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

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 Staffilani
Jeff A. Viaclovsky

2010 Mathematics Subject Classification. 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., 1970-
Title: From Frenet to Cartan : the method of moving frames / Jeanne N. Clelland.
Description: Providence, Rhode Island : American Mathematical Society, [2017] | Series: Gradu-
ate studies in mathematics ; volume 178 | Includes bibliographical references and index.
Identifiers: LCCN 2016041073 | ISBN 9781470429522 (alk. paper)
Subjects: LCSH: Frames (Vector analysis) | Vector analysis. | Exterior differential systems. |
Geometry, Differential. | Mathematical physics. | AMS: Topological groups, Lie groups –
Noncompact transformation groups – Homogeneous spaces. msc | Differential geometry –
Classical differential geometry – Curves in Euclidean space. msc | Differential geometry –
Classical differential geometry – Surfaces in Euclidean space. msc | Differential geometry –
Classical differential geometry – Affine differential geometry. msc | Differential geometry –
Classical differential geometry – Projective differential geometry. msc | Differential geometry –
Classical differential geometry – Differential invariants (local theory), geometric objects. msc
| Differential geometry – Local differential geometry – Local submanifolds. msc | Differential
geometry – Local differential geometry – Lorentz metrics, indefinite metrics. msc | Global
analysis, analysis on manifolds – General theory of differentiable manifolds – Differential forms.
msc | Global analysis, analysis on manifolds – General theory of differentiable manifolds –
Exterior differential systems (Cartan theory). msc
Classification: LCC QA433 .C564 2017 | DDC 515/.63–dc23 LC record available at https://lccn.
loc.gov/2016041073

Copying and reprinting. Individual readers of this publication, and nonprofit libraries acting
for them, are permitted to make fair use of the material, such as to copy 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 Center’s 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
first 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.

∞ 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 differential geometry 3

§1.1. Manifolds 3
§1.2. Tensors, indices, and the Einstein summation convention 9
§1.3. Differentiable maps, tangent spaces, and vector fields 15
§1.4. Lie groups and matrix groups 26
§1.5. Vector bundles and principal bundles 32

Chapter 2. Differential forms 35

§2.1. Introduction 35
§2.2. Dual spaces, the cotangent bundle, and tensor products 35
§2.3. 1-forms on Rn 40
§2.4. p-forms on Rn 41
§2.5. The exterior derivative 43
§2.6. Closed and exact forms and the Poincaré lemma 46
§2.7. Differential forms on manifolds 47
§2.8. Pullbacks 49
§2.9. Integration and Stokes’s theorem 53
§2.10. Cartan’s lemma 55

vii
viii Contents

§2.11. The Lie derivative 56

§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
§3.1. Introduction 69
§3.2. Euclidean space 70
§3.3. Orthonormal frames on Euclidean space 75
§3.4. Homogeneous spaces 84
§3.5. Minkowski space 85
§3.6. Equi-affine space 92
§3.7. Projective space 96
§3.8. Maple computations 103
Chapter 4. Curves and surfaces in Euclidean space 107
§4.1. Introduction 107
§4.2. Equivalence of submanifolds of a homogeneous space 108
§4.3. Moving frames for curves in E3 111
§4.4. Compatibility conditions and existence of submanifolds
with prescribed invariants 115
§4.5. Moving frames for surfaces in E3 117
§4.6. Maple computations 134
Chapter 5. Curves and surfaces in Minkowski space 143
§5.1. Introduction 143
§5.2. Moving frames for timelike curves in M1,2 144
§5.3. Moving frames for timelike surfaces in M1,2 149
§5.4. An alternate construction for timelike surfaces 161
§5.5. Maple computations 166
Chapter 6. Curves and surfaces in equi-affine space 171
§6.1. Introduction 171
§6.2. Moving frames for curves in A3 172
§6.3. Moving frames for surfaces in A3 178
§6.4. Maple computations 191
Contents ix

Chapter 7. Curves and surfaces in projective space 203

§7.1. Introduction 203
§7.2. Moving frames for curves in P2 204
§7.3. Moving frames for curves in P3 214
§7.4. Moving frames for surfaces in P3 220
§7.5. Maple computations 235

Part 3. Applications of moving frames

Chapter 8. Minimal surfaces in E3 and A3 251

§8.1. Introduction 251
§8.2. Minimal surfaces in E3 251
§8.3. Minimal surfaces in A3 268
§8.4. Maple computations 280

Chapter 9. Pseudospherical surfaces and Bäcklund’s theorem 287

§9.1. Introduction 287
§9.2. Line congruences 288
§9.3. Bäcklund’s theorem 289
§9.4. Pseudospherical surfaces and the sine-Gordon equation 293
§9.5. The Bäcklund transformation for the sine-Gordon equation 297
§9.6. Maple computations 303

Chapter 10. Two classical theorems 311

§10.1. Doubly ruled surfaces in R3 311
§10.2. The Cauchy-Crofton formula 324
§10.3. Maple computations 329

Part 4. Beyond the flat case: Moving frames on Riemannian

manifolds

Chapter 11. Curves and surfaces in elliptic and hyperbolic spaces 339
§11.1. Introduction 339
§11.2. The homogeneous spaces Sn and Hn 340
§11.3. A more intrinsic view of Sn and Hn 345
§11.4. Moving frames for curves in S3 and H3 348
§11.5. Moving frames for surfaces in S3 and H3 351
§11.6. Maple computations 357
x Contents

Chapter 12. The nonhomogeneous case: Moving frames on Riemannian

manifolds 361
§12.1. Introduction 361
§12.2. Orthonormal frames and connections on Riemannian
manifolds 362
§12.3. The Levi-Civita connection 370
§12.4. The structure equations 373
§12.5. Moving frames for curves in 3-dimensional Riemannian
manifolds 379
§12.6. Moving frames for surfaces in 3-dimensional Riemannian
manifolds 381
§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 curve—
curvature and torsion—appear 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 construct-
ing moving frames on surfaces in Euclidean space [Dar72a], [Dar72b],
[Dar72c], [Dar72d]. In the early twentieth century, Élie Cartan general-
ized the notion of moving frames to other geometries (for example, affine and
projective geometry) and developed the theory of moving frames extensively.
A very nice introduction to Cartan’s ideas may be found in Guggenheimer’s
text [Gug77].
More recently, Fels and Olver [FO98], [FO99] have introduced the notion
of an “equivariant moving frame”, which expands on Cartan’s 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 Cartan’s 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 back-
ground, including courses in linear algebra, abstract algebra, real analysis,
and topology, as well as a course on the differential geometry of curves and
surfaces. (An appropriate differential geometry course might be based on a
text such as [dC76], [O’N06], or [Opr07].) There are occasional references
to additional topics such as differential equations, but these are less crucial.
The first two chapters contain background material that might typically
be taught in a graduate differential geometry course; Chapter 1 contains
general material from differential geometry, while Chapter 2 focuses more
specifically on differential 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 3–7 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 4–7 show how to apply the
method of moving frames to compute local geometric invariants for curves
and surfaces in 3-dimensional Euclidean, Minkowski, affine, and projective
spaces. These chapters should be read in order (with the possible exception
of Chapter 5), as they build on each other.
Chapters 8–10 show how the method of moving frames may be applied to
several classical problems in differential geometry. The first 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-flat 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 don’t 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 computa-
tions 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 pack-
age Cartan, which was written by myself and Yunliang Yu of Duke Univer-
sity. 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 Cartan’s 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 different syntax, and no attempt will be made
here to translate—but interested readers are encouraged to do so!
Acknowledgments

First and foremost, my deepest thanks go to Robert Bryant—my teacher,

mentor, and friend—for inviting me to teach alongside him at the Math-
ematical 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 unflagging support in more ways than I can
count over the years.
Thanks also to Edward Dunne and Sergei Gelfand at the American Mathe-
matical 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 significant
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 ver-
sions of the manuscript over the last several years: Brian Carlsen, Michael
Schmidt, Edward Estrada, Molly May, Jonah Miller, Sean Peneyra, Duff
Baker-Jarvis, Akaxia Cruz, Rachel Helm, Peter Joeris, Joshua Karpel, An-
drew Jensen, and Michael Mahoney. These independent study courses—and
the research projects that followed—have been, hands down, the most re-
warding 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
MSRI’s 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
 Chapter 1

Assorted notions from

differential geometry

This chapter contains some useful background material from differential ge-
ometry. Much of this material would typically be covered in courses on
multivariable analysis, algebra, and graduate-level differential geometry and
linear algebra. This chapter is not intended to be a comprehensive intro-
duction 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 defined 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 definition, let’s
remind ourselves of some of the fundamental properties of regular surfaces.

1.1.1. Regular surfaces. First, recall the definition of a regular surface

in R3 (see, e.g., [dC76]):
Definition 1.1. A subset Σ ⊂ R3 is a regular surface if for each point q ∈ Σ,
there exists a neighborhood V ⊂ R3 of q and a map x : U → V ∩ Σ from an
open set U ⊂ R2 onto V ∩ Σ with the following properties:

(1) x is differentiable.
(2) x is a homeomorphism.
(3) For each u ∈ U , the differential dxu : R2 → R2 is one-to-one.

3
4 1. Assorted notions from differential geometry

The mapping x is called a parametrization of Σ, or a system of local coordi-

nates on Σ, in a neighborhood of q.

Remark 1.2. Throughout this book, the word “differentiable” will be taken
to mean “infinitely differentiable” (i.e., C ∞ ). We will also use the word
“smooth” as a synonym.

Any regular surface Σ ⊂ R3 can be covered by a family of parametrizations

xi : Ui → Σ, where each Ui is an open set in R2 and each xi is an injective,
differentiable map from Ui to R3 . The image xi (Ui ) of each parametrization
is an open set in Σ (in the subspace topology inherited from R3 ); moreover,
if the images of two parametrizations xi : Ui → Σ and xj : Uj → Σ have
nonempty intersection V ⊂ Σ, then the composite maps

(1.1) x−1 −1 −1
j ◦ xi : xi (V ) → xj (V ), x−1 −1 −1
i ◦ xj : xj (V ) → xi (V )

are differentiable maps between open sets in R2 . (See Figure 1.1.)

.................................................................................
................................................ .........................
......................... ....................
.................... ..................
.................... .............................................. .......................................................
... ....... ....
....
.................. .......... ..
... ..........
. .
........... ..... ........ ...
..
... ..
. ..
.
..... . .
..... ...... Σ .
..
..
... .
.. .
... .
... .
... ...
x (U ) ... x (U ) ..
.. .. i .. V ...
.. j ...
... ..
..
..
.. .... ..... . . ..
.. ... ... .... ...
. ...
.. ... ... .
.. .... .... ... ...
. ...
.. .... .... .... .... ...
.. ..... ...... .... ..... ...
. ........ ......... ............. .......
.... ........... ....
.................................................. ...................................................... ......... ..
..
.. ..
.. ..
..
.. ...............................................................................................................
..
.
.... .................................................................. . .......................... ...
.
................. ...............

 @
I
@ xj
xi
@
@
v @ v̄
6 6
.............................................. ............................ ..... ...............
.............. .... ............. .........
......... .............. ........... .......
....... ...... ......... ....... .......... .....
..... .... ..... .... ....
x−1
. .... .
j ◦ xi
. .... . .... ..
... U . ... ... U j ......
-
. i . . ...
... ... −1 ... ... −1 .... ...
.... x.... (V ) ... x
.... j (V )
... ... i -u ...
. ...
...
.. - ū...
.
..
..
 ..
.. .. ..
... ... .. ... ... ..
... ... .... ... .
..
...
.
x−1
... ... . ... ... ...
i ◦ xj
.
...
.... ....
...
. ...
.... .... ....
..... ....
..... ........ ..... ... ...
........ .............. ................... .
.........
.......... .......... .
.................... ............................. .................... .......................... ....
........... ...........

Figure 1.1. Overlapping parametrizations

When Σ is a regular surface in R3 , the differentiability of the maps (1.1)

is a theorem. But regular surfaces can also be defined as abstract, intrin-
1.1. Manifolds 5

sic objects, not living in any particular Euclidean space. In this case, the
differentiability of (1.1) becomes part of the definition of a regular surface:

Definition 1.3. A regular surface is a set Σ and a family of injective map-

pings xi : Ui → Σ, where each Ui is an open set in R2 , with the following
properties:

(1) xi (Ui ) = Σ.
i

(2) For each pair i, j with V = xi (Ui ) ∩ xj (Uj ) = ∅, the sets x−1
i (V )
−1 −1
and xj (V ) are open sets in R , and the mappings xj ◦ xi and
2

x−1
i ◦ xj are differentiable.

The mappings xi are called parametrizations of Σ, or systems of local coor-

dinates on Σ.

Incorporating condition (2) as an axiom allows us to define objects such as

differentiable functions on surfaces and differentiable maps between surfaces
in a way that is independent of a choice of parametrization. All the usual
notions of differential calculus on R2 can then be extended to analogous
notions on regular surfaces.
For an abstract surface, the maps xi are rarely defined explicitly. (This is
very different from how we view surfaces in R3 , where the parametrizations
are often used to define the surface!) Rather, we think of an abstract surface
as a collection of open sets Ui in R2 , “glued together” via the transition maps
x−1
j ◦ xi . It is these transition maps between open sets in R that may (or
2

may not) be defined 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 , (ū, v̄) are local coordinates on Uj , and
the set V = xi (Ui ) ∩ xj (Uj ) ⊂ Σ is nonempty, then the transition map

x−1 −1 −1
j ◦ xi : xi (V ) → xj (V )

is a local coordinate transformation of the form

x−1
j ◦ xi (u, v) = (ū(u, v), v̄(u, v)).

(See Figure 1.1.)

1.1.2. Manifolds: from 2 to n. A manifold (or, more precisely, a differ-

entiable manifold) of dimension n ≥ 1 is simply what we get by replacing
6 1. Assorted notions from differential geometry

the number 2 in Definition 1.3 with n:

Definition 1.4. A differentiable manifold of dimension n ≥ 1 is a set M
and a family of injective mappings xi : Ui → 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 ) ∩ xj (Uj ) = ∅, the sets x−1
i (V ) and
xj (V ) are open sets in R , and the mappings xj ◦xi and x−1
−1 n −1
i ◦xj
are differentiable. 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 → R3
as ⎡ ⎤
x(u, v)
x(u, v) = ⎣y(u, v)⎦ .
z(u, v)
When we graduate to manifolds of arbitrary dimension, we need to use
variables with indices so that we don’t 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 ⊂ Rn .

Just as for regular surfaces, a manifold M of dimension n can be a subset

of some Euclidean space Rk with k ≥ 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 defined by explicit parametrizations xi : Ui → Rk , where the Ui are open
sets in Rn . When M is an intrinsic manifold, the maps xi are generally not
defined explicitly; we simply think of M as a collection of open sets Ui ⊂ Rn
that are glued together via the transition maps x−1j ◦ xi . As for surfaces,
these transition maps are local coordinate transformations of the form
(1.2) x−1
j ◦ xi (u , . . . , u ) = (ū (u , . . . , u ), . . . , ū (u , . . . , u )).
1 n 1 1 n n 1 n

1.1.3. Examples.
Example 1.6 (The unit sphere). Let Sn ⊂ Rn+1 be the set
Sn = {t[x1 , . . . , xn+1 ] ∈ Rn+1 | (x1 )2 + · · · + (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 specified. 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

⎡ ⎤
x1
⎢ .. ⎥
⎣ . ⎦. 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 ⊂ 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 ] ∈ Sn | xi > 0},

Vi− = {t[x1 , . . . , xn+1 ] ∈ Sn | xi < 0}.

Let
U = {t[u1 , . . . , un ] ∈ Rn | (u1 )2 + · · · + (un )2 < 1};
i.e., U is the open unit ball in Rn . Define maps

i : U → Vi ,
x+ +
x−
i : U → Vi
−

by
t
x+ 1 n
i (u , . . . , u ) = u1 , . . . , ui−1 , 1 − (u1 )2 − · · · − (un )2 , ui , . . . , un ,

t
x− 1 n
i (u , . . . , u ) = u1 , . . . , ui−1 , − 1 − (u1 )2 − · · · − (un )2 , ui , . . . , un .

−
(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.

i (U ) ∩ xj (U ) in S and the open

(b) For i = j, identify the open set Vij+ = x+ + n
+ −1 + + −1 +
sets (xj ) (Vij ) and (xi ) (Vij ) in U , and compute the local coordinate
transformation (x+ −1 ◦ x+ between the latter two open sets.
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})/ ∼,

where ∼ represents the equivalence relation defined by the condition that

two points x, y ∈ Rn+1 \ {0} satisfy x ∼ y if and only if x and y lie on
8 1. Assorted notions from differential 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 λ, we have
[x0 , . . . , xn ] ∼ t[λx0 , . . . , λxn ].
t

The equivalence class of the point x = t[x0 , . . . , xn ] is denoted by

[x] = [x0 : · · · : 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, −x}. So, an
alternative (but equivalent!) way to think of Pn is as the manifold obtained
from Sn by identifying every point x ∈ Sn with its polar opposite −x.
Even better, we can identify Pn with the upper half of Sn (which is topo-
logically an n-dimensional disk), with its boundary glued together so as to
identify opposite points on the boundary. Note that gluing the boundary—
which is a copy of Sn−1 —together in this fashion produces a copy of Pn−1 .
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 Pn−1 attached along its
boundary. (See Figure 1.2 for an idea of how this looks for P1 and P2 .)

P2
..................................
........... .......
....... .....
S1 ..... ....
...
 P1 ..
.. ...
...
............. .
..... .......
...
........... ....
...r .r -
.
..
.... .... r...
......
..... .
.... ..
... .
.... .
..
. .... ... .. ... ..... .. ... ...r
. .. ..
... ..
..... .... ..
...
.
r
........
.........r..... ....... ..
r
............ . ........

. .
..

... .
...........................................................
. .
O 
... . . .
.... .....
...
.. ....
....
.......
K
............ ................. .................
.....................................

Figure 1.2. Construction of P1 and P2

Unlike the unit sphere Sn , Pn is not obviously defined 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 : · · · : xn ] ∈ Pn | xi = 0}.
Let U = Rn , with coordinates u = (u1 , . . . , un ). Define maps
[xi ] : U → Pn
by
[xi (u1 , . . . , un )] = [u1 : · · · : ui : 1 : ui+1 : · · · : 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 )] ∩ [xj (U )] in Pn and the open sets
[xj ]−1 (Vij ) and [xi ]−1 (Vij ) in U , and compute the local coordinate transfor-
mation [xj ]−1 ◦ [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 wondering—why on earth would we do that? So,
before we go any farther, let’s discuss how (and why!) indices are used
throughout this book. Some objects will be indexed with superscripts (“up-
per indices” or, less formally, “up indices”), some by subscripts (“lower
indices” or “down indices”), and some by a combination of both. Most of
these objects will be tensors or tensor fields. Without getting too specific,
a tensor is just an element of a certain type of vector space, and a tensor
field on a manifold is defined by assigning a tensor to each point of the
manifold. (A precise definition will be given in Chapter 2.) For example,
tensors include objects such as vectors and linear transformations, while
tensor fields include objects such as vector fields, metrics, and differential
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 ∈ V can be expressed uniquely as

v = a1 e1 + · · · + an en

for some real numbers a1 , . . . , an ∈ R.

Now, what happens if we express v in terms of a different basis (ē1 , . . . , ēn )?
Suppose that the new basis can be written in terms of the old basis as
n
ēi = rik ek , i = 1, . . . , n,
k=1

or, in more compact matrix notation,

 
ē1 . . . ēn = e1 . . . en R,
10 1. Assorted notions from differential geometry

where ⎡ ⎤
r11 . . . rn1
⎢ .. .. ⎥ .
R=⎣. .⎦
r1n . . . rnn
Then we have  
e1 . . . en = ē1 . . . ēn R−1 ,
and so
v = a1 e1 + · · · + an en
⎡ 1⎤
a
⎢ . ⎥
= e1 . . . en ⎣ .. ⎦
an
⎤⎡
a1
 ⎢ ⎥
= ē1 . . . ēn R−1 ⎣ ... ⎦
an
= ā1 ē1 + · · · + ān ēn ,
where ⎡ ⎤ ⎡ 1⎤
ā1 a
⎢ .. ⎥ −1 ⎢ .. ⎥
⎣ . ⎦ = R ⎣ . ⎦.
ān 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. “Rank 1” means that,

when the vector is expressed in terms of a given basis, the compo-
nents—in this case, (a1 , . . . , an )—each 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-defined indepen-
dently of a choice of basis: When the basis vectors (e1 , . . . , en ) are
changed, the coefficients (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 R−1 . This is not an accident! In
general, the placement—up or down—of 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 → W be a linear transformation. In terms of the given bases for V
and W , T can be represented as an n × m matrix
⎡ 1 ⎤
c1 . . . c1m
⎢ .. ⎥ .
AT = ⎣ ... . ⎦
n
c1 . . . cmn

Let’s think carefully about what this means. We usually write a vector
v ∈ V as a column vector
⎡ 1⎤
a
⎢ .. ⎥
a=⎣ . ⎦
am
(by which we really mean that v = [e1 . . . em ] · a), and we write w = T (v)
as the matrix product
⎡ 1⎤ ⎡ 1 ⎤⎡ 1⎤
b c1 . . . c1m a
⎢ .. ⎥ ⎢ .. .. ⎦ ⎣ ... ⎥
. ⎥ ⎢
b=⎣.⎦=⎣. ⎦
bn n
c1 . . . cm n am

(by which we really mean that w = [f1 . . . fn ] · b). The use of different
notations here for the vectors v and w, which are well-defined independently
of a choice of basis, and their expressions a and b in terms of specific 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

 m

w = T (v) = T ai ei
i=1
n
 m

= cji ai fj .
j=1 i=1

In more compact matrix notation,

⎛ ⎡ 1 ⎤⎞ ⎡ ⎤
a a1
⎜  ⎢ . ⎥⎟  ⎢ .. ⎥
(1.3) T ⎝ e1 . . . em ⎣ .. ⎦⎠ = f1 . . . fn AT ⎣ . ⎦.
am am
12 1. Assorted notions from differential geometry

Again, we ask what happens if we change bases for V and W . Suppose that
we set
 
(1.4) ē1 . . . ēm = e1 . . . em R,
 
f̄1 . . . f̄n = f1 . . . fn S,

where R is an invertible m × m matrix and S is an invertible n × n matrix.

We want to find the matrix AT that satisfies
⎛ ⎡ 1 ⎤⎞ ⎡ 1⎤
ā ā
⎜  ⎢ . ⎥⎟  ⎢ .. ⎥
T ⎝ ē1 . . . ēm ⎣ . ⎦⎠ = f̄1 . . . f̄n AT ⎣ . ⎦ .
.
ām ām
Using (1.3) and (1.4), we can compute:
⎛ ⎡ 1 ⎤⎞ ⎛ ⎡ 1 ⎤⎞
ā ā
⎜  ⎢ . ⎥⎟ ⎜  ⎢ . ⎥⎟
T ⎝ ē1 . . . ēm ⎣ . ⎦⎠ = T ⎝ e1 . . . em R ⎣ .. ⎦⎠
.
ām ām
⎡ 1⎤
ā
 ⎢ .. ⎥
= f1 . . . fn AT R ⎣ . ⎦
ām
⎡ 1⎤
ā
 −1 ⎢ .. ⎥
= f̄1 . . . f̄n (S AT R) ⎣ . ⎦ .
ām
So, in terms of the new bases, the new matrix representation for T is

AT = S −1 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 specific bases
for V and W have an upper index i with range 1 ≤ i ≤ n, corre-
sponding to the vector space W , and a lower index j with range
1 ≤ j ≤ 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 −1 and on the right by R. This illustrates
the general phenomenon that, under a change of basis:
• 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

• Up indices indicate that components will transform by left

multiplication by the inverse of the matrix for the basis trans-
formation.
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 inter-
ested in tensor fields, which consist of an underlying manifold with a tensor
defined at each point. It may sound complicated, but you already know
several examples—e.g., vector fields and metrics on surfaces. The only real
differences 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 man-
ifold.

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
 ∂ with local  coordinates (x1 , . . . , xn ), the partial de-
rivative operators ∂x1 , . . . , ∂x∂n will often be used as a local basis for the
tangent space at each point (the reason for this will be explained in §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 defined by specifying for

each point q ∈ M a symmetric, positive definite bilinear form gq on the
tangent space Tq M . (Tangent spaces will be defined in §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 × Tq M → R

such that for all v, w ∈ Tq M , we have gq (v, w) = gq (w, v) and gq (v, v) ≥ 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 define the components of gq with respect
to this basis to be the real numbers

{gij (q) = gq (ei , ej ) | 1 ≤ i, j ≤ n}.

14 1. Assorted notions from differential 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 ei , w= bj ej
i=1 j=1

in Tq M , we have
n n
gq (v, w) = gij (q)ai bj .
i=1 j=1
For instance, when M is a regular surface Σ ⊂ R3 with a local parametriza-
tion x : U → Σ, 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 · w.
Then we have
g11 = E = xu · xu , g12 = g21 = F = xu · xv , g22 = G = xv · xv .
A metric g is often represented by the symmetric matrix
⎡ ⎤
g11 . . . g1n
⎢ .. ⎥ .
Ag = ⎣ ... . ⎦
gn1 . . . gnn

Suppose that we make a change of basis

 
ē1 . . . ēn = e1 . . . en R
for Tq M . Compute the components (ḡij (q)) of gq with respect to the new
basis (ē1 , . . . , ēn ). How does the matrix Ag transform? Compare the trans-
formation rule for Ag to that given in Example 1.12 for the matrix repre-
sentation 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 fields, there are a lot of sums involved: A
vector is expressed as
n
v= ai ei ;
i=1
n n
i
a metric g acting on a pair of vectors v = a ei , w = bj ej is expressed
i=1 j=1
as
n n
g(v, w) = gij ai bj ,
i=1 j=1
1.3. Differentiable maps, tangent spaces, and vector fields 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 ap-
pears 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 it’s
actually extremely handy. For instance, it makes the multivariable chain
rule look exactly like the single-variable version, where you can “cancel” the
intermediate variables:
∂f ∂f ∂y j
= .
∂xi ∂y j ∂xi
It also gives a quick method for error checking, much like dimensional analy-
sis in chemistry or physics: Once repeated indices are “canceled”, the indices
on both sides of an equation should match.

1.3. Differentiable maps, tangent spaces, and vector fields

Let U ⊂ Rm be an open set, and consider a function F : U → Rn . (Such

a function is often called a mapping, or simply a map, from U to Rn .) We
can write ⎡ 1 1 ⎤
f (x , . . . , xm )
⎢ .. ⎥
F (x1 , . . . , xm ) = ⎣ . ⎦
n 1 m
f (x , . . . , x )
for some real-valued functions f 1 , . . . , f n : U → R. We say that F is contin-
uous if each of the functions f 1 , . . . , f n is continuous, and F is differentiable
if each of the functions f 1 , . . . , f n is differentiable.
Now, if a function is differentiable, 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 ⊂ 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 α : I → Rn defined by
⎡ 1 ⎤
y (t)
⎢ .. ⎥
α(t) = ⎣ . ⎦ ,
y n (t)
16 1. Assorted notions from differential geometry

the derivative of α at any point t ∈ I is often simply defined to be the

column vector
⎡ dy1 ⎤
dt (t)
 ⎢ .. ⎥
(1.5) α (t) = ⎣ . ⎦ ,
dy n
dt (t)

also called the tangent vector to α at the point α(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 definition. A more accurate
definition would be
⎛⎡ 1 ⎤ ⎡ dy1 ⎤⎞
y (t) dt (t)
 ⎜⎢ .. ⎥ ⎢ .. ⎥⎟
(1.6) α (t) = ⎝⎣ . ⎦ , ⎣ . ⎦⎠ ,
y n (t) dy n
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 specific point α(t) in Rn .
Given a point y = t[y 1 , . . . , y n ] ∈ 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 = {α (0) | α : I → Rn is a smooth curve with α(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
“right” one. Specifically, this isomorphism is defined by (1.5), which identi-
1 dy n
fies the tangent vector α (0) ∈ Ty Rn with the vector t dy
dt (0), . . . , dt (0) ∈
Rn .
Now let m be arbitrary, and let F : U ⊂ Rm → Rn be a differentiable map.
Fix a point x ∈ U . The derivative or differential of F (denoted by dF or
F∗ , depending on the context) at x is a linear map from Tx Rm to TF (x) Rn
defined as follows: Given a tangent vector v ∈ Tx Rm , let α : I → Rm be a
smooth curve with α(0) = x and α (0) = v. Then

dFx (v) = (F ◦ α) (0) ∈ TF (x) Rn .

(See Figure 1.3.)

1.3. Differentiable maps, tangent spaces, and vector fields 17

v=α (0) 
....................
............. F◦α
.........
r..
.......
..... rP dF (v)=(F ◦ α) (0)
............................................................
............
PP
....
F .........
....
....
....
x=α(0) - F (x) q
P
........
......
.....
x

.... ....
....
... ...
α .. ...
.. .
..
....
..
..

Figure 1.3. Construction of dFx (v)

While it may not be obvious, it is true that:

(1) dFx (v) is well-defined, independent of the choice of the curve α.

(2) dFx : Tx Rm → 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 ⊂ Rm → Rn be a
differentiable map.
(a) Let x ∈ U , and let v ∈ Tx Rm be a tangent vector. Let α : I → Rm ,
β : I → Rm be two curves with
α(0) = β(0) = x, α (0) = β  (0) = v.
Show by direct computation that
(F ◦ α) (0) = (F ◦ β) (0) ∈ TF (x) Rn .
This shows that the differential dFx (v) is well-defined.
(b) Let x ∈ U , and let v, w ∈ 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 α : I → Rm , β : I → Rm
be two curves with
α(0) = β(0) = x, α (0) = v, β  (0) = w.
Write ⎤ ⎡ ⎤ ⎡
x11 (t) x12 (t)
⎢ ⎥ ⎢ ⎥
α(t) = ⎣ ... ⎦ , β(t) = ⎣ ... ⎦ .
xm
1 (t) xm
2 (t)
18 1. Assorted notions from differential geometry

Let γ be the curve

⎡ 1
 1 ⎤
2 x1 (2t) + x12 (2t)
⎢ .. ⎥
γ(t) = ⎣ . ⎦.
1
2 (xm m
1 (2t) + x2 (2t))

Then γ satisfies γ(0) = x and γ  (0) = v + w.)

(c) Let
⎡ ⎤
y 1 (x1 , . . . , xm )
⎢ .. ⎥
F (x1 , . . . , xm ) = ⎣ . ⎦.
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
∂y
J= .
∂xj
This means that if v has canonical representation a = t[a1 , . . . , am ], then
the canonical representation for dFx (v) is
⎡ ∂y1 j ⎤
∂xj
a
⎢ .. ⎥
Ja = ⎣ . ⎦ .
∂y n j
∂xj
a
(Hint: What is the simplest curve α(t) that you can think of with α(0) = x
and α (0) = v?)

Now let’s get a little more abstract and define the analogous notions for
manifolds.
Definition 1.17. Suppose that M and N are m- and n-dimensional man-
ifolds, respectively. A map F : M → N is called differentiable if for any
parametrizations
x : U ⊂ Rm → M, y : V ⊂ Rn → N
on M and N such that F (x(U )) ⊂ y(V ), the composite map
y−1 ◦ F ◦ x : U → V
is differentiable. (Definition 1.4 implies that this condition is independent
of the choice of parametrizations x and y.)
Remark 1.18. Keeping track of all these different maps can be challenging,
and so they are often described using diagrams. For example, the maps
1.3. Differentiable maps, tangent spaces, and vector fields 19

above might be depicted as follows:

M
F - N

x 6 6y
y−1 ◦F ◦x
Rm ⊇ U - V ⊂ Rn .

This is called a commutative diagram. (Sometimes we simply say that “the

diagram commutes”.) 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. Specifically, if u ∈ U , then
(F ◦ x)(u) = (y ◦ (y−1 ◦ F ◦ x))(u) ∈ N.
Of course, the commutativity of the diagram is obvious in this case, but
sometimes it can indicate something more substantial.

In order to define the derivative of such a map F , we first need to define

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 ∈ 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 ∈ Rm and v ∈ Tx Rm .
If α : I → Rm is any smooth curve in Rm with α(0) = x, α (0) = v and if
f : Rm → R is any differentiable function, then v acts on f as follows:

d 
v[f ] =  f (α(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 definition for tangent vectors.
Definition 1.19. Let α : I → M be a smooth curve in M . The tangent
vector α (t0 ) to α at the point α(t0 ) ∈ M is the differential operator
α (t0 ) : C ∞ (M ) → R
(where C ∞ (M ) denotes the space of all smooth, real-valued functions on
M ) defined by

 d 
α (t0 )[f ] =  f (α(t))
dt t=t0
20 1. Assorted notions from differential geometry

for f ∈ C ∞ (M ). The tangent space Tq M to M at a point q ∈ M is the set

of all tangent vectors to all curves in M passing through q; i.e.,
Tq M = {α (0) | α : I → M is a smooth curve with α(0) = q}.
Example 1.20. Let Σ be a regular surface in R3 and x : U → Σ a
parametrization of Σ defined on some open set U ⊂ R2 , with local coordi-
nates (u, v) on U . The tangent vectors (xu , xv ) form a basis for the tangent
plane Tq Σ at each point q ∈ x(U ). In order to see how these vectors act as
differential operators, let f : Σ → R be a differentiable, real-valued function
on Σ. Let q = x(u0 , v0 ) ∈ x(U ), and let α(t) be the curve
α(t) = x(u0 + t, v0 )
in Σ. Then α (0) = xu , and so

d 
xu [f ] =  f (α(t))
dt t=0

d 
=  f (x(u0 + t, v0 ))
dt t=0

∂(f ◦ x) 
= .
∂u (u,v)=(u0 ,v0 )


Similarly, xv [f ] = ∂(f∂v◦x)  . So, if we think of the parametrization
(u,v)=(u0 ,v0 )
x as identifying its image x(U ) ⊂ Σ with the open set U ⊂ R2 and identifying
a function f ∈ C ∞ (M ) withthe composition f ◦x ∈ C ∞ (U ), then the partial
∂ ∂
derivative operators ∂u , ∂v form a natural basis for the tangent space Tq Σ
at each point q ∈ x(U ).
*Exercise 1.21. Let M be a manifold with local coordinates x = (x1 , . . .,
xm ) on some open set V ⊂ M . (The parametrization x : U ⊂ Rm → V ⊂ M
is implicit in this statement.) Let q ∈ 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 differential operators ∂x∂ 1 , . . . , ∂x∂m as a basis, almost
exactly as in Example 1.20
(a) Let v = α (0) ∈ Tq M , where
α(t) = x(x1 (t), . . . , xm (t)).
Show that for any f ∈ C ∞ (M ), we have

i  ∂(f ◦ x) 
v[f ] = (x ) (0) .
∂xi (x1 ,...,xm )=(x1 ,...,xm )
0 0

Therefore, any tangent vector v ∈ Tq M may be regarded

 as a linear combi-
∂ ∂
nation of the differential operators ∂x1 , . . . , ∂xm .
1.3. Differentiable maps, tangent spaces, and vector fields 21

(b) Show that if ci ∂x∂ 1 m ∈ R, then

i = 0 for some real numbers c , . . . , c
 
c1 = · · · = cm = 0. This shows that the differential operators ∂x∂ 1 , . . . , ∂x∂m
∂
are linearly independent. (Hint: The statement “ci ∂x i = 0” means that
◦x)
ci ∂(f
∂xi
= 0 for every differentiable function f : M → R. Try to concoct
real-valued functions fi on V (which can then be extended to all of M via
standard techniques) with the property that

∂(fi ◦ x) 1, i = j, 
=
∂x j
0, i = j.

*Exercise 1.22. (a) Suppose that we have a local coordinate transforma-

tion of the form

xi = xi (x̄1 , . . . , x̄m ), i = 1, . . . , m,

on M . Use the definition of tangent vectors and the multivariable chain rule
to show that
 
∂
∂ x̄1 · · · ∂
∂ x̄m = ∂
∂x 1 · · · ∂
∂xm J,

where J is the Jacobian matrix of the coordinate transformation, i.e., the

i
matrix whose (i, j)th entry is ∂∂xx̄j .
(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.
q∈M

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 Σ ⊂ R3 be a regular

surface. At each point q ∈ Σ, the tangent plane Tq Σ at q consists of all
tangent vectors to Σ at q. The tangent bundle T Σ of Σ is simply the union
of all these tangent planes:

TΣ = Tq Σ.
q∈Σ
22 1. Assorted notions from differential geometry

Note that the tangent bundle is characterized by

(1) the base space Σ and

(2) for each point q ∈ Σ, a vector space Tq Σ associated to q, called
the fiber at q.

Moreover, the vector spaces associated to each point all have the same
dimension—2, in this case. T Σ 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 ⊂ Σ, it can be
canonicallyextended to a system of local coordinates (x, y) = (x1 , x2 , y 1 , y 2 )
on T V = q∈V Tq V ⊂ T Σ by associating to any tangent vector v ∈ Tq V
its unique representation as
∂ ∂
(1.7) v = y1 1 + y2 2 .
∂x ∂x
*Exercise 1.24. Let T Σ be the tangent bundle of a regular surface, and let
x = (x1 , x2 ) and x̄ = (x̄1 , x̄2 ) be two overlapping systems of local coordinates
on Σ, related by a coordinate transformation of the form
(1.8) x1 = x1 (x̄1 , x̄2 ), x2 = x2 (x̄1 , x̄2 ).
Show that the local coordinate transformation between the canonically asso-
ciated local coordinates (x, y) = (x1 , x2 , y 1 , y 2 ) and (x̄, ȳ) = (x̄1 , x̄2 , ȳ 1 , ȳ 2 )
on T Σ is given by equations (1.8) together with the equations
   
y1 ȳ 1
= J ,
y2 ȳ 2
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 ⊂ Rm → M is a
parametrization of M , then the associated canonical parametrization of T M
is the map (x, y) : U × Rm → T M given by
∂
(x, y)(x1 , . . . , xm , y 1 , . . . , y m ) = y i i ∈ Tx(x1 ,...,xm ) M.
∂x
In other words, for each point q ∈ x(U ), the tangent  space Tq M isidentified
with the vector space Rm by identifying the basis ∂x∂ 1 , . . ., ∂x∂m 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
(x, y) and (x̄, ȳ) is differentiable and, in fact, has the form
∂xi j
xi = xi (x̄1 , . . . , x̄m ), yi = ȳ ;
∂ x̄j
1.3. Differentiable maps, tangent spaces, and vector fields 23

therefore, T M is a smooth manifold of dimension 2m. T M also comes

equipped with the base-point projection map π : T M → M defined by the
condition that for any v ∈ Tq M , π(v) = q.
Here’s the good news: Now that we know how to define tangent spaces for
manifolds, the definition of the derivative of a differentiable map F : M → N
is exactly the same as before.

Definition 1.25. Let M be a manifold of dimension m, N a manifold of

dimension n, and let F : M → N be a differentiable map. Fix a point
q ∈ M . The derivative or differential of F at q is the linear map dFq :
Tq M → TF (q) N defined as follows: Given a tangent vector v ∈ Tq M , let
α : I → M be a smooth curve with α(0) = q and α (0) = v. Then

dFq (v) = (F ◦ α) (0) ∈ 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
 ∂ ∂
 ∂ ∂
∂x1
, . . . , ∂xm and ∂y1 , . . . , ∂yn are used for the tangent spaces Tq M and
TF (q) N , respectively. The main difference is that the isomorphisms between
Tq M, TF (q) N and Rm , Rn defined 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 → N be a differentiable map. Let

x : U ⊂ Rm → M, y : V ⊂ Rn → N

be parametrizations, and let

H = y−1 ◦ F ◦ x : U → V.

We can write

H(x1 , . . . , xm ) = (y 1 (x1 , . . . , xm ), . . . , y n (x1 , . . . , xm )).

(a) Suppose that q ∈ x(U ), and let v = aj ∂x∂ j ∈ Tq M . Show that

" #
∂y i ∂
dFq (v) = aj
∈ TF (q) N ;
∂xj ∂y i
24 1. Assorted notions from differential geometry

in other words,
⎛ ⎡ ⎤⎞ ⎡ ⎤
a1 a1
⎜  ⎢ . ⎥⎟ ⎢ ⎥
dFq ⎝ ∂
∂x1
··· ∂
∂xm ⎣ . ⎦⎠ =
. ∂
∂y 1
··· ∂
∂y n J ⎣ ... ⎦ ,
am am
where J is the n × m Jacobian matrix of H.
BONUS: What happens to this formula if you perform local coordinate
transformations on M and N ?
(b) The differential of F can be extended to a map
dF : T M → T N
in the obvious way: For v ∈ Tq M , define
dF (v) = dFq (v).
Show that dF is a differentiable map from T M to T N .

*Exercise 1.27. Let M , N , and P be manifolds, and suppose that F :

M → N and G : N → P are differentiable maps.
(a) Prove the chain rule: For q ∈ M ,
d(G ◦ F )q = dGF (q) ◦ dFq .

(b) Give an explicit interpretation of the chain rule in terms of local coor-
dinates 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 fields on manifolds. Intuitively, a

vector field 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 field to be differentiable. Fortunately, we now have all the tools that
we need in order to make this idea precise:

Definition 1.28. A (smooth) vector field v on a manifold M is a differen-

tiable map
v : M → TM
with the property that for any q ∈ M , we have v(q) ∈ Tq M. (Equivalently,
the composition π ◦ v : M → M is the identity map.)
1.3. Differentiable maps, tangent spaces, and vector fields 25

A vector field v on a manifold M may be regarded as a first-order, linear,

homogeneous differential operator v : C ∞ (M ) → C ∞ (M ) defined as follows:
For f ∈ C ∞ (M ), the function v[f ] : M → R is given by
v[f ](q) = v(q)[f ].
This means that the value of v[f ] at q ∈ M is just the value of the tangent
vector v(q) ∈ Tq M acting on f at q.
In terms of local coordinates (x1 , . . . , xm ) on M , a vector field v is generally
expressed as
∂
(1.9) v(x1 , . . . , xm ) = ai (x1 , . . . , xm ) i .
∂x
Then for f ∈ C ∞ (M ), the function v[f ] is expressed in terms of these local
coordinates as
∂f
(1.10) v[f ](x1 , . . . , xm ) = ai (x1 , . . . , xm ) i (x1 , . . . , xm ).
∂x
Conversely, any (smooth) first-order, linear, homogeneous differential oper-
ator L : C ∞ (M ) → C ∞ (M ) given in local coordinates by (1.10) defines a
vector field v on M described in local coordinates by (1.9).
We close this section by defining four important types of differentiable maps
between manifolds.
Definition 1.29. Let M be a manifold of dimension m, N a manifold of
dimension n, and let F : M → N be a differentiable map. F is called

(1) a diffeomorphism if F is bijective and the inverse map F −1 : N →

M is also differentiable; if such a map F exists, then we say that
M and N are diffeomorphic;
(2) an immersion if dFq : Tq M → TF (q) N is injective at every point
q ∈ M (note that this requires m ≤ n);
(3) an embedding if F is an injective immersion that is also a homeo-
morphism from M onto its image F (M ) ⊂ N , where F (M ) is given
the subspace topology inherited from the topology of N (note that
this requires m ≤ n);
(4) a submersion if dFq : Tq M → TF (q) N is surjective at every point
q ∈ M (note that this requires m ≥ n).
Remark 1.30. The definition 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 sufficiently small neighborhood of any point in its domain is always an
embedding.
26 1. Assorted notions from differential geometry

1.4. Lie groups and matrix groups

We begin with the following definition.

Definition 1.31. A Lie group is a set G that is both a group and a differen-
tiable manifold, with the additional property that the map μ : G × G → G
given by
μ(g, h) = gh−1
is differentiable.

This is a concise way of stating that group multiplication and group inverse
are both differentiable operations on G, as the following exercise shows.
Exercise 1.32. Let G be a Lie group. Show that the multiplication map
G × G → G defined by
(g, h) → gh
and the inverse map G → G defined by
g → g −1
are both differentiable. Conversely, show that differentiability of both of
these maps implies that the map μ in Definition 1.31 is differentiable. (Hint:
Consider the inverse map first.)

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 ∈ 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. Defining this structure will require a bit of
effort.
For each element g ∈ G, the left translation or left multiplication map Lg :
G → G is defined by
Lg (h) = gh.
(Similarly, the right translation map Rg : G → G is defined by
Rg (h) = hg.)
For any element h ∈ G, the differential (dLg )h of Lg at h is a linear map from
Th G to Tgh G. In particular, the differential 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
associate to any element v ∈ g a left-invariant vector field ṽ on G defined
at each point g ∈ G by
(1.11) ṽ(g) = (dLg )e (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 ∈ g, and
let ṽ be the vector field on G defined by (1.11).
(a) Show that for any elements g, h ∈ G, the value of ṽ at gh ∈ G is given
by
ṽ(gh) = dLg (ṽ(h)),
where
dLg : Th G → Tgh G
is the differential of the left multiplication map Lg . This is why the vector
field ṽ is called left-invariant: It is invariant under the action of left multi-
plication by any element of G. (It is also true that any left-invariant vector
field ṽ on a Lie group G is smooth; see, e.g., [Lee13].)
(b) Conversely, suppose that a smooth vector field V on G has the property
that for any g, h ∈ G,
(1.12) V(gh) = dLg (V(h)).
Show that V = ṽ, where v = V(e).

Recall that a vector field on G is really a first-order, linear, homogeneous

differential operator on smooth functions f : G → R. Given any two smooth
vector fields V, W on G (or indeed, any two smooth vector fields on any
manifold), the Lie bracket [V, W] is the smooth vector field defined by the
property that for any smooth function f : G → R,
[V, W][f ] = V[W[f ]] − W[V[f ]].
Wait a minute—a vector field is a first-order differential operator, but that
looks like a second-order differential operator! The following exercise should
allay your concerns:
*Exercise 1.34. Let V, W be two vector fields 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) , W(x) = bj (x)
∂xi ∂xj
for some functions (ai (x), bj (x)). Show by direct computation that the Lie
bracket [V, W] is a vector field—i.e., a first-order, linear, homogeneous dif-
ferential operator—on M . How are its coefficients related to (ai (x), bj (x))?
(Hint: Compute the action of the operator [V, W] on a smooth function
f : M → R. And since “smooth” means “infinitely differentiable”, you can
safely assume that mixed partial derivatives commute.)
28 1. Assorted notions from differential geometry

Finally, we use the Lie bracket of the left-invariant vector fields ṽ, w̃ to
define the Lie bracket operation on the Lie algebra g:
Definition 1.35. Let G be a Lie group with associated Lie algebra g. For
any two elements v, w ∈ g, the Lie bracket [v, w] of v and w is the unique
element z ∈ g such that
[ṽ, w̃] = z̃,
where tildes denote the extensions to left-invariant vector fields on G, as
defined above.

In order for this definition to make sense, we need to know that the Lie
bracket of two left-invariant vector fields on G is also left-invariant.
*Exercise 1.36. Let ṽ, w̃ be left-invariant vector fields on a Lie group G.
Show that the Lie bracket [ṽ, w̃] satisfies the condition (1.12) and that it
is therefore equal to z̃ for some z ∈ g. Conclude that the Lie bracket is a
well-defined operation on g.

While this all sounds fairly abstract, the most common examples of Lie
groups—and the only ones that we will encounter in this book—are sub-
groups of the group GL(n) of invertible n × n matrices. The differentiable
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 × 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 α(t) be a curve in G with α(0) = I.
Then we can write
α(t) = g(t),
where g(t) is an n × n matrix. The tangent vector to this curve at t = 0 is,
of course,
α (0) = g  (0),
2
which, while it could be thought of as a vector in Rn , is most naturally
regarded as an n × 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 − BA.
*Exercise 1.38. Let G be a subgroup of GL(n).
(a) Let g ∈ G, and let Lg : G → G be the left multiplication map. Show
that the differential dLg : g → Tg G is given explicitly by
dLg (A) = g · A, A ∈ g,
1.4. Lie groups and matrix groups 29

where the right-hand side is just the matrix product of the two n×n matrices
g and A. Therefore, the vector field à on G defined by
Ã(g) = g · A
is left-invariant. (Hint: Consider a curve h(t) in G with h(0) = I, h (0) = A.)
(b) Let’s get explicit about exactly what differential operator the vector field
à represents on G: If we use local coordinates (xij ) on the manifold of n × n
matrices, then a general element g ∈ G is represented by a matrix [xij ]. If
we write A = [aij ], then the tangent vector A ∈ g represents the differential
operator
∂
A = aij i ,
∂xj
and the matrix product à = g · A, whose entries are
Ãij = xir arj ,
represents the differential operator
∂
à = xir arj .
∂xij
Now, let A = [aij ], B = [bkl ] ∈ g. Then we can write
∂ ∂
à = xir arj , B̃ = xks bsl .
∂xij ∂xkl
Use the formula that you computed in Exercise 1.34 (and some very careful
index manipulation!) to show that
! ∂
[Ã, B̃] = xir arl blj − brl alj .
∂xij

(c) Conclude from part (b) that


[Ã, B̃] = g · (AB − BA) = (AB − BA)
and that the Lie bracket on g is therefore just the matrix commutator:
[A, B] = AB − 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 × n matrices. As a manifold, GL(n) is an open subset of the space Mn×n
of all n × n matrices; specifically,
GL(n) = {g ∈ Mn×n | det(g) = 0}.
30 1. Assorted notions from differential geometry

Thus, its Lie algebra gl(n) is the tangent space to the space of all n × n
matrices at the identity In . This tangent space is naturally isomorphic to
Mn×n .

Example 1.40. The special linear group SL(n). SL(n) is defined as

SL(n) = {A ∈ 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 − 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) ∈ SL(n) for all
t, we have

det(A(t)) = sgn(σ)a1σ(1) (t) · · · anσ(n) (t) = 1,

σ∈Sn

where the sum is over all permutations σ in the symmetric group Sn and
sgn(σ) = ±1 is the sign of σ. In order to determine what conditions this
imposes on the tangent vector A (0), we need to differentiate 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.
(Don’t 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) − a12 (t)a21 (t) = 1.

Differentiate 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 verified by comparing dimensions: SL(n) has dimension
n2 − 1, and the set of n × n matrices with trace equal to zero is a vector
space of dimension n2 − 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 ∈ Mn×n | tr(B) = 0}.
Example 1.42. The orthogonal group O(n). O(n) is defined as
O(n) = {A ∈ 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 − 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) ∈ O(n) for all t, we have
t
A(t) A(t) = In .
Differentiating 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 dimen-
sions as in the previous example shows that there are no other restrictions
on A (0), and so
o(n) = {B ∈ Mn×n | tB + B = 0}.
Example 1.43. The special orthogonal group SO(n). SO(n) is defined as
SO(n) = {A ∈ 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 defining equation
for O(n) that every element of O(n) has determinant equal to either 1 or
−1. Moreover, the function
det : O(n) → R
is continuous, with image equal to the set {1, −1}. Therefore, the subsets
SO(n) = {A ∈ O(n) | det(A) = 1},
O(n) \ SO(n) = {A ∈ O(n) | det(A) = −1}
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 −1. 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 differential 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 definition of a vector bundle is a bit complicated, but the general
idea is much like this example.
Definition 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 differentiable
projection map π : E → B such that for each point q ∈ B, the inverse image
π −1 (q) ⊂ E is a vector space of dimension k. This vector space is called
the fiber of E at q. Moreover, every point q ∈ B must have a neighborhood
U ⊂ B such that the set π −1 (U ) ⊂ E is diffeomorphic to U × Rk via a
diffeomorphism that, for every q̃ ∈ U , maps the fiber π −1 (q̃) linearly onto
{q̃} × Rk . Such a diffeomorphism is called a local trivialization of the vector
bundle.

It is important to note that, while vector bundles always have local trivial-
izations, they may not have global trivializations. That is to say, the local
trivializations may patch together in such a way that their union becomes
“twisted” in a way that cannot be straightened out. So, a vector bundle
π : E → B is not necessarily diffeomorphic to B × 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.
Definition 1.45. A (smooth) section of a vector bundle π : E → B is
a differentiable map σ : B → E with the property that the composition
π ◦ σ : B → B is the identity map on B.

A section σ : B → 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 σ has a domain space B, but at each point q ∈ B,
the value of σ(q) must lie in the fiber of E at q. Intuitively, this means
that the range space is different 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
identified with its image in E, which is a submanifold of E consisting of
exactly one point in each fiber. This may sound complicated, but again,
1.5. Vector bundles and principal bundles 33

you already know an example: A vector field on a surface Σ is a section of

the tangent bundle T Σ.
Any vector bundle has a distinguished section called the zero section; this is
the section that assigns to each point q ∈ B the zero vector in the fiber of
E at q. Other than the zero section, there is no canonical way of choosing
“constant” 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 “constant” along a curve, but not on any open set.) A connection
allows sections of vector bundles to be differentiated through a process called
covariant differentiation. The tangent bundle of a regular surface in R3
carries a canonical connection called the Levi-Civita connection, which is
used to define 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 Poincaré-Hopf theorem: Any smooth
vector field 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 weren’t enough, we will also frequently encounter prin-
cipal 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 differentiable projection map π : P → B. The main difference
is that the fiber π −1 (q) at any point q ∈ B is a manifold diffeomorphic
to some Lie group G rather than to a vector space. However, this diffeo-
morphism is generally not a group isomorphism in any canonical way. In
particular, there is usually no well-defined “identity section” 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 fiber π −1 (q) is a manifold
on which G acts freely and transitively.
A principal bundle π : P → B whose fiber at each point is diffeomorphic to
a Lie group G is often represented by the diagram

G - P

π
?
B.
34 1. Assorted notions from differential geometry

The arrow G → P doesn’t really represent a map here; it just indicates that
each fiber of P is diffeomorphic to G and that G acts freely and transitively
on the fibers 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 fibers will
all be diffeomorphic to some subgroup G of P .
Exercise 1.46. Let Σ be the unit sphere S2 ⊂ R3 . Let F (S2 ) denote the
orthonormal frame bundle of S2 ; this is the set of all orthonormal bases
(a.k.a. “frames”) e = (e1 , e2 ) for the tangent space Tq S2 at each point
q ∈ S2 . Given any orthonormal frame e at a point q ∈ S2 , any other frame
at q can be obtained from e by composition of a rotation and (possibly) a
reflection, 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 different frames. Thus, the fiber of F (S2 ) over
any point q ∈ S2 is diffeomorphic to O(2), and F (S2 ) is a principal bundle
over S2 with fiber group O(2).
Prove that F (S2 ) has no global sections. (Hint: Use the Poincaré-Hopf
theorem.)
 Chapter 2

Differential forms

2.1. Introduction

In calculus, you may have seen the differential or exterior derivative df of a

function f : R3 → R defined to be

∂f ∂f ∂f
df = dx + dy + dz.
∂x ∂y ∂z
The expression df is called a 1-form and you probably learned various ways
of manipulating such things—for 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 definitions for 1-forms and then more
generally for p-forms with p ≥ 0 on Rn , and we will explore a bit about how
they work. Once we understand how differential forms behave in Rn , we will
see how to define 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

Definition 2.1. Let V be an n-dimensional vector space. The dual space

V ∗ is the set of all linear mappings α : V → R.

It should be clear that V ∗ is closed under addition and scalar multiplication,

and so V ∗ is a vector space. The following exercise shows that V ∗ has

35
36 2. Differential forms

dimension n:
*Exercise 2.2. Let (e1 , . . . , en ) be a basis for V . Define linear functions
e∗i : V → R, 1 ≤ i ≤ n, by the property that

∗i i 1, i = j,
e (ej ) = δj =
0, i = j.
(Note that requiring e∗i to be linear and specifying the value of e∗i on all of
the basis vectors ej completely determines the function e∗i on V .)
(a) Show that the functions (e∗1 , . . . , e∗n ) are linearly independent. (Hint:
The hypothesis “ci e∗i = 0” means that for every v ∈ V , ci e∗i (v) = 0.)
(b) Show that the functions (e∗1 , . . . , e∗n ) span V ∗ . (Hint: Let α : V → R be
a linear mapping, and let ci = α(ei ). Consider the function ci e∗i : V → R.)
(c) Show that (V ∗ )∗ ∼
= V . (Hint: Associate to any vector v ∈ V the function
∗
v̂ : V → R defined by
v̂(α) = α(v).)

The basis (e∗1 , . . . , e∗n ) for V ∗ constructed in Exercise 2.2 is called the
dual basis to the basis (e1 , . . . , en ) for V . Also, note that the isomorphism
(V ∗ )∗ ∼
= V is canonical: It is completely independent of any choice of basis
for V . Thus it is accurate—and common—to regard (V ∗ )∗ 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 ∈ M , and let
V = Tq M . The dual space V ∗ = Tq∗ M is called the cotangent space to M
at q. The cotangent bundle of M is the union of all these cotangent spaces:

T ∗M = Tq∗ M.
q∈M

T ∗M has the structure of a smooth manifold of dimension 2m: Given any

parametrization x : U ⊂ Rm → M of M , there is an associated canonical
parametrization (x, p) : U × Rm → T ∗ M of T ∗ M , defined in a manner
analogous to the canonical parametrizations of the tangent bundle T M .
∗
Specifically, the element α = (x, p)(x1 , . . . , xm , p1 , . . . , pm ) ∈ Tx(x1 ,...,xm ) M

i ∈ Tx(x1 ,...,xm ) M , we have

∂
is defined by the condition that for any v = y i ∂x
α(v) = pi y i ∈ R.
Remark 2.4. More generally, given a vector space V , the dual space V ∗ is
sometimes referred to as the covector space of V , and the elements of V ∗ 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 ∗ are represented by row vectors, and vice versa.

Most of the tensor fields 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 official definition starts by defining
tensor products for dual spaces:
Definition 2.5. Let V be a vector space of dimension m and W a vector
space of dimension n. Let α ∈ V ∗ , β ∈ W ∗ . The tensor product of α and β,
denoted α ⊗ β, is the bilinear function
α⊗β :V ×W →R
defined by
(α ⊗ β)(v, w) = α(v)β(w).
The tensor product of V ∗ and W ∗ , denoted V ∗ ⊗ W ∗ , is the vector space
consisting of all linear combinations of such tensor products; i.e.,
V ∗ ⊗ W ∗ = span{α ⊗ β | α ∈ V ∗ , β ∈ W ∗ }.
*Exercise 2.6. (a) Show that α ⊗ β is indeed a bilinear function on V × W ,
i.e., that for any vectors v, v1 , v2 ∈ V , w, w1 , w2 ∈ W , and real numbers
a, b,
(α ⊗ β)(av1 + bv2 , w) = a(α ⊗ β)(v1 , w) + b(α ⊗ β)(v2 , w),
(α ⊗ β)(v, aw1 + bw2 ) = a(α ⊗ β)(v, w1 ) + b(α ⊗ β)(v, w2 ).

(b) Show that V ∗ ⊗ W ∗ is a vector space of dimension mn. (Hint: Let

(e1 , . . . , em ) be a basis for V , with dual basis (e∗1 , . . . , e∗m ) for V ∗ , and let
(f1 , . . . , fn ) be a basis for W , with dual basis (f ∗1 , . . . , f ∗n ) for W ∗ . Show
that
{e∗i ⊗ f ∗j | 1 ≤ i ≤ m, 1 ≤ j ≤ n}
is a basis for V ∗ ⊗ W ∗ .)

So, how do we define the tensor product V ⊗ W ? We simply take advantage

of the canonical identification V = (V ∗ )∗ that we proved in Exercise 2.2:
Definition 2.7. Let V be a vector space of dimension m and W a vector
space of dimension n. For any vectors v ∈ V, w ∈ W , define v̂ ∈ (V ∗ )∗ ,
ŵ ∈ (W ∗ )∗ by
v̂(α) = α(v), α ∈ V ∗,
ŵ(β) = β(w), β ∈ W ∗.
38 2. Differential forms

The tensor product of v and w, denoted v ⊗ w, is given by

v ⊗ w = v̂ ⊗ ŵ ∈ (V ∗ )∗ ⊗ (W ∗ )∗ .
The tensor product of V and W , denoted V ⊗ W , is the vector space
V ⊗ W = (V ∗ )∗ ⊗ (W ∗ )∗ .

Of course, this definition 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 ⊗ W is simply a vector space of dimension mn, with basis
given by the formal symbols
{ei ⊗ fj | 1 ≤ i ≤ m, 1 ≤ j ≤ n}.
Remark 2.8. The tensor product is associative; i.e.,
(v ⊗ w) ⊗ x = v ⊗ (w ⊗ x),
and so it makes sense to write v ⊗ w ⊗ x. But it is not commutative; even
when V = W , v ⊗ w is, in general, not equal to w ⊗ v.

Definition 2.9. A rank k tensor is an element of a tensor product of the

form
V1 ⊗ · · · ⊗ 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 ∗ ).
Example 2.10 (Cf. Example 1.12). Let V be an m-dimensional vector
space with basis (e1 , . . . , em ) and dual basis (e∗1 , . . . , e∗m ) for V ∗ , and let
W be an n-dimensional vector space with basis (f1 , . . . , fn ) and dual basis
(f ∗1 , . . . , f ∗n ) for W ∗ . Let T : V → W be a linear transformation, repre-
sented in terms of the given bases for V and W by the n × m matrix
⎡ 1 ⎤
c1 . . . c1m
⎢ .. ⎥ .
AT = ⎣ ... . ⎦
n
c1 . . . cm n

Then T is a rank 2 tensor; it is an element of the vector space W ⊗ V ∗ , and

it can be written in terms of the given bases as
T = cij fi ⊗ e∗j .

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

Definition 2.11. Let V be an n-dimensional vector space, and let V ⊗2 =

V ⊗ V . (Similarly, let V ⊗k 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 ⊗2 defined by
S 2 V = span{v ⊗ v | v ∈ V }.
The space of skew-symmetric 2-tensors of V is the subspace Λ2 V of V ⊗2
defined by
Λ2 V = span{v ⊗ w − w ⊗ v | v, w ∈ V }.
Exercise 2.12. Show that for any vectors v, w ∈ V ,
v ⊗ w + w ⊗ v ∈ S 2 V.
(Hint: Consider (v + w) ⊗ (v + w).)
Definition 2.13. The symmetric product v ◦ w and wedge product v ∧ w
are defined by
v ◦ w = 12 (v ⊗ w + w ⊗ v),
v ∧ w = v ⊗ w − w ⊗ v.
(Note that some authors insert a factor of 12 into the definition of the wedge
product as well.) More generally, if v1 , . . . , vk ∈ V , then
1
v1 ◦ · · · ◦ vk = v ⊗ · · · ⊗ vσ(k) ,
k! σ σ(1)

v1 ∧ · · · ∧ vk = sgn(σ)vσ(1) ⊗ · · · ⊗ vσ(k) ,
σ
where the sum is over all permutations σ in the symmetric group Sk and
sgn(σ) = ±1 is the sign of σ. The spaces of symmetric and skew-symmetric
k-tensors of V are the subspaces S k V , Λk V of V ⊗k defined by
S k V = span{v1 ◦ · · · ◦ vk | v1 , . . . , vk ∈ V },
Λk V = span{v1 ∧ · · · ∧ vk | v1 , . . . , vk ∈ V }.
Exercise 2.14. (a) Show that V ⊗ V = S 2 V ⊕ Λ2 V ; i.e., every element of
V ⊗ 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 ⊗3 . (Hint:
Compute the dimensions of V ⊗3 , S 3 V , and Λ3 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
fiber is isomorphic to a fixed tensor product of vector spaces. In the most
common examples, the fiber over each point q ∈ M is a tensor product of
40 2. Differential forms

some numbers of copies of the tangent space Tq M and the cotangent space
Tq∗ M , or some natural subspace of such a tensor product. Finally, this allows
us to give a formal definition of tensor fields:
Definition 2.15. A rank k tensor field on a manifold M is a section of a
vector bundle π : E → M , where the fiber Eq of E over each point q ∈ M
has the form
Eq = V1 ⊗ · · · ⊗ Vk ,
where V1 , . . . , Vk are vector spaces. (In most cases, each of the vector spaces
V1 , . . . , Vk is equal to either Tq M or Tq∗ M .)
Remark 2.16. In practice, tensor fields on a manifold M are almost always
referred to as “tensors on M ”. For example, a Riemannian manifold M has
a “metric tensor” g and a “Riemann curvature tensor” R. Strictly speaking,
these are both tensor fields on M , but you will probably never encounter
the phrase “Riemann curvature tensor field”.
Example 2.17 (Cf. Exercise 1.15). Let M be a manifold of dimension n,
and let 
S 2 (T ∗ M ) = S 2 (Tq∗ M ).
q∈M
S 2 (T ∗ M ) is a subbundle of the tensor bundle T ∗ M ⊗ T ∗ M , and it is a
bundle of symmetric 2-tensors on M . (Note that I didn’t say “the” 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 ∗ M ) because, at every point q ∈ M , g defines a symmetric, bilinear
function
gq : Tq M × Tq M → R.

Now we’re ready to start talking about differential forms!

2.3. 1-forms on Rn

Definition 2.18. A (smooth) 1-form φ on Rn is a (smooth) section of the

cotangent bundle T ∗ Rn . Equivalently,
φ : T Rn → R
is a real-valued function on the set of all tangent vectors to Rn , with the
properties that:

(1) For each x ∈ Rn , the restriction

φx : Tx Rn → R
of φ to Tx Rn is a linear map.
2.4. p-forms on Rn 41

(2) For any smooth vector field v on Rn , the function φ(v) : Rn → R

is smooth.

In particular, the 1-forms (dx1 , . . . , dxn ) are defined by the property that
for any vector v = t[v 1 , . . . , v n ] ∈ 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 ∈ Rn , the 1-forms (dx1 , . . . , dxn ) defined by (2.1) form
a basis for the cotangent space Tx∗ 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 ∈ Tx Rn . Show that this
implies that c1 = · · · = cn = 0.
(b) In order to show that the 1-forms (dx1 , . . . , dxn ) span Tx∗ Rn , let φx ∈
Tx∗ Rn be arbitrary. Let (e1 , . . . , en ) be the standard basis for Tx Rn , and let
ci = φx (ei ), i = 1, . . . , n.
Show that φx = ci dxi . (It suffices to show that φx (v) = ci dxi (v) for every
v ∈ 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 ∞ (Rn ); this means that any 1-form φ
on Rn can be expressed uniquely as
(2.2) φ = fi (x) dxi
for some smooth, real-valued functions f1 , . . . , fn : Rn → R. If a vector field
v on Rn has the form
v(x) = t[v 1 (x), . . . , v n (x)],
then we have
φ(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 differential forms on Rn . The multiplication in this algebra is the wedge
product of Definition 2.13. The wedge product is multilinear and skew-
42 2. Differential forms

symmetric; i.e., for any two 1-forms φ, ψ, we have

φ ∧ ψ = −ψ ∧ φ.

In particular, φ ∧ φ = 0 for any 1-form φ.

If each summand of a differential form Φ 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 φ1 , . . . , φp be 1-forms.

According to (2.2), we can write

φj = fji (x) dxi , j = 1, . . . , p.

Show by direct computation that if p > n, then φ1 ∧ · · · ∧ φp = 0.)

More formally, we have the following definition:

Definition 2.21. A p-form on Rn is a section of the tensor bundle Λp (T ∗ Rn ).

The algebra of differential forms on Rn consists of all sections of the bundle
$  $
Λp (T ∗ Rn ) = Λp (Tx∗ Rn ),
p≥0 x∈Rn p≥0

with multiplication given by the wedge product

∧ : Λp (T ∗ Rn ) × Λq (T ∗ Rn ) → Λp+q (T ∗ Rn ).

The space of p-forms on Rn is generally denoted by Ωp (Rn ), and the algebra

of all differential forms on Rn is denoted by
$
Ω∗ (Rn ) = Ωp (Rn ).
p≥0

A basis for the p-forms on Rn (as a module over C ∞ (Rn )) is given by the
set
{dxi1 ∧ · · · ∧ dxip | 1 ≤ i1 < i2 < · · · < ip ≤ n};
this simply means that any p-form Φ on Rn can be expressed uniquely as

Φ= fI (x) dxi1 ∧ · · · ∧ dxip ,

|I|=p

where I ranges over all increasing multi-indices I = (i1 , . . . , ip ) of length p

and the coefficients fI (x) represent smooth functions fI : Rn → R.
2.5. The exterior derivative 43

Just as 1-forms act on vector fields to give real-valued functions, so p-forms

act on p-tuples of vector fields to give real-valued functions. For instance, if
φ, ψ are 1-forms and v, w are vector fields, then
(φ ∧ ψ)(v, w) = φ(v)ψ(w) − φ(w)ψ(v).
More generally, if φ1 , . . . , φp are 1-forms and v1 , . . . , vp are vector fields,
then
(φ1 ∧ · · · ∧ φp )(v1 , . . . , vp ) = sgn(σ) φ1 (vσ(1) ) φ2 (vσ(2) ) · · · φn (vσ(n) ),
σ∈Sp

where the sum is over all permutations σ in the symmetric group Sp and
sgn(σ) = ±1 is the sign of σ.
Exercise 2.22. Show that the wedge product is related to the determinant
as follows: If φ1 , . . . , φp are 1-forms and v1 , . . . , vp are vector fields, then
⎡ ⎤
φ1 (v1 ) · · · φ1 (vp )
⎢ ⎥
⎢ .. ⎥ .
(φ1 ∧ · · · ∧ φp )(v1 , . . . , vp ) = det ⎢ ... . ⎥
⎣ ⎦
φp (v1 ) · · · φp (vp )

2.5. The exterior derivative

The exterior derivative is an operator that takes p-forms to (p + 1)-forms.

We will define it first for functions and then extend this definition to higher
degree forms.
Definition 2.23. If f : Rn → R is differentiable, then the exterior derivative
of f is the 1-form df with the property that for any x ∈ Rn , v ∈ 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 defined 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 difficult to show that, as one might expect,

∂f i
(2.3) dx .
df =
∂xi
The exterior derivative also obeys the Leibniz rule
(2.4) d(f g) = g df + f dg
44 2. Differential forms

for functions f, g : Rn → R and the chain rule

(2.5) d(h ◦ f ) = (h ◦ f ) df
for functions f : Rn → R, h : R → R.

*Exercise 2.25. Verify equations (2.3), (2.4), and (2.5).

We extend the definition of the exterior derivative to p-forms on Rn as

follows:

Definition 2.26. Given a p-form Φ = fI (x) dxi1 ∧ · · · ∧ dxip on Rn , the

|I|=p
exterior derivative dΦ of Φ is the (p + 1)-form

(2.6) dΦ = dfI ∧ dxi1 ∧ · · · ∧ dxip .

|I|=p

If Φ is a p-form and Ψ is a q-form, then the Leibniz rule takes the form
(2.7) d(Φ ∧ Ψ) = dΦ ∧ Ψ + (−1)p Φ ∧ dΨ.

*Exercise 2.27. Prove the Leibniz rule (2.7) in the case p = q = 1: If φ, ψ

are 1-forms on Rn , then
d(φ ∧ ψ) = dφ ∧ ψ − φ ∧ dψ.

The following theorem is possibly the most important one in this entire
book!

Theorem 2.28. d ◦ d = 0; i.e., for any differential form Φ on Rn ,

d(dΦ) = 0.

Proof. First, suppose that f is a function (i.e., a 0-form). Then

" #
∂f i
d(df ) = d dx
∂xi
n
∂ 2f
= dxj ∧ dxi
∂xi ∂xj
i,j=1
" 2 #
∂ f ∂2f
= − dxi ∧ dxj
∂xj ∂xi ∂xi ∂xj
i<j
=0
because mixed partials commute.
2.5. The exterior derivative 45

Next, note that by Exercise 2.24 and the argument above, d(dxi ) = 0. Now
suppose that
Φ= fI (x) dxi1 ∧ · · · ∧ dxip .
|I|=p
Then by the Leibniz rule,
⎛ ⎞

d(dΦ) = d ⎝ dfI ∧ dxi1 ∧ · · · ∧ dxip ⎠

|I|=p
 
= d(dfI ) ∧ dxi1 ∧ · · · ∧ dxip − dfI ∧ d(dxi1 ) ∧ · · · ∧ dxip + · · ·
|I|=p
= 0.

*Exercise 2.29. In multivariable calculus, there are three basic operations
based on partial differentiation involving functions and vector fields on R3 :

(1) the gradient of a function f : R3 → R:

3
∂f ∂
grad(f ) = ;
∂xi ∂xi
i=1

∂
(2) the curl of a vector field v(x) = v i (x) i :
∂x
" 3 2
# " 1 3
# " 2 #
∂v ∂v ∂ ∂v ∂v ∂ ∂v ∂v 1 ∂
curl(v) = 2
− 3 1
+ 3
− 1 2
+ 1
− 2 ;
∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x ∂x3
∂
(3) the divergence of a vector field v(x) = v i (x) :
∂xi
3
∂v i
div(v) = .
∂xi
i=1

These operations can all be expressed in terms of differential forms and their
exterior derivatives via the following correspondences between forms on the
one hand and vector fields and functions on the other:
3
∂
f1 (x)dx1 + f2 (x)dx2 + f3 (x)dx3 ←→ fi (x) ,
∂xi
i=1
3
∂
f1 (x)dx ∧ dx + f2 (x)dx ∧ dx + f3 (x)dx ∧ dx ←→
2 3 3 1 1 2
fi (x) ,
∂xi
i=1
f (x) dx ∧ dx ∧ dx ←→ f (x).
1 2 3
46 2. Differential forms

Prove that under this correspondence:

(1) If f is a function on R3 , then df ←→ grad(f ).

(2) If φ is a 1-form on R3 and φ ←→ v, then dφ ←→ curl(v).
(3) If η is a 2-form on R3 and η ←→ v, then dη ←→ div(v).

Show that the identities

curl(grad(f )) = 0,
div(curl(v)) = 0
follow from the fact that d ◦ d = 0.

2.6. Closed and exact forms and the Poincaré lemma

In light of Theorem 2.28, we introduce the following important classes of

differential forms:

Definition 2.30. A p-form Φ on Rn is called closed if dΦ = 0. Φ is called

exact if there exists a (p − 1)-form Ψ on Rn such that Φ = dΨ.

By Theorem 2.28, every exact form is closed. The converse is only partially
true: Every closed form is locally exact. Precisely, we have

Theorem 2.31 (The Poincaré lemma). Given a closed p-form Φ on an open

set U ⊂ Rn that is homeomorphic to an open ball, there exists a (p − 1)-form
Ψ on U with dΨ = Φ.

(For proofs of the results in this section, see Volume 1 of [Spi79].) In

particular, when p = 1 we have

Corollary 2.32. Given a closed 1-form φ on an open set U ⊂ Rn that is

homeomorphic to an open ball, there exists a differentiable function f : U →
R such that φ = df .

Closely related to Corollary 2.32 is the following special case of the Frobenius
theorem:

Theorem 2.33 (Frobenius). Let φ be a nonvanishing 1-form on an open

set U ⊂ Rn , and suppose that φ ∧ dφ = 0. Then for any point x ∈ U , there
exist a neighborhood V ⊂ U of x and differentiable functions f, g : V → R
such that the restriction of φ to V is given by φ = g df .
2.7. Differential forms on manifolds 47

2.7. Differential forms on manifolds

Now that we know how differential forms behave on Rn , we might expect

them to behave similarly in terms of any local coordinate system on an
arbitrary manifold M . But if we want differential forms to be well-defined
objects on M , then they need to be defined independently of any particular
choice of local coordinate system. So, we start with the following coordinate-
free definition:
Definition 2.34. Given a smooth manifold M , a smooth 1-form φ on M
is a smooth section of the cotangent bundle T ∗ M . Equivalently, φ is a
real-valued function φ : T M → R on the tangent bundle of M such that:

(1) For each q ∈ M , the restriction

φ : Tq M → R
of φ to Tq M is a linear map.
(2) For any smooth vector field v on M , the function φ(v) : M → R is
smooth.

Now, suppose that we have a system of local coordinates x = (x1 , . . . , xn )

on anopen set U ⊂  M . Associated to these local coordinates, we have the
basis ∂x∂ 1 , . . . , ∂x∂n for the local sections of the tangent bundle T M over U .
We define the 1-forms (dx1 , . . . , dxn ) on U to be the dual basis for the local
sections of the cotangent bundle T ∗ M over U ; in other words, the 1-forms
(dxi ) are defined by the property that
" # 
i ∂ 0, i = j,
dx =
∂xj 1, i = j.
Any p-form Φ on U then has a unique expression as
Φ= fI (x) dxi1 ∧ · · · ∧ dxip
|I|=p

for some functions fI on U .

Wedge products, p-forms, and exterior derivatives of functions are defined
just as for Rn . (Go back and convince yourself that these definitions are
coordinate-independent, and so they make sense on an arbitrary manifold
without having to specify a particular choice of local coordinates.) However,
Definition 2.26 for the exterior derivative of a p-form with p ≥ 1 does de-
pend on a choice of local coordinates, and so extending this definition to an
arbitrary manifold requires some care. The most straightforward approach
is to go ahead and use Definition 2.26 and to prove that this definition is
consistent under local coordinate transformations.
48 2. Differential forms

Definition 2.35. Let Φ be a p-form on a manifold M , and suppose that Φ

can be expressed as

Φ= fI (x) dxi1 ∧ · · · ∧ dxip

|I|=p

in terms of some system of local coordinates x = (x1 , . . . , xn ) on M . Then

the exterior derivative dΦ of Φ is the (p + 1)-form whose local coordinate
expression is
dΦ = dfI ∧ dxi1 ∧ · · · ∧ dxip .
|I|=p

*Exercise 2.36. Show that Definition 2.35 is well-defined for 1-forms, as

follows: Suppose that x = (x1 , . . . , xn ) and x̄ = (x̄1 , . . . , x̄n ) are two systems
of local coordinates on an open set U ⊂ M , with transition functions
 
(x̄1 , . . . , x̄n ) → x1 (x̄1 , . . . , x̄n ), . . . , xn (x̄1 , . . . , x̄n ) .
Then according to (2.3),
∂xi j
dxi =
dx̄ .
∂ x̄j
(Note that this equation may be expressed as
⎡ 1⎤ ⎡ 1⎤
dx dx̄
⎢ .. ⎥ ⎢ .. ⎥
⎣ . ⎦ = J ⎣ . ⎦,
dxn dx̄n
where J is the Jacobian matrix of the coordinate transformation. Compare
with Exercise 1.22.)
(a) Let φ be a 1-form on U , defined by
φ = fi (x) dxi
with respect to the local coordinates x = (x1 , . . . , xn ). Show that
∂xi k
φ = fi (x(x̄)) dx̄
∂ x̄k
with respect to the local coordinates x̄ = (x̄1 , . . . , x̄n ).
(b) Compute dφ for both of the local coordinate expressions in part (a), and
use the chain rule to show that both expressions for dφ represent the same
2-form on U .

Alternatively, the exterior derivative may be defined independently of any

choice of local coordinates: According to Cartan’s formula, if Φ is a p-form,
2.8. Pullbacks 49

then dΦ is the (p + 1)-form defined by the property that for any p + 1 vector
fields v1 , . . . , vp+1 on M ,
p+1
(2.8) dΦ(v1 , . . . , vp+1 ) = (−1)i+1 vi [Φ(v1 , . . . , v̂i , . . . , vp+1 )]
i=1

+ (−1)i+j Φ([vi , vj ], v1 , . . . , v̂i , . . . , v̂j , . . . , vp+1 ).

i<j

Here a hat over a symbol indicates that that element is omitted from the list;
vi [Φ(v1 , . . . , v̂i , . . . , vp+1 )] denotes the directional derivative of the scalar-
valued function Φ(v1 , . . . , v̂i , . . . , vp+1 ) in the direction of vi , and [vi , vj ]
denotes the Lie bracket of the vector fields vi and vj .

Remark 2.37. This definition is obviously somewhat cumbersome, and it

is rarely used in practice. Its importance lies in its theoretical significance:
Since the exterior derivative is well-defined independently of a choice of local
coordinates, it really should have a coordinate-free definition, and Cartan’s
formula provides one.

