May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws

This page intentionally left blank

Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

Library of Congress Cataloging-in-Publication Data

Differential geometry from a singularity theory viewpoint / by Shyuichi Izumiya
(Hokkaido University, Japan) [and three others].
pages cm
Includes bibliographical references and index.
ISBN 978-9814590440 (hardcover : alk. paper)
1. Surfaces--Areas and volumes. 2. Singularities (Mathematics) 3. Geometry, Differential.
4. Curvature. I. Izumiya, Shyuichi.
QA645.D54 2015
516.3'6--dc23
2015033184

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

Copyright © 2016 by World Scientific Publishing Co. Pte. Ltd.

All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy
is not required from the publisher.

Printed in Singapore
 Preface

The geometry of surfaces is a subject that has fascinated many mathemati-

cians and users of mathematics. This book o↵ers a new look at this classical
subject, namely from the point of view of singularity theory. Robust ge-
ometric features on a surface in the Euclidean 3-space, some of which are
detectable by the naked eye, can be captured by certain types of singulari-
ties of some functions and mappings on the surface. In fact, the mappings
in question come as members of some natural families of mappings on the
surface. The singularities of the individual members of these families of
mappings measure the contact of the surface with model objects such as
lines, circles, planes and spheres.
This book gives a detailed account of the theory of contact between
manifolds and its link with the theory of caustics and wavefronts. It then
uses the powerful techniques of these theories to deduce geometric informa-
tion about surfaces immersed in the Euclidean 3, 4 and 5-spaces as well as
spacelike surfaces in the Minkowski space-time.
In Chapter 1 we argue the case for using singularity theory to study the
extrinsic geometry of submanifolds of Euclidean spaces (or of other spaces).
To make the book self-contained, we devote Chapter 2 to introducing ba-
sic facts about the extrinsic geometry of submanifolds of Euclidean spaces.
Chapter 3 deals with singularities of smooth mappings. We state the re-
sults on finite determinacy and versal unfoldings which are fundamental
in the study of the geometric families of mappings on surfaces treated in
the book. Chapter 4 is about the theory of contact introduced by Mather
and developed by Montaldi. In Chapter 5 we recall some basic concepts in
symplectic and contact geometry and establish the link between the the-
ory of contact and that of Lagrangian and Legendrian singularities. We
apply in Chapters 6, 7 and 8 the singularity theory framework exposed in

v
vi Di↵erential Geometry from a Singularity Theory Viewpoint

the previous chapters to the study of the extrinsic di↵erential geometry of

surfaces in the Euclidean 3, 4 and 5-spaces respectively. The codimension
of the surface in the ambient space is 1, 2 or 3 and this book shows how
some aspects of the geometry of the surface change with its codimension.
In Chapter 9 we chose spacelike surfaces in the Minkowski space-time to
illustrate how to approach the study of submanifolds in Minkowski spaces
using singularity theory. Most of the results in the previous chapters are
local in nature. Chapter 10 gives a flavour of global results on closed sur-
faces using local invariants obtained from the local study of the surfaces in
the previous chapters.
The emphasis in this book is on how to apply singularity theory to the
study of the extrinsic geometry of surfaces. The methods apply to any
smooth submanifolds of higher dimensional Euclidean space as well as to
other settings, such as affine, hyperbolic or Minkowski spaces. However, as
it is shown in Chapters 6, 7 and 8, each pair (m, n) with m the dimension
of the submanifold and n of the ambient space needs to be considered
separately.
This book is unapologetically biased as it focuses on research results
and interests of the authors and their collaborators. We tried to remedy
this by including, in the Notes of each chapter, other aspect and studies
on the topics in question and as many references as we can. Omissions are
inevitable, and we apologise to anyone whose work is unintentionally left
out.
Currently, there is a growing and justified interest in the study of the
di↵erential geometry of singular submanifolds (such as caustics, wavefronts,
images of singular mappings etc) of Euclidean or Minkowski spaces, and of
submanifolds with induced (pseudo) metrics changing signature on some
subsets of the submanifolds. We hope that this book can be used as a
guide to anyone embarking on the study of such objects.
This book has been used (twice so far!) by the last-named author as
lecture notes for a post-graduate course at the University of São Paulo,
in São Carlos. We thank the following students for their thorough read-
ing of the final draft of the book: Alex Paulo Francisco, Leandro Nery
de Oliveira, Lito Edinson Bocanegra Rodrı́guez, Martin Barajas Sichaca,
Mostafa Salarinoghabi and Patricia Tempesta. Thanks are also due to
Catarina Mendes de Jesus for her help with a couple of the book’s figures
and to Asahi Tsuchida, Shunichi Honda and Yutaro Kabata for correcting
some typos. Most of the results in Chapter 4 are due to James Montaldi.
We thank him for allowing us to reproduce some of his proofs in this book.
 Preface vii

We are also very grateful to Masatomo Takahashi for reading the final draft
of the book and for his invaluable comments and corrections.

S. Izumiya, M. C. Romero Fuster, M. A. S. Ruas and F. Tari

August, 2015
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws

This page intentionally left blank

 Contents

Preface v
1. The case for the singularity theory approach 1
1.1 Plane curves . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.1 The evolute of a plane curve . . . . . . . . . . . . 3
1.1.2 Parallels of a plane curve . . . . . . . . . . . . . . 5
1.1.3 The evolute from the singularity theory viewpoint 7
1.1.4 Parallels from the singularity theory viewpoint . . 10
1.2 Surfaces in the Euclidean 3-space . . . . . . . . . . . . . . 11
1.2.1 The focal set . . . . . . . . . . . . . . . . . . . . . 13
1.3 Special surfaces in the Euclidean 3-space . . . . . . . . . . 14
1.3.1 Ruled surfaces . . . . . . . . . . . . . . . . . . . . 15
1.3.2 Developable surfaces . . . . . . . . . . . . . . . . 19
1.4 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2. Submanifolds of the Euclidean space 23

2.1 Hypersurfaces in Rn+1 . . . . . . . . . . . . . . . . . . . . 23
2.1.1 The first fundamental form . . . . . . . . . . . . . 24
2.1.2 The shape operator . . . . . . . . . . . . . . . . . 25
2.1.3 Totally umbilic hypersurfaces . . . . . . . . . . . . 29
2.1.4 Parabolic and umbilic points . . . . . . . . . . . . 32
2.2 Higher codimension submanifolds of Rn+r . . . . . . . . . 36
2.2.1 Totally ⌫-umbilic submanifolds . . . . . . . . . . . 40
2.2.2 ⌫-parabolic and ⌫-umbilic points . . . . . . . . . . 41
2.2.3 The canal hypersurface . . . . . . . . . . . . . . . 41

3. Singularities of germs of smooth mappings 45

ix
x Di↵erential Geometry from a Singularity Theory Viewpoint

3.1 Germs of smooth mappings . . . . . . . . . . . . . . . . . 45

3.2 Multi-germs of smooth mappings . . . . . . . . . . . . . . 47
3.3 Singularities of germs of smooth mappings . . . . . . . . . 47
3.4 The Thom-Boardman symbols . . . . . . . . . . . . . . . 51
3.5 Mather’s groups . . . . . . . . . . . . . . . . . . . . . . . 51
3.6 Tangent spaces to the G-orbits . . . . . . . . . . . . . . . 52
3.7 Finite determinacy . . . . . . . . . . . . . . . . . . . . . . 55
3.8 Versal unfoldings . . . . . . . . . . . . . . . . . . . . . . . 56
3.9 Classification of singularities . . . . . . . . . . . . . . . . . 59
3.9.1 Germs of functions . . . . . . . . . . . . . . . . . 60
3.9.2 Discriminants and bifurcation sets . . . . . . . . . 64
3.10 Damon’s geometric subgroups . . . . . . . . . . . . . . . . 69
3.11 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4. Contact between submanifolds of Rn 73

4.1 Contact between submanifolds . . . . . . . . . . . . . . . 74
4.2 Genericity . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.3 The meaning of generic immersions . . . . . . . . . . . . . 85
4.4 Contact with hyperplanes . . . . . . . . . . . . . . . . . . 88
4.5 The family of distance squared functions . . . . . . . . . . 90
4.6 The family of projections into hyperplanes . . . . . . . . . 92
4.7 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

5. Lagrangian and Legendrian Singularities 97

5.1 Symplectic manifolds . . . . . . . . . . . . . . . . . . . . . 98
5.1.1 Lagrangian submanifolds and Langrangian maps . 100
5.1.2 Lagrangian singularities . . . . . . . . . . . . . . . 101
5.2 Contact manifolds . . . . . . . . . . . . . . . . . . . . . . 105
5.2.1 Legendrian submanifolds and Legendrian maps . . 107
5.2.2 Legendrian singularities . . . . . . . . . . . . . . . 108
5.3 Graph-like Legendrian submanifolds . . . . . . . . . . . . 113
5.4 Versal unfoldings and Morse families of functions . . . . . 117
5.5 Families of functions on hypersurfaces in Rn . . . . . . . . 119
5.5.1 The family of height functions . . . . . . . . . . . 122
5.5.2 The extended family of height functions . . . . . . 124
5.5.3 The family of distance squared functions . . . . . 127
5.6 Contact from the viewpoint of Lagrangian and Legendrian
singularities . . . . . . . . . . . . . . . . . . . . . . . . . . 128
 Contents xi

5.6.1 Contact of hypersurfaces with hyperplanes . . . . 128

5.6.2 Contact of hypersurfaces with hyperspheres . . . . 132
5.6.3 Contact of submanifolds with hyperplanes . . . . 134

6. Surfaces in the Euclidean 3-space 139

6.1 First and second fundamental forms . . . . . . . . . . . . 139
6.2 Surfaces in Monge form . . . . . . . . . . . . . . . . . . . 145
6.3 Contact with planes . . . . . . . . . . . . . . . . . . . . . 146
6.4 Contact with lines . . . . . . . . . . . . . . . . . . . . . . 159
6.4.1 Contour generators and apparent contours . . . . 160
6.4.2 The generic singularities of orthogonal projections 165
6.5 Contact with spheres . . . . . . . . . . . . . . . . . . . . . 178
6.6 Robust features of surfaces . . . . . . . . . . . . . . . . . 183
6.6.1 The parabolic curve . . . . . . . . . . . . . . . . . 184
6.6.2 The flecnodal curve . . . . . . . . . . . . . . . . . 186
6.6.3 The ridge curve . . . . . . . . . . . . . . . . . . . 189
6.6.4 The sub-parabolic curve . . . . . . . . . . . . . . 194
6.7 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

7. Surfaces in the Euclidean 4-space 201

7.1 The curvature ellipse . . . . . . . . . . . . . . . . . . . . . 202
7.2 Second order affine properties . . . . . . . . . . . . . . . . 207
7.2.1 Pencils of quadratic forms . . . . . . . . . . . . . 211
7.3 Asymptotic directions . . . . . . . . . . . . . . . . . . . . 213
7.4 Surfaces in Monge form . . . . . . . . . . . . . . . . . . . 218
7.5 Examples of surfaces in R4 . . . . . . . . . . . . . . . . . 219
7.6 Contact with hyperplanes . . . . . . . . . . . . . . . . . . 221
7.6.1 The canal hypersurface . . . . . . . . . . . . . . . 225
7.6.2 Characterisation of the singularities of the height
function . . . . . . . . . . . . . . . . . . . . . . . 229
7.7 Contact with lines . . . . . . . . . . . . . . . . . . . . . . 232
7.7.1 The geometry of the projections . . . . . . . . . . 237
7.8 Contact with planes . . . . . . . . . . . . . . . . . . . . . 242
7.9 Contact with hyperspheres . . . . . . . . . . . . . . . . . 246
7.10 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

8. Surfaces in the Euclidean 5-space 251

8.1 The second order geometry of surfaces in R5 . . . . . . . . 252
xii Di↵erential Geometry from a Singularity Theory Viewpoint

8.2 Contacts with hyperplanes . . . . . . . . . . . . . . . . . . 259

8.3 Orthogonal projections onto hyperplanes, 3-spaces and
planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
8.3.1 Contact with lines . . . . . . . . . . . . . . . . . . 267
8.3.2 Contact with planes . . . . . . . . . . . . . . . . . 268
8.3.3 Contact with 3-spaces . . . . . . . . . . . . . . . . 270
8.4 Contacts with hyperspheres . . . . . . . . . . . . . . . . . 272
8.5 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277

9. Spacelike surfaces in the Minkowski space-time 281

9.1 Minkowski space-time . . . . . . . . . . . . . . . . . . . . 283
9.1.1 The hyperbolic space and the Poincaré ball model 284
9.2 The lightcone Gauss maps . . . . . . . . . . . . . . . . . . 285
9.3 The normalised lightcone Gauss map . . . . . . . . . . . . 291
9.4 Marginally trapped surfaces . . . . . . . . . . . . . . . . . 292
9.5 The family of lightcone height functions . . . . . . . . . . 293
9.6 The Lagrangian viewpoint . . . . . . . . . . . . . . . . . . 296
9.7 The lightcone pedal and the extended lightcone height
function: the Legendrian viewpoint . . . . . . . . . . . . . 300
9.8 Special cases of spacelike surfaces . . . . . . . . . . . . . . 304
9.8.1 Surfaces in Euclidean 3-space . . . . . . . . . . . . 305
9.8.2 Spacelike surfaces in de Sitter 3-space . . . . . . . 305
9.8.3 Spacelike surfaces in Minkowski 3-space . . . . . . 306
9.8.4 Surfaces in hyperbolic 3-space . . . . . . . . . . . 307
9.9 Lorentzian distance squared functions . . . . . . . . . . . 309
9.9.1 Lightlike hypersurfaces . . . . . . . . . . . . . . . 311
9.9.2 Contact of spacelike surfaces with lightcones . . . 313
9.10 Legendrian dualities between pseudo-spheres . . . . . . . 315
9.11 Spacelike surfaces in the lightcone . . . . . . . . . . . . . 317
9.11.1 The Lightcone Theorema Egregium . . . . . . . . 320
9.12 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325

10. Global viewpoint 329

10.1 Submanifolds of Euclidean space . . . . . . . . . . . . . . 330
10.1.1 Surfaces in R3 . . . . . . . . . . . . . . . . . . . . 330
10.1.2 Wavefronts . . . . . . . . . . . . . . . . . . . . . . 333
10.1.3 Surfaces in R4 . . . . . . . . . . . . . . . . . . . . 335
10.1.4 Semiumbilicity . . . . . . . . . . . . . . . . . . . . 337
 Contents xiii

10.2 Spacelike submanifolds of Minkowski space-time . . . . . 339

10.3 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343

Bibliography 347
Index 363
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws

This page intentionally left blank

 Chapter 1

The case for the singularity theory

approach

The study of curves and surfaces in the Euclidean space is a fascinating and
important subject in di↵erential geometry. We highlight in this chapter how
singularity theory can be used not only to recover classical results on curves
and surfaces in a simpler and more elegant way but also how it reveals the
rich and deep underlying concepts involved.
We start with the evolute and parallels of a plane curve. We first use
classical di↵erential geometry techniques to obtain the shape of the evolute
and parallels. We then define the family of distance squared functions on
the plane curve and recover from the singularities type of the members
of this family geometric information about the curve itself. We outline
how to use the Lagrangian and Legendrian singularity theory framework to
deduce properties of the evolute that are invariant under di↵eomorphisms.
We proceed similarly for surfaces in the Euclidean 3-space and consider
the singularities of their focal sets. We deal in the last section with the
singularities of ruled and developable surfaces.
We refer to [do Carmo (1976)] for a detailed study of the di↵erential
geometry of curves and surfaces.
Throughout this book, a given map is said to be smooth (or C 1 ) if its
partial derivatives of all order exist and are continuous.
The Euclidean n-space is the vector space Rn endowed with the scalar
product
hu, vi = u1 v1 + · · · + un vn
for any u = (u1 , . . . , un ) and v = (v1 , . . . , vn ) in Rn .
We also view the Euclidean n-space as a set of points. The vector
space Rn comes with a standard orthogonal basis e1 = (1, . . . , 0), . . . ,
en = (0, . . . , 1). We choose a point O = (0, . . . , 0) to be the origin and
denote by ⌃ = (O, e1 , . . . , en ) the standard orthonormal coordinates system

1
2 Di↵erential Geometry from a Singularity Theory Viewpoint

in Rn . Then, a point p in the Euclidean n-space is the endpoint of the vector

Op and its coordinates (x1 , . . . , xn ) in the system ⌃ are the coordinates of
the vector Op in the basis {e1 , . . . , en }. Curves, surfaces, submanifolds in
Rn are considered as subsets of points in Rn .
The vector product of n 1 vectors u1 , . . . , un 1 in Rn , is defined by

e1 · · · en
u11 · · · u1n
u1 ⇥ · · · ⇥ un 1 = .. . ,
. · · · ..
un1 1 · · · unn 1

where ui = (ui1 , . . . , uin ). By the property of the determinant, we have

hu, u1 ⇥ · · · ⇥ un 1i = det (u, u1 , . . . , un 1 ).

1.1 Plane curves

A smooth curve in the Euclidean n-space is a smooth map : I ! Rn ,

where I is an open interval of R. The trace of , which we still denote by
, is the set of points (I) in Rn . The curve is said to be regular if 0 (t)
is not the zero vector for any t in I. Points where 0 (t) is the zero vector
are called singular points of .
We consider here smooth and regular plane curves (n = 2 above). We
shall suppose that the curve : I ! R2 is parametrised by arc length and
denote the arc length parameter by s. Then, t(s) = 0 (s) is a unit tangent
vector to . We denote by n(s) the unit normal vector to obtained by
rotating t(s) anti-clockwise by an angle of ⇡/2. It follows from the fact
that ht(s), t(s)i = 1 that ht0 (s), t(s)i = 0, so

t0 (s) = (s)n(s), (1.1)

for some smooth function (s), called the curvature of at s.

We have, similarly, hn0 (s), n(s)i = 0, so n0 (s) = ↵(s)t(s) for some
function ↵(s). Di↵erentiating the identity ht(s), n(s)i = 0 and using (1.1)
gives ↵(s) = (s), so that

n0 (s) = (s)t(s).

We can use (1.1) to deduce that

(s) = ht0 (s), n(s)i.

 The case for the singularity theory approach 3

When the parameter t of the curve is not necessarily the arc length
parameter, the curvature is given by the formula
det( 0 (t), 00 (t))
(t) = .
|| 0 (t)||3
(One can re-parametrise by arc length and use the chain rule to get the
above formula, see for example [Bruce and Giblin (1992)].) If we write
(t) = (x(t), y(t)), then
x0 y 00 x00 y 0
= 3 ,
(x02 + y 02 ) 2
where all the functions are evaluated at t.
The curvature function determines completely the curve up to rigid
motions (i.e., up to translations and rotations about points in the plane).
Indeed,

Theorem 1.1 (Fundamental Theorem of Plane Curves).

Given a smooth function (s) : I ! R, there is a smooth and regular
curve : I ! R2 parametrised by arc length s with curvature (s). The
curve is unique up to rigid motions of R2 .

An inflection point of is a point where (t) = 0. An inflection point

is referred to as an ordinary inflection if (t) = 0 but 0 (t) 6= 0.
A vertex of is a point where (t) 6= 0 and 0 (t) = 0. A vertex is called
an ordinary vertex if (t) 6= 0, 0 (t) = 0 and 00 (t) 6= 0.
We define the following types of singularities of plane curves.

Definition 1.1. (1) A smooth curve : I ! R2 has an ordinary cusp

singularity at t0 2 I if t0 is a singular point of and the vectors 00 (t0 ) and
000
(t0 ) are linearly independent (Figure 1.1, left).
(2) A smooth curve : I ! R2 has a (3, 4)-singularity at t0 2 I if
0
(t0 ) = 00 (t0 ) = (0, 0) and 000 (t0 ) and the fourth derivative vector (4) (t0 )
are linearly independent (Figure 1.1, right).

1.1.1 The evolute of a plane curve

The evolute of a curve : I ! R2 is the plane curve " parametrised by
1
"(t) = (t) + n(t), t 2 I, (1.2)
(t)
4 Di↵erential Geometry from a Singularity Theory Viewpoint

Fig. 1.1 A cusp singularity left and a (3, 4)-singularity right.

where n(t) is the unit normal vector obtained by rotating the unit tangent
vector 0 (t)/|| 0 (t)|| anti-clockwise by ⇡/2.
The evolute is well defined and is a smooth curve away from the inflec-
tion points of . We can use classical di↵erential geometry techniques to
study its geometry.

Proposition 1.1. The evolute of a smooth and regular curve : I ! R2

is a regular curve except at points corresponding to the vertices of . The
evolute has an ordinary cusp singularity at points corresponding to ordinary
vertices of .

Proof. We take parametrised by arc length s. Di↵erentiating (1.2) and

dropping the argument s, we get
0
0
" = n,
2
and this is the zero vector at s0 2 I if and only if 0 (s0 ) = 0, that is, if and
only if has a vertex at s0 .
We obtain by di↵erentiating again
✓ 02 ◆
00 0 2 00 
" = t+ 3
n,

✓ 00  ◆ ✓ 0 4 ◆
000 2  302  000 2 + 600 0  603
" = t+ n.
2 4
At a vertex s0 of , 0 (s0 ) = 0 and the expressions for "00 and "000 at s0
simplify and become
00 (s0 )
"00 (s0 ) = n(s0 ),
2 (s0 )
200 (s0 ) 000 (s0 )
"000 (s0 ) = t(s0 ) n(s0 ).
(s0 ) 2 (s0 )
The vectors "00 (s0 ) and "000 (s0 ) are linearly independent if and only if
00 (s0 ) 6= 0. Thus, the evolute of has an ordinary cusp singularity if and
only if the corresponding point on is an ordinary vertex of . ⇤
 The case for the singularity theory approach 5

An ellipse has four ordinary vertices and Figure 1.2, left, shows the
evolute of an ellipse with its four ordinary cusps. The vertices of a curve
are points where the curve is most or least curved and these can be detected
(approximately) by the naked eye for the ellipse in Figure 1.2, left. It is
not possible to do so for the ellipse in Figure 1.2, right, as its principal
axes have almost the same length. The ellipse in Figure 1.2, right, looks
like a circle but is not a circle as its evolute is not a point. We can find
the vertices of the ellipse in Figure 1.2, right, by considering the limiting
tangent lines to the evolute at its ordinary cusp. These lines intersect the
ellipse at its vertices.

Fig. 1.2 Evolutes of ellipses: the di↵erence between the lengths of the principal axes of
the ellipse on the left is noticeable, whereas that of the ellipse on the right is negligible.
The ellipse on the right looks like a circle but is not a circle.

1.1.2 Parallels of a plane curve

A parallel (or a wavefront) of a curve : I ! R2 is the curve obtained by
moving each point on along its unit normal vector by a fixed distance d.
When is parametrised by arc length, a parametrisation of a parallel is
given by
⇢d (s) = (s) + dn(s), s 2 I.
We have ⇢0d (s) = (1 d(s))t(s), so a parallel is singular at points where
d = 1/(s). This means that the singular points of a parallel are located
on the evolute of . As d varies, the singular points of the parallels of
6 Di↵erential Geometry from a Singularity Theory Viewpoint

trace the evolute of . Figure 1.3 shows the parallels of an ellipse with their
singular points tracing the evolute of the ellipse.

Fig. 1.3 The parallels of an ellipse. The ellipse and its evolute are drawn in thick.

Proposition 1.2. The s of a smooth and regular curve have an ordinary

cusp singularity at regular points on the evolute. The parallel through an
ordinary singularity of the evolute has a (3, 4)-singularity at that point.

Proof. We have the following successive derivatives of the parametrisa-

tion of a parallel:

⇢0d = (1 d)t,
⇢00d = d0 t + (1 d)n,
⇢000
d = (d00 + 2 (1 d))t + 0 (1 3d)n,
(4)
⇢d = (d000 + 30 (1 2d))t
+( (d00 + 2 (1 d)) + 00 (1 3d) 3d02 )n.
At a singularity s0 of the parallel d = 1/(s0 ), so
00 0 (s0 )
⇢d (s0 ) = t(s0 ),
(s0 )
00 (s0 )
000
⇢d (s0 ) = t(s0 ) 20 (s0 )n(s0 ).
(s0 )
The vectors ⇢00d (s0 ) and ⇢000
d (s0 ) are linearly independent if and only if
0
 (s0 ) 6= 0, equivalently, if and only if (s0 ) is not a vertex of . If this is
 The case for the singularity theory approach 7

the case, the parallel ⇢d with d = 1/(s0 ) has an ordinary cusp singularity
at s = s0 (see Definition 1.1).
Suppose that d = 1/(s0 ) and 0 (s0 ) = 0. Then ⇢0d (s0 ) = ⇢00d (s0 ) = 0
and
00 (s0 )
⇢000
d (s0 ) = t(s0 ),
(s0 )
(4) 000 (s0 )
⇢d (s0 ) = t(s0 ) 300 (s0 )n(s0 ).
(s0 )
(4)
The vectors ⇢000
d (s0 ) and ⇢d (s0 ) are linearly independent if and only if
00
 (s0 ) 6= 0, equivalently, if and only if (s0 ) is an ordinary vertex of . If
this is the case, the parallel ⇢d with d = 1/(s0 ) has a (3, 4)-singularity at
s0 (see Definition 1.1). ⇤

1.1.3 The evolute from the singularity theory viewpoint

In sections 1.1.1 and 1.1.2 we obtained geometric information about the
evolute and parallels of a plane curve by direct computation of the succes-
sive derivatives of a parametrisation of the curve. This method has several
limitations. For instance, it does not explain which singularities could ap-
pear in the evolute and parallels and how these bifurcate as the original
curve is deformed. It also misses to capture the deep concepts involved.
We outline these concepts in this section.
We consider the contact (see Chapter 4) of a smooth and regular plane
curve : I ! R2 with circles. A circle of centre a and radius r is the level
set Da (p) = r2 of the distance squared function Da : R2 ! R, given by
Da (p) = ||p a||2 = hp a, p ai.
The contact of with the level sets of Da can be measured by the
vanishing of successive derivatives of the function
g(s) = Da ( (s)) = h (s) a, (s) ai.
A point (s0 ) is on a circle C of centre a and radius r if and only if
g(s0 ) = r2 . The curve and the circle C have an ordinary tangency at
(s0 ) if and only if g(s0 ) = r2 , g 0 (s0 ) = 0 and g 00 (s0 ) 6= 0. Higher orders of
tangency between and C are captured by the vanishing of the successive
derivatives of g at s0 .

Definition 1.2. We say that and C have k + 1-point contact at s0 if

g (i) (s0 ) = 0 for i = 1, . . . , k but g (k+1) (s0 ) 6= 0. Then s0 is said to be a
8 Di↵erential Geometry from a Singularity Theory Viewpoint

Table 1.1 Geometric conditions for the singularities of g.

g Conditions Geometric interpretation
A1 a = (s0 ) + n(s0 ), 6= The centre of the circle C lies on the normal line
1
(s0 )
to at s0 but is not on the evolute of .

1
A2 a = (s0 ) + (s0 )
n(s0 ), The centre of the circle C lies on the evolute of
0 (s0 ) 6= 0 but s0 is not a vertex of .

A3 a = (s0 ) + (s1 ) n(s0 ), The centre of the circle C lies on the evolute of
0
0 (s0 ) = 0, 00 (s0 ) 6= 0 and s0 is an ordinary vertex of .

singularity of g of type Ak . We say that and the circle C have k-point

contact at s0 if g (i) (s0 ) = 0 for i = 1, . . . , k and call s0 a singularity of g of
type A k .

Suppose that g(s0 ) = r2 . We can recover geometric information about

the curve at s0 from the singularity type of the function g at s0 .

Proposition 1.3. Let : I ! R2 be a smooth and regular plane curve and

let C be a circle of centre a and radius r. Suppose that g(s0 ) = r2 for some
s0 2 I. Then g has a singularity of type A1 , A2 or A3 at s0 if and only if
the geometric conditions in Table 1.1 are satisfied.

Proof. We take parametrised by arc length. Then the result follows

by observing that
1 0
2g = ht, ai,
1 00
2g = hn, ai + 1,
1 000 0
2g =  hn, ai 2 ht, ai,
1 (4)
2g = (00 3 )hn, ai 30 ht, ai 2 ,
where all the functions are evaluated at s. ⇤
It is possible to carry on and identify geometrically the singularities
of g of type Ak , with k > 3. However, in general, or to be more precise
for generic curves, the function g has only singularities of type A1 , A2 or
A3 . (The concept of genericity is dealt with in Chapter 4. Intuitively, a
property of an object is generic if it persists when the object is deformed.)
We consider the functions Da , a 2 R2 , all together as members of the
family of distance squared functions D : R2 ⇥ R2 ! R, given by

D(p, a) = Da (p).
 The case for the singularity theory approach 9

The restriction of D to a plane curve is the family D : I ⇥ R2 ! R,

given by
D(s, a) = h (s) a, (s) ai.
The catastrophe set of the family D is defined to be the set
⇢
@D
CD = (s, a) 2 I ⇥ R2 : (s, a) = 0 ,
@s
and the (local) bifurcation set of D is defined as the set
n o
@2D
BD = a 2 R : 9(s, a) 2 CD such that @s2 (s, a) = 0
2

= a 2 R2 : Da has an A 2 singularity at some s 2 I .

Proposition 1.4. (1) The local bifurcation set of the family of distance
squared functions on is the evolute of .
(2) The catastrophe set CD is a regular surface in I ⇥ R2 . The set of
critical values of the catastrophe map ⇡CD : CD ! R2 , with ⇡CD (s, a) = a,
is the local bifurcation set of the family of distance squared functions.

Proof. The proof of (1) follows from the definition of the bifurcation set
and from Proposition 1.3. As for (2), we prove in Chapter 5 a more general
result that shows that CD is a regular surface. ⇤
We can now outline the underlying singularity theory concepts involved
in the study of the evolute. These are developed in subsequent chapters.

1. The cotangent bundle T ⇤ R2 , with the canonical projection ⇡ : T ⇤ R2 !

R2 to the base space, has the canonical symplectic structure ! =
P2
i=1 dqi ^ dpi .
2. There is a Lagrangian immersion L(D) : CD ! T ⇤ R2 , that is, L(D) is
an immersion and L(D)(CD ) is a Lagrangian surface in T ⇤ R2 .
3. The following diagramme commutes

T< ⇤ R2
L(D)
⇡
✏
CD / R2
⇡CD

so the catastrophe map ⇡CD = ⇡ L(D) is a Lagrangian map.

4. It follows from Proposition 1.4 that the evolute is the set of critical
values of the Lagrangian map ⇡CD , i.e., it is a caustic. As a consequence,
it has only Lagrangian singularities.
10 Di↵erential Geometry from a Singularity Theory Viewpoint

5. Lagrangian surfaces are given by generating families and the family D

of distance squared functions is the generating family of L(D)(CD ).
6. There is an equivalence relation between two generating families of
functions such that if two families are equivalent, then their bifurcation
sets are di↵eomorphic. As a consequence, it is enough to compute the
bifurcation sets of the model families (which have simple expressions)
to deduce the di↵eomorphism models of the evolute. We can then
conclude that the evolute of has an ordinary cusp singularity at points
corresponding to ordinary vertices of .

1.1.4 Parallels from the singularity theory viewpoint

Parallels of a plane curve can also be studied using the family of distance
squared function D on . For each r > 0, consider the family of functions
D̃r : I ⇥ R2 ! R, given by

D̃r (s, a) = D(s, a) r2 .

The set
@ D̃r
⇤ (D̃r ) = {(s, a) 2 I ⇥ R : D̃r (s, a) = @s (s, a) = 0}
⌃± 2

= {(s, a) 2 I ⇥ R2 : a = (s) ± rn(s)}

is a smooth and regular surface in I ⇥ R2 . Consider the projection ⇡2 :

⌃±⇤ (D̃) ! R , given by ⇡2 (s, a) = a. Then ⇡2 (⌃⇤ (D̃r )) is the parallel ⇢±r
2 ±

(i.e., the parallel ⇢±r is the discriminant of the family D̃r ).

We outline below the underlying singularity theory concepts involved in
the study of the parallels. These are also developed in subsequent chapters.

1. The projective cotangent bundle P T ⇤ R2 , with the canonical projection

⇡ : P T ⇤ R2 ! R2 to the base space, has a canonical contact structure.
2. There is a Legendrian immersion L (D̃r ) : ⌃± ⇤ (D̃r ) ! P T R , that is,
⇤ 2

L (D̃r ) is an immersion and L (D̃r )(⌃±⇤ (D̃r )) is a Legendrian surface

in P T R .
⇤ 2

3. The following diagramme commutes

P 9 T ⇤ R2
L (D̃r )
⇡
✏
⌃±
⇤ (D̃r )
/ R2
⇡2

so the projection ⇡2 = ⇡ L (Dr ) is a Legendrian map.

 The case for the singularity theory approach 11

4. It follows that the parallels ⇢±r are wavefronts. As a consequence, they

have only Legendrian singularities.
5. Legendrian surfaces are given by generating families and the family D̃r
is the generating family of L (D̃r )(⌃±⇤ (D̃r )).
6. There is an equivalence relation between two generating families of
functions such that if two such families are equivalent, then their dis-
criminant sets are di↵eomorphic. As a consequence, it is enough to
compute the discriminant sets of the model families to deduce the dif-
feomorphism models of the parallels.
7. To study how the parallels evolve as r varies, we consider the big family
D̃ : I ⇥R2 ⇥R+ ! R given by D̃(s, a, r) = D̃r (s, a). The set ⌃±⇤ (D̃) is a
2
regular 3-dimensional manifold in I ⇥R ⇥R+ . We obtain a Legendrian
immersion L (D̃) : ⌃± ⇤ (D̃) ! P T (R ⇥ R+ ) and a big wavefront ⇡
⇤ 2

L (D)(⌃± ⇤ (D̃)). The individual fronts are recovered by slicing the big
front by the planes r = constant.
8. We deduce that the singularities of the parallel ⇢r of the curve are
ordinary cusps at points ⇢r (s0 ) where r = 1/(s0 ) when s0 is not a
vertex of . The parallels undergo the transitions given by the generic
section of the swallowtail surface (big wavefront) at an ordinary vertex
of (Figure 1.4). See also Figure 1.3 which shows how the parallels to
an ellipse are stacked together at the vertices of the ellipse.

Fig. 1.4 A swallowtail surface (big wavefront) left and its generic sections (right).

1.2 Surfaces in the Euclidean 3-space

Let M be a smooth and regular surface in the Euclidean 3-space R3 . We

consider local properties of M and take at a point p 2 M a local parametri-
sation x : U ! R3 , where U is an open subset of R2 with p 2 x(U ). We
denote by u = (u1 , u2 ) the parameters in U .
Let S 2 denote the unit sphere in R3 . The map N : M ! S 2 which
12 Di↵erential Geometry from a Singularity Theory Viewpoint

assigns to each point p 2 M a unit normal vector N (p) to M at p is called

the Gauss map. (We require M to be orientable if we want to define N on
the whole surface.)
The normal vector N (p) is also orthogonal to the tangent plane TN (p) S 2
of the sphere S 2 at N (p), so we can identify TN (p) S 2 with the tangent plane
Tp M of the surface M at p. With this identification, we can consider the
di↵erential map dNp as a linear operator Tp M ! Tp M .
The shape operator or the Weingarten map at p 2 M is the linear map
Wp : Tp M ! Tp M given by
Wp (w) = dNp (w),
for any w 2 Tp M . The map Wp is a self-adjoint operator, that is, Wp is a
linear map which satisfies
hWp (w1 ), w2 i = hw1 , Wp (w2 )i
for any w1 , w2 2 Tp M .
The scalar product h, i in R3 defines a scalar product in Tp M by re-
striction. Therefore, Tp M admits an orthonormal basis of eigenvectors
v1 , v2 of the shape operator Wp , see Theorem 2.1 in Chapter 2. We have
Wp (v1 ) = 1 v1 and Wp (v2 ) = 2 v2 .
The eigenvalues 1 and 2 of Wp , which are functions of p, are called
the principal curvatures of M at p.
The directions given by the eigenvectors v1 and v2 depend on p and are
called the principal directions.
The determinant K(p) = det(Wp ) = 1 (p)2 (p) of the shape operator
is the Gaussian curvature of the surface M at the point p. A point p 2 M
is called elliptic if K(p) > 0; hyperbolic if K(p) < 0; parabolic if K(p) = 0;
umbilic if 1 (p) = 2 (p) (i.e., if Wp is a multiple of the identity map).
The Gaussian curvature is an intrinsic invariant of the surface.

