50, 76, 339, 367 Carlsen, Brian, xvi Cartan package for Maple, xiii, 59ā66 &Ė command, 60 d command, 60 Forder command, 60 Form command, 59
403
404
makebacksub command, 63 pick command, 62 ScalarForm command, 63 Simf command, 61 WedgeProduct command, 60 Cartan structure equations, see Structure equations Cartanās formula for exterior derivative, 48 Cartanās formula for Lie derivative, 58 Cartanās lemma, 55 Ā“ Cartan, Elie, xi, 70, 223, 233, 383 Cartan-Janet isometric embedding theorem, 383 Catenoid, 128, 260 associated family, 268 conjugate surface, 268 Weierstrass-Enneper representation, 268 Cauchy-Crofton formula, 324, 327 Cauchy-Riemann equations, 264 Chain rule, 24 Chern, Shiing-Shen, 297 Clelland, Richard, xvi Codazzi equations for a surface in E3 , 127 for a surface in S3 , 353, 354 for a surface in H3 , 353, 354 for a timelike surface in M1,2 , 156, 165 for a submanifold of En+m , 379 Column vector, see Vector, column vector Commutative diagram, 19 Compatibility equations for a surface in E3 , 127 for a surface in S3 or H3 , 353 for a timelike surface in M1,2 , 156, 165 for an elliptic surface in A3 , 188 for an elliptic surface in P3 , 229, 247 for a hyperbolic surface in P3 , 235 for a submanifold of En+m , 379 Complex analytic function, see Holomorphic function Complex structure, 264 Conformal parametrization of a surface, 265 Conformal structure on a hyperbolic surface in P3 , 232 on an elliptic surface in P3 , 224
Index
Conic section, 177, 212 Conjugate surface of a minimal surface in E3 , 267 Connection, 33 compatibility with a metric, 371 curvature tensor, 376 ļ¬at connection on En , 366 Levi-Civita, see Levi-Civita connection on a vector bundle, 365 on the tangent bundle, 365ā370 horizontal tangent space, 367 vertical tangent space, 366 symmetric, 371 torsion-free, 371 Connection forms on the orthonormal frame bundle of En , 79 on the orthonormal frame bundle of M1,n , 91 on the unimodular frame bundle of An , 95 on the projective frame bundle of Pn , 103 for the Levi-Civita connection on Sn or Hn , 347 determined by a connection, 367, 370 Constant type, 316 Cotangent bundle, 36 Cotangent space, 36 Covariant derivative, 33 for vector ļ¬elds on Sn and Hn , 346ā347 compatibility with the metric, 347 for vector ļ¬elds on a submanifold of En+m , 378 Covector, 37 Covector space, 36 Cruz, Akaxia, xvi Curvature, see also Curve, curvature; Gauss curvature; mean curvature curvature matrix of a connection matrix, 340 curvature matrix of the connection matrix on F (Sn ), 342 curvature matrix of the connection matrix on F (Hn ), 344 curvature tensor of a connection, 376 Curve in E3 arc length, 112
Index
binormal vector, 112 complete set of invariants, 115 curvature, 113 Frenet equations, 114 Frenet frame, 112 nondegenerate curve, 112 orthonormal frame ļ¬eld, 111 regular curve, 111 torsion, 113 unit normal vector, 112 unit tangent vector, 111 in M1,2 , null curve, 165 in M1,2 , timelike curve Frenet equations, 148 Minkowski curvature, 147 Minkowski torsion, 148 nondegenerate curve, 146 orthonormal frame ļ¬eld, 144 proper time, 144 regular curve, 144 unit normal vector, 146 unit tangent vector, 144 in A2 , 176ā178 conic section, 177 equi-aļ¬ne curvature, 177 in A3 equi-aļ¬ne arc length, 174ā175 equi-aļ¬ne curvatures, 176 equi-aļ¬ne Frenet equations, 176 equi-aļ¬ne Frenet frame, 175 nondegenerate curve, 172 rational normal curve, 178 unimodular frame ļ¬eld, 172 in P2 canonical lifting, 205 canonical projective frame ļ¬eld, 205 conic section, 212 nondegenerate curve, 205 projective arc length, 211 projective curvature form, 210 projective frame ļ¬eld, 204 projective Frenet equations, 212 projective parameter, 207 projective parametrization, 207 projective structure, 210 Wilczynski invariants, 206 in P3 canonical lifting, 215 canonical projective frame ļ¬eld, 215
405