Exercise 2.38. Show by direct computation that Cartan’s formula agrees

with Definition 2.35 in the case when Φ is a 1-form: Specifically, let
∂ ∂
φ = fi (x)dxi , v = aj (x) , w = bk (x) ,
∂xj ∂xk
and compute dφ(v, w) both in local coordinates as in (2.6) and according to
Cartan’s formula (2.8). This gives another proof (at least for 1-forms) that
the exterior derivative in Definition 2.35 is, in fact, well-defined.

2.8. Pullbacks

Now, suppose that M, N are smooth manifolds and that F : M → N is

a differentiable map. Recall from Definition 1.25 that for any q ∈ M , the
differential of F at q is the linear map
dFq : Tq M → TF (q) N
defined by the property that for any curve α : R → M with α(0) =
q, α (0) = v,
dFq (v) = (F ◦ α) (0).
The differential dF may be thought of as a vector-valued (more specifically,
T N -valued) 1-form on M because for each q ∈ M , it defines a linear map
on Tq M that takes values in the vector space TF (q) N .
50 2. Differential forms

Remark 2.39. This is why the same notation is used both for the differen-
tial dF of a map F : M → N and for the exterior derivative df of a function
f : M → R: The exterior derivative of a function really is just the differen-
tial of the function regarded as a map between manifolds. (This equivalence
relies on the canonical isomorphism Tf (q) R ∼
= R.)
*Exercise 2.40. Let M be a manifold of dimension m, N a manifold of
dimension n, and F : M → N a differentiable map with differential dF :
T M → T N . Let (f1 , . . . , fn ) be vector fields on N that form a basis for Tq N
at each point q ∈ N . Show that there exist ordinary, scalar-valued 1-forms
(φ1 , . . . , φn ) on M defined by the condition that
dF = fi φi .
(Recall that this means that for any v ∈ Tq M ,
dFq (v) = fi φi (v) ∈ TF (q) N.)
If the basis (f1 , . . . , fn ) is understood, then we may identify dF with the
Rn -valued 1-form ⎡ 1⎤
φ
⎢ .. ⎥
φ=⎣ . ⎦
φn
on M .

Because the differential takes tangent vectors on M to tangent vectors on

N , it is sometimes called the “push-forward” map from T M to T N , and
it is often denoted by F∗ rather than dF . (So, for v ∈ Tq M , the vector
dF (v) ∈ TF (q) N is often written as F∗ (v).) There is an analogous map in
the opposite direction for differential forms, called the pullback and denoted
F ∗ . It is defined as follows.
Definition 2.41. Let F : M → N be a differentiable map.

(1) If f : N → R is a differentiable function, then F ∗ f : M → R is the

composite function defined by
(F ∗ f )(q) = (f ◦ F )(q)
for q ∈ M . This may be represented by the following commutative
diagram:
M
F -N
J
F ∗f J f
^ 
J
R.
2.8. Pullbacks 51

(2) If φ is a p-form on N , then F ∗ φ is the p-form on M defined by

its action on tangent vectors to M as follows: If q ∈ M and
v1 , . . . , vp ∈ Tq M , then

(F ∗ φ)(v1 , . . . , vp ) = φ(F∗ (v1 ), . . . , F∗ (vp )).

(Note that F∗ (v1 ), . . . , F∗ (vp ) are elements of TF (q) N , so the right-

hand side makes sense.) This may be represented by the following
commutative diagram:
Λp (Tq M ) -
F∗ p
Λ (TF (q) N )
J
F ∗φ J φ
^
J 
R.

In this diagram, the map F∗ : Λp (Tq M ) → Λp (TF (q) N ) represents

the obvious extension of the map F∗ : Tq M → TF (q) N ; namely,

F∗ (v1 ∧ · · · ∧ vp ) = F∗ (v1 ) ∧ · · · ∧ F∗ (vp ).

In the special case where F : M → N is a diffeomorphism, the pullback F ∗

defines a differentiable map from T ∗ N to T ∗ M ; in particular, for q̃ ∈ N ,
F ∗ takes the cotangent space Tq̃∗ N to TF∗ −1 (q̃) M . Note that this map goes
in the opposite direction from the differential F∗ , which maps T M to T N ;
the following commutative diagrams show how these maps operate:

F∗ -
v ∈ TM T N  F∗ v

? F ?
M - N

and

F∗
F ∗φ ∈ T ∗M  T ∗N  φ

? F ?
M - N.

Just as the differential F∗ : T M → T N can be extended to a differentiable

map F∗ : Λp (T M ) → Λp (T N ), so the pullback F ∗ : T ∗ N → T ∗ M can be
52 2. Differential forms

extended to a differentiable map F ∗ : Λp (T ∗ N ) → Λp (T ∗ M ) as follows:

F ∗ (φ1 ∧ · · · ∧ φp ) = F ∗ (φ1 ) ∧ · · · ∧ F ∗ (φp ).

For more general differentiable maps F : M → N , the picture is slightly

more complicated: According to Definition 2.41, the pullback F ∗ maps p-
forms on N (i.e., sections of the tensor bundle Λp (T ∗ N )) to p-forms on M
(i.e., sections of the tensor bundle Λp (T ∗ M )). But in general, there may
be no well-defined map F ∗ : Λp (T ∗ N ) → Λp (T ∗ M ), because for any given
point q̃ ∈ N , there may not exist a unique point q ∈ M (or indeed, any
point at all) with F (q) = q̃.
In terms of local coordinates x = (x1 , . . . , xm ) on M and y = (y 1 , . . . , y n )
on N , suppose that the map F is described by
 
F (x1 , . . . , xm ) = y 1 (x1 , . . . , xm ), . . . , y n (x1 , . . . , xm ) .
Recall from Chapter 1 that the differential F∗ at each point q ∈ M !may
 
be represented in terms of the bases ∂x∂ 1 , . . . , ∂x∂m and ∂y∂ 1 , . . . , ∂y∂n for
Tq M and TF (q) N , respectively, by the matrix
 
∂y i
;
∂xj
specifically,
" #
∂ ∂y i ∂
F∗ = .
∂xj ∂xj ∂y i

What about F ∗ ? The pullback map takes forms on N to forms on M , and

it acts on the basis 1-forms dy i by
∂y i j
(2.9) F ∗ (dy i ) = dx .
∂xj
*Exercise 2.42. Use Definition 2.41 to prove the formula (2.9).
Exercise 2.43. Let (r, θ) be coordinates on R2 and (x, y, z) coordinates on
R3 . Let F : R2 → R3 be defined by
F (r, θ) = (cos(θ), sin(θ), r).
Describe the differential F∗ in terms of these coordinates, and compute the
pullbacks F ∗ (dx), F ∗ (dy), F ∗ (dz).

The pullback map behaves as nicely as one could hope with respect to the
various operations on differential forms, as described in the following theo-
rem.
2.9. Integration and Stokes’s theorem 53

Theorem 2.44. Let F : M → N be a differentiable map, and let Φ, Ψ be

differential forms on N . Then:

(1) F ∗ (Φ + Ψ) = F ∗ Φ + F ∗ Ψ.
(2) F ∗ (Φ ∧ Ψ) = (F ∗ Φ) ∧ (F ∗ Ψ).
(3) F ∗ (dΦ) = d(F ∗ Φ).
*Exercise 2.45. Prove Theorem 2.44 in the case where Φ, Ψ are 1-forms.
*Exercise 2.46. Suppose that m = n = 2 and that F : M → N is described
in local coordinates by
 
F (x1 , x2 ) = y 1 (x1 , x2 ), y 2 (x1 , x2 ) .
Show that  1 
 ∂y ∂y 1 
 1   ∂x1 ∂x2 
F dy ∧ dy =  2
∗ 2
dx1 ∧ dx2 .
 ∂y 1 ∂y 2 
∂x ∂x2
(You may recognize this expression from the change of variables formula for
double integrals.)

Exercise 2.46 is a special case of the following generalization of equation

(2.9), which follows from Theorem 2.44:
⎛ ⎞

(2.10) F ∗ ⎝ fI (y) dy i1 ∧ · · · ∧ dy ip ⎠
|I|=p
⎛ ⎛ ⎞⎞
m
∂y i1 ∂y ip ⎠⎠ j1
=⎝ ⎝ fI (y(x)) j1 · · · jp dx ∧ · · · dxjp .
∂x ∂x
|I|=p j1 ,...,jp =1

2.9. Integration and Stokes’s theorem

While we won’t need it for the topics in this book, no discussion of differential
forms is complete without a mention of Stokes’s theorem, which has to do
with the integration of differential forms over manifolds. Generally speaking,
a p-form Φ on a manifold M can be integrated over a p-dimensional sub-
manifold P of M , and pullbacks allow us to define such integrals in terms of
ordinary integrals over regions in Rp . This construction requires the notion
of a “manifold with boundary”, which means more or less exactly what it
sounds like; common examples might include a curve segment in Rn (includ-
ing its endpoints) or a hemisphere in R3 (including the boundary circle). It
also requires that the submanifold P be orientable, which means that P can
54 2. Differential forms

be covered by a collection of parametrizations xi : U ⊂ Rp → P with the

property that all the transition functions x−1
j ◦ xi have Jacobian matrices
with positive determinant. The choice of such a collection of parametriza-
tions is called an orientation on P , and P must be given an orientation
before integrals over P can be defined.
%
In order to define the integral P Φ, let D ⊂ Rp be a closed and bounded
region in Rp , and let x : D → P be a parametrization of P that is com-
patible with the orientation of P . (Differentiability at boundary points of
D is defined via one-sided limits, and “compatible with the orientation”
means that the Jacobian of the transition function between x and any of
the parametrizations xi in the orientation has positive determinant.) For
any p-form Φ on M , the integral of Φ over the region x(D) ⊂ P is defined
to be & &
Φ= x∗ (Φ).
x(D) D
(Integrals over regions in Rpare
% defined exactly as you would expect.) In
order to compute the integral P Φ, simply divide P into regions that can
each be covered by a single parametrization and add the integrals
% of Φ over
each region. The formula (2.10) can be used to show that P Φ is well-
defined, independent of the choice of parametrizations used to compute it.
If P is a manifold with boundary ∂P , then an orientation on P induces
an orientation on the boundary ∂P . The precise definition for the induced
orientation is a bit complicated (see, e.g., [Lee13]), but intuitively it involves
constructing parametrizations for ∂P by choosing oriented parametrizations
xi for neighborhoods V ⊂ P containing points of ∂P so that V ∩∂P is defined
by the equation xp = 0, with the remaining coordinates chosen so that the
tangent vector ∂x∂ p points “outward” from P .
Theorem 2.47 (Stokes). Let M be a closed, oriented, (p + 1)-dimensional
manifold, possibly with boundary, and let ∂M denote the p-dimensional
boundary (possibly empty) of M , with orientation induced from the orienta-
tion of M . Let Φ be a p-form on M . Then
& &
(2.11) Φ= dΦ.
∂M M

*Exercise 2.48. Show that the following theorems are all special cases of
Stokes’s theorem:
(a) The Fundamental Theorem of Calculus: If f : [a, b] → R is a differen-
tiable function, then
& b
f  (x) dx = f (b) − f (a).
a
2.10. Cartan’s lemma 55

(Hint: Let M ⊂ R be the closed interval [a, b]. The oriented boundary of M
is the formal sum (b) − (a), and the “integral” of a function f : [a, b] → R
over ∂M is f (b) − f (a).)
(b) Green’s theorem: Let D ⊂ R2 be a region bounded by a simple closed
curve C (oriented counterclockwise), and let P, Q : D → R be differentiable
functions. Then
' &&
(P dx + Q dy) = (Qx − Py ) dx ∧ dy.
C D

(c) Stokes’s theorem (multivariable calculus version): Let S be an oriented

surface in R3 bounded by a simple closed curve C with the induced orien-
tation, and let P, Q, R : U → R be differentiable functions on an open set
U ⊂ R3 containing S. Then
'
(P dx + Q dy + R dz)
C
&&
= ((Ry − Qz ) dy ∧ dz + (Pz − Rx ) dz ∧ dx + (Qx − Py ) dx ∧ dy) .
S
(Bonus points if you can express this in the traditional vector form as
' &&
F · dr = (∇ × F) · n dA !)
C S

(d) The Divergence theorem: Let V ⊂ R3 be a closed and bounded region

with piecewise smooth boundary S (oriented with outward-pointing normal
vector), and let P, Q, R : V → R be differentiable functions. Then
&& &&&
(P dy ∧ dz + Q dz ∧ dx + R dx ∧ dy) = (Px + Qy + Rz ) dx ∧ dy ∧ dz.
S V
(Bonus points if you can express this in the traditional vector form as
&& &&&
(F · n) dA = (∇ · F) dV !)
S V

2.10. Cartan’s lemma

In this section, we present a technical result known as Cartan’s lemma. We

will make frequent use of this result beginning in Chapter 4.
Lemma 2.49 (Cartan). Suppose that η 1 , . . . , η n are linearly independent,
smooth 1-forms on a manifold M and that φ1 , . . . , φn are smooth 1-forms
on M such that
(2.12) φi ∧ η i = 0.
56 2. Differential forms

Then there exist smooth functions {hij = hji : M → R | 1 ≤ i, j ≤ n},

symmetric in their lower indices, such that
φi = hij η j .

Proof. Suppose that M has dimension s ≥ n, and let η n+1 , . . . , η s be any

smooth 1-forms on M with the property that (η 1 , . . . , η n , η n+1 , . . . , η s ) form
a basis for the cotangent space Tq∗ M at each point q ∈ M . Then we can
write
(2.13) φi = hiα η α , 1 ≤ i ≤ n,
for some smooth functions (hiα ) on M , where the sum is over 1 ≤ α ≤ s.
*Exercise 2.50. Show that substituting these expressions into (2.12) and
taking the skew-symmetry of the wedge product into account yields
hir = 0, n + 1 ≤ r ≤ s,
hji = hij , 1 ≤ i, j ≤ n.
Therefore, φi = hij η j , where the sum is over 1 ≤ j ≤ n.


Remark 2.51. Technically, this proof only works if the tangent bundle
T M has a global trivialization, because otherwise there might not exist
complementary 1-forms η n+1 , . . . , η s that are globally defined on M . But
such 1-forms always exist on some neighborhood of any given point in M ,
so this argument can be used to prove the lemma in general by showing that
equation (2.13) holds in a neighborhood of any point in M and hence on all
of M .

2.11. The Lie derivative

The final operation that we will define on differential forms is the Lie deriv-
ative. This is a generalization of the notion of the directional derivative of
a function.
Suppose that v is a vector field on a manifold M , and let ϕ be the flow of
v. This is, by definition, the unique map ϕ : V → M , defined on some open
neighborhood V of the set M × {0} in M × R, that satisfies the conditions
∂ϕ
(2.14) (q, t) = v(ϕ(q, t)),
∂t
ϕ(q, 0) = q.
2.11. The Lie derivative 57

(The existence and uniqueness of the flow on some neighborhood V of

M × {0} is equivalent to the existence and uniqueness of local solutions
to first-order systems of ordinary differential equations; see, e.g., [Lee13].)
In other words, ϕ(q, t) is the point reached at time t by flowing along the
vector field v starting from the point q at time 0. The expression ϕ(q, t) is
sometimes denoted ϕt (q) to emphasize that, for each fixed t, the flow defines
a differentiable map ϕt : U → M , where U = {q ∈ M | (q, t) ∈ V }.
Recall that if f : M → R is a smooth function, then the directional derivative
of f at q in the direction of v may be defined as

f (ϕt (q)) − f (q)

v[f ] = lim
t→0 t
(ϕ∗t (f ) − f )(q)
= lim .
t→0 t

This motivates the following definition:

Definition 2.52. Given a p-form Φ and a vector field v on M , the Lie

derivative of Φ along the vector field v is the p-form

ϕ∗t Φ − Φ
(2.15) Lv Φ = lim ,
t→0 t

where ϕ∗t Φ denotes the pullback of the p-form Φ via the map ϕt .

It is rarely possible to compute the limit in (2.15) directly; among other

things, finding an explicit formula for the map ϕt : U → M requires solving
the system of differential equations (2.14). But fortunately, there is a prac-
tical way to compute the Lie derivative. First, we need the notion of the
interior product of a differential form with a vector field.

Definition 2.53. Given a p-form Φ and a vector field v on M , the interior

product of Φ with v (also called the left-hook of v with Φ) is the (p − 1)-form
v Φ on M defined by the property that for any vector fields w1 , . . . , wp−1
on M ,

(v Φ)(w1 , . . . , wp−1 ) = Φ(v, w1 , . . . , wp−1 ).

Example 2.54. Let Φ be the 2-form Φ = dx ∧ dy + dz ∧ dx on R3 , and let

∂
v = ∂x . Then v Φ is a 1-form, and we can compute it by computing its
58 2. Differential forms

action on an arbitrary vector field w on R3 :

" # " #
∂ ∂
Φ (w) = Φ ,w
∂x ∂x
" #
∂
= (dx ∧ dy + dz ∧ dx) ,w
∂x
" # " #
∂ ∂
= dx dy(w) − dx(w) dy
∂x ∂x
" # " #
∂ ∂
+ dz dx(w) − dz(w) dx
∂x ∂x
= 1 · dy(w) − 0 · dx(w) + 0 · dx(w) − 1 · dz(w)
= (dy − dz)(w).
Therefore,
∂
(dx ∧ dy + dz ∧ dx) = dy − dz.
∂x

With the interior product in hand, the following theorem tells us how to
compute the Lie derivative:
Theorem 2.55. Given a p-form Φ and a vector field v on a manifold M ,
(2.16) Lv φ = v dφ + d(v φ).

The formula (2.16) is called Cartan’s formula for the Lie derivative.
Exercise 2.56. Use the following steps to prove Cartan’s formula for the
Lie derivative:
(a) Prove that the Lie derivative (Definition 2.52) is a derivation of order
zero; i.e., it takes p-forms to p-forms and satisfies the Leibniz rule
Lv (Φ ∧ Ψ) = (Lv Φ) ∧ Ψ + Φ ∧ (Lv Ψ).

(b) Prove that for any vector field v, the operator

(v ) ◦ d + d ◦ (v )
is also a derivation of order zero; i.e., it takes p-forms to p-forms, and for
any forms Φ, Ψ,

[(v ) ◦ d + d ◦ (v )] (Φ ∧ Ψ)
= [(v ) ◦ d + d ◦ (v )] (Φ) ∧ Ψ + Φ ∧ [(v ) ◦ d + d ◦ (v )] (Ψ).

(c) Show that the operators Lv and [(v ) ◦ d + d ◦ (v )] agree on 0-forms

and 1-forms.
2.12. Introduction to the Cartan package for Maple 59

(d) It is shown in [Wil61] that any two derivations of the same degree that
agree on 0-forms and 1-forms must be equal. Use this fact to conclude that
Cartan’s formula holds.

2.12. Introduction to the Cartan package for Maple

The Cartan package is a descendant (and an enhancement) of the Forms

package, originally written by Yungliang Yu of Duke University for per-
forming computations involving differential forms in Maple. The purpose
of this section is to introduce some of the commands in this package and
to illustrate how they may be used in practice. First, you need to down-
load and install the Cartan package; see the Preface for details on how to
do this, and be sure that you install the appropriate version for your pre-
ferred Maple interface (“Worksheet” or “Document” mode). This section
assumes that you are using the Worksheet mode, with the input mode set
to “Maple Input”, so that your input appears verbatim in red print; if you
prefer the Document mode, you must replace the &ˆ command for wedge
product with the alternative command &w.
Once you have Cartan installed, start up Maple and type
> with(Cartan);
This command loads the package; if you have it properly installed, Maple
should return something like the following:

The symbol for wedge product is &ˆ

[CartanKahler, Forder, Form, L, LieDeriv, ScalarForm, Simf, Wdegree,

WedgeProduct, d, hook, makebacksub, mixpar, pick, prolong]
(From now on, we will always assume that this is the first thing that you do
when you start up Maple.)
The Cartan commands that will be most useful for the purposes of this book
are the following:
The Form command. Before you can do any computations with differ-
ential forms, you have to declare them. For instance, the command
> 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. Differential 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 isn’t necessary; anything which hasn’t 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 −1 (this is purely a Maple
convention and has nothing to do with their degree as differential 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 “=1”
can be omitted; e.g., the command
> Form(omega, theta, eta);
declares the objects omega, theta, and eta to be 1-forms.
The &ˆ command. This is the command for wedge product. For in-
stance, the commands
> omega:= x*d(y);
> theta:= y*d(z);
> omega &ˆ 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) &ˆ d(x);
returns

−(d(x) &ˆ d(y))

If you don’t like the order that Maple chooses, you can change it with the
Forder command; see Maple’s help page for the Cartan package for more
details. Furthermore, if you don’t 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 you’re using the Cartan package, don’t 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
" # " # " #
∂ ∂ ∂
f (x, y, z) d(x) + f (x, y, z) d(y) + f (x, y, z) d(z)
∂x ∂y ∂z

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(xˆ2*d(y) + y*d(z));
returns
2x(d(x) &ˆ d(y)) + (d(y) &ˆ 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 different 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 &ˆ eta;
declare that omega, theta, and eta are 1-forms and that the exterior de-
rivative of omega is the 2-form (theta &ˆ eta).
The Simf command. This is the all-purpose simplification command
for differential forms, and you should use it liberally. (Maple fails to make
some fairly obvious simplifications without it.) For instance, the command
> d(xˆ2*d(y) + yˆ2*d(x));
returns
−2y (d(x) &ˆ d(y)) + 2x (d(x) &ˆ d(y))
62 2. Differential forms

and then the command

> Simf(%);
returns
(−2y + 2x)(d(x) &ˆ 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(xˆ2*d(y) + yˆ2*d(x)));
The pick command. The command
> pick(bigform, omega);
where omega is a 1-form, finds all the terms in the expression bigform that
involve wedge products with omega. It then writes bigform as

bigform = form1 &ˆ omega + form2

where form2 contains no terms involving omega and returns form1. For
instance, the command
> pick(x*d(y) &ˆ d(z) + y*d(z) &ˆ d(x) + z*d(x) &ˆ d(y),
d(x));
returns
−z dy + y dz

This command can also be invoked with additional 1-forms as arguments in

order to find all the terms in bigform that involve specific wedge products
of higher degree. So, for example, the command
> pick(bigform, omega1, omega2);
where omega1, omega2 are 1-forms, finds all the terms in the expression
bigform that involve wedge products with omega1 &ˆ 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) &ˆ d(z) + y*d(z) &ˆ d(x) + z*d(x) &ˆ d(y), d(x),
d(y));
simply returns the coefficient z.
2.12. Introduction to the Cartan package for Maple 63

The ScalarForm command. This command takes the scalar coefficient

of each summand in a differential form and returns them all as a list. For
instance, the command
> ScalarForm(x*d(y) &ˆ d(z) + y*d(z) &ˆ d(x)
+ z*d(x) &ˆ d(y));
returns
[z, y, −x]
(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 differential form that you
want to set equal to zero (often something that you computed as d ◦ d of
something else), and you therefore want to set all its coefficients 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 coefficients of these forms produced by the main command.
For instance, the command
> ScalarForm(x*d(y) &ˆ d(z) + y*d(z) &ˆ d(x) + z*d(x) &ˆ d(y),
’terms’);
returns
[z, y, −x]
as before, but now if you type
> terms;
it returns
[(d(x) &ˆ d(y)), (d(z) &ˆ d(x)), (d(z) &ˆ d(y))]
Again, the order may vary, but the order of the coefficients in the first line
will match the order of the forms in the second line.
The makebacksub command. It often happens that we have two dif-
ferent bases for the 1-forms on a manifold, and it’s 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
θ = dz − p dx − q dy, ω 1 = dx, ω 2 = dy,
π1 = dp − ez dy, π2 = dq − ez dx
64 2. Differential forms

for the 1-forms on R5 . (Don’t worry about why you might want to do such
a thing; it has to do with an exterior differential system representing the
partial differential equation
zxy = ez ,
but that’s 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 dθ and ex-
pressing it in terms of the basis (θ, ω 1 , ω 2 , π1 , π2 ), it’s more effective 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 don’t 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 define 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 (θ, ω 1 , ω 2 , π1 , π2 )
to the coordinate basis; e.g., the command
> Simf(subs(sub1, theta));
yields
d(z) − p d(x) − 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 (θ, ω 1 , ω 2 , π1 , π2 ).
So, e.g., the command
> Simf(subs(backsub1, d(z)));
yields
θ + p ω1 + q ω2
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 define
> 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 dθ in terms of the basis
(θ, ω 1 , ω 2 , π1 , π2 ) via the following sequence of commands:
> Simf(subs(sub1, theta));

d(z) − p d(x) − q d(y)

> Simf(d(%));
(d(x)) &ˆ (d(p)) + (d(y)) &ˆ (d(q))

> Simf(subs(backsub1, %));

ω1 &ˆ π1 + ω2 &ˆ π2

And, of course, these commands can be combined into a single command:

> Simf(subs(backsub1, Simf(d(Simf(subs(sub1, theta))))));
(It’s probably not really necessary to put all of those Simf commands in
there, but it doesn’t hurt, and sometimes it prevents problems.)
Having done all this, you can compute—and then assign—the exterior deri-
vatives 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. Differential 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 deriva-
tives, satisfy the identity d◦d = 0. What goes wrong with d(dπ1 ) and d(dπ2 )?
Why does this happen, and what can you do to fix it?
Part 2

Curves and surfaces

in homogeneous spaces
via the method
of moving frames
 Chapter 3

Homogeneous spaces

3.1. Introduction

In the late nineteenth century, there were several different 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 differently than in Euclidean geometry, and even affine and
projective geometries, where lengths and angles weren’t well-defined notions.
In 1872, Felix Klein published a revolutionary treatise on geometry
([Kle93b]; an English translation is available in [Kle93a]), in which he pro-
posed 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 geome-
try, and it continues to influence 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 κ(s), τ (s) on an interval I ⊂ R with κ(s) > 0, there exists a
smooth, unit-speed curve α : I → R3 with curvature κ(s) and torsion τ (s).
Moreover, α is unique up to rigid motion: Any other such curve β differs
from α by a translation and rotation in R3 .

The rigid motions—translations and rotations—are the symmetries of the

Euclidean space R3 : They are exactly the transformations of R3 that pre-
serve 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 orientation-
preserving symmetries of Euclidean space. Reflections 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 Klein’s approach—and to the remainder of this book—is

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 intro-
duced by Élie Cartan in 1935 ([Car35]). In this chapter, we will start with
a detailed discussion of Euclidean space as a homogeneous space; we will
then give some general definitions and explore several other homogeneous
spaces (Minkowski space, equi-affine 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.

Definition 3.3. An inner product on the vector space Rn is a function
·, · : Rn × Rn → R
with the following properties:

(1) Symmetry: For any vectors v, w ∈ Rn , v, w = w, v.

(2) Bilinearity: For any vectors v, w, z ∈ Rn and any scalars a, b ∈ R,
av + bw, z = av, z + bw, z
and
z, av + bw = az, v + bz, w.
(3) Positive definiteness: For any vector v ∈ Rn , v, v ≥ 0, with
equality if and only if v = 0.
Definition 3.4. The vector space Rn endowed with an inner product ·, ·
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 ∈ En , the
length of v is
|v| = v, v,
and the angle between v and w is the angle θ, 0 ≤ θ ≤ π, satisfying
v, w
cos(θ) = .
|v||w|
Exercise 3.5. Let (e1 , . . . , en ) be any basis for Rn , and let ·, · be any
inner product on Rn . Define constants {gij , 1 ≤ i, j, ≤ 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

⎡ ⎤
g11 . . . g1n
⎢ .. ⎥ ,
Ag = ⎣ ... . ⎦
gn1 . . . gnn
and let v, w be represented by the column vectors
⎡ 1⎤ ⎡ 1⎤
a b
⎢ .. ⎥ ⎢ .. ⎥
a = ⎣ . ⎦, b = ⎣ . ⎦,
a n bn
respectively. Show that
v, w = taAg b,
where we identify the 1 × 1 matrix on the right-hand side with its single
real-valued entry.

Definition 3.4 makes it sound as though there might be many different Eu-
clidean spaces corresponding to different inner products, but in fact they
are all equivalent, as the following exercise shows.
Exercise 3.6. Let ·, · be any inner product on Rn , and construct a basis
(e1 , . . . , en ) for Rn inductively as follows:
(Initial step) Choose any nonzero vector v1 ∈ Rn , and set
v1
e1 = .
v1 , v1 
Let (e1 )⊥ denote the orthogonal complement of e1 with respect to ·, ·; i.e.,
(e1 )⊥ = {v ∈ Rn | v, e1  = 0}.
72 3. Homogeneous spaces

(a) Show that (e1 )⊥ ⊂ Rn is a vector space of dimension n − 1. (Hint:

Consider the linear map f : Rn → R defined by f (v) = v, e1 , and compute
the dimension of ker(f ).)
(Inductive step) Suppose that e1 , . . . , ek have been chosen. Let
(e1 , . . . , ek )⊥ = {v ∈ Rn | v, e1  = · · · = v, ek  = 0}.
Let vk+1 be any nonzero vector in (e1 , . . . , ek )⊥ , 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 )⊥ ⊂ Rn is a vector space of dimension n−k. (Hint:
Consider the linear map F : Rn → Rk defined 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 ·, ·.) Therefore, the isomorphism φ : Rn → En that identifies
(e1 , . . . , en ) with the standard basis (e1 , . . . , en ) for En identifies ·, · with
the standard inner product on En .
Exercise 3.7. Let (e1 , . . . , en ) be an orthonormal basis for En . Show that
the dual basis (e∗1 , . . . , e∗n ) for the dual space (En )∗ consists of the linear
mappings e∗i : En → R defined by
e∗i (v) = ei , v
for v ∈ 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 ϕ : En → En pre-
serve the fundamental properties of lengths of vectors and angles between
vectors—specifically, 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 “rigid motions”. Any such

transformation has the form
ϕ(x) = Ax + b,
where A is an element of the special orthogonal group SO(n) and b ∈ 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) × (n + 1)
matrices as follows: Let
(  )
1 t0
E(n) = : A ∈ SO(n), b ∈ E .
n
b A
Here A is an n × n matrix, b is an n × 1 column vector, and t0 represents
a 1 × n row of 0’s, i.e., the transpose of the n × 1 column vector 0. If we

1
represent a vector x ∈ E by the (n + 1)-dimensional column vector
n ,
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 defining 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 ∈ En , it is natural to ask: Which elements of E(n) leave x

fixed? In other words, which transformations ϕ ∈ E(n) have the property
that ϕ(x) = x? The set of such transformations is a subgroup of E(n),
called the isotropy group of the point x ∈ En and denoted Hx .
 
1 t0
This question is easiest to answer when x = 0: An element fixes the
b A
point x = 0 if and only if b = 0. Thus, the isotropy group H0 of the point
1 t0
0 ∈ 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 ∈ En ? It seems reasonable to expect that there
shouldn’t 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 ∈ En is also isomorphic to SO(n). We can define an explicit
1 t0
isomorphism φ : H0 → Hx as follows. Let tx = (where I represents
x I
74 3. Homogeneous spaces

the n×n identity matrix), so that tx represents the translation tx (y) = y+x.
Then for any element h ∈ H0 , define
φ(h) = tx ht−1
x ∈ E(n).

The following exercise shows that φ is indeed an isomorphism from H0 to Hx :

Exercise 3.9. (a) Show that for any h ∈ H0 , we have φ(h) ∈ Hx ; i.e.,
φ(h)(x) = x. Therefore, φ is a map from H0 to Hx .
(b) Show that φ : H0 → Hx is a homomorphism; i.e., show that for any two
elements h1 , h2 ∈ H0 , we have
φ(h1 h2 ) = φ(h1 )φ(h2 ).

(c) Show that φ : H0 → Hx is injective; i.e., show that if φ(h) is the identity
element in Hx , then h must be the identity element in H0 .
(d) Show that φ : H0 → Hx is surjective; i.e., show that for any h̃ ∈ Hx ,
there exists h ∈ H0 with φ(h) = h̃.

The isomorphism defined by φ is called conjugation; we describe it using the

notation
Hx = tx H0 t−1
x ,

and we say that all the isotropy groups Hx ⊂ E(n) are conjugate in E(n).
Moreover, they are all isomorphic to SO(n).

Definition 3.10. The left coset tx H0 is the subset of E(n) defined by

tx H0 = {tx h | h ∈ 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 ∈ En :

*Exercise 3.11. (a) Show that

(  )
1 t0
tx H0 = : A ∈ 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 ϕ ∈ 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 ∼
= E(n)/SO(n).
The set E(n)/SO(n) can be given a manifold structure so that this corre-
spondence becomes a diffeomorphism.

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 .
Definition 3.12. An (oriented) orthonormal frame f on En is a list of
vectors f = (x; e1 , . . . , en ), where x ∈ 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 ∈ 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 define a projection map π : E(n) → En by
π(x; e1 , . . . , en ) = x.
This map is differentiable, and the fiber over any point x ∈ En is the set of all
oriented, orthonormal frames based at x. SO(n) acts freely and transitively
on each fiber, and so this map gives an explicit description of E(n) as a
principal bundle over En with fiber group SO(n):

SO(n) - E(n)
π
?
En ∼
= 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 derivatives—specifically, the
76 3. Homogeneous spaces

exterior derivatives—of 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 ) → En is just the projection map π.)

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 derivatives—which we will get to
shortly—take 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 ∼ = En , which allow us to regard (e1 , . . . , en )
as E -valued functions.
n

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 differentials of the maps
x, ei : F (En ) → En .) For any point f = (x; e1 , . . . , en ) ∈ F (En ), these are
maps

dx : Tf F (En ) → Tx En ,
dei : Tf F (En ) → Tei En ∼= Tei (Tx En ) ∼
= 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 coefficients
are ordinary scalar-valued 1-forms (cf. Exercise 2.40).
These considerations lead us to define scalar-valued 1-forms (ω i , ωji ) on
F (En ) by the equations

dx = ei ω i ,
(3.1)
dei = ej ωij ,

where 1 ≤ i, j, ≤ 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 first equation may be written as the matrix product
⎡ 1⎤
ω
⎢ . ⎥
(3.2) dx = e1 · · · en ⎣ .. ⎦ ,
ωn

and it wouldn’t 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 (ω i , ωji ) on F (En ) are defined by the property
that for any f = (x; e1 , . . . , en ) ∈ F (En ), v ∈ Tf F (En ),

dx(v) = ei ω i (v) ∈ Tx En ,
dei (v) = ej ωij (v) ∈ Tei (Tx En ) ∼
= Tx En .

Remark 3.15. Note that the 1-forms (ω i , ωji ) are not well-defined on the
base space En since they are defined 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 ) ∈ F (En ), let A ∈ SO(n) denote the matrix


A = e1 · · · en .

Then equation (3.2) can be written as

⎡ ⎤
ω1
⎢ ⎥
dx = A ⎣ ... ⎦ .
ωn

Therefore,

⎡ ⎤ ⎡ 1⎤
ω1 dx
⎢ .. ⎥ −1 −1 ⎢ .. ⎥
(3.3) ⎣ . ⎦ = A dx = A ⎣ . ⎦ .
ωn dxn

From this expression, it is clear that the 1-forms (ω 1 , . . . , ω 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 (ω 1 , . . .,

ω n ) are not well-defined on En , they are linearly independent, linear com-
binations of the 1-forms (dx1 , . . . , dxn )—or, more precisely, of the pullbacks
(π ∗ (dx1 ), . . ., π ∗ (dxn )) of (dx1 , . . . , dxn ) to F (En ). Therefore, (dx1 , . . . , dxn )
can be expressed as linear combinations of (ω 1 , . . . , ω n ), with coefficients
that are functions on F (En )—and indeed, the first 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
⎡ 1 ⎤
ω1 · · · ωn1
 ⎢ .. ⎥
dA = de1 · · · den = e1 · · · en ⎣ ... . ⎦
ω1 · · · ωnn
n
⎡ 1 ⎤
ω1 · · · ωn1
⎢ .. ⎥ .
= A ⎣ ... . ⎦
ω1n · · · ωnn

Therefore,
⎡ ⎤
ω11 · · · ωn1
⎢ .. .. ⎥ = A−1 dA.
(3.4) ⎣ . . ⎦
ω1n · · · ωnn

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
   
cos(θ) − sin(θ)
e1 = , e2 = ,
sin(θ) cos(θ)

where θ is the angle between e1 and the standard basis vector e1 = ∂x∂ 1 .
Thus, θ may be regarded as a local coordinate on the 1-dimensional fibers
of the orthonormal frame bundle F (E2 ). Now, write
 
 cos(θ) − sin(θ)
A = e1 e2 = .
sin(θ) cos(θ)

Then (ω 1 , ω 2 ) are given by

 
ω1
= A−1 dx
ω2
   1
cos(θ) sin(θ) dx
(3.5) =
− sin(θ) cos(θ) dx2
 
cos(θ) dx1 + sin(θ) dx2
= ,
− sin(θ) dx1 + cos(θ) dx2
3.3. Orthonormal frames on Euclidean space 79

and the 1-forms (ωji ) are given by

 1 
ω1 ω21
= A−1 dA
ω12 ω22
  
cos(θ) sin(θ) − sin(θ) dθ − cos(θ) dθ
(3.6) =
− sin(θ) cos(θ) cos(θ) dθ − sin(θ) dθ
 
0 −dθ
= .
dθ 0

The 1-forms (ω 1 , . . . , ω n ) are often called the dual forms on the orthonormal
frame bundle F (En ) because they have the property that at any point f ∈
F (En ),

i i 1, i = j,
(3.7) ω (ej ) = δj =
0, i = j.
Exercise 3.18. Verify equations (3.7) by direct computation.

The dual forms also have the property that ω i (v) = 0 for any vector v ∈
T F (En ) that is tangent to the fibers of the projection π : F (En ) → En . The
technical way to say this is that the pullback of ω i to any fiber π −1 (x0 ) of π
via the inclusion map ι : π −1 (x0 ) → F (En ) vanishes.
Exercise 3.19. Prove this statement. (Hint: Let α : I → F (En ) be a curve
tangent to v. Since v is tangent to some fiber π −1 (x0 ) ⊂ F (En ), you can
assume that α(I) ⊂ π −1 (x
 0 ). So, what is the value of x at any point α(t)?
d
Now compute dx(v) = dt t=0 x(α(t)).)

Forms with this property (i.e., their pullbacks to each fiber of π vanish) are
called semi-basic for the projection π; for this reason, the dual forms are also
sometimes called the semi-basic forms on F (En ). The pullbacks of (ωji ), on
the other hand, form a basis for the 1-forms on each fiber of π. They are
called the connection forms on F (En ).
Now for the fun part: Start differentiating! In order to compute the exterior
derivatives of the dual forms and connection forms, we need to differentiate
equations (3.1).
*Exercise 3.20. Differentiate equations (3.1) (taking the second equation
into account!) and do some careful index-juggling to obtain
0 = ei (ωji ∧ ω j + dω i ),
0 = ei (ωki ∧ ωjk + dωji ).
80 3. Homogeneous spaces

Conclude that these 1-forms satisfy the Cartan structure equations

dω i = −ωji ∧ ω j ,
(3.8)
dωji = −ωki ∧ ωjk .
(Hint: Recall the Leibniz rule for differentiating p-forms, and observe that
the vector field ei is a vector-valued 0-form. Note that if we wrote the terms
in (3.1) in the other order—and some authors do!—the Leibniz rule would
lead to different 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 haven’t 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  = δij =
0, i = j.
*Exercise 3.22. Differentiate equations (3.9) and conclude that the con-
nection forms (ωji ) are skew-symmetric in their indices; that is, they have
the property that ωij = −ωji .
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 fits 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
e(n) = : B ∈ so(n), b ∈ E .
n
b B
(Recall that B ∈ so(n) if and only if B is a skew-symmetric n × n matrix.)

We can define an e(n)-valued 1-form ω on E(n) as follows: Recall that for

h ∈ E(n), the left multiplication map Lh : E(n) → E(n) is defined by
Lh (g) = hg,
and for any g ∈ E(n), we have the differential
(Lh )∗ : Tg E(n) → Thg E(n).
3.3. Orthonormal frames on Euclidean space 81

Now, for any g ∈ E(n), v ∈ Tg E(n), define

ω(v) = (Lg−1 )∗ (v).
Since (Lg−1 )∗ maps Tg E(n) to TI E(n) = e(n), we have ω(v) ∈ e(n) for all
v ∈ T E(n). The form ω is called the Maurer-Cartan form on E(n).
*Exercise 3.25. Prove that the Maurer-Cartan form ω on E(n) is left-
invariant; i.e., show that for any h ∈ E(n), L∗h ω = ω. (Hint: Let g ∈ E(n),
v ∈ Tg E(n), and compute
(L∗h ω)(v) = ω((Lh )∗ (v)).
Try to keep track of which tangent space each object lives in!)

While this definition 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) ω = g −1 dg.
This notation requires some explanation. The variable g here essentially
denotes the identity map g : E(n) → 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) × (n + 1) matrix representation. Specifically,
for any f = (x; e1 , . . . , en ) ∈ E(n),
 
1 0 ··· 0
(3.11) g(f ) = .
x e1 · · · 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 difference in notation is meant to
suggest that f0 is an element of the abstract group and g0 is its matrix
representation.) Differentiating (3.11) shows that the mapping
dg : Tf0 E(n) → Tg0 E(n)
can be written as the matrix-valued 1-form
 
0 0 ··· 0
(3.12) dg = .
dx de1 · · · den
Since g is essentially the identity map, the same is true for dg: It simply
identifies any tangent vector v ∈ Tf0 E(n) with its matrix representation as
a tangent vector to the matrix group E(n) at g0 . So
dg(v) = v ∈ Tg0 E(n).
The left multiplication by g −1 , applied to a vector v ∈ Tg0 E(n), then means
to multiply the matrix representation for v by the matrix representation
for g0−1 .
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 “g −1 dg(v)”
is to act on the vector dg(v) = v ∈ Tg0 E(n) by the differential of the left-
multiplication map Lg−1 . (Note that this is our original definition for ω!)
0
But because everything here is matrix-valued, (Lg−1 )∗ (v) is, in fact, given
0
by the product of the matrices g0−1 and v.

As you might suspect, it is no accident that we have used the letter ω 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
⎡ ⎤
0 0 ··· 0
⎢ω1 ω1 · · · ω1 ⎥
⎢ 1 n⎥
ω=⎢ . .. .. ⎥ .
⎣ .. . . ⎦
ω n ω1n · · · ωnn
(Hint: Use equations (3.1), (3.11), (3.12) and the fact that ω = g −1 dg is
equivalent to dg = gω.) Write out this matrix explicitly in the case n = 3.
(b) Use the skew-symmetry of the forms (ωji ) to confirm that for any v ∈
T E(n), we have ω(v) ∈ e(n). (This is what it means to say that ω is
“e(n)-valued”.)

Because the matrix-valued Maurer-Cartan form ω contains the scalar-valued

dual forms and connection forms as its entries, the dual forms and connec-
tion forms are collectively referred to as “the Maurer-Cartan forms”. The
fact that ω is left-invariant implies that the Maurer-Cartan forms are left-
invariant 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
dω = −ω ∧ ω.
(The wedge product of two matrices of 1-forms is computed just like the
ordinary matrix product, substituting wedge product for ordinary multipli-
cation of the appropriate entries.) Note that, despite the skew-symmetry of
the wedge product for scalar-valued 1-forms, the wedge product of a matrix-
valued 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 ∈ En . This amounts to choosing
a section σ : En → F (En ) of the orthonormal frame bundle π : F (En ) → En ,
also known as an orthonormal frame field on En . The pullbacks (σ ∗ (ω i ),
σ ∗ (ωji )) are then 1-forms on En . In order to reduce notational clutter, we
will denote these pullbacks by (ω̄ i , ω̄ji ). Show that if we set

A(x) = e1 (x) . . . en (x) ,
then the dual forms and connection forms associated to this frame field are
given by
⎡ 1⎤ ⎡ 1⎤ ⎡ ⎤
ω̄ dx ω̄11 . . . ω̄n1
⎢ .. ⎥ −1 ⎢ .. ⎥ ⎢ .. .. ⎥ = A(x)−1 dA(x).
⎣ . ⎦ = A(x) ⎣ . ⎦ , ⎣ . . ⎦
ω̄ n dxn ω̄1n . . . ω̄nn
(Compare with equations (3.3), (3.4).) Show that if we write
⎡ ⎤
  0 0 · · · 0
1 t0 ⎢ ω̄ 1 ω̄ 1 · · · ω̄ 1 ⎥
⎢ 1 n⎥
g(x) = , ω̄ = ⎢ . .. .. ⎥ ,
x A(x) ⎣ .
. . . ⎦
ω̄ ω̄1 · · · ω̄nn
n n

then these equations imply that

ω̄ = g(x)−1 dg(x).
Remark 3.30. The dual forms (ω i ) and connection forms (ωji ) are all lin-
early independent “upstairs” on the frame bundle F (En ), but this is no
longer the case when these forms are pulled back “downstairs” to En via a
section, as in Exercise 3.29. The pulled-back dual forms (ω̄ i ) are linearly
independent, linear combinations of (dx1 , . . . , dxn ), but the pulled-back con-
nection forms (ω̄ji ) are no longer linearly independent from the dual forms;
indeed, they may be expressed as linear combinations of (ω̄ 1 , . . . , ω̄ n ), which
form a basis for the 1-forms on En .
Exercise 3.31. Consider E3 (minus the z-axis) with cylindrical coordinates
(r, θ, z), related to the usual Euclidean coordinates (x, y, z) by
x = r cos(θ), y = r sin(θ), z = z.
Apply the result of Exercise 3.29 to the cylindrical frame field
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
cos(θ) − sin(θ) 0
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
e1 = ⎢ ⎥
⎣ sin(θ) ⎦ , e2 = ⎢ ⎥
⎣ cos(θ) ⎦ , e3 = ⎢
⎣0⎦
⎥

0 0 1
84 3. Homogeneous spaces

to compute the dual forms (ω̄ i ) and the connection forms (ω̄ji ) for this frame
field. Show by direct computation that these forms satisfy the structure
equations (3.8).
Exercise 3.32. Repeat Exercise 3.31 for the spherical frame field
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
cos(ϕ) cos(θ) − sin(θ) − sin(ϕ) cos(θ)
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
e1 = ⎢ ⎥
⎣ cos(ϕ) sin(θ) ⎦ , e2 = ⎢ ⎥
⎣ cos(θ) ⎦ , e3 = ⎢ ⎥
⎣ − sin(ϕ) sin(θ) ⎦ ,
sin(ϕ) 0 cos(ϕ)
where (ρ, ϕ, θ) are spherical coordinates on E3 (minus the z-axis), related to
the usual Euclidean coordinates (x, y, z) by
x = ρ cos(ϕ) cos(θ), y = ρ cos(ϕ) sin(θ), z = ρ sin(ϕ).

3.4. Homogeneous spaces

Let’s take a moment to summarize our work so far: We started with Eu-
clidean space En , i.e., the vector space Rn endowed with an inner product
·, ·. 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:
Definition 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 con-
sequence 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 ∈ G/H, the isotropy group Hx of x is the group
Hx = gHg −1 ,
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 π : G → G/H
3.5. Minkowski space 85

defined by
π(g) = gH
describes G as a principal bundle over G/H with fiber H:

H - G

π
?
G/H.

Just as we identified 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 “frames” on the space G/H. The Maurer-Cartan form on G is
defined by
ω(v) = (Lg−1 )∗ (v)
for any v ∈ Tg G, and if G is a matrix group, then we can write
ω = g −1 dg.
Either way, ω takes values in the Lie algebra g of G, and it satisfies the
Maurer-Cartan equation:
(3.13) dω = −ω ∧ ω.
This equation will turn out to play a crucial role in the geometry of sub-
manifolds of the space G/H.

*Exercise 3.34. Suppose that G is a matrix Lie group, so that ω = g −1 dg.

Prove directly that ω satisfies the Maurer-Cartan equation (3.13). (Hint:
Show by differentiating the equation gg −1 = I that d(g −1 ) = −g −1 dg g −1 .)

In the remainder of this chapter, we will explore some other examples of

homogeneous spaces: Minkowski space, equi-affine 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 Ein-
stein’s theory of special relativity. In Minkowski’s original formulation, the
three dimensions of Euclidean space are combined with a single time dimen-
sion 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 prod-
uct structure: Instead of a positive definite inner product, Minkowski space
is equipped with an indefinite 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 ≥ 0 with the property that for
any v = v i ei ∈ Rn ,
Q(v, v) = (v 1 )2 + · · · + (v p )2 − (v p+1 )2 − · · · − (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 definite form Q has signature (n, 0), whereas Q is called indefinite
if p and q are both greater than zero.
Definition 3.35. The (1 + n)-dimensional Minkowski space M1,n is the
vector space Rn+1 endowed with a nondegenerate, symmetric bilinear form
·, · 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 = [gαβ ] = [eα , eβ ]
that describes the Minkowski inner product ·, · in terms of this basis has
one positive eigenvalue and n negative eigenvalues.
Remark 3.36. Even when n has a fixed value, say n = 3, the Minkowski
space M1,3 is still often referred to as “(1+3)-dimensional Minkowski space”
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 (α, β, 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 ·, · be any Minkowski inner product on Rn+1 . Con-
struct a basis (e0 , . . . , en ) for Rn+1 as follows: Choose any nonzero vector
v0 ∈ Rn+1 with v0 , v0  > 0 , and set
v0
e0 = .
v0 , v0 
3.5. Minkowski space 87

Let (e0 )⊥ denote the orthogonal complement of e0 with respect to ·, ·; i.e.,
(e0 )⊥ = {v ∈ Rn+1 | v, e0  = 0}.

(a) Show that for any nonzero vector v ∈ (e0 )⊥ ,

v, v < 0.
(Hint: Let (v1 , . . . , vn ) be any basis for (e0 )⊥ . 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 × n matrix with all negative eigenvalues. Use this
to prove the result.)
(b) By part (a), the restriction of ·, · to (e0 )⊥ is negative definite. Con-
struct an orthonormal basis (e1 , . . . , en ) for (e0 )⊥ as in Exercise 3.6. (For
purposes of this construction, there is essentially no difference between neg-
ative definite and positive definite inner products.)
(c) Show that with respect to the basis (e0 , . . . , en ), the (gαβ ) are given by
⎧
⎪
⎨1, α = β = 0,
(3.14) gαβ = eα , eβ  = −1, α = β = 1, . . . , n,
⎪
⎩
0, α = β.
We say that the basis (e0 , . . . , en ) is orthonormal with respect to the inner
product ·, ·, 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:
Definition 3.40. A nonzero vector v ∈ M1,n is called

• timelike if v, v > 0;

• spacelike if v, v < 0;
• lightlike or null if v, v = 0.

A timelike vector v ∈ M1,n is called future-pointing if v, e0  > 0 and past-

pointing 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 ∈ M1,n with spacelike separation (i.e., for
which the vector q2 − q1 is spacelike) represent events that cannot have any
causal impact on each other because they cannot be connected by a smooth
timelike curve.
Definition 3.42. The Minkowski norm v of a nonzero vector v ∈ M1,n
is defined to be

(1) v = v, v if v is timelike;

(2) v = −v, v if v is spacelike;
(3) v = 0 if v is lightlike.

(Note that v ≥ 0 for all v ∈ M1,n .)

Exercise 3.43. For n ≥ 2 and any λ ∈ R, show that the “sphere” Sλ
consisting of all vectors v ∈ M1,n satisfying v, v = λ is

(1) a hyperboloid of two sheets when λ > 0;

(2) a cone when λ = 0;
(3) a hyperboloid of one sheet when λ < 0.

(See Figure 3.1; note that the e0 -axis is drawn as the vertial axis.)

Figure 3.1. Minkowski “spheres” Sλ with λ = 1, λ = 0, λ = −1

3.5.2. Symmetries and isotropy groups. Let M1,n denote the Minkow-
ski 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 ∈ GL(n + 1). Show that

Av, Av = v, v
for every v ∈ M1,n if and only if A is an element of the Lie group

O(1, n) =
⎧ ⎡ ⎤ ⎡ ⎤⎫
⎪
⎪ 1 0 ··· 0 1 0 ··· 0 ⎪
⎪
⎪
⎨ ⎢0 −1 · · · ⎥ ⎢0 −1 · · · ⎥⎬⎪
⎢ 0 ⎥ ⎢ 0 ⎥
A ∈ GL(n + 1) : tA ⎢ . .. ..⎥ A = ⎢ .. .. .. ⎥ .
⎪
⎪ ⎣ .. . .⎦ ⎣. . . ⎦⎪⎪
⎪
⎩ ⎪
⎭
0 0 ··· −1 0 0 ··· −1
Furthermore, show that O(1, n) consists of all matrices of the form

A = e 0 · · · 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 transfor-
mations 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 “proper, orthochronous Lorentz group” or, more
succinctly, the “restricted Lorentz group”. 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(θ) sinh(θ)
+
SO (1, 1) = :θ∈R .
sinh(θ) cosh(θ)
What are the other components of O(1, 1)?
(b) Show that the light cone
2t 0 1 3
[x , x ] ∈ M1,1 | (x0 )2 − (x1 )2 = 0
and the hyperbolas
2t 0 1 3
[x , x ] ∈ M1,1 | (x0 )2 − (x1 )2 = ±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 ∈ SO+ (1, 1), for various values
of the group parameter θ.
Exercise 3.47. Show that the Lie algebra so(1, n) of O(1, n) is given by
so(1, n) =
⎧ ⎡ ⎤ ⎡ ⎤ ⎫
⎪
⎪ 1 0 ··· 0 1 0 ··· 0 ⎪
⎪
⎪
⎨ ⎢0 −1 · · · ⎥ ⎢0 −1 · · · ⎥ ⎪
⎬
⎢ 0 ⎥ ⎢ 0 ⎥
B ∈ M(n+1)×(n+1) : tB ⎢ . .. ..⎥ + ⎢ .. .. .. ⎥B = 0 .
⎪
⎪ ⎣ .. . .⎦ ⎣. . . ⎦ ⎪
⎪
⎪
⎩ ⎪
⎭
0 0 ··· −1 0 0 ··· −1
In particular, this implies that the entries (bαβ ) of B satisfy the relations
b00 = 0,
bi0 = b0i , i = 1, . . . , n,
bji = −bij , 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
ϕ(x) = Ax + b,
where A ∈ SO + (1, n) and b ∈ M1,n . These transformations form a Lie
group, called the Poincaré group M (1, n), which can be represented as
(  )
1 t0
M (1, n) = : A ∈ SO (1, n), b ∈ M
+ 1,n
.
b A

 if we represent a vector x ∈ M by the (n + 2)-

As in the Euclideancase, 1,n

1
dimensional vector , then elements of M (1, n) act by matrix multiplica-
x
tion:     
1 t0 1 1
= .
b A x Ax + b

Once again, we ask: Given a vector x ∈ M1,n , what is the isotropy group of
x?
*Exercise 3.48. Show that:
(a) The isotropy group H0 of 0 ∈ M1,n is
( t  )
1 0
H0 = : A ∈ SO (1, n) ∼
+
= SO + (1, n).
0 A
3.5. Minkowski space 91

(b) The isotropy group Hx of any other point x ∈ M1,n is

Hx = tx H0 t−1
x ,
 t0

1
where tx = represents the translation tx (y) = y + x.
x I
(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. Orthonor-

mal frames on M1,n are defined as follows.
Definition 3.49. An orthonormal frame f on M1,n is a list of vectors f =
(x; e0 , . . . , en ), where x ∈ 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 orthonor-
mal frames may be regarded as the Poincaré group M (1, n) via the one-to-
one correspondence
 
1 0 ··· 0
g(x; e0 , . . . , en ) = .
x e0 · · · en
We can define a projection map π : M (1, n) → M1,n by
π(x; e0 , . . . , en ) = x;
the fiber of this map is the set of all orthonormal frames based at x.
SO + (1, n) acts freely and transitively on each fiber, and so this map gives
an explicit description of M (1, n) as a principal bundle over M1,n with fiber
group SO+ (1, n):

SO + (1, n) -M (1, n)

π
?
M1,n ∼
= 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 defined 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 difference is that with the indefinite inner
product, the connection forms (ωβα ) are no longer quite skew-symmetric.
92 3. Homogeneous spaces

*Exercise 3.50. Differentiate equations (3.14) and conclude that

(1) ω00 = 0;
(2) ω0i = ωi0 for i = 1, . . . , n;

(3) ωij = −ωji for i, j = 1, . . . , n.

*Exercise 3.51. Prove that the Lie algebra of M (1, n) is
(  )
0 t0
m(1, n) = : B ∈ so(1, n), b ∈ M 1,n
.
b B

Exercise 3.52. Repeat Exercise 3.23 for M1,2 .

3.6. Equi-affine space

3.6.1. Volume forms. Affine geometry is the study of geometric proper-

ties that are preserved by the action of invertible linear transformations and
translations on Rn . There is less structure in affine geometry than in Eu-
clidean or Minkowski geometry; there are, for instance, no invariant notions
of lengths of vectors or angles between vectors. Affine transformations do,
however, preserve collinearity of points, and the notion of parallel lines is
still well-defined in affine geometry.
In this book, we will study equi-affine geometry (also known as special affine
geometry), which has slightly more structure than general affine geometry,
in that we endow Rn with a volume form.
Definition 3.53. A volume form on Rn is a nonzero element dV of the
1-dimensional vector space Λn (Rn )∗ . Equivalently, dV is a skew-symmetric,
multilinear function
n copies
4 56 7
dV : Rn × · · · × Rn → 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 , defined by
the property that
dV0 (e1 , . . . , en ) = 1,
where (e1 , . . . , en ) is the standard basis for Rn .
3.6. Equi-affine space 93

(a) Show that for any n vectors (v1 , . . . , vn ) ∈ Rn ,


dV0 (v1 , . . . , vn ) = det v1 · · · vn .

(b) Let dV be any other volume form on Rn . Show that

dV = λ dV0
for some nonzero real number λ. (Hint: Let λ = 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 · · · wn = A v1 · · · vn
for some matrix A ∈ SL(n). (Hint: Consider the matrix
 −1
w1 · · · wn v1 · · · vn .)

(d) Let dV = λ dV0 . Show that the isomorphism φ : Rn → Rn defined by

φ(v) = λ(1/n) v
identifies dV with the standard volume form dV0 , in the sense that
dV (v1 , . . . , vn ) = dV0 (φ(v1 ), . . . , φ(vn )).
(If λ < 0 and n is even, first perform a linear transformation that inter-
changes e1 and e2 ; this will have the effect of reversing the sign of λ.)
Therefore, all volume forms on Rn are equivalent.
Definition 3.55. The vector space Rn endowed with a volume form dV is
called equi-affine space and is denoted An .
Remark 3.56. The volume form is not really part of the definition of equi-
affine space; where it comes into play is when we decide what transforma-
tions to allow in our symmetry group. Affine transformations are those
that preserve the linear structure of An ; equi-affine (or special affine) trans-
formations are those that preserve the volume form as well. It is most
common—although not universal—to study geometric properties that are
preserved under the group of equi-affine transformations, and this is the ap-
proach that we will take. It is fairly common to use the terms “affine space”
and “affine transformations” to refer to the equi-affine 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 ) ∈ Rn ,

dV (v1 , . . . , vn ) = det v1 · · · vn .
94 3. Homogeneous spaces

3.6.2. Symmetries and isotropy groups. Now we ask: What are the
symmetries of equi-affine space? (It turns out that “orientation-preserving”
is an unneccesary stipulation here, because any volume-preserving transfor-
mation necessarily preserves orientation.) Since equi-affine 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 ∈ 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 ∈ SL(n).

Since the volume form is also preserved by translation, the full symmetry
group of An consists of all transformations ϕ : An → An of the form
ϕ(x) = Ax + b,
where A ∈ SL(n) and b ∈ An . These transformations form a Lie group,
called the equi-affine group or special affine group A(n), which can be rep-
resented as (  )
1 t0
A(n) = : A ∈ SL(n), b ∈ A .
n
b A

Now, given a vector x ∈ An , what is the isotropy group of x?

*Exercise 3.58. Show that:
(a) The isotropy group H0 of 0 ∈ An is
( t  )
1 0
H0 = : A ∈ SL(n) ∼= SL(n).
0 A

(b) The isotropy group Hx of any other point x ∈ An is

Hx = tx H0 t−1
x ,
 t0

1
where tx = represents the translation tx (y) = y + x.
x I
(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 defined as follows:
Definition 3.59. A unimodular frame f on An is a list of vectors f =
(x; e1 , . . . , en ), where x ∈ An and (e1 , . . . , en ) is a basis for the tangent
3.6. Equi-affine 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 unimodu-
lar frames may be regarded as the equi-affine group A(n) via the one-to-one
correspondence
 
1 0 ··· 0
g(x; e1 , . . . , en ) = .
x e1 · · · en
We can define a projection map π : A(n) → An by
π(x; e1 , . . . , en ) = x;
the fiber of this map is the set of all unimodular frames based at x. SL(n)
acts freely and transitively on each fiber, and so this map gives an explicit
description of A(n) as a principal bundle over An with fiber group SL(n):

SL(n) - A(n)
π
?
An ∼
= 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 defined 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 (ωji ).
There is, however, one relation among the (ωji ) that comes from the uni-
modular condition 
det e1 · · · en = 1.
We can differentiate this condition more easily if we express it as
(3.15) e1 ∧ · · · ∧ en = e1 ∧ · · · ∧ 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 ∧ · · · ∧ en = det e1 · · · en e1 ∧ · · · ∧ en .
*Exercise 3.61. Differentiate equation (3.15) and conclude that the con-
nection forms (ωji ) satisfy the trace condition ωii = 0. (Hint: Don’t 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
a(n) = : B ∈ sl(n), b ∈ A .
n
b B
(Recall that B ∈ 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 perspective—for 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 im-
age. In this geometry, lines that are parallel in 3-dimensional space may
appear to intersect “at infinity”, but this intersection point may appear at
some finite distance in the 2-dimensional projection. Projective transfor-
mations of 2-dimensional space are the result of changing the focal point
in 3-dimensional space. Projective transformations are more general than
affine transformations; among other things, they can map points at a finite
distance from the origin to points “at infinity”, and vice versa. Because of
this, n-dimensional projective space must be constructed so as to include
not only the space Rn of finite-distance points, but also a point “at infinity”
for every direction in Rn .
Recall the definition from Example 1.9:
Definition 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 “Rn plus points at infinity”,
we will use homogeneous coordinates [x0 : · · · : xn ] (cf. Example 1.9) to
represent a point in Pn . Consider the dense, open subset V0 of Pn given by
V0 = {[x0 : · · · : xn ] ∈ Pn | x0 = 0}.
Any point [x] ∈ V0 can be written as
 
x1 xn
[x] = [x : · · · : x ] = 1 : 0 : · · · : 0
0 n
x x
3.7. Projective space 97

!
x1 n
and so may be represented by the point x0
, . . . , xx0 ∈ Rn . We will denote
this point by x̄ = (x̄1 , . . . , x̄n ); these are referred to as affine 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 ⊂ Pn

On the other hand, any point [x] ∈ Pn \ V0 has the form

[x] = [0 : x̄1 : · · · : x̄n ].
This point occurs as a limit point of the line
[x(t)] = [1 : tx̄1 : · · · : tx̄n ]
in V0 as t → ±∞ and so may be thought of as a “point at infinity” corre-
sponding to the direction of the vector x̄ = t[x̄1 , . . . , x̄n ] ∈ Rn .
*Exercise 3.65. Show that the set Pn \ V0 may be naturally identified with
Pn−1 . So inductively, we can think of Pn as
Pn = Rn ∪ Rn−1 ∪ · · · ∪ R ∪ {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 Definition 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-defined
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 ∈ GL(n + 1) fixes every line in Rn+1 —and therefore acts as the identity
transformation on Pn —if and only if A = λI for some λ = 0. Moreover, any
two matrices A, B ∈ GL(n + 1) induce the same transformation on Pn if and
only if A = λB for some λ = 0. Therefore, the most natural choice for the
symmetry group of Pn is the quotient group GL(n + 1)/ ∼, where A ∼ B if
and only if A = λB for some λ = 0.

*Exercise 3.66. Show that GL(m)/ ∼ is isomorphic to

(1) SL(m) if m is odd;

(2) a semidirect product (SL(m)/{±I})  {±1} if m is even. (The

quotient of SL(m) by {±I} reflects the fact that −I is an element
of SL(m), and it acts trivially on Pn . The semidirect product with
{±1} reflects the fact that the sign of the determinant is fixed under
scaling, and so this group has two components: one whose elements
have determinant 1 and one whose elements have determinant −1.)

Remark 3.67. The group GL(m)/ ∼ is called the projective general linear
group, denoted P GL(m); similarly, the group SL(m)/ ∼ is called the pro-
jective 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)/{±I})  {±1} of Exercise 3.66, while P SL(m) is the
proper subgroup SL(m)/{±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 per-
spective. The following exercise will illustrate how this works in the case
of P1 .

*Exercise 3.68. Consider the plane R2 with coordinates (x̄, ȳ), and let the
x̄-axis represent the open set

V0 = {[x0 : x1 ] ∈ P1 | x0 = 0}
1
via the identification x̄ = xx0 . Suppose that an object lies along the line L
with equation ȳ = mx̄+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 x̄-axis. (See Figure 3.3.) Observer p sees the point r as