Theorem 1.2 (Theorema Eugregium of Gauss). Two locally isomet-

ric surfaces in R3 have the same Gaussian curvature.

Remark 1.1. There are non-isometric surfaces in R3 with the same Gaus-
sian curvature. A theorem of Bonnet states that if some compatibility
conditions (the Gauss and Mainardi-Codazi conditions) are imposed on the
coefficients of the first and second fundamental forms of the surface, then
these coefficients determine completely the surface up to isometries. See,
for example, [do Carmo (1976)] for a proof.
 The case for the singularity theory approach 13

1.2.1 The focal set

The focal set of a surface in R3 is the analogue of the evolute of a plane
curve and is defined as follows.
Definition 1.3. Let M be a smooth and regular surface in R3 without
parabolic points, and let x : U ! R3 be a local parametrisation of M . The
focal set (or the evolute) of M is the set "(M ) = "1 (M ) [ "2 (M ) where
1
"i (M ) = {p + N (p), p 2 M }, i = 1, 2.
i (p)
Given a non-umbilic point p 2 M , there are two distinct focal points
corresponding to p, one is on "1 and the other on "2 . These focal points
coincide when p is an umbilic point.
What does the focal set look like?
It is very hard to answer this question satisfactorily using classical di↵er-
ential geometry techniques. Note that away from umbilic points, the focal
set is locally the union of the images of the smooth maps i : U ! R3 ,
given by
1
i (u1 , u2 ) = x(u1 , u2 ) + N (u1 , u2 ), i = 1, 2.
i (u1 , u2 )
The maps i , i = 1, 2, are from a 2-dimensional space to a 3-dimensional
space. Whitney ([Whitney (1944)]) showed that the cross-cap is the only A-
stable local singularity for such maps (A means we are considering smooth
changes of coordinates in the source and target, see Chapter 3). An A-
model for the cross-cap is f (u1 , u2 ) = (u1 , u1 u2 , u22 ), see Figure 1.5. How-
ever, the cross-cap singularity can never appear on the focal set. This
follows, for instance, from the fact that away from the umbilic points of the
surface the focal set has a limiting normal direction at its singular points
but the cross-cap does not have one at its singular point.
Also, at umbilic points, the two focal points coincide and the maps i
are not di↵erentiable at such points. Thus, using the maps i directly will
not lead to a satisfactory understanding of the focal set.
We consider the contact of the surface M with spheres. This is measured
by the singularities of the members of the family distance squared functions
D : U ⇥R3 ! R, with D(u, a) = hx(u) a, x(u) ai. The singularity theory
approach outlined in section 1.1.3 for the study of the evolute of a plane
curve can also be used to study the focal set of a surface in R3 . We give in
Chapter 5 the proof that the focal set is a caustic. As a consequence, it has
only Lagrangian singularities and these are as in Figure 1.6 for a generic
surface in R3 .
14 Differential Geometry from a Singularity Theory Viewpoint

Fig. 1.5 A cross-cap.

!! !$ !# "#% "#"

Fig. 1.6 Models of the focal set of a generic surface in R3 . The fourth and fifth figures
model the focal sets ε1 and ε2 joining at an umbilic point. The first three figures model
the focal sets ε1 or ε2 . One can have the following generic combinations for the pair
(ε1 , ε2 ): (A2 , A2 ), (A2 , A3 ), (A2 , A4 ), (A3 , A3 ) or vice-versa.

1.3 Special surfaces in the Euclidean 3-space

Let γ : I → R3 be a smooth and regular space curve parametrised by arc

length s. Then t(s) = γ ′ (s) is a unit tangent vector to γ at s. The curvature
of γ at s is defined as
κ(s) = ||t′ (s)||.
When κ(s) ̸= 0 for all s ∈ I, the unit principal normal vector of γ at s
is defined as the vector
t′ (s)
n(s) = ′ .
||t (s)||
Then
t′ (s) = κ(s)n(s).
The unit vector b(s) = t(s) × n(s) is called the unit binormal vector of
γ at s. It can be shown that b′ (s) is parallel to n(s), so
b′ (s) = −τ (s)n(s),
 The case for the singularity theory approach 15

for some smooth function ⌧ (s). The scalar ⌧ (s) is called the torsion of
at s.
At each point on where (s) 6= 0, we have a positively oriented Serret-
Frenet frame {t(s), n(s), b(s)} which moves along the curve. This motion
is described by the Frenet-Serret formulae
8 0
< t (s) = (s)n(s)
n0 (s) = (s)t(s) + ⌧ (s)b(s)
: 0
b (s) = ⌧ (s)n(s).
The curvature and torsion functions determine completely the space
curve up to rigid motions. Indeed,

Theorem 1.3 (Fundamental Theorem of Space Curves). Given

smooth functions (s), ⌧ (s) : I ! R, with (s) > 0, there is a smooth
and regular curve : I ! R3 parametrised by arc length s with curvature
(s) and torsion ⌧ (s). The curve is unique up to rigid motions of R3 .

One can use singularity theory, as in the case of plane curves, to obtain
a great deal of geometric information about space curves (see for example
[Bruce and Giblin (1992)]). We consider below surfaces in R3 generated by
space curves.

1.3.1 Ruled surfaces

A ruled surface in R3 is one which is swept out by a family of lines (the
rulings) moving along a space curve (the directrix curve or the base curve).
Ruled surfaces are extensively used in architecture. (The reader can surf
the internet to see pictures of well known buildings parts of which have the
shape of ruled surfaces.) Ruled surfaces are also used in abstract art (see
for example some of Barbara Hepworth’s work, available on the web).
Let : I ! R3 denote the base curve and assume that it is a smooth
curve (it needs not be a regular curve) and let : I ! R3 \ {(0, 0, 0)} be
a smooth map which parametrises the directions of the rulings. We take,
without loss of generality, k (t)k = 1 for all t 2 I. Then the ruled surface
M determined by and is the surface parametrised by
x(t, u) = (t) + u (t)
with t 2 I and u 2 R. The partial derivatives of x are given by
xt (t, u) = 0 (t) + u 0 (t),
xu (t, u) = (t),
16 Di↵erential Geometry from a Singularity Theory Viewpoint

and we get
0
xt ⇥ xu (t, u) = (t) ⇥ (t) + u 0 (t) ⇥ (t).
The surface M is singular at (t0 , u0 ) if and only if xt (t0 , u0 ) and
xu (t0 , u0 ) are linearly dependent, equivalently, if and only if xt ⇥ xu (t0 , u0 )
is the zero vector. This occurs when
0
(t0 ) ⇥ (t0 ) + u0 0 (t0 ) ⇥ (t0 ) = (0, 0, 0).
The surface M can be re-parametrised so as to make the singular points
easier to detect. We shall assume that 0 (t) 6= 0 for all t 2 I; such surfaces
are referred to as non-cylindrical ruled surfaces.
Given a non-cylindrical ruled surface M determined by and , there
is a curve : I ! R3 contained in M with
h 0 (t), 0
(t)i = 0 (1.3)
for all t 2 I. To show this, we observe that as the trace of lies on M , we
have
(t) = (t) + u(t) (t) (1.4)
for some smooth function u : I ! R, which we determine as follows. We
have, by di↵erentiating (1.4),
0 0
(t) = (t) + u0 (t) (t) + u(t) 0 (t),
so equation (1.3) becomes
h 0 (t) + u0 (t) (t) + u(t) 0 (t), 0
(t)i = 0. (1.5)
0
The assumption k (t)k = 1 implies h (t), (t)i = 0, so that equation
(1.5) becomes
h 0 (t), 0
(t)i + u(t)h 0 (t), 0
(t)i = 0.
Therefore,
h 0 (t), 0
(t)i
u(t) = . (1.6)
h 0 (t), 0 (t)i

The curve in (1.3) with u(t) as in (1.6) is called the striction curve
of the ruled surface M . It can be shown that the striction curve does not
depend on the choice of the base curve on M (see [do Carmo (1976)]).
Observe that the striction curve need not be a regular curve.
We can now re-parametrise M by taking the striction curve as the
base curve. The new parametrisation of M is given by
y(t, u) = (t) + u (t)
 The case for the singularity theory approach 17

with t 2 I and u 2 R. Then,

yt (t, u) = 0 (t) + u 0 (t),
yu (t, u) = (t),
and
0
yt ⇥ yu (t, u) = (t) ⇥ (t) + u 0 (t) ⇥ (t).
Since h 0 (t), 0
(t)i = 0 and h (t), 0
(t)i = 0, it follows that
0
(t) ⇥ (t) = (t) 0 (t)
where the function : I ! R is given by
h 0 (t) ⇥ (t), 0 (t)i
(t) = . (1.7)
h 0 (t), 0 (t)i
Now,
kyt ⇥ yu (t, u)k2 = ( (t)2 + u2 )k 0 (t)k2 (1.8)
so the singularities of the surfaces M occur when u = 0 and (t) = 0. That
is, the singularities of a ruled surface are located on its striction curve and
at points where (t) = 0.

Example 1.1. The cross-cap parametrised by f (u1 , u2 ) = (u1 , u1 u2 , u22 ) is

a ruled surface. Indeed, we can rewrite f in the form
f (u1 , u2 ) = (0, 0, u22 ) + u1 (1, u2 , 0) = (u2 ) + u1 (u2 ),
with (u2 ) = (0, 0, u22 ) (a singular base curve) and (u2 ) = (1, u2 , 0).
We leave it as an exercise to show that the striction curve is the double
point curve of the cross-cap.

The Gaussian curvature of a ruled surface M is always non-positive, i.e.

K(p)  0 for any p 2 M . In fact, at the regular points of M , we have
(t)2
K(t, u) = , (1.9)
( (t)2 + u2 )2
and this vanishes along the rulings that pass through the singular points of
the surface. It follows that the singular points of a ruled surface are located
at the intersection of the striction curve with the rulings along which the
Gaussian curvature is zero.
Now that we have the location of the singularities of a ruled surface
using tools from classical di↵erential geometry, we can go further and ask
about the nature of such singularities and the shape of the ruled surface at
its singular points. Here, as in the case of the focal set, a ruled surface is
18 Di↵erential Geometry from a Singularity Theory Viewpoint

the image of a map from a 2-dimensional space to a 3-dimensional space.

However, unlike the focal set, the generic singularities of ruled surfaces are
the same as the generic singularities of maps R2 ! R3 .

Theorem 1.4. Generically, a ruled surface has only a cross-cap singularity

(Figure 1.5).

The proof of Theorem 1.4 relies on a special version of the transversal-

ity theorem in the space of ruled surfaces. The proof depends heavily on
Mather’s results (cf. [Mather (1969b,c); Martinet (1982)]), which is given
in [Izumiya and Takeuchi (2001)].

Remark 1.2. More degenerate singularities of ruled surfaces are studied

in [Martins and Nuño-Ballesteros (2009)].

An example of a ruled surface with two cross-cap singularities is the

Plücker conoid. Its base curve is given by (t) = (0, 0, sin(2t)) and its
rulings are along (t) = (cos(t), sin(t), 0), 0  t  2⇡; Figure 1.7.

Fig. 1.7 Plücker conoid.

Another example of a ruled surface with a cross-cap singularity is the

surface swept out by the principal normal lines to a regular space curve.
Let : I ! R3 be a space curve parametrised by arc length and suppose
that (s) 6= 0, for all s 2 I. The principal normal surface is parametrised
by
x(s, u) = (s) + un(s),
with (s, u) 2 I ⇥ R2 . We have
xs ⇥ xu (s, u) = u⌧ (s)t(s) + (1 n(s))n(s),
 The case for the singularity theory approach 19

so the principal normal surface is singular at (s0 , u0 ) if and only if ⌧ (s0 ) = 0

and u0 = 1/(s0 ). The singularity is of type cross-cap if ⌧ 0 (s0 ) 6= 0 (see
[Izumiya and Takeuchi (2003)]).

1.3.2 Developable surfaces

A ruled surface M is called a developable surface if its Gaussian curvature
at all its regular points is zero. It follows from (1.7) that a ruled surface M
is a developable surface if and only if

h 0 (t) ⇥ (t), 0
(t)i = det( 0 (t), (t), 0
(t)) = 0

for all t 2 I.
An example of a developable surface is the tangent developable of a
space curve, whose rulings are along the tangent directions of the space
curve ( (t) is parallel to 0 (t)).
Classically, non-singular developable surfaces are classified as follows
(see [Vaisman (1984)]).

Theorem 1.5. Let x(t, u) = (t) + u (t) be a parametrisation of non-

singular developable surface M with k (t)k = 1. Then M is one of the
following surfaces:
(i) a subset of a plane.
(ii) a subset of a cylindrical surface ( (t) is constant).
(iii) a subset of a conical surface ( (t) is constant).
(iv) a subset of a tangent developable ( (t) is parallel to 0 (t)).

A developable surface can have (stable) singularities which cannot be

removed by deforming it within the space of developable surfaces. We
deduce from (1.8) and (1.9) that the locus of singular points of a developable
surface is precisely its striction curve. For a non-cylindrical developable
surface, the condition det( 0 (t), (t), 0 (t)) = 0 implies that there exist
smooth functions and µ such that 0 (t) = (t) (t) + µ(t) 0 (t), t 2 I.
The functions and µ determine , so the triple ( , µ, ) determines the
developable surface. Therefore, we can identify the space of non-cylindrical
developable surfaces with the space of mappings C 1 (I, R ⇥ R ⇥ S 2 ). We
endow the space C 1 (I, R ⇥ R ⇥ S 2 ) with the Whitney C 1 -topology (see
Chapter 4).
We require the following definition in order to further our understanding
of the singular points of developable surfaces.
20 Di↵erential Geometry from a Singularity Theory Viewpoint

Fig. 1.8 Generic singularities of developable surfaces: cuspidal edge (left), swallowtail
(centre) and cuspidal cross-cap (right).

Definition 1.4. Let U, U 0 be open subsets of R2 and V, V 0 be open subsets

of R3 . A surface x : U ! V is a cuspidal edge, swallowtail or cuspidal
cross-cap if there exist di↵eomorfisms h : U ! U 0 and k : V ! V 0 such
that k x h 1 is given in the following form:

(i) cuspidal edge: (s, t2 , t3 ) (Figure 1.8, left),

(ii) swallowtail: (3s4 + s2 t, 4s3 + 2st, t) (Figure 1.8, centre),
(iii) cuspidal cross-cap: (s3 , s3 t3 , t2 ) (Figure 1.8, right).

A non-cylindrical developable surface is the tangent developable surface

of its striction curve (even when the striction curve is singular). One can
then use the classification of the generic singularities of tangent developable
surfaces in [Cleave (1980); Ishikawa (1993); Mond (1989); Shcherbak
(1986)] to obtain the classification of the generic singularities of developable
surface.

Theorem 1.6. For a generic ( , µ, ) 2 C 1 (I, R⇥R⇥S 2 ), the singularities

of the corresponding non-cylindrical developable surface are cuspidal edges,
swallowtails and cuspidal cross-caps (Figure 1.8).

Remark 1.3. The result in Theorem 1.6 shows that the generic singulari-
ties of developable surfaces are distinct from those of general ruled surfaces
(compare Theorem 1.4).

We turn now to tangent developable surfaces. Such surfaces have always

singularities along their base curves . As an immediate consequence, the
swallowtail singularity, which is a generic singularity of developable surface
(Theorem 1.6), is not a generic singularity of a tangent developable of a
regular space curve.
 The case for the singularity theory approach 21

In fact, a tangent developable surface is a cuspidal edge surface along

(t) if ⌧ (t) 6= 0 (Figure 1.8, left). It is incredible that the classification of
the generic singularities of tangent developable surfaces was only established
recently. It is shown in [Cleave (1980)] that the tangent developable of a
space curve has a cuspidal cross-cap singularity at (t0 ) if ⌧ (t0 ) = 0 and
⌧ 0 (t0 ) 6= 0. These conditions are generic for space curves, so this makes the
cuspidal cross-cap singularity a generic singularity of tangent developable
surfaces.
One can still define the tangent developable of certain singular space
curves ([Ishikawa (1995)]). Consider for example the space curve (t) =
(t2 , t3 , t4 ), which is singular at t = 0. The direction of 0 (t) is the same as
that of (t) = (2, 3t, 4t2 ) for t 6= 0. The ruled surface

x(t, u) = (t) + u (t) = (t2 + 2u, t3 + 3tu, t4 + 4tu)

is still called the tangent developable surface of the singular space curve
(t) = (t2 , t3 , t4 ) ([Ishikawa (1995)]). This surface has a swallowtail singu-
larity at the origin (this was first observed by Arnol’d in [Arnol’d (1981)]).
We deform the curve within the family of curves ✏ (t) = (t2 , t3 ✏t, t4 ),
with 0 = . The tangent developable of the regular curve ✏ , with ✏ 6= 0,
is

x✏ (t, u) = (t2 + 2tu, t3 ✏t + u(3t2 ✏), t4 + 4t3 u),

and this has a cuspidal cross-cap singularity. Therefore, the swallowtail

singularity is also not a generic singularity of the tangent developable of a
singular space curve.

1.4 Notes

We considered briefly the contact of plane curves with circles and showed
how this gives information about the evolute of the curve. This contact
captures also some infinitesimal symmetry of the curve. The locus of centres
of circles tangent to the curve at two or more points is called the Symmetry
Set (SS) of the curve. The SS is used extensively in computer vision and
shape recognition (see for example [Giblin and Brassett (1985); Siddiqi
and Pizer (2008)] and [Damon (2003, 2004, 2006)]).
The contact of a space curve with lines is measured by the singularities
of its projections to planes. The generic singularities of such projections
are obtained in [David (1983)], and in [Dias and Nuño (2008); Oset Sinha
22 Di↵erential Geometry from a Singularity Theory Viewpoint

and Tari (2010)] is considered the singularities as well as the inflections of

the projected curves.
The contact of a space curve with planes is measured by the singularities
of the height functions. These pick up the points of zero torsion of the
curve. Such points can be counted on some curves (see for example [Nuño
Ballesteros and Romero Fuster (1992, 1993); Sedykh (1992)]).
One question of interest is how to approximate a surface in R3 by a
developable surface. In [Izumiya and Otani (2015)] is considered the ap-
proximation of a surface by the developable surface tangent to it along a
curve.
 Chapter 2

Submanifolds of the Euclidean space

In this chapter, we consider some aspects of the extrinsic geometry of a sub-

manifold M of dimension n of the Euclidean space Rn+r , with r 1. When
r = 1, M is a hypersurface and has locally a well defined unit normal vector
field which can be extended to the whole hypersurface if it is orientable.
This normal vector field defines a map, called the Gauss map, from M to
the unit hypersphere in Rn+1 . A great deal of the extrinsic geometry of
the hypersurface M can be derived from the Gauss map and its derivative
map, the shape operator. The shape operator is, at each point of M , a self
adjoint operator from the tangent space of M at p to itself. Its eigenvalues
are called the principal curvatures and when they all are equal at a given
point p, the point p is called an umbilic point. A hypersurface is totally
umbilic if all its points are umbilic points. Totally umbilic hypersurfaces
form our models of hypersurfaces and the extrinsic geometry of a general
hypersurface in Rn+1 is studied in subsequent chapters by looking at its
contact with these model hypersurfaces.
At each point p of a submanifold M of codimension r > 1, there is an
r-dimensional space of normal vectors to M at p. One can define a shape
operator on M along a fixed unit normal vector field on M . All the concepts
and results on the shape operator on a hypersurface can be carried over for
the shape operator along the chosen normal vector field.

2.1 Hypersurfaces in Rn+1

A hypersurface in Rn+1 is a codimension 1 submanifold. As our study is

local in nature, we consider patches of hypersurfaces given by embeddings
x : U ! Rn of an open subset U of Rn in Rn+1 . We write M = x(U ) and
identify M and U via the embedding x.

23
24 Di↵erential Geometry from a Singularity Theory Viewpoint

Let u = (u1 , . . . , un ) denote the coordinates of a point in U with respect

to the canonical basis of Rn . The tangent space Tp M of M at p = x(u) is an
n-dimensional vector space generated by the linearly independent vectors
of the partial derivatives of x at u. We take these vectors as a basis B(x)
of Tp M , so
B(x) = {xu1 (u), . . . , xun (u)}.

2.1.1 The first fundamental form

The scalar product h, i in Rn+1 induces a scalar product h, ip on Tp M by
restriction. Any tangent vector in Tp M is a vector in Rn+1 , so for two
tangent vectors u and v in Tp M , we have hu, vip = hu, vi. To simplify
notation, we drop the subscript in the notation for induced scalar product
on Tp M and indicate it by h, i.

Definition 2.1. The first fundamental form on M is the quadratic form

Ip : Tp M ! R, given by
Ip (w) = hw, wi = ||w||2 .

The first fundamental form, also called the induced metric on M , ex-
presses the way the hypersurface M inherits the metric of the ambient space
Rn+1 . It is a tool for taking measurements on M , such as the length of
curves and angles between curves on M . For instance, the length of a curve
: [a, b] ! M , is defined as
Z b
l( ) = || 0 (t)||dt.
a
We can express the first fundamental form with respect to the basis
Pi=n
B(x) of Tp M at p = x(u) as follows. For w = i=1 wi xui (u) in Tp M , we
have
Xn
Ip (w) = wi wj gij (u)
i,j=1
with
gij (u) = hxui (u), xuj (u)i.
The functions gij are clearly smooth functions on U and satisfy gij (u) =
gji (u) for all i, j = 1, . . . , n and all u in U . These functions form an n ⇥ n-
symmetric matrix (gij ) and the first fundamental form can be written in
matrix form
Ip (w) = wT (gij (u))w.
 Submanifolds of the Euclidean space 25

Definition 2.2. The functions gij , i, j = 1, . . . , n are called the coefficients

of the first fundamental form of M with respect to the parametrisation
x, and the n ⇥ n-symmetric matrix (gij ) is called the matrix of the first
fundamental form.

It is worth observing that the matrix (gij (u)) is not singular for any u
in U . Indeed, as it is a symmetric matrix, it is conjugate to a diagonal
matrix D. The scalar product is positive definite, so the diagonal entries
of D are all strictly positive numbers. The determinant of (gij (u)), being
equal to the product of the diagonal entries of D, is therefore distinct from
zero.

2.1.2 The shape operator

The way M sits in Rn+1 can be described by the variation of its tangent
spaces, or equivalently, by the variation of a normal vector to M . We fix
an orientation of Rn+1 . Given a hypersurface patch M , we can choose a
unique unit normal vector N (p) on M (i.e., orthogonal to Tp M at all p in
M ) which induces a positive orientation of M . This can always be done
for hypersurface patches. For a general hypersurface, it is necessary that
it is orientable in order to be able to make a consistent choice of a normal
vector field on the whole of M .
Let S n be the unit hypersphere in Rn+1 ,

S n = {x 2 Rn+1 : hx, xi = 1}.

Definition 2.3. Let M be a hypersurface in Rn+1 . The map N : M ! S n

which associates to each point p on M the normal vector N (p) to M at p
is called the Gauss map.

The tangent spaces Tp M and TN (p) S n are n-vector spaces in Rn+1 or-
thogonal to N (p) and can be identified. With this identification, the deriva-
tive map dNp of the Gauss map at p is considered as a linear transformation
Tp M ! Tp M.

Definition 2.4. The linear transformation Wp = dNp : Tp M ! Tp M is

called the shape operator or the Weingarten map of M at p.

Theorem 2.1 (The Principal Axis Theorem).

The transformation Wp : Tp M ! Tp M is a self-adjoint operator, that is,
hWp (u), vip = hu, Wp (v)ip for any u and v in Tp M . As a consequence, it
26 Di↵erential Geometry from a Singularity Theory Viewpoint

has n real eigenvalues i (p), i = 1, . . . , n and Tp M admits an orthonormal

basis {v1 (p), . . . , vn (p)} of eigenvectors of Wp .

Proof. We take a parametrisation x : U ! Rn+1 of M and write p =

x(u). To prove that Wp is a self-adjoint operator it is enough to show
that hWp (xui ), xuj i = hxui , Wp (xuj )i for the vectors in the basis B(x) of
Tp M . Di↵erentiating the identity hN, xui i = 0 with respect to uj gives
hNuj , xui i + hN, xui uj i = 0. Therefore,
hNuj , xui i = hN, xui uj i = hN, xuj ui i = hNui , xuj i,
which implies that Wp is a self-adjoint operator as Wp (xui ) = Nui .
For the remaining part of the proof see for example [Mac Lane and
Birkho↵ (1967)], p.403. ⇤

Definition 2.5. The eigenvalues i (p), i = 1, . . . , n, of Wp are called the

principal curvature of M at p and its eigenvectors are called the principal
directions of M at p. The Gauss-Kronecker curvature of M at p is defined
to be
n
Y
K(p) = det Wp = i (p),
i=1

and its mean curvature at p is defined to be

n
1 1X
H(p) = trace Wp = i (p).
n n i=1

Example 2.1. A hyperplane orthogonal to a non-zero vector u in Rn+1 is

the set of points
H(u, r) = {p 2 Rn+1 : hp, ui = r},
where c is some real number. The Gauss map of H(u, r) is the constant map
N (p) = u/||u||. It follows that the principal curvatures at all points on the
hyperplane are equal to zero. In particular, the Gauss-Kronecker curvature
of a hyperplane is identically zero. We can then describe hyperplanes as
flat objects in Rn+1 .
From the above, one can view the Gauss-Kronecker curvature of a hyper-
surface M as a measurement of the deviation of M from being a hyperplane.

Example 2.2. A hypersphere of centre c in Rn+1 and radius r is the set

of points
S n (c, r) = {p 2 Rn+1 : hp c, p ci = r2 }.
 Submanifolds of the Euclidean space 27

The Gauss map of S n (c, r) is given by N (p) = (p c)/r. Its derivative

map is a multiple of the identity map of Tp M by the scalar 1/r. Therefore,
the principal curvatures at all points on the hypersphere are equal to 1/r.
It follows that the Gauss-Kronecker curvature of the hypersphere is equal
to ( 1)n /rn at all points. In particular, hyperspheres can be described as
round objects.

We can use the fact that Wp is a self-adjoint operator to define a bilinear

symmetric map Tp M ⇥ Tp M ! R by (u, v) 7! hWp (u), vi, and associate to
it a quadratic form IIp : Tp M ! R given by IIp (w) = hWp (w), wi.

Definition 2.6. The quadratic form IIp : Tp M ! R, given by

IIp (w) = hWp (w), wi,

is called the second fundamental form of M at p.

The Gauss map of M parametrised by x : U ! Rn+1 is given by

xu1 (u) ⇥ · · · ⇥ xun (u)
N (u) = , (2.1)
kxu1 (u) ⇥ · · · ⇥ xun (u)k
with u 2 U . The map N is infinitely di↵erentiable. We write dNu (xui ) =
Nui (u).
The second fundamental form is also a smooth function on U and can
be expressed in the basis B(x) of Tp M at p = x(u) as follows. For w =
Pn
i=1 wi xui (u) in Tp M , we have
n
X
IIp (w) = wi wj hij (u)
i,j=1

with

hij (u) = hNui (u), xuj (u)i = hN (u), xui uj (u)i. (2.2)

The functions hij are smooth functions on U for all i, j = 1, . . . , n and

satisfy hij = hji . They form an n ⇥ n-symmetric matrix (hij ) and the
second fundamental form can be written in matrix form

IIp (w) = wT (hij (u))w.

Definition 2.7. The functions hij , i, j = 1, . . . , n are called the coefficients

of the second fundamental form and the matrix (hij ) is called the matrix
of the second fundamental form.
28 Di↵erential Geometry from a Singularity Theory Viewpoint

Theorem 2.2 (The Weingarten formula). The matrix of the shape op-
erator Wp with respect to the basis B(x) of Tp M at p = x(u) is given by
(hji (u)) with
(hji (u)) = (hik (u))(g kj (u))
for (g ij (u)) = (gij (u)) 1 , where the matrices, (gij (u)) and (hij (u)) are,
respectively, those of the first and second fundamental forms of M at p.
That is,
Xn
Nui (u) = hji (u)xuj (u).
j=1

Proof. Since the vectors xu1 (u), . . . , xun (u), N (u) form a basis of Rn+1 ,
there exist scalar functions ↵ij (u) and i (u), i, j = 1, . . . , n such that
Xn
Nui (u) = ↵ij (u)xuj (u) + i (u)N (u).
j=1

Di↵erentiating the identity hN (u), N (u)i = 1, yields hN (u), Nui (u)i =

0, so i (u) = 0 for i = 1, . . . , n and all for u 2 U .
We have, by definition of the coefficients of the second fundamental
form,
Xn n
X
j
hik (u) = hNui (u), xuk (u)i = ↵i (u)hxuj (u), xuk (u)i = ↵ij (u)gjk (u).
j=1 j=1
It follows that
(hik (u)) = (↵ij (u))(gjk (u)),
so
(hji (u)) = (↵ij (u)) = (hik (u))(gjk (u)) 1
= (hik (u))(g kj (u)). ⇤
One consequence of Theorem 2.2 is the following formula for the Gauss-
Kronecker curvature with respect to the parametrisation x of M .

Corollary 2.1. The Gauss-Kronecker curvature at p = x(u) is given by

det(hij (u))
K(u) = .
det(gij (u))
Proof. The matrix of the shape operator in the basis B(x) is ( ij ) in
Theorem 2.2. The Gauss-Kronecker curvature, is by definition, K(u) =
det( ij (u)) and from Theorem 2.2
det(hij (u))
K(u) = det(hik (u)) det(gkj (u)) 1 = .
det(gij (u)) ⇤
 Submanifolds of the Euclidean space 29

2.1.3 Totally umbilic hypersurfaces

René Thom and Ian Porteous ([Thom (1976); Porteous (1983a)]) sug-
gested studying the extrinsic di↵erential geometry of hypersurfaces (and
submanifolds in general) in the Euclidean space by comparing them to
model hypersurfaces. This comparison is done in terms of the contact be-
tween submanifolds and is detailed in Chapter 4. Our model hypersurfaces
are the totally umbilic hypersurfaces which we define below.

Definition 2.8. A point p on a hypersurface M is an umbilic point if all

the principal curvatures at p are equal. Equivalently, a point p is an umbilic
point if the shape operator Wp is a scalar multiple of the identity map of
Tp M . A hypersurface M is called totally umbilic if all its points are umbilic
points.

Example 2.3. It follows from Example 2.1 and Example 2.2 that hyper-
planes and hyperspheres in Rn+1 are totally umbilic hypersurfaces.

Hyperplanes and hypersurfaces are not merely examples of totally um-

bilic hypersurfaces. In fact they are, together with their patches, the only
totally umbilic hypersurfaces in Rn+1 .

Theorem 2.3. Suppose that a hypersurface patch M is totally umbilic with

the principal curvatures at p all equal to (p). Then (p) is independent of
p and is equal to a constant . Furthermore,
(i) if  6= 0, then M is contained in a hypersphere.
(ii) if  = 0, then M is contained in a hyperplane.

Proof. We take a parametrisation x : U ! Rn+1 of the hypersurface and

write p = x(u). Since Wp = (u)1Tp M , the function (u) is di↵erentiable
in U . We have Nui (u) = (u)xui (u), i = 1, . . . , n, so for i, j = 1, . . . , n,
Nui uj (u) = uj (u)xui (u) (u)xui uj (u). (2.3)
The maps x and N are C 1 -maps, therefore Nui uj (u) Nuj ui (u) = 0
and xui uj (u) xuj ui (u) = 0 for i, j = 1, . . . , n. We deduce from this and
from the relations (2.3) that
uj (u)xui (u) ui (u)xuj (u) = 0
for i, j = 1, . . . , n. However, the vectors xu1 (u), . . . , xun (u) are linearly
independent, so ui (u) = 0 for i = 1, . . . , n, that is, (u) is a constant
function . We have two possibilities depending on whether  is zero or
not.
30 Di↵erential Geometry from a Singularity Theory Viewpoint

Suppose that  6= 0 and consider the map from U ! Rn+1 given by

1
u 7! x(u) + N (u).

Di↵erentiating it gives
1 1
xui (u) + Nui (u) = xui (u) + ( xui (u)) = 0.
 
Therefore, x(u)+(1/)N (u) is equal to a constant, say, a. We have then
1
kx(u) ak = ,
||
which means that M is contained in the hypersphere S n (a, 1/||).
If  = 0, Nui (u) = 0 for all u in U , so N is a constant map, say
N (u) = n. We have hxui (u), ni = hxui (u), N (u)i = 0 for all i = 1, . . . , n,
so hx(u), ni is a constant function, that is,
hx(u), ni = c,
for some real number c. This means that M is contained in the hyperplane
H(n, c). ⇤

Totally umbilic hypersurfaces with  6= 0 in Theorem 2.3 can also be

identified via their evolutes. The concept of an evolute of a plane curve
extends naturally to hypersurfaces in Rn+1 . Suppose that all the principal
curvatures of M at all its points are not zero. Then we associate to the
point p the n focal points
1
"i (p) = p + N (p).
i (p)
Definition 2.9. The evolute or focal set of a hypersurface M in Rn+1 with
nowhere vanishing principal curvatures is the set
[n
"(M ) = "i (M ),
i=1
where
1
"i (M ) = {p + N (p), p 2 M }.
i (p)
Remark 2.1. When the principal curvatures are pairwise distinct at a
point p, they are smooth functions near p and the evolute is locally the
union of n disjoint patches of hypersurfaces, parametrised by the smooth
maps "i . (Some components of the evolute can be singular!) If two principal
curvatures i and j coincide at p, then the components of the evolute
"i (M ) and "j (M ) have a non-empty intersection. Also, the maps "i and
"j are no longer di↵erentiable at p.
 Submanifolds of the Euclidean space 31

A parametrisation x : U ! Rn+1 of the hypersurface M induces

parametrisations (not necessarily smooth, see Remark 2.1) "i : U ! Rn+1
of the n components of the evolute, with
1
"i (u) = x(u) + N (u),
i (u)
where we assume i (u) 6= 0 for all u 2 U and i = 1, . . . , n.

Proposition 2.1. Let M be a hypersurface in Rn+1 with nowhere vanishing

principal curvatures. Then the following are equivalent.
(i) M is contained in a hypersphere.
(ii) The evolute "(M ) of M is a point in Rn+1 .

Proof. If M , parametrised by x : U ! Rn+1 , is contained in a hyper-

sphere, then by Theorem 2.3,
1
"i (u) = x(u) + N (u).

In particular, the maps "i are identical for i = 1 . . . , n. The partial
derivatives of "i are identically zero, so "i is a constant map, which implies
that the evolute of M is a point.
Conversely, if the evolute is a point, say a, then "i (u) = a on U . Con-
sequently,
1
= hx(u) a, N (u)i,
i (u)
is independent of i and is a smooth function. We denote i (u) by (u).
Di↵erentiating the map "i and using Theorem 2.2 yields
@"i 1 uj (u)
(u) = xuj (u) + Nuj (u) N (u)
@uj (u) (u)2
n
1 X uj (u)
= ((u) jk kj (u))xuk (u) N (u) = 0,
(u) (u)2
k=1

where jk = 1 if k = j and 0 otherwise and the kj are as in Theorem 2.2.

Thus,  is a constant function and ( kj (u)) = ( jk ). This means that M is
a totally umbilic hypersurface with  6= 0. By Theorem 2.3, M is contained
in a hypersphere.
We remark that for a totally umbilic hypersurface M , the point "(M )
is the centre of the hypersphere which contains M. ⇤
32 Di↵erential Geometry from a Singularity Theory Viewpoint

Another way to identify totally umbilic hypersurfaces with  = 0 in

Theorem 2.3 is via the dual hypersurface of M .

Definition 2.10. The dual of M is the hypersurface in the dual projective

space RP n⇤ formed by all the tangent spaces of M .

We represent the dual hypersurface in the double cover S n ⇥ R of RP n⇤

as follows. Let x : U ! Rn+1 be a parametrisation of M , and define the
map x⇤ : U ! S n ⇥ R by
x⇤ (u) = (N (u), hx(u), N (u)i).
Observe that x⇤ (u) represents the hyperplane H(N (u), hx(u), N (u)i)
which is precisely the tangent space Tp M of M at p = x(u).

Definition 2.11. The image of the map x⇤ is called the cylindrical pedal
of M and represents the dual hypersurface of M in S n ⇥ R ([Bruce (1981);
Romero Fuster (1983)]).