nondegenerate curve, 215 projective curvature forms, 218 projective frame ļ¬eld, 214 projective Frenet equations, 219 projective parameter, 217 projective parametrization, 217 projective structure, 217 rational normal curve, 220 Wilczynski invariants, 216 in S3 binormal vector, 350 curvature, 350 Frenet equations, 350 Frenet frame, 350 geodesic, 349 geodesic equation, 349 nondegenerate curve, 349 orthonormal frame ļ¬eld, 348 regular curve, 348 torsion, 350 unit normal vector, 350 in H3 binormal vector, 350 curvature, 350 Frenet equations, 350 Frenet frame, 350 geodesic, 349 geodesic equation, 349 nondegenerate curve, 349 orthonormal frame ļ¬eld, 348 regular curve, 348 torsion, 350 unit normal vector, 350 in a Riemannian 3-manifold curvature, 381 Frenet equations, 381 Frenet frame, 381 geodesic, 380 nondegenerate curve, 380 orthonormal frame ļ¬eld, 379 regular curve, 379 torsion, 381 Darboux tangents, 227 Darboux, Jean-Gaston, xi De Sitter spacetime, 157ā158 Derivative directional, 19, 43, 57, 120, 347, 365 of a map from Rm to Rn , 16 of a map between manifolds, 23 Diļ¬eomorphism, 25 Diļ¬erentiable manifold, see Manifold
406
Diļ¬erential of a real-valued function, 35 of a map from Rm to Rn , 16 of a map between manifolds, 24, 49 Diļ¬erential form 0-form, 42 1-form, 35 on Rn , 40ā41 on a manifold, 47 p-form on Rn , 42 on a manifold, 47 algebra of diļ¬erential forms on Rn , 41, 42 closed form, 46 exact form, 46 DifferentialGeometry package for Maple, xiii Directional derivative, see Derivative, directional Divergence theorem, 55 Doubly ruled surface, see Ruled surface, doubly ruled surface Dual forms on the orthonormal frame bundle of En , 79 on the projective frame bundle of Pn , 103 associated to an orthonormal frame ļ¬eld, 369 Dual space, 35ā36 Dunne, Edward, xv En , see Euclidean space Einstein summation convention, 14ā15 Einstein, Albert, 85 Elliptic paraboloid, 272 Blaschke representation, 280 Elliptic space, 340ā342, see also Homogeneous space, elliptic space Sn Elliptic surface in A3 , 180ā189 in P3 , 223ā232 Embedding, 25 Enneperās surface, 268 Enneper, Alfred, 261 Equi-aļ¬ne arc length, see Curve in A3 , equi-aļ¬ne arc length Equi-aļ¬ne ļ¬rst fundamental form, see Surface in A3 , equi-aļ¬ne ļ¬rst fundamental form
Index
Equi-aļ¬ne geometry, 92 Equi-aļ¬ne group A(n), 94 as a principal bundle over An , 95 Equi-aļ¬ne mean curvature, see Surface in A3 , equi-aļ¬ne mean curvature Equi-aļ¬ne minimal surface, see Minimal surface, equi-aļ¬ne, in A3 Equi-aļ¬ne normal vector ļ¬eld, see Surface in A3 , equi-aļ¬ne normal vector ļ¬eld Equi-aļ¬ne second fundamental form, see Surface in A3 , equi-aļ¬ne second fundamental form Equi-aļ¬ne space, 93, see also Homogeneous space, equi-aļ¬ne space An volume form, 92 Equi-aļ¬ne sphere improper equi-aļ¬ne sphere, 189 proper equi-aļ¬ne sphere, 189 Equi-aļ¬ne transformation, 93 Equivalence problem, 107 Equivariant, 109 Equivariant moving frame, see Moving frame, equivariant moving frame Estrada, Edward, xvi Euclidean group E(n), 73 as a principal bundle over En , 75 Euclidean space, 70, see also Homogeneous space, Euclidean space En Exterior derivative of a real-valued function, 35 of a p-form on Rn , 43ā46 of a p-form on a manifold, 48ā49 chain rule, 44 Leibniz rule, 43, 44 Extrinsic curvature of a surface in S3 or H3 , 353 Fels, Mark, xi First fundamental form of a surface in E3 , 118ā120 of a surface in S3 or H3 , 352 of a surface in a Riemannian 3-manifold, 382 of a timelike surface in M1,2 , 150, 163 equi-aļ¬ne, of an elliptic surface in A3 , 181 equi-aļ¬ne, of a hyperbolic surface in A3 , 190
Index