sq
p s
Q
Q
```QQ
``Q
`s``
Q `
r Q ``` L
Q
Q
Q
Q
s Qs
T (x̄) x̄

Figure 3.3. Projective transformation of P1

lying at (x̄, 0), and observer q sees r as lying at (T (x̄), 0). Use the following
steps to compute the transformation T (x̄):
(a) Show that the point r of intersection between L and the line joining the
point (x̄, 0) to p has coordinates
" #
(p2 − b)x̄ + bp1 mp2 x̄ + bp2
r= , .
mx̄ + p2 − mp1 mx̄ + p2 − mp1

(b) Show that the line joining q to r intersects the x̄-axis at the point
(T (x̄), 0), where

αx̄ + β
(3.16) T (x̄) =
γ x̄ + δ

for some constants α, β, γ, δ with αδ − βγ = 0. (You can keep up with the

precise constants if you want, but they’re kind of a mess!) The map (3.16)
is called a linear fractional transformation.
(c) For what value(s) of x̄ would it make sense to define T (x̄) = ∞? What
value would you assign to T (∞)? Can you see how to interpret these as-
signments in terms of the observers in Figure 3.3?
1
(d) Recall that x̄ = xx0 , where [x0 : x1 ] are homogeneous coordinates on P1 .
Show that T corresponds to the map

T ([x0 : x1 ]) = [δx0 + γx1 : βx0 + αx1 ]

100 3. Homogeneous spaces

on P1 , which in turn corresponds to the linear transformation

    
x0 δ γ x0
T8 =
x1 β α x1
 
δ γ
on R , where
2 ∈ GL(2).
β α
 
δ γ
(e) Show that the matrix can be modified so as to have determi-
β α
nant equal to ±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 ⊂ P2 be the open set
V0 = {[x0 : x1 : x2 ] ∈ P2 | x0 = 0},
with affine coordinates
" #
x1 x2
(x̄1 , x̄2 ) = , .
x0 x0

(a) Let
⎡ 0 0 0⎤
a0 a1 a2
⎢ 1 1 1⎥
A = ⎣a0 a1 a2 ⎦ ∈ SL(3).
a20 a21 a22
Show that the transformation
T ([x]) = [Ax]
corresponds to the map
" #
1 2 a10 + a11 x̄1 + a12 x̄2 a20 + a21 x̄1 + a22 x̄2
T (x̄ , x̄ ) = ,
a00 + a01 x̄1 + a02 x̄2 a00 + a01 x̄1 + a02 x̄2
on V0 .
(b) Show that if A has the form
⎡ ⎤
1 0 0
⎢ ⎥
A = ⎣b1 a11 a12 ⎦
b2 a21 a22
with a11 a22 − a12 a21 = 1, then T : V0 → V0 is the equi-affine transformation
T (x̄) = Āx̄ + b̄,
   
a11 a12 b1
where Ā = 2 2 , b̄ = 2 . Therefore, the equi-affine group A(2) is a
a1 a2 b
proper subgroup of the group of projective transformations.
3.7. Projective space 101

(c) (Maple recommended) Consider the image of the unit circle

{(x̄1 , x̄2 ) ∈ V0 | (x̄1 )2 + (x̄2 )2 = 1}
in V0 under the transformation corresponding to the matrix
⎡ ⎤
cos(ϕ) − sin(ϕ) 0
⎢ ⎥
A = ⎣ sin(ϕ) cos(ϕ) 0⎦ .
0 0 1
Show that:

(1) If 0 < ϕ < π4 , then the image of the circle is an ellipse in V0 .

(2) If ϕ = π4 , then the image of the circle is a parabola in V0 .
(3) If π
4 < ϕ ≤ π2 , 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 pre-
serve the family of nondegenerate conic sections.)

Now, given a point [x] ∈ Pn , what is the isotropy group of [x]?

*Exercise 3.70. Let [x0 ] = [1 : 0 : · · · : 0] ∈ Pn . Show that:
(a) The isotropy group H[x0 ] of [x0 ] in SL(n + 1) is
⎧⎡ −1 r · · · r
⎤ ⎫
⎪
⎪ (det A) 1 n ⎪
⎪
⎪
⎨⎢ ⎥ ⎪
⎬
⎢ 0 ⎥
(3.17) H[x0 ] = ⎢ .. ⎥ : A ∈ GL(n), r1 , . . . , rn ∈ R .
⎪
⎪ ⎣ . A ⎦ ⎪
⎪
⎪
⎩ ⎪
⎭
0

(b) The isotropy group H[x] of any other point [x] ∈ Pn is

H[x] = t[x] H[x0 ] t−1
[x] ,

where t[x] is any matrix in SL(n + 1) whose first 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 defined as follows:
Definition 3.71. A projective frame f on Pn is a list of vectors f = (e0 ,
. . ., en ), where eα ∈ Rn+1 and det e0 · · · en = 1. We identify [e0 ] with
the position vector [x] ∈ Pn , 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 different
representative for [e0 ] would affect the scaling of the vectors (e1 , . . . , en ).

The same reasoning as in the Euclidean case shows that the set of projec-
tive frames may be regarded as the group SL(n + 1) via the one-to-one
correspondence

g(e0 , . . . , en ) = e0 · · · en .
We can define a projection map π : SL(n + 1) → Pn by

π(e0 , . . . , en ) = [e0 ];

the fiber of this map is the set of all projective frames based at [e0 ]. H[x0 ]
acts freely and transitively on each fiber, and so this map gives an explicit
description of SL(n+1) as a principal bundle over Pn with fiber group H[x0 ] :

H[x0 ] - SL(n + 1)
π
?
Pn ∼
= 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 defined by the equations

(3.18) deα = eβ ωαβ ,

where 0 ≤ α, β ≤ n.
The structure equations are

(3.19) dωαβ = −ωγβ ∧ ωαγ ,

where the sum is over 0 ≤ γ ≤ n. Since the Maurer-Cartan form ω = [ωαβ ]

takes values in the Lie algebra sl(n+1), the only relation among the 1-forms
(ωαβ ) is the trace condition
ωαα = 0.
As for which forms are semi-basic for the projection π : SL(n + 1) → Pn and
which should be regarded as connection forms, recall that the semi-basic
forms are those that pull back to each fiber of π to be zero, i.e., those with
the property that ωαβ (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

ω01 (v) = · · · = ω0n (v) = 0,

and so the semi-basic forms for π are (ω01 , . . . , ω0n ). The remaining (ωαβ ) 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: Define A to be the matrix whose columns are the frame
field 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 (ω̄ 1 , ω̄ 2 , ω̄ 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 (ω̄ji ) 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 (dω̄ i ), check that the fol-
lowing quantities are all equal to zero:
> for i from 1 to 3 do
Simf(d(omega[i]) + add(omega[i,j] &ˆ omega[j], j=1..3));
end do;
In order to verify the structure equations for the (dω̄ji ), check that the follow-
ing 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] &ˆ omega[k,j],
k=1..3)));
end do;
end do;
The following two exercises don’t require the Cartan package, but they nicely
illustrate some of Maple’s 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 fixed value
of θ) is given by
> newcurve:= A.unitcircle;
We can now define plots of this curve for various values of θ; 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): Define 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 ϕ as the time parameter, to see
how the unit circle transforms through a family of quadrics and back again
as ϕ varies from 0 to π:
> 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 “Plot”
menu choose “Animation → Play”. (Alternatively, right-click on the plot
and select “Animation → Play” from the pop-up menu.)
 Chapter 4

Curves and surfaces

in Euclidean space

4.1. Introduction

In this chapter, we get to the heart of the matter: Cartan’s 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 exis-
tence: Given prescribed values for the invariants, does there neces-
sarily exist a submanifold whose invariants coincide with the given
values?

In §4.2, we will address the theory underlying the first question, and in §4.3
we will show how it applies to curves in E3 . Then in §4.4 and §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 homoge-

neous 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 ⊂ G/H and f : U → 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 “restriction to M ”
really means pullback to U . The pullback bundle (or induced bundle) f ∗ G
of the principal bundle π : G → G/H is the bundle over U whose fiber over
a point u ∈ U is just the fiber of G over the point f (u) ∈ G/H:
f ∗ G = {(u, f ) ∈ U × G | f (u) = π(f )}.
The bundle f ∗ G is a principal bundle over U , with fiber group H. There is
a natural map fˆ : f ∗ G → G defined by
fˆ(u, f ) = f .
When π : G → G/H is regarded as a frame bundle, the image fˆ(u, f ) of
any element (u, f ) ∈ f ∗ G may be thought of as a frame based at the point
f (u) ∈ G/H. These maps may be represented by the following commutative
diagram:
fˆ -G
f ∗G
π̄ π
? f ?
U - G/H.

We will generally be interested in choosing a frame at each point of M ⊂

G/H—i.e., a “frame field” on M —according to certain geometric consider-
ations. Technically, this means choosing a section of the bundle f ∗ G over U ,
4.2. Equivalence of submanifolds of a homogeneous space 109

but we will usually regard it as choosing a lifting f˜ : U → G, i.e., a function

f˜ with the property that for any u ∈ U ,
(π ◦ f˜)(u) = f (u) ∈ M ⊂ G/H.
In other words, we choose f˜ so that the following diagram commutes:

G
>

f˜
π

 f ?
U - G/H.

When choosing a lifting f˜ : U → G, we will want to choose frame fields that

are adapted to M . This means that, instead of just choosing arbitrary frame
fields, we will use the geometry of M to choose “nice” frame fields. This is
somewhat analogous to choosing “nice” 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 field, or some-
times simply an adapted frame) f˜ : U → G, we can consider the pullback
f˜∗ ω of the Maurer-Cartan form ω of G and its structure equations to U .
The pulled-back Maurer-Cartan form f˜∗ ω 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
f˜∗ ω under such an action follows from the left-invariance of ω 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 field f˜ : U → G to contain useful information
about the invariants of M , the algorithm for choosing f˜ should be completely
determined in some canonical way by the geometry of M . Moreover, the
adapted frame field itself should be equivariant; this means that

(g · f ) = g · f˜
for any g ∈ G. If such an equivariant adapted frame field exists, then the
question of equivalence is completely answered by the following important
lemma:
Lemma 4.2. Let U ⊂ Rn be a connected, open set, and let f˜1 , f˜2 : U → G
be two immersions. Then there exists an element g ∈ G such that
f˜1 (u) = g · f˜2 (u)
for all u ∈ U if and only if f˜1∗ ω = f˜2∗ ω, where ω is the Maurer-Cartan form
of G.
110 4. Curves and surfaces in Euclidean space

Proof. First, observe that for any map f˜ : U → G, we have

(4.1) f˜∗ ω = f˜−1 df˜.
Remark 4.3. What does equation (4.1) really mean? Recall that for any
g ∈ G, ω is a linear map from Tg G to Te G = g defined by
ω(w) = (Lg−1 )∗ (w)
for w ∈ Tg G. Now, if f˜ : U → G is a differentiable map, then f˜∗ ω is a linear
map from Tu U to g defined by
f˜∗ ω(v) = ω(f˜∗ (v)) = (L(f(u))
˜
˜ ˜
−1 )∗ (f∗ (v)) = (f (u))
−1
· df˜(v)
for v ∈ Tu U . Therefore,
f˜∗ ω = f˜−1 df˜.

Now, given f˜1 , f˜2 : U → G, there exists a unique function g : U → G

satisfying the condition that
(4.2) f˜2 (u) = g(u)f˜1 (u)
for all u ∈ U —specifically, g(u) = f˜2 (u)(f˜1 (u))−1 . Differentiating (4.2)
yields
df˜2 = dg f˜1 + g df˜1 ;
therefore,
f˜2∗ ω = f˜2−1 df˜2
= f˜−1 dg f˜1 + f˜−1 g df˜1
2 2

= f˜2−1 dg f˜1 + (g f˜1 )−1 g df˜1

= f˜2−1 dg f˜1 + f˜−1 g −1 g df˜1
1

= f˜2−1 dg f˜1 + f˜1−1 df˜1

= f˜2−1 dg f˜1 + f˜∗ ω.
1

It follows that f˜1∗ ω = f˜2∗ ω 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 f˜∗ ω remains un-

changed when M is transformed by a symmetry g of the ambient
homogeneous space G/H.
(2) Conversely, f˜∗ ω 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 immer-
sion f : U → G/H, we will look for an equivariant method of constructing
a canonical adapted frame field f˜ : U → G. Then we will examine the
pulled-back Maurer-Cartan form f˜∗ ω, 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 α : I → E3 that maps some open in-

terval I ⊂ R into Euclidean space. E3 has the structure of the homogeneous
space E(3)/SO(3), so an adapted frame field along α should be a lifting
α̃ : I → E(3). Any such lifting can be written as
α̃(t) = (α(t); e1 (t), e2 (t), e3 (t)),
where for each t ∈ I, (e1 (t), e2 (t), e3 (t)) is an oriented, orthonormal basis for
the tangent space Tα(t) E3 . Such an adapted frame field is usually called an
orthonormal frame field along α. If the curve is “nice enough” (the precise
meaning of this will become clear shortly), then we will be able to choose
such a frame field in a canonical way, based on the geometry of the curve.
Remark 4.4. While the orthonormal frame field is technically the image
of the map α̃ and so includes the position vector α(t) at each point, it
is common to refer to the triple of vector fields (e1 (t), e2 (t), e3 (t)) as an
“orthonormal frame field along α”. Hopefully this terminology will not
cause any confusion.

Recall that α is regular if α (t) = 0 for every t ∈ I. The first condition that
we will require in order for α to be “nice enough” is that α must be a regular
curve. With this assumption, we can make our first frame adaptation by
setting
α (t)
e1 (t) =  ;
|α (t)|
i.e., we require that e1 (t) be the unit tangent vector to the curve at α(t).
Exercise 4.5. Show that this choice of e1 (t) is equivariant under the action
of E(3): If we replace α by g · α for some g ∈ E(3), then e1 (t) ∈ Tα(t) E3 will
be replaced by (Lg )∗ (e1 (t)) ∈ Tg·α(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 α in order to determine
how to choose the remainder of the adapted frame field.
Here we make an observation that will simplify the remainder of our com-
putations. Fix t0 ∈ I and define the arc length function along α to be
& t
s(t) = |α (u)| du.
t0

Exercise 4.6. Show that s(t) is invariant under the action of E(3); that is,
for any g ∈ E(3), the curves α and g · α have the same arc length function.

Since α (t) = 0 for all t ∈ I, the inverse function theorem implies that s(t)
has a differentiable inverse function t(s). By setting α(s) = α(t(s)), we
may assume that α is parametrized by arc length, so that |α (s)| = 1 and
e1 (s) = α (s).
In order to make the next adaptation, we need to make another assumption
about the curve. We will say that α is nondegenerate if α is regular and, in
addition, e1 (s) = 0 for all s ∈ I. In this case, differentiating 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 α(s).
Exercise 4.7. Show that e2 (s) is equivariant under the action of E(3): If
we replace α by g ·α for some g ∈ E(3), then e2 (s) ∈ Tα(s) E3 will be replaced
by (Lg )∗ (e2 (s)) ∈ Tg·α(s) E3 .

The adapted frame field 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) × e2 (s).
The vector e3 (s) is called the binormal vector to the curve at α(s). The
adapted frame field (e1 (s), e2 (s), e3 (s)) is called the Frenet frame of the
curve α(s); it determines a canonical, left-invariant lifting α̃ : I → E(3)
given by
α̃(s) = (α(s); e1 (s), e2 (s), e3 (s))
for any nondegenerate curve α 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 α̃. We have

α̃∗ (x) = α(s), α̃∗ (ei ) = ei (s).
Therefore,
α̃∗ (dx) = d(α̃∗ (x)) = α (s)ds,
α̃∗ (dei ) = d(α̃∗ (ei )) = ei (s)ds.

As in Chapter 3, write (ω̄ i , ω̄ji ) for the pulled-back forms (α̃∗ ω i , α̃∗ ωji ). 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
α (s)ds = ei (s) ω̄ i ,
(4.3)
ei (s)ds = ej (s) ω̄ij .
Now recall how we constructed our adapted frame field. First, we chose
e1 (s) so that α (s) = e1 (s); therefore, the first equation in (4.3) implies that
ω̄ 1 = ds, ω̄ 2 = ω̄ 3 = 0.
Then we chose e2 (s) so that e1 (s) is a multiple of e2 (s), say e1 (s) =
κ(s)e2 (s). The function κ(s) is called the curvature of α at s; note that
α is nondegenerate if and only if κ(s) > 0 for all s ∈ I. So the equation for
e1 (s) in (4.3) implies that
ω̄12 = κ(s)ds, ω̄13 = 0.
(Recall that ω̄11 = 0 by the skew-symmetry of the (ωji ).) The only remaining
Maurer-Cartan form is ω̄32 ; it must be equal to some multiple of ds, so define
a function τ (s) by the condition that
ω̄32 = −τ (s)ds.
The function τ (s) is called the torsion of α at s.
114 4. Curves and surfaces in Euclidean space

Remark 4.8. The minus sign in the definition of τ (s) is a convention pre-
ferred by some authors, but it is not universal. This choice of sign has the
feature that it results in a positive value of τ for the standard right-handed
helix
α(s) = t [a cos(s), a sin(s), b] ,
where a, b > 0 and a2 + b2 = 1.

Using the skew-symmetry of the (ω̄ji ), the remaining two equations in (4.3)
become
e2 (s)ds = e1 (s) ω̄21 + e3 (s) ω̄23 = (−e1 (s)κ(s) + e3 (s)τ (s))ds,
e3 (s)ds = e1 (s) ω̄31 + e2 (s) ω̄32 = −e2 (s)τ (s)ds.
Thus, we have the familiar Frenet equations:
⎡ ⎤
0 0 0 0
⎢ ⎥
  ⎢1 0 −κ(s) 0 ⎥
   
α (s) e1 (s) e2 (s) e3 (s) = α(s) e1 (s) e2 (s) e3 (s) ⎢⎢ ⎥.
−τ ⎥
⎣0 κ(s) 0 (s) ⎦
0 0 τ (s) 0
Note that if we regard α̃(s) as the matrix

α̃(s) = α(s) e1 (s) e2 (s) e3 (s) ,
then the matrix on the right multiplied by the 1-form ds is equal to
α̃(s)−1 d(α̃(s)),
and so it is exactly the pullback of the Maurer-Cartan form ω = g −1 dg on
E(3) via α̃.
Applying Lemma 4.2 yields the following theorem:
Theorem 4.9. Two nondegenerate curves α1 , α2 : I → E3 parametrized by
arc length differ by a rigid motion if and only if they have the same curvature
κ(s) and torsion τ (s).

This is the uniqueness portion of the fundamental theorem of space curves

(cf. Theorem 3.1). We will address the existence portion in §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:

• Is there a natural choice of parametrization for the curve?

• How should you choose the vectors (e1 (s), e2 (s), e3 (s), e4 (s)) of the
frame field? (And how do you ensure that these vectors form an
orthonormal frame field?) 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 ,

e2 (s) − e2 (s), e1 (s)e1 (s)

e3 (s) = .
|e2 (s) − e2 (s), e1 (s)e1 (s)|

In other words, e3 (s) is obtained by taking the orthogonal projec-

tion of e2 (s) onto the orthogonal complement of e1 (s) and e2 (s)
and then normalizing it to have unit length.)

• What is the right definition of “nondegenerate” for curves in E4 ?

• Where do invariants appear in the pullbacks of equations (3.1)?

What can you conclude from the skew-symmetry of the connection
forms?

• What is the 4-dimensional analog of the Frenet equations?

• How do you think the analysis would go for curves in En ?

4.4. Compatibility conditions and existence of submanifolds

with prescribed invariants

In §4.3, we saw that a curve α : I → E3 parametrized by arc length s is

completely determined up to rigid motions of E3 by its curvature κ(s) and
torsion τ (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 in-
variants for a “nice” immersion f : U → G/H: Assuming that we can find
a canonical, equivariant way of choosing a lifting f˜ : U → G (this is what
“nice” means), a complete set of invariants is contained in f˜∗ ω, the pullback
via f˜ of the Maurer-Cartan form ω of G.
For curves in E3 , it is now natural to ask whether the functions κ(s) and
τ (s) may be prescribed arbitrarily. In other words, given arbitrary functions
κ(s), τ (s) with κ(s) > 0, does there necessarily exist a curve α : I → E3 that
is parametrized by arc length and has curvature κ(s) and torsion τ (s)?

Exercise 4.11. Why must we require κ(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 ω̄
is a g-valued 1-form on a connected and simply connected manifold U . Then
there exists a smooth map f˜ : U → G with f˜∗ ω = ω̄ if and only if ω̄ satisfies
the Maurer-Cartan equation
(4.4) dω̄ = −ω̄ ∧ ω̄.

Outline of Proof. The full proof of this lemma requires the Frobenius
theorem and is beyond the scope of this book. (If you’re curious, the proof
may be found in [Gri74].) However, the main idea goes something like
this: f˜∗ ω contains quantities involving derivatives of the unknown function
f˜ : U → G, and for any given g-valued 1-form ω̄ on U , the equation f˜∗ ω = ω̄
may be regarded as a system of partial differential equations for f˜. In
general, this system is overdetermined and may have no solutions. However,
equation (4.4) is precisely the compatibility condition that must be satisfied
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 f˜(u0 ) for any u0 ∈ U .) Once we know that local solutions exist,
a patching argument can be used to construct a solution f˜ 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 ∈ U ;
simple connectivity is only necessary to ensure that these local solutions
can be patched together to form a single solution that is globally defined
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 →
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 field 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 satisfied, composing the
map f˜ with the natural projection π : G → G/H gives a smooth map
f : U → 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 ω̄ on U is equivalent to prescribing the values
of a complete set of invariants for an unknown submanifold M ⊂ G/H;
4.5. Moving frames for surfaces in E3 117

Lemma 4.12 then gives a necessary and sufficient condition for the existence
of a smooth map f : U → 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 ∈ G.

*Exercise 4.14. Let (ω 1 , . . . , ω n ) be a basis for the left-invariant 1-forms on

G that are semi-basic for the projection π : G → G/H, and let (ω̄ 1 , . . . , ω̄ n )
be the corresponding components of the g-valued 1-form ω̄ on U . Show that
the map f = π ◦ f˜ of Lemma 4.12 is an immersion if and only if (ω̄ 1 , . . . , ω̄ n )
span the cotangent space Tu∗ U at every point u ∈ U . (Note that typically
the dimension of U is less than n, so the forms (ω̄ 1 , . . . , ω̄ n ) will generally
not be linearly independent.)

Corollary 4.15. Let I ⊂ R be an open interval, and let κ, τ : I → R be any

differentiable functions satisfying κ(s) > 0 for all s ∈ I. Then there exists a
nondegenerate curve α : I → E3 , parametrized by arc length, with curvature
κ(s) and torsion τ (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 fields and find 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 → E3 be an immersion whose image is a regular surface Σ = x(U ).
Just as for curves, an adapted frame field along Σ should be a lifting x̃ :
U → E(3) of the form
x̃(u) = (x(u); e1 (u), e2 (u), e3 (u)) ,
where for each u ∈ 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-defined tangent plane Tx(u) Σ for each
point x(u) ∈ Σ. Thus, we can make our first frame adaptation by requiring
118 4. Curves and surfaces in Euclidean space

that e3 (u) be orthogonal to Tx(u) Σ. This determines e3 (u) uniquely up to

sign, and an orthonormal frame field 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) Σ no matter how we choose them. We will explore how we might refine
our choices later, but for now, we allow (e1 (u), e2 (u)) to be an arbitrary
orthonormal basis of Tx(u) Σ.
Now consider the pullbacks of equations (3.1) to U via x̃. (As in Chapter
3, write (ω̄ i , ω̄ji ) for the pulled-back forms (x̃∗ ω i , x̃∗ ωji ) on U .) Our first
observation about these forms is the following:
Proposition 4.18. Let U ⊂ R2 be an open set, and let x : U → E3 be an
immersion. For any adapted orthonormal frame field (e1 (u), e2 (u), e3 (u))
along Σ = x(U ), the associated dual and connection forms (ω̄ i , ω̄ji ) have the
property that ω̄ 3 = 0.

Proof. The pullback of the first equation in (3.1) is

dx = ei ω̄ i .
Let u ∈ U . Then dxu is a linear map from Tu U to Tx(u) Σ, and so for any
v ∈ Tu U , we must have
dxu (v) = ei (u)ω̄ i (v) ∈ Tx(u) Σ.
Since Tx(u) Σ is spanned by e1 (u) and e2 (u), the e3 (u) term in this sum must
vanish; therefore, ω̄ 3 (v) = 0. And since v ∈ Tu U is arbitrary, it follows that
ω̄ 3 = 0. 
*Exercise 4.19. Show that (ω̄ 1 , ω̄ 2 ) are linearly independent 1-forms on
U . Therefore, they form a basis for the 1-forms on U . (Hint: Evaluate ω̄ 1
and ω̄ 2 on vectors v1 , v2 ∈ 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 first fundamental form of the surface.
Definition 4.20. Let U ⊂ R2 be an open set, and let x : U → E3 be an
immersion. The first fundamental form of Σ = x(U ) is the quadratic form I
on T U defined by
I(v) = dx(v), dx(v)
for v ∈ 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 Σ. Its primary function is to describe how to
compute this metric in terms of the local coordinates u on Σ that are given
by the parametrization x : U → E3 .
*Exercise 4.21. Show that for any v ∈ Tu U ,
 2  2
I(v) = ω̄ 1 (v) + ω̄ 2 (v) .
This is often written more concisely as
I = (ω̄ 1 )2 + (ω̄ 2 )2 ,
and each term in the sum should be interpreted as the symmetric product
ω̄ i ◦ ω̄ i .

While the first fundamental form is defined as a function of a single tangent

vector, it can be used to define an inner product ·, ·u on each tangent space
Tu U through a process called polarization.
Definition 4.22. The inner product ·, ·u is defined by
v, wu = 1
4 (I(v + w) − I(v − w))
for v, w ∈ Tu U .
*Exercise 4.23. (a) Show that
v, wu = ω̄ 1 (v) ω̄ 1 (w) + ω̄ 2 (v) ω̄ 2 (w).

(b) Convince yourself that ·, ·u is a section of the symmetric tensor bundle
S 2 (T ∗ U ). Any section of this bundle defines a symmetric bilinear form
B : T U × T U → R,
which in turn defines a quadratic form
Q : TU → R
by setting Q(v) = B(v, v).
*Exercise 4.24. If you’ve seen the first fundamental form before, you prob-
ably 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 → E3 is an immersion with F = 0. (Such a parametriza-
tion for a given surface Σ always exists, at least locally; the proof is beyond
the scope of this book but can be found in [dC76]. This assumption isn’t
necessary, but it keeps the calculations simpler.)
120 4. Curves and surfaces in Euclidean space

(a) Show that the frame field

1 1
e1 (u) = √ xu , e2 (u) = √ xv , e3 (u) = e1 (u) × e2 (u)
E G
is an oriented, orthonormal frame field along Σ = x(U ), with e3 (u) orthog-
onal to Tx(u) Σ.
(b) Show that the dual forms of this frame field are
√ √
ω̄ 1 = E du, ω̄ 2 = G dv, ω̄ 3 = 0
and that
I = (ω̄ 1 )2 + (ω̄ 2 )2 = E du2 + G dv 2 .

(c) Show that if (e1 (u), e2 (u), e3 (u)) is replaced by another adapted frame
field (ẽ1 (u), ẽ2 (u), ẽ3 (u)) of the form
ẽ1 = cos(θ) e1 + sin(θ) e2 ,
ẽ2 = − sin(θ) e1 + cos(θ) e2 ,
ẽ3 = e3 ,
˜ 1 , ω̄
then the dual forms (ω̄ ˜ 2 ) of the new adapted frame field are
˜ 1 = cos(θ) ω̄ 1 + sin(θ) ω̄ 2 ,
ω̄
˜ 2 = − sin(θ) ω̄ 1 + cos(θ) ω̄ 2 .
ω̄
Moreover,
I = (ω̄ 1 )2 + (ω̄ 2 )2 = (ω̄
˜ 1 )2 + (ω̄
˜ 2 )2 .

Now let’s see what we can learn by differentiating! Since ω̄ 3 = 0, we must

have dω̄ 3 = 0 as well. According to the Cartan structure equations (3.8),
this implies that
dω̄ 3 = −ω̄13 ∧ ω̄ 1 − ω̄23 ∧ ω̄ 2 = 0.
Since (ω̄ 1 , ω̄ 2 ) are linearly independent 1-forms, Cartan’s lemma (cf. Lemma
2.49) implies that there exist real-valued functions h11 , h12 , h22 on U such
that  3   
ω̄1 h11 h12 ω̄ 1
= .
ω̄23 h12 h22 ω̄ 2

How should we interpret the functions (hij )? Recall that

de3 = e1 ω31 + e2 ω32 = −(e1 ω13 + e2 ω23 ).
For any tangent vector w ∈ Tx Σ, de3 (w) measures the directional deriv-
ative of the normal vector field e3 in the direction of w. So, up to sign,
ω13 (w) measures the e1 component of this directional derivative, and ω23 (w)
measures its e2 component. In other words, ωi3 (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 ω̄i3 as a linear combination
of ω̄ 1 and ω̄ 2 , 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 ⊂ 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 defined as follows:

Definition 4.25. Let U ⊂ R2 be an open set, and let x : U → E3 be an

immersion with image Σ = x(U ). The Gauss map of Σ = x(U ) is the map
N : Σ → S2 defined by
N (x(u)) = e3 (u),
where (e1 (u), e2 (u), e3 (u)) is any adapted frame field on Σ = x(U ). (Note
that N is well-defined up to sign.)

Notions of curvature typically associated with surfaces (Gauss curvature,

mean curvature, etc.) arise as linear-algebraic properties of the differential
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.

Definition 4.26. Let U ⊂ R2 be an open set, and let x : U → E3 be an

immersion. The second fundamental form of Σ = x(U ) is the quadratic form
II on T U defined by
II(v) = −de3 (v), dx(v)
for v ∈ Tu U , where (e1 (u), e2 (u), e3 (u)) is any adapted frame field on Σ =
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 ∈ Tu U ,

II(v) = ω̄13 (v) ω̄ 1 (v) + ω̄23 (v) ω̄ 2 (v)
= h11 (ω̄ 1 (v))2 + 2h12 ω̄ 1 (v) ω̄ 2 (v) + h22 (ω̄ 2 (v))2 .
This is often written more concisely as
II = ω̄13 ω̄ 1 + ω̄23 ω̄ 2 = h11 (ω̄ 1 )2 + 2h12 ω̄ 1 ω̄ 2 + h22 (ω̄ 2 )2 .
122 4. Curves and surfaces in Euclidean space

(b) Suppose that x : U → E3 is a parametrization with F = 0, and let

(e1 (u), e2 (u), e3 (u)) be the frame field in part (a) of Exercise 4.24. Show
that
√
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  = −(e3 )u , xu  = −de3 ∂u , dx ∂u ,
∂ ∂
∂ ∂
f = e3 , xuv  = −(e3 )v , xu  = −de3 ∂v , dx ∂u 
∂ ∂
= −(e3 )u , xv  = −de3 ∂u , dx ∂v ,
∂ ∂
g = e3 , xvv  = −(e3 )v , xv  = −de3 ∂v , dx ∂v .

(Some authors use , m, n or L, M, N in place of e, f, g.) Show that this

agrees with Definition 4.26, and conclude that
e f g
h11 = , h12 = √ , h22 = .
E EG G

Now, we still haven’t figured out how we should choose the vectors (e1 (u),
e2 (u)), except that they should form an orthonormal basis for Tx(u) Σ at each
point. In order to refine our adapted frame field 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 field, with associated Maurer-Cartan forms (ω̄ i , ω̄ji ).
Any other adapted frame field (ẽ1 (u), ẽ2 (u), ẽ3 (u)) has the form (up to sign)
⎡ ⎤
cos(θ) − sin(θ) 0
 ⎢ ⎥
ẽ1 (u) ẽ2 (u) ẽ3 (u) = e1 (u) e2 (u) e3 (u) ⎣ sin(θ) cos(θ) 0⎦
0 0 1

˜ i , ω̄
for some function θ on U . Let (ω̄ ˜ ji ) be the Maurer-Cartan forms associ-
 
cos(θ) − sin(θ)
ated to the new frame field, and set B = .
sin(θ) cos(θ)

*Exercise 4.28. (a) Show that the result in part (c) of Exercise 4.24 can
be expressed as
 1  1
˜
ω̄ ω̄
(4.5) = B −1 .
˜2
ω̄ ω̄ 2
4.5. Moving frames for surfaces in E3 123

(b) Show that

     
˜ 13
ω̄ ω̄13 ω̄13
−1 t
(4.6) =B = B .
˜ 23
ω̄ ω̄23 ω̄23
(Hint: Use the equation for de3 in (3.1).)
(c) Cartan’s lemma implies that there exist functions h̃11 , h̃12 , h̃22 on U such
that  3    1
˜1
ω̄ h̃11 h̃12 ω̄˜
= .
˜ 23
ω̄ h̃12 h̃22 ω̄˜2
Show that
     
h̃11 h̃12 h11 h12 h11 h12
−1 t
(4.7) =B B= B B.
h̃12 h̃22 h12 h22 h12 h22

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 ∈ U there exists an adapted frame (e1 (u), e2 (u), e3 (u)) at the
point x(u) ∈ Σ with the property that
   
h11 (u) h12 (u) κ1 (u) 0
(4.8) =
h12 (u) h22 (u) 0 κ2 (u)
for some real numbers κ1 (u), κ2 (u).
*Exercise 4.29. Let (e1 (u), e2 (u), e3 (u)) be an adapted frame satisfying
(4.8), and let v1 , v2 ∈ 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)) = −κi (u)ei (u), i = 1, 2.
This implies that e1 (u) and e2 (u) are eigenvectors for the linear transfor-
mation dNx(u) , the differential of the Gauss map N : Σ → S2 at the point
x(u) ∈ Σ, with eigenvalues −κ1 (u), −κ2 (u), respectively.

You may recall the following definition:

Definition 4.30. The eigenvectors for −dNx(u) are called principal vectors
or principal directions at the point x(u) ∈ Σ. The associated eigenvalues
κ1 (u), κ2 (u) are called the principal curvatures of Σ at x(u).

Therefore, there exists an adapted frame (e1 (u), e2 (u), e3 (u)) at each point
x(u) ∈ Σ 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) ∈ Σ, and an adapted frame field on Σ which has this property at every
point x(u) ∈ Σ will be called a principal adapted frame field on Σ.
124 4. Curves and surfaces in Euclidean space

Definition 4.31. If κ1 (u) = κ2 (u) for some point u ∈ U , then the corre-
sponding point x(u) of Σ is called an umbilic point of Σ.

If Σ has no umbilic points, then a principal adapted frame field can be

determined uniquely (up to sign) by requiring that κ1 > κ2 ; moreover,
this frame field determines a smooth map x̃ : U → E(3). However, it can
happen that a principal adapted frame field 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 Σ has no umbilic points, then the choice
of a principal adapted frame field (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 field, the second fundamental
form is given by
II = κ1 (ω̄ 1 )2 + κ2 (ω̄ 2 )2 .
Remark 4.33. Exactly how much freedom does the phrase “up to sign”
represent? Given any principal adapted frame field (e1 (u), e2 (u), e3 (u)), we
can

(1) replace e3 (u) by −e3 (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 field are

(−e1 (u), e2 (u), −e3 (u)),
(e1 (u), −e2 (u), −e3 (u)),
(−e1 (u), −e2 (u), e3 (u)).
*Exercise 4.34. For each of the principal adapted frame fields in Remark
4.33, how do the sign changes to the frame vectors affect the Maurer-Cartan
forms (ω̄ i , ω̄ji )?

Suppose that Σ = x(U ) has no umbilic points. Now that we (finally!) have
a way of defining a canonical adapted frame field along Σ, we can apply
Lemma 4.2 to find a complete set of invariants for the surface.
Theorem 4.35 (Bonnet). Let U ⊂ R2 be an open set. Two immersions
x1 , x2 : U → E3 without umbilic points differ by a rigid motion if and only
if they have the same first 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 field near umbilic points.

Proof. One direction is clear: Since all our constructions are equivariant
under the action of E(3), any two surfaces that differ by a rigid motion must
have the same first and second fundamental forms.
Conversely, suppose that x1 , x2 have the same first and second fundamental
forms. Let x̃1 , x̃2 : U → E(3) be principal adapted frame fields for x1 , x2 ,
respectively; let (ω̄ i , ω̄ji ) denote the pulled-back dual and connection forms
for x1 and let (Ω̄i , Ω̄ij ) denote those for x2 . By hypothesis,
Ix1 = (ω̄ 1 )2 + (ω̄ 2 )2 = (Ω̄1 )2 + (Ω̄2 )2 = Ix2 ,
IIx1 = (κ1 )x1 (ω̄ 1 )2 + (κ2 )x1 (ω̄ 2 )2 = (κ1 )x2 (Ω̄1 )2 + (κ2 )x2 (Ω̄2 )2 = IIx2 .
Equality of the first fundamental forms implies that
 1   
Ω̄ cos(θ) − sin(θ) ω̄ 1
=
Ω̄2 ± sin(θ) ± cos(θ) ω̄ 2
for some function θ 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
 
(κ1 )x1 (ω̄ 1 )2 + (κ2 )x1 (ω̄ 2 )2 = (κ1 )x2 cos2 (θ) + (κ2 )x2 sin2 (θ) (ω̄ 1 )2
+ 2 (((κ2 )x2 − (κ1 )x2 ) sin(θ) cos(θ)) ω̄ 1 ω̄ 2
 
+ (κ1 )x2 sin2 (θ) + (κ2 )x2 cos2 (θ) (ω̄ 2 )2 .
Since (κ1 )x2 > (κ2 )x2 , the vanishing of the middle term on the right-hand
side implies that θ is a multiple of π2 . Then the remaining terms, together
with the inequality κ1 > κ2 on both sides, imply that θ is a multiple of π.
Therefore, (κ1 )x2 = (κ1 )x1 , (κ2 )x2 = (κ2 )x1 , and
Ω̄1 = ±ω̄ 1 , Ω̄2 = ±ω̄ 2 .
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.
Since we now have Ω̄1 = ω̄ 1 , we must have dΩ̄1 = dω̄ 1 . According to the
Cartan structure equations (3.8), this implies that
(Ω̄12 − ω̄21 ) ∧ ω̄ 2 = 0.
By Cartan’s lemma, Ω̄12 − ω̄21 must be a multiple of ω̄ 2 . But the same
reasoning applied to the equation dΩ̄2 = dω̄ 2 implies that Ω̄12 − ω̄21 must also
126 4. Curves and surfaces in Euclidean space

be a multiple of ω̄ 1 . Since (ω̄ 1 , ω̄ 2 ) are linearly independent, it follows that

Ω̄12 = ω̄21 .

Finally, since x̃1 and x̃2 are both principal adapted frame fields, we have
Ω̄31 = (κ1 )x2 Ω̄1 = (κ1 )x1 ω̄ 1 = ω̄13 ,
Ω̄32 = (κ2 )x2 Ω̄2 = (κ2 )x1 ω̄ 2 = ω̄23 .
The theorem now follows from Lemma 4.2. 

Now we consider the question discussed in §4.4; namely, can the first and
second fundamental forms be prescribed arbitrarily? We must require that
I be a positive definite quadratic form (i.e., that I(v) > 0 for every v = 0 ∈
T U ) in order to define 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 ∈ 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 fundamen-

tal forms determines the 1-forms (ω̄ 1 , ω̄ 2 , ω̄13 , ω̄23 ) associated to a principal
adapted frame field up to sign and that these forms will have the properties
that (ω̄ 1 , ω̄ 2 ) are linearly independent and that ω̄i3 is a multiple of ω̄ i for
i = 1, 2. So, suppose that we are given 1-forms (ω̄ 1 , ω̄ 2 , ω̄13 , ω̄23 ) on an open
set U ⊂ R2 that satisfy these conditions. What additional conditions must
these forms satisfy in order that there exist an embedding x : U → E3 whose
first and second fundamental forms are
I = (ω̄ 1 )2 + (ω̄ 2 )2 ,
II = ω̄13 ω̄ 1 + ω̄23 ω̄ 2 ?
Lemma 4.12 gives the answer: The forms ω̄ 1 , ω̄ 2 , ω̄13 = −ω̄31 , ω̄23 = −ω̄32 ,
together with the form ω̄ 3 = 0 and some additional form ω̄21 = −ω̄12 , must
satisfy the structure equations (3.8) for the Maurer-Cartan forms on E(3).
Because ω̄ 3 = 0, the first three of these equations may be written as
dω̄ 1 = −ω̄21 ∧ ω̄ 2 ,
(4.9) dω̄ 2 = ω̄21 ∧ ω̄ 1 ,
dω̄ 3 = 0 = −ω̄13 ∧ ω̄ 1 − ω̄23 ∧ ω̄ 2 .
*Exercise 4.38. Show that the first two equations in (4.9) uniquely deter-
mine the 1-form ω̄21 . (Hint: ω̄21 must be equal to some linear combination
of (ω̄ 1 , ω̄ 2 ). Show that each of the first two equations determines one of the
unknown coefficients.)
4.5. Moving frames for surfaces in E3 127

The form ω̄21 determined by the first two equations in (4.9) is called the
Levi-Civita connection form of the metric defined by the first fundamental
form I = (ω̄ 1 )2 + (ω̄ 2 )2 . The third equation just says that (ω̄13 , ω̄23 ) must be
symmetric linear combinations of (ω̄ 1 , ω̄ 2 ), which will automatically be true
under our assumptions.
The remaining structure equations may be written as
dω̄21 = ω̄13 ∧ ω̄23 ,
(4.10) dω̄13 = ω̄23 ∧ ω̄21 ,
dω̄23 = −ω̄13 ∧ ω̄21 .
The first 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 (ω̄ 1 , ω̄ 2 , ω̄13 , ω̄23 ) be 1-forms on a connected and
simply connected open set U ⊂ R2 satisfying the conditions that (ω̄ 1 , ω̄ 2 ) are
linearly independent at each point of U and that ω̄i3 is a scalar multiple of
ω̄ i for i = 1, 2. Suppose that, together with the Levi-Civita connection form
ω̄21 determined by ω̄ 1 and ω̄ 2 , these forms satisfy the Gauss and Codazzi
equations (4.10). Then there exists an immersed surface x : U → E3 , unique
up to rigid motion, whose first and second fundamental forms are

I = (ω̄ 1 )2 + (ω̄ 2 )2 ,
II = ω̄13 ω̄ 1 + ω̄23 ω̄ 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 → E3 with first
and second fundamental forms

I = cosh2 (v) (du2 + dv 2 ),

II = du2 − dv 2 .

(a) Show that the 1-forms (ω̄ 1 , ω̄ 2 , ω̄13 , ω̄23 ) determined by I and II according
to the conditions of Theorem 4.39 are

ω̄ 1 = cosh(v) du, ω̄ 2 = cosh(v) dv,

1 1
ω̄13 = du, ω̄23 = − dv.
cosh(v) cosh(v)
128 4. Curves and surfaces in Euclidean space

(b) Show that the Levi-Civita connection form is

ω̄21 = tanh(v) du.
(Hint: Set
ω̄21 = a du + b dv
for some unknown functions a, b on R2 . Use the structure equations for dω̄ 1
and dω̄ 2 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 → E3 is an immersion whose coordinate curves are all principal
curves. (This means that xu , xv are both principal vectors at each point of
Σ = x(U ).)
(a) Show that the frame field in part (a) of Exercise 4.24 is a principal
adapted frame field along Σ.
(b) Show that the condition that all coordinate curves of x are principal
curves is equivalent to the condition that the first 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
ω̄21 = √ (Ev du − Gu dv).
2 EG

(d) Show that

e 1 e
ω̄13 = ω̄ = √ du,
E E
g g
ω̄23 = ω̄ 2 = √ dv.
G G

(e) Show that the Gauss equation is equivalent to

" # " # 
eg 1 Ev Gu
(4.11) =− √ √ + √ .
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

1 e g!
ev = Ev + ,
2 E G
(4.12)
1 e 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 → E3 is totally umbilic.

(a) Show that any adapted frame field (e1 (u), e2 (u), e3 (u)) is a principal
adapted frame field.
(b) Let (ω̄ i , ω̄ji ) be the Maurer-Cartan forms for an adapted frame field on
Σ = x(U ). Show that there exists a smooth function λ : U → R such that
(4.13) ω̄13 = λω̄ 1 , ω̄23 = λω̄ 2 .
Conclude that the second fundamental form of Σ is a scalar multiple of the
first fundamental form, i.e., that
II = λI.

(c) Prove that λ is constant. (Hint: Use the structure equations to differen-
tiate equations (4.13), taking into account the fact that we must have
dλ = λ1 ω̄ 1 + λ2 ω̄ 2
for some functions λ1 , λ2 on U . Then use Cartan’s lemma.)
(d) Show that if λ = 0, then de3 = 0. Conclude that the normal vector field
of Σ is constant and that Σ is therefore contained in a plane.
(e) Show that if λ = 0, then d(x+ λ1 e3 ) = 0. Conclude that the vector-valued
function x + λ1 e3 : U → E3 is equal to some constant point q ∈ E3 and that
1
Σ is therefore contained in the sphere of radius |λ| 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

κ1 , κ2 of a regular surface Σ are equal at every point of Σ, then they must in
fact be constant. This suggests a related question: Can we categorize those
surfaces for which κ1 , κ2 are constants, but not necessarily equal?
130 4. Curves and surfaces in Euclidean space

Exercise 4.43. Suppose that the surface x : U → E3 has the property that
both principal curvatures κ1 , κ2 are constants. We know from Exercise 4.42
that if κ1 = κ2 , then Σ = x(U ) is contained in either a plane or a sphere,
so assume that κ1 = κ2 . Let (e1 (u), e2 (u), e3 (u)) be a principal adapted
frame field on Σ, and let x̃ : U → E(3) denote the corresponding lifting of
x : U → E3 , with associated Maurer-Cartan forms (ω̄ i , ω̄ji ). Then we have

(4.14) ω̄13 = κ1 ω̄ 1 , ω̄23 = κ2 ω̄ 2 .

(a) Differentiate equations (4.14) to obtain

(κ1 − κ2 )ω̄21 ∧ ω̄ 1 = (κ1 − κ2 )ω̄21 ∧ ω̄ 2 = 0.
Use Cartan’s lemma to conclude that ω̄21 = 0.
(b) Differentiate the equation ω̄21 = 0 and show that κ1 κ2 = 0. Without loss
of generality, we may assume that κ1 = 0, κ2 = 0.
In the remainder of this exercise, we will see how the structure equations
can be integrated in order to determine the surface Σ.
(c) Show that dω̄ 1 = dω̄ 2 = 0. Apply the Poincaré lemma (cf. Theorem
2.31) to conclude that there exist functions u, v on U such that
ω̄ 1 = du, ω̄ 2 = dv.
Thus the pullbacks of the structure equations (3.1) to U via x̃ can be written
as
dx = e1 du + e2 dv,
de1 = 0,
(4.15)
de2 = e3 (κ2 dv),
de3 = −e2 (κ2 dv).

(d) Integrate equations (4.15) (beginning with the equations for de1 , de2 , de3
and working backwards) to show that there exist constant vectors ē1 , ē2 , ē3 ,
x̄ ∈ E3 such that
e1 (u, v) = ē1 ,
e2 (u, v) = cos(κ2 v) ē2 + sin(κ2 v) ē3 ,
(4.16) e3 (u, v) = − sin(κ2 v) ē2 + cos(κ2 v) ē3 ,
1 1
x(u, v) = x̄ + u ē1 + sin(κ2 v) ē2 − cos(κ2 v) ē3 .
κ2 κ2

(e) Use equations (4.16) and the fact that (e1 (u), e2 (u), e3 (u)) is an or-
thonormal frame field to show that (ē1 , ē2 , ē3 ) is an orthonormal frame.
4.5. Moving frames for surfaces in E3 131

Conclude that via a Euclidean transformation, we can arrange that x̄ = 0

and ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
1 0 0
ē1 = ⎣0⎦ , ē2 = ⎣1⎦ , ē3 = ⎣0⎦ ,
0 0 1
and hence that
⎡ ⎤
u
⎢ ⎥
(4.17) x(u, v) = ⎣ κ12 sin(κ2 v) ⎦ .
− κ12 cos(κ2 v)
Equation (4.17) describes a parametrization for the cylinder of radius |κ12 |
centered along the x1 -axis; therefore, Σ = x(U ) is contained in a cylinder of
radius |κ12 | .

Together, Exercises 4.42 and 4.43 prove the following classification theorem:
Theorem 4.44. Let Σ be a connected, regular surface in E3 whose principal
curvatures are constant. Then Σ is contained in either a plane, sphere, or
cylinder.

Any invariant of an immersed surface x : U → E3 that can be expressed

purely in terms of the first fundamental form
I = (ω̄ 1 )2 + (ω̄ 2 )2
is called an intrinsic invariant of the surface. For instance, arc length and
area are intrinsic quantities on Σ = x(U ). The principal curvatures κ1 , κ2 ,
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
definition:
Definition 4.45. The function K = κ1 κ2 on Σ is called the Gauss curvature
of Σ. The function H = 12 (κ1 + κ2 ) on Σ is called the mean curvature of Σ.
Remark 4.46. It is not necessary that an adapted frame field be principal
in order to compute the Gauss and mean curvatures. For any adapted frame
field on Σ with associated matrix [hij ], we have
K = det[hij ], H = 12 tr[hij ].

Even though κ1 , κ2 are not intrinsic quantities, Gauss’s “Theorema Egregi-

um” 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 specified.)

Exercise 4.47. Let x : U → E3 be an immersed surface. The 1-forms

(ω̄ 1 , ω̄ 2 ) are determined by the first fundamental form of x up to a transfor-
mation of the form
 1   
˜
ω̄ cos(θ) sin(θ) ω̄ 1
=
˜2
ω̄ − sin(θ) cos(θ) ω̄ 2
for some function θ on U .
(a) Show that the area form

dA = ω̄ 1 ∧ ω̄ 2

is an intrinsic quantity, i.e., that

˜ 1 ∧ ω̄
dà = ω̄ ˜ 2 = dA.

(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 ω̄21 is the Levi-Civita connection form corresponding to
(ω̄ 1 , ω̄ 2 ), then
˜ 21 = ω̄21 − dθ.
ω̄
Conclude that dω̄21 is an intrinsic quantity.
(c) Show that
dω̄21 = ω̄13 ∧ ω̄23 = K ω̄ 1 ∧ ω̄ 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 min-
imal surfaces; these surfaces are of considerable interest and will be treated
in detail in Chapter 8. Surfaces for which the Gauss curvature K is iden-
tically zero are called flat, and we will conclude this chapter with a brief
exploration of their local theory.
Because the Gauss curvature of a regular surface Σ is an intrinsic quantity, it
is not changed by any deformation of Σ that preserves the first fundamental
form of Σ. 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 flat.
4.5. Moving frames for surfaces in E3 133

Since any flat surface Σ must have K = κ1 κ2 = 0, one of the principal

curvatures κ1 , κ2 must be identically zero on Σ. As we saw in Exercise 4.42
that any surface with κ1 = κ2 = 0 must be contained in a plane, we will
disregard this case and, to keep things simple, we will assume that Σ has no
umbilic points. (In practice, this simply means that we restrict our attention
to the open subset of Σ consisting of the non-umbilic points.)

Exercise 4.48. Suppose that the surface x : U → E3 is flat and has no

umbilic points. Without loss of generality, we may assume that the principal
curvatures of Σ = x(U ) satisfy κ1 = 0, κ2 = 0. Let (e1 (u), e2 (u), e3 (u))
be a principal adapted frame field on Σ, and let x̃ : U → E(3) denote the
corresponding lifting of x : U → E3 , with associated Maurer-Cartan forms
(ω̄ i , ω̄ji ). Then since κ1 = 0, we have

(4.18) ω̄13 = 0, ω̄23 = κ2 ω̄ 2 .

(a) Differentiate the equation ω̄13 = 0 and use Cartan’s lemma to conclude
that

(4.19) ω̄21 = μ ω̄ 2

for some function μ : U → R.

(b) Show that dω̄ 1 = 0, and apply the Poincaré lemma (cf. Theorem 2.31)
to conclude that there exists a function u on U such that ω̄ 1 = du.
(c) Use the structure equation for dω̄ 2 and the Frobenius theorem (cf. The-
orem 2.33) to conclude that for any point u ∈ U , there exist a neighborhood
V ⊂ U of u and differentiable functions λ, v : V → R (with λ = 0) such
that the restriction of ω̄ 2 to V is given by ω̄ 2 = λ dv. (For simplicity, we
will shrink U if necessary and assume that these functions are defined on
the entire open set U .)
(d) Since ω̄ 1 ∧ ω̄ 2 = 0, the functions (u, v) form a local coordinate system on
U , and we may regard λ, μ as functions of u and v. Show that the structure
equation for dω̄ 2 implies that μ = − λλu , and therefore

(4.20) ω̄21 = −λu dv.

(e) Use the structure equation for dω̄21 to conclude that λuu = 0, and there-
fore

(4.21) λ(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 dω̄23 to conclude that

(κ2 λ)u = 0.
Integrate and use equation (4.21) to conclude that
f2 (v)
(4.22) κ2 (u, v) =
uf1 (v) + f0 (v)
for some smooth functions f0 (v), f1 (v), f2 (v).
(g) The pullbacks of the structure equations (3.1) to U via x̃ can now be
written as
dx = e1 du + e2 (uf1 (v) + f0 (v))dv,
de1 = e2 f1 (v) dv,
(4.23)
de2 = −e1 f1 (v) dv + e3 f2 (v) dv,
de3 = −e2 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 Σ 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 Σ is given by H = 12 κ2
with κ2 as in equation (4.22):
Theorem 4.49. Let Σ be a flat surface whose mean curvature H is nonzero
everywhere. Then for each point x ∈ Σ, there exists a unique straight line
x in E3 such that x ∈ x and x ∩ Σ is an open neighborhood of x in x .
Moreover, the restriction of the function H1 to the open interval x ∩ Σ is an
affine 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 suffices
to declare a linearly independent subset; we’ll define 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]:= 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];
Tell Maple how to differentiate these forms according to the Cartan struc-
ture equations (3.8):
> for i from 1 to 3 do
d(omega[i]):= -add(’omega[i,j] &ˆ omega[j]’, j=1..3);
end do;
d(omega[1,2]):= -add(’omega[1,k] &ˆ omega[k,2]’, k=1..3);
d(omega[3,1]):= -add(’omega[3,k] &ˆ omega[k,1]’, k=1..3);
d(omega[3,2]):= -add(’omega[3,k] &ˆ omega[k,2]’, k=1..3);
Now consider the pullbacks of the Maurer-Cartan forms to the surface via
an adapted frame field. The first condition that these forms must satisfy is
ω̄ 3 = 0.
In Maple, it’s often useful to impose such conditions via a substitution
rather than by simply setting ω̄ 3 equal to zero. The reason for this is that
if we make the assignment ω̄ 3 = 0, we lose the ability to use the structure
equation for dω̄ 3 because Maple will just evaluate dω̄ 3 as d(0) = 0. Using
a substitution allows us to choose when we want Maple to be aware that
ω̄ 3 = 0 and when we don’t.
So, introduce the following substitution for the Maurer-Cartan forms as-
sociated to an adapted frame field. (We’ll add more information to this
substitution as we learn more about the Maurer-Cartan forms.)
> adaptedsub1:= [omega[3]=0];
Now, since we have ω̄ 3 = 0, we must have dω̄ 3 = 0 as well. So the following
quantity must be zero:
> zero1:= Simf(subs(adaptedsub1, Simf(d(omega[3]))));

zero1 := (ω1 ) &ˆ (ω31 ) + (ω2 ) &ˆ (ω32 )

136 4. Curves and surfaces in Euclidean space

Note that we first computed dω̄ 3 and then applied the substitution to tell
Maple that ω̄ 3 = 0. In this case the knowledge that ω̄ 3 = 0 didn’t affect
the computation of dω̄ 3 , but it’s a good idea to get in the habit of applying
such substitutions when you intend for them to be in effect.
Applying Cartan’s lemma tells us that (ω̄13 , ω̄23 ) must be symmetric linear
combinations of (ω̄ 1 , ω̄ 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
forms (ω̄ i , ω̄ji ) and the transformed forms (ω̄
˜ i , ω̄
˜ ji ), introduce new 1-forms to
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 won’t 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]];

We’ll also need the reverse substitution so that we can go back and forth
between the two sets of Maurer-Cartan forms:
> framechangebacksub:= makebacksub(framechangesub);
In order to compare the functions (h̃ij ) associated to the transformed forms
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]];
Now combine all these substitutions to see how the (h̃ij ) are expressed in
˜ 13 in terms of (ω̄13 , ω̄23 ):
terms of the (hij ): First, write ω̄
> Simf(subs(framechangesub, Omega[3,1]));

cos(θ) ω3,1 + sin(θ) ω3,2

Next, convert this to an expression in terms of the (hij ) and (ω̄ 1 , ω̄ 2 ):

> Simf(subs(adaptedsub1, %));

(cos(θ) h1,1 + sin(θ) h1,2 ) ω1 + (cos(θ) h1,2 + sin(θ) h2,2 ) ω2

˜ 1 , ω̄
Finally, convert this to an expression in terms of (ω̄ ˜ 2 ):

> Simf(subs(framechangebacksub, %));

(cos(θ)2 h1,1 + 2 cos(θ) sin(θ) h1,2 + h2,2 − h2,2 cos(θ)2 ) Ω1

+ (− cos(θ) sin(θ) h1,1 + cos(θ) sin(θ) h2,2 + 2 cos(θ)2 h1,2 − h1,2 )Ω2

Of course, this sequence of operations can be combined into a single com-

mand:
> Simf(subs(framechangebacksub, Simf(subs(adaptedsub1,
Simf(subs(framechangesub, Omega[3,1]))))));
˜ 1 , ω̄
Now, the coefficients of (ω̄ ˜ 2 ) in the output are, of course, equal to h̃11 , h̃12 ,
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
expression to our original expression for ω̄ ˜ 13 :

> zero2:= Simf(subs(adaptedsub2, Omega[3,1]) - %);

The coefficients of this expression must both be zero, which gives us two
equations that can be solved for h̃11 and h̃12 . These equations can be ex-
tracted as follows:
> eqns:= {op(ScalarForm(zero2))};
Before solving these equations, we might as well compute the analogous
˜ 23 . We can add these to our
equations that result from consideration of ω̄
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))};
Now solve these equations for the functions (h̃ij ):
> solve(eqns, {H[1,1], H[1,2], H[2,2]});
Then, we might as well actually assign these values to the (h̃ij ):
> assign(%);
Now, it’s not entirely obvious how to recognize these expressions as those of
equation (4.7), but we can at least check that our computations are consis-
tent with these expressions. First, define matrices [hij ], [h̃ij ], B as follows:
> 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 ω̄21 with coefficients 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 dω̄ 1 in two ways: by first making the substitution into ω̄ 1
and then differentiating, and by applying the structure equations and then
making the substitution. Then the difference of the resulting expressions
must be equal to zero:
> Simf(d(Simf(subs(examplesub, omega[1])))
- subs(examplesub, Simf(d(omega[1]))));

(sinh(v) − cosh(v) a) d(v) &ˆ d(u)

> a:= solve(%, a);

4.6. Maple computations 139

sinh(v)
a :=
cosh(v)

An analogous computation for dω̄ 2 yields b = 0; once we have made this

assignment, we will have ω̄21 = 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 (dω̄ji )
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

√ √
ω̄ 1 = E du, ω̄ 2 = G dv.
Then, in order for the second fundamental form to have the desired form,
we must have
e g
ω̄13 = √ du, ω̄23 = √ dv.
E G
1
Moreover, ω̄2 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 isn’t
strictly necessary, but it will make the output of some computations look
nicer.)
> unassign(’a’, ’b’);
> 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 coor-
dinate 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 coefficients in ω̄21 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 dω̄21 :
> Simf(d(Simf(subs(coordsub, omega[1,2])))
- subs(coordsub, Simf(d(omega[1,2]))));
> Gausseq1:= pick(%, d(u), d(v));
Now, you’ll probably notice that this expression doesn’t look quite like the
one in part (e) of the exercise. But we can ask Maple to compare the two
expressions to confirm 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 first 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));

Similar manipulations involving dω̄13 and dω̄23 will confirm that their struc-
ture equations are equivalent to the Codazzi equations (4.12).
Now, in fact, there’s 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. You’ll 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 field and then compute the dual forms (ω̄ 1 , ω̄ 2 ) for
this frame field. (This part isn’t too bad to do by hand.) Details are given
in the Maple worksheet for this chapter on the AMS webpage.
 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 §3.5, such curves represent the world lines of particles in
spacetime, and special relativity describes how particles behave in the ab-
sence of a gravitational field. The effects 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 field is reflected 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 confine our attention to the study of curves and surfaces in the
Minkowski space M1,2 . In §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 §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 pseudo-
Riemannian 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 α : I → M1,2 , where I ⊂ R

is some open interval. M1,2 has the structure of the homogeneous space
M (1, 2)/SO + (1, 2), so an adapted frame field along α should be a lifting
α̃ : I → M (1, 2). Any such lifting can be written as

α̃(t) = (α(t); e0 (t), e1 (t), e2 (t)),

where for each t ∈ I, (e0 (t), e1 (t), e2 (t)) is an oriented, orthonormal basis
for the tangent space Tα(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 field is usually called an
orthonormal frame field along α.
As in the Euclidean case, we say that α is regular if α (t) = 0 for every
t ∈ I; henceforth, we will only consider regular curves. The construction of
an orthonormal frame field along α is very similar to that for a curve in E3 ;
we begin by setting
α (t)
e0 (t) =  ,
|α (t)|
where |α (t)| = α (t), α (t) is computed using the Minkowski inner prod-
uct; i.e., we take e0 (t) to be the unit tangent vector to the curve at α(t).
(Note that, since α is a timelike curve, e0 (t) is a timelike vector, as desired.)
The Minkowski analog of arc length is the following:

Definition 5.1. Given a fixed point t0 ∈ I, the proper time function along
α is
& t
τ (t) = |α (u)| du.
t0

This terminology arises from the key fact that in relativity, time is not
an absolute quantity; different observers may measure the passage of time
differently, and the proper time along a world line α(t) represents how time
passes for an observer traveling along α.

Remark 5.2. The use of the letter τ to denote proper time is traditional
in relativity; however, it should not be confused with the torsion of a non-
degenerate curve in E3 , which is also denoted by τ ! 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 κ1 , κ2 .
5.2. Moving frames for timelike curves in M1,2 145

Exercise 5.3. Show that τ (t) is invariant under the action of M (1, 2); that
is, for any g ∈ M (1, 2), the curves α and g · α have the same proper time
function.

*Exercise 5.4. Consider two particles with world lines α, β : I → M1,2

given by
α(t) = t[t, 0, 0], β(t) = t[t, vt, 0],
so that α represents a particle that remains stationary at the origin in R2
and β 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 for√α (with t0 = 0) is τα (t) = t and
the proper time function for β is τβ (t) = t 1 − v 2 .
(b) Reparametrize β according to its proper time function:
t 
τ vτ
β(τ ) = β(t(τ )) = √ , √ ,0 .
1 − v2 1 − v2
Let u = tanh−1 (v), so that v = tanh(u), and consider the Lorentz transfor-
mation T : M1,2 → M1,2 defined by
⎡ 0⎤ ⎛⎡ 0 ⎤⎞ ⎡ ⎤ ⎡ 0⎤
x̃ x cosh(u) − sinh(u) 0 x
⎢ 1⎥ ⎜⎢ 1 ⎥⎟ ⎢ ⎥ ⎢ 1⎥
⎣x̃ ⎦ = T ⎝⎣x ⎦⎠ = ⎣− sinh(u) cosh(u) 0⎦ ⎣x ⎦ .
x̃2 x2 0 0 1 x2

Show that with respect to the (x̃0 , x̃1 , x̃2 )-coordinates,

t 
t −vt
α(t) = √ , √ ,0 , β(τ ) = t[τ, 0, 0].
1 − v2 1 − v2
So in this new coordinate system, β represents a stationary particle and α
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 − tanh2 (u).)
(c) Let t1 = τ1 = T . Show that in either coordinate system, the points α(t1 )
and β(τ1 ) both lie on the hyperboloid
{t[x0 , x1 , x2 ] ∈ M1,2 | (x0 )2 − (x1 )2 − (x2 )2 = T 2 }.
This hyperboloid represents the points that can be reached from the origin in
proper time τ = T by particles traveling with any constant spatial velocity
v = v 1 ∂x∂ 1 + v 2 ∂x∂ 2 with (v 1 )2 + (v 2 )2 = v 2 < 1.
146 5. Curves and surfaces in Minkowski space

Henceforth, we will assume that α is parametrized by proper time τ , and so

e0 (τ ) = α (τ ).
As in the Euclidean case, we will say that a timelike curve α is nondegenerate
if α is regular and, in addition, e0 (τ ) = 0 for all τ ∈ I. In this case,
differentiating the equation
e0 (τ ), e0 (τ ) = 1
yields
e0 (τ ), e0 (τ ) = 0.
Thus, e0 (τ ) is orthogonal to e0 (τ ), and we define
e0 (τ )
e1 (τ ) = .
|e0 (τ )|
This vector will be called the unit normal vector to the curve at α(τ ); note
that e1 (τ ) is a spacelike vector (cf. Exercise 3.39).
By analogy with the Euclidean case, at this point we would like to define
e2 (τ ) = e0 (τ ) × e1 (τ ).
But first, we have to define 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 × w = t[v 2 w3 − v 3 w2 , v 3 w1 − v 1 w3 , v 1 w2 − v 2 w1 ],
often written schematically as
 
 e1 e2 e3 
 
 
v × w =  v1 v2 v3  .
 
w 1 w 2 w 3 

The analogous notion for Minkowski space is

Definition 5.5. Let v = t[v 0 , v 1 , v 2 ], w = t[w0 , w1 , w2 ] ∈ M1,2 . The Min-
kowski cross product v × w is
v × w = t[v 1 w2 − v 2 w1 , v 0 w2 − v 2 w0 , v 1 w0 − v 0 w1 ],
written schematically as
 
 e0 e1 e2 

 
v × w =  v 0 −v 1 −v 2  .
 
w0 −w1 −w2 

The following exercise shows that the Minkowski cross product has proper-
ties 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 × w is orthogonal to both v and w.
(b) Show that
v × w, v × w = v, vw, w − v, w2 .
Conclude that if v, w are orthogonal vectors of Minkowski norm 1, then
v × w also has Minkowski norm equal to 1.
(c) Show that for any three vectors u, v, w ∈ M1,2 ,

u, v × w = det u v w .
(This is an orientation condition that dictates how the sign of v × w should
be chosen.)

As a consequence of Exercise 5.6, we can complete our adapted frame field

along α by defining
e2 (τ ) = e0 (τ ) × e1 (τ ).
Exercise 5.7. Prove that this choice of frame field (e0 (τ ), e1 (τ ), e2 (τ )) is
equivariant under the action of M (1, 2): If we replace α by g · α for some
g ∈ M (1, 2), then eα (τ ) ∈ Tα(τ ) M1,2 will be replaced by (Lg )∗ (eα (τ )) ∈
Tg·α(τ ) M1,2 .

We now have a canonical adapted frame field (e0 (τ ), e1 (τ ), e2 (τ )) defined at

each point of α(τ ), which in turn defines a canonical, left-invariant lifting
α̃ : I → M (1, 2) for any nondegenerate timelike curve α, given by
α̃(τ ) = (α(τ ); e0 (τ ), e1 (τ ), e2 (τ )) .
Now consider the pullbacks of equations (3.1) to I via α̃; these can be written
as
α (τ )dτ = eα (τ ) ω̄ α ,
(5.1)
eα (τ )dτ = eβ (τ ) ω̄αβ .
Recall that we constructed our adapted frame field so that α (τ ) = e0 (τ );
therefore, the first equation in (5.1) implies that
ω̄ 0 = dτ, ω̄ 1 = ω̄ 2 = 0.
Then we chose e1 (τ ) so that e0 (τ ) is a multiple of e1 (τ ), say e0 (τ ) =
κ1 (τ )e1 (τ ). The function κ1 (τ ) is the analog of the curvature κ(s) for curves
in E3 ; note that α is nondegenerate if and only if κ1 (τ ) > 0 for all τ ∈ I.
We will call κ1 (τ ) the Minkowski curvature of α. So the equation for e0 (τ )
in (5.1) implies that
ω̄01 = κ1 (τ )dτ, ω̄02 = 0.
148 5. Curves and surfaces in Minkowski space

(Recall from Exercise 3.50 that ω̄00 = 0.) The only remaining Maurer-Cartan
form is ω̄21 ; it must be equal to some multiple of dτ , so define a function κ2 (τ )
by the condition that
ω̄21 = κ2 (τ )dτ.
The function κ2 (τ ) is the analog of the torsion τ (s) for curves in E3 . We
will call κ2 (τ ) the Minkowski torsion of α.
Using the symmetry relations between the (ω̄βα ) from Exercise 3.50, we have
the Minkowski analog of the Frenet equations for timelike curves:
⎡ ⎤
0 0 0 0
⎢ ⎥
  ⎢1 0 κ1 (τ ) 0 ⎥
α (τ ) e0 (τ ) e1 (τ ) e2 (τ ) = α(τ ) e0 (τ ) e1 (τ ) e2 (τ ) ⎢
    ⎥
⎢0 κ (τ ) 0 κ (τ )⎥ .
⎣ 1 2 ⎦
0 0 −κ2 (τ ) 0

Applying Lemma 4.2 yields the following theorem:

Theorem 5.8. Two nondegenerate timelike curves α1 , α2 : I → M1,2 para-
metrized by proper time differ by a Lorentz transformation if and only if they
have the same Minkowski curvature κ1 (τ ) and Minkowski torsion κ2 (τ ).
*Exercise 5.9. Consider the world line α 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) =, (x1 ) (0) = 0,
a
where a > 0 is the constant of proportionality for which
(x1 ) (τ ) = a2 x1 (τ ).
(Note that α and all its derivatives will be contained in the plane spanned
by (e0 , e1 ).)
(a) Show that
t 
1 1
α(τ ) = sinh(aτ ), cosh(aτ ), 0 .
a a
(Hint: Let α(τ ) = t[x0 (τ ), x1 (τ ), 0]. Solve the given initial value problem
for x1 (τ ), and then use the condition that α is timelike and α (τ ) = 1 to
solve for x0 (τ ).)
(b) Show that the world line of α lies on a hyperbola in the (x0 , x1 )-plane.
(c) Show that the Minkowski curvature and torsion of α are
κ1 (τ ) = a, κ2 (τ ) = 0.
5.3. Moving frames for timelike surfaces in M1,2 149

Thus, α 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 α of a particle moving along a
circle of radius r in the (x1 , x2 )-plane with what a stationary observer be-
lieves to be constant angular velocity k > 0. This means that α has a
parametrization of the form
α(t) = t[t, r cos(kt), r sin(kt)].

(a) Show that the proper time function for α is

τ (t) = t 1 − k 2 r2 ;
therefore, the proper time parametrization for α is
t " # " #
τ kτ kτ
α(τ ) = √ , r cos √ , r sin √ .
1 − k2 r2 1 − k2 r2 1 − k2 r2
(Note that this means that an observer traveling along α believes its angular
k
velocity to be k̃ = √1−k 2 r2
.)

(b) Compute the Frenet frame (e0 (τ ), e1 (τ ), e2 (τ )) for α and show that the
Minkowski curvature and torsion of α are
k2r k
κ1 (τ ) = , κ2 (τ ) = .
1−k r
2 2 1 − k2 r2
Thus, α is the Minkowski analog of a helix in E3 , i.e., a nondegenerate
curve with constant nonzero curvature and torsion. (This shouldn’t be too
surprising since the world line of α 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 → M1,2 be an immersion

whose image is a timelike surface Σ = x(U ). Just as for curves, an adapted
frame field along Σ is a lifting x̃ : U → M (1, 2) of the form
x̃(u) = (x(u); e0 (u), e1 (u), e2 (u)) ,
where for each u ∈ 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) Σ. (Note that, since Σ 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) Σ. 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) Σ. An orthonormal frame field satisfying this condition will be called
adapted.
Let (ω̄ α , ω̄βα ) represent the pulled-back forms (x̃∗ ω α , x̃∗ ωβα ) on U . Precisely
the same reasoning as in the Euclidean case can be used to prove the fol-
lowing:
Proposition 5.12. Let U ⊂ R2 be an open set, and let x : U → M1,2
be a timelike immersion. For any adapted orthonormal frame field (e0 (u),
e1 (u), e2 (u)) along Σ = x(U ), the associated dual and connection forms
(ω̄ α , ω̄βα ) have the property that ω̄ 2 = 0. Moreover, (ω̄ 0 , ω̄ 1 ) form a basis for
the 1-forms on U .

The metric properties of the surface Σ are once again contained in the first
fundamental form of the surface:
Definition 5.13. Let U ⊂ R2 be an open set, and let x : U → M1,2 be a
timelike immersion. The first fundamental form of Σ = x(U ) is the quadratic
form I on T U defined by
I(v) = dx(v), dx(v)
for v ∈ Tu U , where ·, · is the Minkowski inner product.
*Exercise 5.14. Show that for any v ∈ Tu U ,
 2  2
I(v) = ω̄ 0 (v) − ω̄ 1 (v) .
This is often written more concisely as
I = (ω̄ 0 )2 − (ω̄ 1 )2 .
Note that I is a quadratic form of signature (1, 1) rather than a positive
definite quadratic form. This reflects the fact that Σ is a timelike surface;
if Σ were spacelike, then its first fundamental form would have signature
(0, 2).

As in the Euclidean case, differentiating the equation ω̄ 2 = 0 yields

dω̄ 2 = −ω̄02 ∧ ω̄ 0 − ω̄12 ∧ ω̄ 1 = 0,
and Cartan’s lemma (cf. Lemma 2.49) implies that there exist functions
h00 , h01 , h11 on U such that
 2   
ω̄0 h00 h01 ω̄ 0
=− .
ω̄12 h01 h11 ω̄ 1
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 (hαβ ) are related to the differential of the Gauss
map. The Gauss map and the second fundamental form are defined essen-
tially as before:
Definition 5.15. Let U ⊂ R2 be an open set, and let x : U → M1,2 be a
timelike immersion with image Σ = x(U ). The Gauss map of Σ = x(U ) is
the map N : Σ → M1,2 defined by
N (x(u)) = e2 (u),
where (e0 (u), e1 (u), e2 (u)) is any adapted frame field on Σ = x(U ). (Note
that N takes values in the “sphere” S−1 .)
Definition 5.16. Let U ⊂ R2 be an open set, and let x : U → M1,2 be
a timelike immersion. The second fundamental form of Σ = x(U ) is the
quadratic form II on T U defined by
II(v) = −de2 (v), dx(v)
for v ∈ Tu U , where (e0 (u), e1 (u), e2 (u)) is any adapted frame field on Σ =
x(U ).
*Exercise 5.17. Show that for any v ∈ Tu U ,
 
II(v) = − ω̄02 (v) ω̄ 0 (v) + ω̄12 (v) ω̄ 1 (v)
= h00 (ω̄ 0 (v))2 + 2h01 ω̄ 0 (v) ω̄ 1 (v) + h11 (ω̄ 1 (v))2 .
This is often written more concisely as
II = −(ω̄02 ω̄ 0 + ω̄12 ω̄ 1 ) = h00 (ω̄ 0 )2 + 2h01 ω̄ 0 ω̄ 1 + h11 (ω̄ 1 )2 .
(Hint: Although the result looks the same as in the Euclidean case, there are
some sign differences in the details of the computation. Note that, because
of the slightly different symmetries in the (ω̄βα ), we have
de2 = e0 ω20 + e1 ω21 = e0 ω02 − e1 ω12 .
And don’t forget to use the Minkowski inner product in the definition of II!)

Now we will examine how the matrix [hαβ ] changes if we vary the frame.
Let (e0 (u), e1 (u), e2 (u)) be any adapted frame field along Σ, with associated
Maurer-Cartan forms (ω̄ α , ω̄βα ). Any other adapted frame field (ẽ0 (u), ẽ1 (u),
ẽ2 (u)) has the form (up to sign)
(5.2) ⎡ ⎤
cosh(θ) sinh(θ) 0
 ⎢ ⎥
ẽ0 (u) ẽ1 (u) ẽ2 (u) = e0 (u) e1 (u) e2 (u) ⎣ sinh(θ) cosh(θ) 0⎦
0 0 1
152 5. Curves and surfaces in Minkowski space

˜ α , ω̄
for some function θ on U . Let (ω̄ ˜ βα ) be the Maurer-Cartan forms associ-
 
cosh(θ) sinh(θ)
ated to the new frame field, and set B = .
sinh(θ) cosh(θ)

*Exercise 5.18 (Cf. Exercise 4.28). (a) Show that

 0  0
˜
ω̄ ω̄
−1
=B .
˜1
ω̄ ω̄ 1

(b) Show that

 0  
˜2
ω̄ ω̄20
= B −1 .
˜ 21
ω̄ ω̄21
(Hint: Use the equation for de2 in (3.1).)
(c) Cartan’s lemma implies that there exist functions h̃00 , h̃01 , h̃11 on U such
that  2    0
˜0
ω̄ h̃00 h̃01 ω̄˜
=− .
˜ 12
ω̄ h̃01 h̃11 ω̄˜1
Show that    
−h̃00 −h̃01 −h00 −h01
−1
=B B.
h̃01 h̃11 h01 h11
(Hint: Use the fact that ω̄20 = ω̄02 , ω̄21 = −ω̄12 .)
(d) Use part (c) to show that
   
h̃00 h̃01 h00 h01
t
(5.3) = B B.
h̃01 h̃11 h01 h11
(Hint: Note that B −1 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) Σ → Tx(u) Σ is self-adjoint; i.e., for any two
vectors v, w ∈ Tx(u) Σ, we have
de3 (v), w = v, de3 (w).
The formal statement of the linear algebra theorem is that a self-adjoint op-
erator 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 ω̄20 + e1 ω̄21 = e0 (−h00 ω̄ 0 − h01 ω̄ 1 ) + e1 (h01 ω̄ 0 + h11 ω̄ 1 ).
(You probably did this as part of Exercise 5.18.) Conclude that the matrix
of the linear transformation de2 : Tx(u) Σ → Tx(u) Σ with respect to the basis
(e0 (u), e1 (u)) for Tx(u) Σ is
 
−h00 −h01
.
h01 h11

(b) Show that the linear transformation de2 : Tx(u) Σ → Tx(u) Σ is self-
adjoint with respect to the Minkowski metric; that is, for any two vectors
v, w ∈ Tx(u) Σ, 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:

• If |h00 + h11 | > 2|h01 |, then de2 has distinct, real eigenvalues and
an orthogonal basis of eigenvectors, one timelike and one spacelike.
• If |h00 + h11 | < 2|h01 |, then de2 has complex eigenvalues and no
real eigenvectors.
• If |h00 + h11 | = 2|h01 | = 0, then de2 is a multiple of the identity
transformation.
• 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.
Definition 5.20. The function K = det(de2 ) on Σ is called the Gauss
curvature of Σ. The function H = − 12 tr(de2 ) on Σ is called the mean
curvature of Σ.
*Exercise 5.21. Show that K = h201 − h00 h11 and H = 12 (h00 − h11 ).
154 5. Curves and surfaces in Minkowski space

√
Definition 5.22. The function H  = H 2 − K on Σ is called the skew
curvature of Σ. (By convention, if H 2 − K < 0, then H  is chosen so that
iH  < 0.)

*Exercise 5.23. (a) Show that for a regular surface Σ 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 − 4h201 .

(c) Conclude from the result of Exercise 5.19 that at any point x(u) ∈ Σ:

• 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.

• If H  (u) is imaginary, then d(e2 )x(u) has complex eigenvalues and

no real eigenvectors.

• If H  (u) = 0 and h01 = 0, then d(e2 )x(u) is a multiple of the identity

transformation.

• 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 refinements to our adapted frame field.

5.3.1. Case 1: H  (u) is real and d(e2 )x(u) has an orthogonal ba-
sis of eigenvectors for all u ∈ U . This assumption covers the first
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) ∈ Σ with the property that
   
−h00 (u) −h01 (u) −κ0 (u) 0
=
h01 (u) h11 (u) 0 −κ1 (u)

for some real numbers κ0 (u), κ1 (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)) = −κ0 e0 (u), d(e2 )u (e1 (u)) = −κ1 e1 (u).
5.3. Moving frames for timelike surfaces in M1,2 155

We have the following analog of Definition 4.30:

Definition 5.24. The vectors e0 (u) and e1 (u) in equations (5.4) are called
principal vectors or principal directions at x(u) ∈ Σ, and κ0 (u), κ1 (u) are
called the principal curvatures of Σ at x(u). Such a frame (e0 (u), e1 (u),
e2 (u)) is called a principal adapted frame at the point x(u) ∈ Σ, and an
adapted frame field on Σ which has this property at every point x(u) ∈ Σ
is called a principal adapted frame field on Σ.
Exercise 5.25. Show that K = κ0 κ1 and H = 12 (κ0 + κ1 ).
Definition 5.26. If κ0 (u) = κ1 (u) for some point u ∈ U , then the corre-
sponding point x(u) of Σ is called an umbilic point of Σ.

If Σ has no umbilic points, then a principal adapted frame field along Σ 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 Σ are precisely those
points where H  = 0.
*Exercise 5.28. (a) Show that if Σ has no umbilic points, then a principal