Proposition 2.2. Let M = x(U ) be a hypersurface in Rn+1 . Then the

following are equivalent.
(i) M is contained in a hyperplane.
(ii) The Gauss map of M is a constant map.
(iii) The cylindrical pedal of M is a point.

Proof. By the Weingarten formula in Theorem 2.2, the Gauss map of M

is a constant map if and only if ij (u) are identically zero, for i, j = 1, . . . , n.
This occurs if and only if all the eigenvalues of Wp are zero at all points
on M , that is, if and only if M is a totally umbilic hypersurface with  in
Theorem 2.3 equal to zero. Thus, (i) and (ii) are equivalent.
Now, if M is contained in a hyperplane H(n, c), then
hx(u), ni = c,
for all u in U , so N (u) is the constant vector n or n. It follows that
x⇤ (u) = (N (u), hx(u), N (u)i) = (±n, c) is a constant map, which implies
that the cylindrical pedal of M is a point. Thus, (i) implies (iii).
If the image of x⇤ is a point, then the Gauss map is a constant map.
Therefore, (iii) implies (ii). ⇤

2.1.4 Parabolic and umbilic points

In the previous section we dealt with the property of the whole hypersurface
patch M being totally umbilic (either flat or round). We restrict now to
 Submanifolds of the Euclidean space 33

local properties of M at a given point and use Theorem 2.3 to classify, as

follows, umbilic points on a general hypersurface.

Definition 2.12. An umbilic point p on a hypersurface M in Rn+1 with

principal curvatures i (p), i = 1, . . . , n, all equal to  is said to be a flat
umbilic point if  = 0.

Proposition 2.3. A point p = x(u) on a hypersurface M = x(U ) is a flat

umbilic point if and only if all the coefficients of the second fundamental
form vanish at u, that is, hij (u) = 0 for all i, j = 1, . . . , n.

Proof. The proof is an immediate consequence of Theorem 2.2. ⇤

Points on a hypersurface where the Gauss-Kronecker curvature vanishes
are also of interest. As the Gauss-Kronecker curvature is the product of
the principal curvature, it vanishes at a given point when at least one of
the principal curvatures is zero at that point.

Definition 2.13. A point p on a hypersurface M in Rn+1 is a parabolic

point if the Gauss-Kronecker curvature vanishes at p, that is, if K(p) = 0.
The locus of all the parabolic points on M is called the parabolic set of M .

When M is a surface in R3 , it is possible to identify the shape of M

at a point p according to the sign of the Gaussian curvature K(p). The
point p is hyperbolic if K(p) < 0 and elliptic if K(p) > 0. The parabolic
set separates these two regions on the surface. At an elliptic point p the
two principal curvatures have the same sign. All the curves on M obtained
by slicing M by planes passing through p and containing the normal vector
N (p) bend in the same direction, that is, all their curvatures have the same
sign. Then the surface M is locally at p on one side of its tangent plane.
At a hyperbolic point, the two principal curvatures have opposite signs and
the surface looks like a horse saddle; it has parts on both sides of its tangent
plane. When n > 3, we can have a positive Gauss-Kronecker curvature at
p with an even number of principal curvatures having opposite signs. For
this reason, the concept of a point being elliptic or hyperbolic is reserved
for surfaces in R3 .
The parabolic, umbilic and flat umbilic points on M can be detected
via the families of height functions and distance squared functions on M ,
parametrised by x : U ! Rn+1 . We start with the family of height functions
H : U ⇥ S n ! R, given by
H(u, v) = hx(u), vi.
34 Di↵erential Geometry from a Singularity Theory Viewpoint

We label hv the function defined on U by hv (u) = H(u, v) and call it

the height function on M along v.

Proposition 2.4. Let M be a hypersurface parametrised by x : U ! Rn+1 .

Then, for a given u in U , @hv /@ui (u) = 0 for i = 1, . . . , n if and only if
v = ±N (u).

Proof. We have @hv /@ui (u) = hxui (u), vi. Thus, @hv /@ui (u) = 0 for
i = 1, . . . , n if and only if v is a unit normal vector of M at p = x(u).
Equivalently, if and only if v = ±N (u). ⇤
We fix now v = N (u) at a given point p = x(u) on M . The second
order partial derivatives of hv at u are, for i, j = 1, . . . , n,
@ 2 hv
(u) = hxui uj (u), vi = hxui uj (u), N (u)i = hij (u), (2.4)
@ui @uj
where hij are the coefficients of the second fundamental form as defined
in Definition 2.7. The second order derivatives of Hv at u form an n ⇥ n-
symmetric matrix, called the Hessian matrix of hv at u, which we denote
by
✓ 2 ◆
@ hv
H(hv )(u) = (u) .
@ui @uj
An immediate consequence of expression (2.4) and Corollary 2.1 is the
following.

Corollary 2.2. The Gauss-Kronecker curvature of M at p = x(u) is

det H(hv )(u)
K(u) = .
det(gij (u))
Proposition 2.5. With notation as above and with v = N (u) at p = x(u),
(i) the point p is a parabolic point if and only if det (Hhv )(u) = 0;
(ii) the point p is a flat umbilic point if and only if rank H(hv )(u) = 0.

Proof. Statement (i) follows from Corollary 2.2 as det H(hv )(u) = 0 if
and only if K(u) = 0. Statement (ii) follows from the expressions (2.4) and
Proposition 2.3. ⇤

We turn now to umbilic points and consider the family of distance

squared functions D : U ⇥ Rn ! R on M , given by
D(u, a) = hx(u) a, x(u) ai.
 Submanifolds of the Euclidean space 35

We label da the function defined on U by da (u) = D(u, a).

Proposition 2.6. Let M be a hypersurface parametrised by x : U ! Rn+1 .

Then, for a given u in U , @da /@ui (u) = 0 for i = 1, . . . , n if and only if
there exists a real number such that a = x(u) + N (u).

Proof. The partial derivatives of the distance squared function Da are

@da
(u) = 2hxui (u), x(u) ai, i = 1, . . . , n.
@ui
Thus, @da /@ui (u) = 0 for i = 1, . . . , n if and only if x(u) a is a normal
vector to M at p = x(u). Equivalently, if and only if there exists a real
number such that a = x(u) + N (u). ⇤
We take now a = x(u) + N (u) at p = x(u). The second order partial
derivatives of the distance squared function da are
@ 2 da
(u) = 2(hxui uj (u), x(u) ai + hxuj (u), xuj (u)i)
@ui @uj
= 2( hij (u) + gij (u)), (2.5)
where gij (resp. hij ) are the coefficients of the first (resp. second) funda-
mental form as defined in Definition 2.2 (resp. 2.7). The hessian matrix of
Da at u is denoted by
✓ 2 ◆
@ da
H(da )(u) = (u) .
@ui @uj
Proposition 2.7. Let p = x(u) be a non-parabolic point on M and let
a = x(u) + N (u) for some 2 R. Then p is an umbilic point if and only
if rank H(da )(u) = 0.

Proof. From the expressions (2.5), rank H(da )(u) = 0 if and only if
6= 0 and hij (u) = gij (u)/ for all i = 1, . . . , n. Using matrix notation,
rank H(da )(u) = 0 if and only if
1
(hij (u)) = (gij (u)).

This is equivalent to the matrix (hji (u)) of the shape operator in the
basis B(x) being given by
1
(hji (u)) = (hij (u))(gij (u)) 1
= In ,

where In is the identity matrix. This in turn is equivalent to p being an

umbilic point with all it principal curvatures equal to 1/ . ⇤
36 Di↵erential Geometry from a Singularity Theory Viewpoint

2.2 Higher codimension submanifolds of Rn+r

Let now M be a submanifold of codimension r, with r > 1. At each point p

on M , the tangent space Tp M is an n-dimensional vector subspace of Rn+r .
Here too we have the induced scalar product of Rn+r in Tp M and define
the first fundamental form of M as for the hypersurfaces case.
Let x : U ! Rn+r be an embedding of an open subset U ⇢ Rn in Rn+r .
We write, as in the case of hypersurfaces, M = x(U ) and identify M and
U via the embedding x. The partial derivatives of x at u form a basis of
the tangent space Tp M which we still denote by B(x).
The first fundamental form of M , with respect to the basis B(x), is
given by Ip (w) = wT (gij (u))w where gij (u) = hxui (u), xuj (u)i.
The tangent space Tp M has an orthogonal complement Np M of dimen-
sion r. The space Np M is called the normal space of M at p and we have
Np M = {v 2 Rn+r : hu, vi = 0 for all u 2 Tp M }.
The unit normal space of M at p is defined as
(Np M )1 = {v 2 Np M : hv, vi = 1}.
The normal bundle and the unit normal bundle over M are defined,
respectively, as the sets
[ [
NM = Np M and N1 M = (Np M )1 .
p2M p2M
We have the Whitney sum decomposition of the restriction of the tan-
gent bundle of Rn+r to M
T Rn+r |M = T M N M.
If v 2 Tp R n+r
, we can write v = v + vN , where vT 2 Tp M and
T

vN 2 Np M.
In what follows we denote by X (U ) (resp. N (U )) the set of di↵erentiable
vector fields tangent to M (resp. normal to M ).
Let r be the Riemannian connection on Rn+r . If !1 and !2 are local
vector fields on M, and ! 1 and ! 2 are local extensions to Rn+r , let
T
r! 1 ! 2 = r ! 1 ! 2
be the Riemannian connection relative to the metric induced on M.
Definition 2.14. The second fundamental form of the immersion x : U !
Rn+r is the mapping
B : X (U ) ⇥ X (U ) ! N (U ),
given by B(!1 , !2 ) = r!1 ! 2 r!1 !2 .
 Submanifolds of the Euclidean space 37

Proposition 2.8 (Proposition 2.1, [do Carmo (1992)]). The map-

ping B is bilinear and symmetric.

Definition 2.15. For each p 2 M, the mapping B induces a quadratic

mapping IIp : Tp M ! Np M, called the second fundamental form at p,
defined by
IIp (w) = B(!, !)(p),
where w = !(p), ! 2 X (U ).

We consider smooth sections of the unit normal bundle N1 M , i.e.

smooth normal vector fields ⌫˜ : M ! N1 M on M given by
⌫˜(p) = (p, ⌫(p)).

Definition 2.16. The quadratic form IIvp : Tp M ! R defined by

IIv (!) = hB(!, !), vi , ! 2 T M,
p p

is called the second fundamental form of x at p along the normal vector

v = ⌫(p).

All the concepts and results in section 2.1 derived from the Gauss map
and its derivative can be carried over to similar concepts along a given
normal vector field ⌫˜ on a higher codimension submanifold M of Rn+r .

Definition 2.17. The map G⌫ : M ! S n+r 1 given by G⌫ (p) = ⌫(p) is

called the Gauss map of M with respect to the normal vector field ⌫.

The derivative of the Gauss map G⌫ at a point p in M is a linear map

(dG⌫ )p : Tp M ! Tp Rn+r = Tp M Np M.
Let ⇡ T : Tp M Np M ! Tp M denote the projection to the first com-
ponent given by ⇡ T (u, v) = u, and consider the linear map
Wp⌫ = ⇡T (dG⌫ )p : Tp M ! Tp M.

Definition 2.18. The linear map Wp⌫ is called the shape operator of M at
p along the normal vector field ⌫, or simply, the ⌫-shape operator of M at
p.

Theorem 2.4. The ⌫-shape operator Wp⌫ : Tp M ! Tp M is a self-adjoint

operator. As a consequence, it has n real eigenvalues ⌫i (p), i = 1, . . . , n,
and Tp M admits an orthonormal basis {v1 (p), . . . , vn (p)} of eigenvectors
of Wp⌫ .
38 Di↵erential Geometry from a Singularity Theory Viewpoint

Proof. The proof is similar to that of Theorem 2.1. ⇤

Definition 2.19. The eigenvalues ⌫i (p), i = 1, . . . n, (resp. eigenvectors)

of the ⌫-shape operator Wp⌫ are called the ⌫-principal curvatures (resp. ⌫-
principal directions) at p. The Lipschitz-Killing curvature of M at p with
Qn
respect to ⌫ is defined to be K ⌫ (p) = det Wp⌫ = i=1 ⌫i (p).

In a similar way to the hypersurface case, the second fundamental form

of M along a normal vector field ⌫ is the quadratic form induced by the
⌫-shape operator Wp⌫ .

Proposition 2.9. The following holds

II⌫p (w, w) = hWp⌫ (w), wip .

Proof. The ⌫-second fundamental form can be written, with respect to

the basis B(x), in the form
II⌫p (w) = wT (h⌫ij (u))w
where
h⌫ij (u) = h⌫(u), xui uj (u)i.
As h⌫(u), xui uj (u)i = h ⌫ui (u), xuj (u)ip , the result follows. ⇤
The functions h⌫ij are called the coefficients of the ⌫-second fundamental
form.

Theorem 2.5. Let x : U ! Rn+r be a parametrisation of a submanifold

M in Rn+r of codimension r, and let {⌫1 (u), . . . , ⌫r (u)} be an orthonormal
frame of M , with ⌫1 (u) = ⌫(u). Then,
n
X r
X
⌫
(G )ui = (hji )⌫ xuj + h⌫k , ⌫ui i⌫k
j=1 k=2

with
(hji )⌫ = (h⌫ik )(g kj ),
where all the functions are evaluated at u. As a consequence, we have the
following Weingarten formula along ⌫
n
X
(⇡ T ⌫
G ) ui = (hji )⌫ xuj .
j=1
 Submanifolds of the Euclidean space 39

Proof. Since {xu1 (u), . . . , xun (u), ⌫1 (u), . . . , ⌫r (u)} is a basis of Tp Rn+r
at p = x(u), there exist smooth functions ↵ij and ik on U such that
n
X r
X
G⌫ui = ↵ij xuj + k
i ⌫k .
j=1 k=1

We have hG⌫ , G⌫ i = 1, so that hG⌫ , G⌫ui i = 0, and consequently

hG⌫ , G⌫ui i = h⌫1 , G⌫ui i = 1

i =0

for all i = 1, . . . , n. Furthermore, we have

r
X r
X
k ` `
i = i k` = i h⌫k , ⌫` i = h⌫k , G⌫ui i = h⌫k , ⌫ui i.
`=1 `=1

We have, from the definition of the coefficients of the ⌫-second funda-

mental form,
s
X s
X
h⌫i` = h⌫ui , xu` i = hG⌫ui , xu` i = ↵ij hxuj , xu` i = ↵ij gj` .
j=1 j=1

Thus,

(↵ij ) = (h⌫i` )(g`j ) 1

= (hji )⌫ .

The expression for (⇡ T G⌫ )ui follows from the fact that the vectors
⌫i (u) are normal vectors. ⇤

Corollary 2.3. The Lipschitz-Killing curvature of M at p = x(u) along

the unit normal vector field ⌫ is given by

⌫
det(h⌫ij (u))
K (u) = .
det(g↵ (u))

Remark 2.2. Any unit normal vector ⌫0 at a fixed point p = x(u0 ) can be
extended locally to a unit normal vector field ⌫ along M with ⌫(u0 ) = ⌫0 .
Di↵erentiating the identity h⌫(u), xuj (u)i = 0 and evaluating at u0 yields

h⌫ij (u0 ) = h⌫0 , xuj uj (u0 )i.

It follows that the coefficients of the ⌫-second fundamental at p are indepen-

dent of the local extension ⌫ of ⌫0 . As a consequence, the shape operator
at p depends only on ⌫0 . From this, we deduce that the Lipschitz-Killing
curvature at p is also independent of the local extension of ⌫0 , and write
K ⌫0 (u0 ) for the Lipschitz-Killing curvature with respect to ⌫0 .
40 Di↵erential Geometry from a Singularity Theory Viewpoint

2.2.1 Totally ⌫-umbilic submanifolds

Following section 2.1.3, we make the following definition, where ⌫ is a unit
normal vector field along the submanifold M .

Definition 2.20. We say that a point p on M is an ⌫-umbilic point if

all the principal curvatures of the ⌫-shape operator Wp⌫ are equal. The
submanifold M is totally ⌫-umbilic if all its points are ⌫-umbilic points.

We seek to characterise, as in section 2.1.3, the totally ⌫-umbilic sub-

manifolds. For this we require the following concept.

Definition 2.21. A unit normal vector field ⌫ is said to be parallel at p

if the Np M -component (dG⌫ )p + Wp⌫ of (dG⌫ )p is zero. The unit normal
vector field ⌫ is parallel if it is parallel at all points on M.

Theorem 2.6. Suppose that M is totally ⌫-umbilic submanifold and that

⌫ is parallel. Then ⌫ (p) is a constant function ⌫ . Moreover,
(i) if ⌫ 6= 0, then M is contained in a hypersphere.
(ii) if ⌫ = 0, then M is contained in a hyperplane.

Proof. The condition on ⌫ to be parallel gives ⌫ui (u) = ⇡ T ⌫ui (u), and
adding to it the condition on the submanifold to be totally ⌫-umbilic yields

⌫ui (u) = ⌫ (u)xui (u),

for i = 1, . . . , n. The proof now follows the same steps as that of

Theorem 2.3.
We deduce from the above expression of ⌫ui (u) that ⌫ui uj (u) =
uj (u)xui (u) + ⌫ (u)xui uj (u). Since ⌫ui uj (u) = ⌫uj ui (u) and xui uj (u) =
⌫

xuj ui (u), we have

⌫uj (u)xui (u) ⌫ui xuj (u) = 0.

The vectors xu1 (u), xu2 (u), . . . , xun (u) are lineally independent, so
⌫ui (u) = 0 for i = 1, . . . , n. This means that ⌫ (u) is a constant, say
⌫
 .
Suppose that ⌫ 6= 0. Then the map x(u) + (1/⌫ )⌫(u) is a constant
map a, which implies that M is contained in the hypersphere of centre a
and radius 1/|⌫ |.
If ⌫ = 0, then ⌫ is a constant vector. It follows that hx(u), ⌫i is a
constant map c, so M is contained in the hyperplane H(⌫, c). ⇤
 Submanifolds of the Euclidean space 41

2.2.2 ⌫-parabolic and ⌫-umbilic points

Following the arguments in Remark 2.2, the property of a point p being
⌫-umbilic (Definition 2.20) depends only on the value of ⌫0 at p0 and not
on the vector field ⌫. We call then p a ⌫0 -umbilic point. Likewise, the
point p is a ⌫0 -parabolic point if K ⌫0 (p0 ) = 0. A ⌫0 -flat umbilic point is
⌫0 -umbilic point with K ⌫0 (p0 ) = 0.
We shall characterise these special points using the families of height
functions and distances squared functions. Here, for M parametrised x :
U ! Rn+r , the family of height functions H : U ⇥ S n+r 1 ! R on M is
given by
H(u, v) = hx(u), vi,
where S n+r 1
is the unit sphere in Rn+r . We define the function hv on U
by hv (u) = H(u, v) and denote its Hessian matrix by H(hv )(u).
The following proposition is similar to Proposition 2.4 and
Proposition 2.5 and the proof follows in the same line.
Proposition 2.10. Let M be a submanifold of codimension r in Rn+r
parametrised by x : U ! Rn+r . Then, for any u 2 U , @hv /@ui (u) = 0 for
i = 1, . . . , n if and only if v 2 (Np M )1 at p = x(u).
For v 2 (Np M )1 ,
(i)K v (u) = det H(hv )(u)/det(gij (u));
(ii) p is a v-parabolic point if and only if det Hv )(u) = 0;
(iii) p is a v-flat umbilic point if and only if rank H(hv )(u) = 0.
We have a similar result for the family of distance squared functions
D : U ⇥ Rn+r ! R, defined by D(u, a) = hx(u) a, x(u) ai. We denote
by da the function defined on U by da (u) = D(u, a) and its Hessian matrix
by H(da )(u).
Proposition 2.11. Let M be a submanifold of codimension r in Rn+r
parametrised by x : U ! Rn+r .
(i) For any u 2 U , @da /@ui (u) = 0 for i = 1, . . . , n if and only if there
exist 2 R and v 2 (Np M )1 at p = x(u) such that a = p + v.
(ii) Suppose that p is not a v-parabolic point and that a = p + v. Then,
p is a v-umbilic point if and only if rank H(da )(u) = 0.

2.2.3 The canal hypersurface

For a given submanifold M of codimension r in Rn+r , we can construct a
hypersurface in Rn+r associated to M by taking the union of the (r 1)-
42 Di↵erential Geometry from a Singularity Theory Viewpoint

dimensional spheres in Np M of radius " > 0 and centre p 2 M .

Definition 2.22. The canal hypersurface of M of radius " is the set

CM (") = {p + "⌫ : p 2 M, ⌫ 2 (Np M )1 }.

Theorem 2.7. For " small enough, the canal hypersurface of M is locally
a smooth codimension 1 submanifold of Rn+r .

Proof. We parametrise the submanifold M (locally) by x : U ! Rn+r

and choose an orthonormal frame {⌫1 (u), . . . , ⌫r (u)} of Np M at p = x(u).
Let S r 1 be the unit sphere in Rr and denote the coordinate of its points
by µ = (µ1 , . . . , µr ), with µ21 + µ22 + · · · + µ2r = 1.
At each point p = x(u), the map N (u, ) : S r 1 ! (Np M )1 , given by
X r
N (u, µ) = µi ⌫i (u),
i=1
is a di↵eomorphism.
Let y : U ⇥ S r 1 ! Rn+r be given by y(u, µ) = x(u) + "N (u, µ). We
claim that y is an embedding for " small enough. To show this, we first
compute the partial derivatives of y with respect to the variables ui . These
are given by
yui (u, µ) = xui (u) + "Nui (u, µ)
Pr Pn ⌫i
= xui (u) " i=1 µi j=1 ( ij ) (u)xuj (u)
where ( ij )⌫i are as in Theorem 2.5. As we are considering local properties
of M , we can take U to be a bounded region in Rn . Then, for sufficiently
small " > 0, the vectors yui (u, µ), i = 1, . . . , n, are linearly independent.
To compute the partial derivatives of y with respect to the variables µj ,
we take µ in one of the coordinate charts
Uk+ = {µ = (µ1 , . . . , µr ) 2 S r 1
| µk > 0},
Uk = {µ = (µ1 , . . . , µr ) 2 S r 1
| µk < 0},
q Pr
+
with k = 1, . . . , r. Suppose that µ 2 U1 . Then µ1 = 1 j=2 µ2j , and
Nµj (u, µ) = ⌫j (u) (µj /µ1 )⌫1 (u) for j = 2, . . . , r. It follows that
µj
yµj (u, µ) = "(⌫j (u) ⌫1 (u)).
µ1
The vectors yui (u, µ), i = 1, . . . , n and yµj (u, µ), j = 2 . . . , r are linearly
independent. We obtain the same result if we take µ in the other coordinate
charts of S r 1 . Therefore, the map y is an embedding for sufficiently
small " > 0 (and shrinking U if necessary). Its image is clearly the canal
hypersurface CM ("). ⇤
 Submanifolds of the Euclidean space 43

Proposition 2.12. For " small enough, the Gauss map G : CM (") ! S n
of the canal hypersurface of CM (") of M is given by
G(p, ⌫) = ⌫.
Proof. Following the proof of Theorem 2.7, we take y : U ⇥ S r 1 !
Rn+r , with y(u, µ) = x(u) + "N (u, µ), as a parametrisation of the canal
hypersurface. Then the map G is given by G(u, µ) = N (u, µ). (Observe
that N (u, µ) 2 (Np M )1 , with p = x(u), and (Np M )1 can be considered a
subset of S n .)
We have hN (u, µ), yui (u, µ)i = 0 as N (u, µ) 2 Np M at p = x(u), and
hN (u, µ), yµj (u, µ)i = 0 as the vectors ⌫i (u), i = 1, . . . , r are orthonormal.
It follows that N (u, µ) is a unit orthogonal vector to CM ("). ⇤
Proposition 2.13. Let K(p, ⌫) denote the Gauss-Kronecker curvature of
the canal hypersurface CM (") of M at (p, ⌫) and let K ⌫ (p) denote the
Lipschitz-Killing curvature of M at p along the normal vector ⌫. Then,
K(p, ⌫) = K ⌫ (p).
Proof. We take a parametrisation of M as before and choose an or-
thonormal frame along M in such a way that ⌫(u0 ) = ⌫1 (u0 ) at the
point p = x(u0 ). Then, with the notation of the proof of Theorem 2.7,
N (u0 , µ0 ) = ⌫(u0 ), with µ0 = (1, 0, . . . , 0) 2 U1+ .
By Theorem 2.5, we have, for i = 1, . . . , n,
X n
Nui (u0 , ⌫(u0 )) = ( ij )⌫ (u0 )xuj (u0 )
j=1
and from the proof of Theorem 2.7, for j = 2, . . . , r,
Nµj (u0 , µ0 ) = ⌫j (u0 ).
The matrix of the shape operator of CM (") at (u0 , µ0 ) with re-
spect to the basis {xu1 (u0 ), . . . , xun (u0 ), ⌫1 (u0 ), . . . , ⌫r (u0 )} has determi-
nant det( ( ij )⌫ (u0 )) which is precisely K ⌫ (u0 ). ⇤
Corollary 2.4. Let p be a point on M and let ⌫0 be a unit normal vector
at p. Then p is a ⌫0 -parabolic point of M if and only if (p, ⌫0 ) is a parabolic
point of the canal hypersurface CM (") of M , with " small enough.

Example 2.4 (The canal surface of a space curve). Consider a

smooth curve : I ! R3 parametrised by arc length. A parametrisation of
the canal surface of of radius " is given by y : I ⇥ R ! R3 , with
y(s, ✓) = (s) + "(cos ✓n(s) + sin ✓b(s)),
44 Di↵erential Geometry from a Singularity Theory Viewpoint

where n(s) and b(s) are, respectively, the principal normal and binormal
vectors of at s.
The first order partial derivatives of y are
ys = (1 " cos ✓)t + "⌧ sin ✓n "⌧ cos ✓b
y✓ = "( sin ✓n + cos ✓b)
where , ⌧, t, n, b are evaluated at s. The matrix of the first fundamental
form is given by
✓ ◆
(1 " cos ✓)2 + "2 ⌧ 2 "2 ⌧
(gij ) = .
"2 ⌧ "2
The second order partial derivatives of y are given by
yss = "(0 cos ✓ + ⌧ 2 sin ✓)t + ("⌧ 0 sin ✓ + (1 " cos ✓) "⌧ 2 cos ✓)n
"(⌧ 0 cos ✓ + ⌧ 2 sin ✓)b,
ys✓ = "( sin ✓t + ⌧ cos ✓n + ⌧ sin ✓t),
y✓✓ = "(cos ✓n + sin ✓b).
The Gauss map of the canal surface is given by N (s, ✓) = cos ✓n(s) +
sin ✓b(s), so we can compute the matrix of the second fundamental form
(hij ) using the relations (2.2) and obtain
✓ ◆
(1 " cos ✓) cos ✓ "⌧ 2 "⌧
(hij ) = .
"2 ⌧ "
It follows by Corollary 2.1 that the Gaussian curvature of the canal
surface is given by
(s) cos ✓
K(s, ✓) = .
"(1 "(s) cos ✓)
In particular, the parabolic set of the canal surface consists of the two
curves (s) ± b(s).
 Chapter 3

Singularities of germs of smooth

mappings

The birth of singularity theory can be traced back to the pioneering work
of Whitney [Whitney (1955)] where he showed that maps from the plane
to the plane have, in general, only folds and cusps singularities. Mather
introduced in his seminal papers [Mather (1968, 1969a,b,c, 1970)] groups
that act on the set of map-germs and set the foundations for the study of
finite determinacy of map-germs, a concept linked to the existence of versal
deformations and versal unfoldings. A versal deformation, is in some sense,
one that contains all possible deformations of the initial map-germ. This
has a wide range of applications in mathematics and other fields of science
(see for example [Koenderink (1990); Poston and Stewart (1996); Thom
(1983)]).
We give in this chapter some basic definitions and state the results that
we need in other chapters. For beginners in singularity theory, we rec-
ommend the books [Arnol’d, Guseı̆n-Zade and Varchenko (1985); Bröcker
(1975); Gibson (1979); Martinet (1982)] and [Bruce and Giblin (1992)] for
application to the geometry of curves. C. T. C. Wall’s survey article [Wall
(1981)] remains the first port of call for people embarking on the study of
finite determinacy of map-germs.
The definitions and results are presented for maps f : U ! Rm of class
C 1 from an open subset U of Rn to Rm , but they also hold for smooth
map f : M ! N , where M and N are any smooth dimensional manifolds.

3.1 Germs of smooth mappings

Let X and Y be two subsets of Rn containing a point p 2 Rn . We say that

X is equivalent to Y if there exists an open set U ⇢ Rn containing p such
that X \ U = Y \ U . This defines an equivalence relation among subsets

45
46 Di↵erential Geometry from a Singularity Theory Viewpoint

of Rn containing the point p. The equivalence class of a subset X is called

the germ of X at p and is denoted by (X, p).
Let U and V be two open subsets of Rn containing a point p 2 Rn , and
let f : U ! Rm and g : V ! Rm be two smooth maps. We say that f ⇠ g
if there exists an open set W ⇢ U \ V containing p such that f = g on W ,
that is f |W = g|W .
The relation ⇠ is an equivalence relation and a germ at p of a smooth
map is by definition an equivalent class under this equivalence relation. A
map-germ at p is denoted by
f : (Rn , p) ! Rm .
We shall denote by f˜ : U ! Rm a representative of a germ f in a
neighbourhood U of p.
Sometimes we require that all the elements of the equivalence classes
have the same value at p, say q. Then we write
f : (Rn , p) ! (Rm , q).
Let En denote the set of germs, at the origin 0 in Rn , of smooth functions
(Rn , 0) ! R,
En = {f : (Rn , 0) ! R | f is the germ of a smooth function}.
With the addition and multiplication operations, En becomes a com-
mutative R-algebra with a unit. It has a maximal ideal Mn which is the
subset of germs of functions that vanish at the origin. We have
Mn = {f 2 En | f˜(0) = 0}.
Since Mn is the unique maximal ideal of En , En is a local algebra. (It
is, however, not a Nöether algebra [Gibson (1979)].) If (x1 , . . . , xn ) is a
system of local coordinates of (Rn , 0), then Mn is generated by the germs
of functions xi , i = 1, . . . , n, that is,
Mn = En · {x1 , . . . , xn }.
For a given positive integer k, the kth-power of the maximal ideal Mn
is denoted by Mkn . It is the set of germs of functions f 2 Mn with zero
partial derivatives of order less or equal to k 1 at the origin. We also have
Mkn = En · {xi11 · · · xinn : i1 + · · · + in = k }.
The set of all smooth map-germs f : (Rn , 0) ! Rm is denoted by
E(n, m). It is the direct product of m-copies of En , that is,
E(n, m) = En ⇥ · · · ⇥ En = (En )m .
| {z }
m
 Singularities of germs of smooth mappings 47

We denote by Mk+1 n · E(n, m) = (Mk+1

n )
m
the set of map-germs f :
n m
(R , 0) ! (R , 0) with vanishing partial derivatives of order less or equal
to k at the origin.
The k-jet space of smooth map-germs (Rn , 0) ! (Rm , 0) is defined as
J k (n, m) = Mn · E(n, m)/Mk+1
n · E(n, m).
k k
The map j : Mn · E(n, m) ! J (n, m) assigns to each map-germ f
its kth-jet. An element in J k (n, m) corresponding to a map-germ f 2
Mn · E(n, m) is denoted by j k f .
The set J k (n, m) can be identified with the set of polynomials of degree
less than or equal to k, without the constant terms. Given f 2 Mn ·E(n, m),
j k f is simply its Taylor polynomial of degree k at the origin.

3.2 Multi-germs of smooth mappings

Let S be a collection of s distinct points p1 , . . . , ps in Rn . The equivalence

relation “⇠” in section 3.1 can be extended in a natural way to the collection
S. Let Ui (resp. Vi ), i = 1, . . . , s, be pairwise disjoint open subsets of Rn
with pi 2 Ui \ Vi . Let fi : Ui ! Rm and gi : Vi ! Rm , i = 1, . . . , s
be a collection of smooth maps. Define f = (fi ) : [si=1 Ui ! Rm and
g = (gi ) : [si=1 Vi ! Rm by f |Ui = fi and g|Ui = gi .
We say that f ⇠S g if and only if there exist open sets Wi ⇢ Ui \ Vi con-
taining pi , i = 1, . . . , s, such that fi |Wi = gi |Wi . This is an equivalence rela-
tion and a multi-germ at S is by definition a representative of an equivalent
class under this equivalence relation. We denote by (Rn , S) = [si=1 (Rn , pi )
and write f = (fi ) : (Rn , S) ! Rm for a multi-germ at S. When we require
the multi-germ to fix a point q in the target, we write f : (Rn , S) ! (Rm , q).
We take s independent local coordinate systems of (Rn , pi ), with all
the pi ’s set to be the origin. We denote a multi-germ in this case by
f : (Rn , 0s ) ! Rm and denote by E(n, m)s the set of all these multi-germs.
The multi-jets space J k (n, m)s is defined as the set
J k (n, m)s = (Mn · E(n, m))s /(Mk+1 n · E(n, m))s .
k
An element j fs is the Taylor expansion of degree k of f 2 (Mn · E(n, m))s .

3.3 Singularities of germs of smooth mappings

Let f : U ⇢ Rn ! Rm be a smooth map and denote by df : T U ! T Rm

its derivative map. The map f is singular at p 2 U if the rank of the linear
48 Di↵erential Geometry from a Singularity Theory Viewpoint

map
(df )p : Rn ! Rm
is not maximal, that is, if rank(df )p < min(n, m). The point p is then said
to be a singular point of f . Otherwise, we say that f is non-singular at p
and p is a regular point of f . The critical set of f , denoted by ⌃(f ), is the
set of singular points of f , that is,
⌃(f ) = {p 2 U | rank(df )p < min(n, m)}.
The criminant of f , denoted by Cr(f ), is
Cr(f ) = {p 2 U | rank(df )p < m }.
When n m, Cr(f ) = ⌃(f ), and when n < m, Cr(f ) = U .
The discriminant of f , denoted by (f ), is the image of Cr(f ) by f :
(f ) = f (Cr(f )).
Observe that when n < m, (f ) = f (U ). The set f (⌃(f )) is called
the set of critical values of f. The above definitions can be localised at a
point p 2 Rn . A germ f : (Rn , p) ! Rm is said to be singular if one of
its representatives is singular at p. This definition does not depend on the
choice of the representative of f at p as any two of these are identical in
some neighbourhood of p.
The critical set (respectively, the criminant) of a map-germ f , still de-
noted by ⌃(f ) (respectively, Cr(f )), is the set germ (⌃(f˜), p) (respectively,
(Cr(f˜), p)), where f˜ is a representative of f in some neighbourhood U of p.
Again, this definition does not depend on the choice of the representative.
Likewise, the discriminant of f is (f ) = ( (f˜), f (p)).

Example 3.1. (1) A germ of a function f : (Rn , 0) ! R is singular if

and only if all the partial derivatives of a representative of f vanish at the
origin. Therefore, the set of all singular germs in En is the ideal M2n .
(2) A germ of a curve f : (R, 0) ! Rn is singular if and only if f 0 (0) is
the zero vector.
(3) Consider a map-germ f : (R2 , 0) ! (R2 , 0) of rank 1 in the form
f (x, y) = (x, g(x, y)). The graph of the germ g is the germ of the surface
M in R3 parametrised by the map-germ Gg : (R2 , 0) ! (R3 , 0) given by
Gg (x, y) = (x, y, g(x, y)). Let (u, v, w) denote the coordinates in R3 and let
⇡(u, v, w) = (u, w) be the projection to the (u, w)-plane. Then f = ⇡ Gg ,
so that f is singular if and only if ⇡ restricted to M is singular. (The map-
germ Gg can also be viewed as a 1-parameter unfolding of the map-germ
(y, g(0, y)), with x being the parameter; see section 3.8 for definition.)
 Singularities of germs of smooth mappings 49

Fig. 3.1 The fold (left) and cusp (right) singularities realised by the projection of a
surface to a plane. The singular sets and discriminants are the thick curves.