projective, of an elliptic surface in P3 , 227 Flat connection on En , 366 Flat homogeneous space, 339 Flat surface in E3 , 132ā134 in S3 , 355ā356 ļ¬at torus, 356 in H3 , 356ā357 ļ¬at cylinder, 356 Frenet, Jean, xi Frobenius theorem, 46 Fubini-Pick form of a hyperbolic surface in A3 , 191 of an elliptic surface in A3 , 185 of an elliptic surface in P3 , 226 Fundamental Theorem of Calculus, 54 Fundamental Theorem of Space Curves, 69 existence, 117 uniqueness, 114 GL(n), 28, 29 gl(n), 30 Gauge, 368 Gauge ļ¬eld, 368 Gauge transformation, 368 Gauss curvature of a surface in E3 , 131 of a surface in S3 or H3 , 353 of a timelike surface in M1,2 , 153, 163 Gauss equation for a surface in E3 , 127 for a surface in S3 , 353, 354 for a surface in H3 , 353, 354 for a timelike surface in M1,2 , 156, 165 for a submanifold of En+m , 379 Gauss map of a surface in E3 , 121 of a surface in S3 or H3 , 352 of a surface in a Riemannian 3-manifold, 382 of a timelike surface in M1,2 , 151 Gauss, Carl Friedrich, 131 Theorema Egregium, 131 Gelfand, Sergei, xv General linear group, see GL(n) General relativity, 143 Geodesic in S3 or H3 , 349 in a Riemannian 3-manifold, 380
407
Geodesic equation for curves in S3 or H3 , 349 for curves in a Riemannian 3-manifold, 380 Geodesic spray, 380ā381 Grassmannian, aļ¬ne, 288, 324 Great hyperboloid in H3 , 351, 355 Great sphere in S3 , 351, 355 Greenās theorem, 55 Guggenheimer, Heinrich, xi Hn , see Hyperbolic space Harmonic function, 264 Helicoid, 261, 268 Helm, Rachel, xvi Hilbertās theorem, 301ā302 Holomorphic function, 263 Homogeneous space, 70, 84, 361 ļ¬at homogeneous space, 339 Euclidean space En , 70ā75 Minkowski space M1,n , 85ā92 equi-aļ¬ne space An , 92ā96 projective space Pn , 96ā103 elliptic space Sn , 340ā342 hyperbolic space Hn , 340, 342ā344 Horizontal tangent space, 367 Horizontal vector ļ¬eld, 380 Hyperbolic paraboloid, 311, 319 Hyperbolic plane, 301 Hyperbolic space, 340, 342ā344, see also Homogeneous space, hyperbolic space Hn Hyperbolic surface in A3 , 180, 189ā191 in P3 , 223, 232ā235 Hyperboloid of one sheet, 311 Immersion, 25 Incidence, of a point and a line, 327 Indices lower index, 9 upper index, 9 in partial derivative operators, 13 Inner product Euclidean, 70 Minkowski, 86 Integrable system, 288 soliton solution, 288 Interior product, 57 Intrinsic curvature of a surface in S3 or H3 , 353
408
Index
Intrinsic invariant, see Invariant, intrinsic invariant for surfaces in E3 Invariant, 107 for curves in E3 , 69 for submanifolds of a homogeneous space, 109 complete set of invariants, 107 for curves in E3 , 115 intrinsic invariant for surfaces in E3 , 131 relative invariant, 226, 315 Isometric embedding, 378ā379, 383 Cartan-Janet theorem, 383 Isotropy group of a point in En , 73 of a point in M1,n , 90 of a point in An , 94 of a point in Pn , 101 of a point in Sn , 340 of a point in Hn , 343
Line congruence, 288ā289 focal surface, 289 normal congruence, 289 pseudospherical congruence, 289ā290 surface of reference, 289 Linear fractional transformation, 99 Liouvilleās equation, 302, 320 BĀØ acklund transformation, 302ā303 Local coordinates on a surface, 4, 5 on a manifold, 6 Local trivialization of a vector bundle, 32 of a tangent bundle, 364 of an orthonormal frame bundle, 369 Lorentz group, 89 proper, orthochronous, 89 Lorentz transformation, 89 orthochronous, 89 proper, 89
Janet, Maurice, 383 Jensen, Andrew, xvi Joeris, Peter, xvi
M1,n , see Minkowski space Mahoney, Michael, xvi Manifold, 5 local coordinates, 6 transition map between, 6 parametrization, 6 Riemannian manifold, 362 Maple, xiii, 59ā66, 103ā106, 134ā141, 166ā169, 191ā201, 235ā247, 280ā286, 303ā309, 329ā335, 357ā360, 388ā395 Mapping continuous, 15 diļ¬erentiable from Rm to Rn , 15 between manifolds, 18 Mathematical Sciences Research Institute, xvi Maurer-Cartan equation, see also Structure equations on a Lie group, 85 on the Euclidean group E(n), 82 on the elliptic symmetry group SO(n + 1), 342 on the hyperbolic symmetry group SO+ (1, n), 344 Maurer-Cartan form on a Lie group, 85 on the Euclidean group E(n), 81ā82 on the PoincarĀ“e group M (1, n), 91 on the equi-aļ¬ne group A(n), 95
Karpel, Joshua, xvi Kaufman, Bryan, xv Klein, Felix, 69 Lagrange, Joseph-Louis, 251 Laplaceās equation, 356 Left-hook, 57 Levi-Civita connection, 33, 370ā372 on En , 366 on Sn or Hn , 347 connection forms, 347 Riemann curvature tensor, 376ā378 Lie algebra, 26ā32 Lie bracket, 26 of vector ļ¬elds, 27 on a Lie algebra, 28ā29 Lie derivative, 56ā59, 258 Cartanās formula, 58 Lie group, 26ā32 left translation map, 26 left-invariant vector ļ¬eld, 26ā27 right translation map, 26 Lifting, 109 Light cone, see Minkowski space, light cone Lightlike vector, see Minkowski space, lightlike vector