adapted frame field (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 field, the second fundamental
form is
II = κ0 (ω̄ 0 )2 − κ1 (ω̄ 1 )2 .

The analog of Bonnet’s theorem (cf. Theorem 4.35) is the following:

Theorem 5.29. Let U ⊂ R2 be an open set. Two timelike immersions
x1 , x2 : U → M1,2 for which the differential of the Gauss map has two
linearly independent real eigenvectors differ by a Lorentz transformation if
and only if they have the same first 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 → M1,2 be immersions as in Theorem 5.29, with
the same first and second fundamental forms and κ0 (u) = κ1 (u) for every
u ∈ U.
(a) Let x̃1 , x̃2 : U → M (1, 2) be principal adapted frame fields for x1 , x2 .
Let (ω̄ α , ω̄βα ) denote the pulled-back dual and connection forms for x1 and
156 5. Curves and surfaces in Minkowski space

let (Ω̄α , Ω̄αβ ) denote those for x2 . Show that up to sign,

Ω̄α = ω̄ α , Ω̄αβ = ω̄βα .
Moreover, by making appropriate sign changes in the principal adapted
frame fields, we can arrange that these equations hold exactly.
(b) Use Lemma 4.2 to conclude that x1 and x2 differ by a Lorentz transfor-
mation.
*Exercise 5.31. Let x : U → M1,2 be a timelike immersion as in Theorem
5.29, and let (u, v) be local coordinates on U . We can write the first 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 Σ = x(U ) is timelike
implies that EG − F 2 < 0.
Given a surface Σ of this type with no umbilic points, there exists a para-
metrization x : U → M1,2 of Σ (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 → M1,2 is a timelike immersion as
in Theorem 5.29, with the property that every point of Σ = x(U ) is umbilic.
(a) Show that any adapted frame field (e0 (u), e1 (u), e2 (u)) is a principal
adapted frame field.
(b) Let (ω̄ α , ω̄βα ) be the Maurer-Cartan forms for an adapted frame field on
Σ = x(U ). Show that there exists a smooth function λ : U → R such that
ω̄02 = −λω̄ 0 , ω̄12 = λω̄ 1 .
Conclude that the second fundamental form of Σ is a scalar multiple of the
first fundamental form, i.e., that
II = λI.

(c) Prove that λ is constant. (Hint: Use the structure equations to dif-
ferentiate the equations above, taking into account the fact that we must
5.3. Moving frames for timelike surfaces in M1,2 157

have
dλ = λ0 ω̄ 0 + λ1 ω̄ 1
for some functions λ0 , λ1 on U . Then use Cartan’s lemma.)
(d) Show that if λ = 0, then de2 = 0. Conclude that the normal vector field
of Σ is constant and that Σ is therefore contained in a plane.
(e) Show that if λ = 0, then d(x+ λ1 e2 ) = 0. Conclude that the vector-valued
function x + λ1 e2 : U → M1,2 is equal to some constant point q ∈ M1,2 and
that Σ is therefore contained in the hyperboloid of one sheet defined by the
equation
1
x − q, x − q = − 2 .
|λ|
(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 “sphere” S− 1 .)
λ2

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-flat model for general relativity. It can be
defined as the set of spacelike unit vectors in the Minkowski space M1,4 :
dS4 = {x ∈ M1,4 | x, x = −1}.
De Sitter space is the maximally symmetric vacuum solution of Einstein’s
field equations with a positive cosmological constant. It represents a universe
that is spatially homogeneous, diffeomorphic 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 “sphere”
S−1 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 diffeomorphic
to a circle.
*Exercise 5.34. We can parametrize dS2 by the map x : R2 → 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 param-
eter.
(a) Show that the metric on dS2 corresponding to the parametrization (5.5)
has
E = 1, F = 0, G = − 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 = τ is

& π
xv (τ, v) dv = 2π cosh(τ ).
−π
Thus, the “radius” of the space at time τ is cosh(τ ), which grows exponen-
tially as τ increases for τ > 0.

Now consider the possible world lines for particles traveling in the de Sitter
space dS2 . Since dS2 is spatially finite, we might think that a moving particle
should be able to “circumnavigate” 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 β of a photon emitted at the point
x(0, 0) and traveling in the positive v direction. We can parametrize the
photon’s world line as
β(t) = x(t, v(t))
for some function v(t) which is determined by the conditions that v(0) = 0,
v  (t) > 0, and β  (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 v± = cosh(u0 ) xu ± xv .
(b) Show that the function v(t) above must satisfy the differential equation
v  (t) = sech(t)
and that the solution to this equation with v(0) = 0 and v  (t) > 0 is
  π
v(t) = 2 tan−1 et − .
2

(c) Show that

π π
lim v(t) = , lim v(t) = − .
t→+∞ 2 t→−∞ 2
Therefore, no photon ever travels more than halfway around the circle.
(And, of course, no other particle can travel farther than a photon!)

5.3.2. Case 2: H  (u) is imaginary for all u ∈ 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),
h̃00 + h̃11 = cosh(2θ)(h00 + h11 ) + 2 sinh(2θ)h01 .
Conclude that there exists an adapted frame field (e0 (u), e1 (u), e2 (u)) with
the property that
h00 + h11 = 0,
5.3. Moving frames for timelike surfaces in M1,2 159

and therefore,    
h00 h01 λ0 λ1
=
h01 h11 λ1 −λ0
for some functions λ0 , λ1 on U with |λ1 | > 0 (cf. Exercise 5.18).
(b) Show that the Maurer-Cartan forms associated to such a frame field
satisfy
ω̄02 = −λ0 ω̄ 0 − λ1 ω̄ 1 ,
ω̄12 = −λ1 ω̄ 0 + λ0 ω̄ 1 .

(c) Show that for such a frame field, the second fundamental form is
 
II = λ0 (ω̄ 0 )2 − (ω̄ 1 )2 + 2λ1 ω̄ 0 ω̄ 1 = λ0 I + 2λ1 ω̄ 0 ω̄ 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 Σ might look like, consider the
case where Σ 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 → M1,2 of
the form
x(u, v) = t[u, v, f (u, v)].
Exercise 5.37. Show that Σ is timelike if and only if |fu | < 1.

Now choose a point (u0 , v0 ) ∈ U , and “rotate” the surface (via a Lorentzian
transformation) if necessary so that
fu (u0 , v0 ) = fv (u0 , v0 ) = 0,
so that the tangent plane to Σ at the point x0 = x(u0 , v0 ) is parallel to the
(x0 , x1 )-plane.
Exercise 5.38. (a) Show that the normal vector field along Σ is
1
e2 (u, v) = [−fu , fv , −1].
t
1 + fv2 − fu2
In particular,
e2 (u0 , v0 ) = t[0, 0, −1].

(b) Choose an adapted frame field along Σ 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 (ω̄ α , ω̄βα ) be the associated Maurer-Cartan forms. Show that at the
point x0 ,
ω̄ 0 = du, ω̄ 1 = dv,
ω̄02 = −fuu du − fuv dv, ω̄12 = −fuv du − fvv dv.
Therefore,
   
h00 h01  fuu (u0 , v0 ) fuv (u0 , v0 )
 = .
h01 h11  fuv (u0 , v0 ) fvv (u0 , v0 )
(u0 ,v0 )

(c) Conclude that the graph of the function

f (u, v) = a(u2 − 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 ∈ U . Note that this
is a degenerate condition, and generically it will only hold along a closed
subset of Σ, similar to the set of umbilic points for a surface in Case 1.
Definition 5.39. A point x(u) in a timelike surface Σ = x(U ) in M1,2 will
be called a quasi-umbilic point if H  (u) = 0 and h01 (u) = 0. If every point
of Σ is quasi-umbilic, then Σ is called totally quasi-umbilic.
Exercise 5.40. Suppose that x : U → M1,2 is a totally quasi-umbilic time-
like 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),

h̃01 = e2θ h01 .
Conclude that there exists an adapted frame field (e0 (u), e1 (u), e2 (u)) with
the property that
|h01 | = 12 |h00 + h11 | = 1,
and therefore,    
h00 h01 ε1 + λ ε2
= ,
h01 h11 ε2 ε1 − λ
where λ is a function on U and εi = ±1, i = 1, 2. (In fact, λ is equal to the
mean curvature H of Σ.)
(c) Show that the Maurer-Cartan forms associated to such a frame field
satisfy
ω̄02 = −(ε1 + λ)ω̄ 0 − ε2 ω̄ 1 ,
ω̄12 = −ε2 ω̄ 0 − (ε1 − λ)ω̄ 1 .

(d) Show that for such a frame field, the second fundamental form is
   
II = λ (ω̄ 0 )2 − (ω̄ 1 )2 + ε1 (ω̄ 0 )2 + (ω̄ 1 )2 + 2ε2 ω̄ 0 ω̄ 1
 
= λI + ε1 (ω̄ 0 )2 + (ω̄ 1 )2 + 2ε2 ω̄ 0 ω̄ 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
infinite-dimensional family of such surfaces! We will explore some examples
in Exercise 5.49.

5.4. An alternate construction for timelike surfaces

In §5.3, we used orthonormal frame fields to study timelike surfaces in M1,2 .

This is in keeping with our general perspective that a choice of frame field
along a surface x : U → G/H in a homogeneous space G/H should define
a lifting x̃ : U → G to a surface in the Lie group G. However, this is not
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 Σ = x(U ) be a timelike surface in M1,2 . Since the first 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) Σ. In this section,
162 5. Curves and surfaces in Minkowski space

we will construct frame fields (f1 (u), f2 (u), f3 (u)) along Σ = 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
field along Σ = x(U ). Show that the vectors
f1 (u) = √1 (e0 (u)
2
+ e1 (u)), f2 (u) = √1 (e0 (u)
2
− e1 (u))

are linearly independent null vectors and that

f1 (u), f2 (u) = 1.

Definition 5.42. A null adapted frame field (f1 (u), f2 (u), f3 (u)) along a
timelike surface Σ = 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) Σ
at each point x(u) ∈ Σ;
(2) f1 (u), f2 (u) = 1;
(3) f3 (u) is the normal vector f3 (u) = f1 (u) × f2 (u) to Tx(u) Σ at each
point x(u) ∈ Σ.

Now, let (η̄ i , η̄ji ) be the Maurer-Cartan forms associated to a null adapted
frame field along Σ. These forms are defined just as in (3.1):
dx = fi η̄ i ,
dfi = fj η̄ij ,
where 1 ≤ i, j ≤ 3, and they still satisfy the structure equations
dη̄ i = −η̄ji ∧ η̄ j ,
dη̄ji = −η̄ki ∧ η̄jk .

The main difference is that the vectors (f1 , f2 , f3 ) satisfy the somewhat un-
usual inner product relations
f1 , f1  = 0, f2 , f2  = 0, f3 , f3  = −1,
(5.6)
f1 , f2  = 1, f1 , f3  = 0, f2 , f3  = 0.

*Exercise 5.43. Differentiate the inner product relations (5.6) to obtain

the following relations among the (η̄ji ):

η̄21 = η̄12 = η̄33 = 0,

(5.7)
η̄22 = −η̄11 , η̄31 = η̄23 , η̄32 = η̄13 .
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 ⊂ R2 be an open set, and let x : U → M1,2 be

a timelike immersion. For any null adapted frame field (f1 (u), f2 (u), f3 (u))
along Σ = x(U ), the associated dual and connection forms (η̄ i , η̄ji ) have the
property that η̄ 3 = 0. Moreover, (η̄ 1 , η̄ 2 ) form a basis for the 1-forms on U .

*Exercise 5.45. Show that the first fundamental form of Σ = x(U ) (cf.
Definition 5.13) is
I = 2η̄ 1 η̄ 2 .

*Exercise 5.46. (a) Differentiate the equation η̄ 3 = 0 and use Cartan’s

lemma to conclude that there exist functions k11 , k12 , k22 on U such that
 3   
η̄1 k11 k12 η̄ 1
(5.8) =− .
η̄23 k12 k22 η̄ 2

(b) The Gauss map (cf. Definition 5.15) of Σ = x(U ) is the map N : Σ →
M1,2 defined by
N (x(u)) = f3 (u).
Show that
df3 = f1 η̄31 + f2 η̄32 = −f1 (k12 η̄ 1 + k22 η̄ 2 ) − f2 (k11 η̄ 1 + k12 η̄ 2 ).
Conclude that with respect to the basis (f1 (u), f2 (u)) for Tx(u) Σ, the matrix
of the linear transformation df3 : Tx(u) Σ → Tx(u) Σ is
 
k12 k22
− .
k11 k12

(c) The second fundamental form (cf. Definition 5.16) of Σ = x(U ) is defined
by
II(v) = −df3 (v), dx(v)
for v ∈ Tu U . Show that
II = k11 (η̄ 1 )2 + 2k12 η̄ 1 η̄ 2 + k22 (η̄ 2 )2 .

(d) Show that the Gauss and mean curvatures (cf. Definition 5.20) of Σ are
2
K = k12 − k11 k22 , H = k12
and that the skew curvature (cf. Definition 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

• k11 k22 > 0 (H  real and H  > 0);

• k11 k22 < 0 (H  imaginary);
• k11 = k22 = 0 (umbilic points);
• 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 field
along Σ, with associated Maurer-Cartan forms (η̄ i , η̄ji ).

(a) Show that any other null adapted frame field (f̃1 (u), f̃2 (u), f̃3 (u)) has the
form (up to sign)
⎡ θ ⎤
e 0 0
 ⎢ ⎥
(5.9) f̃1 (u) f̃2 (u) f̃3 (u) = f1 (u) f2 (u) f3 (u) ⎣ 0 e−θ 0⎦
0 0 1
for some function θ on U . (Hint: Consider the transformation (5.2) for the
adapted orthonormal frame field
e0 (u) = √1 (f1 (u)
2
+ f2 (u)), e1 (u) = √1 (f1 (u)
2
− f2 (u)), e2 (u) = f3 (u).)

(b) Let (η̄˜i , η̄˜ji ) be the Maurer-Cartan forms associated to the new frame
 
eθ 0
field, and set B = . Show that
0 e−θ
 1  1   −θ 1   1  1   −θ 1 
η̄˜ η̄ e η̄ η̄˜3 η̄3 e η̄3
−1 −1
(5.10) =B = , =B = .
η̄˜2 η̄ 2 eθ η̄ 2 η̄˜32 η̄32 eθ η̄32

(c) Cartan’s lemma implies that there exist functions k̃11 , k̃12 , k̃22 on U such
that
 3   
η̄˜1 k̃11 k̃12 η̄˜1
=− .
η̄˜23 k̃12 k̃22 η̄˜2
Show that
   
k̃12 k̃22 k12 k22
= B −1 B
k̃11 k̃12 k11 k12
5.4. An alternate construction for timelike surfaces 165

and that
     
k̃11 k̃12 k11 k12 e2θ k11 k12
(5.11) = tB B= .
k̃12 k̃22 k12 k22 k12 e−2θ k22

The result of Exercise 5.47 provides a guide as to how to choose a canonical

null adapted frame field in most cases:

(1) If H  = 0, then there exists a null adapted frame field (f1 (u),
f2 (u), f3 (u)) (unique up to sign) along Σ with the property that
k11 = ±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 field 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 field (f1 (u), f2 (u), f3 (u)) (unique
up to sign) along Σ with the property that k11 = ±1.

*Exercise 5.48 (Cf. Exercise 5.31). A timelike surface Σ always has a

local parametrization x : U → 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 α : I → M1,2 be a nondegenerate null curve, i.e., a curve

with the property that α (u) is a null vector for all u ∈ I. (The condition that
α is nondegenerate means that α (u) and α (u) are linearly independent for
all u ∈ I.) Let f 0 be any null vector that is linearly independent from α (u)
for all u ∈ I, and consider the cylindrical surface Σ with parametrization
x : I × R → M1,2 given by

x(u, v) = α(u) + vf 0 .

(a) Show that α can be reparametrized so as to satisfy the condition

α (u), f 0  = 1.
166 5. Curves and surfaces in Minkowski space

(b) With α as in part (a), show that the vector fields

f1 (u, v) = α (u), f2 (u, v) = f 0 , f3 (u, v) = f1 (u, v) × f2 (u, v)

form a null adapted frame field along Σ.

(c) Show by direct computation of dx, df2 , and df1 (preferably in that order!)
that the Maurer-Cartan forms associated to this frame field satisfy

η̄ 1 = du, η̄ 2 = dv,
η̄31 = η̄23 = 0, η̄32 = η̄13 = λ du

for some nonvanishing function λ : I → R.

(d) Conclude that k11 = −λ, k12 = 0, k22 = 0 and that Σ is therefore totally
quasi-umbilic.
(e) Show that Σ has Gauss and mean curvatures

K ≡ 0, H ≡ 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 Σ ⊂ M1,2 is a timelike surface with
K ≡ H ≡ 0, then every point of Σ 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 different. 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
modifications 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 field along a timelike surface in M1,2 . After loading the Cartan and
LinearAlgebra packages into Maple, begin by declaring the necessary 1-
forms:
> 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]:= 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];
Tell Maple how to differentiate these forms according to the Cartan struc-
ture equations (3.8):
> for i from 1 to 3 do
d(eta[i]):= -add(’eta[i,j] &ˆ eta[j]’, j=1..3);
end do;
d(eta[1,1]):= -add(’eta[1,k] &ˆ eta[k,1]’, k=1..3);
d(eta[3,1]):= -add(’eta[3,k] &ˆ eta[k,1]’, k=1..3);
d(eta[3,2]):= -add(’eta[3,k] &ˆ eta[k,2]’, k=1..3);
Set up a substitution for the Maurer-Cartan forms of a null adapted frame
field, taking into account the relations (5.8) that result from computing
dη̄ 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 (k̃ij ) 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 (k̃ij ) 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 → M1,2 of a timelike surface with F > 0,
we can define a null adapted frame field by
1 1
f1 (u) = √ xu , f2 (u) = √ xv , f3 (u) = f1 (u) × f2 (u).
F F
The associated dual forms are
√ √
η̄ 1 = F du, η̄ 2 = F dv, η̄ 3 = 0,

and in order for the second fundamental form to have the desired form, we
must have
" # " #
e f f g
η̄1 = − √ du + √ dv ,
3
η̄2 = − √ du + √ dv .
3
F F F F

Introduce a substitution for the Maurer-Cartan forms in terms of the coor-

dinate 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 dη̄ 1 and dη̄ 2 to compute the coefficients in
η11 :
> 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 dη̄11 :
> 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 defined to be K = f F−eg

2
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 − Fu Fv ) = .
F F

Similar manipulations involving dη̄13 and dη̄23 show that the Codazzi equa-
tions take the form
" # " #
F fu − f Fu f F fv − f Fv f
ev = =F , gu = =F .
F F u F F v
 Chapter 6

Curves and surfaces

in equi-affine space

6.1. Introduction

Now we will apply the method of moving frames to study the geometry
of curves and surfaces in equi-affine 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-affine group.

Exercise 6.1. Show by example that an equi-affine transformation T :

R3 → R3 does not necessarily preserve the curvature κ(s) and torsion τ (s)
of a curve α : I → E3 . (Hint: Let α 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-affine 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-affine space

6.2. Moving frames for curves in A3

Consider a smooth, parametrized curve α : I → A3 that maps some open

interval I ⊂ R into equi-affine space. A3 has the structure of the homo-
geneous space A(3)/SL(3), so an adapted frame field along α should be a
lifting α̃ : I → A(3). Any such lifting can be written as

α̃(t) = (α(t); e1 (t), e2 (t), e3 (t)),

where for each t ∈ I, (e1 (t), e2 (t), e3 (t)) is a unimodular basis for the tangent
space Tα(t) A3 . Such an adapted frame field is called a unimodular frame
field along α. The situation is quite different 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 first derivative of α to choose e1 (t)
and the second derivative of α 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 α. In order for this
procedure to work, we had to assume that α was “nondegenerate”, i.e., that
the vectors (α (t), α (t)) were linearly independent for each t ∈ I.
So, how might we use similar reasoning to construct a canonical adapted
frame field in the equi-affine case, and what would be the right notion of
“nondegenerate” for equi-affine curves? Since orthonormality is no longer
required, our first guess towards constructing an adapted frame field might
be to take

e1 (t) = α (t),

e2 (t) = α (t),

e3 (t) = α (t).

In order for this to work, we must assume that the vectors (α (t), α (t),
α (t)) are linearly independent for each t ∈ I; such a curve will be called
nondegenerate. (Note that this word means something different for curves
in equi-affine space than for curves in Euclidean space!) For nondegenerate
curves, the only problem with this choice of frame field is that it is not
6.2. Moving frames for curves in A3 173

necessarily unimodular. But we could fix this by defining the adapted frame

field to be
α (t)
e1 (t) = ,
3
det[α (t) α (t) α (t)]
α (t)
(6.1) e2 (t) = ,
3
det[α (t) α (t) α (t)]
α (t)
e3 (t) = .
3
det[α (t) α (t) α (t)]
Exercise 6.2. Prove that this choice of frame field (e1 (t), e2 (t), e3 (t)) is
equivariant under the action of A(3): If we replace α by g · α for some
g ∈ A(3), then ei (t) ∈ Tα(t) A3 will be replaced by (Lg )∗ (ei (t)) ∈ Tg·α(t) .

Now, wouldn’t it be nice to get rid of that ugly denominator? In the Eu-
clidean case, we were able to get rid of the denominator for e1 (t) by repa-
rametrizing according to arc length. So let’s see if we can find a suitable
reparametrization to do the trick here; specifically, we would like to find a
reparametrization α(s) of α for which
det[α (s) α (s) α (s)] = 1.
This will be a bit more complicated than in the Euclidean case since the
denominators in (6.1) involve the first three derivatives of α instead of just
the first derivative.
Suppose that we reparametrize the curve by setting α(s) = α(t(s)) for some
invertible function t(s). Then
dα dα
= t (s) ,
ds dt
d2 α 
2
2 d α dα
(6.2) ≡ t (s) mod ,
ds2 dt2 ds
d3 α d3 α dα d2 α
3
≡ t (s)3 3 mod , .
ds dt ds ds2
2
Note that it doesn’t matter what multiple of dα d α
ds we are ignoring in ds2 —
2 3
and similarly for the dα d α d α
ds , ds terms in ds2 —because they won’t affect the
determinant det[α (s) α (s) α (s)]. Therefore,
det[α (s) α (s) α (s)] = t (s)6 det[α (t(s)) α (t(s)) α (t(s))].
Note that the sign of det[α (s) α (s) α (s)] is fixed, so the best that we can
hope for is to arrange that det[α (s) α (s) α (s)] = ±1. This suggests that
174 6. Curves and surfaces in equi-affine space

we make the following definition:

Definition 6.3. Let α : I → A3 be a nondegenerate curve, and fix t0 ∈ I.

The function
& t9

6 

s(t) = det[α (u) α (u) α (u)] du
t0

is called the equi-affine arc length function along α.

*Exercise 6.4. (a) Show that any nondegenerate curve α : I → A3 can be

smoothly reparametrized by its equi-affine arc length.
(b) Show that if α is parametrized by equi-affine arc length s, then

det[α (s) α (s) α (s)] = ±1.

Remark 6.5. As for curves in Euclidean space, it may or may not be

possible to find an explicit parametrization by equi-affine 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 α : I → A3 , the equi-

affine arc length s(t) is invariant under the action of A(3); that is, for any
g ∈ A(3), the curves α and g·α have the same equi-affine arc length function.

Equi-affine arc length is a very different notion from Euclidean arc length.
Some of the most notable differences are:

(1) Unlike Euclidean arc length, which depends only on the first de-
rivative of α, the equi-affine arc length depends on the first three
derivatives of α. In fact, this number is dependent on the dimen-
sion of the ambient equi-affine space: The equi-affine arc length of
a curve α : I → An depends on the first n derivatives of α.

(2) The equi-affine arc length is only nonzero for nondegenerate curves;
so, for instance, any curve contained in a plane in A3 has equi-affine
arc length zero according to this definition. It could, however, have
nonzero equi-affine arc length when regarded as a curve in A2 .

(3) There is no ambient metric—Euclidean or otherwise—on A3 whose

restriction to a curve gives its equi-affine arc length, as there is for
curves in Euclidean space. This makes it difficult to develop any
real intuition for what the equi-affine 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 α : I → A3 be a nondegenerate curve, parametrized by

its Euclidean arc length s̄. Show that the equi-affine arc length of α is given
by
& s̄ 9

6 

(6.3) s(s̄) = κ(u)2 τ (u) du,
s̄0

where κ and τ represent the Euclidean curvature and torsion, respectively,

of α. So, the greater a curve’s Euclidean curvature and torsion, the greater
its equi-affine arc length. (Hint: Use the Frenet equations for α to compute
α (s̄), α (s̄), α (s̄).)

Remark 6.8. This exercise implies that, while the curvature κ and torsion
τ of a unit-speed curve in E3 are not preserved by the action of the equi-
affine group, the 1-form 6 |κ2 τ | ds̄ (where s̄ is the Euclidean arc length of
the curve) is invariant under the action of this group!

By Exercise 6.4, we can assume that α is parametrized by equi-affine arc

length, so that
det[α (s) α (s) α (s)] = ±1.
Then the frame field
(6.4) e1 (s) = ±α (s), e2 (s) = ±α (s), e3 (s) = ±α (s)
(with signs chosen to agree with the sign of det[α (s) α (s) α (s)]) is uni-
modular and equivariant under the action of A(3). Moreover, this frame
field is canonical, in that it is uniquely determined by the geometry of α.

Definition 6.9. Let α : I → A3 be a nondegenerate curve, parametrized

by equi-affine arc length s. The unimodular frame field (6.4) is called the
equi-affine Frenet frame of α.

Remark 6.10. The vectors (e1 (s), e2 (s), e3 (s)) of the equi-affine Frenet
frame (6.4) may not be exactly equal to those of the frame field (6.1); they
will differ by precisely the terms that we ignored in the chain rule compu-
tations (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[α (s) α (s) α (s)] = 1, so that
the signs in equation (6.4) are all positive. Inserting the appropriate minus
signs when det[α (s) α (s) α (s)] = −1 is left as an exercise for the reader.
Now that we have a canonical adapted frame field along α, we can compute
invariants for nondegenerate equi-affine curves in A3 . As in the Euclidean
176 6. Curves and surfaces in equi-affine space

case, the pullbacks of equations (3.1) to I via α can be written as

α (s)ds = ei (s) ω̄ i ,
(6.5)
ei (s)ds = ej (s) ω̄ij .
But we constructed our adapted frame field so that
α (s) = e1 (s), e1 (s) = e2 (s), e2 (s) = e3 (s).
So, the first equation in (6.5) implies that
ω̄ 1 = ds, ω̄ 2 = ω̄ 3 = 0,
and the equations for e1 (s), e2 (s) in (6.5) imply that
ω̄12 = ω̄23 = ds, ω̄11 = ω̄13 = ω̄21 = ω̄22 = 0.
Finally, e3 (s) must satisfy
e3 (s) = κ1 (s) e1 (s) + κ2 (s) e2 (s)
for some functions κ1 (s), κ2 (s). These functions are called the equi-affine
curvatures of α 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 ω̄31 , ω̄32 , and
ω̄33 ? What relations must the (ω̄ji ) satisfy?)

Thus, the equi-affine analog of the Frenet equations is

⎡ ⎤
0 0 0 0
⎢ ⎥
  ⎢1 0 0 κ1 (s)⎥
α (s) e1 (s) e2 (s) e3 (s) = α(s) e1 (s) e2 (s) e3 (s) ⎢⎢0
⎥.
⎣ 1 0 κ2 (s)⎥
⎦
0 0 1 0

Applying Lemma 4.2 yields the following theorem:

Theorem 6.12. Two nondegenerate equi-affine curves α1 , α2 : I → A3 pa-
rametrized by equi-affine arc length differ by an equi-affine transformation
if and only if they have the same equi-affine curvatures κ1 (s), κ2 (s).
*Exercise 6.13. Let α : I → A2 be a curve in the equi-affine plane A2 .
(a) How would you define an adapted frame field (e1 (t), e2 (t)) at each point
of the curve?
(b) When should a curve α : I → A2 be called “nondegenerate”?
(c) How would you define equi-affine arc length for a nondegenerate curve?
6.2. Moving frames for curves in A3 177

(d) Use the pullbacks of equations (3.1) to find a complete set of invariants
for nondegenerate equi-affine curves α : I → A2 parametrized by equi-affine
arc length.
*Exercise 6.14. Hopefully you discovered a single invariant κ(s), called
the equi-affine curvature of α, in Exercise 6.13. Suppose that α : I → A2 is
nondegenerate and that its equi-affine curvature κ(s) is equal to a constant
κ. Show that:
(a) If κ = 0, then α is a parabola.
(b) If κ > 0, then α is a hyperbola.
(c) If κ < 0, then α is a circle or an ellipse.
(Hint: In each case, you should be able to show that α(s) satisfies a dif-
ferential equation whose general solution is not difficult to find. Since each
of these conditions is preserved under equi-affine transformations, you can
perform an equi-affine transformation to eliminate most of the arbitrary con-
stants in the general solution. Finally, eliminate the parameter s to show
that α lies on the appropriate conic section.)
*Exercise 6.15. Let α : I → A2 be a curve, parametrized by its Euclidean
arc length s̄, and let κ̄(s̄) denote the Euclidean curvature of α.
(a) Show that the equi-affine arc length of α is given by
&
3
(6.6) s(s̄) = κ̄(u)2 du.
s̄0

(b) Let (ē1 (s̄), ē2 (s̄)) denote the Euclidean Frenet frame field along α. Show
that the equi-affine Frenet frame field along α is given by
e1 (s) = κ̄(s̄(s))−1/3 ē1 (s̄(s)),
(6.7) κ̄ (s̄(s))
e2 (s) = − ē1 (s̄(s)) + κ̄(s̄(s))1/3 ē2 (s̄(s)),
3κ̄(s̄(s))5/3
where s̄(s) is the inverse function of the equi-affine arc length function (6.6)
and primes denote derivatives of κ̄(s̄) with respect to s̄. (Hint: Use the
chain rule very carefully!)
(c) Differentiate the expression (6.7) for e2 (s) (again, being very careful with
the chain rule) and conclude that the equi-affine curvature of α is given by
1  
(6.8) κ(s) = − 8/3
3κ̄(s̄(s))κ̄ (s̄(s)) − 5(κ̄ (s̄(s)))2 + 9κ̄(s̄(s))4 .
9κ̄(s̄(s))
This shows that, while Euclidean curvature is a second-order invariant of α
(i.e., it may be expressed in terms of α and its derivatives up to order 2), the
178 6. Curves and surfaces in equi-affine space

equi-affine curvature is a fourth-order invariant of α. (And, like the equi-

affine arc length, the order of this invariant is dependent on the dimension
of the ambient space.) This expression for the equi-affine 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 α : I → A3 be a nondegenerate curve parametrized
by equi-affine arc length s, and suppose that its equi-affine curvatures κ1 (s),
κ2 (s) are both identically equal to zero.
(a) Show that there exist vectors v0 , v1 , v2 , v3 ∈ A3 , with det[v1 v2 v3 ] = 1,
such that
α(s) = v0 + sv1 + 12 s2 v2 + 16 s3 v3 .

(b) Show that there exists an equi-affine transformation g ∈ A(3) such that

g · α(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 different 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 → A3 be an immersion whose

image is a surface Σ = x(U ). Just as for curves, an adapted frame field
along Σ is a lifting x̃ : U → A(3) of the form
x̃(u) = (x(u); e1 (u), e2 (u), e3 (u)) ,
where for each u ∈ 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) Σ; having done so, it followed that (e1 (u), e2 (u))
must be a basis for Tx(u) Σ. But in equi-affine space, there is no notion of
orthogonality, so we don’t have any obvious way to normalize e3 (u) imme-
diately. However, we can still make our first adaptation by requiring that
(e1 (u), e2 (u)) span Tx(u) Σ. A frame field satisfying this condition will be
called 0-adapted. (The 0 reflects the fact that we will be making further re-
finements to the frame later on; these frame fields will be called 1-adapted,
2-adapted, etc.)
Exercise 6.18. Show that the condition that (e1 (u), e2 (u)) span Tx(u) Σ is
equivariant under the action of A(3).
6.3. Moving frames for surfaces in A3 179

Let (ω̄ i , ω̄ji ) represent the pulled-back forms (x̃∗ ω i , x̃∗ ωji ) on U . Precisely the
same reasoning as in the Euclidean case can be used to prove the following:

Proposition 6.19. Let U ⊂ R2 be an open set, and let x : U → A3 be an

immersion. For any 0-adapted frame field (e1 (u), e2 (u), e3 (u)) along Σ =
x(U ), the associated dual and connection forms (ω̄ i , ω̄ji ) have the property
that ω̄ 3 = 0. Moreover, (ω̄ 1 , ω̄ 2 ) form a basis for the 1-forms on U .

As in the Euclidean case, differentiating the equation ω̄ 3 = 0 yields

dω̄ 3 = −ω̄13 ∧ ω̄ 1 − ω̄23 ∧ ω̄ 2 = 0,
and Cartan’s lemma (cf. Lemma 2.49) implies that there exist functions
h11 , h12 , h22 on U such that
 3   
ω̄1 h11 h12 ω̄ 1
(6.9) = .
ω̄23 h12 h22 ω̄ 2

Once again, with an eye towards making further adaptations to our frame
field, we will investigate how the matrix [hij ] changes if we vary the frame.
At this point things begin to look different 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 field, with associated Mau-
rer-Cartan forms (ω̄ i , ω̄ji ). Any other 0-adapted frame field (ẽ1 (u), ẽ2 (u),
ẽ3 (u)) must have the properties that
span(ẽ1 (u), ẽ2 (u)) = span(e1 (u), e2 (u))
and
 
det ẽ1 (u) ẽ2 (u) ẽ3 (u) = det e1 (u) e2 (u) e3 (u) .
This will be true if and only if
⎡ ⎤
r1
 ⎢ B ⎥
(6.10) ẽ1 (u) ẽ2 (u) ẽ3 (u) = e1 (u) e2 (u) e3 (u) ⎣ r2 ⎦
0 0 (det B)−1
for some GL(2)-valued function B and real-valued functions r1 , r2 on U . Let
˜ i , ω̄
(ω̄ ˜ ji ) be the Maurer-Cartan forms associated to the new frame field.

*Exercise 6.20. (a) Show that

 1  1
˜
ω̄ ω̄
(6.11) = B −1 .
˜2
ω̄ ω̄ 2
180 6. Curves and surfaces in equi-affine space

(b) Show that

   
˜ 13
ω̄ ω̄13
t
(6.12) = (det B) B .
˜ 23
ω̄ ω̄23
(Hint: It is no longer true that ω̄i3 = −ω̄3i , so you must use the equations
for de1 and de2 in (3.1) to compute ω̄13 and ω̄23 . It will be simplest to keep
track of everything if you write them in the form
⎡ 1 1⎤
ω̄1 ω̄2
  ⎢ 2 2⎥
d e1 e2 = e1 e2 e3 ⎢ ⎥
⎣ω̄1 ω̄2 ⎦ .)
ω̄13 ω̄23

(c) Cartan’s lemma implies that there exist functions h̃11 , h̃12 , h̃22 on U such
that  3    1
˜1
ω̄ h̃11 h̃12 ω̄ ˜
= .
˜ 23
ω̄ h̃12 h̃22 ω̄ ˜2
Show that
   
h̃11 h̃12 h11 h12
(6.13) = (det B) tB B.
h̃12 h̃22 h12 h22

The transformation (6.13) has the property that

det[h̃ij ] = (det B)4 det[hij ],
so the sign of det[hij ] is fixed.
Definition 6.21. Assume that the matrix [hij ] is nonsingular at every point
of U . The surface Σ = 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 Σ is elliptic; the
hyperbolic case will be treated in Exercise 6.42.
The transformation (6.13) acts transitively on the set of 2 × 2 matrices with
positive determinant; therefore, there exists a choice of 0-adapted frame field
(e1 (u), e2 (u), e3 (u)) for which
   
h11 h12 1 0
= .
h12 h22 0 1
Such a frame field 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 field for
an elliptic equi-affine surface Σ = x(U ) ⊂ A3 .
(a) Show that any other 1-adapted frame field (ẽ1 (u), ẽ2 (u), ẽ3 (u)) for Σ
must have the form
⎡ ⎤
r1
 ⎢ B ⎥
(6.14) ẽ1 (u) ẽ2 (u) ẽ3 (u) = e1 (u) e2 (u) e3 (u) ⎣ r2 ⎦
0 0 (det B) −1

for some SO(2)-valued function B and real-valued functions r1 , r2 on U .

(b) Let (ω̄ i , ω̄ji ) be the Maurer-Cartan forms associated to a 1-adapted frame
field (e1 (u), e2 (u), e3 (u)), and let (ω̄˜ i , ω̄
˜ ji ) be the Maurer-Cartan forms as-
sociated to the 1-adapted frame field (6.14). Show that
˜ 13 ω̄
ω̄ ˜ 1 + ω̄
˜ 23 ω̄
˜ 2 = (ω̄
˜ 1 )2 + (ω̄
˜ 2 )2 = (ω̄ 1 )2 + (ω̄ 2 )2 = ω̄13 ω̄ 1 + ω̄23 ω̄ 2
(cf. Exercise 6.20).
Definition 6.23. Let U ⊂ R2 be an open set, and let x : U → A3 be
an immersion. Let (e1 (u), e2 (u), e3 (u)) be any 1-adapted frame field along
Σ = x(U ), with associated Maurer-Cartan forms (ω̄ i , ω̄ji ). The quadratic
form
I = ω̄13 ω̄ 1 + ω̄23 ω̄ 2 = (ω̄ 1 )2 + (ω̄ 2 )2
is called the equi-affine first fundamental form of Σ.

The result of Exercise 6.22 implies that the equi-affine first fundamental
form is well-defined, independent of the choice of a particular 1-adapted
frame field on Σ.
Exercise 6.24. Show that the equi-affine first fundamental form is invariant
under the action of A(3).

Because the equi-affine first fundamental form is a positive definite quadratic

form at each point u ∈ U , it defines a metric on Σ. But unlike in the
Euclidean case, this metric is not the restriction of any ambient metric on
A3 to Σ. Moreover, this metric depends on first and second derivatives of
x, whereas in the Euclidean case, the first fundamental form depends only
on first derivatives of x.
Remark 6.25. Suppose that Σ is an elliptic surface in A3 and that α : I →
A3 is a nondegenerate curve such that α(I) ⊂ Σ. We now have two ways of
defining an arc length function on α: We can use the equi-affine arc length
function from §6.2 or we can use the restriction of the metric given by the
equi-affine first fundamental form on Σ to α. 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-affine 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-affine
space, especially considering the fact that the equi-affine arc length of a
curve α : I → A3 depends on the first three derivatives of α, whereas the
equi-affine first fundamental form of a surface x : U → A3 depends only on
the first 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, 2π) × (− π2 , π2 ) ⊂ R2 , and let x : U → 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 α : (− π2 , π2 ) → A3 be the curve

α(t) = x(t, t) = t cos2 (t), sin(t) cos(t), sin(t) .

(a) Show that the equi-affine arc length function for α 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 field (e1 (u), e2 (u), e3 (u)) on S2 satisfy
ω̄13 = ω̄ 1 , ω̄23 = ω̄ 2 .
This means that any such frame field is also a 1-adapted frame field for S2
regarded as a surface in A3 . Therefore, the equi-affine first fundamental
form of S2 is the same as its Euclidean first fundamental form. (Moreover,
the unit sphere is the only surface in A3 with this property!)
(c) By part (b), the restriction of the equi-affine first fundamental form on
S2 to α is the same as the restriction of the Euclidean first fundamental
form on S2 to α, which in turn is the same as the Euclidean arc length of α.
Show that this arc length function for α is given by
& t
s(t) = cos2 (u) + 1 du.
0

(d) (Maple recommended) Plot both arc length functions for α, and com-
pute the arc length of α with respect to both arc length functions. What
happens if you change the pitch of α, i.e., set α(t) = x(ct, t) for various
choices of c?
This raises the question: Given an elliptic surface Σ ⊂ A3 , are there any
curves in Σ 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 don’t have a well-defined normal vector field e3 (u) on Σ because
we still allow transformations between 1-adapted frame fields with
ẽ3 (u) = e3 (u) + r1 (u)e1 (u) + r2 (u)e2 (u)
for arbitrary functions r1 , r2 . In order to address this issue, consider the
connection form ω̄33 .
*Exercise 6.27. (a) Show that under a transformation of the form (6.14),
we have
˜ 33 = ω̄33 + r1 ω̄ 1 + r2 ω̄ 2 .
ω̄

Since (ω̄ 1 , ω̄ 2 ) form a basis for the 1-forms on U , there is a unique choice
of r1 , r2 for which ω̄33 = 0. A 1-adapted frame field satisfying the condition
ω̄33 = 0 will be called 2-adapted.
(b) Give a geometric interpretation of the condition ω̄33 = 0.
*Exercise 6.28. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame field for
an elliptic equi-affine surface Σ = x(U ) ⊂ A3 .
(a) Show that any other 2-adapted frame field (ẽ1 (u), ẽ2 (u), ẽ3 (u)) for Σ
must have the form
⎡ ⎤
0
 ⎢ B ⎥
(6.15) ẽ1 (u) ẽ2 (u) ẽ3 (u) = e1 (u) e2 (u) e3 (u) ⎣ 0⎦
0 0 1
for some SO(2)-valued function B on U .
(b) Conclude that ẽ3 (u) = e3 (u) and that the vector field e3 (u) on Σ is
therefore now well-defined.
(c) Show that this choice of e3 (u) is equivariant under the action of A(3).
Definition 6.29. Let U ⊂ R2 be an open set, and let x : U → A3 be
an immersion. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame field along
Σ = x(U ). The vector field e3 (u) along Σ is called the equi-affine normal
vector field of Σ.
Remark 6.30. The equi-affine normal direction at a point x0 ∈ Σ has the
following geometric interpretation: Consider the family of planes in A3 that
are parallel to the tangent plane Tx0 Σ. Because Σ is elliptic, planes in this
family sufficiently close to Tx0 Σ each intersect Σ 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-affine normal vector to Σ at x0 . And since the notions of parallelism
184 6. Curves and surfaces in equi-affine 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 differentiating. Since we now have

ω̄33 = 0, we must have dω̄33 = 0 as well. According to the Cartan structure
equations (3.8) and the normalizations that we have made thus far, this
implies that
dω̄33 = −ω̄13 ∧ ω̄31 − ω̄23 ∧ ω̄32 = ω̄31 ∧ ω̄ 1 + ω̄32 ∧ ω̄ 2 = 0.
Cartan’s lemma implies that there exist functions 11 , 12 , 22 on U such that
 1   
ω̄3 11 12 ω̄ 1
(6.16) = .
ω̄32 12 22 ω̄ 2

Definition 6.31. Let U ⊂ R2 be an open set, and let x : U → A3 be an

immersion. The equi-affine second fundamental form of Σ = x(U ) is the
quadratic form II on T U defined by
II = ω̄31 ω̄13 + ω̄32 ω̄23
= ω̄31 ω̄ 1 + ω̄32 ω̄ 2
= 11 (ω̄ 1 )2 + 212 ω̄ 1 ω̄ 2 + 22 (ω̄ 2 )2 ,

where (ω̄ i , ω̄ji ) are the Maurer-Cartan forms associated to any 2-adapted
frame field (e1 (u), e2 (u), e3 (u)) on Σ = x(U ).
*Exercise 6.32. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame field
along Σ with associated Maurer-Cartan forms (ω̄ i , ω̄ji ), and let (ẽ1 (u), ẽ2 (u),
ẽ3 (u)) be any other 2-adapted frame field of the form (6.15), with associated
Maurer-Cartan forms (ω̄ ˜ i , ω̄
˜ ji ).

(a) Show that

   
˜ 31
ω̄ ω̄31
(6.17) = B −1 .
˜ 32
ω̄ ω̄32

(b) Show that

   
˜11 ˜12 11 12
t
(6.18) =B B.
˜12 ˜22 12 22

(c) Show that the equi-affine second fundamental form of x is well-defined,

independent of the choice of 2-adapted frame field and associated Maurer-
Cartan forms.
6.3. Moving frames for surfaces in A3 185

Remark 6.33. The quantity L = 12 (11 + 22 ) is called the equi-affine mean
curvature of Σ. It has the property (similar to that of the mean curvature
H in the Euclidean case) that it vanishes identically if and only if Σ is a
critical point of the equi-affine area functional.We will explore this further
in Chapter 8.

We still have a bit more differentiating to do:

*Exercise 6.34. (a) Differentiate the equations
ω̄13 = ω̄ 1 , ω̄23 = ω̄ 2
and use Cartan’s lemma to conclude that there exist functions h111 , h112 ,
h122 , h222 on U such that
⎡ ⎤ ⎡ ⎤
2ω̄11 h111 h112  
⎢ 1 ⎥ ⎢ ⎥ ω̄ 1
⎢ 2 ⎥ ⎢ ⎥
(6.19) ⎣ω̄2 + ω̄1 ⎦ = ⎣h112 h122 ⎦ 2 .
ω̄
2ω̄22 h122 h222

(b) Use the fact that ω̄11 + ω̄22 = 0 (Why is this true for a 2-adapted frame
field? Hint: It is not true for a 1-adapted frame field.) to show that
(6.20) h122 = −h111 , h112 = −h222 .
Remark 6.35. If we were to define functions (hijk ) with i, j, k = 1, 2 by
the equations
2ω̄11 = h111 ω̄ 1 + h112 ω̄ 2 ,
ω̄21 + ω̄12 = h121 ω̄ 1 + h122 ω̄ 2 = h211 ω̄ 1 + h212 ω̄ 2 ,
2ω̄22 = h221 ω̄ 1 + h222 ω̄ 2
(where the second line makes sense because the expression on the left is
symmetric in the first two indices), then the result of Exercise 6.34 says
that the (hijk ) are symmetric in all their indices.
Definition 6.36. Let U ⊂ R2 be an open set, and let x : U → A3 be
an immersion. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame field along
Σ = x(U ), with associated Maurer-Cartan forms (ω̄ i , ω̄ji ). The cubic form

P = hijk ω̄ i ω̄ j ω̄ k = h111 (ω̄ 1 )3 + 3h112 (ω̄ 1 )2 ω̄ 2 + 3h122 ω̄ 1 (ω̄ 2 )2 + h222 (ω̄ 2 )3

   
= h111 (ω̄ 1 )3 − 3ω̄ 1 (ω̄ 2 )2 + h222 (ω̄ 2 )3 − 3(ω̄ 1 )2 ω̄ 2
is called the Fubini-Pick form of Σ. P is also known as the cubic form of Σ.
*Exercise 6.37. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame field
along Σ with associated Maurer-Cartan forms (ω̄ i , ω̄ji ), and let (ẽ1 (u), ẽ2 (u),
186 6. Curves and surfaces in equi-affine space

ẽ3 (u)) be any other 2-adapted frame field of the form (6.15), with associated
Maurer-Cartan forms (ω̄ ˜ i , ω̄
˜ ji ). We can write
 
cos(θ) − sin(θ)
B=
sin(θ) cos(θ)
for some function θ on U .
(a) Show that
     
˜ 11 ω̄
ω̄ ˜ 21 ω̄11 ω̄21 0 −dθ
−1
(6.21) =B B+ .
˜ 12 ω̄
ω̄ ˜ 22 ω̄12 ω̄22 dθ 0
More explicitly:
˜ 11 = cos2 (θ) ω̄11 + sin(θ) cos(θ) (ω̄21 + ω̄12 ) + sin2 (θ) ω̄22 ,
ω̄
˜ 21 = cos2 (θ) ω̄21 + sin(θ) cos(θ) (ω̄22 − ω̄11 ) − sin2 (θ) ω̄12 − dθ,
ω̄
(6.22)
˜ 12 = cos2 (θ) ω̄12 + sin(θ) cos(θ) (ω̄22 − ω̄11 ) − sin2 (θ) ω̄21 + dθ,
ω̄
˜ 22 = cos2 (θ) ω̄22 − sin(θ) cos(θ) (ω̄21 + ω̄12 ) + (sin2 θ)ω̄11 .
ω̄

(b) Show that

   
h̃111 h111
3
=B .
h̃222 h222
Conclude that the quantity h2111 + h2222 is well-defined, independent of the
choice of 2-adapted frame field and associated Maurer-Cartan forms. The
quantity J = 16 (h2111 + h2222 ) is called the Pick invariant of Σ.
(c) Show that the Fubini-Pick form of x is well-defined, independent of the
choice of 2-adapted frame field and associated Maurer-Cartan forms.
*Exercise 6.38. Suppose that the equi-affine second fundamental form is
not a multiple of the equi-affine first 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 field on Σ for which [ij ] is
a diagonal matrix and that this choice of frame field is unique up to signs
(and possibly exchanging e1 (u) and e2 (u)). We will call such a frame field
an equi-affine principal adapted frame field on Σ. (Cf. Exercise 4.32.)
(b) Show that prescribing the first and second equi-affine fundamental forms
and the Fubini-Pick form for an equi-affine principal adapted frame field
uniquely determines all the connection forms (ω̄ji ) up to sign. (Cf. Exercise
4.38.)
6.3. Moving frames for surfaces in A3 187

(c) Conclude that any two elliptic surfaces Σ1 , Σ2 ⊂ A3 with the same first
and second equi-affine fundamental forms and Fubini-Pick form differ by an
equi-affine transformation. In other words, the first and second equi-affine
fundamental forms, together with the Fubini-Pick form, form a complete set
of local invariants for elliptic equi-affine surfaces with no umbilic points.

Exercise 6.39 (Maple recommended). Let Σ be an elliptic equi-affine

surface as in Exercise 6.38, and suppose that x : U → A3 is a parametrization
of Σ with the property that the coordinate curves are all equi-affine principal
curves. (This means that xu , xv are multiples of the vectors e1 (u), e2 (u) in
an equi-affine principal adapted frame field for Σ.) The equi-affine first and
second fundamental forms and the Fubini-Pick form of Σ 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-affine principal vectors tangent to Σ are
1 1
e1 (u) = √ xu , e2 (u) = √ xv .
M N

(b) Show that the Maurer-Cartan forms associated to this equi-affine prin-
cipal adapted frame field are given by
√ √
ω̄ 1 = ω̄13 = M du, ω̄ 2 = ω̄23 = N dv,
m n
ω̄31 = √ du, ω̄32 = √ dv,
M N
1 p111 p112 2 p122 p222
ω̄1 = du + dv, ω̄2 = du + dv,
2M
" 2M # " 2N 2N
#
p p
ω̄21 = √112 + s1 du + √122 + s2 dv,
2 MN 2 MN
" # " #
p112 p122
2
ω̄1 = √ − s1 du + √ − 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-affine space

(c) Use the Cartan structure equations for dω̄ 1 and dω̄ 2 to find 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 satisfied.
Use the Cartan structure equations (3.8) to determine the PDE system that
must be satisfied by these functions in order to guarantee the local existence
of an elliptic equi-affine surface with equi-affine first 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 → A3 be an elliptic equi-

affine 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
e g
κ1 = , κ2 =
E 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 field of Exercise
4.41. Show that the frame field
" #1/8 # "
κ1 κ2 1/8
ẽ1 (u) = e1 (u), ẽ2 (u) = e2 (u),
κ32 κ31
   
(κ1 κ2 )u (κ1 κ2 )v
ẽ3 (u) = (κ1 κ2 ) e3 (u) −
1/4
√ 7/4 3/4 e1 (u) − √ 3/4 7/4 e2 (u)
4 Eκ1 κ2 4 Gκ1 κ2

is a 2-adapted equi-affine frame field along Σ.

(b) Use the Maurer-Cartan equation (3.1) to compute the Maurer-Cartan
˜ i , ω̄
forms (ω̄ ˜ ji ) associated to this frame field. (Hint: The first equation in
(3.1) implies that

dx = e1 ω̄ 1 + e2 ω̄ 2 = ẽ1 ω̄
˜ 1 + ẽ2 ω̄
˜ 2,

so the relationship between (ω̄ ˜ 1 , ω̄

˜ 2 ) and (ω̄ 1 , ω̄ 2 ) is closely related to the
relationship between (ẽ1 , ẽ2 ) and (e1 , e2 ).)
(c) Use the result of part (b) to compute the first and second equi-affine
fundamental forms and Fubini-Pick form of Σ in terms of the Euclidean
6.3. Moving frames for surfaces in A3 189

invariants of Σ. In particular, note that

Iaff = K −1/4 IIEuc ,

where K is the Euclidean Gauss curvature of Σ.

Exercise 6.41. Let x : U → A3 be an elliptic equi-affine surface, and

suppose that the equi-affine second fundamental form is a multiple of the
equi-affine first fundamental form, so that the Maurer-Cartan forms (ω̄ i , ω̄ji )
associated to a 2-adapted frame field satisfy the conditions

ω̄31 = λ ω̄ 1 , ω̄32 = λ ω̄ 2

for some function λ on U . (This is the analog of assuming that a surface in

E3 is totally umbilic.)
(a) Prove that λ is constant. (Hint: Use the structure equations to dif-
ferentiate the equations above, taking into account the fact that we must
have
dλ = λ1 ω̄ 1 + λ2 ω̄ 2

for some functions λ1 , λ2 on U . Then use Cartan’s lemma.)

(b) Show that if λ = 0, then de3 = 0, and therefore the equi-affine normals
of Σ are all parallel. Such surfaces are called improper equi-affine spheres.
(c) Show that if λ = 0, then d(x(u) − λ1 e3 (u)) = 0. Therefore, all the equi-
affine normals of Σ intersect at the point q = x(u) − λ1 e3 (u). Such surfaces
are called proper equi-affine spheres.
Equi-affine spheres, both proper and improper, are much more plentiful than
spheres in Euclidean space; in fact, there is an infinite-dimensional family
of such surfaces.

Exercise 6.42. In this exercise, we will explore the frame adaptation pro-
cess for hyperbolic equi-affine surfaces. Suppose that x : U → A3 is a
parametrization for a hyperbolic equi-affine surface Σ = x(U ) ⊂ A3 . Then
the matrix [hij ] in (6.9) has det[hij ] < 0.
(a) Show that there exists a 0-adapted frame field (f1 (u), f2 (u), f3 (u)) for
which
   
h11 h12 0 1
= .
h12 h22 1 0
190 6. Curves and surfaces in equi-affine space

Such a frame field will be called a 1-adapted null frame field on Σ = x(U ).
The associated Maurer-Cartan forms (η̄ i , η̄ji ) satisfy

η̄13 = η̄ 2 , η̄23 = η̄ 1 ,
and the equi-affine first fundamental form becomes
I = 2 η̄ 1 η̄ 2 .
Thus, it defines an indefinite metric on Σ. (Based on this observation, we
might expect that the geometry of hyperbolic equi-affine 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 field for a hyperbolic
equi-affine surface Σ = x(U ) ⊂ A3 . Show that any other 1-adapted null
frame field (f̃1 (u), f̃2 (u), f̃3 (u)) for Σ must have the form
⎡ θ ⎤
e 0 r1
 ⎢ ⎥
(6.25) f̃1 (u) f̃2 (u) f̃3 (u) = f1 (u) f2 (u) f3 (u) ⎣ 0 e−θ r2 ⎦
0 0 1
for some functions θ, r1 , r2 on U .
(c) Show that under a transformation of the form (6.25), we have
η̄˜33 = η̄33 + r2 η̄ 1 + r1 η̄ 2 .
(No, that’s not a typo in the indices!) Conclude that there is a unique
choice of r1 , r2 for which η̄33 = 0. A 1-adapted null frame field satisfying the
condition η̄33 = 0 will be called a 2-adapted null frame field.
For the remainder of this exercise, assume that (f1 (u), f2 (u), f3 (u)) is a 2-
adapted null frame field along Σ.
(d) Differentiate the equation η̄33 = 0 and use Cartan’s lemma to conclude
that there exist functions , 12 , 21 on U such that
 1   
η̄3  12 η̄ 1
= .
η̄32 21  η̄ 2
The equi-affine second fundamental form of Σ is given by
II = η̄31 η̄13 + η̄32 η̄23
= η̄31 η̄ 2 + η̄32 η̄ 2
= 21 (η̄ 1 )2 + 2η̄ 1 η̄ 2 + 12 (η̄ 2 )2 .
6.4. Maple computations 191

(e) Differentiate the equations

η̄13 = η̄ 2 , η̄23 = η̄ 1
and use Cartan’s lemma to conclude that there exist functions h111 , h112 ,
h122 , h222 on U such that
⎡ ⎤ ⎡ ⎤
2η̄12 h111 h112  
⎢ 1 ⎥ ⎢ ⎥ η̄ 1
⎢η̄ + η̄ 2 ⎥ = ⎢h112 h122 ⎥
⎣ 1 2⎦ ⎣ ⎦ 2 .
η̄
2η̄21 h122 h222

(f) Show that

η̄11 + η̄22 = 0
(Hint: The reasoning is the same as for a 2-adapted frame field for an elliptic
surface.), and conclude that
h112 = h122 = 0.
The Fubini-Pick form of Σ is
(6.26) P = h111 (η̄ 1 )3 + h222 (η̄ 2 )3 .

Further adaptations could be made to the frame in order to further normal-

ize the equi-affine 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, we’ll go ahead and declare them all; it turns out
to be convenient to wait until after we define the structure equations to tell
Maple about the single relation among the connection forms because it
allows us to use a for loop to define 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-affine space

Tell Maple how to differentiate these forms according to the Cartan struc-
ture equations (3.8):
> for i from 1 to 3 do
d(omega[i]):= -add(’omega[i,j] &ˆ omega[j]’, j=1..3);
end do;
for i from 1 to 3 do
for j from 1 to 3 do
d(omega[i,j]):= -add(’omega[i,k] &ˆ omega[k,j]’,
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 field, taking into account the relations (6.9) that result from comput-
ing dω̄ 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);
In order to compare the functions (h̃ij ) to the functions (hij ), introduce
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]];
Now combine all these substitutions to see how the (h̃ij ) are expressed in
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(%);
Check that these expressions for the (h̃ij ) agree with those in equation (6.13):
> 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);

Exercise 6.32: Now suppose that Σ is an elliptic surface and that we

have chosen a 2-adapted frame field, so that [hij ] is the identity matrix,
B ∈ SO(2), and the equi-affine normal vector field e3 (u) is well-defined.
Since we now wish to explore transformations among 2-adapted frame fields,
assign these conditions for both (hij ) and (h̃ij ):
> 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-affine space

b[1,1]:= cos(theta);
b[1,2]:= -sin(theta);
b[2,1]:= sin(theta);
b[2,2]:= cos(theta);
Note that our adapted frame substitution is now as it should be:
> Simf(adaptedsub1);
[ω3 = 0, ω3,1 = ω1 , ω3,2 = ω2 ]

Similarly, our frame change substitution has simplified considerably:

> Simf(framechangesub);
[Ω1 = cos(θ) ω1 + sin(θ) ω2 , Ω2 = − sin(θ) ω1 + cos(θ) ω2 ,
Ω3,1 = cos(θ) ω3,1 + sin(θ) ω3,2 , Ω3,2 = − sin(θ) ω3,1 + cos(θ) ω3,2 ]

At this point, we know that ω̄33 = −(ω̄11 + ω̄22 ) = 0, and from equation (6.16)
we have expressions for ω̄31 and ω̄32 . 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 can’t
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 ∈ 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 (˜ij ) 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 (˜ij ) 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);

Exercise 6.34: Continuing on, we now need to compute the quantities

d(ω̄13 − ω̄ 1 ), d(ω̄23 − ω̄ 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]))));

−2 (ω1 ) &ˆ (ω1,1 ) − (ω2 ) &ˆ (ω1,2 ) − (ω2 ) &ˆ (ω2,1 )

(See the Maple worksheet for this chapter on the AMS webpage for an
illustration of why the inner Simf command is a good idea—or experiment
for yourself and see what happens without it!)
We can recognize this as an expression of the form

φ1 ∧ ω̄ 1 + φ2 ∧ ω̄ 2 ,

to which we should apply Cartan’s lemma. If you want Maple to help you
identify φ1 and φ2 , you can use the pick command:
> pick(zero6, omega[1]);
2 ω1,1
> pick(zero6, omega[2]);

ω1,2 + ω2,1
196 6. Curves and surfaces in equi-affine space

Similarly for d(ω̄23 − ω̄ 2 ):

> zero7:= Simf(subs(adaptedsub1,
Simf(d(omega[3,2] - omega[2]))));
−(ω1 ) &ˆ (ω2,1 ) + 2 (ω2 ) &ˆ (ω1,1 ) − (ω1 ) &ˆ (ω1,2 )
> pick(zero7, omega[1]);
ω1,2 + ω2,1
> pick(zero7, omega[2]);
−2 ω1,1

Applying Cartan’s lemma to both of these expressions yields equations

(6.19), (6.20). We need to add these expressions to adaptedsub1 (say, by
solving for ω̄11 and ω̄12 ), 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 cor-
responding 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);
Now combine all these substitutions to see how the (h̃ijk ) are expressed in
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(%);
˜ 21 and ω̄
If you like, you can perform the analogous computations for ω̄ ˜ 12 and
confirm that they are now identically zero. But be warned that sometimes
Maple can be clumsy about substitutions and require that they be applied
repeatedly. For instance, performing the same computation as above for ω̄ ˜ 21
now yields an apparently nonzero expression, but applying adaptedsub2 to
this expression yet again does, in fact, yield zero.
Finally, check that these expressions for the (h̃ijk ) agree with those in equa-
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 first
and second fundamental forms of Σ. Since we want to express everything
in terms of κ1 , κ2 , 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-affine space

For Maple purposes, use (η̄ i , η̄ji ) to denote the Maurer-Cartan forms as-
sociated to the Euclidean adapted frame field (e1 (u), e2 (u), e3 (u)) and use
(ω̄ i , ω̄ji ) to denote those associated to the 2-adapted equi-affine frame field
(ẽ1 (u), ẽ2 (u), ẽ3 (u)). We have computed (η̄ i , η̄ji ) previously, and we can
simply assign them:
> eta[1]:= E(u,v)ˆ(1/2)*d(u);
eta[2]:= G(u,v)ˆ(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)ˆ(1/2)*G(u,v)ˆ(1/2));
eta[3,1]:= (e(u,v)/E(u,v)ˆ(1/2))*d(u);
eta[3,2]:= (g(u,v)/G(u,v)ˆ(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 fields (ei (u)) and (ẽi (u))
so that we can use their structure equations to compute the associated
Maurer-Cartan forms. There’s no good way to tell Maple that these func-
tions are vector-valued, but it won’t really matter.
First we tell Maple the structure equations for the exterior derivatives of
the Euclidean frame field, 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];
We’ll need the Gauss and Codazzi equations later, and these can be com-
puted directly from the equations d(dei ) = 0. Start with d(de3 ):
zero9:= Simf(d(d(e03)));
Pick off the scalar coefficient of du ∧ 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 coefficients must be zero;
indeed, these are the Codazzi equations. We can collect these into a conve-
nient 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 coefficient of du ∧ 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 κ1 , κ2 later, it’s 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)))));

Rather than just computing the Maurer-Cartan forms for the frame field
given in Exercise 6.40(a), let’s explore where these expressions came from.
First, the fact that the coordinate curves of Σ are principal curves means
that the Maurer-Cartan forms associated to this Euclidean frame field have
the property that η̄13 is a multiple of η̄ 1 and η̄23 is a multiple of η̄ 2 . This means
that this Euclidean frame field is actually a 0-adapted equi-affine frame field
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-affine
frame field for which (ẽ1 (u), ẽ2 (u)) are scalar multiples of (e1 (u), e2 (u)),
respectively. So, set

ẽ1 (u) = λ1 e1 (u), ẽ2 (u) = λ2 e2 (u)

for some (unknown) functions λ1 , λ2 . Then in order to keep the frame uni-
modular, we must have
1
ẽ3 (u) = r1 e1 (u) + r2 e2 (u) + e3
λ1 λ2
200 6. Curves and surfaces in equi-affine space

for some functions r1 , r2 . For Maple purposes, we will call this frame field
(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 field to be 2-adapted, the corresponding Maurer-
Cartan forms must satisfy the conditions

ω̄13 = ω̄ 1 , ω̄23 = ω̄ 2 , ω̄11 + ω̄22 = 0.

Moreover, we must have

√ √
1 1 1 E 2 1 2 G
ω̄ = η̄ = du, ω̄ = η̄ = dv.
λ1 λ1 λ2 λ2
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 λ1 , λ2 , r1 , r2 by using the structure
equations for dẽ1 , dẽ2 , dẽ3 :
> 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, you’ll see that the e1 (u)
and e2 (u) terms are rather a mess because they involve Maurer-Cartan forms
that haven’t 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 dẽ1 and dẽ2 to
6.4. Maple computations 201

solve for λ1 , λ2 :
> 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
λ1 (u, v) = ,
RootOf (−κ1 + Z 8 κ23 )3 κ2
3
λ2 (u, v) = RootOf (−κ1 + Z 8 κ23 )
and we will need to intervene by hand. The expression
RootOf (−κ1 + Z 8 κ23 )
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)
κ
Z = 1(3/8) ,
κ2
which leads to the solution
(1/8) (1/8)
κ2 κ1
λ1 = (3/8)
, λ2 = (3/8)
.
κ1 κ2
So make these assignments:
> lambda1(u,v):= kappa2(u,v)ˆ(1/8)/kappa1(u,v)ˆ(3/8);
lambda2(u,v):= kappa1(u,v)ˆ(1/8)/kappa2(u,v)ˆ(3/8);
Now we can use the e3 (u) term in dẽ3 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.
 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 length—not even
an unconventional one such as the equi-affine arc length that we defined for
curves in An . (It may, however, be possible to define 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 con-
cepts, we will start one dimension lower than in previous chapters and begin
by investigating curves in P2 . This is already a topic with important ap-
plications, 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 [α] : I → P2 that maps some open

interval I ⊂ R into the projective plane. P2 has the structure of the homo-
geneous space SL(3)/H[x0 ] , where H[x0 ] is the isotropy group from equation
(3.17), so an adapted frame field along [α] should be a lifting α̃ : I → SL(3).
Any such lifting can be written as

α̃(t) = (e0 (t), e1 (t), e2 (t)),

where for each t ∈ I,

[e0 (t)] = [α(t)] ∈ P2

and

det e0 (t) e1 (t) e2 (t) = 1.

Such an adapted frame field is called a projective frame field along [α]. How
should we choose such a lifting in some geometrically natural way?
As in the equi-affine 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 α : I → R3 of [α] with the property
that

(7.1) det α(t) α (t) α (t) = 1.

In particular, the vectors (α(t), α (t), α (t)) should be linearly independent
for all t ∈ I.

Exercise 7.1. Given a curve [α] : I → P2 , let α1 , α2 : I → R3 be two liftings

of [α], so that

[α1 (t)] = [α2 (t)] = [α(t)] ∈ P2

for all t ∈ I. Show that the vectors (α1 (t), α1 (t), α1 (t)) are linearly indepen-
dent if and only if the vectors (α2 (t), α2 (t), α2 (t)) are linearly independent.
(Hint: α2 (t) = λ(t)α1 (t) for some nonvanishing, real-valued function λ(t).)

Exercise 7.1 implies that the linear dependence or independence of the vec-
tors (α(t), α (t), α (t)) does not depend on the choice of lifting α : I → R3 .
7.2. Moving frames for curves in P2 205

Therefore, the following definition makes sense:

Definition 7.2. A regular curve [α] : I → P2 will be called nondegenerate

if for any lifting α : I → R3 of [α], the vectors (α(t), α (t), α (t)) are linearly
independent for all t ∈ I.

*Exercise 7.3. Let [α] : I → P2 be a nondegenerate curve. Show that [α]

has a unique lifting α : I → R3 satisfying the determinant condition (7.1).
(Hint: Let α0 : I → R3 be any lifting of [α], and consider an arbitrary lifting
α(t) = λ(t)α0 (t). Show that (7.1) uniquely determines the function λ(t).)

Definition 7.4. Let [α] : I → P2 be a nondegenerate curve. The lifting

α : I → R3 of Exercise 7.3 will be called the canonical lifting of [α]. The
projective frame field

e0 (t) = α(t), e1 (t) = α (t), e2 (t) = α (t)

will be called the canonical projective frame field associated to [α].

Exercise 7.5. Prove that the canonical projective frame field (e0 (t), e1 (t),
e2 (t)) associated to [α] is equivariant under the action of SL(3) on P2 : If
we replace [α] by g · [α] for some g ∈ SL(3), then eγ (t) ∈ Tα(t) R3 will be
replaced by g · eγ (t) ∈ Tg·α(t) R3 for γ = 0, 1, 2.

Now things start to get interesting. The canonical projective frame field de-
pends on the parametrization of [α]. How will it change if we reparametrize
[α]? Is there some particular parametrization of [α]—something akin to
the arc length parametrization for curves in Euclidean space—that is some-
how geometrically natural? In order to answer this question, we will first
compute invariants for parametrized nondegenerate curves. Then we will in-
vestigate how these invariants transform under a change of parametrization
for [α] and look for special parametrizations that normalize the invariants
in some natural way.
The pullbacks of equations (3.1) to I via α can be written as

(7.2) eγ (t)dt = eδ (t) ω̄γδ ,

where the indices γ, δ range from 0 to 2. But we constructed the canonical

projective frame field so that

e0 (t) = e1 (t), e1 (t) = e2 (t).

206 7. Curves and surfaces in projective space

The first of these equations implies that

ω̄01 = dt, ω̄00 = ω̄02 = 0,

and the second equation implies that

ω̄12 = dt, ω̄10 = ω̄11 = 0.

Finally, e2 (t) must satisfy

e2 (t) = κ0 (t) e0 (t) + κ1 (t) e1 (t)

for some functions κ0 (t), κ1 (t). These functions are called the Wilczynski in-
variants of the parametrized curve [α] : I → P2 . They were first introduced
by Wilczynski in the context of studying invariants for linear differential
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 repa-

rametrization of α.

*Exercise 7.7 (Maple recommended). Let J ⊂ R, and let [β] : J → P2 be

a reparametrization of [α] given by [β(s)] = [α(t(s))], with t (s) = 0.
(a) Show that the canonical lifting of [β] is given by

1
β(s) = α(t(s)),
t (s)

where α(t) is the canonical lifting of [α].

(b) Show that the canonical projective frame field (ẽ0 (s), ẽ1 (s), ẽ2 (s)) asso-
ciated to [β] is given by

1
ẽ0 (s) = e0 (t(s)),
t (s)
t (s)
ẽ1 (s) = − e0 (t(s)) + e1 (t(s)),
(t (s))2
"  #
t (s) (t (s))2 t (s)
ẽ2 (s) = − − 2 e 0 (t(s)) − e1 (t(s)) + t (s)e2 (t(s)),
(t (s))2 (t (s))3 t (s)

where (e0 (t), e1 (t), e2 (t)) is the canonical projective frame field associated
to [α].
7.2. Moving frames for curves in P2 207

(c) Show that the Wilczynski invariants κ̃0 (s), κ̃1 (s) associated to [β] are
given by
" #3
t(4) (s) t (s)t (s) t (s)
(7.3) κ̃0 (s) = −  +4 −3
t (s) (t (s))2 t (s)
+ (t (s))3 κ0 (t(s)) + t (s)t (s)κ1 (t(s)),
" #2
t (s) t (s)
(7.4) κ̃1 (s) = −2  +3 + (t (s))2 κ1 (t(s)),
t (s) t (s)

where κ0 (t), κ1 (t) are the Wilczynski invariants associated to [α].

The transformation rule (7.4) for κ̃1 is simpler than the transformation rule
(7.3) for κ̃0 , in that (7.4) involves only κ1 , whereas (7.3) involves both κ0
and κ1 . So, first consider equation (7.4). Local existence theory for ordinary
differential equations guarantees that, given any smooth function κ1 (t) on
an interval I ⊂ R, any point t0 ∈ I, and any point s0 ∈ R, there exists an
interval J ⊂ R containing the point s0 and a function t : J → I with the
properties that

(1) t(s0 ) = t0 ;

(2) t (s) = 0 for all s ∈ J (indeed, t (s0 ) = 0 can be prescribed arbitrar-

ily, and then J can be chosen to be small enough so that t (s) = 0
on the entire interval);

(3) t(s) satisfies the differential equation

" #2
t (s) t (s)
(7.5) −2 +3 + (t (s))2 κ1 (t(s)) = 0.
t (s) t (s)

According to (7.4), the corresponding reparametrization [β(s)] will have

κ̃1 (s) ≡ 0.

Definition 7.8. Let I ⊂ R, and let [α] : I → P2 be a nondegenerate

curve. [α] is called a projective parametrization if the Wilczynski invariants
κ0 (t), κ1 (t) associated to [α] have the property that κ1 (t) ≡ 0. In this case,
the parameter t ∈ I is called a projective parameter for [α].

A projective parametrization for a nondegenerate curve in P2 is the nat-

ural 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 flexibility does this notion admit?
In order to address this question, suppose that [α] : I → P2 is a nondegen-
erate, projectively parametrized curve, and suppose that [β(s)] = [α(t(s))]
is a reparametrization of [α] that is also a projective parametrization. This
means that the Wilczynski invariants κ1 (t), κ̃1 (s) associated to [α], [β], re-
spectively, are both identically zero. According to (7.4), it follows that the
function t(s) must satisfy the differential equation
" #2
t (s) t (s)
(7.6) −2  +3 = 0.
t (s) t (s)

The expression on the left-hand side of (7.6) is equal to −2 times the

Schwarzian derivative S(t) of the function t(s), defined by
" #2
t 3 t
(7.7) S(t) =  − .
t 2 t
The Schwarzian derivative is the fundamental invariant differential operator
of projective differential geometry; it has the property that for any differen-
tiable function f : P1 → P1 and any projective transformation g : P1 → P1 ,

S(g ◦ 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 → P1 has the

form
g([x0 : x1 ]) = [dx0 + cx1 : bx0 + ax1 ],
where
 
a b
det = 1.
c d
x1
In terms of the affine coordinate x̄ = x0
on P1 , g is given by a linear fractional
transformation
ax̄ + b
g(x̄) = .
cx̄ + d

(a) Show by direct computation that if g is a linear fractional transformation,

then S(g) = 0.
(b) Suppose that a differentiable function f : P1 → P1 satisfies 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

" # " #2
f  1 f 
S(f ) = − .
f 2 f
Then solve the differential equation

z  − 12 z 2 = 0
f 
for the function z = f  .)

As a consequence of Exercise 7.9, any projective reparametrization of a

projectively parametrized nondegenerate curve [α(t)] must have the form
[β(s)] = [α(t(s))], where
as + b
t(s) =
cs + d
with ad − 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 trans-
formations.

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 α : I → En , any reparametrization of the
form

(7.8) s̃ = s + c,

where c ∈ R is constant, yields an equally valid arc length parameter for

α. It is even possible—for instance, when α is a closed curve—that there
is no single, continuous arc length parameter defined 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) defined
on the regions where intervals overlap. (Recall the discussion of transition
functions between systems of local coordinates on manifolds in §1.1.) What
we really have is not so much an arc length parameter as a metric structure
on the curve, defined 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-defined 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-defined projective structure on the curve, analogous to the metric
structure on a curve in En .

Now, assume that [α] : I → P2 is projectively parametrized, and consider the

remaining invariant function κ0 (t). How does it transform under a projective
reparametrization?

*Exercise 7.11. Let [α] : I → P2 be a nondegenerate, projectively para-

metrized curve, and let [β(s)] = [α(t(s))] be a projective reparametrization
of [α] given by a linear fractional transformation

as + b
(7.9) t(s) =
cs + d

with ad − bc = 1. Let κ0 (t), κ̃0 (s) be the invariants associated to [α], [β],
respectively. Use equation (7.3) to show that
" #3
κ0 (t(s)) dt
κ̃0 (s) = = κ0 (t(s)) .
(cs + d)6 ds

Conclude that under the transformation (7.9), we have

κ̃0 (s)(ds)3 = κ0 (t)(dt)3 .

Thus, the curvature “function” κ0 is not really a well-defined function on

the curve at all, but rather a “cubic form” that can be represented in terms
of any projective parameter t as

(7.10) κ̂0 = κ0 (t)(dt)3 .

The cubic form (7.10) is called the projective curvature form of [α]. In the
following exercise, we show that the projective curvature form is a well-
defined rank 3 symmetric tensor along [α].

*Exercise 7.12. Let [α] : I → P2 be a nondegenerate curve (not neces-

sarily projectively parametrized), and let [β(s)] = [α(t(s))] be an arbitrary
reparametrization of [α]. Let κ0 (t), κ1 (t) be the invariants associated to [α]
and let κ̃0 (s), κ̃1 (s) be the invariants associated to [β]. Use the transforma-
tion rules (7.3), (7.4) to show that
   
κ̃0 (s) − 12 κ̃1 (s) (ds)3 = κ0 (t) − 12 κ1 (t) (dt)3 .
7.2. Moving frames for curves in P2 211

Conclude that the cubic form

 
(7.11) κ̂0 = κ0 (t) − 12 κ1 (t) (dt)3
is a well-defined rank 3 tensor along [α] that has the form (7.10) for any
projective parametrization of [α].
Remark 7.13. Our terminology here is somewhat nonstandard. In the case
where κ̂0 is nonvanishing, the 1-form
 1
3
κ̂0 = κ0 (t) − 12 κ1 (t) 3 dt
is sometimes referred to as the projective arc length element and the function
& t
3
s(t) = κ̂0
t0
is sometimes referred to as the projective arc length of [α]. Parametrizing
[α] according to this projective arc length function then leads to a notion
of projective curvature that is somewhat different from ours. This approach
has the advantage that the projective curvature so obtained is a well-defined
function on [α] 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 field associated to a projectively

parametrized nondegenerate curve is not quite invariant under a projective
reparametrization.
*Exercise 7.14. Let [α] : I → P2 be a nondegenerate, projectively para-
metrized curve, and let [β(s)] = [α(t(s))] be a projective reparametrization
of [α] given by a linear fractional transformation as in equation (7.9). Let
(e0 (t), e1 (t), e2 (t)) be the canonical projective frame field associated to [α].
Use the result of Exercise 7.7(b) to show that the canonical projective frame
field (ẽ0 (s), ẽ1 (s), ẽ2 (s)) associated to [β] is given by
ẽ0 (s) = (cs + d)2 e0 (t(s)),

ẽ1 (s) = 2c(cs + d)e0 (t(s)) + e1 (t(s)),

2c 1
ẽ2 (s) = 2c2 e0 (t(s)) + e1 (t(s)) + e2 (t(s)).
cs + d (cs + d)2

Thus, the projective structure gives rise to a 2-parameter family of canonical

projective frame fields along a nondegenerate curve [α] : I → 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 parameters’ worth of choices for a projective frame
field along the curve, and choosing a specific projective frame at any point
of the curve uniquely determines the canonical projective frame field along
the entire curve.
For any canonical projective frame field (e0 (t), e1 (t), e2 (t)) associated to a
nondegenerate, projectively parametrized curve [α] : I → P2 , we have the
following analog of the Frenet equations:
⎡ ⎤
0 0 κ0 (t)
 ⎢ ⎥
e0 (t) e1 (t) e2 (t) = e0 (t) e1 (t) e2 (t) ⎣1 0 0 ⎦ .
01 0

Applying Lemma 4.2 yields the following theorem:

Theorem 7.15. Two nondegenerate, projectively parametrized curves [α1 ],

[α2 ] : I → P2 differ by a projective transformation if and only if they have
the same projective curvature form κ̂0 = κ0 (t)(dt)3 .

*Exercise 7.16. Suppose that a nondegenerate, projectively parametrized