Consider the following special cases

(a) f (x, y) = (x, y 2 ),
(b) f (x, y) = (x, xy + y 3 ).
The map-germ f (x, y) = (x, y 2 ) is called the fold and the map-germ
f (x, y) = (x, xy + y 3 ) is called the cusp or pleat ([Whitney (1955)]).
For the fold map-germ, the differential map df at (x, y) is represented
by the matrix
! "
1 0
.
0 2y
This is singular if and only if y = 0, so the critical set Σ(f ) is the germ,
at the origin in the source, of the curve y = 0. The discriminant of f is
the germ of the curve w = 0 in the (u, w)-plane. This can be depicted
by considering the projection of the surface M to the (u, w)-plane. The
singular set is the critical set of the projection π|M ; see Figure 3.1, left.
For the cusp map-germ, the differential map df at (x, y) is represented
by the matrix
! "
1 0
.
y x + 3y 2
This is singular if and only if x + 3y 2 = 0, so the singular set Σ(f ) is the
germ, at the origin in the source, of the parabola x + 3y 2 = 0. Then the
discriminant of f is the germ, at the origin in the target, of the cusp curve
27w2 + 4u3 = 0 (Figure 3.1, right). The cusp in the discriminant can be
50 Di↵erential Geometry from a Singularity Theory Viewpoint

seen in Figure 3.1, right, as the image of the critical set of the projection
⇡|M .
(4) Consider the map-germ f : (R2 , 0) ! (R3 , 0) with f (x, y) =
(x2 , y, xy). The di↵erential map df at (x, y) is represented by the matrix
0 1
2x 0
@ 0 1A.
y x

This is singular if and only if x = y = 0, so the critical set ⌃(f ) is the origin
in the source. Its discriminant, which is its image, is as shown in Figure 3.2
and is called a cross-cap or a Whitney umbrella.

Fig. 3.2 A cross-cap.

Let f = (fi ) : (Rn , S) ! Rm be a multi-germ, where S = {p1 , . . . , ps }.

We say that f is singular if fi : (Rn , pi ) ! Rm is singular for some pi 2 S.
The critical set (respectively, the criminant) of f is the union of the
critical sets (respectively, the criminants) of the map-germs fi , i = 1, . . . , s,
that is

⌃(f ) = [si=1 ⌃(fi ), (respectively, Cr(f ) = [si=1 Cr(fi )),

and its discriminant is (f ) = f (Cr(f )).

Example 3.2. Let f1 (x1 , y1 ) = (x21 , y1 ) and f2 (x2 , y2 ) = (x2 , x2 y2 + y23 )

be two smooth map-germs, at the origin, from the plane to the plane and
consider the bi-germ f = (f1 , f2 ). The critical set of f1 is the germ, at the
origin, of the curve x1 = 0. The critical set of f2 is the germ, at the origin,
of the curve x2 + 3y22 = 0. Thus, the discriminant of f is the germ, at the
origin in the target, of the curve u(27w2 4u3 ) = 0.
 Singularities of germs of smooth mappings 51

3.4 The Thom-Boardman symbols

We shall make use of the Thom-Boardman symbols of a smooth map

f : X ! Y between two smooth manifolds X and Y (see Chapter VI
in [Golubitsky and Guillemin (1973)] for details).
The map f is said to have a singularity at p 2 X of type Sr if (df )p
drops rank by r, that is, if rank(df )p = min(dim X, dim Y ) r.
The subset of X of the singularities of f of type Sr is denoted by Sr (f ).
For a generic map f , the subsets Sr (f ) are submanifolds of X of codi-
mension r2 + er, where e = | dim X dim Y | ([Golubitsky and Guillemin
(1973)]) .
When Sr (f ) is a submanifold, we can consider the restriction f |Sr (f ) :
Sr (f ) ! Y . The subset of Sr (f ) where f |Sr (f ) drops rank by s is denoted by
Sr,s (f ). Under some genericity conditions, these subsets are submanifolds
of Sr (f ). We can continue this process and define inductively a nested
family of submanifolds Si1 ,··· ,iq (f ) of X. The points of Si1 ,··· ,iq (f ) are
called singularities of f with Thom-Boardman symbol Si1 ,··· ,iq .

3.5 Mather’s groups

Let R denote the group of germs of di↵eomorphisms (Rn , 0) ! (Rn , 0).

This group is labelled the “right group” and acts smoothly on E(n, m) by
1
h·f =f h
for any h 2 R and f 2 E(n, m).
The “left group” L of germs of di↵eomorphisms (Rm , 0) ! (Rm , 0) acts
smoothly on Mn .E(n, m) by
k·f =k f
for any k 2 L and f 2 Mn · E(n, m).
We denote by A = R ⇥ L the direct product of R and L. The group A
is referred to as the right-left group. It acts smoothly on Mn · E(n, m) by
1
(h, k) · f = k f h
for any (h, k) 2 A and f 2 Mn · E(n, m).
We have another group of interest, namely the contact group K. The
group K is the set of germs of di↵eomorphisms of (Rn ⇥ Rm , (0, 0)) which
can be written in the form H(x, y) = (h(x), H1 (x, y)), with h 2 R and
H1 (x, 0) = 0 for x near 0. We have ⇡ H = h ⇡ where ⇡ : Rn ⇥
52 Di↵erential Geometry from a Singularity Theory Viewpoint

Rm ! Rn is the canonical projection. Thus H is a fibred mapping over the

di↵eomorphism h and preserves the 0-section Rn ⇥ {0}. The set of germs
of di↵eomorphisms of (Rn ⇥ Rm , (0, 0)) in the form (I, H), where I is the
germ at 0 of the identity map of Rn , is denoted by C. The group K is the
semi-direct product of R and C, and we write K = R o C.
The group K acts on Mn .E(n, m) as follows. Given f, g 2 Mn .E(n, m)
and (h, H) 2 K, g = (h, H).f if and only if
1 1
(x, g(x)) = H(h (x), f (h (x))).
The di↵eomorphism H sends the graph of the map-germ f to that of g.

Remark 3.1. The group K is a natural one to use when one seeks to under-
stand the singularities of the zero level-sets of map-germs in Mn .E(n, m).
If two germs are K-equivalent, then their zero level-sets are di↵eomorphic
The action of the group A is finer than that of K. If two map-germs F and
G are A-equivalent, then G = k F h 1 for some (h, k) 2 A, so the level
sets G 1 (c) and F 1 (k 1 (c)) are di↵eomorphic for any c close to 0 2 Rm .
Therefore, the group A preserves also the smooth structure of nearby level
sets to the zero level set.

The groups R, L, A, C, K are known as the Mather’s groups. The groups

R and L can be considered in a natural way as subgroups of A and the
groups C and A (and so R and L) as subgroups of K.
Let G be one of the Mather’s groups. Two germs f, g are said to be
G-equivalent if they are in the same G-orbit.
Let Gk be the subgroup of a Mather group G whose elements have k-jets
the germ of the identity. The group Gk is a normal subgroup of G. Define
J k G = G/Gk . The elements of J k G are the k-jets of the elements of G.
The action of a Mather group G on Mn .E(n, m) induces an action of
J G on J k (n, m) as follows. For j k f 2 J k (n, m) and j k h 2 J k G,
k

j k h.j k f = j k (h.f ).

3.6 Tangent spaces to the G-orbits

Let G be a Lie group acting smoothly on a smooth manifold M . Assume

that the orbits of the action of G are smooth submanifolds of M . Then
the tangent space Tp (G.p) at p 2 M to the orbit G.p of p is well defined.
Also, there exists locally at p a smooth submanifold S of M containing p
such that Tp (G.p) Tp S = Tp M and a submanifold H of G containing the
 Singularities of germs of smooth mappings 53

#$!
!

"

Fig. 3.3 Orbits of a Lie group action.

identity such that H × S is diffeomorphic to a neighbourhood of p in M .

The submanifold S can then be used to parametrise the orbits of G near
G.p (see Figure 3.3 and [Gibson (1979)] for details).
The Mather groups are not Lie groups and E(n, m) is not a finite dimen-
sional manifold (it is not even a Banach manifold). We proceeded as follows
to define the tangent space to an orbit of one of the Mather’s groups G.
Consider the group R acting on En and let R.f denote the R orbit f ∈ En .
Let f ◦ ht , t ∈ (−ϵ, ϵ), be a path in R.f through f , where ht is a path
in R with h0 being the identity map-germ. The germ of the vector field
∂
∂t (f ◦ ht )|t=0 is defined as a “tangent vector” to R.f at f . The union
of these tangent vectors over all paths in R.f through f is defined to be
the “tangent space” to the orbit R.f at f . One can proceed similarly for
the other groups and define algebraically the tangent space to a G-orbit as
follows.
Let π : T Rm → Rm be the tangent bundle over Rm . A map-germ
ξ : Rn , 0 → T Rm is said to be a germ of vector field along f ∈ E(n, m)
if π ◦ ξ = f . The tangent space θf to E(n, m) at f is defined to be the
En -module of germs of vector fields along f .
Let θn = θid(Rn ,0) and θm = θid(Rm ,0) , where id(Rn ,0) and id(Rm ,0) denote
the germs of the identity maps on (Rn , 0) and (Rm , 0) respectively. We
remark that θn is nothing but the set of germs of the vector field on Rn at
the origin. The map

tf : θn → θf
φ &→ df ◦ φ
54 Di↵erential Geometry from a Singularity Theory Viewpoint

is a En -homomorphism, and the map

wf : ✓m ! ✓f
7! f
is a Em -homomorphism via the pull-back homomorphism f ⇤ : Em ! En ,
where f ⇤ (↵) = ↵ f for ↵ 2 Em . (We remark that f ⇤ is a ring homomor-
phism, but it is not an R-algebra homomorphism.)
Let f ⇤ (Mm ) denote the pull-back of the maximal ideal in Em . The
tangent spaces LG · f to the G-orbits of f at the germ f are defined as
follows:
LR · f = tf (Mn .✓n ) LL · f = wf (Mm .✓m ) LC · f = f ⇤ (Mm ).✓f
LA · f = LR · f + LL · f LK · f = LR · f + LC · f
The di↵eomorphisms in a Mather group keep the origin fixed (in the
source and or in the target). When studying deformations (see §3.8), the
singularity can move away from the origin, so the vector fields involved
in defining the tangent spaces are allowed not to fix the origin. For this
reason, the extended tangent spaces are defined as follows:
Le R · f = tf (✓n ) Le L · f = wf (✓m ) Le C · f = f ⇤ (Mm ).✓f
Le A · f = Le R · f + Le L · f Le K · f = L e R · f + L e C · f
If we choose a system of coordinates (y1 , . . . , ym ) in (Rm , 0), the germs
of the vector fields ✓ ◆ ✓ ◆
@ @
f, . . . , f
@y1 @ym
along f form a free basis of ✓f . Then ✓f can be identified canonically with
E(n, m) (✓f is a free En -module of rank m) and we have
n o n o
@f @f @f @f
LR · f = Mn . @x1 , . . . , @xn , Le R · f = En . @x1 , . . . , @xn ,
LL · f = f ⇤ (Mm ).{e1 , . . . , em }, Le L · f = f ⇤ (Em ).{e1 , . . . , em },
LC · f = f ⇤ (Mm ).En .{e1 , . . . , em }, Le C · f = f ⇤ (Mm ).En .{e1 , . . . , em },
where e1 , . . . , em are the standard basis vectors of Rm considered as ele-
ments of E(n, m) and (x1 , . . . , xn ) is a coordinate system in (Rn , 0).
We remark that when m = 1, LC · f = f ⇤ (M1 ).En = En .{f }, so the
group C acts by multiplication by germs of functions in En .
The codimension of the orbit of f is defined by
cod(f, G) = dimR (Mn .E(n, m)/LG · f )
and the codimension of the extended orbit of f is defined by
code (f, G) = dimR (E(n, m)/Le G · f ) .
See [Wall (1981)] for the relation between cod(f, G) and code (f, G). For
example, cod(f, A) = code (f, A) + m.
 Singularities of germs of smooth mappings 55

3.7 Finite determinacy

A germ f is said to be k G-determined if any g with j k g = j k f is G-

equivalent to f (notation: g ⇠G f ). The k-jet of f is then called a sufficient
jet. The least integer k with this property is called the degree of deter-
minacy of f . A G-determined germ is a k G-determined germ for some
integer k.
Some important properties of finitely determined map-germs are the
following ([Wall (1981)], Theorem 1.2).

Theorem 3.1. For each f and G, the following are equivalent:

(i) f is G-determined,
(ii) for some k, Mkn .E(n, m) ⇢ LG · f ,
(iii) cod(f, A) < 1,
(iv) code (f, A) < 1.

A great deal of work is carried out to find out whether a map-germ

is G-finitely determined and to find its degree of determinacy. There is
a detailed account on determinacy of map-germs in C.T.C. Wall survey
article [Wall (1981)], which we refer to for details and references. The
algebraic structure of the tangent space LG · f plays a key role. When G
is R, C or K, the tangent space LG · f is a En -module and one can apply
Nakayama’s Lemma to prove the following.

Theorem 3.2. Let f 2 Mn .E(n, m) and G = R, C or K. Suppose that

Mkn .E(n, m) ⇢ LG · f + Mk+1

n .E(n, m).

Then f is k-G-determined.

We observe that for the group R, there are no R-finitely determined

map-germs in Mn .E(n, m) if m > 1. Therefore, the group R is useful only
when considering germs of functions.
It is shown in [Mather (1969c)] that two map-germs f and g are C-
equivalent if and only if f ⇤ (Mm ).En = g ⇤ (Mm ).En . That leads to the
following result for the group K.

Theorem 3.3. The following assertions are equivalent.

(i) Two map-germs f and g are K-equivalent.
(ii) There exists a germ of a di↵eomorphism : (Rn , 0) ! (Rn , 0) such
that ⇤ (f ⇤ (Mm ).En ) = g ⇤ (Mm ).En .
56 Di↵erential Geometry from a Singularity Theory Viewpoint

(iii) There exists a germ of a di↵eomorphism : (Rn , 0) ! (Rn , 0) such

that induces an R-algebra isomorphism
f⇤ : En /f ⇤ (Mm ).En ! En /g ⇤ (Mm ).En .

When G = L, the tangent space LG · f is an f ⇤ (Em )-module. In fact, if

a map-germ is L-finite, then m 2n ([Wall (1981)]).
Let Gs denote the subgroup of G whose elements have s-jet the identity.

Theorem 3.4. A map-germ f 2 Mn .E(n, m) is k-L1 -determined if and

only if
⇤
Mk+1 k+1 k+1
n .E(n, m) ⇢ LL1 · f + Mn .(f (Mm ).En + Mn )E(n, m).

When G = A, LA · f has a mixed type module structure and this makes

the estimation of the degree of determinacy much harder to deal with. The
question of determining the exact degree of determinacy of a map-germ
is solved in [Bruce, du Plessis and Wall (1987)] by considering unipotent
actions of subgroups of G.
The following corollary of the main determinacy result in [Bruce, du
Plessis and Wall (1987)] can be used in practice to estimate the degree of
A-determinacy of map-germs.

Corollary 3.1 ([Bruce, du Plessis and Wall (1987)]). If f satisfies

Mln .E(n, m) ⇢ LK · f
Mr+1 l+r+1
n .E(n, m) ⇢ LA1 · f + Mn .E(n, m)
then f is r-A1 -determined.

Proof. See [Bruce, du Plessis and Wall (1987)], Corollary 2.5.2. ⇤

3.8 Versal unfoldings

One of the most important concepts in the singularities of map-germs is

that of versal unfoldings and versal deformations.

Definition 3.1. Let f 2 Mn .E(n, m). An a-parameter unfolding (a, F ) of

f is a map-germ
F : (Rn ⇥ Ra , (0, 0)) ! (Rm ⇥ Ra , (0, 0))
in the form F (x, u) = (f¯(x, u), u), with f¯(x, 0) = f (x). The family
f¯ : (Rn ⇥ Ra , (0, 0)) ! (Rm , 0)
is called an a-parameter deformation of f .
 Singularities of germs of smooth mappings 57

It is important to clarify the following about the deformation f˜ in Def-

inition 3.1. Let f¯ : U ⇥ W ! V be a representative of the map-germ f¯,
where U ⇥ W is a neighbourhood of the origin (0, 0) 2 Rn ⇥ Ra and V is
a neighbourhood of the origin in Rm . Denote by f¯u : U ! V the smooth
map given f¯u (x) = f¯(x, u). Then f¯0 (0) = 0 but f¯u (0) is not necessarily the
origin in Rm for u 6= 0. This means that the fibre 0 ⇥ Ra is not necessarily
preserved by F . Also, the singularities of f¯u may no longer be at the origin.
This is why one needs to consider the extended groups Ge .

Definition 3.2. Let G be a Mather group and I the identity in G.

(i) A morphism between two unfoldings (a, F ) and (b, G) is a pair (↵, ) :
(a, F ) ! (b, G) with ↵ : (Ra , 0) ! (G, I), : (Ra , 0) ! (Rb , 0), such that
f¯u = ↵(u) · ḡ (u) .
The unfolding (a, F ) is then said to be induced from (b, G) by (↵, ).
(ii) Two unfoldings (a, F ) and (b, G) are G-equivalent if there exists a
morphism (↵, ) : (a, F ) ! (b, G) where is invertible.
(iii) An unfolding (a, F ) of a map-germ f is said to be G-versal if any
unfolding (b, G) of f can be induced from (a, F ).

An analogous definition can be made for the extended group Ge by

substituting G with Ge in Definition 3.2.
We need the following notion, which can also be defined for maps be-
tween manifolds.

Definition 3.3. A suspension of a map-germ f : (Rn , 0) ! (Rm , 0) by a

germ of a manifold (Rp , 0) is the map-germ (f, id) : (Rn ⇥ Rp , (0, 0)) !
(Rm ⇥ Rp , (0, 0)) given by (f, id)(x, u) = (f (x), u), where id is the identity
map-germ.

Theorem 3.5. (i) For each map-germ f and a Mather group G, the fol-
lowing are equivalent:
(a) f is G-finite;
(b) f has a G-versal unfolding ;
(c) f has a Ge -versal unfolding.
(ii) The least number a0 of parameters for a G-versal (resp. Ge -versal)
unfolding is cod(f, G) (resp. cod(f, Ge )). A versal unfolding (a0 , F ) is called
miniversal.
(iii) Miniversal unfoldings are unique up to equivalence. Any versal
unfolding is equivalent to a suspension of a miniversal unfolding. Versal
unfoldings (a, F ) and (a, G) of f are equivalent.
58 Di↵erential Geometry from a Singularity Theory Viewpoint

Proof. See [Wall (1981)], Theorem 3.4. ⇤

The importance of a G-miniversal unfolding (resp. Ge miniversal unfold-
ing) of a map-germ f (Theorem 3.5(iii)) is that it provides a G (resp. Ge )
model of all possible local deformations of the map-germ f .
Given an unfolding (a, F ) of a map-germ f in Mn .E(n, m), denote by
Ḟi , i = 1, . . . , a, the map-germs in Mn .E(n, m) given by
@ f¯
Ḟi (x) = (x, 0).
@ui
Theorem 3.6. An unfolding (a, F ) of a map-germ f in Mn .E(n, m) is
G-versal if and only if
LG · f + R.{Ḟ1 , . . . , Ḟa } = Mn .E(n, m)
and Ge -versal if and only if
LGe · f + R.{Ḟ1 , . . . , Ḟa } = E(n, m).

Proof. See [Wall (1981)], Theorem 3.3. ⇤

Theorem 3.5 and Theorem 3.6 o↵er a way of finding miniversal unfold-
ings. Let f be a G-finite map-germ and let h1 , . . . , hc in Mn .E(n, m) be an
R-basis of Mn .E(n, m)/LG · f (so c = cod(f, G)). Define the unfolding
c
X
F (x, u) = (f (x) + ui hi (x), u).
i=1

Similarly, let g1 , . . . , gd in E(n, m) be an R-basis of E(n, m)/LGe · f (so

d = cod(f, Ge )) and define the unfolding
d
X
G(x, u) = (f (x) + ui gi (x), u).
i=1
Pc
Theorem 3.7. The unfolding F (x, u) = (f (x) + i=1 ui hi (x), u) of
the map-germ f is G-miniversal and the unfolding G(x, u) = (f (x) +
Pd
i=1 ui gi (x), u) is Ge -miniversal.

Proof. The proof follows by applying Theorem 3.6. ⇤

An unfolding (a, F ) of f is said to be G-trivial (resp. Ge -trivial) if it is
G-equivalent (resp. Ge -equivalent) to the constant unfolding (a, f ).
A map-germ f is G-stable (resp. Ge -stable) if all its unfoldings are trivial.

Theorem 3.8. A map-germ f is G-stable (resp. Ge -stable) if and only if

cod(f, G) = 0 (resp. cod(f, Ge ) = 0).
 Singularities of germs of smooth mappings 59

Proof. See [Mather (1970)]. ⇤

Mather proved the following results which characterises the A-stable
map-germs.

Theorem 3.9. Let f : (Rn , 0) ! (Rm , 0) be a map-germ and F be an

m-parameter unfolding of f given by F (x, u) = (f (x) u, u). Then,
(i) The map-germ f is A-stable if and only if the unfolding (m, F ) is
K-versal.
(ii) If f is A-stable, then f is (m + 1)-A-determined.
(iii) Suppose that f and g are A-stable map-germs. Then f and g are
A-equivalent if and only if they are K-equivalent.

Proof. See [Mather (1969b,c)]. ⇤

It follows from the statements (ii) and (iii) of Theorem 3.9, that the
R-algebra isomorphism class of the local ring
Qm+1 (f ) = En /(f ⇤ (Mm )En + Mm+2
n )
is a complete invariant for the A-stable map-germs (i.e., it determines the
A-stable orbit).

3.9 Classification of singularities

Finding model objects is one of the major activities in mathematics. Given

an equivalence relation in a set, one seeks representatives of the equivalent
classes under this relation. The representatives are chosen in a simple
form. In topology for example, the 2-sphere S 2 is a model for closed and
simply connected surfaces under homeomorphisms. In matrix algebra, we
know that the diagonal matrices are models for real symmetric matrices
X under the equivalence relation AXAT , where A is an invertible matrix.
Here we seek models of map-germs under G-equivalence, where G is one of
Mather’s group. This activity is called classification and consists of listing
representatives of the orbits of the action of the group G on Mn .E(n, m).
The representatives are sometimes referred to as normal forms.
It is natural to start by finding models for non-singular map-germs. The
following result is a consequence of the implicit functions theorem.

Theorem 3.10. Suppose that f : (Rn , 0) ! (Rm , 0) is non-singular.

(1) If n m, then f is R-equivalent to the germ of the projection
⇡(x1 , . . . , xn ) = (x1 , . . . , xm ).
60 Di↵erential Geometry from a Singularity Theory Viewpoint

(2) If n < m, then f is A-equivalent to the germ of the immersion

i(x1 , . . . , xn ) = (x1 , . . . , xn , 0, . . . , 0).

The listing of representatives of all the G-orbits is an impossible task,

so some restrictions are needed. These are imposed, in general, by the
problems under investigation. For problems arising in di↵erential geometry,
we seek G-finite germs of lower G or Ge -codimension. In other context,
G-finite simple germs are sought. The notion of simple germs is defined
in [Arnol’d, Guseı̆n-Zade and Varchenko (1985)] as follows. Let X be a
manifold and G a Lie group acting on X. The modality of a point x 2 X
under the action of G on X is the least number m such that a sufficiently
small neighbourhood of x may be covered by a finite number of m-parameter
families of orbits. The point x is said to be simple if its modality is 0, that
is, a sufficiently small neighbourhood intersects only a finite number of
orbits. The modality of a finitely determined map-germ is the modality of
a sufficient jet in the jet-space under the action of the jet-group.
Each pair of dimensions (n, m) needs to be considered separately when
carrying out the classification task. There are lists for most cases with
n + m  6. N.P. Kirk developed a Maple 5 computer programme ([Kirk
(2000)]) called “Transversal” to carry out classifications of G-singularities
of map-germs. (The computer programme also deals with Damon’s sub-
groups of G, see section 3.10.) Tables that took months to produce by
hand could be obtained in few days! The programme in [Kirk (2000)] is
based on the complete transversal results established in ([Bruce, Kirk and
du Plessis (1997)]). The classification is carried out inductively on the jet
level and returns sufficient jets and other information such as their G or
Ge -codimensions and G-versal unfoldings.

3.9.1 Germs of functions

We deal here in some details with the case of germs of functions (i.e.,
m = 1). The Hessian matrix of a germ of a function f : (Rn , 0) ! (R, 0) is
given by
✓ 2 ◆
@ f
H(f )(0) = (0) .
@xi @xj
If f is singular at the origin, we say that this singularity is non-
degenerate if rank H(f )(0) = n, equivalently, if and only if det H(f )(0) 6= 0.

Theorem 3.11 (The Morse Lemma). Suppose that f : (Rn , 0) !

 Singularities of germs of smooth mappings 61

(R, 0) has a non-degenerate singularity. Then f is R-equivalent to the

non-degenerate quadratic form

Q(x1 , . . . , xn ) = ±x21 ± · · · ± x2n .

Proof. See for example [Martinet (1982)], pp 18-24. ⇤

We consider next the case where the singularity of f is degenerate and

define the corank of f by

corank(f )(0) = n rankH(f )(0),

In particular, the singularity of f is non-degenerate if and only if

corank(f )(0) = 0.

Lemma 3.1 (Thom’s splitting lemma). Suppose that f : (Rn , 0) !

(R, 0) has a singularity of corank(f )(0) = r at the origin. Then f is R-
equivalent to a germ of the form

g(x1 , . . . , xr ) + Q(xr+1 , . . . , xn ),

where g 2 M3r and Q(xr+1 , . . . , xn ) = ±x2r+1 ± · · · ± x2n .

Proof. See for example [Bröcker (1975)], p 125. ⇤

Given two germs g1 and g2 in M3r and Q as in Lemma 3.1, then g1 + Q

and g2 + Q are R-equivalent if and only if g1 and g2 are R-equivalent.
The Splitting Lemma reduces thus the dimension of the source where the
classification is to be carried out.
An extensive list of R-finite germs of functions is given in V.I. Arnold
[Arnol’d, Guseı̆n-Zade and Varchenko (1985)]. Table 3.1 shows the simple
R-finite germs of functions (Rn , 0) ! (R, 0). We observe that cod(g +
Q, R) = cod(g, R) for any R-finite germ g 2 M3r .
The normal forms in Table 3.1 are also the normal forms for the simple
K-finite germs of functions.
We consider now deformations of germs of functions. We call the defor-
mation F : (Rn ⇥ Ra , (0, 0)) ! (R, 0) of the germ f = F |Rn ⇥{0} a family of
germs of functions. In some textbooks, F is referred to as an unfolding of
f.
For germs of functions, it is important, as we shall see in Chapter 5, to
consider also the direct product of the group R with translations, which we
denote by R+ . (This is denoted by Rauge in [Wall (1981)].)
62 Di↵erential Geometry from a Singularity Theory Viewpoint

Table 3.1 Simple germs of functions ([Arnol’d,

Guseı̆n-Zade and Varchenko (1985)]).
Name Normal form cod(f, R)
Ak , k 0 ±xk+11 + Q(x2 , . . . , xn ) k
Dk , k 4 x21 x2 ± xk2 1 + Q(x3 , . . . , xn ) k
E6 x31 + x42 + Q(x3 , . . . , xn ) 6
E7 x31 + x1 x42 + Q(x3 , . . . , xn ) 7
E8 x31 + x52 + Q(x3 , . . . , xn ) 8

Q(xr , . . . , xn ) = ±x2r ± · · · ± x2n

Definition 3.4. Two families of germs of functions F and G : (Rn ⇥

Ra , (0, 0)) ! (R, 0) are P -R+ -equivalent if there exist a germ of a dif-
feomorphism : (Rn ⇥ Ra , (0, 0)) ! (Rn ⇥ Ra , (0, 0)) of the form (x, u) =
(↵(x, u), (u)) and a germ of a function c : (Ra , 0) ! R such that
G(x, u) = F ( (x, u)) + c(u).

In Definition 3.4, the letter “P ” stands for parametrised (as we have a

family of germs of di↵eomorphisms ↵( , u) of Rn parametrised by u) and
“+” stands for the addition of the term c(u).
Given a family of germs of functions F , we write
@F
Ḟi (x) = (x, 0).
@ui
Theorem 3.6 can be adapted as follows for families of functions.

Theorem 3.12. A deformation F : (Rn ⇥ Ra , (0, 0)) ! (R, 0) of a germ of

a function f in Mn is R+ -versal if and only if
n o
LRe · f + R. 1, Ḟ1 , . . . , Ḟa = En .

Definition 3.4 of P -R+ -equivalence can be extended to families F :

(Rn ⇥ Ra , (0, 0)) ! (R, 0) and G : (Rp ⇥ Ra , (0, 0)) ! (R, 0) with n 6= p. We
add a non-degenerate quadratic form Q(yn+1 , . . . , yn+p ) to F and a non-
degenerate quadratic form Q0 (zp+1 , . . . zn+p ) to G and consider the two
families of germs F + Q and G + Q0 from (Rn+p ⇥ Ra , (0, 0)) ! (R, 0).

Definition 3.5. We say that the two families F and G are stably P -R+ -
equivalent if F + Q and G + Q0 are P -R+ -equivalent.

Theorem 3.13. Let f 2 M2n be an R-finitely determined germ with

1  code (f, R)  4. Then any R+ -miniversal unfolding of f is P -R+ -
equivalent to one of the following families of germs, where Q(xr , . . . , xn ) =
±x2r ± · · · ± x2n :
 Singularities of germs of smooth mappings 63

(1) A2 : Q(x2 , . . . , xn ) + x31 + u1 x1 ,

(2) A3 : Q(x2 , . . . , xn ) ± x41 + u2 x21 + u1 x1 ,
(3) A4 : Q(x2 , . . . , xn ) + x51 + u3 x31 + u2 x21 + u1 x1 ,
(4) A5 : Q(x2 , . . . , xn ) ± x61 + u4 x41 + u3 x31 + u2 x21 + u1 x1 ,
(5) D4 : Q(x3 , . . . , xn ) + x31 x1 x22 + u3 (x21 + x22 ) + u2 x2 + u1 x1 ,
(6) D4+ : Q(x3 , . . . , xn ) + x31 + x32 + u3 x1 x2 + u2 x2 + u1 x1 ,
(7) D5 : Q(x3 , . . . , xn ) + x21 x2 + x42 + u4 x22 + u3 x21 + u2 x2 + u1 x1 .

Proof. All germs of functions f of code (f, R)  4 are simple [Arnol’d,

Guseı̆n-Zade and Varchenko (1985)]. The result follows by applying The-
orem 3.12 and Theorem 3.7 to the normal forms in Table 3.1. ⇤
The singularities in Theorem 3.13 are referred to as the Thom’s seven
elementary catastrophes.
We turn now to the group K.

Definition 3.6. Two families of germs functions F and G : (Rn ⇥

Ra , (0, 0)) ! (R, 0) are P -K-equivalent if there exist a germ of a di↵eo-
morphism : (Rn ⇥ Ra , (0, 0)) ! (Rn ⇥ Ra , (0, 0)) of the form (x, u) =
(↵(x, u), (u)) and a germ of a function : (Rn ⇥ Ra , (0, 0)) ! R, with
(0, 0) 6= 0, such that
(x, u) · G(x, u) = F ( (x, u)).

Theorem 3.6 can be written as follows for families of functions.

Theorem 3.14. A deformation F : (Rn ⇥ Ra , (0, 0)) ! (R, 0) of a germ of

a function f in Mn is K-versal if and only if
⇢ n o
@f @f
En . ,..., , f + R. Ḟ1 , . . . , Ḟa = En .
@x1 @xn
Here too Definition 3.6 can also be extended to families F : (Rn ⇥
Ra , 0) ! (R, 0) and G : (Rp ⇥Ra , (0, 0)) ! (R, 0) with n 6= p. We add a non-
degenerate quadratic form Q(yn+1 , . . . , yn+p ) to F and a non-degenerate
quadratic form Q0 (zp+1 , . . . zn+p ) to G and consider the two families of
germs F + Q and G + Q0 from (Rn+p ⇥ Ra , (0, 0)) ! (R, 0).

Definition 3.7. We say that the two families F and G are stably P -K-
equivalent if F + Q and G + Q0 are P -K-equivalent.

Theorem 3.15. Let f 2 M2n be a K-finitely determined germ with 1 

code (f, K)  4. Then any K-miniversal unfolding of f is P -K-equivalent
64 Di↵erential Geometry from a Singularity Theory Viewpoint

to one of the following families of germs, where Q(xr , . . . , xn ) = ±x2r ±

· · · ± x2n :
(1) A1 : Q(x1 , . . . , xn ) + u1 ,
(2) A2 : Q(x2 , . . . , xn ) + x31 + u2 x1 + u1 ,
(3) A3 : Q(x2 , . . . , xn ) + x41 + u3 x21 + u2 x1 + u1 ,
(4) A4 : Q(x2 , . . . , xn ) + x51 + u4 x31 + u3 x21 + u2 x1 + u1 ,
(5) D4 : Q(x3 , . . . , xn ) + x31 x1 x22 + u4 (x21 + x22 ) + u3 x2 + u2 x1 + u1 ,
(6) D4+ : Q(x3 , . . . , xn ) + x31 + x32 + u4 x1 x2 + u3 x2 + u2 x1 + u1 .

Proof. The result follows by applying Theorem 3.14 and Theorem 3.7 to
the normal forms in Table 3.1. ⇤

3.9.2 Discriminants and bifurcation sets

We associate to a family of germs of functions F some germs of sets. For
the families of height functions and distance squared functions on a hyper-
surface, these sets captures geometric information about the hypersurface.
The catastrophe set CF of a family F : (Rn ⇥ Ra , (0, 0)) ! (R, 0) is
defined by
⇢
@F @F
CF = (x, u) 2 (Rn ⇥ Ra , (0, 0)) | (x, u) = · · · (x, u) = 0 .
@x1 @xn
The bifurcation set of F is defined by
⇢ ✓ ◆
a @2F
BF = u 2 (R , 0) 9(x, u) 2 CF and rank (x, u) <n .
@xi @xj
The discriminant of F is defined as
⇢
@F @F
DF = u 2 (Ra , 0) 9x 2 (Rn , 0) and F = = ··· = 0 at (x, u) .
@x1 @xn
Let ⇡CF = ⇡2 |CF : CF ! (Ra , 0), where ⇡2 : (Rn ⇥ Ra , (0, 0)) ! (Ra , 0)
is the projection to the second component. We call ⇡CF the catastrophe
map-germ of F.

Proposition 3.1. Let F and G be two families of germs of functions

(Rn ⇥Ra , 0) ! (R, 0) such that their catastrophe sets CF and CG are smooth
submanifolds. Suppose that F and G are P -R+ -equivalent. Then the catas-
trophe map-germs ⇡CF and ⇡CG are A-equivalent. Moreover, there exists a
germ of a di↵eomorphism : (Rn , 0) ! (Rn , 0) such that (BF ) = BG .
 Singularities of germs of smooth mappings 65

!"#$% &'() *+,--"+%,#- .--#)%#/ '01#-#/ 23)451"-#/ '01#-#/

Fig. 3.4 Bifurcation sets of miniversal unfoldings of germs of Re -codimension ≤ 3.

Proof. By Definition 3.4, there exist a germ of a diffeomorphism Φ :

(Rn ×Ra , 0) → (Rn ×Ra , 0) of the form Φ(x, u) = (α(x, u), ψ(u)) and a germ
of a function c : (Ra , 0) → (R, 0) such that G(x, u) = F (α(x, u), ψ(u)) +
c(u). We have
n
! ∂F
∂G ∂αℓ
(x, u) = (Φ(x, u)) (x, u).
∂xi ∂xℓ ∂xi
ℓ=1
∂G
The matrix (∂αℓ /∂xi )(0, 0) is regular, so ∂x i
(x, u) = 0, for i = 1, . . . , n,
∂F
if and only if ∂xi (Φ(x, u)) = 0, for i = 1, . . . , n. Therefore, Φ(CG ) = CF and
ψ ◦ πCG = πCF ◦ (Φ|CG ). This means that πCG and πCF are A-equivalent.
At (x, u) ∈ CG , we have
n
" n #
2
∂ G ! ! 2
∂ F ∂αh ∂αℓ
(x, u) = (Φ(x, u)) (x, u) (x, u).
∂xi ∂xj ∂xh ∂xℓ ∂xj ∂xi
ℓ=1 h=1

Again, since the matrix (∂αℓ /∂xi )(0, 0) is regular,

$ 2 % $ 2 %
∂ G ∂ F
rank (x, u) = rank (Φ(x, u)) .
∂xi ∂xj ∂xi ∂xj
Therefore, u ∈ BG if and only if ψ(u) ∈ BF , that is, BF = ψ(BG ). !
The catastrophe sets and bifurcation sets of the Thom’s seven elemen-
tary catastrophes in Theorem 3.13 are computed in [Bröcker (1975)]. We
∂F
re-do the calculations here. We have ∂x i
= ±2xi , i = r, . . . , n for all
the models in Theorem 3.13, so we set the redundant variable xi = 0,
i = r, . . . , n. We draw in Figure 3.4 the bifurcation sets of the miniversal
unfoldings in Theorem 3.13 of germs of Re -codimension ≤ 3.
(1) The A2 -singularity
∂F ∂2F
For F (x, u) = x31 + u1 x1 , ∂x1 = 3x21 + u1 and ∂x21
= 6x1 , so

CF = {(x1 , u1 ) | u1 = −3x21 },
BF = {0}.
66 Di↵erential Geometry from a Singularity Theory Viewpoint

The bifurcation set is a point in this case (Figure 3.4).