Index
on the projective symmetry group SL(n + 1), 102 on the elliptic symmetry group SO(n + 1), 341 on the hyperbolic symmetry group SO+ (1, n), 344 May, Molly, xvi Mean curvature of a surface in E3 , 131 of a surface in S3 or H3 , 353 of a timelike surface in M1,2 , 153, 163 equi-aļ¬ne, of an elliptic surface in A3 , 185 Measure, 324 area measure, 324 Meromorphic function, 266 Method of moving frames, see Moving frame, method of moving frames Metric, 13ā14 Metric structure on a curve in En , 209 Miller, Jonah, xvi Minimal surface in E3 , 132, 251ā268 associated family, 267 catenoid, 128, 260, 268 conjugate surface, 267 Enneperās surface, 268 helicoid, 261, 268 Weierstrass-Enneper representation, 266ā267 equi-aļ¬ne, in A3 , 268ā280 Blaschke representation, 278 elliptic paraboloid, 272, 280 Minkowski cross product, 146 Minkowski norm, 88 Minkowski space, 86, see also Homogeneous space, Minkowski space M1,n future-pointing vector, 87 light cone, 87 lightlike vector, 87 Minkowski norm of a vector, 88 null cone, 87 null vector, 87 past-pointing vector, 87 spacelike vector, 87 timelike vector, 87 world line of a particle, 88 Minkowski, Hermann, 85 Moving frame equivariant moving frame, xi
409
method of moving frames, 70, 107, 111 Nash embedding theorem, 378 National Science Foundation, xvi Nondegenerate curve, see Curve, nondegenerate Null adapted frame ļ¬eld on a timelike surface in M1,2 , 162 on a hyperbolic surface in A3 , 190 on a hyperbolic surface in P3 , 232 Null cone, see Minkowski space, null cone Null coordinates on a timelike surface in M1,2 , 165 Null curve in M1,2 , 165 Null vector, see Minkowski space, null vector O(1, n), 89 O(n), 31 o(n), 31 Olver, Peter, xi Orthogonal group, see O(n) Orthonormal basis for En , 72 for M1,n , 87 Orthonormal frame on En , 75 on M1,n , 91 on Sn , 341, 345 on Hn , 343, 345 on a Riemannian manifold, 363 Orthonormal frame bundle of En , 75 of M1,n , 91 of S2 , 34 of Sn , 341, 345 of Hn , 343, 345 of a Riemannian manifold, 363 local trivialization, 369 Orthonormal frame ļ¬eld on En , 83 along a curve in E3 , 111 along a curve in S3 or H3 , 348 along a curve in a Riemannian 3-manifold, 379 along a timelike curve in M1,2 , 144 p-form on Rn , 42ā43 on a manifold, 47
410
P GL(m), 98 Pn , see Projective space P SL(m), 98 Paraboloid elliptic paraboloid, 272 Blaschke representation, 280 hyperbolic paraboloid, 311, 319 Parametrization of a surface, 4, 5 of a manifold, 6 asymptotic, 295 conformal, 265 principal, 128, 156, 187 Partial derivative operator as a tangent vector, 20 indices in, 13 Peneyra, Sean, xvi Pick invariant of an elliptic surface in A3 , 186 Plateau problem, 251 Plateau, Joseph, 251 PoincarĀ“e group M (1, n), 90 as a principal bundle over M1,n , 91 PoincarĀ“e lemma, 46 PoincarĀ“e-Hopf theorem, 33, 34 Principal adapted frame ļ¬eld on a surface in E3 , 123 on a timelike surface in M1,2 , 155 equi-aļ¬ne, on an elliptic surface in A3 , 186 Principal bundle, 33ā34, 362 base space, 33 base-point projection map, 33 ļ¬ber, 33 section, 33 total space, 33 local trivialization, 369 Principal curvatures of a surface in E3 , 123 of a surface in S3 or H3 , 352 of a timelike surface in M1,2 , 155 surface in E3 with constant principal curvatures, 130ā131 Principal vectors on a surface in E3 , 123 on a surface in S3 or H3 , 352 on a timelike surface in M1,2 , 155 Projective arc length, see Curve in P2 /P3 , projective arc length Projective curvature form, see Curve in P2 /P3 , projective curvature form
Index