curve [α] : I → P2 has projective curvature κ0 (t) = 0. (Note that this
condition is invariant under projective reparametrization.)
(a) Show that the canonical lifting α : I → R3 of [α] is part of a parabola
contained in a plane in R3 .
(b) Conclude that all conic sections in P2 have projective curvature identi-
cally equal to zero.

Exercise 7.17. This exercise demonstrates the result of Exercise 7.16 ex-
plicitly in the case where [α] is a circle. Let [α] : R → P2 be the circle
parametrized in homogeneous coordinates as

[α(t)] = [1 : cos(t) : sin(t)].

(a) Show that the canonical lifting α : R → R3 of [α] is

α(t) = t[1, cos(t), sin(t)].

(b) Show that the invariants κ0 (t), κ1 (t) associated to [α] are

κ0 (t) = 0, κ1 (t) = −1.

Conclude that [α] is not a projective parametrization.

7.2. Moving frames for curves in P2 213

(c) Show that the function

t(s) = 2 tan−1 (s)
satisfies equation (7.5); thus, the curve
[β(s)] = [α(2 tan−1 (s))]
is a projective reparametrization of [α].
(d) Show that the canonical lifting of [β] is
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
1 + s2 1 0 1
1⎢ ⎥ 1 2⎢ ⎥ ⎢ ⎥ 1⎢ ⎥
β(s) = 2 ⎣1 − s ⎦ = 2 s ⎣−1⎦ + s ⎣0⎦ + 2 ⎣1⎦ .
2

2s 0 1 0
Therefore, β is a parabola in the plane spanned by the vectors t[1, −1, 0],
t[0, 0, 1] and passing through the point t[ 1 , 1 , 0].
2 2

Exercise 7.18. We like to think of P2 as the plane R2 , represented as the

open set
2 3
V0 = 1 : x̄1 : x̄2 ⊂ P2 ,
plus points “at infinity”. Thus, we like to think of a curve [α] : I → P2 as
having a parametrization of the form
(7.12) [α(t)] = [1 : x̄1 (t) : x̄2 (t)]
that is valid at all but finitely many points on the curve.
(a) Show that any nondegenerate curve [α] : I → P2 of the form (7.12) has
a reparametrization [β(s̄)] = [α(t(s̄))] whose canonical lifting is
β(s̄) = t[1, x̄1 (t(s̄)), x̄2 (t(s̄))].

(b) Show that the invariants κ0 (s̄), κ1 (s̄) associated to [β] have the property
that κ0 (s̄) ≡ 0.
(c) Conclude that [β(s̄)] is not a projective parametrization unless [α] 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 [β(s̄)] is not a projective parametrization, it has the
property that the curve β̄ : I → A2 defined by
β̄(s̄) = t[x̄1 (t(s̄)), x̄2 (t(s̄))]
is parametrized by its equi-affine arc length, and the Wilczynski invariant
κ1 (s̄) of [β] is equal to the equi-affine curvature κ(s̄) of β̄. This is related to
214 7. Curves and surfaces in projective space

the fact that the equi-affine 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 ] ∈ R3 | x0 = 1} in R3 .
(e) Let [γ(s)] = [β(s̄(s))] be a projective reparametrization of [β(s̄)], so that
the Wilczynski invariants κ̃0 (s), κ̃1 (s) associated to [γ] satisfy κ̃1 (s) = 0.
Use the formulas (7.3), (7.4) and the fact that κ0 (s̄) = 0 to show that

κ̃0 (s) = − 12 (s̄ (s))3 κ (s̄(s)),

where κ(s̄) is the equi-affine curvature of β̄ and κ (s̄(s)) denotes the deriva-
tive of κ(s̄) with respect to s̄, evaluated at s̄ = s̄(s).
(f) Conclude that the projective curvature form of the original curve [α] may
be expressed as
κ̂0 = κ̃0 (s)(ds)3 = − 12 κ (s̄)(ds̄)3 ,
where s̄ is the equi-affine arc length of the curve β̄ : I → A2 and κ(s̄) is the
equi-affine curvature of β̄. Comparing with the result of Exercise 6.15, we
see that the projective curvature form is a fifth-order invariant of [α].

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 [α] : I → P3 that maps some open
interval I ⊂ R into the projective space P3 . The construction of the canoni-
cal projective frame field associated to [α] is analogous to that for curves in
P2 : A projective frame field along [α] is a lifting α̃ : I → SL(4), written as

α̃(t) = (e0 (t), e1 (t), e2 (t), e3 (t)),

with the properties that

[e0 (t)] = [α(t)] ∈ 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 α : I → R4

with the property that

(7.14) det α(t) α (t) α (t) α (t) = 1.

In particular, the vectors (α(t), α (t), α (t), α (t)) should be linearly inde-
pendent for all t ∈ I.
7.3. Moving frames for curves in P3 215

Definition 7.19. A regular curve [α] : I → P3 will be called nondegenerate

if for any lifting α : I → R4 of [α], the vectors (α(t), α (t), α (t), α (t)) are
linearly independent for all t ∈ I.
Exercise 7.20 (Cf. Exercise 7.1). Show that Definition 7.19 makes sense,
as follows. Given a curve [α] : I → P3 , let α1 , α2 : I → R4 be two liftings of
[α], so that
[α1 (t)] = [α2 (t)] = [α(t)] ∈ P3
for all t ∈ I. Show that the vectors (α1 (t), α1 (t), α1 (t), α1 (t)) are linearly
independent if and only if the vectors (α2 (t), α2 (t), α2 (t), α2 (t)) are linearly
independent.
*Exercise 7.21 (Cf. Exercise 7.3). (a) Let [α] : I → P3 be a nondegenerate
curve. Show that [α] has a lifting α : I → R4 satisfying the determinant
condition

det α(t) α (t) α (t) α (t) = ±1
and that this lifting is uniquely determined up to sign (i.e., α may be replaced
by −α).
(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 α(t) α (t) α (t) α (t) = 1.
(This can be achieved by replacing [α] by its reflection through a plane in
P3 if necessary.)
Definition 7.22. Let [α] : I → P3 be a nondegenerate curve. The lifting
α : I → R4 of Exercise 7.21 will be called the canonical lifting of [α]. The
projective frame field
e0 (t) = α(t), e1 (t) = α (t), e2 (t) = α (t), e3 (t) = α (t)
will be called the canonical projective frame field associated to [α].
Exercise 7.23. Prove that the canonical projective frame field (e0 (t), e1 (t),
e2 (t), e3 (t)) associated to [α] is equivariant under the action of SL(4) on
P3 : If we replace [α] by g · [α] for some g ∈ SL(4), then eγ (t) ∈ Tα(t) R4 will
be replaced by g · eγ (t) ∈ Tg·α(t) R4 for γ = 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 field associated
to a parametrized nondegenerate curve [α] : I → P3 satisfies the structure
equations
e0 (t) = e1 (t), e1 (t) = e2 (t), e2 (t) = e3 (t),
e3 (t) = κ0 (t) e0 (t) + κ1 (t) e1 (t) + κ2 (t) e2 (t)
for some functions κ0 (t), κ1 (t), κ2 (t).

As for curves in P2 , the functions κ0 (t), κ1 (t), κ2 (t) are called the Wilczynski
invariants of the parametrized curve [α] : I → P3 . Next, we investigate how
the Wilczynski invariants transform under a reparametrization of [α].
*Exercise 7.25 (Maple recommended; cf. Exercise 7.7). Let J ⊂ R, and
let [β] : J → P3 be a reparametrization of [α] given by [β(s)] = [α(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 [β] is given by
1
β(s) = α(t(s)),
t (s)
where α(t) is the canonical lifting of [α].
(b) Show that the canonical projective frame field (ẽ0 (s), ẽ1 (s), ẽ2 (s), ẽ3 (s))
associated to [β] is given by
1
ẽ0 (s) = e0 (t(s)),
(t (s))3/2

3t (s) 1
ẽ1 (s) = − e 0 (t(s)) + e1 (t(s)),
(2t (s))5/2 (t (s))1/2
" #
3t (s) 15(t (s))2 t (s)
ẽ2 (s) = − − e 0 (t(s)) − 2 e1 (t(s))
2(t (s))5/2 4(t (s))7/2 (t (s))3/2

+ (t (s))1/2 e2 (t(s)),

 
3t(4) (s) 45t (s)t (s) 105(t (s))3
ẽ3 (s) = − − + e0 (t(s))
2(t (s))5/2 4(t (s))7/2 8(t (s))9/2
" #
7t (s) 27(t (s))2 3t (s)
− − e1 (t(s)) − e2 (t(s))
2(t (s))3/2 4(t (s))5/2 2(t (s))1/2

+ (t (s))3/2 e3 (t(s)),

7.3. Moving frames for curves in P3 217

where (e0 (t), e1 (t), e2 (t), e3 (t)) is the canonical projective frame field asso-
ciated to [α].
(c) Show that the Wilczynski invariants κ̃0 (s), κ̃1 (s), κ̃2 (s) associated to [β]
are given by
 "  #2
1 t(5) (s) t(4) (s)t (s) t (s)
κ̃0 (s) = −24  + 120 
+ 60
16 t (s) (t (s)) 2 t (s)
" #4 
t (s)(t (s))2 t (s)
−300 + 135
(t (s))3 t (s)
(7.15)
3
+ (t (s))4 κ0 (t(s)) + t (s)(t (s))2 κ1 (t(s))
2
" #
3   3 
+ t (s)t (s) − (t (s)) κ2 (t(s)),
2
2 4
 "  #3 
t(4) (s) t (s)t (s) t (s)
κ̃1 (s) = 5 −  +4  2
−3 
(7.16) t (s) (t (s)) t (s)
+ (t (s))3 κ1 (t(s)) + 2t (s)t (s)κ2 (t(s)),
 " #2 
5 t (s) t (s)
(7.17) κ̃2 (s) = −2  +3 + (t (s))2 κ2 (t(s)),
2 t (s) t (s)
where κ0 (t), κ1 (t), κ2 (t) are the Wilczynski invariants associated to [α].

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 [β(s)] for which κ̃2 (s) ≡ 0.
Thus, we have the analog of Definition 7.8:
Definition 7.26. Let I ⊂ R, and let [α] : I → P3 be a nondegenerate
curve. [α] is called a projective parametrization if the Wilczynski invariants
κ0 (t), κ1 (t), κ2 (t) associated to [α] have the property that κ2 (t) ≡ 0. In this
case, the parameter t ∈ I is called a projective parameter for [α].

As for curves in P2 , a projective parametrization is unique up to repara-

metrizations of the form [β(s)] = [α(t(s))], where t(s) is a linear fractional
transformation. Therefore, any nondegenerate curve in P3 has a well-defined
projective structure.
Now, assume that [α] : I → P3 is projectively parametrized, and consider
the remaining invariant functions κ0 (t), κ1 (t).
218 7. Curves and surfaces in projective space

*Exercise 7.27 (Cf. Exercise 7.11). Let [α] : I → P3 be a nondegenerate,

projectively parametrized curve, and let [β(s)] = [α(t(s))] be a projective
reparametrization of [α] given by a linear fractional transformation as in
equation (7.9). Let κ0 (t), κ1 (t) and κ̃0 (s), κ̃1 (s) be the invariants associated
to [α], [β], respectively.
(a) Use equations (7.15) and (7.16) to show that
κ1 (t(s))
κ̃1 (s) = ,
(cs + d)6

κ0 (t(s)) 3c κ1 (t(s))
κ̃0 (s) = 8
− .
(cs + d) (cs + d)7

(b) Show that under the transformation (7.9), we have

κ̃1 (s)(ds)3 = κ1 (t)(dt)3 ,
   
κ̃0 (s) − 12 κ̃1 (s) (ds)4 = κ0 (t) − 12 κ1 (t) (dt)4 .
Therefore, the cubic form
(7.18) κ̂1 = κ1 (t)(dt)3
and the quartic form
 
(7.19) κ̂0 = κ0 (t) − 12 κ1 (t) (dt)4
are invariant under projective transformations. We will call these forms the
projective curvature forms of [α].
*Exercise 7.28. Let [α] : I → P3 be a nondegenerate curve (not neces-
sarily projectively parametrized), and let [β(s)] = [α(t(s))] be an arbitrary
reparametrization of [α]. Let κ0 (t), κ1 (t), κ2 (t) be the invariants associated
to [α], and let κ̃0 (s), κ̃1 (s), κ̃2 (s) be the invariants associated to [β].
(a) Use the transformation rules (7.15), (7.16), (7.17) to show that
   
κ̃1 (s) − κ̃2 (s) (ds)3 = κ1 (t) − κ2 (t) (dt)3 .
Conclude that the cubic form
 
(7.20) κ̂1 = κ1 (t) − κ2 (t) (dt)3
is a well-defined rank 3 tensor along [α] that has the form (7.18) for any
projective parametrization of [α].
(b) Use the transformation rules (7.15), (7.16), (7.17) to show that
 
κ̃0 (s) − 12 κ̃1 (s) + 15 κ̃2 (s) + 100
9
(κ̃2 (s))2 (ds)4
 
= κ0 (t) − 12 κ1 (t) + 15 κ2 (t) + 9
100 (κ2 (t))
2
(dt)4 .
7.3. Moving frames for curves in P3 219

Conclude that the quartic form

 
κ̂0 = κ0 (t) − 12 κ1 (t) + 15 κ2 (t) + 9
100 (κ2 (t))
2
(dt)4
is a well-defined rank 4 tensor along [α] that has the form (7.19) for any
projective parametrization of [α].

As for curves in P2 , the canonical projective frame field 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 [α] : I → P3 be a nondegener-
ate, projectively parametrized curve, and let [β(s)] = [α(t(s))] be a projec-
tive reparametrization of [α] 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 field associated to [α]. Use the result of Exercise 7.25(b) to show that
the canonical projective frame field (ẽ0 (s), ẽ1 (s), ẽ2 (s), ẽ3 (s)) associated to
[β] is given by
ẽ0 (s) = (cs + d)3 e0 (t(s)),

ẽ1 (s) = 3c(cs + d)2 e0 (t(s)) + (cs + d)e1 (t(s)),

1
ẽ2 (s) = 6c2 (cs + d)e0 (t(s)) + 4ce1 (t(s)) + e2 (t(s)),
(cs + d)
6c2 3c
ẽ3 (s) = 6c3 e0 (t(s)) + e1 (t(s)) + e2 (t(s))
(cs + d) (cs + d)2
1
+ e3 (t(s)).
(cs + d)3

Thus, the projective structure gives rise to a 2-parameter family of canonical

projective frame fields along a nondegenerate curve [α] : I → P3 , just as it
did for curves in P2 . For any canonical projective frame field (e0 (t), e1 (t),
e2 (t), e3 (t)) associated to such a curve, we have the following analog of the
Frenet equations:
⎡ ⎤
0 0 0 κ0 (t)
⎢ ⎥
  ⎢1 0 0 κ1 (t)⎥
   
e0 (t) e1 (t) e2 (t) e3 (t) = e0 (t) e1 (t) e2 (t) e3 (t) ⎢⎢ ⎥.
⎥
⎣0 1 0 0 ⎦
001 0

Applying Lemma 4.2 yields the following theorem:

Theorem 7.30. Two nondegenerate, projectively parametrized curves [α1 ],
[α2 ] : I → P3 differ by a projective transformation if and only if they have
220 7. Curves and surfaces in projective space

the same projective curvature forms

 
κ̂1 = κ1 (t)(dt)3 , κ̂0 = κ0 (t) − 12 κ1 (t) (dt)4 .
*Exercise 7.31. Suppose that a nondegenerate, projectively parametrized
curve [α] : I → P3 has projective curvatures κ0 (t) = κ1 (t) = 0. (Note that
this condition is invariant under projective reparametrization.)
(a) Show that the canonical lifting α : I → R4 of [α] has the form
α(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 α is contained in a 3-dimensional affine hyperplane
in R4 —namely, the plane passing through v0 and spanned by the vectors
(v1 , v2 , v3 ).
(b) Show that [α] is projectively equivalent to the curve [β] with canonical
lifting
β(t) = t[1, t, 12 t2 , 16 t3 ].
[β] is called the rational normal curve of degree 3 (cf. Exercise 6.17). In gen-
eral, the rational normal curve [β] of degree n in Pn is the curve parametrized
in homogeneous coordinates as
[β(t)] = [1 : t : 12 t2 : · · · : 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 → P3 be an immersion

whose image is a surface [Σ] = [x(U )]. Just as for curves in P3 , an adapted
frame field along [Σ] is a lifting x̃ : U → SL(4) of the form
x̃(u) = (e0 (u), e1 (u), e2 (u), e3 (u)) ,
where for each u ∈ U , we have
[e0 (u)] = [x(u)] ∈ P3
and

det e0 (u) e1 (u) e2 (u) e3 (u) = 1.

Any choice for the function e0 (u) defines a lifting x : U → R4 of [x], given
by
x(u) = e0 (u),
whose image is a surface Σ ⊂ R4 .
7.4. Moving frames for surfaces in P3 221

We might like to proceed as in the equi-affine case, by choosing (e1 (u), e2 (u))
so that they span the tangent space Tx(u) Σ. However, this choice is compli-
cated by the fact that this space is not well-defined: If we replace the lifting
x by an alternate lifting
x̂(u) = λ(u)x(u)
for some nonvanishing function λ : U → R, then the tangent vectors xu , xv ,
which span Tx(u) Σ, are replaced by the tangent vectors
x̂u = λxu + λu x, x̂v = λxv + λv x.
Since the function λ and its partial derivatives at any point are arbitrary
(aside from the requirement that λ = 0), the tangent space Tx(u) Σ is only
well-defined modulo the vector x(u) = e0 (u). Consequently, we can only
require that (e1 (u), e2 (u)) span Tx(u) Σ 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) Σ. In particular, we will require that this space be 3-
dimensional, which means that the position vector of Σ 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 → P3 to be an immersion.) A projective
frame field (e0 (u), e1 (u), e2 (u), e3 (u)) along [Σ] satisfying these conditions
will be called 0-adapted.
Exercise 7.32. Show that, given a smooth map [x] : U → P3 , the condition
that span(x, xu , xv ) be 3-dimensional is independent of the choice of lifting
x : U → R4 of [x].

Now, suppose that x̃ : U → SL(4) is a 0-adapted frame field along [Σ] =

[x(U )], and let (ω̄βα ) represent the pulled-back forms (x̃∗ ωβα ) on U . Recall
from §3.7 that the 1-forms (ω01 , ω02 , ω03 ) are the semi-basic forms for the pro-
jection π : SL(4) → P3 . Thus, these should be regarded as the dual forms,
and the remaining (ωβα ) as the connection forms.
Precisely the same reasoning as in the Euclidean and equi-affine cases can
be used to prove the following:
Proposition 7.33. Let U ⊂ R2 be an open set, and let [x] : U → P3 be
an immersion. For any 0-adapted frame field (e0 (u), e1 (u), e2 (u), e3 (u))
along [Σ] = [x(U )], the associated dual and connection forms (ω̄βα ) have the
property that ω̄03 = 0. Moreover, (ω̄01 , ω̄02 ) form a basis for the 1-forms on U .

Differentiating the equation ω̄03 = 0 yields

dω̄03 = −ω̄13 ∧ ω̄01 − ω̄23 ∧ ω̄02 = 0,
222 7. Curves and surfaces in projective space

and Cartan’s lemma (cf. Lemma 2.49) implies that there exist functions
h11 , h12 , h22 on U such that
 3   
ω̄1 h11 h12 ω̄01
(7.21) = .
ω̄23 h12 h22 ω̄02

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 field, with associated
Maurer-Cartan forms (ω̄βα ). Any other 0-adapted frame field (ẽ0 (u), ẽ1 (u),
ẽ2 (u), ẽ3 (u)) must have the properties that
span(ẽ0 (u)) = span(e0 (u)),
span(ẽ0 (u), ẽ1 (u), ẽ2 (u)) = span(e0 (u), e1 (u), e2 (u)),
and
 
det ẽ0 (u) ẽ1 (u) ẽ2 (u) ẽ3 (u) = det e0 (u) e1 (u) e2 (u) e3 (u) = 1.
This will be true if and only if

ẽ0 (u) ẽ1 (u) ẽ2 (u) ẽ3 (u)
⎡ ⎤
λ r1 r2 s0
(7.22) ⎢
⎢0 s1 ⎥
⎥
= e0 (u) e1 (u) e2 (u) e3 (u) ⎢ B ⎥
⎣0 s2 ⎦
0 0 0 (λ det B)−1
for some GL(2)-valued function B and real-valued functions λ, r1 , r2 , s0 , s1 ,
s2 on U , with λ = 0. Let (ω̄
˜ βα ) be the Maurer-Cartan forms associated to
the new frame field.
The computations in the following exercise should feel familiar from the
equi-affine case:
*Exercise 7.34 (Cf. Exercise 6.20). (a) Show that
 1  1
˜0
ω̄ ω̄0
= λB −1 .
˜ 02
ω̄ ω̄02

(b) Show that

 3  
˜1
ω̄ ω̄13
= λ(det B) tB .
˜ 23
ω̄ ω̄23
7.4. Moving frames for surfaces in P3 223

(c) Cartan’s lemma implies that there exist functions h̃11 , h̃12 , h̃22 on U such
that  3    1
˜1
ω̄ h̃11 h̃12 ω̄ ˜0
= .
˜ 23
ω̄ h̃12 h̃22 ω̄ ˜ 02
Show that
   
h̃11 h̃12 h11 h12
(7.23) = (det B) tB B.
h̃12 h̃22 h12 h22

The transformation (7.23) is precisely the same as in the equi-affine case, so

we use the same terminology here:
Definition 7.35. Assume that the matrix [hij ] is nonsingular at every point
of U . The surface [Σ] = [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 [Σ] 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 × 2 matrices with
positive determinant; therefore, there exists a choice of 0-adapted frame field
(e0 (u), e1 (u), e2 (u), e3 (u)) for which
   
h11 h12 1 0
= .
h12 h22 0 1
Such a frame field will be called 1-adapted.
*Exercise 7.36. Let (e0 (u), e1 (u), e2 (u), e3 (u)) be any 1-adapted frame
field for an elliptic projective surface [Σ] = [x(U )] ⊂ P3 .
(a) Show that any other 1-adapted frame field (ẽ0 (u), ẽ1 (u), ẽ2 (u), ẽ3 (u))
for [Σ] must have the form

ẽ0 (u) ẽ1 (u) ẽ2 (u) ẽ3 (u)
⎡ ⎤
λ r1 r2 s0
(7.24) ⎢
⎢0 s1 ⎥
⎥
= e0 (u) e1 (u) e2 (u) e3 (u) ⎢ B ⎥
⎣0 s2 ⎦
0 0 0 (λ det B) −1

for some SO(2)-valued function B and real-valued functions λ, r1 , r2 , s0 , s1 ,

s2 on U , with λ = 0.
224 7. Curves and surfaces in projective space

(b) Let (ω̄βα ) be the Maurer-Cartan forms associated to a 1-adapted frame

field (e0 (u), e1 (u), e2 (u), e3 (u)), and let (ω̄ ˜ βα ) be the Maurer-Cartan forms
associated to the 1-adapted frame field (7.24). Show that
   
˜ 13 ω̄
ω̄ ˜ 01 + ω̄
˜ 23 ω̄
˜ 02 = (ω̄
˜ 01 )2 + (ω̄
˜ 02 )2 = λ2 (ω̄01 )2 + (ω̄02 )2 = λ2 ω̄13 ω̄01 + ω̄23 ω̄02 .

(Cf. Exercise 7.34.)

Note that the result of Exercise 7.36(b) is slightly different from the equi-
affine case. Now the quadratic form
(7.25) I = ω̄13 ω̄01 + ω̄23 ω̄02 = (ω̄01 )2 + (ω̄02 )2
is well-defined only up to a scalar multiple. Rather than defining a metric,
it defines a conformal structure on [Σ], in which angles between tangent
vectors are well-defined, 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 differ-

entiate the equations
(7.26) ω̄13 = ω̄01 , ω̄23 = ω̄02 ,
which hold for the Maurer-Cartan forms associated to any 1-adapted frame
field.
*Exercise 7.38. Differentiate the equations (7.26) and use Cartan’s lemma
to conclude that there exist functions h111 , h112 , h122 , h222 on U such that
⎡ 1 ⎤ ⎡ ⎤
2ω̄1 − (ω̄00 + ω̄33 ) h111 h112  
⎢ ⎥ ⎢ ⎥ ω̄01
⎢ 1 2
ω̄2 + ω̄1 ⎥ ⎢ ⎥
(7.27) ⎣ ⎦ = ⎣h112 h122 ⎦ 2 .
ω̄0
2ω̄22 − (ω̄00 + ω̄33 ) h122 h222

Next, we need to compute how the functions (hijk ) vary under a transfor-
mation 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 λ = 1, so that
⎡ ⎤
1 r1 r2 s0
  ⎢0 1 0 s1 ⎥
ẽ0 (u) ẽ1 (u) ẽ2 (u) ẽ3 (u) = e0 (u) e1 (u) e2 (u) e3 (u) ⎢ ⎥
⎣0 0 1 s2 ⎦ .
0 0 0 1
7.4. Moving frames for surfaces in P3 225

Show that under such a transformation, we have

h̃111 = h111 + 3(r1 − s1 ),

h̃112 = h112 + (r2 − s2 ),
h̃122 = h122 + (r1 − s1 ),
h̃222 = h222 + 3(r2 − s2 ).

Therefore,

h̃111 + h̃122 = h111 + h122 + 4(r1 − s1 ),

h̃112 + h̃222 = h112 + h222 + 4(r2 − s2 ),

so there exists a choice of 1-adapted frame field (e0 (u), e1 (u), e2 (u), e3 (u))
for which

(7.28) h111 + h122 = h112 + h222 = 0.

Any 1-adapted frame field satisfying the condition (7.28) will be called 2-
adapted.

Remark 7.40. This is certainly not the only way that one might choose to
normalize these functions! But our experience with the equi-affine case—
where a similar relation held for any 2-adapted frame field—suggests 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
field for an elliptic projective surface [Σ] = [x(U )] ⊂ P3 .
(a) Show that any other 2-adapted frame field (ẽ0 (u), ẽ1 (u), ẽ2 (u), ẽ3 (u))
for [Σ] must have the form (7.24), where
   
r1 s1
(7.29) = λ tB .
r2 s2

(b) Show that under a transformation of the form (7.24) satisfying equation
(7.29), we have
   
h̃111 h111
−1 3
(7.30) =λ B .
h̃222 h222

(c) Let (ω̄βα ) be the Maurer-Cartan forms associated to the 2-adapted frame
˜ βα ) be the Maurer-Cartan forms
field (e0 (u), e1 (u), e2 (u), e3 (u)), and let (ω̄
associated to the 2-adapted frame field (ẽ0 (u), ẽ1 (u), ẽ2 (u), ẽ3 (u)). Show
226 7. Curves and surfaces in projective space

that
 1 3   2 3 
˜ 0 ) − 3ω̄
h̃111 (ω̄ ˜ 01 (ω̄ ˜ 0 ) − 3(ω̄
˜ 02 )2 + h̃222 (ω̄ ˜ 01 )2 ω̄
˜ 02
   
= λ2 h111 (ω̄01 )3 − 3ω̄01 (ω̄02 )2 + h222 (ω̄02 )3 − 3(ω̄01 )2 ω̄02 .

Conclude that the cubic form

P = hijk ω̄0i ω̄0j ω̄0k
(7.31) = h111 (ω̄01 )3 + 3h112 (ω̄01 )2 ω̄02 + 3h122 ω̄01 (ω̄02 )2 + h222 (ω̄02 )3
   
= h111 (ω̄01 )3 − 3 ω̄01 (ω̄02 )2 + h222 (ω̄02 )3 − 3(ω̄01 )2 ω̄02
is well-defined up to a scalar multiple, independent of the choice of a par-
ticular 2-adapted frame field on [Σ]. (P is the projective analog of the
Fubini-Pick form for surfaces in equi-affine 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 ∈ 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 find a 2-adapted frame based at u for which
it is equal to any nonzero vector we choose. Any point [x(u)] ∈ [Σ] at which
h111 = h222 = 0 will be called an umbilic point of Σ.

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 ∈ U , i.e., that
Σ contains no umbilic points. (We will treat the totally umbilic case, where
h111 ≡ h222 ≡ 0, in Exercises 7.48 and 7.49.) With this assumption, equation
(7.30) implies that there exists a choice of 2-adapted frame field for which
h111 = 2, h222 = 0.
Such a frame field will be called 3-adapted.
*Exercise 7.42. (a) Let (e0 (u), e1 (u), e2 (u), e3 (u)) be any 3-adapted frame
field for an elliptic projective surface [Σ] = [x(U )] ⊂ P3 with no umbilic
points. Show that any other 3-adapted frame field (ẽ0 (u), ẽ1 (u), ẽ2 (u),
ẽ3 (u)) for [Σ] must have the form described in Exercise 7.41, where either
λ = 1 and B is one of the three matrices
   √   √ 
1 0 1 −1 − 3 1 −1 3
, √ , √
0 1 2 3 −1 2 − 3 −1
or λ = −1 and B is one of the three matrices
   √   √ 
−1 0 1 1 − 3 1 1 3
, √ , √ .
0 −1 2 3 1 2 − 3 1
7.4. Moving frames for surfaces in P3 227

(b) Let (ω̄βα ) be the Maurer-Cartan forms associated to a 3-adapted frame

˜ βα ) be the Maurer-Cartan forms
field (e0 (u), e1 (u), e2 (u), e3 (u)), and let (ω̄
associated to any other 3-adapted frame field (ẽ0 (u), ẽ1 (u), ẽ2 (u), ẽ3 (u)).
Show that
˜ 01 )2 + (ω̄
(ω̄ ˜ 02 )2 = (ω̄01 )2 + (ω̄02 )2 .

Therefore, the conformal structure (7.25) associated to a 3-adapted frame

field provides a well-defined metric for an elliptic surface in P3 with no
umbilic points.

Definition 7.43. Let U ⊂ R2 be an open set, and let [x] : U → P3 be an

elliptic immersion with no umbilic points. Let (e0 (u), e1 (u), e2 (u), e3 (u))
be any 3-adapted frame field along [Σ] = [x(U )], with associated Maurer-
Cartan forms (ω̄βα ). The quadratic form

I = (ω̄01 )2 + (ω̄02 )2

is called the projective first fundamental form of [Σ].

Remark 7.44. For a 3-adapted frame field, the cubic form (7.31) becomes
 
P = 2 (ω̄01 )3 − 3 ω̄01 (ω̄02 )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)] ∈ [Σ]. These directions
are called the Darboux tangents at the point [x(u)] ∈ [Σ]; for more details,
see [Su83].

For simplicity, we will restrict our attention to transformations between 3-

adapted frames with B equal to the identity matrix; thus, we will assume
that any two 3-adapted frame fields vary by a transformation of the form
(7.32) ⎡ ⎤
1 s1 s2 s0
  ⎢0 1 0 s1 ⎥
ẽ0 (u) ẽ1 (u) ẽ2 (u) ẽ3 (u) = e0 (u) e1 (u) e2 (u) e3 (u) ⎢ ⎥
⎣0 0 1 s2 ⎦ .
0 0 0 1

*Exercise 7.45. (a) Show that the Maurer-Cartan forms associated to any
3-adapted frame field satisfy the equations

ω̄00 + ω̄33 = ω̄11 + ω̄22 = 0.

(Hint: Apply the result of Exercise 7.38, keeping in mind condition (7.28),
and recall the defining condition for the Lie algebra sl(4).)
228 7. Curves and surfaces in projective space

(b) Differentiate the equation

ω̄00 + ω̄33 = 0

and use Cartan’s lemma to conclude that there exist functions 11 , 12 , 22
on U such that
     1
ω̄10 − ω̄31 11 12 ω̄0
(7.33) = .
ω̄20 − ω̄32 12 22 ω̄02

*Exercise 7.46. (a) Show that under a transformation of the form (7.32),
we have

˜11 = 11 + s21 + s22 − 2s0 − 2s1 ,

˜12 = 12 + 2s2 ,
˜22 = 22 + s2 + s2 − 2s0 + 2s1 .
1 2

(b) Conclude that there exists a choice of 3-adapted frame field for which

11 = 12 = 22 = 0.

Any frame field satisfying this condition will be called 4-adapted.

(c) Show that any two 4-adapted frame fields vary by a transformation of
the form
⎤ ⎡
λ 0 0 0
 ⎢
⎢0 0 ⎥⎥
ẽ0 (u) ẽ1 (u) ẽ2 (u) ẽ3 (u) = e0 (u) e1 (u) e2 (u) e3 (u) ⎢ B ⎥,
⎣0 0 ⎦
0 0 0 λ−1

where (λ, B) are one of the pairs from Exercise 7.42. Therefore, the 4-
adapted frames at each point form a discrete set, and a continuous 4-adapted
frame field along [Σ] is uniquely determined by its value at any point u ∈ U .

*Exercise 7.47. (a) Show that the Maurer-Cartan forms associated to a

4-adapted frame field satisfy the equations

ω̄11 = ω̄01 , ω̄22 = −ω̄01 , ω̄21 + ω̄12 = −2ω̄02 ,

(7.34)
ω̄10 = ω̄31 , ω̄20 = ω̄32 .
7.4. Moving frames for surfaces in P3 229

(b) Differentiate the equations in part (a) and use Cartan’s lemma to con-
clude that there exist functions p1 , p2 , q11 , q12 , q21 , q22 such that
ω̄21 = p1 ω̄01 + (p2 − 1)ω̄02 ,
ω̄12 = −p1 ω̄01 − (p2 + 1)ω̄02 ,
ω̄31 = q11 ω̄01 + q12 ω̄02 ,
(7.35)
ω̄32 = q21 ω̄01 + q22 ω̄02 ,
ω̄00 = −ω̄33 = 3p2 ω̄01 − 3p1 ω̄02 ,
ω̄30 = (q11 − q22 )ω̄01 − (q12 + q21 )ω̄02 .
The functions p1 , p2 , q11 , q12 , q21 , q22 are the fundamental invariants for el-
liptic 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 field for an elliptic projective surface [Σ] = [x(U )] ⊂
P3 , and suppose that [Σ] 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 field satisfy
(7.36) ω̄11 = ω̄22 = ω̄21 + ω̄12 = ω̄00 + ω̄33 = 0.

(b) Differentiate equations (7.36) to obtain

(ω̄10 − ω̄31 ) ∧ ω̄01 = 0,
(ω̄20 − ω̄32 ) ∧ ω̄02 = 0,
(ω̄20 − ω̄32 ) ∧ ω̄01 + (ω̄10 − ω̄31 ) ∧ ω̄02 = 0.
Use Cartan’s lemma to conclude that there exists a function σ such that
ω̄10 − ω̄31 = σ ω̄01 ,
ω̄20 − ω̄32 = σ ω̄02 .

(c) Show that under a transformation of the form

⎡ ⎤
1 0 0 s0
  ⎢0 1 0 0⎥
ẽ0 (u) ẽ1 (u) ẽ2 (u) ẽ3 (u) = e0 (u) e1 (u) e2 (u) e3 (u) ⎢⎣0
⎥,
0 1 0⎦
0 0 0 1
we have
σ̃ = σ − 2s0 .
230 7. Curves and surfaces in projective space

Conclude that there exists a 2-adapted frame field with σ = 0 and that for
such a frame field we have
(7.37) ω̄31 = ω̄10 , ω̄32 = ω̄20 .
We will call such a frame field fully adapted.
(d) Show that any two fully adapted frame fields differ by a transformation
of the form

ẽ0 (u) ẽ1 (u) ẽ2 (u) ẽ3 (u)
⎡ ⎤
λ r1 r2 s0
(7.38) ⎢
⎢ 0 s1 ⎥⎥
= e0 (u) e1 (u) e2 (u) e3 (u) ⎢ B ⎥,
⎣0 s2 ⎦
0 0 0 λ−1
where
1
B ∈ SO(2), [r1 r2 ] = λ[s1 s2 ]B, s0 = λ(s21 + s22 ).
2

(e) Differentiate equations (7.37) to obtain

2ω̄30 ∧ ω̄01 = 0,
2ω̄30 ∧ ω̄02 = 0.

Use Cartan’s lemma to conclude that ω̄30 = 0.

(f) Show that differentiating the equation ω̄30 = 0 yields an identity.
At this point, the Maurer-Cartan form for the bundle of fully adapted frames
over [Σ] is
⎡ 0 0 0 ⎤
ω̄0 ω̄1 ω̄2 0
⎢ 1 ⎥
⎢ω̄0 0 ω̄21 ω̄10 ⎥
⎢ ⎥
ω̄ = ⎢ ⎥.
⎢ω̄ 2 −ω̄ 1 0 ω̄ 0 ⎥
⎣ 0 2 2 ⎦

0 ω̄01 ω̄02 −ω̄00

We have not yet found a unique frame over each point of [Σ], but since
differentiating 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 [Σ] 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 [Σ] 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

⎡ ⎤
0 0 0 −1
⎢0 10 0⎥
Q=⎢ ⎣ 0 0 1 0 ⎦.
⎥

−1 0 0 0
Q represents the quadratic form
Q(x) = (x1 )2 + (x2 )2 − 2x0 x3 ,
which is an indefinite quadratic form of signature (3, 1) on R4 . The Lie
group SO(Q) is defined to be the group of matrices of determinant 1 that
preserve this quadratic form; i.e.,
SO(Q) = {A ∈ 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 defined by
so(Q) = {B ∈ sl(4) | tBQ + QB = 0}.
(Hint: Let A(t) be a curve in SO(Q) with A(0) = I. Differentiate 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
⎡ 0 0 0 ⎤
a0 a1 a2 0
⎢ 1 ⎥
⎢a0 0 a12 a01 ⎥
B=⎢ ⎢a2 −a1 0 a0 ⎥ .
⎥
⎣ 0 2 2 ⎦
0 a10 a20 −a00
Conclude that the reduced Maurer-Cartan form on the bundle of fully a-
dapted frames in Exercise 7.48 takes values in so(Q).
(c) Recall that for any fully adapted frame field
x̃(u) = (e0 (u), e1 (u), e2 (u), e3 (u))
as in Exercise 7.48, the pullback of the Maurer-Cartan form ω̄ to U via x̃
is given by x̃∗ (ω̄) = x̃−1 dx̃. Use this fact to show that any fully adapted
frame field has the form
x̃(u) = A0 A(u),
where A0 ∈ SL(4) is a constant matrix and A(u) ∈ SO(Q) for all u ∈ U .
232 7. Curves and surfaces in projective space

Since A0 ∈ SL(4), the frame field x̃(u) is projectively equivalent to the

frame field A−1
0 x̃(u). (This equivalence replaces the surface

[Σ] = [x(U )] = [e0 (U )] ⊂ P3

with the projectively equivalent surface

[A−1 −1 −1
0 Σ] = [A0 x(U )] = [A0 e0 (U )] ⊂ P .)
3

Therefore, up to projective equivalence, we can assume that x̃(u) ∈ SO(Q)

for all u ∈ U .

(d) Show that for any matrix e0 e1 e2 e3 ∈ SO(Q), the vector e0 is a null
vector for Q; i.e., Q(e0 ) = 0. Therefore, the surface x(U ) = e0 (U ) ⊂ R4 \{0}
must be contained in the 3-dimensional hypersurface S ⊂ R4 \ {0} defined
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} → P3
may be regarded as a sphere in P3 , and [Σ] = [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 → P3 be a smooth immersion
whose image is a hyperbolic projective surface [Σ]. Then the matrix [hij ] in
(7.21) has det[hij ] < 0.
(a) Show that there exists a 0-adapted frame field (f0 (u), f1 (u), f2 (u), f3 (u))
along Σ for which
   
h11 h12 0 1
= .
h12 h22 1 0
Such a frame field will be called a 1-adapted null frame field along [Σ]. The
associated Maurer-Cartan forms (η̄βα ) satisfy

(7.39) η̄13 = η̄02 , η̄23 = η̄01 ,

and the conformal structure on [Σ] is the indefinite quadratic form

I = η̄13 η̄01 + η̄23 η̄02 = 2η̄01 η̄02 .

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 field for a
hyperbolic projective surface [Σ] = [x(U )] ⊂ P3 . Show that any other 1-
adapted null frame field (f̃0 (u), f̃1 (u), f̃2 (u), f̃3 (u)) for Σ must have the form
(7.40) ⎡ ⎤
λ r1 r2 s0
⎢ ⎥
  ⎢ 0 eθ 0 s1 ⎥
f̃0 (u) f̃1 (u) f̃2 (u) f̃3 (u) = f0 (u) f1 (u) f2 (u) f3 (u) ⎢ ⎢ 0 0 e−θ s ⎥
⎥
⎣ 2 ⎦
0 0 0 λ−1
for some real-valued functions λ, θ, r1 , r2 , s0 , s1 , s2 on U , with λ = 0.
(c) Differentiate equations (7.39) and use Cartan’s lemma to conclude that
there exist functions h111 , h112 , h122 , h222 on U such that
⎡ ⎤ ⎡ ⎤
2η̄12 h111 h112  
⎢ 1 ⎥ ⎢ ⎥ η̄ 1
⎢(η̄ + η̄ 2 ) − (η̄ 0 + η̄ 3 )⎥ = ⎢h112 h122 ⎥ 0 .
⎣ 1 2 0 3 ⎦ ⎣ ⎦ 2
η̄0
2η̄21 h122 h222

(d) Show that under a transformation of the form (7.40), we have

h̃111 = λ−1 e3θ h111 ,
h̃112 = λ−1 eθ h112 + 2(λ−1 r1 − eθ s2 ),
(7.41)
h̃122 = λ−1 e−θ h122 + 2(λ−1 r2 − e−θ s1 ),
h̃222 = λ−1 e−3θ h222 .
Therefore, there exists a choice of 1-adapted null frame field (f0 (u), f1 (u),
f2 (u), f3 (u)) for which
h112 = h122 = 0.
Any 1-adapted null frame field satisfying this condition will be called a 2-
adapted null frame field.
(e) Show that any other 2-adapted null frame field (f̃0 (u), f̃1 (u), f̃2 (u), f̃3 (u))
for [Σ] must have the form (7.40), where
(7.42) r1 = λeθ s2 , r2 = λe−θ 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 [Σ] is a ruled surface in P3 .) Equations (7.41) then
imply that there exists a choice of 2-adapted null frame field 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
fα (u) → −fα (u)
for an even number of indices α.) Any 2-adapted null frame field satisfying
this condition will be called a 3-adapted null frame field.
(f) Show that, up to signs, any two 3-adapted null frame fields vary by a
transformation of the form
⎡ ⎤
1 s2 s1 s0
  ⎢0 1 0 s1 ⎥
(7.43) f̃0 (u) f̃1 (u) f̃2 (u) f̃3 (u) = f0 (u) f1 (u) f2 (u) f3 (u) ⎢⎣0 0 1 s2 ⎦ .
⎥

0 0 0 1

(g) Show that the Maurer-Cartan forms associated to any 3-adapted null
frame field satisfy the equations
η̄00 + η̄33 = η̄11 + η̄22 = 0.
Differentiate the equation
η̄00 + η̄33 = 0
and use Cartan’s lemma to conclude that there exist functions 11 , 12 , 22
on U such that  0    
η̄1 − η̄32 11 12 η̄01
= .
η̄20 − η̄31 12 22 η̄02

(h) Show that under a transformation of the form (7.43), we have

˜11 = 11 − 2s1 ,
˜12 = 12 + 2s1 s2 − 2s0 ,
˜22 = 22 − 2s2 .
Conclude that there exists a choice of 3-adapted null frame field for which
11 = 12 = 22 = 0.
Any frame field satisfying this condition will be called a 4-adapted null frame
field. 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 field along [Σ]
is uniquely determined by its value at any point u ∈ U .
(i) Show that the Maurer-Cartan forms associated to a 4-adapted null frame
field satisfy the equations
η̄12 = η̄01 , η̄21 = η̄02 , η̄10 = η̄32 , η̄20 = η̄31 .
7.5. Maple computations 235

Differentiate these equations and use Cartan’s lemma to conclude that there
exist functions p1 , p2 , q11 , q12 , q21 , q22 such that
η̄11 = −η̄22 = p1 η̄01 + p2 η̄02 ,
η̄00 = −η̄33 = −3p1 η̄01 + 3p2 η̄02 ,
η̄31 = q11 η̄01 + q12 η̄02 ,
η̄32 = q21 η̄01 + q22 η̄02 ,
η̄30 = q12 η̄01 + q21 η̄02 .
The functions p1 , p2 , q11 , q12 , q21 , q22 are the fundamental invariants for non-
ruled, hyperbolic surfaces in P3 .

7.5. Maple computations

In this chapter, the computations for curves in P2 and P3 are already suffi-
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.
Exercise 7.25: The canonical lifting ẽ0 (s) of [β(s)] must be equal to a
scalar multiple of e0 (t(s)); say
ẽ0 (s) = λ(s) e0 (t(s)).
For Maple purposes, we will denote the canonical projective frame field for
[β(s)] by (f0(s), f1(s), f2(s), f3(s)).
> f0(s):= lambda(s)*e0(t(s));
Compute the first three derivatives of ẽ0 (s):
> 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 deriva-
tives of e0 (t(s)); for instance, it returns
f 1(s) := λs e0(t(s)) + λ(s) D(e0)(t(s)) ts
d
(The expression D(e0)(t(s)) refers to the derivative dt (e0 (t)) evaluated at
t = t(s).)
We don’t have any good way of telling Maple that these derivatives are all
linearly independent, but we can still collect terms and extract the coeffi-
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))});

Upon inspection, it becomes clear that det ẽ0 (s) ẽ1 (s) ẽ2 (s) ẽ3 (s) can be
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 := λ(s)4 t6s

Since this determinant must be equal to 1, we must have

1
λ(s) = .
(t (s))(3/2)

Now computing the rest of the canonical projective frame field for [β] is
simply a matter of differentiating the expression

1
ẽ0 (s) = e0 (t(s)).
(t (s))(3/2)

But keep in mind that we still need to compute the Wilczynski invariants,
which requires expressing ẽ3 (s) as a linear combination of (ẽ0 (s), ẽ1 (s),
ẽ2 (s)). This can get a bit messy, but we can keep things better organized if,
instead of explicitly expressing everything as functions of t and s, we instead
work with appropriately chosen differential forms. In this context, we assign
the derivatives (and the Wilczynski invariants) of the original frame field as
follows:
> d(e0):= e1*d(t);
d(e1):= e2*d(t);
d(e2):= e3*d(t);
d(e3):= 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
relate dt to ds and ẽ0 to e0 :
> PDETools[declare](r(s));
tsub:= [d(t) = r(s)*d(s)];
fsub:= [f0 = e0/r(s)ˆ(3/2)];
7.5. Maple computations 237

Compute the derivative of ẽ0 with respect to s by computing dẽ0 and then
substituting for dt in terms of ds:
> Simf(subs(tsub, Simf(d(Simf(subs(fsub, f0))))));
The coefficient of ds in this expression should be ẽ1 (s), so add it to fsub:
> fsub:= [op(fsub), f1 = pick(%, d(s))];
Similarly for ẽ2 (s) and ẽ3 (s):
> 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,
we need to express dẽ3 in terms of (ẽ0 , ẽ1 , ẽ2 ). The makebacksub command
won’t 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 dẽ3 and express the result in terms of the canonical
projective frame field for [β]:
> Simf(subs(esub, Simf(subs(tsub,
Simf(d(Simf(subs(fsub, f3))))))));
df3:= collect(pick(%, d(s)), {f0, f1, f2, f3});
As expected, we see that the output contains no ẽ3 term, and the invariants
κ̃0 (s), κ̃1 (s), κ̃2 (s) associated to [β] 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 κ0 (t), κ1 (t), κ2 (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 κ̃0 (s), κ̃1 (s), κ̃2 (s)
(with respect to s) in terms of κ0 (t), κ1 (t), κ2 (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
κ̃0 (s), κ̃1 (s), κ̃2 (s) and their derivatives that don’t contain any derivatives
of r(s), so that they transform tensorially under the change of variables
t → t(s). The first one is relatively easy to spot: The computations above
show that

κ̃1 (s) = r(s)3 κ1 (t) + 2 r(s)r (s)κ2 (t)

1    2  

+ 20 r(s)r (s)r (s) − 5 r(s) r (s) − 15 r (s) 3
r(s)3
(compare with equation (7.16)), while

κ̃2 (s) = r(s)3 κ2 (t) + 2 r(s)r (s)κ2 (t)

1  
+ 3
20 r(s)r (s)r (s) − 5 r(s)2 r (s) − 15 r (s)3 .
r(s)
Therefore, we have
 
κ̃1 (s) − κ̃2 (s) = r(s)3 κ1 (t) − κ2 (t) ,

and hence
     
κ̃1 (s) − κ̃2 (s) ds3 = κ1 (t) − κ2 (t) (r(s) ds)3 = κ1 (t) − κ2 (t) dt3 .

The quartic form requires a bit more finesse 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)*Kappa2ˆ2);
1 4 
r 100 κ0 + 9 κ22 + 20 kappa2 tt − 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 different 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(’omega[i,k] &ˆ omega[k,j]’,
k=0..3);
end do;
end do;
It’s a little bit clumsy to introduce the relation

ω̄00 + ω̄11 + ω̄22 + ω̄33 = 0

at this point because doing so would require solving for one of the forms in
terms of the others, and it’s not yet clear which form would be best to solve
for. So for now, we’ll simply keep the relation in mind, and we’ll tell Maple
about it when it becomes convenient to do so.
Next, set up the initial substitution for the Maurer-Cartan forms of a 0-
adapted frame field:
> 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, we’re 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) ˜ βα = A−1 dA + A−1 ω̄βα A,
ω̄

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 once—although you will
probably want to suppress the output so as not to have Maple spew out
pages and pages of nasty expressions that we don’t 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 you’ll definitely
want to suppress the output):
> RHSmatrix:= map(Simf,
MatrixInverse(groupmatrix).map(d, groupmatrix)
+ MatrixInverse(groupmatrix).omegamatrix.groupmatrix):
Now, here’s the desired substitution:
> framechangesub:= [
Omega[0,0] = RHSmatrix[1,1], Omega[0,1] = RHSmatrix[1,2],
Omega[0,2] = RHSmatrix[1,3], Omega[0,3] = RHSmatrix[1,4],
Omega[1,0] = RHSmatrix[2,1], Omega[1,1] = RHSmatrix[2,2],
Omega[1,2] = RHSmatrix[2,3], Omega[1,3] = RHSmatrix[2,4],
Omega[2,0] = RHSmatrix[3,1], Omega[2,1] = RHSmatrix[3,2],
Omega[2,2] = RHSmatrix[3,3], Omega[2,3] = RHSmatrix[3,4],
Omega[3,0] = RHSmatrix[4,1], Omega[3,1] = RHSmatrix[4,2],
Omega[3,2] = RHSmatrix[4,3], Omega[3,3] = RHSmatrix[4,4]]:
We need to be careful when constructing the reverse substitution; the expres-
sions on the right-hand sides in framechangesub involve exterior derivatives
of the group parameters as well as the forms (ω̄βα ), and this may prevent the
makebacksub command from working as desired. It’s safer to do the follow-
ing 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]];
Next, determine how the (h̃ij ) may be expressed in terms of the (hij ). (Note
that we need one extra substitution here that wasn’t needed in Exercise 6.20;
if you wonder why it’s 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 [Σ] is an elliptic surface and that we
have chosen a 1-adapted frame field, so that [hij ] is the identity matrix and
B ∈ SO(2). Since we now wish to explore transformations among 1-adapted
frame fields, assign these conditions for both (hij ) and (h̃ij ):
> 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);
Differentiate the first 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
φ1 ∧ ω̄01 + φ2 ∧ ω̄02
so that we can apply Cartan’s lemma:
pick(zero4, omega[1,0]);
−ω0,0 + 2 ω1,1 − ω3,3
pick(zero4, omega[2,0]);
ω1,2 + ω2,1

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]);
ω1,2 + ω2,1
> pick(zero5, omega[2,0]);
−ω0,0 + 2 ω2,2 − ω3,3

Equation (7.27) then follows from Cartan’s lemma. Now add these condi-
tions 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 func-
tions (h̃ijk ) as follows:
> 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 consid-
erably if we take θ = 0, λ = 1:
> Simf(subs([theta=0, lambda=1], H[1,1,1]));

h1,1,1 − 3 s1 + 3 r1

> Simf(subs([theta=0, lambda=1], H[1,1,2]));

h1,1,2 − s2 + r2

> Simf(subs([theta=0, lambda=1], H[1,2,2]));

h1,2,2 − s1 + r1

> Simf(subs([theta=0, lambda=1], H[2,2,2]));

h2,2,2 − 3 s2 + 3 r2

Exercise 7.41: Now we assume that both frame fields 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
be satisfied by the group parameters in order for it to hold for the (h̃ijk ) as
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(θ)λ s2 + λ s1 cos(θ), r2 = − sin(θ)λ s1 + λ cos(θ)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 fields, 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
λ = 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]));

ω0,0 + ω3,3

Therefore, since ω̄00 + ω̄11 + ω̄22 + ω̄33 = 0, we must have

ω̄00 + ω̄33 = ω̄11 + ω̄22 = 0.

We need to add these relations to adaptedsub1 (and give it a new name

since we’re working with a subcase) via the same two-step process that we
used in Chapter 6. Since the substitution already contains expressions for
ω̄11 and ω̄22 , we only need to add the relation ω̄00 + ω̄33 = 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 differentiate this relation:

> zero8:= Simf(subs(adaptedsub13,
Simf(d(omega[0,0] + omega[3,3]))));
Identify the appropriate 1-forms so that we can apply Cartan’s lemma:
> pick(zero8, omega[1,0]);
−ω0,1 + ω1,3
> pick(zero8, omega[2,0]);
−ω0,2 + ω2,3

It follows from Cartan’s 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 trans-
formation (7.32) between 3-adapted frame fields.
> 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 find a 3-adapted frame for which the
(ij ) are all equal to zero, it suffices to show that we can solve the equations
˜11 = ˜12 = ˜22 = 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 = − ell1,1 − ell2,2 + ell1,2 + ell2,2 − ell2,2 ell1,1 + ell1,1 ,
4 4 8 32 16 32
1 1 1
s1 = ell2,2 − ell1,1 , s2 = ell1,2 }
4 4 2
246 7. Curves and surfaces in projective space

Exercise 7.47: For a 4-adapted frame field, we have the following condi-
tions:
> 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 refine our 3-adapted frame adaptation by incorporating these con-
ditions as follows:
> adaptedsub14:= Simf(subs(fouradaptedsub, adaptedsub13));
We can now read off the relations (7.34) directly from adaptedsub14.
At this point, all the remaining Maurer-Cartan forms must be linear com-
binations of (ω̄01 , ω̄02 ). It follows from equations (7.34) that the first four
equations in (7.35) must hold for some functions (pi , qij ). Before we con-
tinue, 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 differentiate the first two equations in (7.34):
> Simf(subs(adaptedsub14, Simf(d(omega[1,1] - omega[1,0]))));
> pick(%, omega[1,0]);

−ω0,0 − 3 p1 ω2,0

By Cartan’s lemma, (ω̄00 + 3p1 ω̄02 ) must be equal to a multiple of ω̄01 .

> Simf(subs(adaptedsub14,
Simf(d(omega[1,2] + omega[2,1] + 2*omega[2,0]))));
> pick(%, omega[2,0]);

2 ω0,0 − 6 p2 ω1,0

By Cartan’s lemma, (2ω̄00 − 6p2 ω̄01 ) must be equal to a multiple of ω̄02 . To-
gether, these conditions imply that

ω̄00 = 3p2 ω̄01 − 3p1 ω̄02 .

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, differentiating the last two equations in (7.34) and applying Cartan’s
lemma in this same fashion yields
ω̄30 = (q11 − q22 )ω̄01 − (q12 + q21 )ω̄02 ,
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
 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 find a surface of min-
imum 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 films in order to study the properties of such surfaces.
Despite the long history, rigorous existence results for area-minimizing sur-
faces 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 “locally area minimizing”.

More precisely, Σ ⊂ E3 is minimal if for any sufficiently small open set
V ⊂ Σ, 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
252 8. Minimal surfaces in E3 and A3

calculus of variations, see, e.g., [Dac08] or [Olv00].) Intuitively, we consider

the set S of closed and bounded surfaces Σ ⊂ E3 to be an infinite-dimensional
space. (Making this notion precise would require that we define a topology
on this space, but we won’t need this level of technicality.) We then define
the area functional A on this space to be
&
A(Σ) = dA;
Σ

i.e., A(Σ) is simply the area of Σ. The key idea is that minimal surfaces
should be critical points of this functional.
Unfortunately, since the space S is infinite-dimensional (with a rather com-
plicated topology!), it isn’t entirely obvious how to go about finding 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 →R

is a smooth function on a finite-dimensional manifold M and q0 ∈ M is a

critical point of f , then for any smooth curve α : I → M with α(0) = q0 ,
we must have 
d 
(f ◦ α) = 0.
dt 
t=0
Conversely, if q0 ∈ M is not a critical point
 of f , then there exists a smooth
d
curve α : I → M with α(0) = q0 and dt t=0 (f ◦ α) = 0.
If we could come up with a good definition for a “smooth curve” in the
space S, then we could apply this same idea: In order for a surface Σ to
be a critical point of A, it should have the property that for any “smooth
curve” α : I → S with α(0) = Σ,

d 
(8.1) (A ◦ α) = 0.
dt t=0
Conversely, if Σ is not a critical point of A,
 then there should exist a “smooth
d
curve” α : I → S with α(0) = Σ and dt t=0
(A ◦ α) = 0.

Remark 8.1. It is customary to denote the surface α(t) ∈ S by Σt , so that

Σ0 = Σ and the condition (8.1) becomes

d 
A(Σt ) = 0.
dt t=0

The intuitive idea of a “smooth curve” in S can be defined rigorously as

follows.
8.2. Minimal surfaces in E3 253

Definition 8.2. Let U ⊂ R2 be an open set, and let x : U → E3 be a

smooth immersion whose image is a surface Σ = x(U ). A variation of x is
a smooth map
X : U × (−ε, ε) → E3
for some ε > 0, with the properties that:

(1) For each t ∈ (−ε, ε), the set Σt = X(U, t) is a regular surface in E3 .
(2) For each u ∈ U , X(u, 0) = x(u). (In particular, Σ0 = Σ.)

So intuitively, a variation defines a 1-parameter family of surfaces Σt ⊂ E3

that “vary smoothly” with t.
Definition 8.3. A variation X of an immersion x : U → E3 is called

(1) compactly supported if there exists a compact set D ⊂ U such that

for all t ∈ (−ε, ε) and all u ∈ U \ D,
X(u, t) = X(u, 0) = x(u);
(2) normal if the vector ∂X∂t (u, t) is parallel to the unit normal vector
to the surface Σt for all t ∈ (−ε, ε) and all u ∈ U .

We will only consider compactly supported, normal variations. For surfaces

with a given boundary, the restriction to compactly supported variations cor-
responds to considering only variations that leave the boundary fixed, 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
finite.) The restriction to normal variations will simplify the analysis of
minimal surfaces, and the following exercise shows that any compactly sup-
ported variation can be made normal via a reparametrization; hence, this
assumption is not really a significant restriction.
*Exercise 8.4. Let x : U → E3 be an immersion, and let X : U × (−ε, ε) →
E3 be a compactly supported variation of x. For each t ∈ (−ε, ε), the
function xt : U → E3 defined by
xt (u) = X(u, t)
defines a parametrization of the surface Σt .
Consider a reparametrization X̄ : U  × (−ε, ε) → E3 of X of the form
X̄(ū, t) = X(u(ū, t), t),
with u(ū, 0) = ū for all ū ∈ U  . If we write u = (u1 , u2 ), then we can write
this reparametrization as
 
X̄ ū1 , ū2 , t) = X(u1 (ū1 , ū2 , t), u2 (ū1 , ū2 , t), t .
254 8. Minimal surfaces in E3 and A3

(a) Show that for each t ∈ (−ε, ε), the function

x̄t (ū) = X̄(ū, t)

is a reparametrization of the surface Σt ; i.e., the image of x̄t is the same as

the image of xt . Moreover, x̄0 = x0 = x.
(b) Show that
∂ X̄ ∂u1 ∂X ∂u2 ∂X ∂X
= + + .
∂t ∂t ∂u1 ∂t ∂u2 ∂t

(c) Let nt (u) denote the unit normal vector to the surface Σt at the point
xt (u). Then we can decompose ∂X ∂t as

∂X ∂X ∂X
= a(u, t) 1 + b(u, t) 2 + c(u, t)nt
∂t ∂u ∂u
∂ X̄
for some functions a(u, t), b(u, t), c(u, t). Show that the condition that ∂t
is parallel to nt (u) is equivalent to the system of differential equations

∂u1 ∂u2
(8.2) = −a(u1 , u2 , t), = −b(u1 , u2 , t)
∂t ∂t

for the functions u1 (ū, t), u2 (ū, t). (These equations should be regarded as
ordinary differential equations, with t as the independent variable and ū =
(ū1 , ū2 ) as parameters.) Moreover, the condition u(ū, 0) = ū is equivalent
to the initial conditions

(8.3) u1 (ū1 , ū2 , 0) = ū1 , u2 (ū1 , ū2 , 0) = ū2 .

(d) Conclude from the existence/uniqueness theorem for ordinary differen-

tial equations that for each choice of (ū1 , ū2 ), the system (8.2), (8.3) has
a unique solution for t in some interval (−ε, ε). Therefore, there exists a
reparametrization X̄ of X (possibly defined for a smaller value of ε than
the original variation) with the property that the variation defined by X̄ is
normal.

Remark 8.5. You might worry that, because the value of ε in the previous
exercise depends on ū, it could happen that there is no single ε > 0 that
suffices for all ū ∈ U  . Fortunately, the hypothesis that the variation is com-
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
ε > 0 such that the desired variation X̄ is well-defined on U  × (−ε, ε).
8.2. Minimal surfaces in E3 255

Now we can give a rigorous definition that corresponds to our intuitive idea
of a critical point for the area functional A on S:

Definition 8.6. Let U ⊂ R2 be an open set, and let x : U → E3 be an

immersion whose image is a surface Σ = x(U ). Σ is called a minimal surface
if for every compactly supported variation X : U × (−ε, ε) of x, we have

d 
(8.4) A(Σt ) = 0.
dt t=0
Remark 8.7. Since the area functional is invariant under reparametriza-
tions, minimality is a property of the surface Σ, independent of the choice of
parametrization x : U → E3 of Σ. Moreover, in order to show that a given
immersion x is minimal, it suffices 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 different areas spanning a given boundary curve or
curves. However, it can be shown that any minimal surface Σ is locally
area-minimizing in the sense that, given any point q ∈ Σ, there exists a
neighborhood V ⊂ Σ of q that is area-minimizing among all surfaces with
the same boundary curve as V .

Now, it still isn’t entirely obvious how to use this definition to find minimal
surfaces. In order to find critical points q0 for a function f : M → R on a
finite-dimensional manifold M , it suffices to identify those points q0 ∈ M
where all the partial derivatives of f vanish; however, for the functional
A on the infinite-dimensional space S, there is no finite set of compactly
supported variations that plays a role analogous to the partial derivatives of
f . Instead, we will need to find a way to work directly with Definition 8.6.
Let x : U → E3 be an immersion whose image is a surface Σ ⊂ E3 . We
begin by choosing an adapted orthonormal frame field (e1 (u), e2 (u), e3 (u))
along Σ, so that (e1 (u), e2 (u)) span the tangent plane Tx(u) Σ and e3 (u) is
orthogonal to Tx(u) Σ. Recall that choosing an orthonormal frame field along
Σ is equivalent to choosing a lifting x̃ : U → E(3) defined by

x̃(u) = (x(u); e1 (u), e2 (u), e3 (u)) .

The pullbacks (ω̄ i , ω̄ji ) to U of the Maurer-Cartan forms (ω i , ωji ) on E(3) via
the lifting x̃ satisfy the conditions that ω̄ 3 = 0 and
 3   
ω̄1 h11 h12 ω̄ 1
=
ω̄23 h12 h22 ω̄ 2
256 8. Minimal surfaces in E3 and A3

for some functions h11 , h12 , h22 on U , and the Gauss and mean curvature
functions of Σ (cf. Definition 4.45) are given by
K = h11 h22 − h212 , H = 12 (h11 + h22 ).

Now, consider a compactly supported, normal variation

X : U × (−ε, ε) → E3

d
of x. In order to compute dt t=0
A(Σt ), we will choose an orthonormal frame
field on the variation X, i.e., a lifting X 8 : U × (−ε, ε) → E(3)—and consider
the pullbacks (ω̄ , ω̄j ) of the Maurer-Cartan forms on E(3) to U × (−ε, ε)
i i

8
via X.
We can define such a frame field as follows: For each (u, t) ∈ U × (−ε, ε),
let (e1 (u, t), e2 (u, t), e3 (u, t)) be an orthonormal frame for the surface Σt at
the point xt (u), with e3 (u, t) normal to the tangent plane Txt (u) Σt . Then
the pullbacks (ω̄ i ) of the (ω i ) to U via X 8 must satisfy the equation

dX = ei ω̄ i .
*Exercise 8.9. (a) For each t ∈ (−ε, ε), let ıt : U → U × (−ε, ε) denote
the inclusion map defined by ıt (u) = (u, t). Show that the pullbacks of the
1-forms
ω̄ 1 = dX, e1 , ω̄ 2 = dX, e2 
to U via ıt are the usual dual forms on the surface Σt = xt (U ). (Hint: This
isn’t 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 ◦ ıt
and the fact that pullbacks behave well with respect to composition—spe-
cifically,
(xt )∗ = (ıt )∗ ◦ (X)∗ .)

(b) Show that

 
 ∂X 
ω̄ = dX, e3  = ± 
3  dt.
∂t 
(The sign is dependent on the choice of orientation for Σ, as determined by
the choice of the vector field e3 (u) on Σ.)

In particular, note that ω̄ 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
(ω̄ 1 , ω̄ 2 , dt) are linearly independent and form a basis for the 1-forms on the
3-dimensional manifold U × (−ε, ε).
8.2. Minimal surfaces in E3 257

 
For convenience, let χ(u, t) = ±  ∂X 
∂t (u, t) , with the sign chosen so that
3
ω̄ = χ(u, t) dt.
*Exercise 8.10. (a) Differentiate the equation ω̄ 3 = χ dt to obtain
ω̄13 ∧ ω̄ 1 + ω̄23 ∧ ω̄ 2 + dχ ∧ dt = 0.

(b) Apply Cartan’s lemma to conclude that

⎡ 3⎤ ⎡ ⎤ ⎡ 1⎤
ω̄1 h11 h12 χ1 ω̄
⎢ 3⎥ ⎢ ⎥ ⎢ 2⎥
(8.5) ⎣ω̄2 ⎦ = ⎣h12 h22 χ2 ⎦ ⎣ω̄ ⎦
dχ χ1 χ2 χ3 dt
for some functions h11 , h12 , h22 , χ1 , χ2 , χ3 on U × (−ε, ε).
(c) Show that the functions (hij ) are precisely the coefficients of the sec-
ond fundamental form of the surface Σt , while the functions (χi ) are the
directional derivatives of χ in the directions of the vectors ei (u, t).

Now, recall that for any adapted orthonormal frame field x̃ : U → E(3) on
the surface Σ = x(U ), the area of Σ is given by
& &
A(Σ) = ω̄ ∧ ω̄ =
1 2
x̃∗ (ω 1 ∧ ω 2 )
U U
(cf. Exercise 4.47). Applying this formula to each surface Σt in the variation
yields &
A(Σt ) = x̃∗t (ω 1 ∧ ω 2 ).
U
The surface Σ = Σ0 is minimal if and only if for every compactly supported
normal variation, we have

d 
0 =  A(Σt )
dt t=0
 &
d 
(8.6) =  x̃∗ (ω 1 ∧ ω 2 )
dt t=0 U t
& 
d   ∗ 1 
=  x̃t (ω ∧ ω 2
) .
U dt t=0

*Exercise 8.11. Show that

x̃∗t (ω 1 ∧ ω 2 ) = ı∗t (ω̄ 1 ∧ ω̄ 2 ),
8 (cf. Exercise
where (ω̄ 1 , ω̄ 2 ) are the pullbacks of (ω 1 , ω 2 ) to U ×(−ε, ε) via X
8.9). Therefore, the condition (8.6) can be written as
& 
d   ∗ 1 
(8.7)  ıt (ω̄ ∧ ω̄ 2 ) = 0.
U dt t=0
258 8. Minimal surfaces in E3 and A3

In order to evaluate this integral, recall from §2.11 that the Lie derivative
of a p-form Φ along a vector field v is the p-form
ϕ∗t Φ − Φ
Lv Φ = lim ,
t→0 t
where ϕt is the flow of the vector field v. In other words,

d 
Lv Φ =  ϕ∗t Φ.
dt t=0

*Exercise 8.12. (a) Show that the inclusion map ıt : U → U × (−ε, ε) may
∂
be regarded as the flow of the vector field ∂t on U × (−ε, ε), restricted to
the set U × {0} ⊂ U × (−ε, ε).
(b) Use part (a) to show that

d   ∗ 1 
 ıt (ω̄ ∧ ω̄ 2 ) = L∂/∂t (ω̄ 1 ∧ ω̄ 2 ).
dt t=0
(See §2.11 for the relevant definitions.)
(c) Use equation (8.7) to conclude that if the immersion x : U → E3 is
minimal, then for any compactly supported, normal variation of x, we must
have
&
(8.8) L∂/∂t (ω̄ 1 ∧ ω̄ 2 ) = 0.
U

Now we’re ready to put all the pieces together!

*Exercise 8.13. Let x : U → E3 be an immersion, and let X : U ×(−ε, ε) →
E3 be a compactly supported, normal variation of x.
(a) Use Cartan’s formula for the Lie derivative (cf. Theorem 2.55) to show
that
∂  
L∂/∂t (ω̄ 1 ∧ ω̄ 2 ) = d(ω̄ 1 ∧ ω̄ 2 ) .
∂t
(Hint: Since X is a normal variation of x, the vector
" #
∂X ∂
= X∗
∂t ∂t
is a multiple of the frame vector e3 (u, t). Therefore,
" # " #
∂X ∂X
ω1 = ω2 = 0,
∂t ∂t
and pulling these equations back via X∗ yields
" # " #
1 ∂ 2 ∂
ω̄ = ω̄ = 0.
∂t ∂t
8.2. Minimal surfaces in E3 259

(b) Use the Cartan structure equations, the equation ω̄ 3 = χ dt, and equa-
tion (8.5) to show that
d(ω̄ 1 ∧ ω̄ 2 ) = −(h11 + h22 )χ ω̄ 1 ∧ ω̄ 2 ∧ dt = −2Hχ ω̄ 1 ∧ ω̄ 2 ∧ dt.

(c) Use parts (a) and (b) to show that

& &
L∂/∂t (ω̄ 1 ∧ ω̄ 2 ) = −2Hχ ω̄ 1 ∧ ω̄ 2 .
U U

(d) Conclude that if the mean curvature H of Σ is identically equal to zero,

then &
L∂/∂t (ω̄ 1 ∧ ω̄ 2 ) = 0
U
for every compactly supported, normal variation of x, and hence Σ is mini-
mal.

Thus, we have proved the following proposition:

Proposition 8.14. If a regular surface Σ ⊂ E3 has mean curvature H
identically equal to zero, then Σ is minimal.

In the following exercise, we will prove the converse of Propsition 8.14: If

Σ = x(U ) is a surface whose mean curvature is not identically zero, then Σ
is not a critical point for the area functional A, and hence Σ is not minimal.
*Exercise 8.15. Let Σ = x(U ) be a surface whose mean curvature is not
identically zero.
(a) Let u0 ∈ U be a point where H(u0 ) = 0. Show that there exists a
neighborhood U  ⊂ U of u0 whose closure Ū  is contained in U and such
that H is nonzero and does not change sign on U  .
(b) Let X : U × (−ε, ε) be a normal variation of x supported on Ū  ; i.e.,
X(u, t) = x(u)
for all u ∈ U \ Ū  ,
chosen so that ω̄ 3 = χ dt for some function χ that is
nonzero in a neighborhood of {u0 } × (−ε, ε) and has the same sign as H
wherever it is nonzero. Show that for this variation,

d 
A(Σt ) < 0.
dt t=0
Conclude that Σ = Σ0 is not a critical point for A.

Together with Proposition 8.14, this proves the following theorem:

Theorem 8.16. A regular surface Σ ⊂ E3 is minimal if and only if its mean
curvature H is identically equal to zero.
260 8. Minimal surfaces in E3 and A3

Remark 8.17. Theorem 8.16 is often taken as a definition for minimal

surfaces because it is much easier to work with than Definition 8.6.

In the following two exercises, we explore two classical minimal surfaces.

Exercise 8.18. The catenoid is the surface Σ ⊂ 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 field

xu
e1 (u, v) = = t[− sin(u), cos(u), 0],
|xu |
xv 1 t
e2 (u, v) = = [cos(u) sinh(v), sin(u) sinh(v), 1],
|xv | cosh(v)
1 t
e3 (u, v) = e1 (u, v) × e2 (u, v) = [cos(u), sin(u), − sinh(v)]
cosh(v)
is orthonormal and that (e1 (u, v), e2 (u, v)) span the tangent space to Σ at
each point x(u, v) ∈ Σ.
(b) Show that the dual forms of this frame field are

ω̄ 1 = cosh(v) du, ω̄ 2 = cosh(v) dv.

(c) Compute de3 , and show that

1 1
ω̄13 = − du, ω̄23 = dv.
cosh(v) cosh(v)

(d) Use the results of parts (b) and (c) to compute the matrix [hij ], and
show that the mean curvature of Σ is H = 0. Therefore, the catenoid is a
minimal surface.
(e) (Maple recommended) Repeat the computations of parts (a)–(d) for an
arbitrary non-planar surface of revolution, parametrized by

x(u, v) = t[ρ(v) cos(u), ρ(v) sin(u), v].

Show that the surface is minimal if and only if the function ρ(v) satisfies
the differential equation

(8.9) ρρ = (ρ )2 + 1.

8.2. Minimal surfaces in E3 261

(f) Show that the only solutions of equation (8.9) are

1
cosh(av + b),
ρ(v) =
a
where a, b are constants. Conclude that catenoids are the only non-planar
minimal surfaces of revolution.
Exercise 8.19. The helicoid is the ruled surface Σ ⊂ E3 parametrized by
x(u, v) = t[v cos(u), v sin(u), u].

(a) Show that the frame field

xu 1
e1 (u, v) = =√ t
[−v sin(u), v cos(u), 1],
|xu | v2 + 1
xv
e2 (u, v) = = t[cos(u), sin(u), 0],
|xv |
1
e3 (u, v) = e1 (u, v) × e2 (u, v) = √ t
[− sin(u), cos(u), −v]
2
v +1
is orthonormal and that (e1 (u, v), e2 (u, v)) span the tangent space to Σ at
each point x(u, v) ∈ Σ.
(b) Show that the dual forms of this frame field are
ω̄ 1 = v 2 + 1 du, ω̄ 2 = dv.

(c) Compute de3 , and show that

1 1
ω̄13 = dv, ω̄23 = √ du.
v2 +1 2
v +1

(d) Use the results of parts (b) and (c) to compute the matrix [hij ], and
show that the mean curvature of Σ is H = 0. Therefore, the helicoid is a
minimal surface.

8.2.2. The Weierstrass-Enneper representation for minimal sur-

faces. There is a beautiful connection between minimal surfaces in E3 and
the theory of holomorphic functions of a complex variable, which was first de-
scribed 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 Σ ⊂ E3 be a regular surface with parametrization x : U → E3 , and let
(e1 (u), e2 (u), e3 (u)) be an orthonormal frame field on Σ, with e3 (u) normal
to the tangent plane Tx(u) Σ at each point x(u) ∈ Σ. As we saw in the
262 8. Minimal surfaces in E3 and A3

previous subsection, the associated Maurer-Cartan forms (ω̄ i , ω̄ji ) have the
property that
 3   
ω̄1 h11 h12 ω̄ 1
=
ω̄23 h12 h22 ω̄ 2
for some functions (hij ) on U , and Σ is minimal if and only if h11 + h22 = 0.
Consider the complex, vector-valued 1-form

ξ = (e1 − ie2 )(ω̄ 1 + iω̄ 2 )

= (e1 ω̄ 1 + e2 ω̄ 2 ) + i(e1 ω̄ 2 − e2 ω̄ 1 )
√
on U , where i = −1.

Remark 8.20. In keeping with the notation used throughout this book,
we will continue to use (ω̄ i , ω̄ji ) to denote the (real-valued!) Maurer-Cartan
forms associated to the orthonormal frame field (e1 (u), e2 (u), e3 (u)). In
order to avoid confusion, we will use the notation z ∗ rather than z̄ to denote
complex conjugation.

*Exercise 8.21. (a) Show that ξ is a well-defined 1-form on U , indepen-

dent of the choice of orthonormal frame field (e1 (u), e2 (u), e3 (u)). (Hint:
This is the same sort of computation that you used to show that the first
fundamental form was well-defined in Exercise 4.24.)
(b) Show that
dξ = i(h11 + h22 ) e3 ω̄ 1 ∧ ω̄ 2 .
Therefore, dξ = 0 if and only if Σ is minimal.

Now suppose that Σ is minimal. Let (e1 (u), e2 (u), e3 (u)) be any adapted or-
thonormal frame field along Σ, with associated Maurer-Cartan forms (ω̄ i , ω̄ji ),
and consider the complex, scalar-valued 1-form ω̄ 1 + iω̄ 2 on U .

*Exercise 8.22. (a) Use the Frobenius theorem (cf. Theorem 2.33) to show
that every point u ∈ U has a neighborhood V ⊂ U on which there exist
complex-valued functions z, ϕ : V → C such that

(8.10) ω̄ 1 + iω̄ 2 = ϕ dz.

(Hint: Show that the hypothesis of Theorem 2.33 is automatically satisfied

for any 1-form on a 2-dimensional surface. Moreover, this theorem holds for
complex-valued 1-forms Φ, 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 ϕ must be nonzero at every point of V . (Hint:
The 2-form

(ω̄ 1 + iω̄ 2 ) ∧ (ω̄ 1 + iω̄ 2 )∗ = (ω̄ 1 + iω̄ 2 ) ∧ (ω̄ 1 − iω̄ 2 ) = −2iω̄ 1 ∧ ω̄ 2

is nonvanishing on Σ.)
(c) Write the function z as
z = u + iv,

where u, v : V → R are real-valued functions on V , and let z ∗ = u − iv

denote the complex conjugate of z. Use your calculation from part (b) to
show that
dz ∧ dz ∗ = −2i du ∧ 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 ⊂ 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 ∗ = u − iv form a complex-valued local
coordinate system (z, z ∗ ) on V .
(a) Show that we can write the original coordinates (u, v) as

1 1
u = (z + z ∗ ), v= (z − z ∗ ).
2 2i

Now, let w : V → C be a complex-valued, differentiable function on V . The

function w may be considered as a function of either the coordinates (u, v)
or the coordinates (z, z ∗ ). The function w is called holomorphic or complex
analytic if, when considered as a function w(z, z ∗ ), it satisfies the condition

∂w
(8.11) = 0,
∂z ∗
i.e., if w is a function of z alone.
(b) Show that equation (8.11) is equivalent to

(8.12) wu + iwv = 0.
264 8. Minimal surfaces in E3 and A3

(c) Write
w(u, v) = x(u, v) + iy(u, v),
where x, y : V → R are real-valued, differentiable functions on V . Show
that equation (8.12) is equivalent to the pair of equations
(8.13) xu = yv , xv = −yu .
Equations (8.13) are called the Cauchy-Riemann equations.
(d) Show that any differentiable functions x, y : V → 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 satisfies
dz ∧ dz ∗ = 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 ⊂ U in the form x(z), thereby defining a complex structure
on Σ; in other words, it gives Σ the structure of a 1-dimensional complex
manifold.
Next, with ϕ as in equation (8.10), let f : V → C3 be the complex, vector-
valued function
f (z, z ∗ ) = (e1 (z, z ∗ ) − ie2 (z, z ∗ ))ϕ(z, z ∗ ).
Then we can write ξ as
ξ = f (z, z ∗ ) dz.
*Exercise 8.24. (a) Show that the assumption that Σ is minimal, and
hence dξ = 0, is equivalent to
∂f
= 0.
∂z ∗
Therefore, if Σ is minimal, then f (z, z ∗ ) = 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 Poincaré lemma (cf. Theorem 2.31) to the closed 1-form ξ =
f (z) dz to conclude that there exists a vector-valued, holomorphic function
8.2. Minimal surfaces in E3 265

z : V → C3 such that
ξ = dz,
and therefore, f (z) = z (z).
(d) Show that
Re(dz) = Re(ξ) = e1 ω̄ 1 + e2 ω̄ 2 = dx,
where x : U → Σ is the original parametrization of Σ. (Here Re denotes
the “real part”; 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 → 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 ∈ V .)

Exercise 8.24 shows that any minimal surface x : U → E3 can locally

be written as the real part of a holomorphic function z : U → C3 with
z (z), z (z) = 0. The next exercise shows that the converse is true as well.
*Exercise 8.25. Let V ⊂ C be an open set with local coordinates (u, v) and
complex coordinate z = u + iv. Let z : V → 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 → R3 are real vector-valued functions on V .
(a) Use the Cauchy-Riemann equations (8.13) to show that
dz = (xu − ixv )(du + i dv) = (xu − ixv ) dz.
Therefore, z (z) = xu − ixv .
(b) Use the condition z (z), z (z) = 0 to show that the parametrization
x : V → 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 Σ = x(V ) has mean curvature H = 0 and therefore Σ is minimal.
(Hint: Exercise 4.27 may be helpful.)
266 8. Minimal surfaces in E3 and A3

Together, Exercises 8.24 and 8.25 imply the following proposition:

Proposition 8.26. A regular surface Σ ⊂ E3 is minimal if and only if

every point of Σ has a neighborhood that can be parametrized by a smooth
immersion x : V → E3 that is the real part of a holomorphic function
z : V → C3 with z (z), z (z) = 0.

This brings us to the Weierstrass-Enneper representation for minimal sur-

faces, which is defined as follows. Let U ⊂ C be open, and suppose that we
are given

(1) a meromorphic function g : U → C, i.e., a holomorphic function

that may have isolated singularities z0 ∈ U called poles, where
g(z) = (z−z1 0 )k h(z) for some holomorphic function h(z) and some
integer k ≥ 1, called the order of the pole at z0 ;
(2) a holomorphic function f : U → C with the properties that when-
ever g has a pole of order k at z0 ∈ 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 ∈ U , and define z : U → C3 by

(8.15)
t& z & z & z 
2 f (ζ)(1 − g(ζ) ) dζ,
1 2 i 2
z(z) = 2 f (ζ)(1 + g(ζ) ) dζ, f (ζ)g(ζ) dζ .
z0 z0 z0

The following exercise shows that for any choice of functions f, g : U → C as

above, the real part of z(z) defines 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 ∈ U and that z (z), z (z) = 0. Conclude
that x = Re(z) : U → 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 ⊂ R2 , and let x : U → E3 be a parametrization

for an immersed minimal surface Σ ⊂ R3 . Choose a complex coordinate z
on U as in Exercise 8.22 and a holomorphic function z : U → C3 such that
z (z), z (z) = 0 and x = Re(z).
8.2. Minimal surfaces in E3 267

(a) Write z (z) = t[ξ1 , ξ2 , ξ3 ], where ξ1 , ξ2 , ξ3 : U → C3 are holomorphic

functions. Show that the condition z (z), z (z) = 0 is equivalent to
ξ12 + ξ22 + ξ32 = 0.

(b) Define
ξ3 (z)
f (z) = ξ1 (z) − iξ2 (z), g(z) = .
ξ1 (z) − iξ2 (z)
Show that f is holomorphic and that if g has a pole of order k at z0 ∈ 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 ∈ U where ξ1 (z0 ) − iξ2 (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 →

E3 be a parametrization for a minimal surface Σ ⊂ E3 with Weierstrass-
Enneper representation x = Re(z), with z as in (8.15).
(a) Show that the first fundamental form of Σ can be written as
I = 12 dz, dz∗ .
(Hint: Recall that dz = ω̄ 1 + iω̄ 2 .)
(b) For each t ∈ [0, 2π], let
zt (z) = eit z(z).
Show that the function zt is holomorphic, with zt (z), zt (z) = 0, and there-
fore, the function xt = Re(zt ) defines a parametrization for a minimal sur-
face Σt ⊂ E3 . The 1-parameter family of minimal surfaces Σt is called the
associated family of Σ.
(c) Use part (a) to show that the first fundamental form of Σt is independent
of t; therefore, the minimal surfaces Σt are all isometric. In particular, the
surface Σ−π/2 with parametrization
y = Im(z)
is isometric to Σ; this surface is called the conjugate surface of Σ.

Exercise 8.30 (Enneper’s surface). (a) In the Weierstrass-Enneper repre-

sentation (8.15), let
f (z) = 2, g(z) = z.
268 8. Minimal surfaces in E3 and A3

Show that the resulting minimal surface Σ is parametrized by


x(u, v) = t u − 13 u3 + uv 2 , −v + 13 v 3 − vu2 , u2 − v 2 ,
where z = u + iv. This surface is called Enneper’s surface.
(b) Use Maple to plot Enneper’s 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 Enneper’s
surface. (Hint: The Weierstrass-Enneper representation for Σt can be ob-
tained by taking
f (z) = 2eit , g(z) = z.)

(d) Use Maple to create an animation of plots of the family of surfaces Σt ,

with t ∈ [0, 2π] as the time parameter.
Exercise 8.31. Let Σ ⊂ E3 be the catenoid, parametrized as in Exercise
8.18.
(a) Show that the Weierstrass-Enneper representation for Σ is obtained (up
to a translation) by taking
f (z) = −ie−iz , g(z) = eiz
and that the conjugate surface of the catenoid is the helicoid. (Hint: The
formula that you find for the conjugate surface will require a slight repa-
rametrization 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 Σt
in the associated family, with t ∈ [0, 2π] as the time parameter.

8.3. Minimal surfaces in A3

Recall that in the process of constructing adapted frame fields for elliptic
surfaces Σ ⊂ A3 , we found two invariant quadratic forms that we referred to
as the first and second equi-affine fundamental forms. These were defined
in terms of the Maurer-Cartan forms (ω̄ i , ω̄ji ) associated to any 2-adapted
frame field (e1 (u), e2 (u), e3 (u)) along Σ by
I = (ω̄ 1 )2 + (ω̄ 2 )2 ,
II = ω̄31 ω̄ 1 + ω̄32 ω̄ 2 .
These quadratic forms are equi-affine analogs of the Euclidean first and sec-
ond fundamental forms for an elliptic surface Σ in E3 , but despite the appar-
ent similarities, they are quite different from their Euclidean counterparts.
For instance, if x : U → A3 is a parametrization of Σ, then the coefficients of
8.3. Minimal surfaces in A3 269

the equi-affine first fundamental form are defined in terms of second deriva-
tives of x, whereas the Euclidean first fundamental form is defined in terms
of first derivatives of x. (In fact, the equi-affine first fundamental form of a
surface Σ is a scalar multiple of the Euclidean second fundamental form of