(2) The A3 -singularity
@F @2F
For F (x, u) = ±x41 + u2 x21 + u1 x1 , @x1 = ±4x31 + 2u2 x1 + u1 and @x21
=
±12x21 + 2u2 , so
CF = {(x1 , u1 , u2 ) | u1 = ⌥4x31 2u2 x1 },
BF = {(u1 , u2 ) | 27u21 + 8u32 = 0}.
It follows that the bifurcation set is a cusp (Figure 3.4).
(3) The A4 -singularity
@F
For F (x, u) = x51 + u3 x31 + u2 x21 + u1 x1 , @x1 = 5x41 + 3u3 x21 + 2u2 x1 + u1
@2F
and @x21
= 20x31 + 6u3 x1 + 2u2 , so
CF = {(x1 , u1 , u2 , u3 ) | u1 = 5x41 3u3 x21 2u2 x1 },
and BF is the germ of the surface parametrised by
(x1 , u3 ) 7! (u1 , u2 , u3 ) = (15x41 + 3u3 x21 , 10x31 3u3 x1 , u3 ),
and called a swallowtail surface (Figure 3.4).
(4) The A5 -singularity
For F (x, u) = ±x61 + u4 x41 + u3 x31 + u2 x21 + u1 x1 , we have
@F
@x1 = ±6x51 + 4u4 x31 + 3u3 x21 + 2u2 x1 + u1 ,
@2F
@x21
= ±30x41 + 12u4 x21 + 6u3 x1 + 2u2 ,
so
CF = {(x1 , u1 , u2 , u3 ) | u1 = ⌥6x51 4u4 x31 3u3 x21 2u2 x1 },
and BF is the germ of the 3-dimensional variety parametrised by (x1 , u3 , u4 )
and given by
(u1 , u2 , u3 , u4 ) = (±24x41 + 8u4 x31 + 3u3 x21 , ⌥15x31 6u4 x21 3u3 x1 , u3 , u4 ).

(5) The D4 -singularity

To simplify calculations, we take a slightly di↵erent normal form here
and consider F (x, u) = 13 x31 x1 x22 + u3 (x21 + x22 ) + u2 x2 + u1 x1 . Then we
have
@F 2
@x1 = x1 x22 + 2u3 x1 + u1 ,
@F
@x2 = 2x1 x2 + 2u3 x2 + u2 ,
and
@2F @2F
! ✓ ◆
@x21 @x1 @x2 2x1 + 2u3 2x2
@2F @2F
= .
@x2 @x1 @x22
2x2 2x1 + 2u3
 Singularities of germs of smooth mappings 67

It follows that
CF = {(x1 , x2 , u1 , u2 , u3 ) | u1 = x21 + x22 2u3 x1 ,
u2 = 2x1 x2 2u3 x2 }
and
BF = {(u1 , u2 , u3 ) | u1 = x21 + x22 2u3 x1 ,
u2 = 2x1 x2 2u3 x2 ,
u23 = x21 + x22 , with (x1 , x2 ) 2 (R2 , 0) }.
The singularity of F (x, 0) is called elliptic umbilic and the bifurcation
set of F is as in Figure 3.4 (called pyramid).
(6) The D4+ -singularity
For F (x, u) = x31 + x23 + u3 x1 x2 + u2 x2 + u1 x1 , we have
@F
@x1 = 3x21 + u3 x2 + u1 ,
@F
@x1 = 3x22 + u3 x1 + u2 ,
and
@2F @2F
! ✓ ◆
@x1 2 @x1 @x2 6x1 u3
@2F @2F
= .
@x2 @x2 @x22
u3 6x2

Then,
CF = {(x1 , x2 , u1 , u2 , u3 ) | u1 = 3x21 u3 x2 ,
u2 = 3x22 u3 x1 }
and
BF = {(u1 , u2 , u3 ) | u1 = 3x21 u3 x2 ,
u2 = 3x22 u3 x1
u23 = 36x1 x2 , with (x1 , x2 ) 2 (R2 , 0) }.
The singularity of F (x, 0) is called hyperbolic umbilic and the bifurcation
set of F is as in Figure 3.4 (called purse).
(7) The D5 -singularity
Here too we take a di↵erent normal form F (x, u) = x21 x2 + 16 x42 + u4 x22 +
u3 x21 + u2 x2 + u1 x1 . We have
@F
@x1 = 2x1 x2 + 2u3 x1 + u1 ,
@F
@x2 = x21 + 23 x32 + 2u4 x2 + u2 ,
and
@2F @2F
! ✓ ◆
@x21 @x1 @x2 2x2 + 2u3 2x1
@2F @2F
= 2 .
@x2 @x1 @x22
2x1 2x2 + 2u4
68 Di↵erential Geometry from a Singularity Theory Viewpoint

We deduce that
CF = {(x1 , x2 , u1 , u2 , u3 , u4 ) | u1 = 2x1 x2 2u3 x1 ,
u2 = x21 23 x32 2u4 x2 }
and
BF = {(u1 , u2 , u3 , u4 ) | u1 = 2x1 x2 2u3 x1 ,
u2 = x21 23 x32 2u4 x2 ,
(x2 + u3 )(x22 + u4 ) x21 = 0, with (x1 , x2 ) 2 (R2 , 0) }.
Theorem 3.16. Let F and G be two P -K-equivalent families of germs of
functions. Then the discriminants DF and DG are di↵eomorphic.

Proof. By Definition 3.6, there exist a germ of a di↵eomorphism :

(R ⇥ R , 0) ! (R ⇥ R , 0) of the form (x, u) = (↵(x, u), (u)) and
n a n a

a germ of a function : (Rn ⇥ Ra , (0, 0)) ! (R, 0), with (0, 0) 6= 0,

such that (x, u) · G(x, u) = F ( (x, u)). Then, G(x, u) = 0 if and only if
F ( (x, u)) = 0.
We have
Xn
@ @G @F @↵j
(x, u) · G(x, u) + (x, u) · (x, u) = ( (x, u)) (x, u).
@xi @xi j=1
@x j @x i

The matrix (@↵` /@xi )(0, 0) is regular, so G(x, u) = 0 and

@G/@xi (x, u) = 0 for i = 1, . . . , n, if and only if F ( (x, u)) = 0 and
@F/@xi ( (x, u)) = 0 for i = 1, . . . , n. This means that DF = (DG ). ⇤
We make use of the calculations for the bifurcation sets of the families
in Theorem 3.13 to compute the discriminants of the families in Theorem
3.15.
(1) A1 -singularity
For F (x, u) = u1 , DF = {0}.
(2) A2 -singularity
For F (x, u) = x31 + u2 x1 + u1 ,
DF = {(u1 , u2 ) | 27u21 + 8u32 = 0},
which is a cusp.
(3) A3 -singularity
For F (x, u) = x41 + u3 x21 + u2 x1 + u1 ,
DF = {(u1 , u2 , u3 ) | u1 = 3x41 + u3 x1 ,
u2 = 4x31 2u3 x21 , with (x1 , u3 ) 2 (R2 , 0) },
 Singularities of germs of smooth mappings 69

which is a swallowtail surface.

(4) A4 -singularity
For F (x, u) = x51 + u4 x31 + u3 x21 + u2 x1 + u1 ,
DF = {(u1 , u2 , u3 , u4 ) | u1 = 4x51 + 2u4 x31 + u3 x21 ,
u2 = 5x41 3u4 x21 2u3 x1 ,
with (x1 , u3 , u4 ) 2 (R3 , 0) }.

(5) D4 -singularity
For F (x, u) = 13 x31 x1 x22 + u4 (x21 + x22 ) + u3 x2 + u2 x1 + u1 ,

DF = {(u1 , u2 , u3 , u4 ) |u1 = 23 x31 2x1 x22 + u4 (x21 + x22 ),

u2 = x21 + x22 2u4 x1 ,
u3 = 2x1 x2 2u4 x2 , with (x1 , x2 , u4 ) 2 (R3 , 0) }.

(6) D4+ -singularity For F (x, u) = x31 + x23 + u4 x1 x2 + u3 x2 + u2 x1 + u1 ,

DF = {(u1 , u2 , u3 , u4 ) |u1 = 2x31 + 2x23 + u4 x1 x2 ,
u2 = 3x21 u4 x2 ,
u3 = 3x22 u4 x1 , with (x1 , x2 , u4 ) 2 (R3 , 0) }.

Remark 3.2. It is not difficult to check that the Boardman symbol of the
catastrophe map-germ associated to an Ak -singularity is S1k 1 , where the
subindex k 1 refers to the number of entries of the subindex 1 in the sym-
bol, so S11 = S1,0 , S12 = S1,1,0 , S13 = S1,1,1,0 and so on. The Boardman
symbol of the catastrophe map-germ associated to a Dk -singularity is S2 .

3.10 Damon’s geometric subgroups

The results on finite determinacy, complete transversal and versal unfold-

ings are stated here for the Mather groups R, L, C, K and A. However, in
various situations one has to deal with subgroups of one of these groups.
For example, consider germs of a function on a variety (X, 0) in Rn . Any
relevant di↵eomorphism in R should preserve the germ of the variety (X, 0).
J. Damon showed that the results we presented in this chapter are valid
for a large class of subgroups of K and A, which he called geometric sub-
groups of K and A ([Damon (1984)]). For instance, the subgroup of R of
di↵eomorphisms preserving a variety (X, 0) in Rn is a geometric subgroup.
Another example of a geometric subgroup is the group of equivariant dif-
feomorphisms with respect to a compact Lie group action.
70 Di↵erential Geometry from a Singularity Theory Viewpoint

3.11 Notes

Our aim in this chapter is to set the singularity theory notation and state
the results we need in this book. It is far short from being a survey of
results in the area of singularities of map-germs. We give below some of
the research directions in this area. The few mentioned references (mainly
books) are meant as appetisers.
We stated the results for C 1 (real) map-germs as these arise in applica-
tions to di↵erential geometry. The study of C r map-germs and of complex
holomorphic map-germs is very rich. The survey article of Wall ([Wall
(1981)]) touches on this.
J. Milnor ([Milnor (1968)]) considered the topology of the fibre of a
holomorphic function. If the germ of a map f at its singular point is R-
finite, then the intersection of the singular fibre with a small sphere has the
homotopy type of the wedge of µ(f )-spheres, where

µ(f ) = cod(f, R)

(µ(f ) is called the Milnor number). Furthermore, given an R-finite holo-

morphic map-germ f : (Cn , 0) ! (C, 0), one has a locally trivial fibre bundle

f /kf k : S"n 1
\ K ! S1

on any sufficiently small sphere S"n 1 , outside some small tubular neigh-
bourhood NK of the link K = f 1 (0) \ S"n 1 . This is known as Milnor
Fibration Theorem. (See [Brieskorn and Knörrer (1986); Wall (2004)] for
the case n = 2, and [Seade (2007)] for the real case.)
A classical example of a family of non-equivalent R-finite germs is
f (x, y) = xy(x y)(x + y), for 6= 0, 1. The parameter is a mod-
ulus and represents the cross-ratio of the four lines f 1 (0). If one relaxes
the equivalence relation and considers germs of homeomorphisms in the
source instead of di↵eomorphisms, then all the members of the family f
become topologically equivalent. We say that the family f is topologi-
cally trivial. The problem of determining whether a family of map-germs
is topologically trivial was and still remains a major problem in singularity
theory ([du Plessis and Wall (1995)]; see also [Damon (1988, 1992)] for
topological triviality for Damon’s geometric subgroups).
A G-invariant of a map-germ f is a number that depends only on the
G-orbit of f . For example, the Milnor number µ is an R-invariant of germs
of functions (it is also a topological invariant). It turns out that the con-
stancy of µ in a family of germs of functions implies that the family is
 Singularities of germs of smooth mappings 71

topologically trivial (in fact it implies the equisingularity of the family)

when n 6= 3 ([Lê and Ramanujam (1976)]). One approach to tackle the
problem of topological triviality or equisingularity is to determine a collec-
tion of invariants whose constancy in the family implies the triviality of the
family (see [Ga↵ney and Massey (1999)]).
Metric properties of singular sets (semi-algebraic, sub-analytic etc) is
also well developed. Here, bi-Lipschitz maps play a key role, and one has the
notion of bi-Lipschitz equivalence and bi-Lipschitz equisingularity ([Birbrair
(2007); Mostowski (1985)]).
There is also a great interest in map-germs with non-isolated singulari-
ties, see for example [Massey (1995)].
H. Whitney showed that maps from the plane to the plane have only
singularities of type fold, cusps or double folds ([Whitney (1955)]). This is
the first result on the singularities of maps (rather than map-germs). See
[Saeki (2004)] for more recent results.
There is also a branch of research of global topological invariants which
measure how the global topology of spaces forces the singularities of map-
pings. Each singularity type determines a universal characteristic class,
which is a polynomial in the generators of the cohomology ring of a clas-
sifying space. The resulting polynomial is the Thom polynomial of the
singularity. See for example, [Kazarian (2006, 2001)].
There are applications of singularity theory to bifurcations of station-
ary solutions of ordinary di↵erential equations in [Golubitsky and Schae↵er
(1985)] and to solutions of quasi-linear first order partial di↵erential equa-
tions in [Izumiya and Kossioris (1997b)].
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws

This page intentionally left blank

 Chapter 4

Contact between submanifolds of Rn

Following Felix Klein’s approach, a geometry of submanifolds in an ambient

space X depends on the description of their properties which are invariant
under the action of a transformation group of X ([Klein (1974)]). Each
geometry has its class of model submanifolds which is invariant under the
action of the transformation group of X. These submanifolds are homo-
geneous spaces (they look locally the same at all points). We can then
associate geometrical invariants to a submanifold M by comparing it with
the models in such a way that an invariant at a given point p in M is de-
fined as that of the model that better approximates M at p. For instance,
the curvature of a plane curve at a point p is the curvature of the osculat-
ing circle at p. The osculating circle has the highest possible contact with
the curve at p among all circles passing through the point p. The model
submanifolds do not need to be unique at each point. This is illustrated by
the focal hyperspheres at a point p on a hypersurface in Rn , n 3. There
are n focal hyperspheres at p and this leads to the hypersurface having n-
principal curvatures, where each principal curvature is the curvature of an
osculating hypersphere. The osculating hyperspheres at a point p are the
hyperspheres with highest contact with a hypersurface at p. We investigate
in this chapter the concept of contact between submanifolds as a singularity
theory tool for the study of di↵erential geometry of submanifold of Rn (or
of any manifold).
The relation between the contact of equidimensional submanifolds of a
given manifold and K-singularities of maps was introduced by Mather in
[Mather (1969b)] in his investigation of properties of the contact group K.
The general theory of contact between submanifolds of any dimensions of
a given manifold was developed by Montaldi in his PhD thesis [Montaldi
(1983)]. The results that we need in this book can also be found in [Montaldi

73
74 Di↵erential Geometry from a Singularity Theory Viewpoint

(1986a)] (see also [Montaldi (1991)]) and some of them are reproduced in
section 4.1.
The maps defining the contact between a submanifold of Rn and model
submanifolds come naturally in a family of map-germs. One can ask if such
a family is versal and what possible singularities one can expect in the mem-
bers of the family. This leads to the notion of transversality and genericity
which are dealt with in section 4.2. The remaining sections give applica-
tions of Montaldi’s genericity theorems to the study of the families of height
functions, distance squared functions and projections of hypersurfaces to
Rn .

4.1 Contact between submanifolds

We start with the example of plane curves. Let ↵(t) = (x(t), y(t)) be a
regular plane curve and let be another plane curve given as the zero set
of a smooth function F : R2 ! R. We say that the curve ↵ has (k + 1)-
point contact at t0 with the curve if t0 is a zero of order k of the function
g(t) = F (↵(t)) = F (x(t), y(t)), that is,

g(t0 ) = g 0 (t0 ) = . . . = g (k) (t0 ) = 0 and g (k+1) (t0 ) 6= 0,

where g (i) denotes the ith -derivative of the function g. Using the singularity
theory terminology in Chapter 3, the curve ↵ has (k + 1)-point contact at
t0 with if and only if the function g has an Ak -singularity at t0 .
A particular case of interest is when the curve is a circle or a line.
If for instance is a circle, then the singularities of g reveal geometric
information about the curve ↵ (see Proposition 1.3 in Chapter 1).
We extend in this chapter the notion of contact between plane curves
and define the contact between two submanifolds of an n-dimensional man-
ifold.

We start with the following definition.

Definition 4.1 ([Montaldi (1983, 1986a)]). Let Mi , Ni , i = 1, 2, be

submanifolds of Rn with dim(M1 ) = dim(M2 ) = m and dim(N1 ) =
dim(N2 ) = d. We say that the contact of M1 and N1 at y1 is of the same
type as the contact of M2 and N2 at y2 if there is a germ of a di↵eomor-
phism : (Rn , y1 ) ! (Rn , y2 ) such that (M1 ) = M2 and (N1 ) = N2 . In
this case we write K(M1 , N1 ; y1 ) = K(M2 , N2 ; y2 ).
 Contact between submanifolds of Rn 75

The definition of contact between submanifolds in Definition 4.1 is local

in nature, so the ambient space Rn can be replaced by any manifold Z.
Following the setting for plane curves, suppose that a submanifold of
R is given locally as the image of an immersion-germ g : (M, x) ! (Rn , 0)
n

of some manifold M and that another submanifold N is given locally as the

zero set of a submersion-germ f : (Rn , 0) ! (Rk , 0), that is, N = f 1 (0).
We say that f cuts out N. We consider, as in the case of plane curves, the
germ of the composite map f g at x with g(x) = y 2 f 1 (0) and analyse
its K-singularities, where K is the contact group (see Chapter 3). We call
the composite map-germ f g the contact map-germ.
One can ask if di↵erent choices of immersions and/or submersions of
the submanifolds lead to K-equivalent contact map-germs. Also, one can
choose to immerse N and cut out M. Following [Montaldi (1986a)], we get
a positive answer to these questions from the following lemma (Symmetry
Lemma, in [Montaldi (1986a)]).

Lemma 4.1 (Symmetry Lemma in [Montaldi (1986a)]). Let M

and N be submanifold-germs of Rn at 0. Let
g : (M, x0 ) ! (Rn , 0) and ḡ : (N, 0) ! (Rn , 0)
be immersion-germs, and let
f : (Rn , 0) ! (Rk1 , 0) and f¯ : (Rn , 0) ! (Rk2 , 0)
be submersion-germs such that M = f¯ 1 (0) and N = f 1 (0). If k2 k1 ,
then f¯ ḡ is K-equivalent to a suspension of f g by Rk2 k1 .

Proof. Consider the following commutative diagram

(M, x0 )
=f g
- (Rk1 , 0)
H *
HH
g H ⇡1
H
j
H =(f,f¯)
(Rn , 0) - (Rk1 , 0) ⇥ (Rk2 , 0)
* HH
ḡ ⇡2HH
¯=f¯ ḡ
H
j
(N, 0) - (Rk2 , 0)

where ⇡1 and ⇡2 are the canonical projections to the first and second com-
ponents respectively.
76 Di↵erential Geometry from a Singularity Theory Viewpoint

With convenient choices of coordinates in M and (Rn , 0) the immersion

g : (M, x0 ) ! (Rn , 0) can be written in the form g(x) = (x, 0). Since
f¯ g = 0, we also get f¯(x, z) = z. Then, (x, z) = (f (x, z), z) and (x, 0) =
(f g(x), 0). It follows that (x, z) can be considered as an unfolding of
= f g. Since an unfolding of any map is K-equivalent to a suspension
of that map, is K-equivalent to a suspension of . Similarly, can also
be considered as an unfolding of ¯ = f¯ ḡ : (M, x0 ) ! (Rk2 , 0), and we get
is K-equivalent to a suspension of ¯. Since k2 k1 , it follows that ¯ is
K-equivalent to a suspension of as required. ⇤
Lemma 4.2 follows from Lemma 4.1.

Lemma 4.2 ([Montaldi (1986a)], p. 196). For any pair of germs of

submanifold in Rn , the K-class of the contact map depends only on the
submanifold-germs themselves and not on the contact map.

Montaldi proved the following result which shows that the K-equivalence
class of the contact map-germ f g determines the contact class of the pair
of submanifolds M and N.

Theorem 4.1 ([Montaldi (1986a)], p. 195). Let gi : (Mi , xi ) !

(Rn , 0) be immersion-germs and fi : (Rn , 0) ! (Rk , 0) submersion-germs,
with Ni = fi 1 (0), i = 1, 2. Then the pairs (M1 , N1 ) and (M2 , N2 ) have the
same contact type if and only if f1 g1 and f2 g2 are K-equivalent.

The proof of Theorem 4.1 is done in two steps. The first step deals with
the equidimensional case, i.e., when dim Mi = dim Ni = m, and the second
deals with the general case.
Proof of Theorem 4.1 for the equidimensional case. Suppose
that the pairs (M1 , N1 ) and (M2 , N2 ) have the same contact type. Let
H be the di↵eomorphism of (Rn , 0) taking g1 (M1 ) to g2 (M2 ) and N1 to
N2 . As H|g1 (M1 ) : g1 (M1 ) ! g2 (M2 ) is a di↵eomorphism, there exists
a di↵eomorphism h : M1 ! M2 , such that H g1 = g2 h. We also
have (f2 H) 1 (0) = f1 1 (0). Then we can write each coordinate function
(f2 H)j of f2 H as

(f2 H)j (y) = ⌃ki=1 f1i (y)aij (y), (4.1)

where f1 = (f11 , . . . , f1k ) and aij : Rk ! R, i = 1, . . . , k, are germs of

smooth functions. To show this, we use the hypothesis that f1 : (Rn , 0) !
(Rk , 0) is a submersion-germ, and choose a coordinate system in (Rn , 0)
 Contact between submanifolds of Rn 77

such that f 1 (0) is the linear subspace Rn k ⇥ {0}. Then, equation (4.1)
follows from Hadamard’s Lemma. Now, the map f2 H is also a submersion,
and as f1 (0) = 0, it follows that the k ⇥ k matrix [a1 (y), . . . , ak (y)] is
invertible in a neighbourhood of 0, where the vectors ai (y) have coordinates
(aij (y)).
We now define ✓ : Rn ⇥ Rk ! Rk by

✓(y, z) = ⌃ki=1 zi ai (y),

and ✓0 : M1 ⇥ Rk ! Rk by

✓0 (x, z) = ✓(g1 (x), z).

Then, ✓0 (x, f1 g1 (x)) = f2 g2 h(x). It follows that f1 g1 ⇠K f2 g2

with (h, ✓0 ) the di↵eomorphism in K taking f1 g1 to f2 g2 .

Suppose now that f1 g1 and f2 g2 are K-equivalent and assume that

dim Mi = dim Ni = m. We shall write each gi (Mi ) as the graph of some
map i : Rm ! Rk , k = n m, and the Ni as fi 1 (0), where fi : Rn ! Rk is
the projection fi (x1 , . . . , xm , xm+1 , . . . , xn ) = (xm+1 , . . . , xn ), i = 1, 2. To
do this we first choose a coordinate system in Rn so that N1 = Rm ⇥{0}. We
then choose a k-dimensional vector subspace V1 ⇢ Rn transverse to both
g1 (M1 ) and N1 and write Rn = N1 ⇥ V1 . Let ⇡ : Rn ! N1 be the canonical
projection to the first factor. Then ⇡|M1 : g1 (M1 ) ! N1 is a di↵eomorphism
which induces a coordinate system on g1 (M1 ), hence on M1 . With respect
to these coordinate system N1 = Rm ⇥ {0}, while M1 is the graph of the
map f1 g1 (by considering f1 as the projection: N1 ⇥ V1 ! V1 , which
in the chosen coordinate systems are respectively Rm ⇥ Rk and Rk ). A
similar construction can be done for M2 and N2 . Then any di↵eomorphism
H : Rm ⇥ Rk ! Rm ⇥ Rk preserving Rm ⇥ {0} and taking the graph of
f1 g1 to the graph of f2 g2 is a di↵eomorphism taking M1 to M2 and N1
to N2 , and this concludes the proof for the equidimensional case. ⇤

The proof of the general case of Theorem 4.1 requires the following two
lemmas which relate the contact between two submanifolds in Rn with that
of their suspensions.

Lemma 4.3 (Lemma A in [Montaldi (1986a)]). For any positive in-

teger a, let Mi0 = Mi ⇥ Ra , Ni0 = Ni ⇥ {0}, with yi0 = (yi , 0) all in
Rn ⇥ Ra for i = 1, 2. Then K(M1 , N1 ; y1 ) = K(M2 , N2 ; y2 ) if and only
if K(M10 , N10 ; y10 ) = K(M20 , N20 ; y20 ).
78 Di↵erential Geometry from a Singularity Theory Viewpoint

Proof. The sufficient part of the statement follows from the fact that the
suspension of the di↵eomorphism taking M1 to M2 and N1 to N2 takes M10
to M20 and N10 to N20 .
For the converse, let H 0 : (Rn ⇥ Ra , y10 ) ! (Rn ⇥ Ra , y20 ) be the
di↵eomorphism-germ such that H 0 (M10 , y10 ) = (M20 , y20 ), H 0 (N10 , y10 ) =
(N20 , y20 ). We write H 0 = (H1 , H2 ), where H1 : (Rn+a , y10 ) ! (Rn , y1 ) and
H2 : (Rn+a , y10 ) ! (Ra , 0). As H 0 is a di↵eomorphism it follows that H1
and H2 are submersions.
Suppose that there exists a map-germ ⌘ : (Rn , y1 ) ! (Ra , 0) such that

(a) ⌘|N1 = 0 and

(b) the map H : (Rn , y1 ) ! (Rn , y2 ), given by H(x) = H1 (x, ⌘(x)) is a
di↵eomorphism-germ.

Then the map H will be the required map. Indeed, for all x 2 (M1 , y1 )
we get (x, ⌘(x)) 2 M10 , then H 0 (x, ⌘(x)) 2 M20 and hence H(x) =
H1 (x, ⌘(x)) 2 M2 . Similarly, y 2 N1 gives (y, ⌘(y)) = (y, 0) 2 N10 , hence
H 0 (y, 0) 2 N20 and then H(y) 2 N2 .
We show now that the map ⌘ does exist. For (b) it is enough to show that
the derivative map dH is injective. We write U = d⌘ and dH1 = (A, B),
where A : Rn ! Rn and B : Ra ! Rn are linear mappings. Since (A, B)
has rank n, we require U so that A + BU has rank n.
For condition (a) we require U to be zero on T N1 . Since A restricted to
T N1 is injective, we can show that such U exists. ⇤

Lemma 4.4 (Lemma B in [Montaldi (1986a)]). Let Mi , Ni , fi and gi

as above, with i = 1, 2. Let a and b be non-negative integers. Consider the
submanifold-germs of Rn ⇥ Ra ⇥ Rb at the origin Mi0 = Mi ⇥ Ra ⇥ {0} and
Ni0 = Ni ⇥ {0} ⇥ Rb .
Let gi0 be immersion-germs with image Mi0 and fi0 submersion-germs
with zero-set Ni0 . Then f10 g10 and f20 g20 are K-equivalent if and only if
f1 g1 and f2 g2 are K-equivalent.

Proof. By Lemma 4.2 we can choose

gi0 (x, u) !
7 (gi (x), u, 0),
fi0 (y, u, v) ! 7 (fi (y), u),

without changing the K class of the contact maps. Then fi0 gi0 (x, u) =
(fi gi (x), u), so fi0 gi0 is a suspension of fi gi , and the result follows by
Theorem 3.3 in Chapter 3. ⇤
 Contact between submanifolds of Rn 79

Proof of Theorem 4.1 of the general case. The proof necessary part
is the same as that of the equidimensional case. For the sufficient part, if
dim(Mi ) 6= dim(Ni ), we can suspend the submanifold of lower dimension,
say Mi , with Ra , where a = dim(Ni ) dim(Mi ). This gives Mi0 = Mi ⇥ Ra
and Ni0 = Ni ⇥ {0} in Rn ⇥ Ra and the map-germs gi0 (x, a) = (gi (x), a) and
fi0 (y, a) = (fi (y), a).
We then have the following implications, which give the result:
(b)
f10 g10 ⇠K f20 g20 =) K(M10 , N10 ) = K(M20 , N20 )
*(a) +(c)
f1 g 1 ⇠ K f2 g 2 K(M1 , N1 ) = K(M2 , N2 )

where (a) follows by Lemma 4.4, (b) by the proof of the result in the
equidimensional case and (c) by Lemma 4.3. ⇤

Example 4.1. Let C be a curve in Rn given as the image of an immersion

g : R ! Rn and let N be a submanifold given by the zero set of a submersion
f : Rn ! Rk . The resulting contact map is the map h = f g : R ! Rk .
Any K-finitely determined map-germ (R, 0) ! Rk is K-equivalent to a map-
germ h(t) = (ts+1 , 0, . . . , 0), for some s. The local ring Q(h) is isomorphic
to R{1, t, . . . , ts }, and we say that h has an As -singularity. As in the case
of plane curves, we say that the curve C has an (s + 1)-point contact with
N. This coincides with the notion of (s + 1)-point contact in the case of
plane curves.

We consider now the contact of a submanifold in Rn with foliations.

This is motivated by problems involving contact of the submanifold and a
family of model submanifolds. In many situations, it is necessary to replace
the group K by a group that gives information not only about the contact
of a submanifold with the zero fibre of a submersion, but also on its contact
with nearby fibres.
Here we consider the relationship between the contact of a submanifold
with foliations and the R+ -class of functions. Let Mi (i = 1, 2) be mani-
folds with dim M1 = dim M2 , and let gi : (Mi , xi ) ! (Rn , yi ) be germs of
immersions. Let fi : (Rn , yi ) ! (R, 0) be germs of submersions.
For a germ of a submersion f : (Rn , 0) ! (R, 0), we denote by Ff
the regular foliation Ff = {f 1 (c) | c 2 (R, 0)}. We say that the contact
of g1 (M1 ) with the foliation Ff1 at y1 is of the same type as the contact
of g2 (M2 ) with the foliation Ff2 at y2 if there exists a germ of a di↵eo-
morphism : (Rn , y1 ) ! (Rn , y2 ) such that (g1 (M1 )) = g2 (M2 ) and
80 Di↵erential Geometry from a Singularity Theory Viewpoint

(Y1 (c)) = Y2 (c), where Yi (c) = fi 1 (c) for all c 2 (R, 0). In this case we
write K(g1 (M1 ), Ff1 ; y1 ) = K(g2 (M2 ), Ff2 ; y2 ).
Here too we can replace the ambient space Rn by any manifold. The
following result characterises the contact of a hypersurface with a foliation
in terms of the R+ -singularities of functions.

Proposition 4.1 ([Goryunov (1990)], Appendix). Let Mi , i = 1, 2,

be two manifolds with dim(M1 ) = dim(M2 ) = n 1, and gi : (Mi , xi ) !
(Rn , yi ) be germs of immersion. Let fi : (Rn , yi ) ! (R, 0) be germs of
submersions. Then K(g1 (M1 ), Ff1 ; y1 ) = K(g2 (M2 ), Ff2 ; y2 ) if and only if
f1 g1 and f2 g2 are R+ -equivalent.

Proof. See [Goryunov (1990)]. ⇤

4.2 Genericity

The space C 1 (X, Y ) of smooth maps between two manifolds X and Y

is endowed with the so-called Whitney C 1 -topology which is defined as
follows (more details can be found in [Golubitsky and Guillemin (1973)]).
For p 2 X and q 2 Y , and for a non-negative integer k, denote by
k
J (X, Y )p,q the set of k-jets of map-germs (X, p) ! (Y, q). The k-jet space
of mappings from X to Y is defined as J k (X, Y ) = [p2X,q2Y J k (X, Y )p,q .
The set J k (X, Y ) is a smooth manifold (Theorem 2.7 in [Golubitsky and
Guillemin (1973)]). The topology of J k (X, Y ) is used to define a topology
on C 1 (X, Y ) as follows. Let U be an open set in J k (X, Y ) and denote by
M (U ) = {f 2 C 1 (X, Y ) | j k f (X) ⇢ U }.
The family of sets {M (U )} where U is an open set of J k (X, Y ) forms a
basis for a topology on C 1 (X, Y ) (note that M (U ) \ M (V ) = M (U \ V )).
This topology is called the Whitney C k -topology. Denote by Wk the set
of open subsets of C 1 (X, Y ) in the Whitney C k -topology. The Whitney
S1
C 1 -topology on C 1 (X, Y ) is the topology whose basis is W = k=0 Wk .
Let Imm(X, Y ) denote the subset of C 1 (X, Y ) whose elements are
proper C 1 -immersions from X to Y , and Emb(X, Y ) the space of proper
C 1 -embeddings of X into Y. The sets Imm(X, Y ) and Emb(X, Y ) are given
the induced Whitney C 1 -topology.
With this topology, the set Imm(X, Y ) is an open subset of C 1 (X, Y )
and when dim(Y ) 2 dim(X), Imm(X, Y ) is also dense in C 1 (X, Y )
(Whitney Immersion Theorem [Golubitsky and Guillemin (1973)]). When
 Contact between submanifolds of Rn 81

dim(Y ) 2 dim(X) + 1, the set Emb(X, Y ) is dense in C 1 (X, Y ), and

when X is compact, it is also open. See [Golubitsky and Guillemin (1973)]
for proofs.
A property P in the topological space Imm(X, Y ) (resp. Emb(X, Y ))
is said to be generic if it is satisfied by a residual subset of Imm(X, Y )
(resp. Emb(X, Y )). A residual subset of a topological space is a countable
intersection of open dense subsets. Since Imm(X, Y ) is a Baire space, a
residual subset of Imm(X, Y ) is dense set. Similarly, a residual subset of
Emb(X, Y ) is also dense ([Golubitsky and Guillemin (1973)] p. 44).
Given a generic property P , we call immersions (resp. embeddings) that
satisfy P generic immersions (resp. generic embeddings).

We shall apply the results in section 4.1 in the following way. We con-
sider the contact of a manifold M immersed in Rn with families of subman-
ifolds (more specifically, k-spheres and k-planes) and show that this contact
is generic for a dense subset of immersions of M in Rn . We also list the
generic contacts.
First we recall the notion of transversality and Thom’s transversality
theorem ([Golubitsky and Guillemin (1973)]).

Definition 4.2. Let f : X ! Y be a C 1 -map and let Z ⇢ Y be a

submanifold. The map f is transverse to Z at x 2 X if one of the following
conditions holds:

(i) f (x) is not in Z.

(ii) f (x) 2 Z and dfx (Tx X) + Tf (x) Z = Tf (x) Y.

If f is transverse to Z for every x 2 X, we say that f is transverse to Z

and write f t Z.

Theorem 4.2. Let g : X ! Y be a smooth map between smooth manifolds.

Let Z ⇢ J r (X, Y ) be a submanifold. Then, the set

TZ = {g 2 C 1 (X, Y ) | j r g t Z}

is a dense subset of C 1 (X, Y ). Moreover, if Z is a closed subset of

J r (X, Y ), then TZ is open in C 1 (X, Y ).