Projective ļ¬rst fundamental form, see Surface in P3 , projective ļ¬rst fundamental form Projective frame bundle of Pn , 102 Projective frame ļ¬eld along a curve in P2 , 204 canonical projective frame ļ¬eld, 205 along a curve in P3 , 214 canonical projective frame ļ¬eld, 215 Projective frame on Pn , 101 Projective general linear group, 98 Projective parametrization, see Curve in P2 /P3 , projective parametrization Projective space, 7ā9, 96, see also Homogeneous space, projective space Pn aļ¬ne coordinates, 97 homogeneous coordinates, 8 Projective special linear group, 98 Projective sphere, 229ā232 Projective structure on a curve in P2 , 210 on a curve in P3 , 217 on a curve in Pn , 203 Projective transformation, 96, 97 Schwarzian derivative, 208 Proper time, see Curve in M1,2 , proper time Pseudosphere, 287 Pseudospherical line congruence, 289ā290 Pseudospherical surface, 287 1-soliton pseudospherical surface, 301 asymptotic coordinates, 295 asymptotic parametrization, 295 Pullback for diļ¬erential forms, 50ā53 for bundles, 108 Push-forward, 50 Quasi-umbilic point on a timelike surface in M1,2 , 160 Rational normal curve in A3 , 178 in P3 , 220 Regular curve, see Curve, regular Regular surface, see Surface Relative invariant, 226, 315
Index
Relativity special relativity, 85, 143 general relativity, 143 Reyes, Enrique, 297 Ricci equations for a submanifold of En+m , 379 Riemann curvature tensor, 376ā378 ļ¬rst Bianchi identity, 377 on a Riemannian 3-manifold, 385 Riemannian manifold, 362 Row vector, see Vector, row vector Ruled surface, 311 doubly ruled surface, 311 0-adapted frame ļ¬eld, 314 1-adapted frame ļ¬eld, 316 2-adapted frame ļ¬eld, 317 classiļ¬cation theorem, 313 hyperbolic paraboloid, 311, 319 hyperboloid of one sheet, 311 SL(n), 30ā31 sl(n), 30 SL(n + 1) as a principal bundle over Pn , 102 as the symmetry group of Pn , 98 Sn , 30 Sn , see Elliptic space; Unit sphere SO+ (1, n), 89 as a principal bundle over Hn , 344 as the symmetry group of Hn , 342 so(1, n), 90 SO(n), 31 SO(n + 1) as a principal bundle over Sn , 341 as the symmetry group of Sn , 340 Schmidt, Michael, xvi Schwarzian derivative, 208ā209 of a projective transformation, 208 Second fundamental form of a surface in E3 , 121ā122 of a surface in S3 or H3 , 352 of a surface in a Riemannian 3-manifold, 382 of a timelike surface in M1,2 , 151, 163 equi-aļ¬ne, of an elliptic surface in A3 , 184 equi-aļ¬ne, of a hyperbolic surface in A3 , 190 of a submanifold of En+m , 378 Self-adjoint linear operator, 152 Semi-basic forms
411
on the orthonormal frame bundle of En , 79 on the projective frame bundle of Pn , 103 Serret, Joseph, xi Simple connectivity, 116 Sine-Gordon equation, 288 1-soliton solution, 300 BĀØ acklund transformation, 288, 298 in characteristic/null coordinates, 296 in space-time coordinates, 296 Skew curvature of a timelike surface in M1,2 , 154, 163 Smooth manifold, see Manifold Soliton, 288 1-soliton pseudospherical surface, 301 1-soliton solution of the sine-Gordon equation, 300 Spacelike surface, see Surface in M1,2 , spacelike surface Spacelike vector, see Minkowski space, spacelike vector Special aļ¬ne geometry, see Equi-aļ¬ne geometry Special linear cross product, 277 Special linear group, see SL(n) Special orthogonal group, see SO(n) Special relativity, 85, 143 Stokesās theorem, 53ā55 Divergence theorem, 55 Fundamental Theorem of Calculus, 54 Greenās theorem, 55 Stokesās theorem, multivariable calculus version, 55 Structure equations on the orthonormal frame bundle of En , 80 on the orthonormal frame bundle of M1,n , 91 on the unimodular frame bundle of An , 95 on the projective frame bundle of Pn , 102 on the orthonormal frame bundle of Sn , 341 on the orthonormal frame bundle of Hn , 344 on the orthonormal frame bundle of a Riemannian manifold, 374, 377 Submersion, 25
412