Σ; cf. Exercise 6.40(c).) Nevertheless, we will see that an equi-affine ana-
log of Theorem 8.16 is true: If we define an area functional for equi-affine
surfaces based on the equi-affine first 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 → A3 be a smooth immersion

whose image is an elliptic surface Σ = x(U ). Let (e1 (u), e2 (u), e3 (u)) be
a 2-adapted frame field along Σ, and let (ω̄ i , ω̄ji ) be the associated Maurer-
Cartan forms.

*Exercise 8.32. Show that the 2-form

dA = ω̄ 1 ∧ ω̄ 2

is well-defined, independent of the choice of 2-adapted frame field and associ-

ated Maurer-Cartan forms. (Hint: This computation makes use of Exercises
6.20 and 6.22.) This 2-form is called the equi-affine area form of Σ.

By analogy with the Euclidean case, we define the equi-affine area functional
A on the set S of closed and bounded elliptic surfaces Σ ⊂ A3 to be
&
A(Σ) = dA.
Σ

If x : U → A3 is a parametrization of Σ and x̃ : U → A(3) is a 2-adapted

frame field along the surface Σ = x(U ), then the equi-affine area of Σ is
given by
& &
A(Σ) = ω̄ ∧ ω̄ =
1 2
x̃∗ (ω 1 ∧ ω 2 ).
U U

In order to look for critical points of the equi-affine 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 difference is that “normal” refers to the
equi-affine normal vector, which is well-defined for any 2-adapted frame field
on Σ. Thus, the equi-affine analog of Definition 8.6 is as follows:

Definition 8.33. Let U ⊂ R2 be an open set, and let x : U → A3 be

an immersion whose image is an elliptic surface Σ = x(U ). Σ is called
270 8. Minimal surfaces in E3 and A3

an equi-affine minimal surface if for every compactly supported variation

X : U × (−ε, ε) of x, we have

d 
(8.16) A(Σt ) = 0.
dt t=0

*Exercise 8.34. Convince yourself that the results of Exercise 8.4 are
equally valid in the equi-affine case; the only change is that the unit normal
vector nt (u) to the surface Σt should be replaced with the equi-affine nor-
mal vector to Σt at the point xt (u). Therefore, as in the Euclidean case, it
suffices to consider normal variations of x.

Now, let U ⊂ R2 be an open set; let x : U → A3 be an immersion whose

image is an elliptic surface Σ = x(U ); and let X : U × (−ε, ε) → A3 be a
compactly supported, normal variation of x. By analogy with the Euclidean
case, define a 2-adapted frame field on the variation X as follows: For each
(u, t) ∈ U × (−ε, ε), let (e1 (u, t), e2 (u, t), e3 (u, t)) be a 2-adapted frame for
the surface Σt at the point xt (u), so that e3 (u, t) is the equi-affine normal
to the surface Σt at the point xt (u). Let (ω̄ i , ω̄ji ) denote the pullbacks of the
Maurer-Cartan forms on A(3) to U × (−ε, ε) via X. 8

*Exercise 8.35 (Cf. Exercise 8.9). (a) For each t ∈ (−ε, ε), let ıt : U →
U × (−ε, ε) denote the inclusion map defined by ıt (u) = (u, t). Show that
the pullbacks of the 1-forms (ω̄ 1 , ω̄ 2 ) to U via ıt are the usual dual forms on
the surface Σt = xt (U ).
(b) Show that ω̄ 3 = χ dt, where the function χ(u, t) is determined by the
condition that
∂X
e3 (u, t) ω̄ 3 = dt.
∂t

(c) Show that the pullbacks of the connection forms (ω̄ji ) to U via ıt are
the usual connection forms on the surface Σt = xt (U ). In particular, the
relations
 1   
ω̄3 11 12 ω̄ 1
(8.17) = ,
ω̄32 12 22 ω̄ 2

which hold for a 2-adapted frame field on an elliptic equi-affine surface, still
hold modulo dt for the corresponding forms on U × (−ε, ε). (This means,
e.g., that the 1-form
ω̄31 − (11 ω̄ 1 + 12 ω̄ 2 )
on U × (−ε, ε) is equal to a multiple of dt.)
8.3. Minimal surfaces in A3 271

As in the Euclidean case, the equi-affine area functional of the surface Σt is

given by
&
A(Σt ) = x̃∗t (ω 1 ∧ ω 2 ),
U
and the surface Σ = Σ0 is minimal if and only if for every compactly sup-
ported normal variation of x, we have
 & 
d  d   ∗ 1 
0 =  A(Σt ) =  x̃t (ω ∧ ω 2 ) .
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-affine case.

*Exercise 8.37 (Cf. Exercise 8.13). Let x : U → A3 be an immersion, and

let X : U × (−ε, ε) → A3 be a compactly supported, normal variation of x.
(a) Use Cartan’s formula for the Lie derivative to show that
∂  
L∂/∂t (ω̄ 1 ∧ ω̄ 2 ) = d(ω̄ 1 ∧ ω̄ 2 ) .
∂t

(b) Use the Cartan structure equations for a 2-adapted coframing, the equa-
tion ω̄ 3 = χ dt, and equation (8.17) to show that
d(ω̄ 1 ∧ ω̄ 2 ) = 2Lχω̄ 1 ∧ ω̄ 2 ∧ dt,
where L = 12 (11 + 22 ) is the equi-affine mean curvature of Σ. (Note that
you only need to know that equation (8.17) holds modulo dt for this com-
putation.)
(c) Use parts (a) and (b) to show that
& &
L∂/∂t (ω̄ ∧ ω̄ ) =
1 2
2Lχ ω̄ 1 ∧ ω̄ 2 .
U U

(d) Conclude that if the equi-affine mean curvature L of Σ is identically

equal to zero, then
&
L∂/∂t (ω̄ 1 ∧ ω̄ 2 ) = 0
U
for every compactly supported, normal variation of x and hence Σ is equi-
affine minimal.

Thus, we have proved the following proposition:

Proposition 8.38. If a regular elliptic surface Σ ⊂ A3 has mean curvature

L identically equal to zero, then Σ is equi-affine minimal.
272 8. Minimal surfaces in E3 and A3

*Exercise 8.39. Adapt the argument of Exercise 8.15 to the equi-affine

case to prove the converse of Proposition 8.38: If Σ ⊂ A3 is an elliptic
surface whose equi-affine mean curvature is not identically zero, then Σ is
not a critical point for the equi-affine area functional A, and hence Σ is not
equi-affine minimal.

Together with Proposition 8.38, this proves the following equi-affine analog
of Theorem 8.16:

Theorem 8.40. A regular elliptic surface Σ ⊂ A3 is equi-affine minimal if

and only if its equi-affine mean curvature L is identically equal to zero.

Exercise 8.41. (a) Show that for any values of a, b, c with ac − b2 > 0, the
elliptic paraboloid
z = ax2 + bxy + cy 2
is equi-affinely equivalent to the paraboloid Σ ⊂ A3 defined by the equation
(8.18) z = 12 (x2 + y 2 ).

(b) Consider the parametrization x : R2 → A3 of Σ given by

 
x(u, v) = t u, v, 12 u2 + v 2 .

Show that the equi-affine frame field

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 field on Σ by computing its dual and connection forms
and showing that they satisfy the defining conditions
ω̄13 = ω̄ 1 , ω̄23 = ω̄ 2 , ω̄33 = 0
for a 2-adapted frame field.
(c) Show that the equi-affine mean curvature of Σ is identically zero. Con-
clude that any elliptic paraboloid is equi-affine 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-affine minimal surface Σ ⊂ A3 . Consider the
parametrization x : R2 → A3 of Σ 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-affine frame field on Σ
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 field are
ω̄ 1 = du, ω̄ 2 = dv
and that the only nonzero connection forms are
ω̄13 = fuu du + fuv dv,
ω̄23 = fuv du + fvv dv.
Thus, we have
   
h11 h12 fuu fuv
= .
h12 h22 fuv fvv

Assume that fuu fvv −fuv 2 > 0, so that Σ is elliptic, and for simplicity assume

that fuu , fvv > 0. In order to compute the equi-affine mean curvature of
Σ, we need to construct a 2-adapted frame field on Σ and compute the
associated Maurer-Cartan forms. Recall from Chapter 6 that any other
0-adapted frame field (ẽ1 (u), ẽ2 (u), ẽ3 (u)) on Σ has the form
⎡ ⎤
r1
 ⎢ B ⎥
(8.20) ẽ1 (u) ẽ2 (u) ẽ3 (u) = e1 (u) e2 (u) e3 (u) ⎣ r2 ⎦
0 0 (det B)−1
for some GL(2)-valued function B and real-valued functions r1 , r2 on U and
that the Maurer-Cartan forms (ω̄ ˜ i , ω̄
˜ ji ) associated to this frame field satisfy
the relations
 1  1  3  3
˜
ω̄ ω̄ ˜1
ω̄ ω̄1
(8.21) = B −1 , = (det B) tB .
˜2
ω̄ ω̄ 2 ˜ 23
ω̄ ω̄23

(b) Show that if we take

⎡ ⎤
(fuu fvv − fuv
2 )1/8 −fuv
⎢ √ √ 2 )3/8 ⎥
⎢ fuu fuu (fuu fvv − fuv ⎥
(8.22) B=⎢ ⎢ ⎥,
√ ⎥
⎣ fuu ⎦
0
(fuu fvv − fuv
2 )3/8
274 8. Minimal surfaces in E3 and A3

then    
h̃11 h̃12 10
= ,
h̃12 h̃22 01
˜ 13 = ω̄
and therefore ω̄ ˜ 1 , ω̄
˜ 23 = ω̄
˜ 2.

(c) Show that for this frame field (with r1 , r2 still arbitrary), we have
" #
(fuu r1 + fuv r2 ) (fuu fuvv − 2fuv fuuv + fvv fuuu )
˜ 33 =
ω̄ + du
(fuu fvv − fuv
2 )1/4 4(fuu fvv − fuv
2 )

" #
(fuv r1 + fvv r2 ) (fuu fvvv − 2fuv fuvv + fvv fuuv )
+ + dv.
(fuu fvv − fuv )
2 1/4 4(fuu fvv − fuv
2 )

Conclude that by choosing

⎡ ⎤
(fuu fuvv − 2fuv fuuv + fvv fuuu )
   −1 ⎢ ⎥
r1 fuu fuv ⎢ 4(fuu fvv − fuv
2 )3/4
⎥
(8.23) =− ⎢ ⎥,
⎢ ⎥
r2 fuv fvv ⎣ (fuu fvvv − 2fuv fuvv + fvv fuuv ) ⎦
4(fuu fvv − fuv
2 )3/4

we can arrange that ω̄˜ 33 = 0 and hence (together with the result of part (b))
that the resulting frame field (ẽ1 (u), ẽ2 (u), ẽ3 (u)) is 2-adapted.
(d) Show that the dual forms associated to this frame field are
√
fuu fuv
˜ 1
ω̄ = du + √ dv,
(fuu fvv − fuv )
2 1/8 fuu (fuu fvv − fuv
2 )1/8

(fuu fvv − fuv

2 )3/8
˜2 =
ω̄ √ dv.
fuu
˜ 31 and ω̄
Then compute ω̄ ˜ 32 , and find the functions (ij ) such that
 1    1
˜3
ω̄ 11 12 ω̄˜
= .
˜ 32
ω̄ 12 22 ω̄˜2
Finally, write the equi-affine mean curvature equation
1
L= 2 (11 + 22 ) = 0
as a (rather nasty!) fourth-order differential equation for f .

8.3.2. A Weierstrass-Enneper-type representation for elliptic equi-

affine minimal surfaces in A3 . In this section, we will derive a Weier-
strass-Enneper-type representation for elliptic equi-affine minimal surfaces.
This formula is originally due to Blaschke and may be found in [Bla85].
8.3. Minimal surfaces in A3 275

Let Σ ⊂ A3 be an elliptic surface with parametrization x : U → A3 , and

let (e1 (u), e2 (u), e3 (u)) be a 2-adapted frame field along Σ, so that the
associated Maurer-Cartan forms (ω̄ i , ω̄ji ) satisfy

ω̄13 = ω̄ 1 , ω̄23 = ω̄ 2 , ω̄33 = 0.

Recall that for such a frame field, we have
 1   
ω̄3 11 12 ω̄ 1
= ,
ω̄32 12 22 ω̄ 2
⎡ ⎤ ⎡ ⎤
2ω̄11 h1 −h2  
⎢ 1 ⎥ ⎢ ⎥ ω̄ 1
⎢ω̄ + ω̄ 2 ⎥ = ⎢−h2 −h1 ⎥
⎣ 2 1⎦ ⎣ ⎦ 2 ,
ω̄
2ω̄22 −h1 h2
where we have set
h1 = h111 = −h122 , h2 = h222 = −h112 ,
and that Σ is equi-affine minimal if and only if 11 + 22 = 0.
Now, let A3C denote the complexified equi-affine space A3 ⊗ C. (This is
simply the vector space C3 , but with an equi-affine structure rather than a
Euclidean structure.) Consider the Λ2 A3C -valued 1-form (cf. Definition 2.11)
ξ = 12 e3 ∧ (e1 − ie2 )(ω̄ 1 + iω̄ 2 )
= 12 e3 ∧ [(e1 ω̄ 1 + e2 ω̄ 2 ) + i(e1 ω̄ 2 − e2 ω̄ 1 )]
on U .
*Exercise 8.43. (a) Show that ξ is a well-defined 1-form on U , independent
of the choice of 2-adapted frame field (e1 (u), e2 (u), e3 (u)).
(b) Show that
dξ = 12 (11 + 22 )(e1 ∧ e2 ) ω̄ 1 ∧ ω̄ 2 .
Therefore, dξ = 0 if and only if Σ is equi-affine minimal.

Now, suppose that Σ is equi-affine minimal. Let (e1 (u), e2 (u), e3 (u)) be any
2-adapted frame field on Σ, with associated Maurer-Cartan forms (ω̄ i , ω̄ji ),
and consider the scalar-valued, complex 1-form ω̄ 1 + iω̄ 2 on U .
*Exercise 8.44. Convince yourself that the results of Exercise 8.22 hold in
the equi-affine case. Thus, every point u ∈ U has a neighborhood V ⊂ U
on which there exist complex-valued functions z, ϕ : V → C such that
ω̄ 1 + iω̄ 2 = ϕ dz.
276 8. Minimal surfaces in E3 and A3

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 → A3 of Σ to V ⊂ U in the form x(z), thereby
defining a complex structure on Σ.

Let F : V → Λ2 A3C be the function

F(z, z ∗ ) = 12 e3 ∧ (e1 − ie2 )ϕ,
so that
ξ = F(z, z ∗ ) dz.
*Exercise 8.45. (a) Show that the assumption that Σ is equi-affine mini-
mal, and hence dξ = 0, is equivalent to
∂F
= 0.
∂z ∗
Therefore, F(z, z ∗ ) = F(z) is a holomorphic, Λ2 A3C -valued function on V .
For the remainder of this exercise, assume that this condition holds.
(b) Apply the Poincaré lemma (cf. Theorem 2.31) to the closed 1-form ξ =
F(z) dz to conclude that there exists a Λ2 A3C -valued holomorphic function
Z : V → Λ2 A3C such that
ξ = dZ,
and therefore, F(z) = Z (z).
For ease of notation, let
e = 12 (e1 − ie2 ), ω̄ = ω̄ 1 + iω̄ 2 ,
so that
ξ = dZ = e3 ∧ e ω̄.
Then the complex conjugate of ξ is
ξ ∗ = dZ∗ = e3 ∧ e∗ ω̄ ∗ .

(c) Show that

d(e ∧ e∗ ) = 12 (ξ ∗ − ξ) = 12 (dZ∗ − dZ).
Conclude that
Z∗ − Z = 2e ∧ e∗ + 2iC
for some real-valued constant C ∈ Λ2 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 effect on the condition that ξ = 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 different from the usual cross product on
R3 , in that it operates on a pair of elements of Λ2 A3 and produces an element
of A3 .
Definition 8.46. The special linear cross product is the unique skew-sym-
metric, bilinear map
×sl : Λ2 A3 × Λ2 A3 → A3
that is SL(3)-equivariant and satisfies
(8.24) (e1 ∧ e2 ) ×sl (e1 ∧ e3 ) = e1
for any unimodular basis (e1 , e2 , e3 ) of A3 .

A few comments on this definition are in order:

(1) “SL(3)-equivariant” means that for any matrix A ∈ SL(3), we have

(Ae1 ∧ Ae2 ) ×sl (Ae1 ∧ 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 ∈ 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 ele-
ments of the complexified space Λ2 A3C .

For a geometric interpretation of this cross product, think of an element

v1 ∧ v2 ∈ Λ2 A3 as representing the oriented plane spanned by the vectors
(v1 , v2 ) in A3 . The cross product of two such planes (v1 ∧ v2 , w1 ∧ 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 ∧ v2 ) ×sl (w1 ∧ w2 ) = (v1 × v2 ) × (w1 × w2 ).
(Hint: Since both sides are skew-symmetric and bilinear, it suffices 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.)
278 8. Minimal surfaces in E3 and A3

We are now ready to derive Blaschke’s formula.

*Exercise 8.48. (a) Use the results of Exercise 8.45 to show that

(Z∗ − Z) ×sl d(Z∗ + Z) = −i(e ω̄ + e∗ ω̄ ∗ ) = −i dx,

where x : V → A3 is the given parametrization of Σ.

(b) Conclude that

dx = i[(Z∗ − Z) ×sl d(Z∗ + Z)]

= i[Z∗ ×sl dZ∗ − Z ×sl dZ + d(Z∗ ×sl Z)].

Therefore, up to translation, the parametrization x : V → A3 of the Σ is

given in terms of the complex coordinate z on V by
(8.25) " & z #
∗
 ∗  ∗ 

x(z) = i Z(z) ×sl Z(z) + Z(ζ) ×sl Z (ζ) − Z(ζ) ×sl Z (ζ) dζ
z0

for any choice of base point z0 ∈ V . This formula is called the Blaschke
representation for Σ.

Therefore, for any equi-affine minimal surface x : U → A3 and any point

u ∈ U , there exists a neighborhood V ⊂ 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 → Λ2 A3C .

*Exercise 8.49. Write

Z = 12 (X + iY)
for some smooth functions X, Y : V → Λ2 A3R . Recall that the functions
X, Y must satisfy the Cauchy-Riemann equations

(8.26) Xu = Yv , Xv = −Yu ,

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 ×sl Yv ) du − (Y ×sl 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 Λ2 A3R for each z ∈
V.
8.3. Minimal surfaces in A3 279

Conversely, let V ⊂ C be an open set, and let Z = 12 (X + iY) : V → Λ2 A3C

be a holomorphic function such that the vectors (Y(z), Yu (z), Yv (z)) are
linearly independent for each z ∈ V . Define an immersion x : V → A3 by
(8.25).

*Exercise 8.50. Let λ : V → R be a smooth function, and consider the

frame field along Σ = x(V ) defined by
e1 (u) = λxu = λ(Y ×sl Yv ),
(8.28) e2 (u) = ±λxv = ∓λ(Y ×sl Yu ),
e3 (u) = λ2 (Yu ×sl 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 λ (up to sign) for
which (e1 (u), e2 (u), e3 (u)) is a unimodular frame field. For the remainder
of this exercise, assume that λ has been chosen accordingly.
(b) Show that the dual forms associated to the frame field (8.28) are

ω̄ 1 = λ−1 du, ω̄ 2 = ±λ−1 dv,

with the sign of ω̄ 2 chosen according to the sign of e2 (u).

In order to compute the connection forms associated to the frame field (8.28),
observe that, since (Y, Yu , Yv ) are linearly independent, the second deriva-
tives of Y can be written as

Yuu = h0 Y + h1 Yu + h2 Yv ,
Yuv = k0 Y + k1 Yu + k2 Yv ,
Yvv = −h0 Y − h1 Yu − h2 Yv

for some functions hi , ki : V → R.

(c) Differentiate equations (8.28) and show that the connection forms asso-
ciated to the frame field (8.28) satisfy the conditions

ω̄13 = ω̄ 1 , ω̄23 = ω̄ 2 , ω̄33 = 0.

Conclude that the frame field (8.28) is 2-adapted. (Hint: For the last con-
dition, first show that

ω̄11 + ω̄22 = ω̄33 = 2λ−1 dλ + λ(h1 + k2 )ω̄ 1 ∓ λ(h2 + k1 )ω̄ 2 .

Then use the assumption that the frame field is unimodular to conclude that
ω̄33 = 0.)
280 8. Minimal surfaces in E3 and A3

(d) Show that

    
ω̄31 h0 k 0 ω̄ 1
2
=λ .
ω̄32 k0 −h0 ω̄ 2
Conclude that Σ is equi-affine minimal.

Together, Exercises 8.48 and 8.50 imply the following proposition:

Proposition 8.51. A regular elliptic surface Σ ⊂ A3 is equi-affine minimal
if and only if every point of Σ has a neighborhood that can be parametrized
by a smooth immersion x : V → A3 of the form (8.25) for some holomorphic
function Z : V → Λ2 A3C .
*Exercise 8.52. Let Σ be the elliptic paraboloid (8.18) of Exercise 8.41,
with the 2-adapted frame field (8.19).
Show that the Blaschke representation for Σ may be obtained by taking
Z(z) = 1
2 (−ie1 ∧ e2 + iz e2 ∧ e3 + z e3 ∧ 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 ∧ e2 ) × (e1 ∧ e3 ) = e1 ,
(e2 ∧ e3 ) × (e2 ∧ e1 ) = e2 ,
(e3 ∧ e1 ) × (e3 ∧ e2 ) = e3 .)

8.4. Maple computations

As usual, begin by loading the Cartan and LinearAlgebra packages into

Maple.
Exercise 8.18 (e): Define the parametrization for Σ and the orthonormal
frame field (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 (ω̄ 1 , ω̄ 2 ) basis as necessary (but first we need to declare
the 1-forms (ω̄ 1 , ω̄ 2 )):
> 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 coefficients of (e1 (u), e2 (u)) will be the 1-forms ω̄31 = −ω̄13 , ω̄32 = −ω̄23 ,
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
ω1 ρv,v ω2 1 + ρ2v
ω3,1 = − , ω3,2 =
1 + ρ2v ρ 1 + 2 ρ2v + ρ4v

> assign(%);
We can read off the (hij ) directly from ω̄13 and ω̄23 :
> 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, Σ is minimal if and only if H = 12 (h11 + h22 ) = 0:
> minsurfeq:= numer(simplify(Trace(h)));

minsurf eq := −ρ2v − 1 + ρv,v ρ

Now, Maple can solve this equation with the dsolve command. Unfor-
tunately, Maple can be rather clumsy about simplifying exponentials and
282 8. Minimal surfaces in E3 and A3

trigonometric functions, so it requires a bit of manipulation to get the solu-

tion into a nice form.
> soln:= dsolve(minsurfeq, rho(v));
⎛ ⎞
1 ⎜ 1 ⎟ v C2
soln := ρ = C1 ⎝ !2 !2 + 1⎠ e C1 e C1 ,
2 v C2
e C1 e C1

" !2 !2 #
v C2
C1 e C1 e C1 +1
1
f= v C2
2 e C1 e C1

Maple apparently doesn’t 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: Define the Weierstrass parametrization associated to func-

tions f (z), g(z):
> PDETools[declare](f(z), g(z));
> dZ:= Vector([(1/2)*f(z)*(1 - g(z)ˆ2),
(I/2)*f(z)*(1 + g(z)ˆ2), 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 → 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, let’s remove the assumptions on (u, v), just
because the trailing tildes in the output are distracting:
> unassign(’u’, ’v’);
Define the parametrization for Σ and the unimodular frame field (e1 (u),
e2 (u), e3 (u)):
> X:= Vector([u, v, (1/2)*(uˆ2 + vˆ2)]);
> e1:= map(diff, X, u);
e2:= map(diff, X, v);
e3:= Vector([0,0,1]);
284 8. Minimal surfaces in E3 and A3

Since we have e1 (u) = xu (u), e2 (u) = xv (u), it follows that the associated
Maurer-Cartan forms ω̄ 1 , ω̄ 2 are equal to

ω̄ 1 = du, ω̄ 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 (ω̄ji ), differentiate the frame fields:
> 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 off that

de1 = e3 ω̄ 1 , de2 = e3 ω̄ 2 , de3 = 0.

Therefore, we have
ω̄13 = ω̄ 1 , ω̄23 = ω̄ 2 ,

and the remaining connection forms (ω̄ji ) are equal to zero. It follows that
this frame field is 2-adapted. Moreover, since ω̄31 = ω̄32 = 0, Σ is an equi-
affine minimal surface.
Exercise 8.42: Define the parametrization for Σ and the 0-adapted frame
field of part (a) (we’ll call this frame field (e10, e20, e30) because it will
be refined 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 field we have ω̄ 1 = du, ω̄ 2 = dv:
> framesub0:= [omega[1] = d(u), omega[2] = d(v)];
> framebacksub0:= makebacksub(framesub0);
In order to compute the connection forms (ω̄ji ), differentiate the frame fields:
> 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

ω̄13 = fuu ω̄ 1 + fuv ω̄ 2 , ω̄23 = fuv ω̄ 1 + fvv ω̄ 2 ,

and the remaining connection forms (ω̄ji ) are equal to zero.

Now we need to find a matrix-valued function B and real-valued functions
r1 , r2 on U that will make the frame field (8.20) 2-adapted. Rather than
simply verifying that the expressions given in equations (8.22) and (8.23)
will do the trick, let’s see if we can figure out where they came from. First,
look for a matrix-valued function B such that the transformed Maurer-
Cartan forms in equation (8.21) will satisfy the 1-adapted condition ω̄˜ 13 = ω̄
˜ 1,
˜ 23 = ω̄
ω̄ ˜ 2 . We can make this problem slightly simpler by observing that, since
B is only determined up to multiplication by a rotation matrix, we should be
able to find a suitable matrix B whose lower left-hand entry is zero. With
this assumption, define the new frame field 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 coefficients of the 1-
forms in their difference 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)ˆ2)ˆ(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)
286 8. Minimal surfaces in E3 and A3

- diff(f(u,v),u,v)ˆ2)ˆ(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)ˆ2)ˆ(3/8);
Set up a substitution for the Maurer-Cartan forms for this new frame field:
> 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 dẽ3 so that we can
˜ 33 and set it equal to zero. We can do this fairly compactly as follows:
find ω̄
If we let A be the matrix

A = ẽ1 (u) ẽ2 (u) ẽ3 (u) ,
then we have ⎡ ⎤
˜ 31
ω̄
⎢ ˜ 2⎥
dẽ3 = A ⎣ω̄ 3⎦ .
˜ 33
ω̄
Therefore, ω̄˜ 33 is the last entry of the vector A−1 dẽ3 , and we can solve for
r1 , r2 by setting the scalar coefficients of ω̄˜ 33 equal to zero.

> 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).
Finally, the equi-affine mean curvature L of Σ can be computed from ω̄ ˜ 31 and
2
˜ 3 , which are the first two entries of connformsvec2. First, we need to use
ω̄
the substitution to express these forms as linear combinations of (ω̄˜ 1 , ω̄
˜ 2)
and then compute the trace of the associated coefficient 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, aren’t you glad that you didn’t have to compute L by hand?
 Chapter 9

Pseudospherical
surfaces and
Bäcklund’s theorem

9.1. Introduction

In this chapter, we will show how moving frames may be used to prove
Bäcklund’s theorem. Bäcklund’s theorem concerns surfaces of constant neg-
ative Gauss curvature, also known as pseudospherical surfaces. The best-
known pseudospherical surface is, of course, the pseudosphere, which is the
surface of revolution obtained by revolving the tractrix

α(t) = t[t − tanh t, sech t, 0]

about the x-axis. But there are infinitely many other pseudospherical sur-
faces as well.
Bäcklund’s theorem is based on a beautiful geometric construction that
starts with a given pseudospherical surface and produces from it a 2-pa-
rameter family of new pseudospherical surfaces. (See [RS82] for a discus-
sion of Bäcklund’s original construction.) The construction can be iterated,
thereby producing an arbitrary number of increasingly complicated fami-
lies of pseudospherical surfaces from a single starting surface. For example,
we might take the pseudosphere as our initial surface and use Bäcklund’s
construction to generate new families of surfaces.

287
288 9. Pseudospherical surfaces and Bäcklund’s theorem

In §9.4, we will see how pseudospherical surfaces are intimately connected

with solutions φ(x, y) of the partial differential equation known as the sine-
Gordon equation:
(9.1) φxy = sin(φ).
This is a nonlinear partial differential equation, and it is one of a number of
nonlinear PDEs known as “integrable systems”. PDEs in this class share a
number of important features, including special families of solutions known
as “soliton” solutions. As we will see, Bäcklund’s geometric construction for
pseudospherical surfaces gives rise to a corresponding transformation be-
tween solutions of equation (9.1). This transformation is called a Bäcklund
transformation, and, just as Bäcklund’s construction generates new pseu-
dospherical surfaces from a known pseudospherical surface, the Bäcklund
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 φ(x, y) = 0, produces nontrivial new
solutions; in fact, the soliton solutions of (9.1) can be constructed in pre-
cisely this way.

9.2. Line congruences

Bäcklund’s construction begins with the notion of a line congruence. Rough-

ly, a line congruence in E3 is simply a 2-parameter family of lines in E3 . A
more formal definition requires that the set of lines in E3 be given a topolog-
ical structure so that it becomes a manifold, called the affine 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 × S2 . Now define an
equivalence relation ∼ on E3 × S2 by the condition that (x, e) ∼ (y, f ) if and
only if the pairs (x, e) and (y, f ) determine the same line.
Exercise 9.1. Show that (x, e) ∼ (y, f ) if and only if f = ±e and y = x+te
for some t ∈ R.

The affine Grassmannian G1 (E3 ) is then defined to be the set of equivalence

classes  
G1 (E3 ) = E3 × S2 / ∼,
with the line  ∈ 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 definition in hand, the formal definition of a
line congruence is as follows:
Definition 9.2. A line congruence in E3 is an immersed surface in G1 (E3 ).
9.3. Bäcklund’s 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 → E3 be an
immersion whose image is a regular surface Σ ⊂ E3 . For each u ∈ 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 ∈ U }
is a line congruence in E3 . A line congruence of this type is called a normal
congruence.

Given an open set U ⊂ R2 and a line congruence  : U → G1 (E3 ), we can

express the congruence—in infinitely many different ways—as
(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
Σ = x(U ) is called a surface of reference for the line congruence. Any other
8 = x̃(U ) for the congruence can then be parametrized
surface of reference Σ
as
x̃(u) = x(u) + λ(u)e(u)
for some smooth, real-valued function λ on U .
For a generic line congruence  : U → G1 (E3 ), there are two distinguished
surfaces of reference Σ = x(U ), Σ 8 = x̃(U ), called focal surfaces of the
congruence. The definition 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
parametrized as above, so that for each u ∈ U , the points x(u), x̃(u) are
contained in the line (u) and the line (u) is tangent to the surfaces Σ, Σ 8
at the points x(u), x̃(u), respectively.

9.3. Bäcklund’s theorem

Bäcklund’s theorem concerns a special category of line congruences known

as pseudospherical congruences.
Definition 9.4. Let U ⊂ R2 , and let  : U → G1 (E3 ) be a line congruence
8 parametrized as above. The congruence is
in E3 with focal surfaces Σ, Σ,
called pseudospherical if the following two conditions hold:

(1) The distance r = |x̃(u) − x(u)| is a constant, independent of u.

290 9. Pseudospherical surfaces and Bäcklund’s theorem

(2) The angle α (assumed to be nonzero) between the surface normal

8 at the points x(u), x̃(u),
vectors e3 (u), ẽ3 (u) to the surfaces Σ, Σ
respectively, is a constant, independent of u.

Theorem 9.5 (Bäcklund).

(1) Let U ⊂ R2 , and suppose that  : U → G1 (E3 ) is a pseudospherical

8 Then both Σ and Σ
line congruence in E3 with focal surfaces Σ, Σ. 8
sin2 (α)
have constant negative Gauss curvature K = − r2 , where r, α
are as in Definition 9.4.
(2) If Σ ⊂ E3 is any surface of constant negative Gauss curvature
2
K = − sinr2(α) , then given any point x0 ∈ Σ and any unit tangent
vector e0 ∈ Tx0 Σ that is not a principal direction at x0 , there exists
a unique surface Σ 8 ⊂ E3 and a pseudospherical line congruence with
focal surfaces Σ, Σ 8 such that if x̃0 is the point in Σ 8 corresponding
to x0 , then x̃0 − x0 = re0 and the angle between the surface normal
vectors to Σ, Σ8 at the points x0 , x̃0 , respectively, is α.

The construction in part (2) of Theorem 9.5 is due to Bianchi and Bäcklund
([B8̈3], [Bia79]) and is called a Bäcklund transformation; the terminology
8 is a “transformation” of the original
refers to the idea that the new surface Σ
surface Σ.
Bäcklund’s theorem can be proved using local coordinates on the surfaces
8 but the computations are rather ugly. The proof can be greatly sim-
Σ, Σ,
plified by using the method of moving frames because frame fields can be
adapted to the geometry of the problem in a way that local coordinates can-
not. 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
guide our choices of orthonormal frame fields on the surfaces Σ, Σ.8 (This
proof is taken from [CT80].)

8 respectively,
Proof. (1) Let x, x̃ : U → E3 be the parametrizations of Σ, Σ,
induced from the line congruence  : U → G1 (E ) as above. We can choose
3

orthonormal frame fields (e1 (u), e2 (u), e3 (u)) along Σ and (ẽ1 (u), ẽ2 (u),
8 such that:
ẽ3 (u)) along Σ

(1) e3 (u) is the unit normal vector to Σ at x(u) and ẽ3 (u) is the unit
normal vector to Σ 8 at x̃(u); therefore, (e (u), e (u)) span the tan-
1 2
8
gent space Tx(u) Σ and (ẽ1 (u), ẽ2 (u)) span the tangent space Tx̃(u) Σ.
9.3. Bäcklund’s theorem 291

(2) e1 (u) = ẽ1 (u) is the common unit tangent vector to both surfaces
in the direction of x̃(u) − x(u).
Remark 9.6. The notation ei (u) is intended to distinguish these orthonor-
mal frame vectors from the principal orthonormal frame vectors ei (u) that
will be introduced in §9.4. This notational distinction will become impor-
tant in §9.5, when we need to consider both of these orthonormal frame
fields simultaneously! Similarly, we will denote the Maurer-Cartan forms
associated to the frame field (e1 (u), e2 (u), e3 (u)) by (ω̄ i , ω̄ ij ), and those
˜ i , ω̄
associated to the frame field (ẽ1 (u), ẽ2 (u), ẽ3 (u)) by (ω̄ ˜ ji ).

*Exercise 9.7. Show that it follows from these conditions (and the defini-
tion of α as the angle between e3 (u), ẽ3 (u)) that
ẽ1 (u) = e1 (u),
(9.2) ẽ2 (u) = cos(α)e2 (u) + sin(α)e3 (u),
ẽ3 (u) = − sin(α)e2 (u) + cos(α)e3 (u)
and that
(9.3) x̃(u) = x(u) + re1 (u).

Let (ω̄ i , ω̄ ij ) denote the pullbacks of the Maurer-Cartan forms (ω i , ωji ) on

˜ i , ω̄
E(3) to U via the frame field (x(u); e1 (u), e2 (u), e3 (u)) for Σ, and let (ω̄ ˜ ji )
i i
denote the pullbacks of (ω , ωj ) to U via the frame field (x̃(u); ẽ1 (u), ẽ2 (u),
ẽ3 (u)) for Σ. 8

*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) ˜ 1 + ẽ2 ω̄
ẽ1 ω̄ ˜ 2 = e1 ω̄ 1 + e2 ω̄ 2 + r(e2 ω̄ 21 + e3 ω̄ 31 ).
Then substitute the expressions (9.2) for (ẽ1 (u), ẽ2 (u)) into equation (9.4)
to obtain
˜ 1 + e2 cos(α) ω̄
e1 ω̄ ˜ 2 + e3 sin(α) ω̄
˜ 2 = e1 ω̄ 1 + e2 (ω̄ 2 + r ω̄ 21 ) + e3 (r ω̄ 31 ).
Conclude that we have the following relationships between the Maurer-
8
Cartan forms on Σ and those on Σ:
˜ 1 = ω̄ 1 ,
ω̄
(9.5) ˜ 2 = ω̄ 2 + r ω̄ 21 ,
cos(α) ω̄
˜ 2 = r ω̄ 31 .
sin(α) ω̄

(b) Show that the last two equations in (9.5) imply that
(9.6) ω̄ 2 + r ω̄ 21 = r cot(α) ω̄ 31 .
292 9. Pseudospherical surfaces and Bäcklund’s theorem

(c) Use the fact that ω̄ ij = dej , ei  (and similarly for ω̄

˜ ji ), the expressions
(9.2), and equation (9.6) to show that
sin(α) 2
(9.7) ˜ 13 =
ω̄ ω̄ , ˜ 23 = ω̄ 32 .
ω̄
r

Next, recall that the coefficients (hij ) of the second fundamental form of Σ
are defined by the equations
ω̄ 31 = h11 ω̄ 1 + h12 ω̄ 2 ,
(9.8)
ω̄ 32 = h12 ω̄ 1 + h22 ω̄ 2 .
Because we have not made any attempt to arrange for e1 (u) and e2 (u) to
be principal vector fields, 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 first and third equations in (9.5), the first equation
˜ 1 , ω̄
in (9.8), and the fact that (ω̄ ˜ 2 ) are linearly independent 1-forms on U to
conclude that h12 = 0.

8 Recall that
Finally, consider the Gauss curvature K̃ of Σ.
(9.9) ˜ 13 ∧ ω̄
ω̄ ˜ 1 ∧ ω̄
˜ 23 = K̃ ω̄ ˜2
(cf. Exericse 4.47).
*Exercise 9.10. (a) Use equations (9.7) and (9.8) to show that
sin(α)
˜ 13 ∧ ω̄
ω̄ ˜ 23 = − h12 ω̄ 1 ∧ ω̄ 2 .
r
(b) Use equation (9.9), the first and third equations of (9.5), and (9.8) to
show that
r
˜ 13 ∧ ω̄
ω̄ ˜ 23 = K̃ h12 ω̄ 1 ∧ ω̄ 2 .
sin(α)
(c) Conclude from parts (a) and (b) and the fact that h12 = 0 that
sin2 (α)
K̃ = − .
r2
2
An analogous argument shows that K = − sinr2(α) as well.
The proof of part (2) of Theorem 9.5 involves concepts from the theory of
exterior differential systems similar to those needed for the proof of Lemma
4.12. The key step involves constructing an adapted orthonormal frame
field (e1 (u), e2 (u), e3 (u)) along Σ with the property that a parametrization
x̃ : U → E3 for the desired surface Σ 8 will be given by
(9.10) x̃(u) = x(u) + re1 (u).
9.4. Pseudospherical surfaces and the sine-Gordon equation 293

The Maurer-Cartan forms for such a frame field must satisfy equation (9.6);
in fact (as we will see in §9.5), equation (9.6) is equivalent to an overde-
termined system of partial differential equations for the frame field (e1 (u),
e2 (u), e3 (u)), and the compatibility condition for this system is precisely the
2
condition that Σ has Gauss curvature K = − sinr2(α) . The initial condition
e1 (u0 ) = e0 ∈ Tx0 Σ then determines the desired frame field uniquely, and
the remainder of the proof consists of showing that the surface Σ 8 defined by
(9.10) satisfies 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
Σ.
follows: One parameter comes from the choice of constants r, α such that
2
K = − sinr2(α) , and one comes from the choice of a non-principal unit vector
e0 ∈ Tx0 Σ. The choice of r, α determines the coefficients 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 Σ be a pseudospherical surface, and for simplicity assume that its Gauss
curvature is K = −1. Since the Gauss curvature of Σ is negative, Σ cannot
have any umbilic points; consequently, it can be shown (for a proof, see
[dC76]) that every point x ∈ Σ has a neighborhood for which there exists a
local parametrization x : U → E3 of Σ whose coordinate curves are principal
curves in Σ (cf. Exercises 4.24 and 4.27). As in Exercise 4.24, we can then
choose the adapted orthonormal frame field
1 1
e1 (u) = √ xu , e2 (u) = √ xv , e3 (u) = e1 (u) × e2 (u)
E G
along Σ. (Note that, unlike in §9.3, we are only considering a single pseu-
dospherical surface, so there is no line congruence to take into consideration
when choosing an adapted frame field.) Then, as we saw in Exercises 4.24
and 4.27, the coefficients of the first and second fundamental forms satisfy
F = f = 0, and the associated Maurer-Cartan forms (ω̄ i , ω̄ji ) are given by
√ √
ω̄ 1 = E du, ω̄ 2 = G dv,
e √ g √
ω̄13 = √ du = κ1 E du, ω̄23 = √ dv = κ2 G dv,
(9.11) E G
1
ω̄21 = √ (Ev du − Gu dv),
2 EG
where κ1 = Ee , κ2 = Gg are the principal curvatures of Σ.
294 9. Pseudospherical surfaces and Bäcklund’s theorem

The following exercises will show how the surface Σ = x(U ) gives rise to a
solution of the sine-Gordon equation (9.1). First, we investigate the Gauss
and Codazzi equations for Σ and show that the parametrization x : U → E3
can be fine-tuned to arrange that the first and second fundamental forms of
Σ can be expressed nicely in terms of a single function ψ : U → R.
*Exercise 9.11. (a) Show that the Codazzi equations of Exercise 4.41(f)
can be written in the form
∂κ1 ∂ √ ! ∂κ2 ∂ √ !
(9.12) = (κ2 − κ1 ) ln( E) , = (κ1 − κ2 ) ln( G) .
∂v ∂v ∂u ∂u

(b) Divide equations (9.12) by (κ1 − κ2 ), multiply the left-hand sides by κκ11
and κκ22 , respectively, and use the Gauss equation κ1 κ2 = −1 to show that
∂   ∂ √ !
ln(κ21 + 1) = −2 ln( E) ,
∂v ∂v
(9.13)
∂   ∂ √ !
ln(κ22 + 1) = −2 ln( G) .
∂u ∂u

(c) Integrate equations (9.13) and conclude that

c1 (u) c2 (v)
(9.14) κ21 + 1 = , κ22 + 1 =
E G
for some functions c1 (u), c2 (v) > 0.
(d) Show that under a change of coordinates of the form
(9.15) ũ = h(u), ṽ = k(v),
the coefficients of the first fundamental form with respect to the coordinates
(ũ, ṽ) become
1 1
Ẽ =  2
E, G̃ =  G.
(h (u)) (k (v))2
(The functions κ1 , κ2 , 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
c̃1 (ũ) = c̃2 (ṽ) = 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 ψ(u, v) such that
κ1 = tan(ψ), κ2 = − cot(ψ),
2
E = cos (ψ), G = sin2 (ψ).
9.4. Pseudospherical surfaces and the sine-Gordon equation 295

Thus, the first and second fundamental forms of Σ are

I = cos2 (ψ) du2 + sin2 (ψ) dv 2 ,
II = sin(ψ) cos(ψ) (du2 − dv 2 ).
(Hint: Recall that κ1 κ2 = −1 and that the tangent and cotangent functions
are surjective onto R.)

Next, we give a geometric interpretation of the function ψ and show that

the function φ = 2ψ 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 2ψ(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 ± xv .)
(b) Show that ω̄21 (cf. equation (9.11)) is given by
ω̄21 = −ψv du − ψu dv.

(c) Use the Gauss equation

dω̄21 = K ω̄ 1 ∧ ω̄ 2
to show that the function ψ satisfes the PDE
(9.16) ψuu − ψvv = sin(ψ) cos(ψ).

(d) Let φ = 2ψ, so that φ(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) φuu − φvv = sin(φ)
for the function φ.
(e) Consider the change of coordinates
1 1
x = (u + v), y = (u − v).
2 2
Show that in terms of the (x, y)-coordinates, the first and second fundamen-
tal forms of Σ are given by
I = dx2 + 2 cos(2ψ) dx dy + dy 2 ,
(9.18)
II = 2 sin(2ψ) dx dy.
Note that the x- and y-coordinate directions are now the asymptotic direc-
tions at each point of Σ; for this reason, (x, y) are called asymptotic coordi-
nates on Σ, and the corresponding parametrization is called an asymptotic
parametrization of Σ.
(f) Show that equation (9.17) is equivalent to equation (9.1).
296 9. Pseudospherical surfaces and Bäcklund’s theorem

Remark 9.13. Equations (9.17) and (9.1) are both referred to as the sine-
Gordon 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
φuu − φvv = 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 “characteristic” comes from the
fact that the x- and y-coordinate curves are the characteristic curves for the
PDE (9.1), while the term “null” 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 Σ deter-

mines a solution φ of the sine-Gordon equation (9.1). The following exercise
shows that the converse is true as well.

*Exercise 9.14. Let φ : U → R be any solution of the sine-Gordon equation

(9.1), and let ψ(x, y) = 12 φ(x, y). Define 1-forms (ω̄ 1 , ω̄ 2 , ω̄13 , ω̄23 , ω̄21 ) by

ω̄ 1 = cos(ψ)(dx + dy), ω̄ 2 = sin(ψ)(dx − dy),

(9.19) ω̄13 = sin(ψ)(dx + dy), ω̄23 = − cos(ψ)(dx − dy),
ω̄21 = −ψx dx + ψy dy.

(a) Check that the first and second fundamental forms

I = (ω̄ 1 )2 + (ω̄ 2 )2 ,
II = ω̄13 ω̄ 1 + ω̄23 ω̄ 2

agree with those in equations (9.18).

(b) Check that these 1-forms (together with ω̄ 3 = 0) satisfy the Cartan
structure equations (3.8). (Hint: This result depends on the fact that φ is
a solution of equation (9.1).)
(c) Conclude from Bonnet’s theorem (cf. Theorem 4.39) that there exists an
immersed surface x : U → E3 whose first and second fundamental forms are
given by (9.18). In particular, the Gauss curvature of the surface Σ = x(U )
is K = −1, and the angle between the asymptotic directions at the point
x(x, y) is equal to φ(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. Bäcklund transformation for the sine-Gordon equation 297

provides an important tool for studying the space of solutions to these equa-
tions. This connection was first considered by Chern and Tenenblat [CT86]
and further explored by Reyes [Rey98] and others.

9.5. The Bäcklund transformation for the sine-Gordon

equation

In this section, we will see how the geometric Bäcklund transformation be-
tween pseudospherical surfaces gives rise to a corresponding analytic trans-
formation between solutions of the sine-Gordon equation (9.1).
Suppose that we have a Bäcklund transformation between two pseudospher-
ical surfaces Σ, Σ8 of Gauss curvature K = −1. (Note that the condition
K = −1 implies that r = sin(α).) Let (e1 (u), e2 (u), e3 (u)) be the orthonor-
mal frame field on Σ adapted to the Bäcklund transformation as in §9.3,
and let (e1 (u), e2 (u), e3 (u)) be the principal adapted frame field on Σ as in
§9.4.
Let η(u) denote the angle between e1 (u) and e1 (u). The following exercise
shows how the function η is related to the function ψ of §9.4.

*Exercise 9.16. (a) Show that

 
  cos(η(u)) −sin(η(u))
(9.20) e1 (u) e2 (u) = e1 (u) e2 (u) .
sin(η(u)) cos(η(u))

(b) Show that

 1      
ω̄ cos(η) sin(η) ω̄ 1 cos(ψ − η) cos(ψ + η) dx
= = ,
ω̄ 2 −sin(η) cos(η) ω̄ 2 sin(ψ − η) −sin(ψ + η) dy
     3   
(9.21) ω̄ 31 cos(η) sin(η) ω̄1 sin(ψ − η) sin(ψ + η) dx
= = ,
ω̄ 32 −sin(η) cos(η) ω̄23 −cos(ψ − η) cos(ψ + η) dy

ω̄ 12 = ω̄21 − dη = −(ψx + ηx ) dx + (ψy − ηy ) dy.

(Hint: Cf. Exercises 4.28 and 4.47 and equations (9.19).)
(c) Show that the Bäcklund equation (9.6) is equivalent to the first-order
system of partial differential equations
ψx + ηx = λ sin(ψ − η),
(9.22) 1
ψy − ηy = sin(ψ + η),
λ
298 9. Pseudospherical surfaces and Bäcklund’s theorem

where λ = cot(α) − csc(α). (Hint: Recall that, since K = −1, we have

r = sin(α), and you will need the trigonometric identity
1
−(cot(α) + csc(α)) = .)
(cot(α) − csc(α))

The system (9.22) is a coupled system of partial differential equations for

the pair of functions (ψ(x, y), η(x, y)). The following exercise explores some
properties of solutions of this system.

*Exercise 9.17. (a) Suppose that the pair of functions (ψ(x, y), η(x, y))
satisfies the PDE system (9.22), where λ is any nonzero constant. Show
that the functions 2ψ, 2η must each be solutions of the sine-Gordon equa-
tion (9.1). (Hint: Differentiate the first equation in (9.22) with respect to y
and the second equation with respect to x; then add and subtract the result-
ing equations and apply trigonometric identities to simplify the right-hand
sides.)
(b) Now, suppose that 2ψ is any known solution of (9.1). Then the system
(9.22) can be regarded as the overdetermined PDE system
ηx = −ψx + λ sin(ψ − η),
1
ηy = ψy − sin(ψ + η)
λ
for the unknown function η(x, y). Show that this system is compatible, i.e.,
that (ηx )y = (ηy )x —precisely because 2ψ satisfies (9.1). It follows that this
system has a 1-parameter family of solutions η(x, y) and that these solutions
can be constructed using only techniques of ordinary differential equations.

The PDE system (9.22) is called a Bäcklund transformation for the sine-
Gordon 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 2ψ of the sine-
Gordon equation (9.1), the PDE system (9.22) can be used to construct
a 2-parameter family of new solutions 2η: One parameter comes from the
choice of the constant λ = 0, and one comes from the 1-parameter family of
solutions η to the system (9.22).
At this point, we have established several relationships involving pseudo-
spherical surfaces and solutions of the sine-Gordon equation:

(1) Every pseudospherical surface Σ of Gauss curvature K = −1 is

associated with a solution 2ψ to the sine-Gordon equation (9.1)
that describes the angle between its asymptotic directions at each
point (and vice versa).
9.5. Bäcklund transformation for the sine-Gordon equation 299

(2) Every pseudospherical surface Σ of Gauss curvature K = −1 has a

8 gener-
2-parameter family of associated pseudospherical surfaces Σ
ated by the construction in part (2) of Theorem 9.5.
(3) Every solution 2ψ to the sine-Gordon equation (9.1) has a 2-para-
meter family of associated solutions 2η to (9.1) generated by the
PDE system (9.22).

It seems natural to ask: Given a pseudospherical surface Σ with associated

solution 2ψ to (9.1), what is the relationship between the new surfaces Σ8
associated to Σ and the new solutions 2η to (9.1) associated to 2ψ? The
following exercise answers this question.
*Exercise 9.18. Let K = −1 (so that r = sin(α)), and let (ω̄ i , ω̄ ij ) be as in
Exercise 9.16.
˜ 1,
(a) Use equations (9.5), (9.7) to show that the Maurer-Cartan forms (ω̄
˜ 2 , ω̄
ω̄ ˜ 13 , ω̄ 8 are given by
˜ 23 ) on Σ
˜ 1 = cos(ψ − η) dx + cos(ψ + η) dy,
ω̄
˜ 2 = sin(ψ − η) dx + sin(ψ + η) dy,
ω̄
(9.23)
˜ 13 = sin(ψ − η) dx − sin(ψ + η) dy,
ω̄
˜ 23 = − cos(ψ − η) dx + cos(ψ + η) dy.
ω̄

8 are
(b) Show that the first and second fundamental forms of Σ
˜ 1 )2 + (ω̄
I = (ω̄ ˜ 2 )2 = dx2 + 2 cos(2η) dx dy + dy 2 ,
(9.24)
˜ 13 ω̄
II = ω̄ ˜ 1 + ω̄ ˜ 2 = −2 sin(2η) dx dy.
˜ 23 ω̄

(c) Conclude that the angle between the asymptotic directions of Σ 8 = x̃(U )
at the point x̃(x, y) is equal to 2η(x, y). (The sign of the second fundamental
form is unimportant, as it can be reversed by reversing the orientation of
8
Σ.)

The result of Exercise 9.18 completes the picture: The analytic Bäcklund
transformation (9.22) between solutions of the sine-Gordon equation (9.1)
is the precise analog of the geometric Bäcklund transformation between
pseudospherical surfaces of Gauss curvature K = −1.
Exercise 9.19. Let 2ψ(x, y) ≡ 0 be the trivial solution of the sine-Gordon
equation (9.1). (This solution corresponds to the degenerate “surface” con-
sisting of a straight line in E3 .) Show that the corresponding solutions
2η(x, y) generated by the Bäcklund transformation (9.22) are given by
η(x, y) = 2 tan−1 (Ce−(λx+ λ y) ),
1
300 9. Pseudospherical surfaces and Bäcklund’s theorem

where C = 0 is constant. (Hint: You may find the trig identity csc(η) +
cot(η) = cot( 12 η) useful.) The functions

2η = 4 tan−1 (Ce−(λx+ λ 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 Σ be the pseudosphere, with parametrization

⎡ ⎤
x + y − tanh(x + y)
⎢ ⎥
x(x, y) = ⎢ ⎥
⎣sech(x + y) cos(x − y)⎦ .
sech(x + y) sin(x − y)

(a) Show that the first and second fundamental forms of Σ are given by

2(cosh2 (x + y) − 2)
I = dx2 + 2 dx dy + dy 2 ,
cosh (x + y)

4 sinh(x + y)
II = dx dy.
cosh2 (x + y)

(b) Conclude that:

(1) (x, y) are asymptotic coordinates for Σ.

(2) The Gauss curvature of Σ is K = −1.

(3) The angle 2ψ(x, y) between the asymptotic directions of Σ at the

point x(x, y) satisfies the conditions

cosh2 (x + y) − 2 2 sinh(x + y)
(9.26) cos(2ψ(x, y)) = 2 , sin(2ψ(x, y)) = .
cosh (x + y) cosh2 (x + y)

(In particular, check that

" #2 " #2
cosh2 (x + y) − 2 2 sinh(x + y)
+ = 1.)
cosh2 (x + y) cosh2 (x + y)

(c) Show that the function ψ(x, y) defined by equation (9.26) is

ψ(x, y) = 2 tan−1 (ex+y ).

9.5. Bäcklund 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 “1-soliton” pseudo-
spherical surfaces.
Exercise 9.21. In this exercise, we will give a proof of Hilbert’s 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 infinite area,
diffeomorphic to R2 , with constant Gauss curvature K = −1. We will
explore hyperbolic spaces in more detail in Chapter 11.)
Suppose that x : H2 → E3 is an isometric immersion. Recall that H2
has Gauss curvature K = −1; therefore the image Σ = x(H2 ) is an im-
mersed pseudospherical surface. (“Immersed” allows for the possibility of
self-intersection; an immersion x : H2 → 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 infinite area, the same must be
true of Σ.
(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 Σ and
for which the first and second fundamental forms of Σ 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
(x̃, ỹ), respectively, we must have
x̃ = x + x0 , ỹ = y + y0
for some constants x0 , y0 . Since each pair of coordinates is only determined
up to additive constants, we can set x̃ = x, ỹ = y. A topological argument
shows that this patching construction works globally because H2 is simply
connected.)
(b) Let 2ψ : H2 → R be the solution of the sine-Gordon equation (9.1)
associated to Σ. Show that, because x is an immersion, the function ψ must
satisfy
π
0 < ψ(x, y) <
2
for all (x, y) ∈ H . (Hint: An immersed pseudospherical surface must have
2

linearly independent asymptotic directions at each point.)

(c) Let R ⊂ H2 be a rectangle of the form
a ≤ x ≤ b, c ≤ y ≤ d.
302 9. Pseudospherical surfaces and Bäcklund’s theorem

Use the sine-Gordon equation (9.1) and the Fundamental Theorem of Cal-
culus to show that the area of x(R) ⊂ Σ is given by
&
A(x(R)) = ω̄ 1 ∧ ω̄ 2
R
= 2 (ψ(b, d) − ψ(b, c) − ψ(a, d) + ψ(a, c)) .
Conclude from part (b) that A(x(R)) < 2π.
(d) Observe that, because H2 is complete and has infinite 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 Bäcklund transformation (9.22) relates two dif-
ferent solutions of the same PDE (9.1), it is also possible for a Bäcklund
transformation to relate solutions of two different PDEs. For example, con-
sider the first-order system of partial differential equations
(ψ−η)
ψx + ηx = 2λe 2 ,
(9.27) 1 (ψ+η)
ψy − ηy = e 2 .
λ

(a) Suppose that the pair of functions (ψ(x, y), η(x, y)) satisfies the PDE
system (9.27), where λ is any nonzero constant. Show that η(x, y) must be
a solution of the wave equation (in characteristic coordinates)
(9.28) ηxy = 0,
while ψ(x, y) must be a solution of Liouville’s equation
(9.29) ψxy = eψ .

(b) Show by integration that the general solution of the wave equation (9.28)
is
(9.30) η(x, y) = ρ(x) + σ(y),
where ρ(x), σ(y) are arbitrary functions of a single variable.
(c) Substitute equation (9.30) into the system (9.27) to obtain a system of
two first-order PDEs for the function ψ(x, y). By treating each of these
equations as a separable ODE in the appropriate variable, show that
" & & #
− 21 ψ 1
−ρ(x) 1
(9.31) e = −e 2 (ρ(x)−σ(y))
λ e dx + e σ(y)
dy .
2λ
9.6. Maple computations 303

(Hint: Write the right-hand sides of equations (9.27) as

1 1 1 (ψ−σ(y)+ρ(x)+2σ(y))
2λe 2 (ψ+ρ(x)−2ρ(x)−σ(y)) and e2 .)
λ
(d) Set
& &
−ρ(x) 1
X(x) = −λ e dx, Y (y) = − eσ(y) dy.
2λ
Solve equation (9.31) for ψ to obtain the general solution of Liouville’s equa-
tion:
" #
2X  (x)Y  (y)
(9.32) ψ(x, y) = ln ,
(X(x) + Y (y))2
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 different adapted frame fields in play in
this chapter (one frame field on each of the surfaces Σ, Σ 8 adapted to the
Bäcklund transformation and one principal adapted frame field on Σ), we’ll
need to keep track of three different sets of Maurer-Cartan forms. In or-
der to distinguish them a bit, we’ll use (theta1[i], theta1[i,j]) and
(theta2[i], theta2[i,j]) for the Maurer-Cartan forms on Σ, Σ, 8 respec-
tively, corresponding to the Bäcklund-adapted frame fields, and (omega[i],
omega[i,j]) for the Maurer-Cartan forms corresponding to the principal
frame field on Σ. 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]:= 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];
304 9. Pseudospherical surfaces and Bäcklund’s theorem

theta1[1,1]:= 0;
theta1[2,2]:= 0;
theta1[3,3]:= 0;
theta1[2,1]:= -theta1[1,2];
theta1[1,3]:= -theta1[3,1];
theta1[2,3]:= -theta1[3,2];
theta2[1,1]:= 0;
theta2[2,2]:= 0;
theta2[3,3]:= 0;
theta2[2,1]:= -theta2[1,2];
theta2[1,3]:= -theta2[3,1];
theta2[2,3]:= -theta2[3,2];
> for i from 1 to 3 do
d(omega[i]):= -add(’omega[i,j] &ˆ omega[j]’, j=1..3);
end do;
d(omega[1,2]):= -add(’omega[1,k] &ˆ omega[k,2]’, k=1..3);
d(omega[3,1]):= -add(’omega[3,k] &ˆ omega[k,1]’, k=1..3);
d(omega[3,2]):= -add(’omega[3,k] &ˆ omega[k,2]’, k=1..3);
for i from 1 to 3 do
d(theta1[i]):= -add(’theta1[i,j] &ˆ theta1[j]’, j=1..3);
end do;
d(theta1[1,2]):= -add(’theta1[1,k] &ˆ theta1[k,2]’, k=1..3);
d(theta1[3,1]):= -add(’theta1[3,k] &ˆ theta1[k,1]’, k=1..3);
d(theta1[3,2]):= -add(’theta1[3,k] &ˆ theta1[k,2]’, k=1..3);
for i from 1 to 3 do
d(theta2[i]):= -add(’theta2[i,j] &ˆ theta2[j]’, j=1..3);
end do;
d(theta2[1,2]):= -add(’theta2[1,k] &ˆ theta2[k,2]’, k=1..3);
d(theta2[3,1]):= -add(’theta2[3,k] &ˆ theta2[k,1]’, k=1..3);
d(theta2[3,2]):= -add(’theta2[3,k] &ˆ theta2[k,2]’, k=1..3);
We’ll also need the vector form of the structure equations; we’ll use x1 and
8 respectively, and (e11, e12, e13),
x2 for the parametrizations of Σ and Σ,
(e21, e22, e23) for the frame fields on the two surfaces that are adapted
to the Bäcklund 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 fields
8 as in equations
(e1 (u), e2 (u), e3 (u)) on Σ and (ẽ1 (u), ẽ2 (u), ẽ3 (u)) on Σ
(9.2) and that equation (9.3) holds. First, declare α and r to be constants,
and set up a substitution to move from one frame field 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 differentiate equation (9.3), and write the results in terms of the first
adapted frame field:
> 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 coefficients 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 Σ:
> zero1d:= Simf(zero1b - cot(alpha)*zero1c);
The result yields equation (9.6).
Next, we need to compute the 1-forms

˜ 13 = dẽ1 , ẽ3 ,
ω̄ ˜ 23 = dẽ2 , ẽ3 .
ω̄

Since we haven’t told Maple that (e11, e12, e13) are vectors, we’ll have
to define 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 Bäcklund’s theorem

+ coeff(exp1, e13)*coeff(exp2, e13));

end proc;
˜ 13 and ω̄
Now we can compute ω̄ ˜ 23 via the formulas above. First, compute ω̄
˜ 13 :

> Simf(innprod1(d(Simf(subs(frame2to1sub, e21))),

Simf(subs(frame2to1sub, e23))));

θ11,2 sin(α) + θ13,1 cos(α)

Apply the relation (9.6):

> Simf(subs(solve({zero1d}, {theta1[3,1]}), %));
This yields the expression in the first equation of (9.7). Similarly, the com-
putation
> 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
φ(x, y) = 2ψ(x, y) satisfies the sine-Gordon equation (9.1) and one to define
the desired Maurer-Cartan forms associated to the principal adapted frame
field (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 doesn’t really know how to compute symmetric products of differ-
ential forms, but the following commands will work for computing the first
and second fundamental forms:
> collect(simplify(Simf(subs(SGEformsub, omega[1]))ˆ2
+ Simf(subs(SGEformsub, omega[2]))ˆ2), {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 satisfied, check that com-
putations 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 doesn’t immediately reduce to zero is the one for ω̄21 ,
which requires an application of SGEsub in order to see that it vanishes.
Exercise 9.16: Equation (9.20) is immediate from the definition of η(u),
and the results of Chapter 4 yield the expressions for (ω̄ 1 , ω̄ 2 , ω̄ 31 , ω̄ 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 ω̄ 12 in terms of ω̄21 as follows: Compute dω̄ 1 and dω̄ 2
by differentiating the expressions for ω̄ 1 and ω̄ 2 in equation (9.21), and then
use the reverse substitution to express the result in terms of (ω̄ 1 , ω̄ 2 ):
> dtheta1[1]:= Simf(subs([omega[3]=0],
Simf(subs(rotationbacksub, Simf(d(Simf(subs(rotationsub,
theta1[1]))))))));
> pick(dtheta1[1], theta1[2]);
−ω1,2 + ηx d(x) + ηy d(y)
> dtheta1[2]:= Simf(subs([omega[3]=0],
Simf(subs(rotationbacksub, Simf(d(Simf(subs(rotationsub,
theta1[2]))))))));
> pick(dtheta1[2], theta1[1]);
ω1,2 − ηx d(x) − ηy d(y)
It follows from the structure equations for dω̄ 1 and dω̄ 2 that ω̄ 12 = ω̄21 − dη.
Add this to our substitution:
> rotationsub:= [op(rotationsub),
theta1[1,2] = omega[1,2] - d(eta(x,y))];
308 9. Pseudospherical surfaces and Bäcklund’s theorem

We can combine this with the expressions for (ω̄ i , ω̄ji ) in SGEformsub in order
to write (ω̄ i , ω̄ ij ) in terms of (dx, dy):
> SGErotationsub:= Simf(subs(SGEformsub, rotationsub));
Since K = −1, we have r = sin(α):
> r:= sin(alpha);
Now consider the Bäcklund 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 coefficients of dx and dy in this expression are PDEs involving the
functions ψ and η:
> PDE1:= pick(Backlundzero, d(x));
PDE2:= pick(Backlundzero, d(y));
These PDEs contain some coefficients involving trigonometric functions of
α, and they will look nicer if we rename them. In the first equation, divide
by sin(α) and set λ = cot(α) − csc(α) via the substitution:
> lambdasub:= [cos(alpha) = lambda*sin(alpha) + 1];
> Simf(subs(lambdasub, PDE1/sin(alpha)));

−λ sin(ψ) cos(η) + λ sin(η) cos(ψ) + ψx + ηx

> combine(%, trig);

λ sin(η − ψ) + ψx + ηx

Similarly, for the second equation divide by sin(α) and set μ = −(cot(α) +
csc(α)):
> musub:= [cos(alpha) = -mu*sin(alpha) - 1];
> Simf(subs(musub, PDE2/sin(alpha)));

μ sin(ψ) cos(η) + μ sin(η) cos(ψ) − ψy + ηy

> combine(%, trig);

μ sin(η + ψ) − ψy + ηy
9.6. Maple computations 309

All that remains is to show that λμ = 1, so that μ = λ1 :

> 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

that we have already computed in order to write the forms (ω̄ ˜ i , ω̄
˜ ij ) in terms
of (dx, dy). Equations (9.7) and the first and third equations in (9.5) allow
˜ 1 , ω̄
us to write (ω̄ ˜ 2 , ω̄
˜ 31 , ω̄
˜ 32 ) in terms of (ω̄ 1 , ω̄ 2 , ω̄ 31 , ω̄ 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
first and second fundamental forms in equation (9.24) as follows:
> collect(simplify(Simf(subs(theta2coordsub, theta2[1]))ˆ2
+ Simf(subs(theta2coordsub, theta2[2]))ˆ2), {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)});
 Chapter 10

Two classical theorems

In this chapter, we will see how moving frames may be used to prove two
classical results in differential geometry: the classification 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 Σ ⊂ R3 is called ruled if Σ contains a straight line segment

passing through each point x ∈ Σ. Ruled surfaces have been the subject of
much study in classical differential geometry; see, e.g., [Eis60] or [Wil62].
A surface Σ is called doubly ruled if it can be realized as a ruled surface in
two distinct ways, i.e., if Σ contains two linearly independent line segments
passing through each point x ∈ Σ.
There are two well-known examples of non-planar doubly ruled surfaces in
R3 : the hyperboloid of one sheet
2 3
(10.1) Σ = t[x, y, z] ∈ R3 | x2 + y 2 − z 2 = r2 ,
where r > 0 is a positive constant, and the hyperbolic paraboloid
2 3
(10.2) Σ = t[x, y, z] ∈ R3 | z = xy .
*Exercise 10.1. Recall that a regular surface Σ ⊂ R3 is ruled if and only if
every point x ∈ Σ has a neighborhood that can be given a parametrization
of the form
x(u, v) = α(u) + vβ(u),
where α is a regular curve in R3 and β(u) = 0.

311
312 10. Two classical theorems

(a) Show that the following maps x1 , x2 : R2 → R3 are parametrizations of

the hyperboloid (10.1):
x1 (u, v) = t [r cos(u), r sin(u), 0] + v t[− sin(u), cos(u), 1] ,

x2 (u, v) = t [r cos(u), r sin(u), 0] + v t[− sin(u), cos(u), −1] .

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 → 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 α(u), β(u), γ(v), δ(v) in R3 such that
x(u, v) = α(u) + vβ(u) = γ(v) + uδ(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-affine group A(3). Specifically, if Σ ⊂ R3 is a
doubly ruled surface and g ∈ A(3) is an equi-affine transformation, then the
surface g · Σ ⊂ 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-affine space A3 .
The goal of this section is to prove the following theorem [HCV52]:
Theorem 10.3. Any non-planar doubly ruled regular surface Σ ⊂ A3 is
equivalent via an equi-affine 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 affine group without the equi-affine restriction. Under this
group action, the hyperboloids of one sheet (10.1) for all positive
values of r are all equivalent, whereas different values of r give
surfaces that are inequivalent under the action of the equi-affine
group.
(2) If we regard R3 as an open subset of P3 via the identification

(x1 , x2 , x3 ) ↔ 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 → A3

be an immersion whose image is a regular surface Σ ⊂ A3 that is doubly
ruled. In order to apply the method of moving frames, it seems natural
to choose a frame field (e1 (u), e2 (u), e3 (u)) along Σ with the property that
for each u ∈ 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 field on Σ (as opposed to, say, an orthonormal frame
field) allows us sufficient flexibility to adapt our choice of frame field to the
surface in precisely this way. So, in this context, a unimodular frame field
314 10. Two classical theorems

(e1 (u), e2 (u), e3 (u)) along Σ will be called 0-adapted if for each u ∈ 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 “0-adapted” is more
restrictive than that in Chapter 6 because here we have additional geometric
information to guide our initial choice for an adapted frame field.)

Exercise 10.5. Explain why orthonormal frame fields would generally be

too restrictive to allow for an adapted frame field satisfying this tangency
condition.

Recall that choosing such a unimodular frame field along Σ is equivalent to

choosing a lifting x̃ : U → A(3) defined by

x̃(u) = (x(u); e1 (u), e2 (u), e3 (u))

and that the pullbacks (ω̄ i , ω̄ji ) to U of the Maurer-Cartan forms (ω i , ωji ) on
A(3) via the lifting x̃ satisfy the conditions that ω̄ 3 = 0 and

dx = e1 ω̄ 1 + e2 ω̄ 2 ,
(10.3)
dei = e1 ω̄i1 + e2 ω̄i2 + e3 ω̄i3 , i = 1, 2, 3.

*Exercise 10.6. (a) Show that the condition that the vector fields (e1 (u),
e2 (u)) are tangent to the rulings implies that

(10.4) de1 (e1 ) ≡ 0 mod e1 , de2 (e2 ) ≡ 0 mod e2 .

(Hint: Recall that de1 (e1 ) computes the directional derivative of the vector
field e1 (u) in the direction of e1 (u). What limitations are imposed on this
derivative by the condition that the vector field e1 (u) is tangent to a straight
line along this direction?)
(b) Show that the condition (10.4) implies that

(10.5) de1 ≡ 0 mod (e1 , ω̄ 2 ), de2 ≡ 0 mod (e2 , ω̄ 1 ).

Conclude from equations (10.3) and (10.5) that

ω̄12 = a212 ω̄ 2 , ω̄13 = a312 ω̄ 2 ,

(10.6)
ω̄21 = a121 ω̄ 1 , ω̄23 = a321 ω̄ 1

for some functions a121 , a212 , a312 , a321 on U .

(c) Use the Cartan structure equation

0 = dω̄ 3 = −ω̄13 ∧ ω̄ 1 − ω̄23 ∧ ω̄ 2

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
ω̄12 = a ω̄ 2 , ω̄13 = c ω̄ 2 ,
(10.7)
ω̄21 = b ω̄ 1 , ω̄23 = c ω̄ 1 .

Now we will investigate how the functions a, b, c transform if we vary our

choice of 0-adapted frame field on Σ.
*Exercise 10.7. Let (e1 (u), e2 (u), e3 (u)), (ẽ1 (u), ẽ2 (u), ẽ3 (u)) be any two
0-adapted frame fields on Σ, with associated Maurer-Cartan forms (ω̄ i , ω̄ji )
˜ i , ω̄
and (ω̄ ˜ ji ), respectively. For simplicity, assume that the vector field ẽ1 (u)
is tangent to the same family of rulings as e1 (u), and similarly for the vector
fields e2 (u) and ẽ2 (u).
(a) Show that
⎡ ⎤
λ1 0 r1
 ⎢ ⎥
(10.8) ẽ1 (u) ẽ2 (u) ẽ3 (u) = e1 (u) e2 (u) e3 (u) ⎢
⎣ 0 λ2 r2 ⎦
⎥

0 0 λ11λ2
for some functions λ1 , λ2 , r1 , r2 on U , with λ1 , λ2 = 0.
(b) Show that under a transformation of the form (10.8), we have
 1  1 1  3  2 
˜
ω̄ λ1 ω̄
˜1
ω̄ λ1 λ2 ω̄13
(10.9) = , = .
˜2 1 2 ˜ 23
ω̄ λ2 ω̄ ω̄ λ1 λ22 ω̄23
Conclude that the transformed function c̃ defined by the conditions
˜ 13 = c̃ ω̄
ω̄ ˜ 2, ˜ 23 = c̃ ω̄
ω̄ ˜1
is given by
(10.10) c̃ = λ21 λ22 c.

Thus, the function c is a relative invariant: If it vanishes for any 0-adapted

frame based at a point u ∈ 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 field (e1 (u), e2 (u),
e3 (u)) along Σ, the function c is identically equal to zero on U . Show that
this assumption implies that
de1 ≡ de2 ≡ 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 Σ 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 field (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 → A3 of Σ 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 Σ 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 field

(e1 (u), e2 (u), e3 (u)) for which c = ±1. If c = −1, we can exchange e1 (u)
and e2 (u) and replace e3 (u) by −e3 (u) to arrive at a new 0-adapted frame
field for which c = 1; thus, without loss of generality, we will assume that
we can choose a 0-adapted frame field for which c = 1. Such a frame field
will be called 1-adapted.
*Exercise 10.10. Let (e1 (u), e2 (u), e3 (u)) be any 1-adapted frame field
along Σ.
(a) Show that any other 1-adapted frame field (ẽ1 (u), ẽ2 (u), ẽ3 (u)) along Σ
must have the form
⎡ ⎤
λ 0 r1
 ⎢ ⎥
(10.11) ẽ1 (u) ẽ2 (u) ẽ3 (u) = e1 (u) e2 (u) e3 (u) ⎢
⎣ 0 ± λ r2 ⎦
1 ⎥

0 0 ±1
for some functions λ, r1 , r2 on U , with λ = 0.
(b) Note that the condition c = 1 is equivalent to the relations
(10.12) ω̄13 = ω̄ 2 , ω̄23 = ω̄ 1
among the associated Maurer-Cartan forms. Compare with Exercise 6.42(a),
and conclude that any 1-adapted frame field along Σ satisfies the defining
conditions for what we called a “1-adapted null frame” there. In particular,
Σ must be a hyperbolic equi-affine surface; otherwise no such frame field
could exist along Σ.
(c) Apply the result of Exercise 6.42(c) to show that there exists a choice of
1-adapted frame field along Σ for which ω̄33 = 0. (This condition corresponds
10.1. Doubly ruled surfaces in R3 317

to choosing e3 (u) to be parallel to the equi-affine normal direction.) Such a

frame field will be called 2-adapted.
(d) Show that any two 2-adapted frame fields (e1 (u), e2 (u), e3 (u)), (ẽ1 (u),
ẽ2 (u), ẽ3 (u)) along Σ must be related by a transformation of the form
⎡ ⎤
λ 0 0
 ⎢ ⎥
(10.13) ẽ1 (u) ẽ2 (u) ẽ3 (u) = e1 (u) e2 (u) e3 (u) ⎢ ⎥
⎣0 ± λ 0 ⎦
1

0 0 ±1

for some function λ = 0 on U .

*Exercise 10.11. Let (e1 (u), e2 (u), e3 (u)) be any 2-adapted frame field
along Σ.
(a) Apply the results of Exercises 6.42(e) and 6.42(f) and equations (10.7)
to show that
a = b = 0.

(Alternatively, differentiate equations (10.12) to arrive at the same result.)

Consequently, the Maurer-Cartan forms associated to any 2-adapted frame
field on Σ must satisfy the conditions

(10.14) ω̄21 = ω̄12 = 0.

Conclude from equation (10.14) that the Fubini-Pick form (6.26) of Σ is

identically zero.
(b) Conversely, show that if Σ is a hyperbolic equi-affine surface whose
Fubini-Pick form is identically zero, then Σ is doubly ruled. (Hint: It suffices
to show that the condition (10.5) holds.)
(c) Differentiate equations (10.14) and apply Cartan’s lemma to conclude
that there exist functions f, g on U such that

(10.15) ω̄31 = f ω̄ 1 , ω̄32 = g ω̄ 2 .

(d) Differentiate the equation ω̄33 = 0 and conclude that g = f .

(e) Substitute equations (10.15) (with g = f ) into the Cartan structure
equations (3.8) for dω̄31 and dω̄32 to obtain

df ∧ ω̄ 1 = df ∧ ω̄ 2 = 0.

Conclude that df = 0, and hence f is equal to a constant C ∈ 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 field e3 (u) is constant
on U . (Note that this implies that Σ is an improper equi-affine sphere; cf.
Exercise 6.41.)
(b) Recall that the Maurer-Cartan forms associated to a 2-adapted frame
field on Σ satisfy
ω̄11 + ω̄22 = 0.
Show that dω̄11 = dω̄22 = 0, and use the Poincaré lemma (cf. Theorem 2.31)
to conclude that there exists a function μ on U such that
(10.16) ω̄11 = dμ = −ω̄22 .
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 → A3 to each of these open subsets of U . The
theorem can then be obtained for the entire surface Σ = x(U ) via a patching
argument.

(c) Use equations (10.14) and (10.16) to show that

(10.17) dω̄ 1 = −dμ ∧ ω̄ 1 , dω̄ 2 = dμ ∧ ω̄ 2 .

(d) Show that    

d eμ ω̄ 1 = d e−μ ω̄ 2 = 0.
Conclude that by a transformation of the form (10.13) with λ = e−μ , we
can arrange that
(10.18) dω̄ 1 = dω̄ 2 = 0.
Then use equation (10.17) to conclude that dμ = 0, and hence
(10.19) ω̄11 = ω̄22 = 0.

To summarize, we have shown that when C = 0, there exists a 2-adapted

frame field along Σ for which the associated Maurer-Cartan forms (ω̄ i , ω̄ji )
satisfy the following conditions:
dω̄ 1 = dω̄ 2 = 0,
(10.20) ω̄11 = ω̄21 = ω̄12 = ω̄22 = ω̄31 = ω̄32 = ω̄33 = 0,
ω̄13 = ω̄ 2 , ω̄23 = ω̄ 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 Σ.

*Exercise 10.14. (a) Apply the Poincaré lemma to the first equations in
(10.20) to show that there exist functions u, v on U such that

(10.21) ω̄ 1 = du, ω̄ 2 = dv

(cf. Remark 10.13). Note that the condition ω̄ 1 ∧ ω̄ 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
working backwards) to show that there exist constant vectors (x̄, ē1 , ē2 , ē3 )
such that
e3 (u, v) = ē3 ,
e1 (u, v) = ē1 + vē3 ,
(10.23)
e2 (u, v) = ē2 + uē3 ,
x(u, v) = x̄ + uē1 + vē2 + uvē3 .

(d) Use equations (10.23) and the fact that (e1 (u), e2 (u), e3 (u)) is a uni-
modular frame field to show that

det ē1 ē2 ē3 = 1.

Conclude that via an equi-affine transformation, we can arrange that x̄ =

t [0, 0, 0] and

ē1 = t [1, 0, 0] , ē2 = t [0, 1, 0] , ē3 = t[0, 0, 1] ,

and hence
x(u, v) = t [u, v, uv] .
Therefore, up to equi-affine equivalence, Σ 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-affine transformation of the form
 
x = t x1 , x2 , x3 → t −x1 , x2 , −x3
will reverse the sign of C.
*Exercise 10.15. (a) Suppose that C > 0. Show that
(10.24) dω̄ 1 = −ω̄11 ∧ ω̄ 1 , dω̄ 2 = −ω̄22 ∧ ω̄ 2 ,
and use the Frobenius theorem (cf. Theorem 2.33) to conclude that every
point u ∈ U has a neighborhood V ⊂ U on which there exist functions
u, v, g1 , g2 such that
(10.25) ω̄ 1 = eg1 du, ω̄ 2 = eg2 dv.
It suffices 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 ω̄ 1 ∧ ω̄ 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 λ = e 2 (g1 −g2 ) ,
1

we can arrange that

(10.26) ω̄ 1 = eh du, ω̄ 2 = eh dv,
where h = 12 (g1 + g2 ).
(c) Use equations (10.24) (together with the fact that ω̄11 + ω̄22 = 0) to show
that
(10.27) ω̄11 = −ω̄22 = hu du − hv dv.
Then substitute this expression into the structure equation for dω̄11 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
Liouville’s equation
(10.29) zuv = ez .
*Exercise 10.16. (a) The general solution to Liouville’s 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

ũ = U (u), ṽ = V (v),

we can arrange that

" #
2
z(ũ, ṽ) = ln .
(ũ + ṽ)2
Dropping the tildes, this yields
 √ 
1 2
(10.31) h(u, v) = (z(u, v) − ln C) = ln √ .
2 C(u + v)

(b) Use equations (10.12), (10.14), (10.15), (10.26), (10.27), and (10.31) to
show that
√ √
2 2
ω̄ = √
1
du, ω̄ = √
2
dv,
C(u + v) C(u + v)
ω̄21 = ω̄12 = ω̄33 = 0,
1
ω̄11 = −ω̄22 = (−du + dv),
(10.32) (u + v)
√ √
2 2
ω̄1 = √
3
dv, ω̄2 = √
3
du,
C(u + v) C(u + v)
√ √
1 2C 2 2C
ω̄3 = du, ω̄3 = dv.
(u + v) (u + v)

(c) From equations (10.32), conclude that the structure equations (10.3) now
take the form
 √   √ 
2 2
dx = e1 √ du + e2 √ dv ,
C(u + v) C(u + v)
" #  √ 
1 2
de1 = e1 (−du + dv) + e3 √ dv ,
(u + v) C(u + v)
(10.33) " #  √ 
1 2
de2 = e2 (du − dv) + e3 √ du ,
(u + v) C(u + v)
 √   √ 
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) √ √
2 2
xu = √ e1 , xv = √ e2 ,
C(u + v) C(u + v)
√
1 1 2
(e1 )u = − e1 , (e1 )v = e1 + √ e3 ,
(u + v) (u + v) C(u + v)
√
1 2 1
(e2 )u = e2 + √ e3 , (e2 )v = − e2 ,
(u + v) C(u + v) (u + v)
√ √
2C 2C
(e3 )u = e1 , (e3 )v = 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

1 1
(10.35) e1 (u, v) = f (v), e2 (u, v) = g(u).
(u + v) (u + v)

(c) Use the equations for (e1 )v and (e2 )u to write

√ √
C C
(10.36) e3 (u, v) = √ ((u + v)(e2 )u − e2 ) = √ ((u + v)(e1 )v − e1 ) .
2 2

Substitute equations (10.35) into this equation to show that

(10.37) (u + v)f  (v) − 2f (v) = (u + v)g (u) − 2g(u).

(d) Differentiate 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 = −c11 , c20 = c10 . Set

c1 = c11 = −c21 , c0 = c10 = c20 ,

so that (10.38) becomes

(10.39) f (v) = v 2 c2 + vc1 + c0 , g(u) = u2 c2 − uc1 + c0 .

10.1. Doubly ruled surfaces in R3 323

(e) Substitute (10.39) into equations (10.35), (10.36) to obtain

1
e1 (u, v) = (v 2 c2 + vc1 + c0 ),
(u + v)
1
(10.40) e2 (u, v) = (u2 c2 − uc1 + c0 ),
(u + v)
√
C
e3 (u, v) = √ (2uvc2 + (u − v)c1 − 2c0 ).
2(u + v)

Observe from equations (10.33) that d(x − C1 e3 ) = 0; this implies that

x − C1 e3 is constant along Σ (in particular, Σ is a proper equi-affine sphere;
cf. Exercise 6.41), and by a translation we can arrange that x − C1 e3 = 0.
Then we have
1 1
(10.41) x(u, v) = e3 (u, v) = √ (2uvc2 + (u − v)c1 − 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
√
 C 
det e1 (u, v) e2 (u, v) e3 (u, v) = √ det c2 c1 c0 .
2
Since (e1 , e2 , e3 ) must be a unimodular frame, conclude that
√
 2
(10.42) det c2 c1 c0 = √ .
C

(h) By an equi-affine transformation, we can arrange that

√
1 t 2t 1 t
c2 = √ √6
[1, 0, 1] , c1 = √
6
[0, 1, 0] , c0 = √ √ [−1, 0, 1] .
2 C C 26C
Check that these vectors satisfy the determinant condition (10.42) and that
with this choice, we have
1
x(u, v) = 2
t
[uv + 1, u − v, uv − 1] .
C (u + v)
3

Finally, check that the coordinates of x(u, v) satisfy the defining equation
(10.1) for the hyperboloid, with r = C − 3 . Therefore, up to equi-affine
2

equivalence, Σ 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 “size” 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 “size” 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 affine