The key to the results in this section is the next proposition which is a
variant of Thom’s transversality theorem ([Thom (1956)], [Golubitsky and
Guillemin (1973)]).
82 Di↵erential Geometry from a Singularity Theory Viewpoint

Proposition 4.2. Let B, X and Y be any smooth manifolds, and let Z be

a submanifold of J r (X, Y ) ⇥ B. Then the set

TZ = {g 2 C 1 (X, Y ) | (j r g, id) t Z}

is residual in C 1 (X, Y ), where (j r g, id) : X ⇥ B ! J r (X, Y ) ⇥ B is the

product map and id : B ! B is the identity map. Moreover, if Z is closed
and B is compact, then TZ is open and dense.

Let (Nb )b2B denote our chosen model submanifolds of Rn , where B is a

smooth manifold and let g 2 Imm(M, Rn ).
To study the contact between g(M ) and Nb , b 2 B, for each b 2 B, we
suppose that there is a submersion fb : Rn ! Rk , with Nb = fb 1 (0). We
also suppose that the family of maps F : Rn ⇥ B ! Rk , given by

F (y, b) = fb (y) (4.2)

is smooth. By Theorem 4.1, the contact between g(M ) and a model sub-
manifold Nb at the point g(x) is given by the K-singularities of the com-
posite map fb g at x.
We denote by g,b : M ! Rk the map g,b (x) = fb g(x). We also
denote by g : M ⇥ B ! Rk the family of maps given by

g (x, b) = g,b (x) = fb g(x).

Denote by Jyr (M, Rk ) the subset of the jet space J r (M, Rk ) of jets with
target y. In our application, 0 2 Rk is a preferred target as it is the target
of fb g. We consider all maps with non-zero target as being K-equivalent,
indeed if two map-germs have non-zero target their local algebras are iso-
morphic and equal to E(n). Moreover if one map has target 0, and another
does not, then they are not K-equivalent. Thus, any K-invariant subman-
ifold of J r (M, Rk ) is either all of the complement of J0r (M, Rk ) or is a
submanifold of J0r (M, Rk ).

Theorem 4.3 (Theorem 2.2 in [Montaldi (1983)]). Let M, B, fb , g

and g be as above. Let W be either a K-invariant submanifold of
J0r (M, Rk ) or all of its complement. Then the set

RW = {g 2 Imm(M, Rn ) : j1r g t W}

is residual in Imm (M, Rn ), where j1r is the r-jet with respect to the first
variable, so j1r g maps M ⇥ B to J r (M, Rk ). Moreover, if B is compact and
W is closed then RW is open and dense.
 Contact between submanifolds of Rn 83

Proof. Define the map : M ⇥ C 1 (M, Rn ) ⇥ B ! M ⇥ C 1 (M, Rk ) by

(x, g, b) = (x, fb g)
and let r
: J r (M, Rn ) ⇥ B ! J r (M, Rk ) be the induced map in jet space
given by
r
(j r g(x), b) = j r (fb g)(x) = j r g,b (x).

The main idea of the proof is as follows. If r t W, then Z =

( r ) 1 (W ) ⇢ J r (M, Rn ) ⇥ B is a smooth submanifold ([Golubitsky and
Guillemin (1973)]) and the theorem follows from Proposition 4.2 as RW =
TZ \ Imm(M, Rn ). Note that since Imm (M, Rn ) is open in C 1 (M, Rn ), the
set TZ \ Imm(M, Rn ) is a residual set in Imm(M, Rn ).
We now prove that the condition that fb : Rn ! Rk is a submersion
imply that r t W. If W is the complement of J0r (M, Rk ), then it is open
and the transversality condition r t W holds trivially. One therefore
needs to show that r t W whenever W is a submanifold of J0r (M, Rk ).
For this, it is enough to show that for any b 2 B, rb is a submersion at x
if rb (x) 2 J0r (M, Rk ).
This is a local condition, so we can suppose x to be the origin and the
tangent space to the fibre over x in J r (M, Rk ) at j r g(x) can be identified
with ✓g /Mr+1 .✓g , where ✓g is the set of germs at x of vector fields along g.
(Observe that ✓g is isomorphic to E(m, n), hence ✓g /Mr+1 m .✓g is isomorphic
r+1
to E(m, n)/Mm .E(m, n)).
Now for each b 2 B, the map b = (. , . , b) induces a map
b⇤ : ✓g ! ✓
⇠ 7 ! dfb (⇠),
where = g,b and the derivative map is taken at g(x) = g(0) = 0. The
map
r
b⇤ : ✓g /Mr+1 .✓g ! ✓ /Mr+1 .✓
is the derivative of rb restricted to the fibre over x.
Then b⇤ is surjective if fb is a submersion at g(x). Indeed, suppose
that fb is a submersion at y = g(x), and choose coordinates in Rn and Rk
so that fb takes the form
fb (y1 , . . . , yn ) = (y1 , . . . , yk ).
Let ⇠ = (h1 (x), . . . , hn (x)), then b⇤ (⇠) = (h1 (x), . . . , hk (x)), so b⇤ is
clearly surjective. It follows that b is a submersion, and consequently so
is rb as required.
84 Di↵erential Geometry from a Singularity Theory Viewpoint

To complete the proof, we consider the following diagram, where

I (M, Rn ) is the set of r-jets of immersions g : M ! Rn :
r

j r g⇥id
- I r (M ⇥ Rn ) ⇥ B - J r (M, Rk )
r
M ⇥B

? jr g
?
M - I r (M, Rn )

We now suppose that W is closed and B is compact. Then the restriction

of the projection ⇡ : I r (M, Rn ) ⇥ B ! I r (M, Rn ) to the closed set W 0 :=
r 1
(W ) is a proper map. Hence RW is open and dense. ⇤
The results in Theorem 4.3 are extended in [Montaldi (1986a)] to cover
the Mather’s subgroups of K, G = R, L, A, C, K. (The case G = R+ is dealt
with in [Looijenga (1974)].)
We need the following definition before stating Montaldi’s Theorem for
the Mather’s groups.

Definition 4.3. Let G be one of Mather’s subgroups of K. A family of

maps F : Rn ⇥ B ! Rk , given by F (y, b) = fb (y), is said to be locally
G-versal if for every (y, b) 2 Rn ⇥ B, the germ of F at (y, b) is a G-versal
unfolding of fb at y. The family F is said to be G-versal if for every b 2 B
and every finite subset S ⇢ Rn , the multi-germ of F at S ⇥ {b} is a G-versal
unfolding of the multigerm of fb at S.

Let g : M ! Rn be an immersion and denote, as before, by g : M ⇥B !

Rk the map given by
g (x, b) = F (g(x), b),
with F as in Definition 4.3.

Theorem 4.4 ([Montaldi (1986a)]). Let G = R, R+ , L, A, C, K and let

M , F and be as above.
(i) Suppose that F is locally G-versal and let W ⇢ J r (M, Rk ) be a G-
invariant submanifold. Then the set
RW = {g 2 Imm(M, Rn ) | j1r g t W}
is a residual set in Imm(M, Rn ).
(ii) Suppose that F is G-versal and let W ⇢ J r (M, Rk )s be a G-invariant
submanifold, with s 1. Then the set
RW = {g 2 Emb(M, Rn ) | j1r g t W}
 Contact between submanifolds of Rn 85

is residual in Emb(M, Rn ).
If W is closed and M is compact, then in (i) and (ii) RW is open and
dense.

Proof. See [Montaldi (1986a)]. The arguments of the proof of (ii) hold
only for embeddings g : M ! Rn . ⇤

Remark 4.1. Theorem 4.4 still holds if one replaces the ambient space Rn
by a manifold Z and the target space by another manifold Q.

4.3 The meaning of generic immersions

We clarify here what is meant by a generic immersion of a given m-

dimensional manifold M in Rn with respect to a given family of model
submanifolds of Rn of codimension k parametrized by a manifold B, with
dim B = d. Let F : Rn ⇥ B ! Rk be the family of submersions defining the
model manifolds, i.e. Nb = Fb 1 (0), and g : M ⇥ B ! Rk the composite
g = F g, where g : M ! Rn is an immersion of M in Rn . For each
fixed b 2 B, g measures the contact of M and Nb at the given point.
The desirable properties of a generic immersion are described in terms of
the Ke -versality of the family g . More precisely, we say that g is generic
(with respect to its contacts with the family of model submanifolds) if g is
Ke -versal. A versal unfolding of a map-germ provides a model of all possi-
ble singularities appearing under small perturbations of the map-germ. In
particular, a Ke -versal family gives information on all local contacts of M
and the elements of the family (Nb )b2B .
The applications of this setting to di↵erential geometry of M may re-
quire finer equivalence relations than contact equivalence. This is the case,
for instance, when we are concerned with geometric information derived
from the contact of M not only with Fb 1 (0), but also on nearby level sets
of Fb . In this case, we use A (or Ae ) equivalence of the composite map g,b ,
when the dimension of the target is bigger than 1, and R+ -equivalence
otherwise. If two map-germs are A (respectively R+ ) equivalent, the folia-
tions defined by their level sets are isomorphic. In this book we are mainly
concerned with the action of the three groups G = K, A, R+ . A generic
immersion with respect to G-equivalence is an immersion g : M ! Rn for
which the corresponding composite map g,b is Ge -versal.
An unfolding is G-versal if and only if it is transversal to the G-orbits,
(see Theorem 3.6), this transversality condition being verified at the level
86 Di↵erential Geometry from a Singularity Theory Viewpoint

of jet spaces. Mather in a series of papers found the tools to build the
bridge between versality and transversality. His results give a method to
describe the generic singularities in terms of transversality of the r jet of
the mapping to a J r G invariant stratification of J r (M, Rk ). The idea is to
determine the pair of dimensions (m, k) for which the relevant strata of
this stratification are the r-jets of simple G-orbits. When G = K, Mather
computed in [Mather (1971)] the codimension (m, k) of the set of K non-
simple singularities in J0r (M, Rk ) - the fibre of J r (M, Rk ) over a point
(x, 0) in M ⇥ Rk - for sufficiently large r. As we will see, when (m, k)
is sufficiently large, transversality to the K-invariant stratification in jet
space means avoiding non-simple singularities. The pairs (m, k) satisfying
this condition form the nice dimensions.
The action of the group K does not move the origin in Rk , so the Ke -
tangent space to any singular jet is contained in J0r (M, Rk ). Hence, the
Ke -codimension of the set of all non-simple singularities in the jet space
J r (M, Rk ) is (m, k) + k.
For a given immersion g : M ! Rn , it also follows that the dimension of
the image of the associated jet-extension map j1r g : M ⇥ B ! J r (M, Rk )
is m + d, where d = dim B, and g is the composite F g.
Suppose that m + d < (m, k) + k (this will be the case for all the
applications we consider in this book). Let {W1 , . . . , Ws } be the finite set
of all the K-orbits in J r (M, Rk ) of Ke -codimension less than m + d and let
{Ws+1 , . . . , Wt } be a finite stratification of the complement of W1 [. . .[Ws .
Let R be the residual set of immersions given by the intersection of the sets
RWi , i = 1, . . . , t in Theorem 4.3. For g 2 R, it follows from the condition
m + d < (m, k) + k that the associated jet-extension map j1r g misses
the strata Wi for i > s, and is transverse to the strata Wi for i  s. We
call the immersions g 2 R generic immersions.
When m+d (m, k)+k, a generic embedding does not avoid in general
the non-simple singularities. In this case, a stratification of J0r (M, Rk ) is
given by the strata which are the simple K-orbits together with strata
which are the union of the non-simple K-orbits parametrised by the moduli
(usually excluding some exceptional values of the moduli).
When the dimensions m, d, k are such that the non-simple singularities
are not encountered for a generic immersion, the strata of codimension
smaller than or equal to m + d are K-orbits, and the transversality of j1r g
to the stratification in jet space, for sufficiently high values of r gives that
g is Ke -versal. If, on the other hand, the non-simple singularities are
present, then all the singularities that arise for a generic immersion cannot
 Contact between submanifolds of Rn 87

be presented transversely.
The formulae for the codimension r (m, k) of the algebraic variety con-
sisting of all non-simple singularities in J0r (M, Rk ), are given by Mather
in [Mather (1971)]. The number r (m, k) is a decreasing function of r,
and (m, k) is defined to be infr r (m, k). The formulae for (m, k) are as
follows:

Case I: m  k
(m, k) = 6(k m) + 8 if k m 4 and m 4
= 6(k m) + 9 if 0 < k m < 3 and m > 4, or
if m = 3
= 7(k m) + 10 if m = 2
=1 if m = 1

Case II: m > k

(m, k) = 9 if m k=1
=8 if m k=2
=m k+7 if m k 3

Remark 4.2. In some applications in this book, for instance when con-
sidering the family of projections into linear spaces (section 4.6), we take
G = A or R+ if k = 1. The orbits of the extended actions of these groups
are invariant by translations y ! y + c, y and c in Rk . Then, for G = A, R+ ,
and f : (Rm , 0) ! (Rk , 0), the Ge -tangent space of f contains the vector
@ @
subspace of ✓f generated by { @y i
f }, i = 1, . . . , k where { @y i
, i = 1, . . . , k}
are the generators of ✓k .
When k = 1, the R+ -simple germs coincide with the K-simple germs
coincide. When k > 1 and G = A, a formula for the codimension A (m, k)
of the set of the Ae -non-simple singularities is not known in general (see
[Rieger and Ruas (2005)], [Oset Sinha, Ruas and Atique (2015)] for
some partial results). In this case, we proceed as follows. With similar
arguments as above, for each pair (m, k), if the A-classification of map-
germs (Rm , 0) ! (Rk , 0) of Ae -codimension  d = dim(B) is finite, that
is, all orbits in this classification are A-simple, then we let {W1 , . . . , Ws }
to be the finite set of A-orbits in J r (M, Rk ) of Ae -codimension  d,
and let {Ws+1 , . . . , Wt } to be a finite stratification of the complement of
W1 [ . . . [ Ws . In this case, we set W to be this stratification. Again, g is
a generic immersion if j r g t W.
88 Di↵erential Geometry from a Singularity Theory Viewpoint

4.4 Contact with hyperplanes

We take the model submanifolds to be the hyperplanes in Rn . A hyperplane

in Rn is determined, in a unique way, by a unit vector v in Rn and a scalar
r. If H(v, r) denotes such a hyperplane, then
H(v, r) = {y 2 Rn | hy, vi r = 0}.
e : Rn ⇥ S n
The family (4.2) can be written as H 1
⇥ R ! R with
e v, r) = hy, vi
H(y, r. (4.3)
The family He is called the (universal) extended family of height func-
tions. Here the manifold B in section 4.1 is S n 1 ⇥ R and the function
e
h(v,r) : Rn ! R, given by e e v, r), is clearly a submersion
h(v,r) (y) = H(y,
for any (v, r) 2 S n 1 ⇥ R. The zero fibre of this function is precisely the
hyperplane H(v, r).
We are also interested in the contact of submanifolds with families of
parallel hyperplanes. This is why we also consider the (universal) family of
height functions H : Rn ⇥ S n 1 ! R given by
H(y, v) = hy, vi.
Given an immersion g : M ! Rn of a submanifold M into Rn , we
consider the family of height functions H : M ⇥ S n 1 ! R on M defined
by
H(p, v) = H(g(p), v) = hg(p), vi. (4.4)
e : M ⇥ Sn
The extended family of height functions H 1
⇥ R ! R on M
is defined by
e v, r) = H(g(p),
H(p, e v, r) = hg(p), vi r.
For v fixed, we denote by hv : M ! R the function given by
hv (p) = H(p, v). Following Theorem 4.1, the contact of g(M ) with the
family of parallel hyperplanes determined by v 2 S n 1 is measured by the
K-singularities of the function hv .
For low values of m and n, the singularities of hv of a generic immersed
m-dimensional submanifold in Rn can be obtained using Theorem 4.3. We
give below the singularities of the height function in the case of generic
immersions of curves and surfaces (i.e., m = 1 or m = 2).

Theorem 4.5. (i) For an open and dense set of immersions of a smooth
curve C in Rn , n e (resp. H) is locally Ke -versal (resp.
2, the family H
P-R+ -versal).
 Contact between submanifolds of Rn 89

(ii) For an open and dense set of immersions of a 2-dimensional surface

M in Rn , with 3  n  7, the family He (resp. H) is locally Ke -versal (resp.
P-R+ -versal).

Proof. Let G = K. The extended height function family H e is a n para-

metric family. According to the discussion in section 4.3, we should find n
such that the condition m + d < (m, k) + k holds, where m = 1 in (i)
and m = 2 in (ii), d = n and k = 1. Now, (1, 1) = 1 and (2, 1) = 9, and
the result follows from Theorem 4.3.
The arguments for G = R+ are similar. Notice that in this case d = n 1
and the condition m + d < (m, k) gives the same inequality as above. ⇤

Theorem 4.6. For a generic immersed curve C in Rn , the local K-

singularities of hv are of type Ak , k = 1, . . . , n.

Proof. The simple singularities in this case are the Ak -singularities.

These describe the simple orbits of G = K and R+ . The strata of the strat-
ification of the jet space J r (1, 1) are the Ak -orbits. Hence, for a generic
embedding of a curve in Rn , the r-jet of the extended family of the height
e : C ⇥ S n 1 ⇥ R ! J r (C, R) avoids all strata of codimension
functions, J1r H
bigger than n. ⇤

Theorem 4.7. (i) For a generic immersed surface M in R3 , the local K-

singularities of hv are of type Ak , k = 1, 2, 3.
(ii) For a generic immersed surface M in R4 , the local K-singularities
of hv are of type Ak , k = 1, 2, 3, 4 or D4 .
(iii) For a generic immersed surface M in R5 , the local K-singularities
of hv are Ak , k = 1, 2, 3, 4, 5, D4 or D5 .

Proof. The argument is similar to that in the proof of Theorem 4.6. Here
m = 2 and k = 1 and one needs to stratify J0r (M, R). From Theorem 4.5
the relevant strata of the stratification in jet space are the orbits of the
simple singularities Ak , Dk , E6 , E7 , E8 .
Now, the generic singularities of the height function must have codi-
mension less than or equal to n. Hence, when n = 3, 4, 5 we get respectively
those in (i), (ii) and (iii). ⇤

Remark 4.3. Theorem 4.7 gives the possible singularities of the height
function on an immersed surface M in Rn , n = 3, 4, 5. The singularities
describe the contact of g(M ) with hyperplanes, where g is a generic immer-
sion M ! Rn . We extract in Chapters 6, 7, 8 extrinsic geometric properties
90 Di↵erential Geometry from a Singularity Theory Viewpoint

of g(M ) from each of the generic singularities of the height function. We

also do the same for the families of functions and maps in the subsequent
sections.

4.5 The family of distance squared functions

In this section, the model submanifolds in Rn are taken to be hyperspheres.

A hypersphere in Rn is determined, in a unique way, by its centre a 2 Rn
and its radius r. Let
S(a, r) = {y 2 Rn | hy a, y ai r2 = 0}
e :
denote such a hypersphere. Then the family (4.2) can be written as D
Rn ⇥ Rn ⇥ R ! R, with
e a, r) = hy
D(y, a, y ai r2 . (4.5)
The family De is called the (universal) extended family of distance squared
functions. Here, the manifold B in section 4.1 is Rn ⇥ R and the function
e
da,r : Rn ! R, given by e e a, r), is a submersion if and only if
da,r (y) = D(y,
r 6= 0. The zero fibre of this function is the hypersphere S(a, r), including
the degenerate hypersphere when r = 0 (which is just the point a). This is
the reason why we chose in section 4.2 to cut out Nb as we can in this way
include singular fibres.
We are interested in the contact of a submanifold with a family of hyper-
spheres with the same centre and consider the (universal) family of distance
squared functions D : Rn ⇥ Rn ! R, given by
D(y, a) = hy a, y ai.
Given an immersion g : M ! Rn of a submanifold M into Rn , we
consider the family of distance squared functions D : M ⇥ Rn ! R on M
defined by
D(p, a) = D(g(p), a) = hg(p) a, g(p) ai. (4.6)
e : M ⇥ Rn ⇥ R ! R
The extended family of distance squared functions D
on M is defined by
e a, r) = D(g(p),
D(p, e a, r) = hg(p) a, g(p) ai r2 .
For v fixed, we denote by da : M ! R the function given by
da (p) = D(p, a). Following Proposition 4.1, the contact of g(M ) with the
family of hyperspheres with the same centre a 2 Rn is measured by the
 Contact between submanifolds of Rn 91

R+ -singularities of the function da . Here too, the singularities of the dis-

tance squared function da of a generic immersed m-dimensional submani-
fold in Rn can be obtained using Theorem 4.3. We give below the possible
singularities of the distance squared function in the case of curves and
surfaces.
The proofs of Theorems 4.8, 4.9, 4.10 are analogous to those of Theorems
4.6, 4.7 and are omitted.
Theorem 4.8. (i) For an open and dense set of immersions of a curve C
in Rn , n 2 the family D̃ (resp. D ). is locally Ke -versal (resp. P-R+ -
versal).
(ii) For an open and dense set of immersions of a 2-dimensional surface
M in Rn , with 3  n  6, the family D̃ (resp. D ) is locally Ke -versal (resp.
P-R+ -versal).
Theorem 4.9. For a generic immersed curve C in Rn , the K-singularities
of Da are of type Ak , k = 1, . . . , n + 1.
Theorem 4.10. For a generic immersed surface M in R3 , the K-
singularities of Da are of type Ak , k = 1, 2, 3, 4 or D4 .
One can relate the contact of a submanifold in Rn with hyperplanes
and the contact between submanifolds in S n with hyperspheres using the
stereographic projection.
Denote by N p = (0, · · · , 0, 1) the north pole on the unit n-sphere S n of
Rn+1 . The stereographic projection : S n \ {N p} ! Rn ⇥ {0} assigns to
each point q 2 S n \ {N p} the point x 2 Rn ⇥ {0} ⌘ Rn determined by the
intersection of the hyperplane Rn with the line in Rn+1 that passes through
the points N p and q. The map is a di↵eomorphism and we denote its
inverse by ' : Rn ! S n \ {N p}. The map ' maps a hypersphere in Rn to
a hypersphere in S n , and maps a hyperplane in Rn to a hypersphere in S n
passing through the north pole N p.
The image '(S(a, r)) of the hypersphere of radius r and centre a 2 Rn
is a hypersphere in S n and there is a unique hyperplane H(a, r) of Rn+1
whose intersection with S n is '(S(a, r)).
Given an immersed manifold M in Rn , its image '(M ) is a submanifold
of Rn+1 contained in the hypersphere S n . Since ' is a di↵eomorphism, it
follows from Definition 4.1 that it preserves the contacts between subman-
ifolds. Hence the contact of M with any hypersphere S(a, r) at a point p
in Rn is the same as the contact of '(M ) with the hypersphere '(S(a, r))
at '(p) in S n .
92 Di↵erential Geometry from a Singularity Theory Viewpoint

Theorem 4.11. Given the hypersphere S(a, r) in Rn , suppose that n(a, r)

and ⇢(a, r) are respectively a normal unit vector and the distance to the
origin of the hyperplane H(a, r) determining '(S(a, r)) in Rn+1 . Suppose
that g : M ! Rn is an immersion of a manifold M in Rn and that S(a, r)
is tangent to g(M ) at the point p0 = g(q0 ). Then the germs at q0 of the
extended distance squared function D̃(a,r) on M and of the extended height
function H̃(n(a,r),⇢(a,r)) on '(M ) are K-equivalent.
Proof. Consider the function a,r : Rn+1 ! R given by
a,r = hu, n(a, r)i ⇢(a, r).
1
We have ( a,r ) (0) = H(a, r), so a,r i ' x is a contact map of
the pair ('(M ), H(a, r)) at p in Rn+1 , where i is the canonical inclusion of
the hypersphere S n in Rn+1 .
If we denote h̄a,r = a,r |S n , we get (h̄a,r ) 1 (0) = '(S(a, r)) = Ha,r \S n .
Hence h̄a,r ' x is the contact map of the pair ('(M ), '(S(a, r))) in S n .
As h̄a,r ' x = a,r i ' x, it follows that
ha,r ' x ⇠K a,r i ' x.
Since d˜a,r is also the contact map of M and S(a, r) too and h̃a,r =
a,r i ' x, it follows that
d˜a,r ⇠K a,r i ' x ⇠K h̃a,r . ⇤
Remark 4.4. It is shown in [Romero Fuster (1997)] that the extended
family of distance squared functions D̃ on M and the extended family
height functions H̃ on '(M ) are K-equivalent as unfoldings.

4.6 The family of projections into hyperplanes

We consider here the contact of a submanifold in Rn with lines. We shall

bundle together all parallel lines and represent them by their unit direction
vectors v 2 S n 1 . A vector v 2 S n 1 is the kernel of a linear map from Rn
to an (n 1)-dimensional linear space V . We choose V to be the hyperplane
orthogonal to v. This is precisely Tv S n 1 , the tangent space to the unit
sphere S n 1 at v. We take the linear map to be the orthogonal projection
in Rn to Tv S n 1 parallel to v. Thus, a point y 2 Rn is projected to a point
q = y + v. As q 2 Tv S n 1 , hq, vi = 0 and this gives = hy, vi. By
varying v in S n 1 , we obtain the (universal) family of orthogonal projections
P : Rn ⇥ S n 1 ! T S n 1 given by
P(y, v) = (v, y hy, viv).
 Contact between submanifolds of Rn 93

Here the manifold B in section 4.1 is S n 1 and the map Pv : Rn !

Tv S n 1 , given by Pv (y) = y hy, viv, is clearly a submersion for any
v 2 S n 1 . The zero fibre of this map is the line through the origin parallel
to v.
Given an immersion g : M ! Rn of a submanifold M into Rn , we
consider the family of orthogonal projections P : M ⇥ S n 1 ! T S n 1 on
M defined by
P (p, v) = P(g(p), v) = (v, g(p) hg(p), vi). (4.7)
Following Theorem 4.1, the contact of g(M ) with the family of parallel
lines to v 2 S n 1 is measured by the K-singularities the mapping Pv given
by
Pv (p) = g(p) hg(p), vi.
Given a point p0 2 M , we choose a local parametrisation x : U ⇢
R ! M of M at p0 with x(0) = p0 (we assume of course that 0 2 U ). We
m

also identify Tv S n 1 with Rn 1 and suppose that Pv (p0 ) = 0. Then the

composite map Pv x is locally a map-germ (Rm , 0) ! (Rn 1 , 0).
There is an advantage in considering the A-singularities of the map-
germ Pv x instead of its K-singularities when n 3 as the dimension of
the target is greater than 1. In fact, not only the zero sets, but also the
singular sets and the discriminants of two A-equivalent contact maps Pv x
and Pv 0 x0 are di↵eomorphic.
We consider in some detail the case of surfaces (dim(M ) = 2) immersed
in Rn with n = 3, 4, 5.
The family of projections of surfaces in R3 to planes was investigated by
Koenderink and van Doorn [Koenderink and van Doorn (1976)], Ga↵ney
and Ruas [Ga↵ney and Ruas (1979)] and Arnol’d [Arnol’d (1979)]. The
family of projections of surfaces in R4 into 3-dimensional spaces was studied
by D. Mond [Mond (1985)]. These results can be also be proved using
Theorem 4.4.

Theorem 4.12. For an open and dense set of immersions of a surface M

in Rn , n = 3, 4, 5, the family of orthogonal projections is locally Ae -versal.
The local A-singularities of Pv are those in Table 4.1 for n = 3, Table 4.2
for n = 4 and Table 4.3 for n = 5.

Proof. Following Remark 4.2, we proceed as follows. First, we verify

if the A-classification of map-germs (R2 , 0) ! (Rn 1 , 0), n = 3, 4, 5 of
Ae -codimension less than or equal to d = n 1 is finite, that is, if all
94 Di↵erential Geometry from a Singularity Theory Viewpoint

Table 4.1 Local singularities of projections of surfaces in

R3 to planes ([Rieger (1987)]).
Name Normal form Ae -codimension
Immersion (x, y) 0
Fold (x, y 2 ) 0
Cusp (x, xy + y 3 ) 0
42 (Lips/Beaks) (x, y 3 ± x2 y) 1
43 (Goose) (x, y 3 ± x3 y) 2
5 (Swallowtail) (x, xy + y 4 ) 1
6 (Butterfly) (x, xy + y 5 ± y 7 ) 2
115 (Gulls) (x, xy 2 + y 4 + y 5 ) 2

Table 4.2 Local singularities of projections of surfaces in R4

to 3-spaces ([Mond (1985)]).
Name Normal form Ae -codimension
Immersion (x, y, 0) 0
Cross-cap (x, xy, y 2 ), 0
Sk (x, y 2 , y 3 + xk+1 y), k > 0 k
Bk (x, y 2 , x2 y + y 2k+1 ), k > 2 k
Ck (x, y 2 , xy 3 ± xk y), k 2 k
Hk (x, xy + ±y 3k 1 , y 3 ), k 2 k

Table 4.3 Local singularities of projections of surfaces in R5 to 4-s-

paces ([Kirk (2000)] and [Rieger (2007)]).
Name Normal form Ae -codimension
Immersion (x, y, 0, 0) 0
Ik (x, xy, y 2 , y 2k+1 ), k = 1, 2, 3, 4 k
II2 (x, y 2 , y 3 , xk y), k = 2 4
III2,3 (x, y 2 , y 3 ± xk y, xl y), k = 2, l = 3 4
VII1 (x, xy, xy 2 ± y 3k+1 , xy 3 ), k = 1 4

orbits in this classification are A-simple. Then we let {W1 , . . . , Ws } to

be the finite set of A-orbits in J r (M 2 , Rn 1 ) of Ae -codimension  n 1,
and let {Ws+1 , . . . , Wt } to be a finite stratification of the complement of
W1 [ . . . [ Ws . The result then follows applying Theorem 4.4.
In fact, for n = 3, the relevant A-orbits are given by the corank 1
singularities from the plane to the plane of Ae -codimension less than or
equal to 2, see Table 4.1. Similarly, for n = 4, the relevant singularities
are those in Table 4.2. The classification of singularities R2 ! R5 of Ae -
codimension 4 can be found in Table 4.3. ⇤
 Contact between submanifolds of Rn 95

4.7 Notes

We considered in this chapter the notion of contact and Thom’s transver-

sality theorem as main tools for investigating generic properties of subman-
ifolds of Rn .
The application of singularity theory to the study of the extrinsic geom-
etry of submanifolds in Euclidean spaces started in the late 1960’s. René
Thom, in his investigation of parabolic points on surfaces in R3 , pointed
at the importance of the family of height functions as a tool for studying
the di↵erential geometry of submanifolds of Rn . The idea of René Thom
was taken up by Porteous in [Porteous (1983a,b)] where he considered the
generic singularities of the distance squared function.
The subject has had a great development since then, and problems
on extrinsic geometry of submanifolds in Euclidean spaces motivated sev-
eral extensions of Thom’s transversality theorems (see for example [Abra-
ham (1963)], [Buchner (1974)], [Looijenga (1974)], [Mather (1970)], [Wall
(1977)] for an account of the earlier works on this subject).
Bruce in [Bruce (1986)] extended the notion of contact between mani-
folds to the notion of contact of a submanifold with an eventually singular
space, and defined weak transversality. With weak transversality, one can
consider K-finitely determined germs and not merely submersion-germs.
A more general result is given in [Montaldi (1986a)], in which it is
allowed contact of submanifolds of Rn with model submanifolds given by
an G-versal family of mappings F. Montaldi’s Theorem 4.4 holds for any
Damon’s geometric subgroup, allowing a wider range of applications to
di↵erential geometry.
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws

This page intentionally left blank

 Chapter 5

Lagrangian and Legendrian

Singularities

A caustic in Euclidean space is the envelope of light rays reflected on a

concave surface. It is the set of points in the space which are mostly illumi-
nated by these rays. Caustics can be visible (the reader may try to catch
one on a surface of a cup of co↵ee or tea).
A phenomenon closely related to caustics, but which is not visible, is
wavefronts. If we think of light as particles propagating at unit speed along
the normal lines to a surface, then these particles trace another surface at
any given time t. This surface is called the wavefront of the original surface
(which is called the initial front). Wavefronts may acquire singularities,
and their singularities form the caustic.
The generic singularities that can occur in caustics and wavefronts and
the way they deform as the original front is deformed are described by
Arnol’d and Zakalyukin using the theory of Lagrangian and Legendrian sin-
gularities [Arnol’d, Guseı̆n-Zade and Varchenko (1985); Zakalyukin (1976,
1984)]. This theory was initiated by L. Hörmander [Hörmander (1971)]
and has wide applications [Arnol’d (1983); Izumiya (1993, 1995); Izumiya
and Kossioris (1995, 1997a,b); Izumiya, Kossioris and Makrakis (2001);
Izumiya and Janeczko (2003); Izumiya, Pei and Sano (2003); Zakalyukin
(1995)]. We apply it here to study some aspects of the extrinsic geometry
of a submanifold of Euclidean spaces.
We start by recalling some basic concepts in symplectic and contact
geometries. We state a fundamental result that says that any Lagrangian
or Legendrian manifold is defined locally by a generating family of func-
tions. The generating families of interest here are the families of distance
squared functions, the family of height functions and the extended family
of height functions on a submanifold in an Euclidean space. We showed in
Chapter 4 how these families measure the contact of a submanifold with

97
98 Di↵erential Geometry from a Singularity Theory Viewpoint

model hypersurfaces (totally umbilic hypersurfaces). We consider here the

theory of contact from the view point of the Lagrangian and Legendrian
singularities.

5.1 Symplectic manifolds

A symplectic form ! on a smooth manifold M is a closed, non-degenerate

and skew-symmetric 2-form.
It follows from the above definition of a symplectic form ! that d! = 0,
n
! is a volume form and that dim M = 2n, for some positive integer n.
A manifold M equipped with a symplectic form ! is called a symplectic
manifold. We also call ! a symplectic structure on M .

Example 5.1. Let R2n = {(x1 , . . . , xn , p1 , . . . , pn ) | xi , pi 2 R } and con-

sider the 1-form
X n
= pi dxi .
i=1

This yields the symplectic form

n
X
!= d = dxi ^ dpi
i=1

on R2n . The vector space R2n equipped with this symplectic form is the
standard model of a linear symplectic manifold.

Example 5.2 (The cotangent bundle). Let N be a smooth manifold

and T ⇤ N its cotangent bundle. There is a canonical symplectic structure
on T ⇤ N which is defined as follows (see for example [McDu↵ and Salamon
(1995)] for details). Denote by ⇢ : T ⇤ N ! N the canonical projection
defined by ⇢(q, v) = q for any (q, v) 2 T ⇤ N (so q 2 N and v : Tq N ! R is
an element of Tq⇤ N ).
The di↵erential map d⇢ : T (T ⇤ N ) ! T N determines at each point
(q, v) 2 T ⇤ N a linear map
d⇢(q,v) : T(q,v) (T ⇤ N ) ! Tq N.
The canonical 1-form (or, the Liouville form) on T ⇤ N is defined
at each point (q, v) 2 T ⇤ N by the linear map (q,v) = v d⇢(q,v) :
T(q,v) (T ⇤ N ) ! R given by

(q,v) (w) = v(d⇢(q,v) (w)).

 Lagrangian and Legendrian Singularities 99

The canonical symplectic structure on T ⇤ N is given by the 2-form

!= d .

Let x : U ! Rn be a local system of coordinates of N and write x =

(x1 , . . . , xn ), where xi : U ! R, i = 1, . . . , n, are smooth functions. The
1-forms dxi (q) : Tq N ! R, i = 1, . . . , n, form a basis of Tq⇤ N , so any v in
Tq⇤ N can be written in a unique way in the form
n
X
v= pi (q, v)dxi (q).
i=1

The dual coordinates pi are completely determined by q and v and yield a

smooth map p : T ⇤ U ! Rn given by

p(q, v) = (p1 (q, v), . . . , pn (q, v)).

We obtain a local system of coordinates : T ⇤ U ! Rn ⇥ Rn of T ⇤ N

given by

(q, v)) = (x(q), p(q, v)).

In this system of coordinates, the 1-form is given by

n
X
= pi dxi
i=1

and the symplectic form ! by

n
X
!= dxi ^ dpi .
i=1

Definition 5.1. Let (M1 , !1 ) and (M2 , !2 ) be two symplectic manifolds.

A symplectomorphism between (M1 , !1 ) and (M2 , !2 ) is a di↵eomorphism
: M1 ! M2 which sends the symplectic structure !2 on M2 to the
symplectic structure !1 on M1 , that is, '⇤ !2 = !1 . More specifically,
!1 (q, v1 , v2 ) = !2 ( (q), d q (v1 ), d q (v2 )) for any q 2 M1 and v1 , v2 in Tq M1 .