Surface, 3, 5 parametrization, 4, 5 local coordinates, 4, 5 transition map between, 5 ruled surface, see Ruled surface doubly ruled surface, see Ruled surface, doubly ruled surface in E3 adapted frame ļ¬eld, 118 area functional, 252 Bonnetās theorem, 127 catenoid, 128, 260, 268 Codazzi equations, 127 compatibility equations, 127 Enneperās surface, 268 ļ¬rst fundamental form, 118ā120 ļ¬at surface, 132ā134 Gauss curvature, 131 Gauss equation, 127 Gauss map, 121 helicoid, 261, 268 mean curvature, 131 minimal surface, 132, 251ā268 principal adapted frame ļ¬eld, 123 principal curvatures, 123 principal vectors, 123 pseudosphere, 287 pseudospherical surface, 287 second fundamental form, 121ā122 shape operator, 121 surface with constant principal curvatures, 130ā131 totally umbilic surface, 129 umbilic point, 124 variation, 252ā255 in A3 0-adapted frame ļ¬eld, 178 in A3 , elliptic surface, 180ā189 1-adapted frame ļ¬eld, 180 2-adapted frame ļ¬eld, 183 compatibility equations, 188 cubic form, 185 elliptic paraboloid, 272, 280 equi-aļ¬ne area functional, 269 equi-aļ¬ne ļ¬rst fundamental form, 181 equi-aļ¬ne mean curvature, 185 equi-aļ¬ne normal vector ļ¬eld, 183 equi-aļ¬ne principal adapted frame ļ¬eld, 186
Index
equi-aļ¬ne second fundamental form, 184 Fubini-Pick form, 185 improper equi-aļ¬ne sphere, 189 minimal surface, 268ā280 Pick invariant, 186 proper equi-aļ¬ne sphere, 189 variation, 269 in A3 , hyperbolic surface, 180, 189ā191 1-adapted null frame ļ¬eld, 190 2-adapted null frame ļ¬eld, 190 equi-aļ¬ne ļ¬rst fundamental form, 190 equi-aļ¬ne second fundamental form, 190 Fubini-Pick form, 191 hyperbolic paraboloid, 311, 319 hyperboloid of one sheet, 311 in M1,2 , spacelike surface, 143 in M1,2 , timelike surface, 143 adapted frame ļ¬eld, 150 Codazzi equations, 156, 165 compatibility equations, 156, 165 de Sitter spacetime, 157ā158 ļ¬rst fundamental form, 150, 163 Gauss curvature, 153, 163 Gauss equation, 156, 165 Gauss map, 151 mean curvature, 153, 163 null adapted frame ļ¬eld, 162 null coordinates, 165 principal adapted frame ļ¬eld, 155 principal curvatures, 155 principal vectors, 155 quasi-umbilic point, 160 second fundamental form, 151, 163 skew curvature, 154, 163 totally quasi-umbilic surface, 160, 165ā166 totally umbilic surface, 156ā157 umbilic point, 155 in P3 0-adapted frame ļ¬eld, 221 in P3 , elliptic surface, 223ā232 1-adapted frame ļ¬eld, 223 2-adapted frame ļ¬eld, 225 3-adapted frame ļ¬eld, 226 4-adapted frame ļ¬eld, 228 compatibility equations, 229, 247 conformal structure, 224
Index
cubic form, 226 Darboux tangents, 227 Fubini-Pick form, 226 projective ļ¬rst fundamental form, 227 projective sphere, 229ā232 totally umbilic surface, 229ā232 umbilic point, 226 in P3 , hyperbolic surface, 223, 232ā235 1-adapted null frame ļ¬eld, 232 2-adapted null frame ļ¬eld, 233 3-adapted null frame ļ¬eld, 234 4-adapted null frame ļ¬eld, 234 compatibility equations, 235 conformal structure, 232 in S3 Bonnetās theorem, 354 Codazzi equations, 353, 354 compatibility equations, 353 extrinsic curvature, 353 ļ¬rst fundamental form, 352 ļ¬at surface, 355ā356 ļ¬at torus, 356 Gauss curvature, 353 Gauss equation, 353, 354 Gauss map, 352 great sphere, 351, 355 intrinsic curvature, 353 mean curvature, 353 principal curvatures, 352 principal vectors, 352 second fundamental form, 352 totally geodesic surface, 355 in H3 Bonnetās theorem, 354 Codazzi equations, 353, 354 compatibility equations, 353 extrinsic curvature, 353 ļ¬rst fundamental form, 352 ļ¬at cylinder, 356 ļ¬at surface, 356ā357 Gauss curvature, 353 Gauss equation, 353, 354 Gauss map, 352 great hyperboloid, 351, 355 intrinsic curvature, 353 mean curvature, 353 principal curvatures, 352 principal vectors, 352 second fundamental form, 352
413
totally geodesic surface, 355 in a Riemannian 3-manifold ļ¬rst fundamental form, 382 Gauss map, 382 second fundamental form, 382 totally geodesic surface, 383ā388 Symmetric group, see Sn Symmetric product of vectors, 39 of 1-forms, 119 Symmetry group of En , 73 of M1,n , 90 of An , 94 of Pn , 98 of Sn , 340 of Hn , 342 of a homogeneous space G/H, 84 as a principal bundle over G/H, 85 as the set of frames on G/H, 85 Tangent bundle, 21ā23 of a surface, 21ā23 of a manifold, 21 base space, 22 total space, 22 ļ¬ber, 22 base-point projection map, 23 canonical parametrization, 22 transition map between, 22 local trivialization, 364 Tangent space, 16, 20 tangent plane, 21 Tangent vector, 16, 19 Tenenblat, Keti, 297 Tensor, 9ā14 change of basis, 9ā10, 12ā13 components, 10, 12, 13 metric, 13 rank 1, 10 rank 2, 12 rank k, 38 skew-symmetric, 38ā39 symmetric, 38ā39 Tensor bundle, 39 Tensor ļ¬eld, 9, 13 rank k, 40 Tensor product, 37ā38 symmetric product, 39 wedge product, 39 Theorema Egregium (Gauss), 131