Grassmannian G1 (E2 ) (cf. §9.2). In general, a measure on a manifold M is a
differential form dμ on M (which, despite the notation, is not necessarily an
exact form) that is used to define the size of any subset of M via integration.
Specifically, the measure of a subset Ω ⊂ M is given by
&
μ(Ω) = dμ.
Ω

One familiar example is the area measure on the Euclidean plane E2 : The
area measure is the 2-form dA = dx ∧ dy, and the area of any (measurable)
subset Ω ⊂ E2 is given by
&
A(Ω) = dA.
Ω
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 ∈ E(2) and any
measurable subset Ω ⊂ E2 , we have
A(g · Ω) = A(Ω).
We can see how this invariance comes about as follows: Recall that we can
regard E2 as the homogeneous space E2 ∼= E(2)/SO(2); i.e., E2 is the set of
left cosets of the subgroup SO(2) ⊂ E(2):

SO(2) - E(2)
π
?
E2 ∼
= E(2)/SO(2).

The Maurer-Cartan forms (ω 1 , ω 2 , ω21 ) form a basis for the left-invariant 1-

forms on the Lie group E(2), and of these, ω 1 and ω 2 are semi-basic for
the projection π : E(2) → 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 = ω 1 ∧ ω 2 .
10.2. The Cauchy-Crofton formula 325

(While the 1-forms (ω 1 , ω 2 ) are not individually well-defined 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 Maurer-
Cartan forms: Given any measurable subset Ω ⊂ E2 and any element g ∈
E(2), we have
&
A(g · Ω) = ω1 ∧ ω2
g·Ω
&
= L∗g (ω 1 ∧ ω 2 )
&Ω
= ω1 ∧ ω2
Ω
= A(Ω).

Now consider the space of lines in E2 . As we saw in §9.2 when we constructed

the affine 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 = ±e1 and y = x + te1
for some t ∈ R (cf. Exercise 9.1). In this case, we write (x, e1 ) ∼ (y, f1 ),
and the affine Grassmannian is defined to be the quotient space
 
G1 (E2 ) = E2 × S1 / ∼ .

*Exercise 10.18. (a) Show that the product E2 ×S1 is diffeomorphic to the
oriented, orthonormal frame bundle F (E2 ), which in turn is diffeomorphic
to the Lie group E(2). (Hint: Choosing one unit vector e1 ∈ Tx E2 uniquely
determines a second unit vector e2 ∈ 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 ≥ 3.) Consequently, we can write

G1 (E2 ) = E(2)/ ∼ .

(b) Show that the equivalence relation (x, e1 ) ∼ (y, f1 ) on E2 × S1 corre-

sponds to the following equivalence relation on E(2): If
   
1 0 0 1 0 0
x̃ = , ỹ = ∈ E(2),
x e1 e2 y f1 f2

then x̃ ∼ ỹ if and only if

⎡ ⎤
1 0 0
ỹ = x̃ ⎣ t ±1 0 ⎦
0 0 ±1
326 10. Two classical theorems

for some t ∈ 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
⎧⎡ ⎤ ⎫
⎨ 1 0 0 ⎬
H = ⎣ t ±1 0 ⎦ : t ∈ R ⊂ E(2),
⎩ ⎭
0 0 ±1

and E(2) may be regarded as a principal bundle over G1 (E2 ) with fiber
group H:

H - E(2)

π
?
G1 (E2 ) ∼
= E(2)/H.

(c) Show that the Maurer-Cartan forms (ω 2 , ω21 ) on E(2) are semi-basic for
the projection π : E(2) → G1 (E2 ).

This suggests that a reasonable way to define a measure dμ on G1 (E2 ) might

be to set
dμ = |ω 2 ∧ ω21 |.
Like the area measure dA = ω 1 ∧ ω 2 , the 2-form dμ is well-defined on the
quotient G1 (E2 ), even though the individual 1-forms (ω 2 , ω21 ) are not. (The
absolute value sign is necessary because we have defined G1 (E2 ) to be the
space of unoriented lines, but reversing the orientation changes the sign of
the 2-form ω 2 ∧ ω21 ; therefore, this 2-form is only well-defined up to sign on
G1 (E2 ).) Then, given any (measurable) subset Ω ⊂ G1 (E2 ), we define the
measure μ(Ω) to be
& &
μ(Ω) = dμ = |ω 2 ∧ ω21 |.
Ω Ω

Exercise 10.19. Another way to determine a line  in E2 is to specify

two parameters: the angle θ with 0 ≤ θ < π such that the vector e1 =
t[cos(θ), sin(θ)] is parallel to  and the shortest distance ρ from the origin

to . This identification gives rise to a diffeomorphism

G1 (E2 ) ∼
= R × S1

(where S1 represents R/πZ rather than the more typical R/2πZ), with co-
ordinates
{(ρ, θ) | ρ ∈ R, 0 ≤ θ < π} .
10.2. The Cauchy-Crofton formula 327

Show that the measure dμ on G1 (E2 ) is then given by the 2-form

dμ = |dρ ∧ dθ|
on R × S1 . (Hint: How are the Maurer-Cartan forms (ω 2 , ω21 ) related to
(dρ, dθ)?)

Before we state the Cauchy-Crofton formula, we need to introduce the notion

of incidence:
Definition 10.20. A point x ∈ E2 and a line  ∈ G1 (E2 ) are called incident
if the point x lies on the line . A curve α in E2 and a line  ∈ G1 (E2 ) are
called incident if any point of α is incident with .
*Exercise 10.21. Let π1 : E(2) → E2 denote the projection from E(2) to
E2 and let π2 : E(2) → G1 (E2 ) denote the projection from E(2) to G1 (E2 ),
as in the following diagram:

SO(2) - E(2)  H
π1  S π2

/ SS
w
E2 G1 (E2 ).

Show that a point x ∈ E2 and a line  ∈ G1 (E2 ) are incident if and only if
the left cosets π1−1 (x), π2−1 () ⊂ E(2) have nonempty intersection.

We are now ready to state the Cauchy-Crofton formula.

Theorem 10.22. Let α : [a, b] → E2 be a curve in the Euclidean plane, and
let Ω ⊂ G1 (E2 ) be the set of lines incident with α, counted with multiplicity
(e.g., if a line  intersects α at two distinct points, then it counts twice).
Then
μ(Ω) = 2L,
where L is the length of α.

To begin the proof, let

8 = π −1 (α([a, b])) ⊂ E(2).
Ω 1
8 consists of all oriented, orthonormal frames (x; e1 , e2 ) based at all
The set Ω
points x ∈ α, each of which corresponds to a line  passing through the point
x and parallel to the frame vector e1 . Moreover, any line  that intersects
α in k distinct points is represented by 2k distinct points in Ω: 8 Each line
appears twice in Ω8 for each point x where it intersects α, represented by
frames of the form (x; e1 , e2 ) and (x; −e1 , −e2 ).
328 10. Two classical theorems

8 is a 2-dimensional submanifold
*Exercise 10.23. Convince yourself that Ω
of E(2).

Now, let
8 ⊂ G1 (E2 ),
Ω = π2 (Ω)

and observe that Ω consists of all lines incident with α. Moreover, for any
8 is precisely twice the number of
 ∈ Ω, the cardinality of the set π2−1 () ∩ Ω
distinct points in which  intersects α. So the set Ω 8 represents all the lines
in E that are incident with α, counted with double multiplicity.
2

The restriction of the projection π2 : E(2) → G1 (E2 ) to the set Ω 8 is an

8 8
immersion π2 : Ω → Ω; in fact, this map realizes Ω as a double cover of Ω.
Because the measure dμ = |ω 2 ∧ ω21 | is semi-basic for the projection π2 , it
follows that
& & &
μ(Ω) = dμ = 21 ∗
π2 (dμ) = 21 8
dμ = 12 μ(Ω),
Ω 
Ω 
Ω

where the integral on the left must be counted with multiplicity. (We are
abusing notation slightly here by writing dμ for both the 2-form |ω 2 ∧ ω21 |
on E(2) and the well-defined 2-form dμ on G1 (E2 ) whose pullback to E(2)
via π2 is equal to |ω 2 ∧ ω21 |.)

*Exercise 10.24. Let α : [a, b] → E2 be an arc-length parametrization of

α, given by α(s) = t[x(s), y(s)].
(a) Show that the vectors
 
ē1 (s) = t x (s), y  (s) , ē2 (s) = t −y  (s), x (s)

form an oriented, orthonormal frame at the point α(s).

8 can be parametrized by the map
(b) Show that the set Ω

α̃ : [a, b] × [0, 2π] → E(2)

defined by

α̃(s, θ) = (x(s, θ); e1 (s, θ), e2 (s, θ))

= (α(s); cos(θ) ē1 (s) + sin(θ) ē2 (s), − sin (θ)ē1 (s) + cos(θ) ē2 (s))
     
x(s) x (s) cos(θ)−y  (s) sin(θ) −x (s) sin(θ)−y  (s) cos(θ)
= ;  , .
y(s) y (s) cos(θ)+x (s) sin(θ) −y  (s) sin(θ)+x (s) cos(θ)
10.3. Maple computations 329

(c) Pull back the Cartan structure equation

dx = e1 ω 1 + e2 ω 2
on E(2) via α̃ to show that the 1-forms ω̄ 1 = α̃∗ ω 1 and ω̄ 2 = α̃∗ ω 2 are given
by
(10.43) ω̄ 1 = cos(θ) ds, ω̄ 2 = − sin(θ) ds.
Then use the structure equations for dω̄ 1 and dω̄ 2 to show that
(10.44) ω̄21 = −dθ + λ ds
for some function λ(s, θ). (In fact, the structure equation for dω̄21 implies
that λ is a function of s alone.)
(d) Use equations (10.43) and (10.44), together with the fact that
& & &
|ω ∧ ω2 | =
2 1
|ω ∧ ω2 | =
2 1
α̃∗ (|ω 2 ∧ ω21 |),

Ω α̃([a,b]×[0,2π]) [a,b]×[0,2π]

to show that & &

2π b
8 =
μ(Ω) | sin(θ)| ds dθ = 4L,
0 a
where L = (b − a) is the length of α. Conclude that μ(Ω) = 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
(ω̄ i , ω̄ji ) associated to a unimodular frame field on Σ, and tell Maple about
the structure equations and the relation ω̄11 + ω̄22 + ω̄33 = 0.
Now, suppose that Σ is doubly ruled and that we have chosen an adapted
frame field 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]];
˜ i , ω̄
Exercise 10.7: Begin by introducing new 1-forms (ω̄ ˜ ji ) to represent the
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] = lambda1ˆ2*lambda2*omega[3,1],
Omega[3,2] = lambda1*lambda2ˆ2*omega[3,2]];
> framechangebacksub:= makebacksub(framechangesub);
In order to find how the function c transforms, combine the substitutions
framechangesub and ruledsub to see how ω̄ ˜ 13 and ω̄
˜ 23 are related to ω̄
˜ 2 and
˜ 1 , respectively:
ω̄
> Simf(subs(framechangebacksub, Simf(subs(ruledsub,
Simf(subs(framechangesub, Omega[3,1]))))));

λ12 λ22 c Ω2

> Simf(subs(framechangebacksub, Simf(subs(ruledsub,

Simf(subs(framechangesub, Omega[3,2]))))));

λ12 λ22 c Ω1

So we have c̃ = λ21 λ22 c, as in equation (10.10).

Exercise 10.11: Since (e1 (u), e2 (u), e3 (u)) is a 2-adapted frame field, we
have c = 1, and since we have already set ω̄33 = −(ω̄11 + ω̄22 ), the condition
ω̄33 = 0 is equivalent to ω̄11 +ω̄22 = 0. Add these conditions to our substitution:
> c:= 1;
> ruledsub:= [op(ruledsub), omega[2,2] = -omega[1,1]];
Now differentiate equations (10.12):
> Simf(subs(ruledsub, Simf(d(omega[3,1] - omega[2]))));

−2 a (ω1 ) &ˆ (ω2 )

> Simf(subs(ruledsub, Simf(d(omega[3,2] - omega[1]))));

2 b (ω1 ) &ˆ (ω2 )

Since ω̄ 1 ∧ ω̄ 2 = 0, it follows that a = b = 0.

> a:= 0;
b:= 0;
10.3. Maple computations 331

Now differentiate equations (10.14):

> Simf(subs(ruledsub, Simf(d(omega[1,2]))));

(ω1 ) &ˆ (ω1,3 )

> Simf(subs(ruledsub, Simf(d(omega[2,1]))));

(ω2 ) &ˆ (ω2,3 )

By Cartan’s 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 differentiate the equation ω̄11 + ω̄22 = 0:
> Simf(subs(ruledsub, Simf(d(omega[1,1] + omega[2,2]))));

(−f + g) (ω1 ) &ˆ (ω2 )

Therefore, g = f .
> g:= f;
Finally, differentiate equations (10.15):
> Simf(subs(ruledsub, Simf(d(omega[1,3]))
- d(Simf(subs(ruledsub, omega[1,3])))));

(ω1 ) &ˆ (d(f ))

> Simf(subs(ruledsub, Simf(d(omega[2,3]))

- d(Simf(subs(ruledsub, omega[2,3])))));

(ω2 ) &ˆ (d(f ))

It follows from Cartan’s lemma that df = 0, and hence f is equal to a

constant C ∈ 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 dω̄ 1 and dω̄ 2 :

> 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 (ω̄ 1 , ω̄ 2 ) have the form (10.26). Add these conditions to our substi-
tution, and note that since ω̄ 1 and ω̄ 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 ω̄11 = −ω̄22 :
> zero1:= Simf(subs(ruledsubcase2, Simf(d(omega[1]))
- d(Simf(subs(ruledsubcase2, omega[1])))));
> factor(pick(zero1, d(u)));

−eh (ω11 + hv d(v))

> zero2:= Simf(subs(ruledsubcase2, Simf(d(omega[2]))

- d(Simf(subs(ruledsubcase2, omega[2])))));
> factor(pick(zero1, d(v)));

−eh (−ω11 + hu d(u))

Equation (10.27) then follows from Cartan’s 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 dω̄11 :
> Simf(subs(ruledsubcase2, Simf(d(omega[1,1]))
- d(Simf(subs(ruledsubcase2, omega[1,1])))));

(C e2 h − 2 hu,v ) (d(v)) &ˆ (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 off 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 coefficients of du and dv
in each of these expressions:
> pde1a:= pick(zero1, d(u));
pde1b:= pick(zero1, d(v));
pde2a:= pick(zero2, d(u));
pde2b:= pick(zero2, d(v));
pde3a:= pick(zero3, d(u));
pde3b:= pick(zero3, d(v));
pde4a:= pick(zero4, d(u));
pde4b:= pick(zero4, d(v));
In this case, it turns out that Maple can solve the entire system in one
step. (This often isn’t 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 transla-
tion to eliminate one of them, this gives the expressions in equations (10.40),
(10.41).
Exercise 10.24: Since this exercise requires different Maurer-Cartan forms
with different structure equations from the previous exercises, it’s 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] &ˆ omega[2];
d(omega[2]):= -omega[2,1] &ˆ omega[1];
d(omega[1,2]):= 0;
Declare the functions (x(s), y(s)), and define the vectors (x(s, θ), e1 (s, θ),
e2 (s, θ)):
> 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, θ), e2 (s, θ)) in order to
determine (ω̄ 1 , ω̄ 2 ) via the structure equation
dx = e1 ω̄ 1 + e2 ω̄ 2 :
> 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 (ω̄ 1 , ω̄ 2 ). Now use the structure
equations for dω̄ 1 and dω̄ 2 to determine ω̄21 :
> zero2:= Simf(subs(coordsub, Simf(d(omega[1]))
- d(Simf(subs(coordsub, omega[1])))));
> factor(pick(zero2, d(s)));
sin(θ) (ω1,2 + d(θ))
> 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(θ) (ω1,2 + d(θ))
Since sin(θ) and cos(θ) cannot vanish simultaneously, Cartan’s lemma im-
plies that ω̄21 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 α̃∗ (ω 2 ∧ ω21 ) as follows:
> Simf(subs(coordsub, omega[2] &ˆ omega[1,2]));
− sin(θ) (d(θ) &ˆ d(s))
Part 4

Beyond the flat case:

Moving frames on
Riemannian manifolds
 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 defined (Euclidean, Minkowski,
equi-affine, projective) are flat. This property is encoded in the Cartan
structure equations (3.8): According to the second equation in (3.8), the
matrix of connection forms ωc = [ωji ] satisfies the structure equation

(11.1) dωc + ωc ∧ ωc = 0.

(Note that this is not quite the same thing as the Maurer-Cartan equa-
tion (3.13) because ωc is a submatrix of the matrix-valued Maurer-Cartan
form ω.)
This might not seem like such a big deal, but in fact flatness is a necessary
condition for the existence of canonical isomorphisms Tx Rn ∼ = Rn (cf. Re-
mark 3.13 and the discussion in §3.3.2). It is these canonical isomorphisms
that allow us to think of the components (e1 (x), . . . , en (x)) of a frame field
on Rn as functions from Rn to Rn rather than as sections of the tangent
bundle T Rn , which in turn allows us to define 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 “curved” versions of En : elliptic space Sn (which is diffeomorphic to
an n-dimensional sphere) and hyperbolic space Hn (which is diffeomorphic
to Rn , but with a different metric structure than that of En ). We will see
that for these spaces, the flatness condition (11.1) no longer holds; rather,
the matrix of 2-forms
(11.2) Ω = dωc + ωc ∧ ωc ,
called the curvature of the connection matrix ωc , is nonzero and is an im-
portant 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 flat homogeneous spaces in Chapter
3. Fortunately, Sn and Hn may naturally be regarded as submanifolds of
the flat 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 .

Definition 11.1. The n-dimensional elliptic space Sn is the unit sphere in
En+1 ; i.e.,
Sn = {x ∈ En+1 | x, x = 1}.
The symmetry group of Sn is defined 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 ma-
trices  ∈ GL(n + 2) of the form
 t 
1 0
(11.3) Â = ,
0A
where A ∈ SO(n+1). Therefore, the symmetry group of Sn is isomorphic to
SO + (n + 1), and we will generally identify the element  in equation (11.3)
with the corresponding element A ∈ 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 ∈ Sn .
*Exercise 11.3. Let x0 = t[1, 0, . . . , 0] ∈ Sn . Show that:
(a) The isotropy group Hx0 of x0 in SO(n + 1) is
( t  )
1 0
(11.4) Hx0 = : Ā ∈ SO(n) .
0 Ā
11.2. The homogeneous spaces Sn and Hn 341

(b) The isotropy group Hx of any other point x ∈ Sn is

Hx = tx Hx0 t−1
x ,
where tx is any matrix in SO(n + 1) whose first column is x. (The transfor-
mation tx will then have the property that tx (x0 ) = x.)

Orthonormal frames on Sn are defined as follows:

Definition 11.4. An orthonormal frame f on Sn is a list of vectors f =
(e0 , . . . , en ), where e0 ∈ Sn ⊂ En+1 , e1 , . . . , en ∈ En+1 , and e0 · · · en ∈
SO(n+1). We identify  e0 with the position vector x ∈ Sn , and the condition
that e0 · · · en ∈ SO(n + 1) implies that the vectors (e1 , . . . , en ) may be
regarded as an orthonormal basis for the tangent space Te0 Sn ⊂ Te0 En+1 ∼ =
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 · · · en .
The projection map π : SO(n + 1) → Sn defined by
π([e0 . . . en ]) = e0
describes SO(n + 1) as a principal bundle over Sn with fiber group SO(n);
therefore, we have a natural correspondence Sn ∼
= SO(n + 1)/SO(n).
Because we have defined the components (e0 , . . . , en ) of a frame as elements
of the vector space En+1 , we can regard them as functions eα : F (Sn ) →
En+1 and define the Maurer-Cartan forms (ωβα ) on SO(n + 1) as usual by
the equations
(11.5) deα = eβ ωαβ ,
where 0 ≤ α, β ≤ n and ωαβ = −ωβα .
*Exercise 11.5. (a) Show that the forms (ω0i ) for 1 ≤ i ≤ n are semi-basic
for the projection π : SO(n + 1) → Sn and so may be regarded as the dual
forms of any orthonormal frame (e1 , . . . , en ) based at e0 ∈ Sn . The forms
(ωji ) for 1 ≤ i, j ≤ n may then be regarded as the connection forms.

(b) Set ω i = ω0i . Show that the dual forms (ω i ) and the connection forms
(ωji ) satisfy the structure equations
dω i = −ωji ∧ ω j ,
(11.6)
dωji = −ωki ∧ ωjk + ω i ∧ ω 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 con-
nection matrix ωc = [ωji ] is given by
Ω = [Ωij ] = [ω i ∧ ω j ].
The nonzero entries of Ω reflect 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 Maurer-
Cartan equation
dω = −ω ∧ ω,
where ⎡ ⎤
0 −ω 1 · · · −ω n
⎢ω1 ω1 · · · ω1 ⎥
⎢ 1 n ⎥
ω=⎢ . .. .. ⎥
⎣ .. . . ⎦
ω n ω1n · · · ωnn
is the so(n + 1)-valued Maurer-Cartan form on SO(n + 1).

11.2.2. Hyperbolic space Hn .

Definition 11.6. The n-dimensional hyperbolic space Hn is the upper sheet
of the two-sheeted hyperboloid
(x0 )2 − (x1 )2 − · · · − (xn )2 = 1
in M1,n ; i.e.,
Hn = {x ∈ 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 defines a Riemannian metric on Hn . (Technically, this re-
quires reversing signs in the definition of the Minkowski inner product so
that its restriction to each tangent space Tx Hn yields a quadratic form that
is positive definite rather than negative definite, but for the sake of con-
sistency we will retain the sign conventions of Chapter 5.) The symmetry
group of Hn is defined 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 ma-
trices  ∈ GL(n + 2) of the form
 t 
1 0
(11.7) Â = ,
0A
where A ∈ SO+ (1, n). Therefore, the symmetry group of Hn is isomorphic
to SO + (1, n), and we will generally identify the element  in equation (11.7)
with the corresponding element A ∈ 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 ∈ Hn .
*Exercise 11.8. Let x0 = t[1, 0, . . . , 0] ∈ Hn . Show that:
(a) The isotropy group Hx0 of x0 in SO+ (1, n) is
( t  )
1 0
(11.8) Hx0 = : Ā ∈ SO(n) .
0 Ā

(b) The isotropy group Hx of any other point x ∈ Hn is

Hx = tx Hx0 t−1
x ,
where tx is any matrix in SO+ (1, n) whose first column is x. (The transfor-
mation tx will then have the property that tx (x0 ) = x.)

Orthonormal frames on Hn are defined as follows:

Definition 11.9. An orthonormal frame f on Hn is a list of vectors f =
(e0 , . . . , en ), where e0 ∈ Hn ⊂ M1,n , e1 , . . . , en ∈ M1,n , and e0 · · · en ∈
SO + (1, n). We identify  e0 with the position vector x ∈ Hn , and the con-
dition that e0 · · · en ∈ 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 eα ∈ En+1 satisfies eα , eα  = 1, and so could, if desired, be
identified with a point of Sn . But for an orthonormal frame on Hn , only
the vector e0 satisfies the defining condition e0 , e0  = 1 for points in Hn ,
while the vectors (e1 , . . . , en ) each satisfy ei , ei  = −1. This illustrates the
fact that for non-flat 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 flat 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 · · · en .
The projection map π : SO+ (1, n) → Hn defined by
π([e0 . . . en ]) = e0
344 11. Curves and surfaces in elliptic and hyperbolic spaces

describes SO+ (1, n) as a principal bundle over Hn with fiber group SO(n);
therefore, we have a natural correspondence Hn ∼ = SO+ (1, n)/SO(n).
Because we have defined the components (e0 , . . . , en ) of a frame as elements
of the vector space M1,n , we can regard them as functions eα : F (Hn ) →
M1,n and define the Maurer-Cartan forms (ωβα ) on SO+ (1, n) as usual by
the equations
(11.9) deα = eβ ωαβ ,
where 0 ≤ α, β ≤ n. Recall from Exercise 3.50 that these forms satisfy the
relations ⎧
⎪
⎨0, α = β,
β α
ωα = ωβ , α = 0 or β = 0,
⎪
⎩ α
−ωβ , α, β ≥ 1.

*Exercise 11.11. (a) Show that the forms (ω0i ) for 1 ≤ i ≤ n are semi-basic
for the projection π : SO+ (1, n) → Hn and so may be regarded as the dual
forms of any orthonormal frame (e1 , . . . , en ) based at e0 ∈ Hn . The forms
(ωji ) for 1 ≤ i, j ≤ n may then be regarded as the connection forms.

(b) Set ω i = ω0i . Show that the dual forms (ω i ) and the connection forms
(ωji ) satisfy the structure equations
dω i = −ωji ∧ ω j ,
(11.10)
dωji = −ωki ∧ ωjk − ω i ∧ ω j .
(Compare with equations (11.6).) The second equation in (11.10) implies
that the curvature (11.2) of the connection matrix ωc = [ωji ] is given by
Ω = [Ωij ] = [−ω i ∧ ω j ].
The nonzero entries of Ω reflect the fact that all sectional curvatures of Hn
are identically equal to −1.
(c) Show that the structure equations (11.10) are equivalent to the Maurer-
Cartan equation
dω = −ω ∧ ω,
where ⎡ ⎤
0 ω1 · · · ωn
⎢ω1 ω11 · · · ωn1 ⎥
⎢ ⎥
ω=⎢ . .. .. ⎥
⎣ .. . . ⎦
ω n ω1n · · · ωnn
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 flat

metric spaces En+1 and M1,n , it is also somewhat artificial. These manifolds
are perfectly well-defined as intrinsic n-dimensional homogeneous spaces of
the Lie groups SO(n + 1) and SO+ (1, n), respectively, and they shouldn’t
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
flat homogeneous space in which Xn is defined 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 ⊂ Tx Vn+1 ∼= Vn+1 inherits an inner
product ·, · 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 definition for orthonormal frames on Xn that is
equivalent to Definitions
 11.4 and 11.9, as follows. (Note that the condition
det e0 · · · en = 1 implied by these definitions induces an orientation on
each tangent space Te0 Xn .)

Definition 11.12. An (oriented) orthonormal frame f on Xn is a list f =

(e0 , . . . , en ), where e0 ∈ Xn and (e1 , . . ., en ) is an oriented, orthonormal ba-
sis 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 identified with the Lie group G.

The Maurer-Cartan form ω and its components (ω i , ωji ) are defined 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
differentiating vector fields on Xn . We can no longer think of the compo-
nents (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 ) ∈ F (Xn ), we have

ei (f ) ∈ Te0 (f ) Xn

for 1 ≤ i ≤ n. Moreover, the derivatives of the functions ei : F (Xn ) → T Xn

at a point f ∈ 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 defined 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 or-
thogonal projection of these extrinsically defined exterior derivatives onto
the tangent plane Te0 Xn . This idea leads to the notion of the covariant
derivative for vector fields on Xn :

Definition 11.13. Let e0 ∈ Xn ⊂ Vn+1 , and let w ∈ Te0 Xn ⊂ Te0 Vn+1 = ∼

V n+1 . Let v be a (tangent) vector field on X , regarded as a function
n

v : Xn → Vn+1 with the property that for every x ∈ Xn , we have v(x) ∈

Tx Xn ⊂ Vn+1 . Let πe0 : Vn+1 → 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 ∇w v ∈ Te0 Xn defined by
(11.11) ∇w v = πe0 (dv(w)).

In particular, if we take v(x) = ei (x) (1 ≤ i ≤ n) for some orthonormal

frame field (e1 (x), . . . , en (x)) on Xn , then we have

∇w ei = πe0 (dei (w)) = ej ω̄ij (w),

where the repeated index j is summed from 1 to n and ω̄ji represents the
pullback of the Maurer-Cartan form ωji on F (Xn ) to Xn via the frame field
(e1 (x), . . . , en (x)). Moreover, the Leibniz rule for exterior derivatives im-
plies that for any vector field v(x) = v i (x)ei (x) on Xn , we have
∇w v = w(v i )ei (x) + v i ∇w ei (x)
(11.12) = w(v i )ei (x) + v i ej (x)ω̄ij (w)
 
= ei (x) w(v i ) + v j ω̄ji (w) .

The remarkable fact about equation (11.12) is that, even though we used
extrinsic objects to define the covariant derivative, the result is described
entirely in intrinsic terms: The tangent vector fields (ei (x)) and the pulled-
back Maurer-Cartan forms (ω̄ji ) are well-defined 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 defined on Xn .
The following exercise shows how the covariant derivative may be thought
of as an analog to the exterior derivative for vector fields on Xn .

*Exercise 11.14. Given a vector field v on Xn , consider the map

∇v : T Xn → T Xn
11.3. A more intrinsic view of Sn and Hn 347

defined by
(11.13) ∇v(w) = ∇w v.
Use the explicit formula (11.12) to show that ∇v is a T Xn -valued 1-form on
Xn (cf. Definition 2.18). Specifically, if (e1 (x), . . . , en (x)) is an orthonormal
frame field on Xn with associated connection forms (ω̄ji ), then
 
∇v = ei dv i + v j ω̄ji .
In particular, if v(x) = ei (x), then we have
(11.14) ∇ei (x) = ej (x) ω̄ij .
Thus, we can think of ∇ 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 fields v1 , v2 on Xn
and any vector w ∈ Tx Xn , we have
(11.15) w(v1 , v2 ) = ∇w v1 , v2  + v1 , ∇w v2 .
(You can think of this as a Leibniz rule for computing the directional deriv-
ative 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 ω̄ij = −ω̄ji for 1 ≤ i, j ≤ n, which
is true for the connection forms on both Sn and Hn .)

Definition 11.16. Let Γ(T Xn ) denote the space of smooth local sections
of T Xn (i.e., smooth vector fields on open sets in Xn ). The map
∇ : Γ(T Xn ) × Γ(T Xn ) → Γ(T Xn )
defined by
∇(v, w)(x) = ∇w(x) v ∈ Tx Xn
is called the Levi-Civita connection on Xn . The 1-forms (ω̄ji ) determined by
the frame field (e1 (x), . . . , en (x)) and equation (11.14) are called the con-
nection forms associated to ∇ and the frame field (e1 (x), . . . , en (x)). (Note
that this terminology is consistent with their definition as the pullbacks of
the connection forms on the frame bundle G → Xn via the frame field.)
Remark 11.17. Although the Levi-Civita connection is defined as an op-
erator on vector fields on Xn , equation (11.14) suggests—correctly!—that
there should be a related operator (also denoted ∇) on an appropriate class
of smooth maps from F (Xn ) to T Xn , determined by the condition
(11.16) ∇ei = ej ωij
348 11. Curves and surfaces in elliptic and hyperbolic spaces

together with an extension to appropriate maps ṽ : F (Xn ) → T Xn via

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 differences between the Levi-Civita operator ∇

and the exterior derivative d is that there is no analog of the identity d◦d = 0
for ∇, and so it is not immediately clear how to differentiate equation (11.16)
in order to obtain structure equations for the derivatives (dωji ). 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 α : I → X3 that maps some open

interval I ⊂ R into X3 . Since X3 has the structure of the homogeneous space
G/SO(3), an adapted frame field along α should be a lifting α̃ : I → G. Any
such lifting can be written as
α̃(t) = (e0 (t), e1 (t), e2 (t), e3 (t)),
where for each t ∈ I, e0 (t) = α(t) and (e1 (t), e2 (t), e3 (t)) is an oriented,
orthonormal basis for the tangent space Tα(t) X3 . Such an adapted frame
field is usually called an orthonormal frame field along α.
As in the Euclidean case, we say that α is regular if α (t) = 0 for every t ∈ I;
as usual, we will only consider regular curves. We begin our construction of
an adapted orthonormal frame field along α by setting
α (t)
e1 (t) = ;
|α (t)|
i.e., we require that e1 (t) be the unit tangent vector to the curve at α(t).
The same argument as in the Euclidean case shows that α can be smoothly
reparametrized by its arc length function s(t), so henceforth we will assume
that α = α(s) is parametrized by arc length and that
e1 (s) = α (s).
Note that this makes sense even in the intrinsic setting: Since α(s) ∈ X3 , the
derivative e1 (s) = α (s) is an element of Tα(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 different from the Euclidean
case: Since e1 (s) is a vector field along α, we have to use the covariant
derivative to differentiate it. In particular, differentiating any vector field
along the curve α means taking its covariant derivative with respect to the
unit tangent vector field along α. So the natural analog for the Euclidean
derivative e1 (s) is the covariant derivative ∇e1 (s) e1 (s). We will say that α is
nondegenerate if α is regular and, in addition, ∇e1 (s) e1 (s) = 0 for all s ∈ I.
*Exercise 11.18. In the Euclidean case, any regular curve α : I → E3 with
e1 (s) = α (s) = 0 for all s ∈ I is contained in a straight line in E3 . Consider
the analogous condition for curves in X3 : Let α : I → X3 be a regular curve
parametrized by arc length; let e0 (s) = α(s), e1 (s) = α (s), and suppose
that ∇e1 (s) e1 (s) = 0 for all s ∈ I.
(a) Use the extrinsic definition (11.11) together with the structure equations
(11.5) and (11.9) to show that, when regarded as functions e0 , e1 : I → V4 ,
we have
(11.17) e0 (s) = e1 (s), e1 (s) = k(s)e0 (s),
where k(s) = ω10 (e1 (s)).
(b) Use the fact that ω10 = ±ω01 = ±ω 1 (with the sign depending on whether
X3 = S3 or H3 ) to show that

−1, X3 = S3 ,
k(s) =
1, X3 = H3 .

(c) Solve the differential equations (11.17) and show that:

(1) If X3 = S3 , then there exist orthogonal unit vectors ē0 , ē1 ∈ E4

such that
(11.18) α(s) = e0 (s) = cos(s) ē0 + sin(s) ē1 .
In particular, α is contained in the great circle determined by the
intersection of S3 with the plane spanned by (ē0 , ē1 ).
(2) If X3 = H3 , then there exist orthogonal unit vectors ē0 , ē1 ∈ M1,3 ,
with ē0 timelike and ē1 spacelike, such that
(11.19) α(s) = e0 (s) = cosh(s) ē0 + sinh(s) ē1 .
In particular, α is contained in the “great hyperbola” determined by
the intersection of H3 with the timelike plane spanned by (ē0 , ē1 ).

The differential equation ∇α (s) α (s) = 0 is called the geodesic equation, and
the curves α 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 α : I → X3 is a nondegenerate curve parametrized by

arc length. According to equation (11.15), differentiating the equation
e1 (s), e1 (s) = 1
along α yields
∇e1 (s) e1 (s), e1 (s) = 0.
Thus, ∇e1 (s) e1 (s) is orthogonal to e1 (s), and we define
∇e1 (s) e1 (s)
e2 (s) = .
|∇e1 (s) e1 (s)|
This vector will be called the unit normal vector to the curve at α(s).
The adapted frame field 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) × e2 (s).
This makes sense because e1 (s) and e2 (s) are elements of the oriented, 3-di-
mensional Euclidean vector space Tα(s) X3 , where the cross product is well-
defined. The vector e3 (s) is called the binormal vector to the curve at α(s).
The adapted frame field (e1 (s), e2 (s), e3 (s)) is called the Frenet frame of the
curve α(s); it determines a canonical, left-invariant lifting α̃ : I → G given
by
α̃(s) = (e0 (s), e1 (s), e2 (s), e3 (s)),
where e0 (s) = α(s), for any nondegenerate curve α in X3 parametrized by
arc length.

*Exercise 11.19. Show that we have the following analog of the Frenet
equations for nondegenerate curves α : I → X3 parametrized by arc length:

(11.20) α (s) ∇e1 (s) e1 (s) ∇e1 (s) e2 (s) ∇e1 (s) e3 (s)
⎡ ⎤
0 0 0 0
⎢ ⎥
 ⎢1 0 −κ(s) 0 ⎥
= α(s) e1 (s) e2 (s) e3 (s) ⎢ ⎢ ⎥,
−τ ⎥
⎣ 0 κ(s) 0 (s) ⎦
0 0 τ (s) 0
where κ, τ : I → R are smooth functions along α with
κ(s) = |∇e1 (s) e1 (s)| > 0.
(Hint: Equation (11.14) might be helpful.) As in the Euclidean case, the
functions κ(s), τ (s) are called the curvature and torsion, respectively, of α.
11.5. Moving frames for surfaces in S3 and H3 351

Exercise 11.20. Let α : I → X3 ⊂ V4 be a nondegenerate curve parame-

trized by arc length.
(a) Show that, when regarded as functions e0 , e1 , e2 , e3 : I → V4 , we have
⎡ ⎤
0 ±1 0 0
⎢ ⎥
  ⎢1 0 −κ(s) 0 ⎥
e0 (s) e1 (s) e2 (s) e3 (s) = e0 (s) e1 (s) e2 (s) e3 (s) ⎢⎢0
⎥,
⎣ κ(s) 0 −τ (s)⎥ ⎦
0 0 τ (s) 0
with the sign depending on whether X3 = S3 or H3 .
(b) Show that there exist orthogonal unit vectors ē0 , ē1 , ē2 ∈ V4 such that α
is contained in the intersection of X3 with the plane spanned by (ē0 , ē1 , ē2 ) if
and only if the torsion τ (s) is identically zero. (This intersection is a “great
sphere” when X3 = S3 and a “great hyperboloid” 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 τ (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 → X3 be an immersion whose

image is a surface Σ = x(U ). Just as for curves, an adapted frame field
along Σ is a lifting x̃ : U → G of the form
x̃(u) = (e0 (u), e1 (u), e2 (u), e3 (u)) ,
where for each u ∈ 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 Σ =
x(U ) ⊂ X3 . (In particular, this means that none of the following construc-
tions relied on the flatness of E3 or involved differentiating vector fields
on E3 .)
(a) We can choose an orthonormal frame (e1 (u), e2 (u), e3 (u)) for each tan-
gent space Tx(u) X3 so that that e3 (u) is orthogonal to Tx(u) Σ and (e1 (u),
e2 (u)) span Tx(u) Σ. Any such choice defines a lifting x̃ : U → G by
x̃(u) = (e0 (u), e1 (u), e2 (u), e3 (u)) ,
where e0 (u) = x(u).
(b) Let (ω̄ i , ω̄ji ) denote the pullbacks of the Maurer-Cartan forms (ω i , ωji )
on G to U via x̃. Then we have ω̄ 3 = 0, and the 1-forms (ω̄ 1 , ω̄ 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 Σ = x(U ) ⊂ X3 , given

by the first fundamental form
I = (ω̄ 1 )2 + (ω̄ 2 )2 .

(d) Differentiating the equation ω̄ 3 = 0 and applying Cartan’s lemma implies

that there exist functions h11 , h12 , h22 on U such that
 3   
ω̄1 h11 h12 ω̄ 1
= .
ω̄23 h12 h22 ω̄ 2

Once again, the functions (hij ) are related to the derivative of the Gauss
map on Σ. However, there are two important differences:

(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 Σ to the unit sphere in E3 ; rather, it is a map from Σ to the
tangent bundle T X3 .
(2) In order to define the second fundamental form, we must use the
covariant derivative to differentiate the Gauss map.
Definition 11.22. Let U ⊂ R2 be an open set, and let x : U → X3 be an
immersion with image Σ = x(U ). The Gauss map of Σ = x(U ) is the map
N : Σ → T X3 defined by
N (x(u)) = e3 (u) ∈ Tx(u) X3 ,
where (e1 (u), e2 (u), e3 (u)) is any adapted frame field on Σ = x(U ).
Definition 11.23. Let U ⊂ R2 be an open set, and let x : U → X3 be
an immersion. The second fundamental form of Σ = x(U ) is the quadratic
form II on T U defined by
II(v) = −∇v e3 , dx(v)
for v ∈ Tu U , where (e1 (u), e2 (u), e3 (u)) is any adapted frame field on Σ =
x(U ).
*Exercise 11.24. Show that
II = ω̄13 ω̄ 1 + ω̄23 ω̄ 2 = h11 (ω̄ 1 )2 + 2h12 ω̄ 1 ω̄ 2 + h22 (ω̄ 2 )2 .

As in the Euclidean case, the eigenvalues κ1 (u), κ2 (u) of the matrix [hij (u)]
are called the principal curvatures of Σ at the point x(u), and the eigen-
vectors of the self-adjoint map −dNx(u) : Tx(u) Σ → Tx(u) Σ are called the
principal vectors or principal directions of Σ at the point x(u).
11.5. Moving frames for surfaces in S3 and H3 353

Definition 11.25. The functions

(11.21) K̄ = κ1 κ2 , H = 12 (κ1 + κ2 )
on U are called the extrinsic curvature and the mean curvature, respectively,
of Σ. The Gauss curvature K of Σ (also called the intrinsic curvature of
Σ) is the function on U defined by the condition that
dω̄21 = K ω̄ 1 ∧ ω̄ 2 ,
where (ω̄ i , ω̄ji ) are the Maurer-Cartan forms associated to any adapted frame
field on Σ.
*Exercise 11.26. (a) Show that the functions K, K̄, and H are all well-
defined on U , independent of the choice of adapted frame field.
(b) Use the structure equations (11.6) and (11.10) to show that:

(1) If X3 = S3 , then K = K̄ + 1.
(2) If X3 = H3 , then K = K̄ − 1.

So, unlike in the Euclidean case where the Gauss equation implies that
K = K̄, here these two notions of curvature differ by the sectional curvature
of the underlying homogeneous space.

As in the Euclidean case, the pullbacks to U of the structure equations for

the derivatives (dω̄ji ) 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
dω̄21 = ω̄13 ∧ ω̄23 ± ω̄ 1 ∧ ω̄ 2 ,
(11.22) dω̄13 = ω̄23 ∧ ω̄21 ,
dω̄23 = −ω̄13 ∧ ω̄21 ,
with the sign in the first equation depending on whether X3 = S3 or H3 .
*Exercise 11.28. Suppose that x : U → X3 is an immersion whose coordi-
nate curves are all principal curves. Then the first 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 field
1 1
e1 (u) = √ xu , e2 (u) = √ xv , e3 (u) = e1 (u) × e2 (u)
E G
354 11. Curves and surfaces in elliptic and hyperbolic spaces

has associated Maurer-Cartan forms

√ √
ω̄ 1 = E du, ω̄ 2 = G dv, ω̄ 3 = 0,
1
ω̄21 = √ (Ev du − Gu dv),
2 EG
e e
ω̄13 = ω̄ 1 = √ du,
E E
g g
ω̄23 = ω̄ 2 = √ dv.
G G

(b) Show that the Gauss equation is equivalent to

" # " # 
eg 1 Ev Gu
(11.23) ±1=− √ √ + √ ,
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
1 e g!
ev = Ev + ,
2 E G
(11.24)
1 e g!
g u = Gu +
2 E G
(cf. Exercises 4.24 and 4.41).

As for surfaces in E3 , the first and second fundamental forms are invariants
of the surface. Consequently, if two surfaces have different first 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 Bonnet’s theorem:
Theorem 11.29. Let (ω̄ 1 , ω̄ 2 , ω̄13 , ω̄23 ) be 1-forms on a connected and simply
connected open set U ⊂ R2 satisfying the conditions that (ω̄ 1 , ω̄ 2 ) are lin-
early independent at each point of U and that ω̄i3 is a scalar multiple of ω̄ i
for i = 1, 2. Suppose that, together with the Levi-Civita connection form ω̄21
determined by ω̄ 1 and ω̄ 2 , these forms satisfy the Gauss and Codazzi equa-
tions (11.22). Then there exists an immersed surface x : U → X3 , unique up
to transformation by an element of G, whose first and second fundamental
forms are
I = (ω̄ 1 )2 + (ω̄ 2 )2 ,
II = ω̄13 ω̄ 1 + ω̄23 ω̄ 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

Definition 11.30. A regular surface Σ = x(U ) ⊂ X3 is called totally geo-

desic if every geodesic in Σ is also a geodesic in X3 .

It turns out that Σ is totally geodesic if and only if

(11.25) ∇w e3 = 0 for all w ∈ T Σ,
where e3 is a unit normal vector field along Σ.
Remark 11.31. By way of comparison, for a surface Σ ⊂ E3 , the totally
geodesic condition says that every geodesic in Σ is a straight line in E3 ,
which implies that Σ is contained in a plane in E3 . Meanwhile, the condition
(11.25) says that the normal vector field e3 to Σ is constant along Σ, which
also implies that Σ is contained in a plane in E3 .
Exercise 11.32. Suppose that Σ = x(U ) ⊂ 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 field along Σ satisfy
(11.26) ω̄13 = ω̄23 = 0.

(b) Use the extrinsic definition (11.11) together with the structure equations
(11.5) and (11.9) to show that, when regarded as functions e0 , e1 , e2 , e3 :
U → V4 , equation (11.26) implies that
de3 = 0.
Conclude that there exist orthogonal unit vectors ē0 , ē1 , ē2 ∈ V4 such that
Σ = x(U ) is contained in the “great sphere” of S3 or the “great hyper-
boloid” of H3 determined by the intersection of X3 with the plane spanned
by (ē0 , ē1 , ē2 ) (cf. Exercise 11.20). This is the analog of the fact that any
totally geodesic surface in E3 is contained in a plane.

Definition 11.33. A surface Σ = x(U ) ⊂ X3 is called flat if its Gauss

curvature K is identically zero.
Exercise 11.34. Let Σ be a flat surface in S3 .
(a) Show that every point x0 ∈ Σ has a neighborhood for which there exists
an asymptotic parametrization x : U → S3 of Σ such that the first and
second fundamental forms of Σ = x(U ) are given by
I = dx2 + 2 cos(2ψ) dx dy + dy 2 ,
(11.27)
II = 2 sin(2ψ) dx dy,
where the function ψ : U → R, called the angle function, satisfies the wave
equation ψxy = 0. (Hint: Observe that κ1 κ2 = −1, and adapt the construc-
tion of §9.4.) Conversely, Theorem 11.29 implies that any solution ψ(x, y)
356 11. Curves and surfaces in elliptic and hyperbolic spaces

of the wave equation gives rise to a flat surface in S3 whose first and second
fundamental forms are given by (11.27).
(b) Let a, b > 0 with a2 +b2 = 1, and consider the map x : S1 ×S1 → S3 ⊂ E4
defined by
x(θ, ϕ) = t [a cos(θ), a sin(θ), −b cos(ϕ), b sin(ϕ)] ,
where θ and ϕ denote angle coordinates on the two copies of S1 . Show that
Σ = x(S1 × S1 ) is a flat torus in S3 , with angle function equal to a constant
ψ0 such that
cos(ψ0 ) = a, sin(ψ0 ) = b.
(Hint: Begin by considering the principal adapted frame field
e0 (θ, ϕ) = x(θ, ϕ) = t[a cos(θ), a sin(θ), −b cos(ϕ), b sin(ϕ)],
xθ (θ, ϕ)
e1 (θ, ϕ) = = t[− sin(θ), cos(θ), 0, 0],
|xθ (θ, ϕ)|
xϕ (θ, ϕ)
e2 (θ, ϕ) = = t[0, 0, sin(ϕ), cos(ϕ)],
|xϕ (θ, ϕ)|
e3 (θ, ϕ) = t[−b cos(θ), −b sin(θ), −a cos(ϕ), a sin(ϕ)]
and its associated Maurer-Cartan forms.)

Exercise 11.35. Let Σ be a flat surface in H3 , and assume that Σ has no

umbilic points (i.e., points where κ1 = κ2 ).
(a) Show that every point x ∈ Σ has a neighborhood for which there exists
a principal parametrization x : U → H3 of Σ such that the first and second
fundamental forms of Σ = x(U ) are given by
I = cosh2 (ψ) du2 + sinh2 (ψ) dv 2 ,
(11.28)
II = sinh(ψ) cosh(ψ)(du2 + dv 2 ),
where the angle function ψ : U → R satisfies Laplace’s equation ψuu + ψvv =
0. (Hint: Observe that κ1 κ2 = 1, and adapt the construction of §9.4.)
Conversely, Theorem 11.29 implies that any solution ψ(u, v) of Laplace’s
equation gives rise to a flat surface in H3 whose first and second fundamental
forms are given by (11.28).
(b) Let a, b > 0 with a2 − b2 = 1, and consider the map x : R × S1 → H3 ⊂
M1,3 defined by
x(t, ϕ) = t[a cosh(t), a sinh(t), b cos(ϕ), b sin(ϕ)] ,
where t denotes a standard coordinate on R and ϕ denotes an angle coor-
dinate on S1 . Show that Σ = x(R × S1 ) is a flat cylinder in H3 , with angle
11.6. Maple computations 357

function equal to a constant ψ0 such that

cosh(ψ0 ) = a, sinh(ψ0 ) = b.

(Hint: Begin by considering the principal adapted frame field

e0 (t, ϕ) = x(t, ϕ) = t[a cosh(t), a sinh(t), b cos(ϕ), b sin(ϕ)],

xt (t, ϕ)
e1 (t, ϕ) = = t[sinh(t), cosh(t), 0, 0],
|xt (t, ϕ)|
xϕ (t, ϕ)
e2 (t, ϕ) = = t[0, 0, − sin(ϕ), cos(ϕ)],
|xϕ (t, ϕ)|
e3 (t, ϕ) = t[−b cosh(t), −b sinh(t), −a cos(ϕ), −a sin(ϕ)]

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 different due to the
curvature terms. Here we will work through Exercise 11.34 regarding flat
surfaces in S3 ; only minor modifications are required for the computations
for Exercise 11.35 regarding flat 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(’omega[i,j] &ˆ omega[j]’, j=1..3);
end do;
358 11. Curves and surfaces in elliptic and hyperbolic spaces

> d(omega[1,2]):= -omega[1,3] &ˆ omega[3,2]

+ omega[1] &ˆ omega[2];
d(omega[3,1]):= -omega[3,2] &ˆ omega[2,1]
+ omega[3] &ˆ omega[1];
d(omega[3,2]):= -omega[3,1] &ˆ omega[1,2]
+ omega[3] &ˆ omega[2];
Now, suppose that x : U → S3 is a principal parametrization of a flat surface
Σ ⊂ S3 and that (e1 (u), e2 (u), e3 (u)) is a principal adapted frame field along
Σ. Since Σ is flat, the principal curvatures κ1 (u), κ2 (u) satisfy the condition
κ1 (u)κ2 (u) = −1.
The same argument as that in §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 ψ0 (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 dω̄13 and dω̄23 . Now check the Gauss equation:
> Simf(d(Simf(subs(adaptedsub, omega[1,2])))
+ Simf(subs(adaptedsub, omega[1,3] &ˆ omega[3,2]
- omega[1] &ˆ omega[2])));
(−ψ0v,v + ψ0u,u ) (d(v)) &ˆ (d(u))
So the Gauss equation is satisfied if and only if the function ψ0 (u, v) satisfies
the wave equation (ψ0 )uu − (ψ0 )vv = 0.
Now make the change to asymptotic coordinates
(u + v) (u − v)
x= , y= ,
2 2
or, equivalently,
u = x + y, v = x − y.

We’ll need to remove ω̄21 from the list in adaptedsub because Maple
won’t like this substitution in the expressions diff(psi0(u,v), u) and
11.6. Maple computations 359

diff(psi0(u,v), v). We’ll also introduce a new name for the function

ψ(x, y) = ψ0 (x + y, x − 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 doesn’t really know how to compute symmetric products of differ-
ential forms, but the following commands will work for computing the first
and second fundamental forms:
> collect(simplify(Simf(subs(adaptedsub asymp, omega[1]))ˆ2
+ Simf(subs(adaptedsub asymp, omega[2]))ˆ2), {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 field, as follows. First, declare a, b to be constants and
define 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 define the cor-
responding group element

g = e0 e1 e2 e3 ∈ SO(4)

and compute the matrix-valued Maurer-Cartan form

ω̄ = g −1 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

⎡ ⎤
d(θ)a d(φ)
⎢ 0 − −b 0 ⎥
⎢ a2 + b2 a2 + b2 ⎥
⎢ d(θ)a 0 0 −b d(θ)⎥
⎢ ⎥
connection matrix = ⎢
⎢b d(φ)
⎥
⎢ 0 0 a d(φ) ⎥
⎥
⎢ ⎥
⎣ d(θ) d(φ)a ⎦
0 b 2 2
− 2 2
0
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(ψ0 ) = a, sin(ψ0 )
= b, as desired.
 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 Σ of homogeneous spaces G/H. In this context,
the geometry of Σ is determined by the geometry of the ambient homoge-
neous space G/H and the particular way that Σ 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 geo-
metric structure of a nonhomogeneous manifold that is defined 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
Σ of some manifold M that is not homogeneous.
Recall that for a homogeneous space G/H, the natural projection map

π : G → 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 ∈ 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 ∈ 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 diffeomorphic to G, and G acts freely and transitively on
F (G/H).
When we replace the homogeneous space G/H with a more general n-
dimensional manifold M , the situation is a bit different. We can still as-
sociate to each point x ∈ M a collection of frames (x; e1 , . . . , en ) for the
tangent space Tx M , and the specific 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 ∈ 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 ∈ 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 different 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 fiber group H = O(n)
(cf. §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 Σ of M . Similar constructions can be applied
to manifolds with other types of geometry—e.g., Lorentzian, equi-affine, or
projective manifolds.

12.2. Orthonormal frames and connections on Riemannian

manifolds

Definition 12.1. A Riemannian manifold of dimension n is a smooth man-

ifold of dimension n, together with a smoothly varying metric g on M (cf.
Exercise 1.15). The metric g determines an inner product ·, · on each
tangent space Tx M defined by

v, w = g(v, w)

for v, w ∈ Tx M .
12.2. Orthonormal frames and connections 363

Exercise 12.2. The phrase “smoothly varying” in Definition 12.1 means

that for any local coordinate system 1 n
2  3 x = (x , . . . , x ) on M , the functions
gij (x) = g ∂xi , ∂xj | 1 ≤ i, j, ≤ n are smooth functions of x. Show that
∂ ∂

this condition is independent of the choice of local coordinates on M : If

x̄ = (x̄1 , . . . , x̄n ) is another local coordinate system on M with smooth local
coordinate transformation functions xi = xi (x̄1 , . . . , x̄n ), then the functions
" #
∂ ∂
ḡk (x̄) = g ,
∂ x̄k ∂ x̄
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 defined almost exactly as they were on En (cf. Definition
3.12); the only difference is that now x is simply a point of M and, in general,
not an element of a vector space.

Definition 12.3. An (oriented) orthonormal frame f on an oriented Rie-

mannian manifold M is a list f = (x; e1 , . . . , en ) where x ∈ M and (e1 , . . .,
en ) is an oriented, orthonormal basis for the tangent space Tx M . Alterna-
tively, 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 fiber SO(n), called the (oriented) orthonormal
frame bundle of M and denoted F (M ):

SO(n) - F (M )
π
?
M.

There are several important differences between the orthonormal frame bun-
dle of the homogeneous space En and that of a general Riemannian manifold
M . First, while each fiber of the frame bundle is acted on freely and tran-
sitively by SO(n), there is no larger group that acts transitively on the
entire frame bundle F (M ). But the most significant change is that, given
an orthonormal frame field (e1 (x), . . . , en (x)) on an open set in M , there
is no natural way of thinking of the frame vector fields (ei (x)) as functions
from M to a fixed vector space—not 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 ∈ M , the vectors (e1 (x), . . . , en (x)) take values in the vector
space Tx M .
In the case of En , we were able to differentiate vector fields by using the fact
that each tangent space Tx En is canonically isomorphic to En and regarding
the vector fields (e1 (x), . . . , en (x)) as functions into this fixed 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 define the covariant derivatives of vector fields. But for a more general
Riemannian manifold M , there is no such canonical isomorphism between
each tangent space Tx M and a fixed vector space En , or even a canonical
embedding of Tx M into a larger fixed vector space; indeed, there are infin-
itely many ways of identifying each tangent space Tx M with a fixed 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 field (e1 (x), . . . , en (x))
on an open set U ⊂ M , we can associate a local trivialization (cf. §1.5)
φ : T U → U × En of the tangent bundle T M by defining
   
(12.1) φ x, ai ei (x) = x, t[a1 , . . . , an ] .

(a) Show that the map φ defines an isometry between each tangent space
Tx M and the vector space En .
(b) We may regard a local trivialization φ as a local choice of basis vector
fields (e1 (x), . . ., en (x)) for sections of T M , and the identification (12.1)
makes it tempting to think that we might be able to regard these vector
fields as “constant” for purposes of differentiation. But what happens when
we choose a different basis? Let (ẽ1 (x), . . . , ẽn (x)) be any other orthonormal
frame field on U , related to the original frame field by
 
(12.2) ẽ1 (x) . . . ẽn (x) = e1 (x) . . . en (x) Ā(x),
where Ā(x) is an SO(n)-valued function on U . Show that under the analo-
gous local trivialization φ̃ associated to the orthonormal frame field (ẽ1 (x),
. . ., ẽn (x)), the vector fields (e1 (x), . . . , en (x)) are identified with the col-
 −1
umns of the matrix Ā(x) . In particular, vector fields that appear “con-
stant” with respect to one trivialization do not necessarily remain “constant”
with respect to a different trivialization.

This exercises raises a crucial question: If there is no canonical trivializa-

tion of T M , and hence no consistent notion of a “constant” vector field on
M , then how can we possibly differentiate vector fields on M ? (This is a
special case of the more general question of how to differentiate 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 differentiation for vector fields on M , we need to introduce an additional
structure, called a connection, on the tangent bundle Tx M .

Definition 12.5. Let Γ(T M ) denote the space of smooth local sections of
T M (i.e., smooth vector fields on open sets in M ). An affine connection (or,
more succinctly, a connection) ∇ on T M is a map
∇ : Γ(T M ) × Γ(T M ) → Γ(T M ),
with ∇(w, v) denoted by ∇w v, such that for any vector fields v, v1 , v2 , w,
w1 , w2 ∈ Γ(T M ), any smooth, real-valued functions f, g on M , and any
real numbers a, b ∈ R, we have

(1) ∇f w1 +gw2 v = f ∇w1 v + g∇w2 v;

(2) ∇w (av1 + bv2 ) = a∇w v1 + b∇w v2 ;
(3) ∇w (f v) = w(f )v + f ∇w v.

Remark 12.6. The Levi-Civita connection of Definition 11.16 is, of course,

a connection according to this definition. It also has certain additional
properties, which we will discuss shortly.

Conditions (1) and (2) are linearity properties: They say that the map ∇ is
linear over smooth functions in its first input and linear over real numbers in
its second input. Condition (3) is an analog of the Leibniz rule that describes
how ∇ behaves when its second input is multiplied by a smooth function.
The vector field ∇w v should be regarded as defining a sort of “directional
derivative” of the vector field v in the direction of w, and conditions (1)–(3)
are precisely the conditions that such a directional derivative must satisfy.

Remark 12.7. This definition applies more generally to any smooth vector
bundle B over a manifold M : If π : B → M is a vector bundle with fibers
isomorphic to a fixed k-dimensional vector space V (cf. §1.5), then an affine
connection ∇ on B is a map
∇ : Γ(T M ) × Γ(B) → Γ(B)
satisfying the properties of Definition 12.5. This more general setting il-
lustrates the fact that the two copies of Γ(T M ) used for the inputs of ∇
in Definition 12.5 play significantly different roles: The first input is the
direction along which differentiation should occur (which must be a tangent
vector to M ), and the second input is the object to be differentiated. The
output is the resulting differentiated 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 fields on En .

By taking advantage of the usual canonical identification of each tangent
space Tx En with the vector space En , we may regard these vector fields as
smooth functions v, w : En → En . Having done so, define
∇w v = dv(w).

(a) Show that ∇ 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 ∇w v is a vector field on En —
in fact, the components of ∇w v are obtained by computing the directional
derivatives of the components of v in the direction w.)
(b) The canonical isomorphism Tx En ∼
= En allows us to write the tangent
bundle T E as the product manifold
n

(12.3) T En ∼
= En × En ,
where the first factor represents the base manifold En and the second factor
represents the fibers 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 ) ∼
= T En × T En
in the obvious way. The vector field v is a section of T En , and via the
identification (12.3), we can write it as a function σ : En → En × En defined
by
σ(x) = (x, v(x)) .
Show that ∇w v is given by the composition
(12.5) ∇w v = π2 ◦ dσ(w),
where dσ : T En → T (T En ) is the differential of the map σ (cf. §1.3) and
π2 : T En × T En → T En is the projection onto the second factor. This
connection is called the flat 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) ∈ 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 different
ways. The n-dimensional subspace
V(x,v) = T(x,v) (Tx M ) ⊂ T(x,v) (T M )
(corresponding to the second factor in (12.4)) is canonically defined: It is
the tangent space to the fiber Tx M ⊂ 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 first factor in (12.4) (called the horizontal tangent
space); in fact, any n-dimensional subspace H(x,v) ⊂ T(x,v) (T M ) for which
H(x,v) ∩ 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 ∇ on T M determines a projection map

π2 : T(x,v) (T M ) → V(x,v)

as in equation (12.5), and this in turn determines the n-dimensional hor-

izontal subspace H(x,v) = ker(π2 ) ⊂ T(x,v) (T M ) and the corresponding
decomposition
T(x,v) (T M ) = H(x,v) ⊕ V(x,v) .
Conversely, a choice of such a decomposition for each (x, v) ∈ T M (subject
to certain consistency conditions implied by Definition 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 flat connection ∇ on T En .

As a consequence of properties (1)–(3) of Definition 12.5, a connection ∇ is

completely determined by its action on any given orthonormal frame field
on M . Given an orthonormal frame field (e1 (x), . . . , en (x)) on M , the con-
nection ∇ determines scalar-valued 1-forms (ω̄ji ), with 1 ≤ i, j ≤ n, on M ,
called the connection forms associated to this frame field, defined by the
condition that for any w ∈ Tx M ,

(12.6) (∇w ei ) (x) = ej (x) ω̄ij (w).

The following exercise shows how a connection on T M may be thought of

as an analog to the exterior derivative for vector fields on M (cf. Exercise
11.14).

*Exercise 12.10. Given a connection ∇ on T M and a vector field v on

M , consider the map
∇v : T M → T M
defined by

(12.7) ∇v(w) = ∇w v.

(a) Use the linearity properties of Definition 12.5 to show that ∇v is a

T M -valued 1-form on M (cf. Definition 2.18).
368 12. Moving frames on Riemannian manifolds

(b) Show that if (e1 (x), . . . , en (x)) is an orthonormal frame field on an open
set U ⊂ M , then ∇ei is the T M -valued 1-form

(12.8) ∇ei = ej ω̄ij ,

where (ω̄ij ) are the scalar-valued 1-forms on M defined by equation (12.6).

Thus, we can think of ∇ as a generalization of the exterior derivative d that
appears in the structure equations (3.1).

*Exercise 12.11. Let ∇ be a connection on T M , and let (e1 (x), . . . , en (x))

be an orthonormal frame field on an open set U ⊂ M , with associated
connection forms (ω̄ji ) on U . (Think of the frame field as determining a
local trivialization of T M , as in Exercise 12.4.)
: i
(a) Show that for any vector field v(x) = v (x)ei (x) on U ,

(12.9) (∇v) (x) = ei (x)(dv i + v j ω̄ji ).

(Note that, while the exterior derivative of a vector field on M is not well-
defined, 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

∇ = d + ω̄,

where ω̄ is the matrix of 1-forms ω̄ = [ω̄ji ]. The notation means that if

: i
the vector field v(x) = v (x)ei (x) is expressed as the column vector
v̄(x) = t[v 1 (x), . . . , v n (x)], then the vector field ∇v should be expressed as
the column vector dv̄ + ω̄v̄, as indicated by equation (12.9).
(b) Let (ẽ1 (x), . . . , ẽn (x)) be any other orthonormal frame field on U , related
to the original frame field as in equation (12.2). Use equation (12.8) to show
that the matrix ω̄ ˜ = [ω̄ ˜ ji ] of connection forms associated to the new frame
field is given by

(12.10) ˜ = Ā−1 dĀ + Ā−1 ω̄ Ā.

ω̄

Remark 12.12. Connections on fiber bundles play an important role in

theoretical physics, particularly in field 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 field, and equation (12.10) describes how the gauge field transforms
under a gauge transformation.
12.2. Orthonormal frames and connections 369

In addition to the connection forms (ω̄ji ), we can associate to an orthonormal

frame field (e1 (x), . . . , en (x)) on M the dual forms (ω̄ 1 , . . . , ω̄ n ). These are
defined in essentially the same way as for orthonormal frame fields on En :
Definition 12.13. Let M be a Riemannian manifold, and let (e1 (x), . . .,
en (x)) be an orthonormal frame field on an open set U ⊂ M . The dual
forms (ω̄ 1 , . . . , ω̄ n ) associated to this frame field are the unique scalar-valued
1-forms on U defined by the property that

i i 1, i = j,
ω̄ (ej ) = δj =
0, i = j
(cf. (3.7)).
*Exercise 12.14. (a) Show that the dual forms (ω̄ 1 , . . . , ω̄ n ) associated to
an orthonormal frame field (e1 (x), . . . , en (x)) form a basis for the 1-forms
on U .
(b) Let (ẽ1 (x), . . . , ẽn (x)) be any other orthonormal frame field on U , related
to the original frame field as in equation (12.2). Show that the dual forms
˜ 1 , . . . , ω̄
(ω̄ ˜ n ) associated to the new frame field are given by
⎡ 1⎤ ⎡ 1⎤
˜
ω̄ ω̄
⎢ .. ⎥ −1 ⎢ .. ⎥
(12.11) ⎣ . ⎦ = Ā ⎣ . ⎦ .
˜n
ω̄ ω̄ n

As you might have guessed from the notation, the dual forms and connection
forms (ω̄ i , ω̄ji ) on M associated to an orthonormal frame field (e1 (x), . . .,
en (x)) on M are the pullbacks to M of certain 1-forms (ω i , ωji ) 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 defined. (It might
be helpful to review §3.3.2 at this point, particularly the derivations of equa-
tions (3.3) and (3.4).)
*Exercise 12.15. Let (e1 (x), . . ., en (x)) be a local orthonormal frame field
on an open set U ⊂ M , with associated dual and connection forms (ω̄ i , ω̄ji ).
Just as this orthonormal frame field determines a local trivialization of T M
(cf. Exercise 12.4), it can also be used to define a local trivialization of
the principal bundle F (M ): For any x ∈ U and any orthonormal frame
f = (x; e1 , . . . , en ) based at x, we can write
 
(12.12) e1 · · · en = e1 · · · en A
370 12. Moving frames on Riemannian manifolds

for some unique matrix A = [aij ] ∈ SO(n). Define a map φ : F (U ) →

U × SO(n) by
φ (x; e1 , . . . , en ) = (x, A) ,
where A is the matrix defined by equation (12.12).
(a) Show that the map φ defines a diffeomorphism between F (U ) = π −1 (U )
⊂ F (M ) and U × SO(n).
(b) Let (ẽ1 (x), . . . , ẽn (x)) be any other orthonormal frame field on U , related
to the original frame field as in equation (12.2). Show that the analogous
local trivialization φ̃ associated to the orthonormal frame field (ẽ1 (x), . . .,
ẽn (x)) is given by
!  −1 !
(12.13) φ̃ (x; e1 , . . . , en ) = x, Ã = x, Ā(x) A .

(c) Define 1-forms (ω i , ωji ) on F (U ) by

⎡ 1⎤ ⎡ 1⎤
ω ω̄
⎢ .. ⎥ −1 ⎢ .. ⎥
⎣ . ⎦ = A ⎣ . ⎦,
(12.14) ωn ω̄ n

ω = [ωji ] = A−1 dA + A−1 ω̄A.

Note the distinction between these 1-forms and those in equations (12.10)
and (12.11): In those equations, Ā is an SO(n)-valued function on the open
set U ⊂ M , whereas in equations (12.14), the entries of A represent lo-
cal coordinates on SO(n), independent of the local coordinates on M . So,
while the 1-forms (ω̄ i , ω̄ji ) are 1-forms on M and the (ω̄ji ) are linear combi-
nations of the (ω̄ i ), the 1-forms (ω i , ωji ) 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-defined, independent of
the choice of orthonormal frame field (e1 (x), . . . , en (x)) used to define the
1-forms (ω̄ i , ω̄ji ) and the local trivialization φ 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 field on an open set U ⊂ M , any n × n matrix ω̄ of
1-forms on M can be used to define a connection on T U via the equation
(12.8). But some connections have nicer properties than others; the following
definition describes two properties that are often considered desirable.
12.3. The Levi-Civita connection 371

Definition 12.16. An affine connection ∇ on T M is called

(1) torsion-free or symmetric if for any vector fields v, w on M , we

have
∇v w − ∇w v = [v, w],
where [v, w] denotes the usual Lie bracket of vector fields (cf. §1.4);

(2) compatible with the metric g on M if for any vector fields v, w on

M , we have

dv, w = ∇v, w + v, ∇w.

(This condition is often written as ∇g = 0.)

A given connection ∇ on T M may have one, both, or neither of these prop-

erties. 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 ∇ on T M that is both torsion-free and compat-
ible with the metric. This connection is called the Levi-Civita connection
on T M .

Remark 12.18. The existence of a canonical “nice” 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-affine, projective), there is often no single “best” 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 field
on an open set U ⊂ M , with associated dual forms (ω̄ 1 , . . . , ω̄ n ). (At this
point, we will drop the underscore notation on the vector fields ei (x) since
we no longer need to think of them as “basis” vector fields.) Let ∇ be an
affine connection on M , and let (ω̄ji ) be the corresponding connection forms
associated to the given frame field. 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

(12.15) ω̄ji = aijk ω̄ k

for some functions (aijk ) on U .

*Exercise 12.19. Suppose that ∇ is compatible with the metric.

(a) Show that the connection forms must satisfy the condition ω̄ij = −ω̄ji ;
i.e., the connection forms are skew-symmetric in their indices. (Hint: Dif-
ferentiate the equations ei , ej  = δij .)
(b) Conclude that if ∇ is compatible with the metric, then the functions
(aijk ) must satisfy

(12.16) aijk = −ajik .

*Exercise 12.20. Suppose that ∇ is torsion-free, and let ckij (x) = −ckji (x)
be the functions on U defined by the Lie bracket relations

[ei , ej ] = ckij ek .

(a) Show that for each i, j, k,

ω̄jk (ei ) − ω̄ik (ej ) = ckij .

(Hint: Apply the torsion-free condition with v = ei , w = ej .)

(b) Conclude that if ∇ is torsion-free, then the functions (aijk ) must satisfy

(12.17) akji − akij = ckij .

*Exercise 12.21. Now, suppose that ∇ 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 − cijk − ckij .

(Hint: This is an exercise in index juggling. Start with akij and apply equa-
tions (12.16) and (12.17) alternately until you come back to the index ar-
rangement that you started with. And keep in mind that ckij = −ckji !)
(b) Conclude that Theorem 12.17 is true and that the Levi-Civita connection
is defined by equation (12.15), with (akij ) as in equation (12.18).

From now on, we will assume that ∇ is the Levi-Civita connection on T M .

12.4. The structure equations 373

12.4. The structure equations

Let (ω i , ωji ) 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 first need to compute the structure equations of these
1-forms. The process is analogous to that of §3.3.2, but it is complicated
by the fact that we must use the connection ∇ rather than the exterior
derivative d to differentiate vector-valued quantities.
As in the case of En , we can define projection maps x : F (M ) → M and
ei : F (M ) → T M in the obvious way: If f = (x; e1 , . . . , en ) ∈ F (M ), then
x(f ) = x ∈ M,
ei (f ) = (x, ei ) ∈ Tx M.
The differentials of (x, ei ) are maps
dx : Tf F (M ) → Tx M,
dei : Tf F (M ) → 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 defined
precisely so that
(12.19) dx = ei ω i .

The maps (dei ) are a bit more complicated. For each f ∈ F (M ), the image
of (dei )f takes values in the 2n-dimensional tangent space T(x,ei ) (T M ). The
Levi-Civita connection ∇ determines a linear projection operator
π2 : T(x,ei ) (T M ) → T(x,ei ) (Tx M ) ∼
= Tx M,
defined by the condition that
(12.20) (π2 ◦ d)(ei ) = ∇ei = ej ωij
(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 (ω i , ωji ), we will
first need to differentiate 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 ∇ to differentiate them. And in general, it is not necessarily
true that ∇ ◦ d = 0, so we cannot directly apply any obvious analog of
the identity d ◦ d = 0. Fortunately, we can get around this problem by
considering two different 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 ⊂ M , we can write
∂
(12.21) dx = dxi ,
∂xi
 
where ∂x∂ 1 , . . . , ∂x∂n are the coordinate vector fields on U .
 
(a) Show that the coordinate vector fields ∂x∂ 1 , . . . , ∂x∂n have pairwise Lie
∂ ∂

brackets equal to zero; i.e., ∂x i , ∂xj = 0 (cf. Exercise 1.34).
(b) Apply ∇ to equation (12.21) and use an argument similar to that of
Exercise 12.11, part (a), to show that
" #
∂ ∂
∇ (dx) = ∇ i
∧ dxi + i d(dxi ).
∂x ∂x
Conclude that
" #
∂
(12.22) ∇ (dx) = ∇ ∧ dxi .
∂xi
 
(c) Use the definition (12.7) for the T M -valued 1-form ∇ ∂
∂xi
to show that
" # " #
∂ ∂
(12.23) ∇ =∇ ∂ dxj .
∂xi ∂xj ∂xi

(d) Use equations (12.22) and (12.23) to show that

" " # " ##
∂ ∂
∇ (dx) = ∇ ∂ −∇ ∂ dxi ∧ dxj .
∂xi ∂xj ∂xj ∂xi
i<j

(e) Conclude from part (a) and the fact that the Levi-Civita connection is
torsion-free that ∇ (dx) = 0.
*Exercise 12.23. Now that we know that ∇ (dx) = 0, we can differentiate
equation (12.19) to obtain the structure equations for the dual forms.
(a) Apply ∇ to equation (12.19) to show that
(12.24) ∇ei ∧ ω i + ei dω i = 0.

(b) Apply equation (12.20) (and a bit of index juggling) to equation (12.24)
to obtain  
ei dω i + ωji ∧ ω j = 0.

(c) Use the linear independence of the vectors (e1 , . . . , en ) in each tangent
space Tx M to conclude that the dual forms (ω 1 , . . . , ω n ) satisfy the structure
equations
(12.25) dω i = −ωji ∧ ω j .
12.4. The structure equations 375

*Exercise 12.24. In this exercise, we will show that the structure equations
(12.25) can be used to compute the connection forms directly from the dual
forms associated to any orthonormal frame field on M . To this end, let
(e1 (x), . . . , en (x)) be an orthonormal frame field on an open set U ⊂ M ,
with associated dual forms (ω̄ 1 , . . . , ω̄ n ).
(a) Show that if there exist 1-forms (ω̄ji ) on U that satisfy the metric com-
patibility condition ω̄ij = −ω̄ji and the equations

(12.26) dω̄ i = −ω̄ji ∧ ω̄ j ,

then the 1-forms (ω i , ωji ) on F (U ) defined by equations (12.14) satisfy the

structure equations (12.25).
(b) Show that if there exist 1-forms (ω̄ji ) on U that satisfy the condition
ω̄ji = −ω̄ij and equations (12.26), then they are unique. (Hint: Suppose that
there are two such sets of 1-forms (ω̄ji ), (ω̄
˜ ji ). Subtract the two equations,
and then use Cartan’s lemma and the skew-symmetry of the (ω̄ji ) to show
that ω̄˜ ji = ω̄ji .)

(c) In order to show that there exist such 1-forms (ω̄ji ), let ckij = −ckji be the
functions on U defined by the conditions
(12.27) dω̄ k = −ckij ω̄ i ∧ ω̄ j .

Suppose that the desired 1-forms (ω̄ji ) exist; then we must have

ω̄ji = aijk ω̄ k

for some functions (aijk ) on U . Since ω̄ji = −ω̄ij , the functions (aijk ) must
satisfy the conditions
aijk = −ajik .
Show that the structure equations (12.26) are equivalent to the equations
akji − akij = ckij .
Observe that these are precisely the same conditions as equations (12.16),
(12.17), and conclude that the functions (akij ) defined by equation (12.18)
produce the desired connection forms (ω̄ji ).

Our next task is to compute the structure equations for the connection
forms (ωji ). Differentiating equation (12.20) as we did for the analogous
equation in the Euclidean case turns out to be impractical: We would have
to use the connection ∇ to differentiate the equation, and there is no useful
identity for ∇ ◦ ∇ that would allow us to differentiate the left-hand side
376 12. Moving frames on Riemannian manifolds

effectively. Instead, we will differentiate the structure equations (12.25);

these are scalar-valued equations and can be differentiated with the usual
exterior derivative d.
*Exercise 12.25. Differentiate equation (12.25) (and do some index jug-
gling) to show that for i = 1, . . . , n,
(12.28) (dωji + ωki ∧ ωjk ) ∧ ω j = 0.
Definition 12.26. Given any affine connection ∇ on T M , the matrix-
valued 2-form
(12.29) Ω = [Ωij ] = [dωji + ωki ∧ ωjk ]
on F (M ) is called the curvature tensor of ∇. When ∇ is the Levi-Civita
connection of a Riemannian manifold M , Ω is also called the Riemann cur-
vature tensor of M .
*Exercise 12.27. Let (e1 (x), . . . , en (x)) be an orthonormal frame field on
an open set U ⊂ M , with ω̄ = [ω̄ji ] the matrix of associated connection
forms on U . Let (ẽ1 (x), . . . , ẽn (x)) be any other orthonormal frame field
on U , related to the original frame field as in equation (12.2). Equation
(12.10) shows how the matrix ω̄ ˜ ji ] of connection forms associated to
˜ = [ω̄
the new frame field is related to ω̄; in particular, the presence of the exterior
derivative dĀ indicates that the connection matrix ω̄ is not a tensor on
U because the transformation rule (12.10) is not an algebraic function of
the components of Ā. Use equations (12.10) and (12.29) to show that, by
contrast, the pullbacks Ω̄, Ω̄˜ of Ω to U via the two frame fields are related
by the equation
(12.30) ˜ = Ā−1 Ω̄Ā.
Ω̄
This transformation rule indicates that, unlike the matrix of connection
forms ω, the curvature tensor Ω is, in fact, a tensor on U .

Now, your instinctive reaction is probably to apply Cartan’s lemma to equa-

tion (12.28). But Cartan’s lemma only applies to equations of the form
φj ∧ ω j = 0,
where φ1 , . . . , φn are 1-forms (cf. Lemma 2.49), whereas equation (12.28)
has the form
Ωij ∧ ω j = 0,
where Ωi1 , . . . , Ωin are 2-forms. So Cartan’s lemma cannot be applied directly,
and we will need to explore the implications of equation (12.28) via direct
computation.
12.4. The structure equations 377

*Exercise 12.28. Let Ωij = dωji + ωki ∧ ωjk .

(a) Use the fact that ωji = −ωij to show that Ωij = −Ωji .
(b) Show that we can write

(12.31) Ωij = 12 Rjk

i
ω k ∧ ω  + Sjm
ik
ωk ∧ ω m + Tjn ωk ∧ ωm
ikm  n

i ), (S ik ), (T ikm ) on F (M ) that satisfy the symmetry

for some functions (Rjk jm jn
conditions
j
i
Rjk = −Rik = −Rjk
i
,
jk
(12.32) ik
Sjm = −Sim = −Sjkm
i
,
jkm
ikm
Tjn = −Tin = −Tjkn
im
= −Tjm
ikn
= −Tjn