Theorem 5.1 (The Darboux Theorem). Any two symplectic mani-

folds of the same dimension are locally symplectomorphic.

Proof. See [Arnol’d, Guseı̆n-Zade and Varchenko (1985)]. ⇤

100 Di↵erential Geometry from a Singularity Theory Viewpoint

5.1.1 Lagrangian submanifolds and Langrangian maps

Definition 5.2. Let M be a 2n-dimensional smooth manifold and let ! be

a symplectic form on M . We say that a smooth submanifold L of M is a
Lagrangian submanifold if dim L = n and !|L = 0.

Example 5.3. Consider the cotangent bundle T ⇤ N of a smooth manifold

N endowed with the canonical symplectic structure (Example 5.2). Then
the fibres of the bundle ⇢ : T ⇤ N ! N and the zero section of this bundle are
Lagrangian submanifolds. We can construct other Lagrangian submanifolds
of T ⇤ N as follows.
Let f : N ! R be a smooth function. Then the di↵erential of f can be
considered as an embedding Df : N ! T ⇤ N , given by Df (x) = (x, dfx ). In
the local system of coordinates x = (x1 , . . . , xn ) and their dual coordinates
p = (p1 , . . . , pn ), we have
✓ ◆
@f @f
Df (x) = x, (x), . . . , (x) ,
@x1 @xn
so that the pull-back of the canonical 1-form by Df is given by
Xn
⇤ @f
Df (x) = (x)dxi = dfx .
i=1
@x i

Then Df ⇤ ! = ddf = 0, so Df (N ) is a Lagrangian submanifold of T ⇤ N.

Definition 5.3. Let ⇡ : E ! N be a fibre bundle such that E is a sym-

plectic manifold. We say that ⇡ : E ! N is a Lagrangian fibration if its
fibres are Lagrangian submanifolds of E.

Example 5.4. Consider the linear symplectic space R2n and let ⇡ : R2n !
Rn be defined by ⇡(x1 , . . . , xn , p1 , . . . , pn ) = (x1 , . . . , xn ). Then ⇡ is a
Lagrangian fibration.

Example 5.5. The cotangent bundle ⇢ : T ⇤ N ! N is a Lagrangian fi-

bration. Indeed, the canonical 1-form is zero along its fibres, and hence
! = d is also zero along the fibres.

Let ⇡ : E ! N and ⇡ 0 : E 0 ! N 0 be Lagrangian fibrations. A symplec-

tomorphism : E ! E 0 is said to be a Lagrangian di↵eomorphism if there
exists a di↵eomorphism : N ! N 0 such that ⇡ 0 = ⇡.

Theorem 5.2. All Lagrangian fibrations of a fixed dimension are locally

Lagrangian di↵eomorphic.
 Lagrangian and Legendrian Singularities 101

Proof. See [Arnol’d, Guseı̆n-Zade and Varchenko (1985)]. ⇤

Let ⇡ : E ! N be a Lagrangian fibration and consider a Lagrangian
immersion i : L ! E (i.e., i⇤ ! = 0). The restriction of the projection ⇡ to
i(L), that is ⇡ i : L ! N, is called a Lagrangian map.
A caustic is the set of critical values of a Lagrangian map. We denote
the caustic of the Lagrangian map ⇡ i : L ! N by C(i(L)).
Two Lagrangian maps ⇡ i : L ! N and ⇡ 0 i0 : L0 ! N 0 are said
to be Lagrangian equivalent if there is a Lagrangian di↵eomorphism :
0 0 0 0 0 0
E ! E such that (i(L)) = i (L ). If ⇡ i : L ! N and ⇡ i : L !
N 0 are Lagrangian equivalent, then the caustics C(i(L)) and C(i0 (L0 )) are
di↵eomorphic.

5.1.2 Lagrangian singularities

We consider in this section the local singularities of Lagragian maps. All
the concepts in §5.1.1 can be defined for germs. We start by stating a
key result in the theory of Lagrangian maps which describes the germ of a
Lagrangian immersion i : L ! E using germs of families of functions.
From Theorem 5.2, all Lagrangian fibrations are locally Lagrangian dif-
feomorphic. Therefore, we can work on the cotangent bundle ⇡ : T ⇤ Rn !
Rn and all the results there will be valid on any Lagrangian fibration.
Let (x, p) = (x1 , . . . , xn , p1 , . . . , pn ) denote the canonical coordinates on
T R , the canonical 1-form and ! the canonical symplectic form on T ⇤ Rn
⇤ n

(Example 5.1).
Let F : (Rk ⇥Rn , 0) ! (R, 0) be an n-parameter family of germs of func-
tions from (Rk , 0) to (R, 0), and denote by (q, x) = (q1 , . . . , qk , x1 , . . . , xn )
the coordinates in Rk ⇥ Rn . In Chapter 3, we associated some set germs
the family of functions F . The germ of the catastrophe set of F is the set
germ
⇢
@F @F
(CF , 0) = (q, x) 2 (Rk ⇥ Rn , 0) (q, x) = · · · = (q, x) = 0
@q1 @qk
and its bifurcation set is the set germ
⇢ ⇣ @2F ⌘
n
(BF , 0) = x 2 (R , 0) 9(q, x) 2 CF such that rank (q, x) < k .
@qi @qj
Let ⇡2 : (Rk ⇥Rn , 0) ! (Rn , 0) denote the canonical projection and con-
sider the map-germ ⇡CF which is given by the restriction of the projection
⇡2 to (CF , 0). Thus,
⇡CF : (CF , 0) ! (Rn , 0)
102 Di↵erential Geometry from a Singularity Theory Viewpoint

with ⇡CF (q, x) = x for any (q, x) 2 (CF , 0). The map-germ ⇡CF is the
catastrophe map of F .
We say that F is a Morse family of functions if the map-germ F :
(R ⇥ Rn , 0) ! (Rk , 0) given by
k

✓ ◆
@F @F
F (q, x) = ,..., (q, x)
@q1 @qk
is not singular. When F is a Morse family of functions, (CF , 0) is a germ of
a smooth submanifold of (Rk ⇥ Rn , 0) of dimension n. We immerse (CF , 0)
in the cotangent bundle T ⇤ Rn by the map-germ L(F ) : (CF , 0) ! T ⇤ Rn
defined by
✓ ◆
@F @F
L(F )(q, x) = x, (q, x), . . . , (q, x) .
@x1 @xn
(One can check that this is indeed an immersion.) We have
Xn
⇤ @F
L(F ) = dxi |CF = dF |CF
i=1
@x i

so that
L(F )⇤ ! = L(F )⇤ d = dL(F )⇤ = d(dF |CF ) = (ddF )|CF = 0.
This proves that L(F ) is a germ of a Lagrangian immersion.
The family of map-germs F is called the generating family of the germ
of the Lagrangian submanifold L(F )(CF ) of T ⇤ Rn .
We have the following fundamental theorem.

Theorem 5.3. Let L be a germ of a Lagrangian submanifold of T ⇤ Rn .

Then there exists a germ of a Morse family of functions F such that
L(F )(CF ) = L.

Proof. See [Arnol’d, Guseı̆n-Zade and Varchenko (1985)]. ⇤

Remark 5.1. The bifurcation set of F is the set of critical values of the
catastrophe map-germ ⇡CF . Since ⇡CF (q, x) = ⇡ L(F )(q, x), it coincides
with the caustic C(L(F )).

A germ of a Lagrangian immersion i : (L, x) ! (T ⇤ Rn , p) (or a germ of

Lagrangian map ⇡ i : (L, x) ! (Rn , ⇡(p))) is said to be Lagrangian stable
if for any representative ī : V ! T ⇤ Rn of i, there exist a neighbourhood W
of ī (in the Whitney C 1 -topology on the subset of Lagrangian immersions
considered as a subspace of C 1 (Rn , T ⇤ Rn )) and a neighbourhood V of x
 Lagrangian and Legendrian Singularities 103

with the following property: For any Lagrangian immersion j̄ in W , there

exists x0 2 V such that ⇡ i and ⇡ j are Lagrangian equivalent, where
j : (L, x0 ) ! (T ⇤ Rn , p0 ) is the germ of j̄ at x0 .
Theorem 5.3 states that any germ of a Lagrangian submanifold of T ⇤ Rn
can be constructed from a generating family of functions. The notion of
Lagrangian equivalence and of Lagrangian stability can be formulated in
terms of generating families.

Theorem 5.4. (1) Let F : (Rk ⇥ Rn , 0) ! (R, 0) and G : (Rl ⇥ Rn , 0) !

(R, 0) be two Morse families of functions. Then the Lagrangian map-germs
⇡ L(F ) and ⇡ L(G) are Lagrangian equivalent if and only if F and G
are stably P -R+ -equivalent.
(2) The Lagrangian map-germ ⇡ L(F ) is Lagrangian stable if and only
if F is an R+ - versal unfolding of f (q) = F (q, 0).

Proof. See [Arnol’d, Guseı̆n-Zade and Varchenko (1985)]. ⇤

If k = l in Theorem 5.4(1), we replace stably P -R+ -equivalent by P -
R+ -equivalent.
The classification of singularities of germs of functions and of their R+ -
versal unfoldings can now be used to obtain a list of normal forms of the
singularities of Lagrangian map-germs in low dimensions.

Theorem 5.5. Let F : (Rk ⇥ Rn , 0) ! (R, 0) be a Morse family of func-

tions. Suppose that L(F ) : (C(F ), 0) ! T ⇤ Rn is Lagrangian stable and
n  4. Then L(F ) is Lagrangian equivalent to a germ of a Lagrangian
submanifold whose generating family G(q1 , . . . , qk , x1 , . . . , xn ) is one of the
following germs, where Q(qr , . . . , qk ) = ±qr2 ± · · · ± qk2 ,
(1) Q(q2 , . . . , qk ) + q13 + x1 q1 .
(2) Q(q2 , . . . , qk ) ± q14 + x1 q1 + x2 q12 .
(3) Q(q2 , . . . , qk ) + q15 + x1 q1 + x2 q12 + x3 q13 .
(4) Q(q2 , . . . , qk ) ± q16 + x1 q1 + x2 q12 + x3 q13 + x4 q14 .
(5) Q(q3 , . . . , qk ) + q13 q1 q22 + x1 q1 + x2 q2 + x3 (q12 + q22 ).
(6) Q(q3 , . . . , qk ) + q13 + q23 + x1 q1 + x2 q2 + x3 q1 q2 .
(7) Q(q3 , . . . , qk ) + q12 q2 + q24 + x1 q1 + x2 q2 + x3 q12 + x4 q22 .

Proof. It follows from Theorem 5.4 that F is an R+ -versal unfolding of

f = F |Rk ⇥{0} . Since n  4, code (f, R)  4 and by Theorem 3.13 we have
that F is P -R+ -equivalent to one of the germs in the above list. ⇤
The normal forms of the Lagrangian stable map-germs ⇡ L(G) when
n  4 are given in Table 5.1.
104 Di↵erential Geometry from a Singularity Theory Viewpoint

Table 5.1 Lagrangian stable singularities in Rn , n  4.

G singularity type ⇡ L(G) singularity type Normal form
A2 Fold q12
A3 Cusp (q13 + x2 q1 , x2 )
A4 Swallowtail (q14 + x2 q1 + x3 q12 , x2 , x3 )
A5 Butterfly (q15 + x2 q1 + x3 q12 + x4 q13 , x2 , x3 , x4 )
D4+ Elliptic umbilic (q12 q22 + x3 q1 , q12 q22 + x3 q2 )
D4 Hyperbolic Umbilic (q12 + x3 q2 , q22 + x3 q1 )
D5 Parabolic Umbilic (q1 q2 + x3 q1 , q12 + q23 + x4 q2 )

On the other hand, Golubitsky and Guillemin [Golubitsky and

Guillemin (1975)] gave an algebraic characterisation for the R+ -equivalence
among function germs. For any f 2 En , we denote by
⇢
@f @f
Jf = LRe · f = En ,..., ,
@x1 @xn
which is called the Jacobian ideal of f in En . We define a local ring by
R(k) (f ) = En /Jfk . Let [f ] be the image of f in this local ring. We say that
f satisfies the Milnor Condition if dimR R(1) (f ) < 1.

Proposition 5.1 ([Golubitsky and Guillemin (1975)]). Let f and g

be germs of functions in En satisfying the Milnor condition with df (0) =
dg(0) = 0. Then f and g are R+ -equivalent if and only if the ranks and
the signatures of the Hessians H(f )(0) and H(g)(0) are equal, and there is
an isomorphism : R(2) (f ) ! R(2) (g) such that ([f ]) = [g].

For Lagrangian stable Lagrangian map-germs, we have the following

classification theorem.

Theorem 5.6. Let F : (Rk ⇥ Rn , 0) ! (R, 0) and G : (Rk ⇥ Rn , 0) ! (R, 0)

be two Morse families of functions. Suppose that the Lagrangian map-germs
⇡ L(F ) and ⇡ L(G) are Lagrangian stable. Then the following statements
are equivalent:
(1) ⇡ L(F ) and ⇡ L(G) are Lagrangian equivalent,
(2) F and G are P -R+ -equivalent,
(3) f = F |Rk ⇥ {0} and g = G|Rk ⇥ {0} are R-equivalent,
(4) (a) the ranks and the signatures of the Hessians H(f )(0) and H(g)(0)
are equal, and
(b) there is an isomorphism : R2 (f ) ! R2 (g) such that ([f ]) = [g].

Proof. Statements (1) and (2) are equivalent by Theorem 5.4. By defi-
nition, statement (2) implies statement (3). Since ⇡ L(F ) and ⇡ L(G)
 Lagrangian and Legendrian Singularities 105

are Lagrangian stable, F and G are R+ -versal unfoldings of f and g, re-

spectively. The uniqueness of the versal deformation shows that statement
(3) implies statement (2). Since F and G are R+ -versal unfoldings of f
and g, f and g satisfy the Milnor condition, so Proposition 5.1 implies that
statements (3) and (4) are equivalent. ⇤

5.2 Contact manifolds

A contact structure on a manifold M is a maximally non-integrable field of

hyperplanes K in the tangent spaces of M . If ↵ is a 1-form that defines
K locally (i.e., ↵ 1 (0) = K|U in an open set U of M ), then the maximal
non-integrability condition is equivalent to ↵ ^ d↵n 6= 0 in U . It is also
equivalent to the 2-form d↵ being non-degenerate on each plane ↵ = 0. A
consequence of this is that M must be an odd dimensional manifold.
The local 1-form ↵ is called a contact form. A manifold M with a
contact structure K is denoted by (M, K) and is called a contact manifold.

Example 5.6. Let R2n+1 = {(x1 , . . . , xn , y, pn , . . . , pn ) | xi , y, pi 2 R } and

consider the 1-form
X n
↵ = dy pi dxi .
i=1

Then
↵ ^ d↵n = ( 1)n(n+1)/2 dx1 ^ . . . ^ dxn ^ dy ^ dp1 ^ . . . ^ dpn ,
which is the standard volume form in R2n+1 . Thus ↵ is a contact form. It
defines the standard contact structure on M = R2n+1 .

Example 5.7. Let N be a C 1 -manifold of dimension n and let J 1 (N, R)

be the manifold of 1-jets of functions on N. This is a (2n + 1)-dimensional
manifold and has a natural contact structure constructed as follows. We
can identify canonically J 1 (N, R) with T ⇤ N ⇥ R, so that we have a 1-form
↵ = dy , where y is a coordinate of R and is the Liouville form on
⇤
T N. Let x = (x1 , . . . , xn ) be a local system of coordinates in an open set
U of N . We obtain local coordinates (x1 , . . . , xn , p1 . . . , pn , y) in J 1 (U, R)
where y = f (x) and pi = @f /@xi (x), i = 1, . . . , n, for any f : N ! R and
Pn
x 2 U . The local contact form ↵ = dy i=1 pi dxi defines the natural
contact structure on J (N, R). This contact structure is independent of the
1

local system of coordinates.

106 Di↵erential Geometry from a Singularity Theory Viewpoint

Example 5.8. Let ⇡ : P T ⇤ N ! N be the projective cotangent bundle over

an n-dimensional manifold N. Consider the tangent bundle ⌧ : T (P T ⇤ N ) !
P T ⇤ N of P T ⇤ N and the di↵erential map d⇡ : T (P T ⇤ N ) ! T N of ⇡.
For any X in T (P T ⇤ N ), there exists an element ⇠ 2 T ⇤ N such that
⌧ (X) = [⇠], where [⇠] denotes the equivalence class in the projective space
represented by ⇠.
For an element v 2 T N, the property ⇠(v) = 0 does not depend on the
choice of a representative of the class [⇠]. Thus, we can define the canonical
contact structure on P T ⇤ N by the hyperplanes
K = {X 2 T (P T ⇤ N ) | ⌧ (X)(d⇡(X)) = 0}.
Let (x1 , . . . , xn ) be a local system of coordinates in an open set U of N .
Then we have a local trivialization P T ⇤ U ⇠ = U ⇥ P (Rn )⇤ If [⇠1 : . . . : ⇠n ]
denotes the homogeneous coordinates of the dual projective space P (Rn )⇤ ,
then ((x1 , . . . , xn ), [⇠1 : . . . : ⇠n ]) is a local system of coordinates in P T ⇤ U .
In the above system of coordinates, we have ⌧ (X) = [⇠] = [⇠1 : · · · : ⇠n ],
Pn Pn
where ⇠ = i=1 ⇠i dxi . Therefore, X 2 K(x,[⇠]) if and only if i=1 µi ⇠i = 0,
Pn @
where d⇡(X) = i=1 µi @x i
.
Consider the open subset V1 = {[⇠1 : · · · : ⇠n ] | ⇠1 6= 0} of P (Rn )⇤ . Then
we have the affine local system of coordinates 1 : U ⇥ V1 ! U ⇥ Rn 1 of
P T ⇤ U defined by
✓ ✓ ◆◆
⇠2 ⇠n
1 ((x1 , . . . , xn ), [⇠1 : · · · : ⇠n ]) = (x1 , . . . , xn ), ,..., .
⇠1 ⇠1
In this affine local system of coordinates, a contact form which defines
locally the contact structure K is given by
Xn
⇠i
↵ = dx1 + dxi .
i=2
⇠ 1

We obtain similar expressions for the contact forms in the coordinate

neighbourhoods U ⇥ Vi , where Vi = {[⇠1 : · · · : ⇠n ] | ⇠i 6= 0}, i = 2, . . . n.

For the projective cotangent bundle ⇡ : P T ⇤ (Rn ⇥ R) ! Rn ⇥ R, we

have the special open set V⌘ = {[⇠1 : · · · : ⇠n : ⌘] 2 P (Rn ⇥ R)⇤ |⌘ 6= 0}.
We then choose the system of coordinates ((x1 , . . . , xn , y), (p1 , . . . pn )) of
Rn ⇥ R ⇥ V⌘ , where pi = ⇠i /⌘, i = 1, . . . , n. In this local system of
coordinates, the contact form in P T ⇤ (Rn ⇥ R) is given by
n
X
↵ = dy pi dxi .
i=1
 Lagrangian and Legendrian Singularities 107

The contact manifold Rn ⇥ R ⇥ V⌘ can thus be identified with the 1-jet

space J 1 (Rn , R). Consequently, the 1-jet space J 1 (Rn , R) can be considered
as a special affine coordinate neighbourhood of P T ⇤ (Rn ⇥ R).

Definition 5.4. Let (M1 , K1 ) and (M2 , K2 ) be two contact manifolds. A

di↵eomorphism : M1 ! M2 which sends the contact structure K1 to the
contact structure K2 , that is, d (K1 ) = K2 , is called a contactomorphism
between (M1 , K1 ) and (M2 , K2 ). The two contact manifolds are then said
to be contactomorphic.

We have the contact version of the Darboux theorem.

Theorem 5.7 (The contact Darboux Theorem). Any two contact

manifolds of the same dimension are locally contactomorphic.

Proof. See [Arnol’d, Guseı̆n-Zade and Varchenko (1985)]. ⇤

5.2.1 Legendrian submanifolds and Legendrian maps

Definition 5.5. Let (M, K) be a (2n + 1)-dimensional contact manifold.

We say that a smooth submanifold L of M is a Legendrian submanifold if
dim L = n and Tp L ⇢ Kp for any p 2 L .

Example 5.9. Let N be a smooth manifold and f : N ! R a smooth

function. Then the 1-jet map j 1 f : N ! J 1 (N, R) is an embedding
of N in J 1 (N, R). This map is given in a local system of coordinates
(x1 , . . . , xn , p1 , . . . , pn , y) of J 1 (N, R) by
✓ ◆
1 @f @f
j f (x) = x, (x), . . . (x), f (x) .
@x1 @xn
Pn
Then the pull-back of the natural contact structure ↵ = dy i=1 pi dxi
in J (N, R) by j f is
1 1

Xn
1 ⇤ @f
j f ↵ = df (x) (x)dxi = df (x) df (x) = 0.
i=1
@x i

Therefore j 1 f (N ) is a Legendrian submanifold of J 1 (N, R).

Definition 5.6. Let ⇡ : E ! N be a fibre bundle such that E is a contact

manifold. We say that ⇡ : E ! N is a Legendrian fibration if its fibres are
Legendrian submanifolds of E.
108 Di↵erential Geometry from a Singularity Theory Viewpoint

Example 5.10. Consider the standard contact manifold R2n+1 (Example

5.6) and let ⇡ : R2n+1 ! Rn+1 be defined by ⇡(x1 , . . . , xn , p1 , . . . , pn , y) =
(x1 , . . . , xn , y). Then ⇡ is a Legendrian fibration.

Example 5.11. The canonical projection ⇡ : P T ⇤ N ! N is a Legendrian

fibration with respect to the canonical contact structure of P T ⇤ N (Example
5.8).

Let ⇡ : E ! N and ⇡ 0 : E 0 ! N 0 be two Legendrian fibrations. A

: E ! E 0 is said to be a Legendrian di↵eomorphism if there exists a
di↵eomorphism : N ! N 0 such that ⇡ 0 = ⇡.

Theorem 5.8. All Legendrian fibrations of a fixed dimension are locally

Legendrian di↵eomorphic.

Proof. See [Arnol’d, Guseı̆n-Zade and Varchenko (1985)]. ⇤

Let ⇡ : E ! N be a Legendrian fibration and consider a Legendrian
immersion i : L ! E. The composition of the projection with ⇡ i : L !
N is called a Legendrian map.
A wavefront is the image of a Legendrian map. The wavefront of the
Legendrian map ⇡ i : L ! N is denoted by W (i(L )). The map i : L ! E
is called the Legendrian lift of WL . The Legendrian submanifold i(L ) is
also referred to as the Legendrian lift of W (i(L )).
Two Legendrian immersions i : L ! E and i0 : L 0 ! E 0 (or, Legen-
drian maps) are said to be Legendrian equivalent if there is a Legendrian
di↵eomorphism : E ! E 0 such that (i(L )) = i0 (L 0 ). If i : L ! E and
i0 : L 0 ! E 0 are Legendrian equivalent, then their wavefronts W (i(L ))
and W (i0 (L 0 )) are di↵eomorphic.

5.2.2 Legendrian singularities

We consider the local singularities of Legendrian maps and work with E =
P T ⇤ Rn (see Theorem 5.8). Here too we have a result which describes a
germ of Legendrian immersion i : L ! E in terms of a germ of families of
functions.
Let F : (Rk ⇥ Rn , 0) ! (R, 0) be an n-parameter family of germs of
smooth function from (Rk , 0) to (R, 0). We say that F is a Morse family of
hypersurfaces if the map-germ ⇤ F : (Rk ⇥ Rn , 0) ! (R ⇥ Rk , 0) given by
✓ ◆
@F @F
⇤ F (q, x) = F, ,..., (q, x)
@q1 @qk
 Lagrangian and Legendrian Singularities 109

is not singular.
When F is a Morse family of hypersurfaces, the set-germ
⇢
@F @F
⌃F = (q, x) 2 (Rk ⇥ Rn , 0) | F (q, x) = (q, x) = · · · = (q, x) = 0
@q1 @qk
is a germ of a smooth (n 1)-dimensional submanifold of (Rk ⇥ Rn , 0).
Then we have the map-germ L (F ) : (⌃F , 0) ! P T ⇤ Rn defined by
✓  ◆
@F @F
L (F )(q, x) = x, (q, x) : · · · : (q, x) .
@x1 @xn
Pn
By definition the local contact form for K is given by ↵ = i=1 ⇠i dxi , where
((x1 , . . . , xn ), [⇠1 : · · · : ⇠n ]) are homogeneous coordinates of P T ⇤ (Rn ).
Therefore, we have
Xn
⇤ @F
L (F ) ↵ = dxi |⌃F = dF |⌃F = d(F |⌃F ) = 0.
i=1
@xi

We can show that L (F ) is an immersion-germ, so that it is a Legendrian

immersion-germ.
The family of function-germs F is called the generating family of the
germ of the Legendrian submanifold L (F )(⌃F ) of P T ⇤ Rn .

Theorem 5.9. Let L be a germ of a Legendrian submanifold of P T ⇤ Rn .

Then there exists a germ of a Morse family of hypersurfaces F such that
L (F )(⌃F ) = L .

Proof. See [Arnol’d, Guseı̆n-Zade and Varchenko (1985)] and [Zaka-

lyukin (1976)]. ⇤
The wavefront of the germ of the Legendrian map ⇡ L (F ) : (⌃F , 0) !
R is the set-germ
n

W (L (F )) = x 2 (Rn , 0) |9q 2 (Rk , 0) such that (q, x) 2 (⌃F , 0) .

This is precisely the discriminant of the family F (see Chapter 3).
Let i : (L , x) ! (P T ⇤ Rn , p) and i0 : (L 0 , x0 ) ! (P T ⇤ Rn , p0 ) be two
germs of Legendrian immersions. We say that i and i0 (or, the Legendrian
map-germs ⇡ i and ⇡ i0 ) are Legendrian equivalent if there exists a germ
of a Legendrian di↵eomorphism : (P T ⇤ Rn , p) ! (P T ⇤ Rn , p0 ) such that
(i(L )) = i0 (L 0 ).
A germ of a Legendrian immersion i : (L , x) ! (P T ⇤ Rn , p) (or, the
Legendrian map-germ ⇡ i) is said to be Legendrian stable if for any rep-
resentative ī : V ! P T ⇤ Rn of i, there exist a neighbourhood W of ī (in
110 Di↵erential Geometry from a Singularity Theory Viewpoint

the Whitney C 1 -topology on the space of Legendrian immersions consid-

ered as a subspace of C 1 (Rn 1 , P T ⇤ Rn )) and a neighbourhood W of ī
with the following property: For any Legendrian immersion j̄ in W , there
exists x0 2 V such that ⇡ i and ⇡ j are Legendrian equivalent, where
j : (L, x0 ) ! (P T ⇤ Rn , p0 ) is the germ of j̄ at x0 .
Since the Legendrian lift i : (L , p) ! (P T ⇤ Rn , p) is uniquely deter-
mined on the regular part of the wavefront W (i(L )), we have the following
property of germs of Legendrian immersions [Zakalyukin (1984)].

Theorem 5.10. Let ij : (Lj , pj ) ! (P T ⇤ Rn , p` ) be a germ of a Leg-

endrian immersion, j = 1, 2. Suppose that there exist representatives
ij : Uj ! P T ⇤ Rn of the germs such that the set of the regular points
of ⇡ ij is dense in Uj for each j = 1, 2. Then the following statements are
equivalent.
(1) The germs of Legendrian immersions i1 and i2 are Legendrian equiv-
alent.
(2) The germs of the wavefront sets W (i1 (L1 )), W (i2 (L2 )) are di↵eo-
morphic.

Proof. Statement (1) implies statement (2) by the definition of Legen-

drian equivalence. We now show that statement (2) implies statement (1).
By the uniqueness of the contact lift of a local di↵eomorphism on Rn to
P T ⇤ Rn , we may assume that there are open neighbourhoods Vj ⇢ Uj of pj
such that ⇡ i1 (V1 ) = ⇡ i2 (V2 ). We may also assume that Vj is relatively
compact and Vj ⇢ Uj for each j = 1, 2. Then, we have
⇡ i1 (V1 ) = ⇡ i1 (V1 )) = ⇡ i2 (V2 ) = ⇡ i2 (V2 ).
By hypothesis, the set of regular points of ⇡ ij |Vj is dense for each j = 1, 2.
We set S = ⇡ i1 (V1 ) = ⇡ i2 (V2 ), Zj = {⇡ ij (u) 2 S | u 2
Vj is a singular point of ⇡ ij },j = 1, 2, Z = Z1 [ Z2 and R = S \ Z. Then
we show that (⇡ ij ) 1 (a) is a finite set for each a 2 S \ Zj . Otherwise,
there exists a sequence {pn } such that ⇡ ij (pn ) = a, j = 1, 2. By taking
a subsequence, we may assume that {pn } converges to a point p 2 Vj . It
follows that ⇡ ij (p) = a. Since a 2 S \ Z` , there exists a neighbourhood
V of p such that ⇡ i` |V is an embedding. For sufficient large n, we have
pn 2 V and pn 6= p. However, we have ⇡ i` (pn ) = a = ⇡ i` (p). This
contradicts the fact that ⇡ i` |V is an embedding. Therefore (⇡ i` ) 1 (a)
is a finite set.
For any a 2 R, we have
1 1
(⇡ i1 ) (a) = {p1 , . . . , pm }, (⇡ i2 ) (a) = {q1 , . . . , ql }.
 Lagrangian and Legendrian Singularities 111

We have P T ⇤ Rn ⌘ Rn ⇥ (RP n 1 ⇤
) , so we get
i1 (p` ) = (a, [⌫` ]) and i2 (qk ) = (a, [⇠k ]),
where ⌫` , ⇠k 2 (Rn )⇤ = HomR (Rn , R) and [⌫` ], [⇠k ] denote the homogeneous
coordinates of the dual projective space. Since ij , j = 1, 2 are embed-
dings, ⌫1 , . . . , ⌫m (respectively, ⇠1 , . . . , ⇠l ) are mutually distinct. Here, a is
a regular value of ij (j = 1, 2), so that ⌫` (respectively, ⇠k ) is considered
to be the tangent hyperplane of one of the components of the hypersurface
⇡ i1 (V1 ) (respectively, ⇡ i2 (V2 )) through a. Since ⇡ i1 (V1 ) = ⇡ i2 (V2 ),
we may conclude that m = l and i1 (p` ) = i2 (q` ) (` = 1, . . . , m). We set
Wj = (⇡ ij |Vj ) 1 (R), j = 1, 2. It follows that i1 (W1 ) = i2 (W2 ). By the
continuity of ij , j = 1, 2, we have i1 (W1 ) = i2 (W2 ). Therefore, it is enough
to show that Wj is dense in Vj . Suppose that (⇡ ij |Vj ) 1 (Z) has an in-
terior point. Since the set of regular points of ⇡ ij is dense in Vj , there
exists an open subset O` ⇢ Vj such that ⇡ ij (Oj ) ⇢ Z and ⇡ ij |Oj is
an immersion. For a point qj 2 Oj , let Tj be the tangent hyperplane of
the regular hypersurface ⇡ ij (Oj ) at qj . It follows that we have a local
di↵eomorphism
j : T` ! ⇡ Lj (Oj )
around qj . Since Zj is the critical value set of ⇡ ij , j (Zj ) is the critical
value set of j ⇡ ij , j = 1, 2. By Sard’s theorem, j (Z) is a measure
zero set. However, we have j ⇡ ij (Oj ) ⇢ j (Z). This is a contradiction,
therefore (⇡ ij |Vj ) 1 (Z) does not have interior points. Since
1 1
V j = (⇡ ij|Vj ) (S) = (⇡ ij |Vj ) (R [ Z)
1 1
= (⇡ ij |Vj ) (R) [ (⇡ ij |Vj ) (Z),
we get that Wj = (⇡ ij |Vj ) 1
(R) is dense in Vj . ⇤
We remark that in the original proof in [Zakalyukin (1984)] it is assumed
that ⇡ ij |U are proper mappings for j = 1, 2. This assumption is not
needed. The idea of the above proof for removing this assumption is given
in [Kokubu, Rossman, Saji, Umehara, and Yamada (2005)].
The assumption on the density of the regular sets of ⇡ i in Theorem
5.10 is a generic condition on i. If i is Legendrian stable, then there exist a
neighbourhood W of a representative i : V ! P T ⇤ Rn of i (in the Whitney
C 1 -topology) such that for any Legendrian immersion j 2 W , there exists
a point x0 2 V such that the germ j : (V, x0 ) ! P T ⇤ Rn is Legendrian
equivalent to i. Since the above property is generic, there exists a Legen-
drian immersion j : V ! P T ⇤ Rn 2 W and x0 2 V such that the germ
112 Di↵erential Geometry from a Singularity Theory Viewpoint

j : (V, x0 ) ! P T ⇤ Rn is Legendrian equivalent to i. Therefore, i satisfies

this property.
The Legendrian equivalence can be interpreted using the generating
families.

Theorem 5.11. (i) Let F : (Rk ⇥ Rn , 0) ! (R, 0) and G : (Rl ⇥ Rn , 0) !

(R, 0) be two Morse families of hypersurfaces. Then the Legendrian map-
germs ⇡ L (F ) and ⇡ L (G) are Legendrian equivalent if and only if F
and G are stably P -K-equivalent.
(ii) The Legendrian map-germ ⇡ L (F ) is Legendrian stable if and only
if F is a K-versal deformation of f (q) = F (q, 0).

Proof. See [Arnol’d, Guseı̆n-Zade and Varchenko (1985)] and [Zaka-

lyukin (1976)]. ⇤

If k = l in Theorem 5.11 (i), we replace stably P -K-equivalence by

P -K-equivalence.
For a given map-germ f : (Rn , 0) ! (Rp , 0), we have the local ring of f
given by
En
Qr (f ) = .
f ⇤ (Mp )En + Mr+1
n

Proposition 5.2. Let F, G : (Rk ⇥ Rn , (0, 0)) ! (R, 0) be two Morse fami-
lies of hypersurfaces. Suppose that ⇡ L (F ) and ⇡ L (G) are Legendrian
stable. Then the following statements are equivalent.

(1) (WL (F ) , 0) and (WL (G) , 0) are di↵eomorphic as germs.

(2) L (F ) and L (G) are Legendrian equivalent.
(3) Qn+1 (f ) and Qn+1 (g) are isomorphic as R-algebras, where f =
F |Rk ⇥{0} and g = G|Rk ⇥{0} .

Proof. By Theorem 5.11, F and G are K-versal deformations of f and

g respectively. Since ⇡ L (F ) and ⇡ L (G) are Legendrian stable, they
satisfy the generic condition of Theorem 5.10, so statements (1) and (2) are
equivalent. If F and G are K-versal deformations of f and g respectively,
then f and g are (n + 1)-K-determined (see [Martinet (1982); Mather
(1969b)]) so statement (3) implies that f and g are K-equivalent. It follows
by the uniqueness of the K-versal deformation of a germ of a function that
F and G are P -K-equivalent. This shows that statement (3) implies (2).
Now statement (2) implies statement (3) by Theorem 5.11. ⇤
 Lagrangian and Legendrian Singularities 113

The classification of singularities of germs of functions and of their K-

versal unfoldings (Table 3.1 and Theorem 3.15) can now be used to obtain
a list the normal forms of the singularities of Legendrian map-germs in low
dimensions.

Theorem 5.12. Let F : (Rk ⇥ Rn , 0) ! (R, 0) be a Morse family of hy-

persurfaces. Suppose that L (F ) : (⌃F , 0) ! P T ⇤ Rn is Legendrian stable
and n  4. Then L (F ) is Legendrian equivalent to a Legendrian subman-
ifold germ whose generating family G(q1 , . . . , qk , x1 , . . . , xn ) is one of the
following germs, where Q(qr , . . . , qk ) = ±qr2 ± · · · ± qk2 .
(1) Q(q1 , . . . , qk ) + x1 .
(2) Q(q2 , . . . , qk ) + q13 + x1 + x2 q1 .
(3) Q(q2 , . . . , qk ) + q14 + x1 + x2 q1 + x3 q12 .
(4) Q(q2 , . . . , qk ) + q15 + x1 + x2 q1 + x3 q12 + x4 q13 .
(5) Q(q3 , . . . , qk ) + q13 q1 q22 + x1 + x2 q1 + x3 q2 + x4 (q12 + q22 ).
(6) Q(q3 , . . . , qk ) + q13 + q23 + x1 + x2 q1 + x3 q2 + x4 q1 q2 .