414
Timelike curve, see Curve in M1,2 , timelike curve Timelike surface, see Surface in M1,2 , timelike surface Timelike vector, see Minkowski space, timelike vector Totally geodesic surface in S3 or H3 , 355 in a Riemannian 3-manifold, 383ā388 Totally quasi-umbilic timelike surface in M1,2 , 160, 165ā166 Totally umbilic surface in E3 , 129 in M1,2 , timelike surface, 156ā157 in P3 , elliptic surface, 229ā232 Transition map between local coordinates on a surface, 5 between local coordinates on a manifold, 6 Transpose notation for matrices, 31 for vectors, 6 Umbilic point on a surface in E3 , 124 on a timelike surface in M1,2 , 155 on an elliptic surface in P3 , 226 Unimodular frame bundle of An , 95 Unimodular frame ļ¬eld along a curve in A3 , 172 Unimodular frame on An , 94 Unit sphere Sn , 6ā7 Variation of a surface in E3 , 252ā255 compactly supported, 253 normal, 253 of an elliptic surface in A3 , 269 compactly supported, 269 normal, 269 Vatuk, Sunita, xv, xvi Vector column vector, 6 row vector, 6 tangent vector, 16, 19 transpose notation for, 6 Vector bundle, 32ā33 base space, 32 total space, 32 ļ¬ber, 32 base-point projection map, 32
Index
rank k, 32 section, 32ā33 global section, 32 local section, 32 zero section, 33 trivialization global trivialization, 32 local trivialization, 32 Vector ļ¬eld, 24ā25 in local coordinates, 25 left-invariant vector ļ¬eld on a Lie group, 26ā27 horizontal vector ļ¬eld, 380 Vertical tangent space, 366 Volume form, 92 Wave equation in characteristic/null coordinates, 302, 355 in space-time coordinates, 296 Wedge product of vectors, 39 of 1-forms, 41 Weierstrass, Karl, 261 Weierstrass-Enneper representation for a minimal surface in E3 , 266ā267 Wilczynski invariants of a curve in P2 , 206 of a curve in P3 , 216 Wilczynski, Ernest, 206 Wilkens, George, xvi World line, see Minkowski space, world line of a particle Yu, Yunliang, xiii
Selected Published Titles in This Series 178 Jeanne N. Clelland, From Frenet to Cartan: The Method of Moving Frames, 2017 177 Jacques Sauloy, Diļ¬erential Galois Theory through Riemann-Hilbert Correspondence, 2016 176 Adam Clay and Dale Rolfsen, Ordered Groups and Topology, 2016 175 Thomas A. Ivey and Joseph M. Landsberg, Cartan for Beginners: Diļ¬erential Geometry via Moving Frames and Exterior Diļ¬erential Systems, Second Edition, 2016 174 Alexander Kirillov Jr., Quiver Representations and Quiver Varieties, 2016 173 Lan Wen, Diļ¬erentiable Dynamical Systems, 2016 172 Jinho Baik, Percy Deift, and Touļ¬c Suidan, Combinatorics and Random Matrix Theory, 2016 171 170 169 168
Qing Han, Nonlinear Elliptic Equations of the Second Order, 2016 Donald Yau, Colored Operads, 2016 AndrĀ“ as Vasy, Partial Diļ¬erential Equations, 2015 Michael Aizenman and Simone Warzel, Random Operators, 2015
167 166 165 164
John C. Neu, Singular Perturbation in the Physical Sciences, 2015 Alberto Torchinsky, Problems in Real and Functional Analysis, 2015 Joseph J. Rotman, Advanced Modern Algebra: Third Edition, Part 1, 2015 Terence Tao, Expansion in Finite Simple Groups of Lie Type, 2015
163 GĀ“ erald Tenenbaum, Introduction to Analytic and Probabilistic Number Theory, Third Edition, 2015 162 Firas Rassoul-Agha and Timo SeppĀØ alĀØ ainen, A Course on Large Deviations with an Introduction to Gibbs Measures, 2015 161 Diane Maclagan and Bernd Sturmfels, Introduction to Tropical Geometry, 2015 160 Marius Overholt, A Course in Analytic Number Theory, 2014 159 John R. Faulkner, The Role of Nonassociative Algebra in Projective Geometry, 2014 158 Fritz Colonius and Wolfgang Kliemann, Dynamical Systems and Linear Algebra, 2014 157 Gerald Teschl, Mathematical Methods in Quantum Mechanics: With Applications to SchrĀØ odinger Operators, Second Edition, 2014 156 155 154 153