imk
.

(Hint: Ωij is a 2-form on F (M ), and the 1-forms (ω i , ωji ) are a basis for the
1-forms on F (M ).)
(c) Equation (12.28) can now be written as
!
1 i
2 Rjk ω
k
∧ ω  + Sjm
ik
ωk ∧ ω m + Tjn ωk ∧ ωm
ikm  n
∧ ω j = 0.

By considering the various types of 3-forms appearing in this expression,

ikm = 0 and that
show that Tjn
ik ik
(12.33) Sjm = Smj ,
i i i
(12.34) Rjk + Rkj + Rjk = 0.

(d) Use equations (12.32) and (12.33) to show that Sjm ik = 0. (Hint: This

is an exercise in index juggling, very similar to that in Exercise 12.21.)

(e) Conclude that the Riemann curvature tensor (cf. Definition 12.26) has
the form

Ω = Ωij = 12 Rjk
i
ωk ∧ ω ,
i ) satisfy the skew-symmetry relations (12.32) and
where the functions (Rjk
the condition (12.34). Equivalently, the connection forms (ωji ) satisfy the
structure equations
(12.35) dωji = −ωki ∧ ωjk + 12 Rjk
i
ωk ∧ ω.
Equation (12.34) is known as the first Bianchi identity, and together with
i
the other symmetries, it implies that the functions Rjk satisfy the additional
relation
i k
Rjk = Rij .
378 12. Moving frames on Riemannian manifolds

Remark 12.29. The reason for including the factor of 12 in equations (12.31)
and (12.35) is so that the curvature 2-forms (Ωij ) may be written as
(12.36) Ωij = 12 Rjk
i
ωk ∧ ω = i
Rjk ωk ∧ ω.
k<
In particular, when n = 2, the Riemann curvature tensor has a single non-
1 , which is equal to the Gauss curvature K.
trivial component R212

Exercise 12.30. The Nash embedding theorem [Nas56] states that any
n-dimensional Riemannian manifold M can be isometrically embedded into
a Euclidean space En+m of sufficiently large dimension. The structure equa-
tions for the Maurer-Cartan forms on M can then be derived from those on
En+m , as follows.
Suppose that M ⊂ En+m is a smooth submanifold of En+m . Choose an
orthonormal frame field (e1 (x), . . ., en+m (x)) for Tx En+m along M so
that for each x ∈ M , the tangent space Tx M is spanned by the vectors
(e1 (x), . . . , en (x)). Let (ω̄ α , ω̄βα ), where 1 ≤ α, β ≤ (n + m), denote the
Maurer-Cartan forms associated to this frame field on M .
(a) Show that
(12.37) ω̄ n+1 = · · · = ω̄ n+m = 0
(cf. Proposition 4.18).
(b) Differentiate equations (12.37) and use Cartan’s lemma to show that
there exist functions (haij ) on M , with 1 ≤ i ≤ n, (n + 1) ≤ a ≤ (n + m),
and haij = haji , such that
(12.38) ω̄ia = haij ω̄ j .
These are the coefficients of the second fundamental form
II = haij ea ⊗ ω̄ i ω̄ j
of M ⊂ En+m (cf. Exercise 4.27).
(c) We can define a connection ∇ on M in much the same way that we
did for Sn and Hn (cf. §11.3), as follows: For any vector field v on M
and any vector w ∈ Tx M , define the covariant derivative of v with respect
to w to be the vector ∇w v ∈ Tx M given by the orthogonal projection of
the Euclidean directional derivative dv(w) in En+m onto the tangent plane
Tx M ⊂ Tx En+m ∼ = En+m . Then let ∇v : T M → T M be the T M -valued
1-form on M defined by
∇v(w) = ∇w v,
as in Exercise 12.10. Show that for 1 ≤ i ≤ n,
∇ei (x) = ej (x) ω̄ij ,
12.5. Moving frames for curves in Riemannian manifolds 379

where the sum ranges over 1 ≤ j ≤ n. Conclude that the 1-forms (ω̄ji )
with 1 ≤ i, j ≤ n are the connection forms associated to the connection ∇
and the orthonormal frame field (e1 (x), . . . , en (x)) on M . Use the fact that
ω̄ji = −ω̄ij and the structure equations for the (dω̄ i ) on En+m to argue that
∇ is, in fact, the Levi-Civita connection on M .
(d) Substitute the result of part (b) into the structure equations
dω̄ji = −ω̄αi ∧ ω̄jα , 1 ≤ i, j ≤ n,
on En+m (where the sum ranges over 1 ≤ α ≤ (n + m)), and compare with
the structure equations (12.35) to obtain the Gauss equations
n+m
 a a 
(12.39) i
Rjk = hik hj − hai hajk .
a=n+1
These equations are the higher-dimensional analog of the Gauss equation
(4.11) for surfaces in E3 .
Similarly, the structure equations for (dω̄ia ) with 1 ≤ i ≤ n, (n + 1) ≤ a ≤
(n + m) are the higher-dimensional analog of the Codazzi equations (4.12),
while the structure equations for (dω̄ba ) with (n + 1) ≤ a, b ≤ (n + m)} are
called the Ricci equations. (The Ricci equations have no analog for surfaces
in E3 because they only appear for embeddings of codmension greater than
or equal to 2.)

The remainder of this chapter will be devoted to the study of curves and
surfaces in a 3-dimensional Riemannian manifold M .

12.5. Moving frames for curves in 3-dimensional Riemannian

manifolds

Let M be a 3-dimensional Riemannian manifold, and consider a smooth,

parametrized curve α : I → M that maps some open interval I ⊂ R into
M . An adapted frame field along α should be a lifting α̃ : I → F (M ). Any
such lifting can be written as
α̃(t) = (α(t); e1 (t), e2 (t), e3 (t)),
where for each t ∈ I, (e1 (t), e2 (t), e3 (t)) is an oriented, orthonormal basis
for the tangent space Tα(t) M . Such an adapted frame field is usually called
an orthonormal frame field along α.
The construction of a canonical adapted frame field along α is very similar
to that for curves in S3 and H3 (cf. §11.4), in that we must use the covariant
derivative to differentiate vector fields along α. As in the Euclidean case,
we say that α is regular if α (t) = 0 for every t ∈ I; as usual, we will only
380 12. Moving frames on Riemannian manifolds

consider regular curves, and we will assume that α = α(s) is parametrized by

arc length. In addition, we will say that a regular curve α is nondegenerate
if ∇α (s) α (s) = 0 for all s ∈ I.

*Exercise 12.31. A regular curve α : I → M is called a geodesic in M if

∇α (t) α (t) = 0 for all t ∈ I, where ∇ denotes the Levi-Civita connection on
M (cf. Exercise 11.18). In this exercise, we will show that the geodesics on
M may be described in terms of the integral curves of a vector field called
the geodesic spray on the tangent bundle T M .
(a) Recall (cf. Remark 12.9) that the connection ∇ determines a decompo-
sition

(12.40) T(x,v) (T M ) = H(x,v) ⊕ V(x,v)

of each tangent space T(x,v) (T M ) into horizontal and vertical tangent spaces,
where V(x,v) = T(x,v) (Tx M ) ⊂ T(x,v) (T M ) is canonically defined and the
complementary subspace H(x,v) is determined by ∇.
Let π : T M → M denote the standard projection operator from T M to
M , and let dπ : T(x,v) (T M ) → Tx M denote its differential at the point
(x, v) ∈ T M . Show that for each vector v ∈ Tx M , there exists a unique
vector v̂ ∈ H(x,v) such that dπ(v̂) = v.
(b) Any regular curve α : I → M has a unique lifting to a curve α :
I → T M defined by the condition that for each t ∈ I, α (t) ∈ Tα(t) is the
tangent vector to α at the point α(t). This curve, in turn, has a unique
lifting to a curve (α ) : I → T (T M ) defined analogously: For each t ∈ I,
(α ) (t) ∈ Tα (t) (T M ) is the tangent vector to α at the point α (t) ∈ T M .
Show that under the decomposition (12.40) of the tangent space Tα (t) (T M ),
we have

(α ) (t) = α  (t) + ∇  α (t),
α (t)

where α   (t) ∈ H  
α (t) and ∇α (t) α (t) ∈ Vα (t) . (Hint: It follows from the
discussion in Remark 12.9 that ∇α (t) α (t) is the projection of (α ) (t) onto
the vertical tangent space Vα (t) , so you only need to show that the projection
of (α ) (t) onto the horizontal tangent space Hα (t) is equal to α  (t).)

(c) Conclude from part (b) that α is a geodesic if and only if (α ) (t) ∈ Hα (t)
for all t ∈ I.
(d) A vector field w on T M is called horizontal if w(x, v) ∈ H(x,v) for all
(x, v) ∈ T M . The horizontal vector field

w(x, v) = v̂
12.6. Moving frames for surfaces in Riemannian manifolds 381

(where v̂ is as in part (a)) is called the geodesic spray on T M . Show that

a curve γ : I → T M is an integral curve of w (i.e., γ  (t) = w(γ(t)) for all
t ∈ I) if and only if the curve α = π ◦ γ : I → M is a geodesic in M .

Remark 12.32. It is not necessary to assume that the curve α in Exer-

cise 12.31 is parametrized by arc length; it turns out that any geodesic is
automatically parametrized by a constant multiple of arc length, sometimes
called the speed of the curve.

*Exercise 12.33. Now suppose that α : I → M is a nondegenerate curve,

parametrized by arc length. Show that the construction in §11.4 for the
Frenet frame may be carried out in exactly the same way and that we have
the following analog of the Frenet equations:

(12.41) α (s) ∇e1 (s) e1 (s) ∇e1 (s) e2 (s) ∇e1 (s) e3 (s)
⎡ ⎤
0 0 0 0
⎢ ⎥
 ⎢1 0 −κ(s) 0 ⎥
= α(s) e1 (s) e2 (s) e3 (s) ⎢ ⎥
⎢0 κ(s) 0 −τ (s)⎥ ,
⎣ ⎦
0 0 τ (s) 0
where ∇ is the Levi-Civita connection on M and κ, τ : I → R are smooth
functions along α with
κ(s) = |∇e1 (s) e1 (s)| > 0.
As in the Euclidean case, the functions κ(s), τ (s) are called the curvature
and torsion, respectively, of α. In general, there is no analog of the result in
Exercise 11.20(b); a curve with torsion τ (s) identically zero is not necessarily
contained in any special hypersurface of M . Indeed, even a geodesic need
not be contained in any special hypersurface of M ; we will explore this
question in Exercises 12.41 and 12.42.

12.6. Moving frames for surfaces in 3-dimensional

Riemannian manifolds

Now, let U be an open set in R2 , and let x : U → M be an immersion whose

image is a surface Σ = x(U ). Just as for curves, an adapted frame field
along Σ is a lifting x̃ : U → F (M ) of the form
x̃(u) = (x(u); e1 (u), e2 (u), e3 (u)) ,
where for each u ∈ U , (e1 (u), e2 (u), e3 (u)) is an oriented, orthonormal basis
for the tangent space Tx(u) M .
382 12. Moving frames on Riemannian manifolds

*Exercise 12.34. Convince yourself that the following statements, which

we proved for surfaces in E3 in Chapter 4 and for surfaces in S3 and H3 in
Chapter 11, remain true for a surface Σ = x(U ) ⊂ M :
(a) We can choose an orthonormal frame (e1 (u), e2 (u), e3 (u)) for each tan-
gent space Tx(u) M so that that e3 (u) is orthogonal to Tx(u) Σ and (e1 (u),
e2 (u)) span Tx(u) Σ. Any such choice defines a lifting x̃ : U → F (M ) by
x̃(u) = (x(u); e1 (u), e2 (u), e3 (u)) .

(b) Let (ω̄ i , ω̄ji ) denote the pullbacks of the Maurer-Cartan forms (ω i , ωji )
on F (M ) to U via x̃. Then we have ω̄ 3 = 0, and the 1-forms (ω̄ 1 , ω̄ 2 ) form
a basis for the 1-forms on U .
(c) The metric on M naturally induces a metric on Σ = x(U ) ⊂ M , given
by the first fundamental form
I = (ω̄ 1 )2 + (ω̄ 2 )2 .

(d) Differentiating the equation ω̄ 3 = 0 and applying Cartan’s lemma implies

that there exist functions h11 , h12 , h22 on U such that
 3   
ω̄1 h11 h12 ω̄ 1
= .
ω̄23 h12 h22 ω̄ 2

As for surfaces in S3 and H3 , the Gauss map of Σ must be defined as a map

from Σ to the tangent bundle T M :
Definition 12.35. Let U ⊂ R2 be an open set, and let x : U → M be an
immersion with image Σ = x(U ). The Gauss map of Σ = x(U ) is the map
N : Σ → T M defined by
N (x(u)) = e3 (u) ∈ Tx(u) M,
where (e1 (u), e2 (u), e3 (u)) is any adapted frame field on Σ = x(U ).
Definition 12.36. Let U ⊂ R2 be an open set, and let x : U → M be
an immersion. The second fundamental form of Σ = x(U ) is the quadratic
form II on T U defined by
II(v) = −∇v e3 , dx(v)
for v ∈ Tu U , where (e1 (u), e2 (u), e3 (u)) is any adapted frame field on Σ =
x(U ).
*Exercise 12.37 (Cf. Exercise 11.24). Show that
II = ω̄13 ω̄ 1 + ω̄23 ω̄ 2 = h11 (ω̄ 1 )2 + 2h12 ω̄ 1 ω̄ 2 + h22 (ω̄ 2 )2 .
12.6. Moving frames for surfaces in Riemannian manifolds 383

As for surfaces in E3 , the first and second fundamental forms are invariants
of the surface. Consequently, if two surfaces have different first and second
fundamental forms, then they cannot be equivalent via an isometry of M .
However, the converse question is not so straightforward: In general, when
there is no group structure on F (M ), there are no analogs of Lemmas 4.2
and 4.12 that tell us when we have a complete set of invariants for a surface,
or when two surfaces are equivalent via a symmetry of M . In particular,
given arbitrary quadratic forms I and II on an open set U ⊂ R2 , checking
that the structure equations are satisfied is not sufficient to guarantee that
there exists an immersion x : U → M whose first and second fundamental
forms are I and II. The best that we can do is to arbitrarily specify one
of these quadratic forms, and even then the existence question may be very
difficult. For instance, we have the following local isometric embedding
theorem, which is due independently to Cartan [Car27] and Janet [Jan26].

Theorem 12.38 (Cartan-Janet). Let M be a 3-dimensional real analytic

Riemannian manifold; let U ⊂ R2 be an open set, and let I be a real analytic
metric on U (i.e., a positive definite quadratic form on U with real analytic
coefficients). Then every point u ∈ U has a neighborhood V ⊂ U on which
there exists an embedding x : V → M for which the metric on Σ = x(V )
induced from M agrees with the given quadratic form I.

The proof of this theorem uses exterior differential systems; it may be found
in [BCG+ 91]. The assumption of real analyticity is a crucial ingredient of
the proof; without this hypothesis, much less is known about the existence
of isometric embeddings, even locally. Even the local existence question for
embedding a surface with a C ∞ metric into E3 is not completely solved. (For
a comprehensive treatment of the current status of the isometric embedding
problem, see [HH06].)
For another example of how the geometry of surfaces in a general 3-dimen-
sional Riemannian manifold M differs from the geometry of surfaces in the
homogeneous spaces E3 , S3 , and H3 , consider the family of totally geodesic
surfaces in M , which are defined exactly as in Definition 11.30:

Definition 12.39. A regular surface Σ = x(U ) ⊂ M is called totally geo-

desic if every geodesic in Σ is also a geodesic in M .

As for surfaces in E3 , S3 , and H3 , a surface Σ ⊂ M is totally geodesic if and

only if

(12.42) ∇w e3 = 0 for all w ∈ T Σ,

where e3 is a unit normal vector field along Σ.

384 12. Moving frames on Riemannian manifolds

As we saw in Remark 11.31 and Exercise 11.32, totally geodesic surfaces

in the homogeneous spaces X = E3 , S3 , and H3 are plentiful: Given any
point x ∈ X and any 2-dimensional subspace P ⊂ Tx X, there exists a
totally geodesic surface Σ in X (a plane in E3 , a great sphere in S3 , or a
great hyperboloid in H3 ) with x ∈ Σ and Tx Σ = P . And for a general
3-dimensional Riemannian manifold M , the results of Exercise 12.31 imply
that a similar statement is true for curves: Given a point x ∈ M and a
vector v ∈ Tx M , there exists a geodesic curve α : I → M with α(0) = x
and α (0) = v. However, the existence of totally geodesic surfaces in a
general 3-dimensional Riemannian manifold M is a bit more complicated,
and the Riemann curvature tensor plays a fundamental role.
In order to understand how the Riemann curvature tensor Ω is related to the
existence of totally geodesic surfaces, it will be helpful to understand how
its component functions transform under a change of orthonormal frame
field. So, let (e1 (x), . . . , en (x)) be an orthonormal frame field on an open
set U ⊂ M , and let (ẽ1 (x), . . . , ẽn (x)) be any other orthonormal frame
field on U , related to the original frame field as in equation (12.2). From
Exercise 12.27, we know that the curvature Ω̄ ˜ associated to the transformed
frame field is related to the curvature Ω̄ associated to the original frame field
by equation (12.30), and the dual forms (ω̄ ˜ i ) associated to the transformed
frame field are related to the dual forms (ω̄ i ) associated to the original frame
field by equation (12.11). The combination of these two transformations
makes for a fairly complicated transformation rule for the functions (R̄jk i );

fortunately, in dimension 3 it can be simplified considerably.

*Exercise 12.40. (a) Let Ā ∈ SO(3), and write Ā as


Ā = a1 a2 a3 ,

so that the vectors (a1 , a2 , a3 ) in R3 represent the columns of Ā. Show that
for any cyclic permutation (i, j, k) of the indices (1, 2, 3), we have
ai × aj = ak .
Similarly, if (a1 , a2 , a3 ) represent the rows of Ā, then
ai × aj = ak .

˜ † denote the vectors of curvature 2-forms

(b) Let Ω̄† , Ω̄
⎡ 2⎤ ⎡ ⎤
Ω̄3 ˜2
Ω̄ 3
⎢ ⎥ ⎢ ⎥
⎢ ˜ 3⎥
Ω̄† = ⎢ 3⎥
⎣Ω̄1 ⎦ ,
˜ † = ⎢Ω̄
Ω̄ ⎥.
⎣ 1⎦
Ω̄12 ˜1
Ω̄ 2
12.6. Moving frames for surfaces in Riemannian manifolds 385

Use equation (12.30) and part (a) to show that

˜ † = Ā−1 Ω̄† .
Ω̄
(Hint: Recall that for Ā ∈ SO(3), we have Ā−1 = tĀ.)
(c) Let ω̄ † , ω̄
˜ † denote the vectors of 2-forms
⎡ 2 ⎤ ⎡ 2 ⎤
ω̄ ∧ ω̄ 3 ˜ ∧ ω̄
ω̄ ˜3
⎢ 3 ⎥ ⎢ 3 ⎥
ω̄ † = ⎢ 1⎥
⎣ω̄ ∧ ω̄ ⎦ ,
˜ † = ⎢ω̄
ω̄ ˜ 1⎥
⎣ ∧ ω̄ ⎦ .
˜

ω̄ 1 ∧ ω̄ 2 ˜ 1 ∧ ω̄
ω̄ ˜2
Use equation (12.11) and part (a) to show that
˜ † = Ā−1 ω̄ † .
ω̄

(d) For a 3-dimensional Riemannian manifold M , the Riemann curvature

tensor has six nontrivial components, represented by the functions
2 3 1
R323 , R131 , R212 ,
(12.43) 3 1 1 2 2 3
R112 = R231 , R223 = R312 , R331 = R123 .
˜ denote the matrices
Let R̄, R̄
⎡ 2 ⎡ ⎤
2
R̄323 R̄331 2 ⎤
R̄321 ˜2
R̄ ˜ 2 R̄
R̄ ˜2
323 331 321
⎢ 3 ⎥ ⎢ ⎥
R̄ = ⎢ 3 ⎥ ˜=⎢ ˜3
⎢R̄ ˜ 3 R̄˜3 ⎥
112 ⎥
3
⎣R̄123 R̄131 R̄112 ⎦ , R̄
⎣ 123
R̄ 131 ⎦
.
1
R̄223 1
R̄231 1
R̄212 ˜1 ˜ 1 R̄˜1
R̄ 223 R̄ 231 212

The symmetries in equation (12.43) imply that these matrices are symmetric,
and by definition, we have
(12.44) Ω̄† = R̄ ω̄ † , ˜ † = R̄
Ω̄ ˜ †.
˜ ω̄

Use parts (b), (c), and equation (12.44) to show that

(12.45) ˜ = Ā−1 R̄Ā.
R̄

Next, we will see how the curvature tensor can impose constraints on the
existence of totally geodesic surfaces.
*Exercise 12.41. Suppose that Σ = x(U ) ⊂ M is a totally geodesic surface
in M . Let (e1 (u), e2 (u), e3 (u)) be an adapted frame field along Σ, with
associated dual and connection forms (ω̄ i , ω̄ji ) on U .
(a) Show that the condition (12.42) is equivalent to the condition that
(12.46) ω̄13 = ω̄23 = 0.
386 12. Moving frames on Riemannian manifolds

(b) Differentiate equations (12.46) and show that

(12.47) 3
R̄112 ω̄ 1 ∧ ω̄ 2 = R̄312
2
ω̄ 1 ∧ ω̄ 2 = 0,
3 , R̄2
where R̄112 312 denote the pullbacks to U of the corresponding compo-
nents of the Riemann curvature tensor of M on F (M ) via the frame field
(e1 (u), e2 (u), e3 (u)). Since we must have ω̄ 1 ∧ ω̄ 2 = 0, conclude that
3 2
(12.48) R̄112 = R̄312 = 0.

(c) The transformation rule (12.45) indicates that R may be regarded as

a well-defined, self-adjoint linear transformation on the tangent space Tx M
and R̄ is its matrix expression with respect to the basis (e1 (u), e2 (u), e3 (u)).
Recall that any symmetric 3×3 matrix can be diagonalized by a transforma-
tion of the form (12.45) with Ā ∈ SO(3). Conclude from part (b) that, given
an orthonormal frame (e1 , e2 , e3 ) at a point x ∈ M , the plane P ⊂ Tx M
spanned by (e1 , e2 ) cannot be tangent to a totally geodesic surface in M
unless e3 is an eigenvector for R.

Finally, we will consider an example to show that the necessary condition

developed in Exercise 12.41(c) is not sufficient to guarantee the existence
of a totally geodesic surface in M passing through the point x ∈ M and
tangent to the plane P ⊂ Tx M ; indeed, a given Riemannian 3-manifold M
may contain very few totally geodesic surfaces, or even none at all.
*Exercise 12.42 (Maple recommended). Let a > b > c > 0, and let
M = R3 , with metric
1  2 
g= 2 2 2 2
dx + dy 2 + dz 2 .
(ax + by + cz + 1)

(a) Show that the frame field

∂
e1 (x, y, z) = (ax2 + by 2 + cz 2 + 1) ,
∂x
∂
(12.49) e2 (x, y, z) = (ax2 + by 2 + cz 2 + 1) ,
∂y
∂
e3 (x, y, z) = (ax2 + by 2 + cz 2 + 1)
∂z
on M is an orthonormal frame field for g and that its dual forms are
dx
ω̄ 1 = ,
(ax2 + by 2 + cz 2 + 1)
dy
(12.50) ω̄ 2 = ,
(ax + by + cz 2 + 1)
2 2

dz
ω̄ 3 = .
(ax + by + cz 2 + 1)
2 2
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
ω̄32 = (by dz − cz dy),
(ax2 + by 2 + cz 2 + 1)
2
(12.51) ω̄13 = (cz dx − ax dz),
(ax2 + by 2 + cz 2 + 1)
2
ω̄21 = (ax dy − by dx).
(ax + by + cz 2 + 1)
2 2

(c) Show that the curvature matrix R̄ has the form

⎡ 2 ⎤
R̄323 0 0
⎢ ⎥
R̄ = ⎢
⎣ 0 R̄ 3
131 0 ⎥,
⎦
0 1
0 R̄212
where
 
2
R̄323 = 2 (b + c − 2a)ax2 + (b − c)(cz 2 − by 2 ) + (b + c) ,
 
(12.52) 3
R̄131 = 2 (c + a − 2b)by 2 + (c − a)(ax2 − cz 2 ) + (c + a) ,
 
1
R̄212 = 2 (a + b − 2c)cz 2 + (a − b)(by 2 − ax2 ) + (a + b) .
2 , R̄3 , R̄1 ) of R̄ are distinct for all x =
Verify that the eigenvalues (R̄323 131 212
(x, y, z) ∈ M , and conclude that the corresponding eigenvectors of R̄ in each
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 ⊂ 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 Σ ⊂ M , it must be a
level surface for one of the coordinate functions (x, y, z) on M .
(e) Let U ⊂ R2 , and consider a candidate surface Σ ⊂ M with parametriza-
tion x : U → M given by
x(u, v) = (u, v, z0 )
for some z0 ∈ R. The frame field (12.49) is an adapted frame field along Σ.
Compute the pullbacks of the connection forms (12.51) to U ; in particular,
show that
2cz0 du 2cz0 dv
ω̄13 = 2 , ω̄23 = .
2 2
(au + bv + cz0 + 1) (au + bv 2 + cz02 + 1)
2

Conclude from the result of Exercise 12.41(a) that Σ 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 “planes”
Σ1 = {(0, y, z) | y, z ∈ R} ⊂ M,
Σ2 = {(x, 0, z) | x, z ∈ R} ⊂ M,
Σ3 = {(x, y, 0) | x, y ∈ R} ⊂ M.
Any open submanifold M0 ⊂ M that does not intersect these surfaces—for
instance, the first octant M0 = {(x, y, z) ∈ M | x, y, z > 0}—contains no
totally geodesic surfaces whatsoever.

12.7. Maple computations

Exercise 12.30: It’s not so simple to tell Maple how to do computations

on a manifold of arbitrary dimension, so we’ll go through this exercise for
the case n = m = 3, which represents a 3-dimensional Riemannian mani-
fold 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 first
need to declare the Maurer-Cartan forms on F (E6 ). This is rather a lot of
forms, and it’s 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 (ω̄ia )
with 1 ≤ i ≤ 3, 4 ≤ a ≤ 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 differentiate these forms according to the Cartan struc-
ture equations (3.8), and be sure to define structure equations only for those
connection forms that have not been defined in terms of other forms:
> for i from 1 to 6 do
d(omega[i]):= sum(’-omega[i,j] &ˆ omega[j]’, ’j’=1..6);
end do;
d(omega[1,2]):= sum(’-omega[1,k] &ˆ omega[k,2]’, ’k’=1..6);
d(omega[2,3]):= sum(’-omega[2,k] &ˆ omega[k,3]’, ’k’=1..6);
d(omega[3,1]):= sum(’-omega[3,k] &ˆ omega[k,1]’, ’k’=1..6);
d(omega[4,5]):= sum(’-omega[4,k] &ˆ omega[k,5]’, ’k’=1..6);
d(omega[5,6]):= sum(’-omega[5,k] &ˆ omega[k,6]’, ’k’=1..6);
d(omega[6,4]):= sum(’-omega[6,k] &ˆ omega[k,4]’, ’k’=1..6);
for a from 4 to 6 do
for j from 1 to 3 do
d(omega[a,j]):= sum(’-omega[a,k] &ˆ omega[k,j]’,
’k’=1..6);
end do;
end do;
Now, consider an adapted orthonormal frame field along M for which the
frame vectors (e1 (u), e2 (u), e3 (u)) are tangent to M at each point. Accord-
ing to equation (12.37), for such a frame field, we must have ω̄ 4 = ω̄ 5 =
ω̄ 6 = 0. Set up a substitution for the Maurer-Cartan forms associated to an
adapted frame field:
> adaptedsub:= [omega[4]=0, omega[5]=0, omega[6]=0];
Now differentiate these equations:
> Simf(subs(adaptedsub, Simf(d(omega[4]))));
Simf(subs(adaptedsub, Simf(d(omega[5]))));
Simf(subs(adaptedsub, Simf(d(omega[6]))));
Applying Cartan’s 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(’i’, ’j’, ’a’);
> 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 dω̄ 1 , dω̄ 2 , dω̄ 3 :
> Simf(subs(adaptedsub, Simf(d(omega[1]))));

(ω2 ) &ˆ (ω1,2 ) − (ω3 ) &ˆ (ω3,1 )

> Simf(subs(adaptedsub, Simf(d(omega[2]))));

−(ω1 ) &ˆ (ω1,2 ) + (ω3 ) &ˆ (ω2,3 )

> Simf(subs(adaptedsub, Simf(d(omega[3]))));

(ω1 ) &ˆ (ω3,1 ) − (ω2 ) &ˆ (ω2,3 )

It follows from these equations that the 1-forms (ω̄ji ) with 1 ≤ i, j ≤ 3 are
the Levi-Civita connection forms associated to the metric
 2  2  2
I = ω̄ 1 + ω̄ 2 + ω̄ 3

on M . Therefore, they satisfy the structure equations (12.35), where (Rjki )

are the components of the Riemann curvature tensor of M . Now compare

equations (12.35) to the structure equations for the (dω̄ji ) as Maurer-Cartan
forms on E6 :
> zero1:= Simf(subs(adaptedsub, Simf(d(omega[1,2])
+ omega[1,3] &ˆ omega[3,2] - R[1,2,1,2]*omega[1] &ˆ omega[2]
- R[1,2,2,3]*omega[2] &ˆ omega[3]
- R[1,2,3,1]*omega[3] &ˆ omega[1])));
> zero2:= Simf(subs(adaptedsub, Simf(d(omega[2,3])
+ omega[2,1] &ˆ omega[1,3] - R[2,3,1,2]*omega[1] &ˆ omega[2]
- R[2,3,2,3]*omega[2] &ˆ omega[3]
- R[2,3,3,1]*omega[3] &ˆ omega[1])));
> zero3:= Simf(subs(adaptedsub, Simf(d(omega[3,1])
+ omega[3,2] &ˆ omega[2,1] - R[3,1,1,2]*omega[1] &ˆ omega[2]
- R[3,1,2,3]*omega[2] &ˆ omega[3]
- R[3,1,3,1]*omega[3] &ˆ omega[1])));
12.7. Maple computations 391

The scalar coefficients in these quantities are equivalent to the Gauss equa-
tions (12.39).
Next, look at the structure equations for the connection forms (ω̄ia ) with
1 ≤ i ≤ 3, 4 ≤ a ≤ 6:
> zero41:= Simf(subs(adaptedsub, Simf(d(omega[4,1]))
- d(Simf(subs(adaptedsub, omega[4,1])))));
The resulting expression has the form

φ1 ∧ ω̄ 1 + φ2 ∧ ω̄ 2 + φ3 ∧ ω̄ 3 ,

where φ1 , φ2 , φ3 are the following 1-forms:

> pick(zero41, omega[1]);

−2 h4,1,2 ω1,2 + 2 h4,3,1 ω3,1 − h5,1,1 ω4,5 + h6,1,1 ω6,4 − d(h4,1,1 )

> pick(zero41, omega[2]);

− (h4,2,2 − h4,1,1 ) ω1,2 − h4,3,1 ω2,3 + h4,2,3 ω3,1

− h5,1,2 ω4,5 + h6,1,2 ω6,4 − d(h4,1,2 )

> pick(zero41, omega[3]);

− h4,2,3 ω1,2 + h4,1,2 ω2,3 − (−h4,3,3 + h4,1,1 ) ω3,1

− h5,3,1 ω4,5 + h6,3,1 ω6,4 − d(h4,3,1 )

Applying Cartan’s lemma to all 9 structure equations for the (dω̄ia ) 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 (ω̄ji , ω̄ba )) + haijk ω̄ k .

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
∂haij ∂haik
− = (lower order terms).
∂xk ∂xj
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 (ω̄ba ) with
4 ≤ a, b ≤ 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

dω̄ba = −ω̄ca ∧ ω̄bc + Sbij

a
ω̄ i ∧ ω̄ j ,

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 coefficients (Sbij
the connection on the normal bundle of M induced from the Euclidean
connection on E6 .
Exercise 12.42: Since this exercise requires different dual and connection
forms with different structure equations from the previous exercise, it’s prob-
ably 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 field
on M , as well as the reverse substitution:
> dualformssub:= [
omega[1] = d(x)/(a*xˆ2 + b*yˆ2 + c*zˆ2 + 1),
omega[2] = d(y)/(a*xˆ2 + b*yˆ2 + c*zˆ2 + 1),
omega[3] = d(z)/(a*xˆ2 + b*yˆ2 + c*zˆ2 + 1)];
> dualformsbacksub:= makebacksub(dualformssub);
12.7. Maple computations 393

In order to compute the connection forms, differentiate the dual forms and
compare the result to the structure equations. (It’s helpful to go ahead
and write the (ω̄ji ) as linear combinations of the (ω̄ i ) with undetermined
coefficients and then solve for these coefficients.)
> 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] &ˆ omega[2] + omega[1,3] &ˆ omega[3])))));
zero2:= Simf(d(Simf(subs(dualformssub, omega[2])))
+ Simf(subs(dualformssub, Simf(subs(connectionformssub,
omega[2,1] &ˆ omega[1] + omega[2,3] &ˆ omega[3])))));
zero3:= Simf(d(Simf(subs(dualformssub, omega[3])))
+ Simf(subs(dualformssub, Simf(subs(connectionformssub,
omega[3,1] &ˆ omega[1] + omega[3,2] &ˆ omega[2])))));

(2cz − h3,1,1 ) (d(x)) &ˆ(d(z)) (2by + h1,2,1 ) (d(y)) &ˆ(d(x))

zero1 := −
(ax2 + by 2 + cz 2 + 1)2 (ax2 + by 2 + cz 2 + 1)2
(h1,2,3 + h3,1,2 ) (d(y)) &ˆ(d(z))
−
(ax2 + by 2 + cz 2 + 1)2

(h1,2,3 − h3,2,1 ) (d(x)) &ˆ(d(z)) (2ax − h1,2,2 ) (d(y)) &ˆ(d(x))

zero2 := +
(ax2 + by 2 + cz 2 + 1)2 (ax2 + by 2 + cz 2 + 1)2
(2cz − h3,2,2 ) (d(y)) &ˆ(d(z))
+
(ax2 + by 2 + cz 2 + 1)2

(2ax + h3,1,3 ) (d(x)) &ˆ(d(z))

zero3 := −
(ax2 + by 2 + cz 2 + 1)2
(h3,1,2 − h3,2,1 ) (d(y)) &ˆ(d(x)) (2by + h3,2,3 ) (d(y)) &ˆ(d(z))
+ −
(ax2 + by 2 + cz 2 + 1)2 (ax2 + by 2 + cz 2 + 1)2
394 12. Moving frames on Riemannian manifolds

We can set the scalar coefficients 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 = −2by, h1,2,2 = 2ax, h1,2,3 = 0, h3,1,1 = 2cz, h3,1,2 = 0, h3,1,3 = −2ax,
h3,2,1 = 0, h3,2,2 = 2cz, h3,2,3 = −2by}

> assign(%);
So here are the connection forms, expressed in terms of the coordinate 1-
forms:
> connectionformssub:= Simf(subs(dualformssub,
Simf(connectionformssub)));


connectionf ormssub :=

2by d(x) 2ay d(y)

ω1,2 = − + 2 ,
ax2 + by + cz + 1 ax + by 2 + cz 2 + 1
2 2

2cz d(x) 2ax d(z)

ω3,1 = − 2 ,
ax2
+ by + cz + 1 ax + by 2 + cz 2 + 1
2 2


2cz d(y) 2by d(z)
ω3,2 = 2 −
ax + by 2 + cz 2 + 1 ax2 + by 2 + cz 2 + 1

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] &ˆ omega[1,3])))));
Omega[3,1]:= Simf(subs(dualformsbacksub,
Simf(d(Simf(subs(connectionformssub, omega[3,1])))
+ Simf(subs(connectionformssub,
omega[3,2] &ˆ 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] &ˆ omega[3,2])))));


Ω2,3 := 2cax2 + 2cby 2 − 2c2 z 2 + 2c + 2bax2 − 2b2 y 2 + 2bcz 2

+ 2b − 4a2 x2 (ω2 ) &ˆ (ω3 )

Ω3,1 := − 2cax2 − 2cby 2 + 2c2 z 2 − 2c + 2a2 x2 − 2aby 2 − 2acz 2

− 2a + 4b2 y 2 (ω1 ) &ˆ (ω3 )

Ω1,2 := 2bax2 − 2b2 y 2 + 2bcz 2 + 2b − 2a2 x2 + 2aby 2 + 2acz 2

+ 2a − 4c2 z 2 (ω1 ) &ˆ (ω2 )

So the matrix R̄ is diagonal, with the following diagonal entries:

> 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]);
Check that the diagonal entries of R̄ are distinct:
> factor(R[2,3,2,3] - R[3,1,3,1]);
−2 (−b + a) (ax2 + by 2 + cz 2 + 1)

> factor(R[3,1,3,1] - R[1,2,1,2]);

−2 (−c + b) (ax2 + by 2 + cz 2 + 1)

> factor(R[1,2,1,2] - R[2,3,2,3]);

2 (a − 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 Σ is contained in the level
set {(x, y, z) ∈ M | z = z0 } for some constant z0 and compute the pullbacks
of the connection forms to Σ:
> Form(z0=-1);
> Simf(subs([x=u, y=v, z=z0], connectionformssub));

2bv d(u) 2au d(v)
ω1,2 = − 2 + ,
au + bv 2 + cz02 + 1 au2 + bv 2 + cz02 + 1

2cz0 d(u) 2cz0 d(v)
ω3,1 = 2 2 2
, ω3,2 = 2
au + bv + cz0 + 1 au + bv 2 + cz02 + 1

Since a totally geodesic surface must have ω̄13 = ω̄23 = 0, Σ is not a totally
geodesic surface unless z0 = 0.
Bibliography

[AEG06] Juan A. Aledo, José M. Espinar, and José A. Gálvez, Timelike surfaces in the Lorentz-
Minkowski space with prescribed Gaussian curvature and Gauss map, J. Geom. Phys.
56 (2006), no. 8, 1357–1369.

[AI79] Robert L. Anderson and Nail H. Ibragimov, Lie-Bäcklund Transformations in Ap-

plications, SIAM Studies in Applied Mathematics, vol. 1, Society for Industrial and
Applied Mathematics (SIAM), Philadelphia, PA, 1979.

[AR93] M. A. Akivis and B. A. Rosenfeld, Élie Cartan (1869–1951), Translations of Mathe-

matical Monographs, vol. 123, American Mathematical Society, Providence, RI, 1993,
translated from the Russian manuscript by V. V. Goldberg.

[B8̈3] A. V. Bäcklund, Om ytor med konstant negativ krökning, Lunds Universitets Årsskrift
19 (1883), 1–48.

[BCG+ 91] R. L. Bryant, S. S. Chern, R. B. Gardner, H. L. Goldschmidt, and P. A. Griffiths,

Exterior Differential 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 superficie elicoidali e sulle superficie a curvatura
costante, Ann. Scuola Norm. Sup. Pisa Cl. Sci. 2 (1879), 285–341.

[Bla23] Wilhelm Blaschke, Vorlesungen über Differentialgeometrie, vol. II, Springer, Berlin,
1923.

[Bla85] , Gesammelte Werke. Band 4, Thales-Verlag, Essen, 1985, Affine Differential-

geometrie. Differentialgeometrie der Kreis- und Kugelgruppen. [Affine differential ge-
ometry. Differential geometry of circle and ball groups], with commentaries by Werner
Burau and Udo Simon, edited by Burau, S. S. Chern, K. Leichtweiß, H. R. Müller, L.
A. Santaló, Simon and K. Strubecker.

[Cal00] James J. Callahan, The Geometry of Spacetime: An Introduction to Special and Gen-
eral Relativity, Undergraduate Texts in Mathematics, Springer-Verlag, New York,
2000.

397
398 Bibliography

[Car04] Élie Cartan, Sur la structure des groupes infinis de transformation, Ann. Sci. École
Norm. Sup. (3) 21 (1904), 153–206.

[Car20] , Sur la déformation projective des surfaces, Ann. Sci. École Norm. Sup. (3)
37 (1920), 259–356.

[Car27] , Sur la possibilité de plonger un espace riemannien donné dans un espace

euclidéen, Ann. Soc. Polon. Math. 6 (1927), 1–7.

[Car30] , La théorie des groupes finis et continus et l’analysis situs, Mémorial des
sciences mathématiques, no. 42, Gauthier-Villars et Cie, 1930.

[Car35] , La Méthode du Repère Mobile, la Théorie des Groupes Continus, et les Es-
paces Généralisés, Exposés de Géométrie, no. 5, Hermann, Paris, 1935.

[Car46] , Leçons sur la Géométrie des Espaces de Riemann, 2nd ed., Gauthier-Villars,
Paris, 1946.

[Car92] , Leçons sur la Géométrie Projective Complexe. La Théorie des Groupes Finis
et Continus et la Géométrie Différentielle Traitées Par la Méthode du Repère Mobile.
Leçons sur la Théorie des Espaces à Connexion Projective, Les Grands Classiques
Gauthier-Villars. [Gauthier-Villars Great Classics], Éditions 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 Equiaffine Space,
Proc. Amer. Math. Soc. 142 (2014), 2543–2558.

[Che42] Shiing-shen Chern, On integral geometry in Klein spaces, Ann. of Math. (2) 43 (1942),
178–189.

[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),

189–208.

[CT80] Shiing Shen Chern and Chuu Lian Terng, An analogue of Bäcklund’s theorem in affine
geometry, Rocky Mountain J. Math. 10 (1980), no. 1, 105–124.

[CT86] Shiing-shen Chern and Keti Tenenblat, Pseudo-spherical surfaces and evolution equa-
tions, Stud. Appl. Math. 74 (1986), 55–83.

[Dac08] Bernard Dacorogna, Introduction to the Calculus of Variations, 2nd ed., Imperial
College Press, London, 2008.

[Dar72a] Gaston Darboux, Leçons sur la Théorie Générale des Surfaces et les Applications
Géométriques du Calcul Infinitésimal. Deuxième partie, Chelsea Publishing Co.,
Bronx, N.Y., 1972, Les Congruences et les Équations Linéaires aux Dérivées Par-
tielles. Les Lignes Tracées sur les Surfaces, réimpression de la deuxième édition de
1915.

[Dar72b] , Leçons sur la Théorie Générale des Surfaces et les Applications Géométriques
du Calcul Infinitésimal. Première partie, Chelsea Publishing Co., Bronx, N.Y., 1972,
Généralités. Coordonnées curvilignes. Surfaces minima, réimpression de la deuxième
édition de 1914.

[Dar72c] , Leçons sur la Théorie Générale des Surfaces et les Applications Géométriques
du Calcul Infinitésimal. Quatrième partie, Chelsea Publishing Co., Bronx, N.Y.,
Bibliography 399

1972, Déformation Infiniment Petite et Representation Sphérique, réimpression de la

première édition de 1896.

[Dar72d] , Leçons sur la Théorie Générale des Surfaces et les Applications Géométriques
du Calcul Infinitésimal. Troisième partie, Chelsea Publishing Co., Bronx, N.Y., 1972,
Lignes Géodésiques et Courbure Géodésique. Paramètres Différentiels. Déformation
des Surfaces, réimpression de la première édition de 1894.

[dC76] Manfredo P. do Carmo, Differential Geometry of Curves and Surfaces, Prentice-Hall

Inc., Englewood Cliffs, N.J., 1976, translated from the Portuguese.

[dC92] , Riemannian Geometry, Mathematics: Theory & Applications, Birkhäuser

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 Geometry—Methods 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, 263–321.

[Eis60] Luther Pfahler Eisenhart, A Treatise on the Differential Geometry of Curves and
Surfaces, Dover Publications Inc., New York, 1960.

[Fav57] J. Favard, Cours de géométrie différentielle 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, 161–213.

[FO99] , Moving coframes. II. Regularization and theoretical foundations, Acta Appl.
Math. 55 (1999), no. 2, 127–208.

[Fom90] A. T. Fomenko, The Plateau Problem. Part I, Studies in the Development of Mod-
ern Mathematics, vol. 1, Gordon and Breach Science Publishers, New York, 1990,
Historical survey, translated from the Russian.

[Fre47] J. F. Frenet, Sur les Courbes à Double Courbure, Ph.D. thesis, Toulouse, 1847.

[Gál09] José A. Gálvez, Surfaces of constant curvature in 3-dimensional space forms, Mat.
Contemp. 37 (2009), 1–42.

[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, differential invariants and rigidity theorems for
curves in homogeneous spaces, Duke Math. J. 45 (1978), no. 4, 735–779.

[Gri74] P. Griffiths, On Cartan’s method of Lie groups and moving frames as applied to
uniqueness and existence questions in differential geometry, Duke Math. J. 41 (1974),
775–814.

[Gug77] Heinrich W. Guggenheimer, Differential Geometry, Dover Publications, Inc., New

York, 1977, corrected reprint of the 1963 edition, Dover Books on Advanced Mathe-
matics.

[HCV52] D. Hilbert and S. Cohn-Vossen, Geometry and the Imagination, Chelsea Publishing
Company, New York, N.Y., 1952, translated by P. Neményi.
400 Bibliography

[HH06] Qing Han and Jia-Xing Hong, Isometric Embedding of Riemannian Manifolds in Eu-
clidean Spaces, Mathematical Surveys and Monographs, vol. 130, American Mathe-
matical Society, Providence, RI, 2006.

[IL03] Thomas A. Ivey and J. M. Landsberg, Cartan for Beginners: Differential Geometry via
Moving Frames and Exterior Differential Systems, Graduate Studies in Mathematics,
vol. 61, American Mathematical Society, Providence, RI, 2003.

[Jan26] M. Janet, Sur la possibilité de plonger un espace Riemannien donné dans un espace
euclidien, Ann. Soc. Polon. Math. 5 (1926), 38–43.

[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, 215–249.

[Kle93b] , Vergleichende Betrachtungen über neuere geometrische Forschungen, Math.

Ann. 43 (1893), no. 1, 63–100.

[Kob95] Shoshichi Kobayashi, Transformation Groups in Differential 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), 266–291.

[Lee13] John M. Lee, Introduction to Smooth Manifolds, 2nd ed., Graduate Texts in Mathe-
matics, vol. 218, Springer, New York, 2013.

[Man10] Elizabeth Louise Mansfield, A Practical Guide to the Invariant Calculus, Cambridge
Monographs on Applied and Computational Mathematics, vol. 26, Cambridge Univer-
sity 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, 161–185.

[Min78] Hermann Minkowski, Die Grundgleichungen für die elektromagnetischen Vorgänge in

bewegten Körpern, Nachrichten von der Gesellschaft der Wissenschaften zu Göttingen,
Mathematisch-Physikalische Klasse (1907/8), 53–111.

[Min89] , Raum und Zeit, Jahresbericht der Deutschen Mathematiker-Vereinigung

(1908/9), 75–88.

[MR05] Sebastián Montiel and Antonio Ros, Curves and surfaces, Graduate Studies in Math-
ematics, vol. 69, American Mathematical Society, Providence, RI; Real Sociedad
Matemática Española, Madrid, 2005, translated and updated from the 1998 Span-
ish edition by the authors.

[Nas56] John Nash, The imbedding problem for Riemannian manifolds, Ann. of Math. (2) 63
(1956), 20–63.

[NMS08] Mehdi Nadjafikhah and Ali Mahdipour Sh., Affine classification of n-curves, Balkan
J. Geom. Appl. 13 (2008), no. 2, 66–73.

[NS94] Katsumi Nomizu and Takeshi Sasaki, Affine Differential Geometry, Geometry of affine
immersions, Cambridge Tracts in Mathematics, vol. 111, Cambridge University Press,
Cambridge, 1994.

[Olv00] Peter Olver, Applications of Lie Groups to Differential 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. 35–63.

[Olv11a] , Lectures on Moving Frames, Symmetries and Integrability of Difference Equa-

tions (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 (Séminaire de Máthematiques
Supérieures) held at the Université de Montréal, Montréal, QC, June 8–21, 2008.

[Olv11b] , Recursive moving frames, Results Math. 60 (2011), 423–452.

[O’N83] Barrett O’Neill, Semi-Riemannian Geometry, with applications to relativity, Pure

and Applied Mathematics, vol. 103, Academic Press Inc. [Harcourt Brace Jovanovich
Publishers], New York, 1983.

[O’N06] , Elementary Differential Geometry, 2nd ed., Elsevier/Academic Press, Ams-

terdam, 2006.

[Opr07] John Oprea, Differential Geometry and Its Applications, 2nd ed., Classroom Resource
Materials Series, Mathematical Association of America, Washington, DC, 2007.

[OT05] V. Ovsienko and S. Tabachnikov, Projective Differential Geometry Old and New, Cam-
bridge Tracts in Mathematics, vol. 165, Cambridge University Press, Cambridge, 2005.

[Rad30] Tibor Radó, On Plateau’s problem, Ann. of Math. (2) 31 (1930), no. 3, 457–469.

[Rey98] E. G. Reyes, Pseudo-spherical surfaces and integrability of evolution equations, J. Diff.

Eq. 147 (1998), 195–230.

[RS82] C. Rogers and W. F. Shadwick, Bäcklund 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, Bäcklund and Darboux Transformations, Geometry and
modern applications in soliton theory, Cambridge Texts in Applied Mathematics, Cam-
bridge University Press, Cambridge, 2002.

[Sen31] C. E. Senff, Theoremata Princioalia e Theoria Curvarum et Su Perficierum, Dorpat

Univ., 1831.

[Ser51] J. Serret, Mémoire sur quelques formules relatives à la théorie des courbes à double
courbure, J. Math. Pures Appl. 16 (1851), 193–207.

[She99] Mary D. Shepherd, Line congruences as surfaces in the space of lines, Differential
Geom. Appl. 10 (1999), no. 1, 1–26.

[Spi79] Michael Spivak, A Comprehensive Introduction to Differential Geometry. 5 Vols., 2nd

ed., Publish or Perish Inc., Wilmington, DE, 1979.

[Ste07] James Stewart, Calculus: Early Transcendentals, 6th ed., Stewart’s Calculus Series,
Brooks Cole, Belmont, CA, 2007.

[Su83] Bu Chin Su, Affine Differential 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, 395–437.

[Wei66] K. Weierstrass, Über die Flächen deren mittlere Krümmung überall gleich null ist.,
Monatsber. Berliner Akad. (1866), 612–625, 855–856.
402 Bibliography

[Wil61] T. J. Willmore, The definition of Lie derivative, Proc. Edinburgh Math. Soc. (2) 12
(1960/1961), 27–29.

[Wil62] E. J. Wilczynski, Projective Differential Geometry of Curves and Ruled Surfaces,

Chelsea Publishing Co., New York, 1962.
Index

0-form, 42 Bäcklund transformation

1-form, 35 for Liouville’s equation, 302–303
on Rn , 40–41 for pseudospherical surfaces, 290
on a manifold, 47 for the sine-Gordon equation, 288,
298
Bäcklund’s theorem, 287, 290
An , see Equi-affine space Bäcklund, Albert, 290
Adapted frame field, 109 Baker-Jarvis, Duff, xvi
on a surface in E3 , 118 Bartels, Martin, xi
on a surface in A3 , 178 Bianchi, Luigi, 290
on a surface in P3 , 221 Blaschke representation for an elliptic
on a timelike surface in M1,2 , 150 equi-affine minimal surface in A3 ,
equi-affine principal adapted frame 278
field on an elliptic surface in A3 , Blaschke, Wilhelm, 178, 274
186 Bonnet’s theorem
null adapted frame field for a surface in E3
on a hyperbolic surface in A3 , 190 existence, 127
on a hyperbolic surface in P3 , 232 uniqueness, 124
on a timelike surface in M1,2 , 162 for a surface in S3 or H3 , 354
principal adapted frame field for a timelike surface in M1,2 , 155
on a surface in E3 , 123 Bryant, Robert, xv
on a timelike surface in M1,2 , 155 Bushek, Nathaniel, xv
Affine connection, see Connection
Affine geometry, 92 Canonical isomorphism
Affine Grassmannian, 288, 324 for dual spaces, 36
Affine transformation, 93 for tangent spaces, 16, 50, 76, 339,
Arc length, see Curve, arc length 367
Area functional Carlsen, Brian, xvi
on surfaces in E3 , 252 Cartan package for Maple, xiii, 59–66
equi-affine, on surfaces in A3 , 269 &ˆ command, 60
Area measure, 324 d command, 60
Associated family of a minimal surface Forder command, 60
in E3 , 267 Form command, 59

403
404 Index

makebacksub command, 63 Conic section, 177, 212

pick command, 62 Conjugate surface of a minimal surface
ScalarForm command, 63 in E3 , 267
Simf command, 61 Connection, 33
WedgeProduct command, 60 compatibility with a metric, 371
Cartan structure equations, see curvature tensor, 376
Structure equations flat connection on En , 366
Cartan’s formula for exterior derivative, Levi-Civita, see Levi-Civita
48 connection
Cartan’s formula for Lie derivative, 58 on a vector bundle, 365
Cartan’s lemma, 55 on the tangent bundle, 365–370
Cartan, Élie, xi, 70, 223, 233, 383 horizontal tangent space, 367
Cartan-Janet isometric embedding vertical tangent space, 366
theorem, 383 symmetric, 371
Catenoid, 128, 260 torsion-free, 371
associated family, 268 Connection forms
conjugate surface, 268 on the orthonormal frame bundle of
Weierstrass-Enneper representation, En , 79
268 on the orthonormal frame bundle of
Cauchy-Crofton formula, 324, 327 M1,n , 91
Cauchy-Riemann equations, 264 on the unimodular frame bundle of
Chain rule, 24 An , 95
Chern, Shiing-Shen, 297 on the projective frame bundle of Pn ,
Clelland, Richard, xvi 103
Codazzi equations for the Levi-Civita connection on Sn
for a surface in E3 , 127 or Hn , 347
for a surface in S3 , 353, 354 determined by a connection, 367, 370
for a surface in H3 , 353, 354 Constant type, 316
for a timelike surface in M1,2 , 156, Cotangent bundle, 36
165 Cotangent space, 36
for a submanifold of En+m , 379 Covariant derivative, 33
Column vector, see Vector, column for vector fields on Sn and Hn ,
vector 346–347
Commutative diagram, 19 compatibility with the metric, 347
Compatibility equations for vector fields on a submanifold of
for a surface in E3 , 127 En+m , 378
for a surface in S3 or H3 , 353 Covector, 37
for a timelike surface in M1,2 , 156, Covector space, 36
165 Cruz, Akaxia, xvi
for an elliptic surface in A3 , 188 Curvature, see also Curve, curvature;
for an elliptic surface in P3 , 229, 247 Gauss curvature; mean curvature
for a hyperbolic surface in P3 , 235 curvature matrix of a connection
for a submanifold of En+m , 379 matrix, 340
Complex analytic function, see curvature matrix of the connection
Holomorphic function matrix on F (Sn ), 342
Complex structure, 264 curvature matrix of the connection
Conformal parametrization of a surface, matrix on F (Hn ), 344
265 curvature tensor of a connection, 376
Conformal structure Curve
on a hyperbolic surface in P3 , 232 in E3
on an elliptic surface in P3 , 224 arc length, 112
Index 405

binormal vector, 112 nondegenerate curve, 215

complete set of invariants, 115 projective curvature forms, 218
curvature, 113 projective frame field, 214
Frenet equations, 114 projective Frenet equations, 219
Frenet frame, 112 projective parameter, 217
nondegenerate curve, 112 projective parametrization, 217
orthonormal frame field, 111 projective structure, 217
regular curve, 111 rational normal curve, 220
torsion, 113 Wilczynski invariants, 216
unit normal vector, 112 in S3
unit tangent vector, 111 binormal vector, 350
in M1,2 , null curve, 165 curvature, 350
in M1,2 , timelike curve Frenet equations, 350
Frenet equations, 148 Frenet frame, 350
Minkowski curvature, 147 geodesic, 349
Minkowski torsion, 148 geodesic equation, 349
nondegenerate curve, 146 nondegenerate curve, 349
orthonormal frame field, 144 orthonormal frame field, 348
proper time, 144 regular curve, 348
regular curve, 144 torsion, 350
unit normal vector, 146 unit normal vector, 350
unit tangent vector, 144 in H3
in A2 , 176–178 binormal vector, 350
conic section, 177 curvature, 350
equi-affine curvature, 177 Frenet equations, 350
in A3 Frenet frame, 350
geodesic, 349
equi-affine arc length, 174–175
geodesic equation, 349
equi-affine curvatures, 176
nondegenerate curve, 349
equi-affine Frenet equations, 176
orthonormal frame field, 348
equi-affine Frenet frame, 175
regular curve, 348
nondegenerate curve, 172
torsion, 350
rational normal curve, 178
unit normal vector, 350
unimodular frame field, 172
in a Riemannian 3-manifold
in P2
curvature, 381
canonical lifting, 205
Frenet equations, 381
canonical projective frame field,
Frenet frame, 381
205
geodesic, 380
conic section, 212
nondegenerate curve, 380
nondegenerate curve, 205
orthonormal frame field, 379
projective arc length, 211 regular curve, 379
projective curvature form, 210 torsion, 381
projective frame field, 204
projective Frenet equations, 212 Darboux tangents, 227
projective parameter, 207 Darboux, Jean-Gaston, xi
projective parametrization, 207 De Sitter spacetime, 157–158
projective structure, 210 Derivative
Wilczynski invariants, 206 directional, 19, 43, 57, 120, 347, 365
in P3 of a map from Rm to Rn , 16
canonical lifting, 215 of a map between manifolds, 23
canonical projective frame field, Diffeomorphism, 25
215 Differentiable manifold, see Manifold
406 Index

Differential Equi-affine geometry, 92

of a real-valued function, 35 Equi-affine group A(n), 94
of a map from Rm to Rn , 16 as a principal bundle over An , 95
of a map between manifolds, 24, 49 Equi-affine mean curvature, see Surface
Differential form in A3 , equi-affine mean curvature
0-form, 42 Equi-affine minimal surface, see
1-form, 35 Minimal surface, equi-affine, in A3
on Rn , 40–41 Equi-affine normal vector field, see
on a manifold, 47 Surface in A3 , equi-affine normal
p-form vector field
on Rn , 42 Equi-affine second fundamental form,
on a manifold, 47 see Surface in A3 , equi-affine
algebra of differential forms on Rn , second fundamental form
41, 42 Equi-affine space, 93, see also
closed form, 46 Homogeneous space, equi-affine
exact form, 46 space An
DifferentialGeometry package for volume form, 92
Maple, xiii Equi-affine sphere
Directional derivative, see Derivative, improper equi-affine sphere, 189
directional proper equi-affine sphere, 189
Divergence theorem, 55 Equi-affine transformation, 93
Doubly ruled surface, see Ruled surface, Equivalence problem, 107
doubly ruled surface Equivariant, 109
Dual forms Equivariant moving frame, see Moving
on the orthonormal frame bundle of frame, equivariant moving frame
En , 79 Estrada, Edward, xvi
on the projective frame bundle of Pn , Euclidean group E(n), 73
103 as a principal bundle over En , 75
associated to an orthonormal frame Euclidean space, 70, see also
field, 369 Homogeneous space, Euclidean
Dual space, 35–36 space En
Dunne, Edward, xv Exterior derivative
of a real-valued function, 35
En , see Euclidean space
of a p-form on Rn , 43–46
Einstein summation convention, 14–15
Einstein, Albert, 85 of a p-form on a manifold, 48–49
Elliptic paraboloid, 272 chain rule, 44
Blaschke representation, 280 Leibniz rule, 43, 44
Elliptic space, 340–342, see also Extrinsic curvature of a surface in S3 or
Homogeneous space, elliptic space H3 , 353
Sn
Elliptic surface Fels, Mark, xi
in A3 , 180–189 First fundamental form
in P3 , 223–232 of a surface in E3 , 118–120
Embedding, 25 of a surface in S3 or H3 , 352
Enneper’s surface, 268 of a surface in a Riemannian
Enneper, Alfred, 261 3-manifold, 382
Equi-affine arc length, see Curve in A3 , of a timelike surface in M1,2 , 150, 163
equi-affine arc length equi-affine, of an elliptic surface in
Equi-affine first fundamental form, see A3 , 181
Surface in A3 , equi-affine first equi-affine, of a hyperbolic surface in
fundamental form A3 , 190
Index 407

projective, of an elliptic surface in P3 , Geodesic equation

227 for curves in S3 or H3 , 349
Flat connection on En , 366 for curves in a Riemannian
Flat homogeneous space, 339 3-manifold, 380
Flat surface Geodesic spray, 380–381
in E3 , 132–134 Grassmannian, affine, 288, 324
in S3 , 355–356 Great hyperboloid in H3 , 351, 355
flat torus, 356 Great sphere in S3 , 351, 355
in H3 , 356–357 Green’s theorem, 55
flat cylinder, 356 Guggenheimer, Heinrich, xi
Frenet, Jean, xi
Frobenius theorem, 46 Hn , see Hyperbolic space
Fubini-Pick form Harmonic function, 264
of a hyperbolic surface in A3 , 191 Helicoid, 261, 268
of an elliptic surface in A3 , 185 Helm, Rachel, xvi
of an elliptic surface in P3 , 226 Hilbert’s theorem, 301–302
Fundamental Theorem of Calculus, 54 Holomorphic function, 263
Fundamental Theorem of Space Curves, Homogeneous space, 70, 84, 361
69 flat homogeneous space, 339
existence, 117 Euclidean space En , 70–75
uniqueness, 114 Minkowski space M1,n , 85–92
equi-affine space An , 92–96
GL(n), 28, 29
projective space Pn , 96–103
gl(n), 30
Gauge, 368 elliptic space Sn , 340–342
Gauge field, 368 hyperbolic space Hn , 340, 342–344
Gauge transformation, 368 Horizontal tangent space, 367
Gauss curvature Horizontal vector field, 380
of a surface in E3 , 131 Hyperbolic paraboloid, 311, 319
of a surface in S3 or H3 , 353 Hyperbolic plane, 301
of a timelike surface in M1,2 , 153, 163 Hyperbolic space, 340, 342–344, see also
Gauss equation Homogeneous space, hyperbolic
for a surface in E3 , 127 space Hn
for a surface in S3 , 353, 354 Hyperbolic surface
for a surface in H3 , 353, 354 in A3 , 180, 189–191
for a timelike surface in M1,2 , 156, in P3 , 223, 232–235
165 Hyperboloid of one sheet, 311
for a submanifold of En+m , 379
Gauss map Immersion, 25
of a surface in E3 , 121 Incidence, of a point and a line, 327
of a surface in S3 or H3 , 352 Indices
of a surface in a Riemannian lower index, 9
3-manifold, 382 upper index, 9
of a timelike surface in M1,2 , 151 in partial derivative operators, 13
Gauss, Carl Friedrich, 131 Inner product
Theorema Egregium, 131 Euclidean, 70
Gelfand, Sergei, xv Minkowski, 86
General linear group, see GL(n) Integrable system, 288
General relativity, 143 soliton solution, 288
Geodesic Interior product, 57
in S3 or H3 , 349 Intrinsic curvature of a surface in S3 or
in a Riemannian 3-manifold, 380 H3 , 353
408 Index

Intrinsic invariant, see Invariant, Line congruence, 288–289

intrinsic invariant for surfaces in E3 focal surface, 289
Invariant, 107 normal congruence, 289
for curves in E3 , 69 pseudospherical congruence, 289–290
for submanifolds of a homogeneous surface of reference, 289
space, 109 Linear fractional transformation, 99
complete set of invariants, 107 Liouville’s equation, 302, 320
for curves in E3 , 115 Bäcklund transformation, 302–303
intrinsic invariant for surfaces in E3 , Local coordinates
131 on a surface, 4, 5
relative invariant, 226, 315 on a manifold, 6
Isometric embedding, 378–379, 383 Local trivialization
Cartan-Janet theorem, 383 of a vector bundle, 32
Isotropy group of a tangent bundle, 364
of a point in En , 73 of an orthonormal frame bundle, 369
of a point in M1,n , 90 Lorentz group, 89
of a point in An , 94 proper, orthochronous, 89
of a point in Pn , 101 Lorentz transformation, 89
of a point in Sn , 340 orthochronous, 89
of a point in Hn , 343 proper, 89

Janet, Maurice, 383 M1,n , see Minkowski space

Jensen, Andrew, xvi Mahoney, Michael, xvi
Joeris, Peter, xvi Manifold, 5
local coordinates, 6
Karpel, Joshua, xvi transition map between, 6
Kaufman, Bryan, xv parametrization, 6
Klein, Felix, 69 Riemannian manifold, 362
Maple, xiii, 59–66, 103–106, 134–141,
Lagrange, Joseph-Louis, 251 166–169, 191–201, 235–247,
Laplace’s equation, 356 280–286, 303–309, 329–335,
Left-hook, 57 357–360, 388–395
Levi-Civita connection, 33, 370–372 Mapping
on En , 366 continuous, 15
on Sn or Hn , 347 differentiable
connection forms, 347 from Rm to Rn , 15
Riemann curvature tensor, 376–378 between manifolds, 18
Lie algebra, 26–32 Mathematical Sciences Research
Lie bracket, 26 Institute, xvi
of vector fields, 27 Maurer-Cartan equation, see also
on a Lie algebra, 28–29 Structure equations
Lie derivative, 56–59, 258 on a Lie group, 85
Cartan’s formula, 58 on the Euclidean group E(n), 82
Lie group, 26–32 on the elliptic symmetry group
left translation map, 26 SO(n + 1), 342
left-invariant vector field, 26–27 on the hyperbolic symmetry group
right translation map, 26 SO+ (1, n), 344
Lifting, 109 Maurer-Cartan form
Light cone, see Minkowski space, light on a Lie group, 85
cone on the Euclidean group E(n), 81–82
Lightlike vector, see Minkowski space, on the Poincaré group M (1, n), 91
lightlike vector on the equi-affine group A(n), 95
Index 409

on the projective symmetry group method of moving frames, 70, 107,

SL(n + 1), 102 111
on the elliptic symmetry group
SO(n + 1), 341 Nash embedding theorem, 378
on the hyperbolic symmetry group National Science Foundation, xvi
SO+ (1, n), 344 Nondegenerate curve, see Curve,
May, Molly, xvi nondegenerate
Mean curvature Null adapted frame field
of a surface in E3 , 131 on a timelike surface in M1,2 , 162
of a surface in S3 or H3 , 353 on a hyperbolic surface in A3 , 190
of a timelike surface in M1,2 , 153, 163 on a hyperbolic surface in P3 , 232
equi-affine, of an elliptic surface in Null cone, see Minkowski space, null
A3 , 185 cone
Measure, 324 Null coordinates on a timelike surface
area measure, 324 in M1,2 , 165
Meromorphic function, 266 Null curve in M1,2 , 165
Method of moving frames, see Moving Null vector, see Minkowski space, null
frame, method of moving frames vector
Metric, 13–14
O(1, n), 89
Metric structure on a curve in En , 209
O(n), 31
Miller, Jonah, xvi
o(n), 31
Minimal surface
Olver, Peter, xi
in E3 , 132, 251–268
Orthogonal group, see O(n)
associated family, 267
Orthonormal basis
catenoid, 128, 260, 268
for En , 72
conjugate surface, 267
for M1,n , 87
Enneper’s surface, 268
Orthonormal frame
helicoid, 261, 268
on En , 75
Weierstrass-Enneper on M1,n , 91
representation, 266–267
on Sn , 341, 345
equi-affine, in A3 , 268–280 on Hn , 343, 345
Blaschke representation, 278 on a Riemannian manifold, 363
elliptic paraboloid, 272, 280 Orthonormal frame bundle
Minkowski cross product, 146 of En , 75
Minkowski norm, 88 of M1,n , 91
Minkowski space, 86, see also of S2 , 34
Homogeneous space, Minkowski of Sn , 341, 345
space M1,n of Hn , 343, 345
future-pointing vector, 87 of a Riemannian manifold, 363
light cone, 87 local trivialization, 369
lightlike vector, 87 Orthonormal frame field
Minkowski norm of a vector, 88 on En , 83
null cone, 87 along a curve in E3 , 111
null vector, 87 along a curve in S3 or H3 , 348
past-pointing vector, 87 along a curve in a Riemannian
spacelike vector, 87 3-manifold, 379
timelike vector, 87 along a timelike curve in M1,2 , 144
world line of a particle, 88
Minkowski, Hermann, 85 p-form
Moving frame on Rn , 42–43
equivariant moving frame, xi on a manifold, 47
410 Index

P GL(m), 98 Projective first fundamental form, see

Pn , see Projective space Surface in P3 , projective first
P SL(m), 98 fundamental form
Paraboloid Projective frame bundle of Pn , 102
elliptic paraboloid, 272 Projective frame field
Blaschke representation, 280 along a curve in P2 , 204
hyperbolic paraboloid, 311, 319 canonical projective frame field,
Parametrization 205
of a surface, 4, 5 along a curve in P3 , 214
of a manifold, 6 canonical projective frame field,
asymptotic, 295 215
conformal, 265 Projective frame on Pn , 101
principal, 128, 156, 187 Projective general linear group, 98
Partial derivative operator Projective parametrization, see Curve
as a tangent vector, 20 in P2 /P3 , projective
indices in, 13 parametrization
Peneyra, Sean, xvi Projective space, 7–9, 96, see also
Pick invariant of an elliptic surface in Homogeneous space, projective
A3 , 186 space Pn
Plateau problem, 251 affine coordinates, 97
Plateau, Joseph, 251 homogeneous coordinates, 8
Poincaré group M (1, n), 90 Projective special linear group, 98
as a principal bundle over M1,n , 91 Projective sphere, 229–232
Poincaré lemma, 46 Projective structure
Poincaré-Hopf theorem, 33, 34 on a curve in P2 , 210
Principal adapted frame field on a curve in P3 , 217
on a surface in E3 , 123 on a curve in Pn , 203
Projective transformation, 96, 97
on a timelike surface in M1,2 , 155
Schwarzian derivative, 208
equi-affine, on an elliptic surface in
Proper time, see Curve in M1,2 , proper
A3 , 186
time
Principal bundle, 33–34, 362
Pseudosphere, 287
base space, 33
Pseudospherical line congruence,
base-point projection map, 33
289–290
fiber, 33
Pseudospherical surface, 287
section, 33
1-soliton pseudospherical surface, 301
total space, 33
asymptotic coordinates, 295
local trivialization, 369
asymptotic parametrization, 295
Principal curvatures
Pullback
of a surface in E3 , 123
for differential forms, 50–53
of a surface in S3 or H3 , 352
for bundles, 108
of a timelike surface in M1,2 , 155
Push-forward, 50
surface in E3 with constant principal
curvatures, 130–131 Quasi-umbilic point on a timelike
Principal vectors surface in M1,2 , 160
on a surface in E3 , 123
on a surface in S3 or H3 , 352 Rational normal curve
on a timelike surface in M1,2 , 155 in A3 , 178
Projective arc length, see Curve in in P3 , 220
P2 /P3 , projective arc length Regular curve, see Curve, regular
Projective curvature form, see Curve in Regular surface, see Surface
P2 /P3 , projective curvature form Relative invariant, 226, 315
Index 411

Relativity on the orthonormal frame bundle of

special relativity, 85, 143 En , 79
general relativity, 143 on the projective frame bundle of Pn ,
Reyes, Enrique, 297 103
Ricci equations for a submanifold of Serret, Joseph, xi
En+m , 379 Simple connectivity, 116
Riemann curvature tensor, 376–378 Sine-Gordon equation, 288
first Bianchi identity, 377 1-soliton solution, 300
on a Riemannian 3-manifold, 385 Bäcklund transformation, 288, 298
Riemannian manifold, 362 in characteristic/null coordinates, 296
Row vector, see Vector, row vector in space-time coordinates, 296
Ruled surface, 311 Skew curvature of a timelike surface in
doubly ruled surface, 311 M1,2 , 154, 163
0-adapted frame field, 314 Smooth manifold, see Manifold
1-adapted frame field, 316 Soliton, 288
2-adapted frame field, 317 1-soliton pseudospherical surface, 301
classification theorem, 313 1-soliton solution of the sine-Gordon
hyperbolic paraboloid, 311, 319 equation, 300
hyperboloid of one sheet, 311 Spacelike surface, see Surface in M1,2 ,
spacelike surface
SL(n), 30–31 Spacelike vector, see Minkowski space,
sl(n), 30 spacelike vector
SL(n + 1) Special affine geometry, see Equi-affine
as a principal bundle over Pn , 102 geometry
as the symmetry group of Pn , 98 Special linear cross product, 277
Sn , 30 Special linear group, see SL(n)
Sn , see Elliptic space; Unit sphere Special orthogonal group, see SO(n)
SO+ (1, n), 89 Special relativity, 85, 143
as a principal bundle over Hn , 344 Stokes’s theorem, 53–55
as the symmetry group of Hn , 342 Divergence theorem, 55
so(1, n), 90 Fundamental Theorem of Calculus,
SO(n), 31 54
SO(n + 1) Green’s theorem, 55
as a principal bundle over Sn , 341 Stokes’s theorem, multivariable
as the symmetry group of Sn , 340 calculus version, 55
Schmidt, Michael, xvi Structure equations
Schwarzian derivative, 208–209 on the orthonormal frame bundle of
of a projective transformation, 208 En , 80
Second fundamental form on the orthonormal frame bundle of
of a surface in E3 , 121–122 M1,n , 91
of a surface in S3 or H3 , 352 on the unimodular frame bundle of
of a surface in a Riemannian An , 95
3-manifold, 382 on the projective frame bundle of Pn ,
of a timelike surface in M1,2 , 151, 163 102
equi-affine, of an elliptic surface in on the orthonormal frame bundle of
A3 , 184 Sn , 341
equi-affine, of a hyperbolic surface in on the orthonormal frame bundle of
A3 , 190 Hn , 344
of a submanifold of En+m , 378 on the orthonormal frame bundle of a
Self-adjoint linear operator, 152 Riemannian manifold, 374, 377
Semi-basic forms Submersion, 25
412 Index

Surface, 3, 5 equi-affine second fundamental

parametrization, 4, 5 form, 184
local coordinates, 4, 5 Fubini-Pick form, 185
transition map between, 5 improper equi-affine sphere, 189
ruled surface, see Ruled surface minimal surface, 268–280
doubly ruled surface, see Ruled Pick invariant, 186
surface, doubly ruled surface proper equi-affine sphere, 189
in E3 variation, 269
adapted frame field, 118 in A3 , hyperbolic surface, 180,
area functional, 252 189–191
Bonnet’s theorem, 127 1-adapted null frame field, 190
catenoid, 128, 260, 268 2-adapted null frame field, 190
Codazzi equations, 127 equi-affine first fundamental form,
compatibility equations, 127 190
Enneper’s surface, 268 equi-affine second fundamental
first fundamental form, 118–120 form, 190
flat surface, 132–134 Fubini-Pick form, 191
hyperbolic paraboloid, 311, 319
Gauss curvature, 131
hyperboloid of one sheet, 311
Gauss equation, 127
in M1,2 , spacelike surface, 143
Gauss map, 121
in M1,2 , timelike surface, 143
helicoid, 261, 268
adapted frame field, 150
mean curvature, 131
Codazzi equations, 156, 165
minimal surface, 132, 251–268
compatibility equations, 156, 165
principal adapted frame field, 123
de Sitter spacetime, 157–158
principal curvatures, 123
first fundamental form, 150, 163
principal vectors, 123
Gauss curvature, 153, 163
pseudosphere, 287 Gauss equation, 156, 165
pseudospherical surface, 287 Gauss map, 151
second fundamental form, 121–122 mean curvature, 153, 163
shape operator, 121 null adapted frame field, 162
surface with constant principal null coordinates, 165
curvatures, 130–131 principal adapted frame field, 155
totally umbilic surface, 129 principal curvatures, 155
umbilic point, 124 principal vectors, 155
variation, 252–255 quasi-umbilic point, 160
in A3 second fundamental form, 151, 163
0-adapted frame field, 178 skew curvature, 154, 163
in A3 , elliptic surface, 180–189 totally quasi-umbilic surface, 160,
1-adapted frame field, 180 165–166
2-adapted frame field, 183 totally umbilic surface, 156–157
compatibility equations, 188 umbilic point, 155
cubic form, 185 in P3
elliptic paraboloid, 272, 280 0-adapted frame field, 221
equi-affine area functional, 269 in P3 , elliptic surface, 223–232
equi-affine first fundamental form, 1-adapted frame field, 223
181 2-adapted frame field, 225
equi-affine mean curvature, 185 3-adapted frame field, 226
equi-affine normal vector field, 183 4-adapted frame field, 228
equi-affine principal adapted frame compatibility equations, 229, 247
field, 186 conformal structure, 224
Index 413

cubic form, 226 totally geodesic surface, 355

Darboux tangents, 227 in a Riemannian 3-manifold
Fubini-Pick form, 226 first fundamental form, 382
projective first fundamental form, Gauss map, 382
227 second fundamental form, 382
projective sphere, 229–232 totally geodesic surface, 383–388
totally umbilic surface, 229–232 Symmetric group, see Sn
umbilic point, 226 Symmetric product
in P3 , hyperbolic surface, 223, of vectors, 39
232–235 of 1-forms, 119
1-adapted null frame field, 232 Symmetry group
2-adapted null frame field, 233 of En , 73
3-adapted null frame field, 234 of M1,n , 90
4-adapted null frame field, 234 of An , 94
compatibility equations, 235 of Pn , 98
conformal structure, 232 of Sn , 340
in S3 of Hn , 342
Bonnet’s theorem, 354 of a homogeneous space G/H, 84
Codazzi equations, 353, 354 as a principal bundle over G/H, 85
compatibility equations, 353 as the set of frames on G/H, 85
extrinsic curvature, 353
first fundamental form, 352 Tangent bundle, 21–23
flat surface, 355–356 of a surface, 21–23
flat torus, 356 of a manifold, 21
Gauss curvature, 353 base space, 22
Gauss equation, 353, 354 total space, 22
Gauss map, 352 fiber, 22
great sphere, 351, 355 base-point projection map, 23
intrinsic curvature, 353 canonical parametrization, 22
mean curvature, 353 transition map between, 22
principal curvatures, 352 local trivialization, 364
principal vectors, 352 Tangent space, 16, 20
second fundamental form, 352 tangent plane, 21
totally geodesic surface, 355 Tangent vector, 16, 19
in H3 Tenenblat, Keti, 297
Bonnet’s theorem, 354 Tensor, 9–14
Codazzi equations, 353, 354 change of basis, 9–10, 12–13
compatibility equations, 353 components, 10, 12, 13
extrinsic curvature, 353 metric, 13
first fundamental form, 352 rank 1, 10
flat cylinder, 356 rank 2, 12
flat surface, 356–357 rank k, 38
Gauss curvature, 353 skew-symmetric, 38–39
Gauss equation, 353, 354 symmetric, 38–39
Gauss map, 352 Tensor bundle, 39
great hyperboloid, 351, 355 Tensor field, 9, 13
intrinsic curvature, 353 rank k, 40
mean curvature, 353 Tensor product, 37–38
principal curvatures, 352 symmetric product, 39
principal vectors, 352 wedge product, 39
second fundamental form, 352 Theorema Egregium (Gauss), 131
414 Index

Timelike curve, see Curve in M1,2 , rank k, 32

timelike curve section, 32–33
Timelike surface, see Surface in M1,2 , global section, 32
timelike surface local section, 32
Timelike vector, see Minkowski space, zero section, 33
timelike vector trivialization
Totally geodesic surface global trivialization, 32
in S3 or H3 , 355 local trivialization, 32
in a Riemannian 3-manifold, 383–388 Vector field, 24–25
Totally quasi-umbilic timelike surface in in local coordinates, 25
M1,2 , 160, 165–166 left-invariant vector field on a Lie
Totally umbilic surface group, 26–27
in E3 , 129 horizontal vector field, 380
in M1,2 , timelike surface, 156–157 Vertical tangent space, 366
in P3 , elliptic surface, 229–232 Volume form, 92
Transition map
between local coordinates on a Wave equation
surface, 5 in characteristic/null coordinates,
between local coordinates on a 302, 355
manifold, 6 in space-time coordinates, 296
Transpose notation Wedge product
for matrices, 31 of vectors, 39
for vectors, 6 of 1-forms, 41
Weierstrass, Karl, 261
Umbilic point Weierstrass-Enneper representation for
on a surface in E3 , 124 a minimal surface in E3 , 266–267
on a timelike surface in M1,2 , 155 Wilczynski invariants
on an elliptic surface in P3 , 226 of a curve in P2 , 206
Unimodular frame bundle of An , 95 of a curve in P3 , 216
Unimodular frame field along a curve in Wilczynski, Ernest, 206
A3 , 172 Wilkens, George, xvi
Unimodular frame on An , 94 World line, see Minkowski space, world
Unit sphere Sn , 6–7 line of a particle

Yu, Yunliang, xiii

Variation
of a surface in E3 , 252–255
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, 32–33
base space, 32
total space, 32
fiber, 32
base-point projection map, 32
Selected Published Titles in This Series
178 Jeanne N. Clelland, From Frenet to Cartan: The Method of Moving Frames, 2017
177 Jacques Sauloy, Differential 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: Differential
Geometry via Moving Frames and Exterior Differential Systems, Second Edition, 2016
174 Alexander Kirillov Jr., Quiver Representations and Quiver Varieties, 2016
173 Lan Wen, Differentiable Dynamical Systems, 2016
172 Jinho Baik, Percy Deift, and Toufic Suidan, Combinatorics and Random Matrix
Theory, 2016
171 Qing Han, Nonlinear Elliptic Equations of the Second Order, 2016
170 Donald Yau, Colored Operads, 2016
169 András Vasy, Partial Differential Equations, 2015
168 Michael Aizenman and Simone Warzel, Random Operators, 2015
167 John C. Neu, Singular Perturbation in the Physical Sciences, 2015
166 Alberto Torchinsky, Problems in Real and Functional Analysis, 2015
165 Joseph J. Rotman, Advanced Modern Algebra: Third Edition, Part 1, 2015
164 Terence Tao, Expansion in Finite Simple Groups of Lie Type, 2015
163 Gérald Tenenbaum, Introduction to Analytic and Probabilistic Number Theory, Third
Edition, 2015
162 Firas Rassoul-Agha and Timo Seppäläinen, 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
Schrödinger Operators, Second Edition, 2014
156 Markus Haase, Functional Analysis, 2014
155 Emmanuel Kowalski, An Introduction to the Representation Theory of Groups, 2014
154 Wilhelm Schlag, A Course in Complex Analysis and Riemann Surfaces, 2014
153 Terence Tao, Hilbert’s Fifth Problem and Related Topics, 2014
152 Gábor Székelyhidi, An Introduction to Extremal Kähler Metrics, 2014
151 Jennifer Schultens, Introduction to 3-Manifolds, 2014
150 Joe Diestel and Angela Spalsbury, The Joys of Haar Measure, 2013
149 Daniel W. Stroock, Mathematics of Probability, 2013
148 Luis Barreira and Yakov Pesin, Introduction to Smooth Ergodic Theory, 2013
147 Xingzhi Zhan, Matrix Theory, 2013
146 Aaron N. Siegel, Combinatorial Game Theory, 2013
145 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/.
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

Photo by Jenna A. Rice

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, equi-
EJ½RI 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 inclu-
sion of detailed guidance regarding the use of the computer algebra system Maple™
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.
—Niky 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 pseudo-
groups and beyond. Cartan theory remains a touchstone in modern differential geometry, and
Clelland’s 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.
—Peter Olver, University of Minnesota

For additional information

and updates on this book, visit
www.ams.org/bookpages/gsm-178

www.ams.org
GSM/178