Markus Haase, Functional Analysis, 2014 Emmanuel Kowalski, An Introduction to the Representation Theory of Groups, 2014 Wilhelm Schlag, A Course in Complex Analysis and Riemann Surfaces, 2014 Terence Tao, Hilbertās Fifth Problem and Related Topics, 2014
152 151 150 149
GĀ“ abor SzĀ“ ekelyhidi, An Introduction to Extremal KĀØ ahler Metrics, 2014 Jennifer Schultens, Introduction to 3-Manifolds, 2014 Joe Diestel and Angela Spalsbury, The Joys of Haar Measure, 2013 Daniel W. Stroock, Mathematics of Probability, 2013
148 147 146 145
Luis Barreira and Yakov Pesin, Introduction to Smooth Ergodic Theory, 2013 Xingzhi Zhan, Matrix Theory, 2013 Aaron N. Siegel, Combinatorial Game Theory, 2013 Charles A. Weibel, The K-book, 2013
144 Shun-Jen Cheng and Weiqiang Wang, Dualities and Representations of Lie Superalgebras, 2012 143 Alberto Bressan, Lecture Notes on Functional Analysis, 2013 142 Terence Tao, Higher Order Fourier Analysis, 2012 141 John B. Conway, A Course in Abstract Analysis, 2012
For a complete list of titles in this series, visit the AMS Bookstore at www.ams.org/bookstore/gsmseries/.
Photo by Jenna A. Rice
The method of moving frames originated in the early nineteenth century with the notion of the Frenet frame along a curve in Euclidean space. Later, Darboux expanded this idea to the study of surfaces. The method was brought to its full power in the early twentieth century by Elie Cartan, and its development continues today with the work of Fels, Olver, and others.
This book is an introduction to the method of moving frames as developed by Cartan, at a level suitable for beginning graduate students familiar with the geometry of curves and surfaces in Euclidean space. The main focus is on the use of this method to compute local geometric invariants for curves and surfaces in various 3-dimensional homogeneous spaces, including Euclidean, Minkowski, equiEJĀ½RI ERH TVSNIGXMZI WTEGIW 0EXIV GLETXIVW MRGPYHI ETTPMGEXMSRW XS WIZIVEP GPEWWMGEP problems in differential geometry, as well as an introduction to the nonhomogeneous case via moving frames on Riemannian manifolds. The book is written in a reader-friendly style, building on already familiar concepts from curves and surfaces in Euclidean space. A special feature of this book is the inclusion of detailed guidance regarding the use of the computer algebra system Mapleā¢ to perform many of the computations involved in the exercises. An excellent and unique graduate level exposition of the differential geometry of curves, surfaces and higher-dimensional submanifolds of homogeneous spaces based on the powerful and elegant method of moving frames.The treatment is self-contained and illustrated through a large number of examples and exercises, augmented by Maple code to assist in both concrete calculations and plotting. Highly recommended. āNiky Kamran, McGill University The method of moving frames has seen a tremendous explosion of research activity in recent years, expanding into many new areas of applications, from computer vision to the calculus of variations to geometric partial differential equations to geometric numerical integration schemes to classical invariant theory to integrable systems to infinite-dimensional Lie pseudogroups and beyond. Cartan theory remains a touchstone in modern differential geometry, and Clellandās book provides a fine new introduction that includes both classic and contemporary geometric developments and is supplemented by Maple symbolic software routines that enable the reader to both tackle the exercises and delve further into this fascinating and important field of contemporary mathematics. Recommended for students and researchers wishing to expand their geometric horizons. āPeter Olver, University of Minnesota For additional information and updates on this book, visit
www.ams.org/bookpages/gsm-178
www.ams.org
GSM/178