Proof. By the assumption and Theorem 5.11 (2), F is a K-versal defor-

mation of f = F |Rk ⇥{0} . Since n  4, code (f, K)  4. By Theorem 3.15, F
is P -K-equivalent to one of the germs in the above list. ⇤

5.3 Graph-like Legendrian submanifolds

We consider some special Legendrian submanifolds in P T ⇤ (Rn ⇥R), namely

the so-called graph-like Legendrian submanifolds. These were introduced in
[Izumiya (1993)] and have also application in geometrical optics. Recall
from Example 5.8 that the manifold of 1-jets of functions J 1 (Rn , R) can be
considered as a coordinate neighbourhood of P T ⇤ (Rn ⇥ R).
A Legendrian immersion i : L ! P T ⇤ (Rn ⇥R) is said to be a graph-like
Legendrian immersion if i(L ) ⇢ J 1 (Rn , R) ⇢ P T ⇤ (Rn ⇥ R).
The manifold J 1 (Rn , R) can be identified canonically with T ⇤ Rn ⇥ R
and we have the canonical projection
e : J 1 (Rn , R) ! T ⇤ Rn
⇡
e(x, p, y) = (x, p).
defined by ⇡
In what follows ⇡ : P T ⇤ (Rn ⇥ R) ! Rn ⇥ R and ⇢ : T ⇤ Rn ! Rn are the
bundle projections.

Proposition 5.3. Let i : L ! P T ⇤ (Rn ⇥ R) be a graph-like Legendrian

e i : L ! T ⇤ Rn is a Lagrangian immersion.
immersion. Then ⇡
114 Di↵erential Geometry from a Singularity Theory Viewpoint

Proof. Given a tangent vector v 2 Tz i(L ), there exist ` , ⌘` , ` =

1, . . . , n, and µ in R, such that
n
X Xn
@ @ @
v= ` + ⌘` +µ
@x` @p` @y
`=1 `=1

as a tangent vector of J 1 (Rn , R). Then

n
X X n
@ @
de
⇡z (v) = i ` + ⌘` .
@x` @p`
`=1 `=1
Pn
Since i(L ) is a Legendrian submanifold, µ = `=1 p` ` . If de ⇡z (v) = 0
then i = ⌘` = 0, i = `, . . . , n, which implies µ = 0. Therefore, de ⇡z (v) = 0
implies v = 0. This proves that ⇡ e i : L ! T ⇤ Rn
e|i(L ) is an immersion, so ⇡
is an immersion.
Let ↵ denote the canonical contact 1-form in P T ⇤ (Rn ⇥ R). The condi-
tion ↵|i(L ) = 0 implies ! = d↵ = 0 on ⇡ e i(L ), where ! is the canonical
symplectic form in T R . This shows that ⇡
⇤ n
e i(L ) is a Lagrangian sub-
manifold of T R .
⇤ n
⇤
We call ⇢ ⇡e i the induced Lagrangian map from the graph-like Leg-
endrian immersion i : L ! P T ⇤ (Rn ⇥ R).

Proposition 5.4. Let i : L ! P T ⇤ (Rn ⇥ R) be a graph-like Legendrian

immersion. Then the Legendrian map ⇡ i : L ! Rn ⇥ R is singular at
e i : L ! Rn is singular at
x 2 L if and only if the Lagrangian map ⇢ ⇡
x 2 L.

Proof. We consider the canonical projection ⇡1 : Rn ⇥ R ! Rn and

remark that
e(x, p, y).
⇡1 ⇡|J 1 (Rn ,R) (x, p, y) = ⇡1 (p, y) = p = ⇢(x, p) = ⇢ ⇡
Therefore, it is enough to show that ⇡|i(L ) is singular at z = (x, p, y) if
and only if ⇡1 ⇡|i(L ) is singular at z.
Suppose that ⇡|i(L ) is not singular at z 2 i(L ). Then d⇡z |Tz i(L ) is a
monomorphism. Any v 2 Tx i(L ) can be written in the form
n
X Xn
@ @ @
v= i + ⌘i +µ
@x` @pj @y
`=1 `=1

for some ` , ⌘` , µ 2 R, ` = 1, . . . , n. Then

n
X @
d(⇡1 ⇡)z (v) = i .
@x`
`=1
 Lagrangian and Legendrian Singularities 115

Suppose that d(⇡1 ⇡)z (v) = 0. Then ` = 0, ` = 1, . . . , n, and this

implies µ = 0 since L is a Legendrian submanifold (see details in the proof
of Proposition 5.3). Therefore
n
X @ @
d⇡z (v) = i +µ = 0,
@x` @y
i`=1

which implies v = 0 as we assumed ⇡|i(L ) to be non-singular at z. This

means that d(⇡1 ⇡)z |Tz i(L ) is a monomorphism, so ⇡1 ⇡|i(L ) is not sin-
gular at z.
For the converse of the assertion, suppose that rank(d(⇡|i(L ) )z ) < n.
Then we have rank(d(⇢ ⇡ e|i(L ) )z ) = rank(d(⇡1 ⇡|i(L ) )z ) < n. This com-
pletes the proof. ⇤
A Legendrian submanifold is given locally by a generating family (The-
orem 5.9). We shall show the generating family of a graph-like Legendrian
submanifold has a special form.
Let F : (Rk ⇥ (Rn ⇥ R), 0) ! (R, 0) be a Morse family of hypersurfaces
and let (q, x, y) denote the parameters in Rk ⇥ (Rn ⇥ R). We say that F is a
graph-like Morse family of hypersurfaces if @F @y (0) 6= 0. In the homogeneous
system of coordinates ((x1 , . . . , xn , y), [⇠1 : · · · : ⇠n : ⌘]) of P T ⇤ (Rn ⇥ R),
the germ L (F)(⌃F ) belongs to the affine coordinate neighbourhood
Rn ⇥ V⌘ = {((x1 , . . . , xn , y), [⇠1 : · · · : ⇠n : ⌘]) | ⌘ 6= 0} = J 1 (Rn , R).
Therefore, the germ L (F)(⌃F ) of the Legendrian submanifold determined
by the graph-like Morse family of hypersurfaces F is a graph-like Legendrian
submanifold germ. We then say that F is a graph-like generating family
of L (F)(⌃F ). In fact, any germ of a graph-like Legendrian submanifold
can be identified with L (F)(⌃F ), for some graph-like Morse family of
hypersurfaces F.
Let F be a graph-like Morse family of hypersurfaces. By the implicit
function theorem, there is a Morse family of functions F : (Rk ⇥ Rn , 0) !
(R, 0) such that
1
F (0) = {(q, x, F (q, x)) | (q, x) 2 (Rk ⇥ Rn , (0, 0)) }.
1
Since F (0) is a regular hypersurface, we have
Ek+n+1 .{F(q, x, t)} = Ek+n+1 .{F (q, x) y}.
This means that there exists a germ of function : (Rk ⇥Rn ⇥R, (0, 0, 0)) !
R) with (0, 0, 0) 6= 0 such that
(q, x, t) · F(q, x, t) = F (q, x) y.
116 Di↵erential Geometry from a Singularity Theory Viewpoint

The families F(q, x, y) and F (q, x) y define the same germ of a graph-like
Legendrian submanifold. Therefore F (q, x) y is a graph-like generating
family of L (F)(⌃F ). In this case,
⌃F = {(q, x, F (q, x)) 2 (Rk ⇥ (Rn ⇥ R), 0) | (q, x) 2 CF }
and L (F) : (⌃F , 0) ! J 1 (Rn , R) is given by
L (F)(q, x, F (q, x)) = (L(F )(q, x), F (q, x)) 2 J 1 (Rn , R) ⌘ T ⇤ Rn ⇥ R.

Define a map L(F ) : (CF , 0) ! J 1 (Rn , R) by

✓ ◆
@F @F
L(F )(q, x) = x, (q, x), . . . , (q, x), F (q, x) .
@x1 @xn
Then L(F )(CF ) = L (F)(⌃F ).
We call W (L(F )) = ⇡(L(F )(CF )) the graph-like wavefront of the germ
of the graph-like Legendrian submanifold L(F )(CF ) and call F a generating
familygenerating family of the germ of the graph-like Legendrian submani-
fold L(F )(CF ).
For a given Morse family of functions F, we set Fe (q, x, y) = F (q, x) y.
e(L (F)(q, x, F (q, x))) = L(F )(q, x), so that F is a
We have, by definition, ⇡
generating family of the germ of the Lagrangian submanifold ⇡ e(L(F )(CF )).

Proposition 5.5. Let F : (Rk ⇥ Rn , 0) ! (R, 0) and G : (Rl ⇥ Rn , 0) !

e(LF (CF )) and ⇡
(R, 0) be two Morse families of functions. If ⇡ e(LG (CG ))
are Lagrangian equivalent, then L(F )(CF ) and L(G)(CG ) are Legendrian
equivalent.

Proof. By the observations following Theorems 5.4 and 5.11, it is enough

to consider the case when k = l. By Theorem 5.4, if ⇡ e(LF (CF )) and
e(LG (CG )) are Lagrangian equivalent then F and G are P -R+ -equivalent.
⇡
Then there exist a germ of a di↵eomorphism : (Rk ⇥Rn , 0) ! (Rk ⇥Rn , 0)
of the form (q, x) = ( 1 (q, x), 2 (x)) and a germ of a function ↵ :
(Rn , 0) ! R such that F (q, x) = G(q, x) + ↵(x). We define the germ of
e
a di↵eomorphism : (R ⇥ Rn ⇥ R, 0) ! (Rk ⇥ Rn ⇥ R, 0) by
k

e (q, x, y) = ( 1 (q, x), 2 (x), y + ↵(x)).

We have
Fe e (q, x, y) = F (q, x) y ↵(x) = G(q, x) e x, y).
y = G(q,
It follows by Theorem 5.11, that LF (CF ) and LG (CG ) are Legendrian
equivalent. ⇤
 Lagrangian and Legendrian Singularities 117

We have the following result as a consequence of the classification the-

orem of Lagrangian singularities (Theorem 5.5).
Theorem 5.13. Let F : (Rk ⇥ Rn , 0) ! (R, 0) be a Morse fam-
ily of functions. Suppose that ⇡ e LF : CF ! T ⇤ Rn is Lagrangian
stable and n  4. Then ⇡ e L(F ) is Lagrangian equivalent to ⇡ e
L(G) where G(q1 , . . . , qk , x1 , . . . , xn ) is one of the following germs, where
Q(qr , . . . , qk ) = ±qr2 ± · · · ± qk2 .
(1) Q(q2 , . . . , qk ) + q13 + x1 q1 .
(2) Q(q2 , . . . , qk ) ± q14 + x1 q1 + x2 q12 .
(3) Q(q2 , . . . , qk ) + q15 + x1 q1 + x2 q12 + x3 q13 .
(4) Q(q2 , . . . , qk ) ± q16 + x1 q1 + x2 q12 + x3 q13 + x4 q14 + x5 q15 .
(5) Q(q3 , . . . , qk ) + q13 q1 q22 + x1 q1 + x2 q2 + x3 (q12 + q22 ).
(6) Q(q3 , . . . , qk ) + q13 + q23 + x1 q1 + x2 q2 + x3 q1 q2 .
(7) Q(q3 , . . . , qk ) + q12 q2 + q24 + x1 q1 + x2 q2 + x3 q12 + x4 q22 .
As a consequence, the germ of the graph-like Legendrian submanifold
LF is Legendrian equivalent to LG , where G is one of the above germs.
In the list of the above theorem, the germ of the Lagrangian map corre-
sponding to the germ (1) is the fold map-germ and the germ of the corre-
sponding graph-like wavefront is the cuspidal edge. In the case of germ (2)
the Lagrangian map-germ is the Whitney cusp and the graph-like wavefront
is the swallowtail.

5.4 Versal unfoldings and Morse families of functions

We summarise here the results in the previous sections which will be used
to describe generic properties of submanifolds in Euclidean and Minkowski
spaces. We start with the following observation.
Proposition 5.6. If F : (Rk ⇥ Rn , 0) ! (R, (0, 0)) is an R+ -versal defor-
mation of the germ f = F |Rk ⇥{0} , then F is a Morse family of functions.
Proof. By Thom’s splitting lemma (Lemma 3.1), we may assume that
f 2 M3k . By Theorem 3.12, we have
n o
LRe · f + R · 1, Ḟ1 , . . . , Ḟn = Ek ,
@F
where Ḟi (q) = @q i
(q, 0). By Theorem 3.1, f is R+ -finitely determined so
that there exists r 2 N such that Mrk ⇢ LRe · f. Then we have
n o
R · j r 1 Ḟ1 , . . . , j r 1 Ḟn = Mk /(LRe · f + Mrk ).
118 Di↵erential Geometry from a Singularity Theory Viewpoint

Since f 2 M3k , we have LRe · f + Mrk ⇢ M2k . Therefore, we have

Mk /(LRe · f + Mrk ) = (Mk /M2k ) (M2k /(LRe · f + Mrk ))

and

n code (f, R) 1 = dimR Mk /(LRe · f + Mrk ) dimR Mk /M2k = k.

On the other hand, the Jacobian matrix of F at (0, 0) is

✓ 2 ◆ ✓ ◆
@ f @2F @2F
J F (0, 0) = (0), (0, 0) = 0, (0, 0) .
@qi @qj @qi @x` @qi @x`
Here, we have
X k
1 @2F
j Ḟ` = (0, 0)qi 2 Mk /M2k .
i=1
@qi @x`
n o
It follows that j 1 Ḟ1 , . . . , j 1 Ḟn generates the R-vector space Mk /M2k .
This means that
✓ 2 ◆
@ F
rank J F (0, 0) = rank (0, 0) = k.
@qi @x`
Therefore F is a Morse family of functions. ⇤

Let F : (Rk ⇥ Rn , (0, 0)) ! (R, 0) be an R+ -versal unfolding of f.

Then we have a Lagrangian submanifold of T ⇤ Rn through the following
Lagrangian embedding
L(F ) : (CF , 0) ! T ⇤ Rn
(q, x) 7! (q, dFx (q, x))).
The bifurcation set BF of F coincides with the caustic of the Lagrangian
submanifold L(F )(CF , 0). Then the catastrophe map ⇡CF : (CF , 0) 7! Rn
can be identified with the Lagrangian map. Thus, the stable singularities
of catastrophe map under the perturbations of F are the stable Lagrangian
singularities.
Consider now the germ of the extended family of functions associated
to F ,

Fe : (Rk ⇥ Rn ⇥ R, 0) ! (R, 0)
(q, x, y) 7! F (q, x) y.

We denote by GF the set Fe 1

(0) = Graph(F ).
 Lagrangian and Legendrian Singularities 119

Consider the contact manifold J 1 (Rn , R) with its standard contact

structure and the legendre fibration ⇡ : J 1 (Rn , R) ! Rn ⇥ R. By using the
natural identification of J 1 (Rn , R) with T ⇤ Rn ⇥ R we have an immersion
L(F ) : (CF , 0) ! T ⇤ Rn ⇥ R
(q, x) 7! ((q, dq Fx ), F (q, x)),
whose image is a graph-like Legendrian submanifold of J 1 (Rn , R). More-
over, the discriminant (or level bifurcation set) DF of F is its corresponding
(graph-like) wavefront W (L(F )).
We define
⌃W (L(F )) = {(x, z) 2 WF | det Hess(fx )(q) = 0, (q, x) 2 CF },
which is the singular set of W (L(F )). Observe that we have the following
commutative diagram
i ⇡
L(F )(CF ) ⇢ T ⇤ Rn ⇥ R ! Rn ⇥ R
L(F ) " #⇡˜ # ⇡1
L(F ) ⇢
C F ! T ⇤ Rn ! Rn
where iLF is the inclusion, ⇡ and ⇡ ¯ are the canonical Legendre fibration,
0
and ⇡˜1 and ⇡1 are the natural projections onto the first factor. Also note
that ⇡CF = ⇢ L(F ). Therefore,
(a) ⇡(L(F )(CF )) = W (L(F )),
(b) ⇡1 (⌃(W (L(F ))) = BF .
Figures 5.1, 5.2 and 5.3 illustrate the above relations for the fold (A2 )
and cusp (A3 ) singularities.

5.5 Families of functions on hypersurfaces in Rn

In the subsequent section (unless stated otherwise), M denotes a hypersur-

face patch in the Euclidean space Rn parametrised by x : U ! Rn . We
defined in Chapter 2 three families of functions on M which measure its
contact with totally umbilic hypersurfaces in Rn . These are
(1) The family of height functions H : U ⇥ S n 1 ! R given by
H(u, v) = hx(u), vi.
(2) The extended family of height functions H e : U ⇥ (S n 1 ⇥ R) ! R
given by
e
H(u, v, r) = hx(u), vi r.
120 Differential Geometry from a Singularity Theory Viewpoint

q1 y
CF WF

~

.(Id x F)
x1 x1

XF
1'

BF x1

Fig. 5.1 Catastrophe set, wavefront and bifurcation set of a 1-parameter versal unfold-
ing of the fold singularity.

y
q1 WF

CF

~

.(Id x F)

x2 x2

XF
1'

x1
x1

BF
x2

x1

Fig. 5.2 Catastrophe set, wavefront and bifurcation set of a 2-parameter versal unfold-
ing of the fold singularity.

(3) The family of distance squared functions D : U × Rn → R, given by

D(u, a) = ∥x(u) − a∥2 .
It follows from Chapter 4 that the germs of the families of height and
distance squared functions associated to any immersion lying in a residual
subset of Imm(Rn−1 , Rn ) (referred to as generic immersion) can be seen as
 Lagrangian and Legendrian Singularities 121

q1 y
WF

CF
~

.(Id x F)

x2 x2

XF
1' x1
x1

BF x2

x1

Fig. 5.3 Catastrophe set, wavefront and bifurcation set of a 2-parameter versal unfold-
ing of the cusp singularity.

versal unfoldings of the germs of the corresponding function at each one of

their points. Clearly, in such case, they must be Morse families of functions.
We prove in the subsections below that the families of height functions and
of distance squared functions are Morse families of functions at each point
in their domain, regardless of the embeddings that induce them be generic
or not. They are thus generating families of some Lagrangian manifolds (at
least locally). We show that their associated caustics are geometric objects
related to the hypersurface M . In the case of generic embeddings, these
Lagrangian manifolds correspond to stable Lagrangian maps and the local
behaviour of their caustics is well known. We also prove that the extended
family of height functions is a Morse family of hypersurfaces at each point
in their domain. It is therefore a generating family of a Legendrian manifold
(at least locally), and the associated wavefront is also a geometric object
related to the hypersurface M .
122 Di↵erential Geometry from a Singularity Theory Viewpoint

5.5.1 The family of height functions

Proposition 5.7. The germ of the family of height functions H on M at

each point (u, v) 2 U ⇥ S n 1 is a Morse family of functions.

Proof. ) 2 Sn
Let v = (v1 , . . . , vnq 1
and assume, without loss of general-
ity, that vn > 0. Then vn = 1 v12 ··· vn2 1 and
q
H(u, v) = x1 (u)v1 + . . . + xn 1 (u)vn 1 + xn (u) 1 v12 ··· vn2 1.

We need to prove that the mapping

✓ ◆
@H @H
H(u, v) = ,..., (u, v)
@u1 @un 1
is not singular at any point (u, v) 2 CH . Its Jacobian matrix is given by
0 1
hxu1 u1 (u), vi · · · hxu1 un 1 (u), vi
B .. .. .. C
@ . . . AA,
hxun 1 u1 )
(u), vi · · · hxun 1 un 1 (u), vi

with
0 v1 vn 1 1
(x1 )u1 (u) (xn )u1 (u) ... (xn 1 )u1 (u) (xn )u1 (u)
B vn vnC
B .. .. .. C
A=B . . . C
@ v1 vn 1 A
(x1 )un 1 (u) (xn )un 1 (u) . . . (xn 1 ) un 1
(u) (xn )un 1 (u)
vn vn
We show that the rank of the matrix A is n 1 at (u, v) 2 C(H). We
set
0 1
(xi )u1 (u)
B .. C
ai = @ . A for i = 0, . . . , n
(xi )un 1 (u)

and rewrite the matrix A in the form

✓ ◆
v1 vn 1
A = a1 an , . . . , a n 1 an .
vn vn
By Proposition 2.4, we have

CH = {(u, ±N (u)) |u 2 U }.
 Lagrangian and Legendrian Singularities 123

Therefore,
v1 vn
detA = ( 1)n+1 det(a2 , . . . , an ) + · · · + ( 1)2n det(a1 , . . . , an 1)
vn vn
⌧✓ ◆
v1 vn
= ( 1)n 1
,..., , x u1 ⇥ · · · ⇥ x un 1
vn vn
( 1)n 1
= h±N, xu1 ⇥ · · · ⇥ xun 1 i
vn
( 1)n 1
=± kxu1 ⇥ · · · ⇥ xun 1 k
vn
and this is not zero for any (u, v) 2 CH . This proves that H is a Morse
family of functions. ⇤

As a consequence of Proposition 5.7, we can define a germ of a

Lagrangian immersion whose generating family is the family of height func-
tions on the hypersurface patch M (Theorem 5.3).
We consider the standard coordinate neighbourhoods

Ui+ = {v = (v1 , . . . , vn ) 2 S n 1
| vi > 0 }, i = 1, . . . , n

and

Ui = {v = (v1 , . . . , vn ) 2 S n 1
| vi < 0 }, i = 1, . . . , n

of the (n 1)-sphere S n 1
. Since T ⇤ S n 1
|Ui± is a trivial bundle, we define
the maps

L± ⇤ n
i (H) : CH ! T S
1
|Ui± , i = 1, . . . , n

by

L±
i✓(H)(u, ±N (u)) ◆
= ±N (u), x1 (u) xi (u) vv1i , . . . , xi (u)\xi (u) vvii , . . . , xn (u) xi (u) vvni ,

where (x1 , . . . , x̂i , . . . , xn ) denotes the point in Rn 1

obtained by removing
the i-th component of (x1 , . . . , xi , . . . , xn ) in Rn .

We have the following consequence of Proposition 5.7.

Corollary 5.1. The maps L± i (H), i = 1, . . . , n, are Lagrangian immersions

and the family of height functions H is their generating family at each point
of U ⇥ Ui± .
124 Di↵erential Geometry from a Singularity Theory Viewpoint

q P
Proof. Since we have vi = ± 1 j6=i vj2 on Ui± ,
@H vj
(u, v) = xj (u) xi (u) , (j 6= i),
@vj vi
at (u, v) 2 U ⇥ Ui± . By the construction of the germ of the Lagrangian im-
mersion from the generating family in Section 5.1.2, L±
i (H) is a Lagrangian
immersion. ⇤
Remark 5.2. 1. The subset CH = {(u, ±N (u)) |u 2 U } is a double cover
of the hypersurface M . It is not difficult to see that CH is the catastrophe
set of H and the corresponding catastrophe map can be identified with
the normal Gauss map (up to a sign) on the hypersurface M . Thus, for
a generic embedding x, the Gauss map of M = x(U ) behaves as a stable
Lagrangian map.
2. We can apply the same construction to a submanifold M of codi-
mension higher than one. In this case, the catastrophe set CH is the unit
normal bundle and it can be identified with a canal hypersurface around
M in Rn . The catastrophe map coincides with the generalised Gauss map
on CH .

5.5.2 The extended family of height functions

e on M is a
Proposition 5.8. The extended family of height functions H
graph-like Morse family of hypersurfaces.
Proof. Since the height function H is a Morse family of fucntions at each
e
point, the extended height function H(u, v, r) = H(u, v) r is a graph-like
Morse family of hypersurface at each point in U ⇥ (S n 1 ⇥ R) (see §5.3 and
the discussion before Proposition 5.5). ⇤
We can define now a graph-like Legendrian immersion whose generating
family at each point in the domain is the germ of the extended family of
e
height functions H.
We consider the contact manifold Rn ⇥ S n 1 whose contact structure is
given by the 1-form ✓ = hv, dxi|{kvk=1} , where (x, v) 2 Rn ⇥ S n 1 . Since
the tangent bundle T Rn is a trivial bundle, the above contact manifold
is identified canonically with the unit tangent bundle T1 Rn . The contact
structure on T1 Rn is given by the above 1-form ([Blair (1976)]).
We consider the projection : Rn ⇥ S n 1 ! S n 1 ⇥ R definded by
(x, v) = (v, hx, vi).
 Lagrangian and Legendrian Singularities 125

Since hv, dxi|{kvk=1} = hdv, xi|{kvk=1} , ✓ = 0 on the set v = constant,

so ⇡ is a Legendrian fibration. We now define the map : Rn ⇥ S n 1 !
T ⇤ S n 1 ⇥ R by
(v, x) = (v, dv hx, vi, hx, vi),
where dv hx, vi is defined as follows: For any v 2 S n 1 , we may consider
that Tv S n 1 ⇢ Tv Rn . Define a linear mapping dv hx, vi : Tv S n 1 ! R by
Xn Xn
@hx, vi
dv hx, vi(⌘) = ⌘i = xi ⌘i
i=1
@v i i=1
where x = (x1 , . . . , xn ), v = (v1 , . . . , vn ) and ⌘ = (⌘1 , . . . , ⌘n ) in the basis
{@/@v1 , . . . , @/@vn } of Tv Rn .
We have the projection ⇡ e : T ⇤ S n 1 ! S n 1 of the cotangent bundle.
By the canonical identification T ⇤ S n 1 ⇥ R ⌘ J 1 (S n 1 , R), T ⇤ S n 1 ⇥ R is
a contact manifold whose contact form is dy , where is the Liouville
⇤ n 1
form of T S . We consider the standard coordinate neighbourhoods
Ui+ = {v = (v1 , . . . , vn ) 2 S n 1
| vi > 0 }, i = 1, . . . , n
and
Ui = {v = (v1 , . . . , vn ) 2 S n 1
| vi < 0 }, i = 1, . . . , n
of the (n 1)-sphere S n 1
.

Proposition 5.9. The map |Rn ⇥U ± : Rn ⇥ Ui± ! T ⇤ S n 1

|U ± ⇥ R is a
i i
Legendrian di↵eomorphism for i = 1, . . . , n.

Proof. We prove the proposition on the local coordinate neighbourhood

U1+ = {v = (v1 , . . . , vn ) 2 S n 1
| v1 > 0}.
On U1+ , we have
n
X
dv hx, vi = hx, dvi = xi dvi ||{kvk=1} .
i=1
q Pn Pn
Since v1 = 1 j=2 vj2 , we have dv1 = i=2 (vi /v1 )dvi , so that
Xn ✓ ◆
vi
hx, dvi|U + = xi x1 dvj .
1
j=2
v1
Therefore, the local representation of is given as
0 1
Xn ✓ ◆
vj
(x, v) = @v, xj x1 dvj , hx, viA .
j=2
v1
126 Di↵erential Geometry from a Singularity Theory Viewpoint

It follows that
n
X n ✓
X ◆
⇤ vj
(dy pj dvj ) = dhx, vi|U + xj x1 dvj
j=2
1
j=1
v1
= hdx, vi|U + + hx, dvi|U + hx, dvi|U +
1 1 1

= hdx, vi|U + = ✓|Rn ⇥U + .

1 1

Let : (T ⇤ S n 1
|U + ) ⇥ R ! Rn ⇥ U1+ be defined by
1

n
X n
X
(v, p, y) = (v1 (y pj vj ), . . . , vn (y pj vj ), v).
j=2 j=2

Then we have |Rn ⇥U + = 1Rn ⇥U + and |Rn ⇥U + = 1(T ⇤ S n 1|

+ )⇥R
.
1 1 1 U1

We proceed similarly for the other coordinate neighbourhoods. ⇤

We define a map LM : U ! Rn ⇥ S n 1 by LM (u) = (x(u), N (u)).
Then we have x⇤ (u) = (LM (u)).
Let Ui± , i = 1, . . . , n, be the standard coordinate neighbourhoods of
S n 1 . Since P T ⇤ (S n 1 ⇥ R)|(U ± ⇥R) is a trivial bundle, we define for i =
i
1, . . . , n, the map

LH,i : CH |U ⇥U ± ! T ⇤ S n 1
⇥ R|U ± ⇥R ⌘ J 1 (Ui± , R)
i i

by LH,i (u, N (u)) = (N (u), ⇠(u), hN (u), x(u)i), where

✓ ◆
v1 \ vi vn
⇠(u) = x1 (u) xi (u) : · · · : xi (u) xi (u) : · · · : xn (u) xi (u)
vi vi vi
and N (u) = (v1 , . . . .vn ). It follows from Corollary 5.1 that LH,i is a graph-
like Legendrian immersion. We have

LM (u) = LH,i (u, N (u)) for u 2 {u 2 U |N (u) 2 Ui± }.

We remark that J 1 (S n 1 , R) = [ni=1 J 1 (Ui± , R). Since is an immersion,

LM : U ! J 1 (S n 1 , R) is a graph-like Legendrian immersion.

Theorem 5.14. Let M be a hypersurface patch in Rn . The map LM

is a graph-like Legendrian immersion and the germ of the extended family
e is its graph-like generating family at each point. The
of height functions H
graph-like wavefront W (LM ) of LM is the cylindrical pedal M ⇤ = x⇤ (U )
of M. Moreover, the Gauss map N : M ! S n 1 is the corresponding
Lagrangian map. (As a consequence, the cylindrical pedal of a hypersurface
in Rn has Legendrian singularities.)
 Lagrangian and Legendrian Singularities 127

5.5.3 The family of distance squared functions

Proposition 5.10. The germ defined by the family of the distance squared
functions D on M at each point (u0 , a0 ) 2 U ⇥ Rn is a Morse family of
functions.
Proof. We have
n
X
D(u, a) = (xi (u) ai ) 2 ,
i=1
where x(u) = (x1 (u), . . . , xn (u)), u = (u1 , . . . , un 1 ) 2 U and a =
(a1 . . . , an ) 2 Rn . We shall prove that the mapping
✓ ◆
@D @D
D(u, a) = ,..., (u, a)
@u1 @un 1
is not singular at any point (u, a). Its Jacobian matrix is given by
0 1
A11 · · · A1(n 1) 2(x1 )u1 (u) · · · 2(xn )u1 (u)
B .. .. .. .. .. .. C
@ . . . . . . A,
A(n 1)1 · · · A(n 1)(n 1) 2(x1 )un 1 (u) · · · 2(xn )un 1 (u)
where Aij = 2(hxui uj (u), x(u) ai + hxui (u), xuj (u)i). Since x : U ! Rn
is an embedding, the rank of the matrix
0 1
2(x1 )u1 (u) · · · 2(xn )u1 (u)
B .. .. .. C
x=@ . . . A
2(x1 )un 1 (u) · · · 2(xn )un 1 (u)
is n 1 at each point u0 in U. Therefore, the rank of the Jacobian matrix
of D is n 1, and this proves that the germ of D at (u0 , a0 ) is a Morse
family of functions. ⇤
As a consequence of Proposition 5.10, we can define a germ of a La-
grangian immersion whose generating family is the family distance squared
functions on the hypersurface patch M (Theorem 5.3).
Consider the set,
CD = {(u, x(u) + N (u)) | 2 R, u 2 U }
and define the smooth mapping L(D) : CD ! T ⇤ Rn by
L(D)(u, x(u) + N (u)) = (x(u) + N (u), 2 N (u)) .
Corollary 5.2. The map L(D) is a Lagrangian immersion and the family
of distance squared functions D is its generating family at each point in
U ⇥ Rn . The caustic of the Lagrangian map ⇡ L(D) is precisely the focal
set (or evolute) of M . (As a consequence, the focal set of a hypersurface in
Rn has Lagrangian singularities.)
128 Di↵erential Geometry from a Singularity Theory Viewpoint

Proof. By definition,
@D
(u, a) = 2(xi (u) ai ),
@ai
so that
✓ ◆
@D @D
(u, x(u) + N (u)), . . . , (u, x(u) + N (u)) = 2 N (u).
@a1 @an
By the construction of the Lagrangian immersion from the generating fam-
ily in §5.1, we have a Lagrangian immersion
L(D)(u, x(u) + N (u)) = (x(u) + N (u), 2 N (u)) .
Therefore the Lagrangian map is given by
⇡ L(D)(u, x(u) + N (u)) = x(u) + N (u). ⇤

5.6 Contact from the viewpoint of Lagrangian and Legen-

drian singularities

We considered in Chapter 4 the theory of contact between two subman-

ifolds and showed that the height (resp. distance squared) functions on
a hypersurface in the Euclidean space measures the contact between the
hypersurface and hyperplanes (resp. hyperspheres). In this section we ap-
ply the Lagrangian and Legendrian singularity theory to the study of this
contact.

5.6.1 Contact of hypersurfaces with hyperplanes

First we consider the contact of hypersurfaces with hyperplanes. Let x⇤i :
(U, ui ) ! (S n 1 ⇥ R, (vi , ri )), i = 1, 2, be two germs of cylindrical pedal
hypersurface patches Mi parametrised by xi : (U, ui ) ! (Rn , pi ).
Suppose that the regular set of Mi⇤ = x⇤i (U ),i = 1, 2, is dense in
some neighbourhood Vi of ui in U . Then, by Theorem 5.10, the germs
(M1⇤ , (v1 , r1 )) and (M2⇤ , (v2 , r2 )) are di↵eomorphic if and only if the corre-
sponding Legendrian immersion germs LM1 : (U, u1 ) ! (Rn ⇥S n 1 (p1 , v1 ))
and LM2 : (U, u2 ) ! (Rn ⇥ S n 1 , (p2 , v2 )) are Legendrian equivalent.
Since the germ of the extended family of height functions H e i on Mi ,
i = 1, 2, is a generating family of the germ of the graph-like Legen-
drian immersion LMi , it follows by Theorem 5.11 that the condition
that (LM1 (U ), (p1 , v1 )) and (LM2 (U ), (p2 , v2 )) are Legendrian equivalent is
equivalent to the two generating families H e 1 and H
e 2 being P -K-equivalent.
 Lagrangian and Legendrian Singularities 129

We have the (universal) family of height functions H : Rn ⇥ S n 1 ! R

given by H(y, v) = hy, vi = hv (y), and an affine hyperplane H(v, r) =
hv 1 (r).
We consider the case when vi = Ni (ui ) and ri = hxi (ui ), Ni (ui )i. We
write e e i (u, vi , ri ) = hv xi (u) ri .
hi,(vi ,ri ) (u) = H i

By Proposition 2.4, @(hvi xi )/@uj (ui ) = 0 for j = 1, . . . , n 1 and vi =

Ni (ui ). This means that Mi = xi (U ) is tangent to the affine hypersurface
H(vi , ri ) at pi = xi (ui ). We call H(vi , ri ) the tangent affine hyperplane of
Mi at pi and denote it by T (Mi )p .
By Theorem 4.1,
K(M1 , T (M1 )p1 , p1 ) = K(M2 , T (M2 )p2 , p2 )
if and only if e
h1,(v1 ,r1 ) and e
h1,(v2 ,r2 ) are K-equivalent. Therefore, we can
apply the arguments in Section 5.2.2.
We denote by Qr (x(U ), u0 ) the local ring of the germ e hv0 ,r0 : (U, u0 ) !
R, where (v0 , r0 ) = CPeM (u0 ). We have
Cu10 (U )
Qr (x(U ), u0 ) = ,
hhx(u), N (u0 )i r0 iCu1 (U ) + Mr+1
u0
0

where r0 = hx(u0 ), N (u0 )i and Cu10 (U )

is the local ring of function germs
at u0 with the unique maximal ideal Mu0 .

Theorem 5.15. Let xi : (U, ui ) ! (Rn , pi ), i = 1, 2, be local parametrisa-

tions of hypersurfaces germs Mi such that the germs of Legendrian immer-
sions
LMi : (U, ui ) ! (Rn ⇥ S n 1
, (pi , vi )), i = 1, 2,
are Legendrian stable. Then the following statements are equivalent:
(1) The germs of the cylindrical pedals (M1⇤ , (v1 , r1 )) and (M2⇤ , (v2 , r2 ))
are di↵eomorphic.
(2) The germs of the Legendrian immersions LM1 and LM2 are Legendrian
equivalent.
(3) He 1 and H e 2 are P -K-equivalent.
(4) e
h1,(v1 ,r1 ) and eh1,(v2 ,r2 ) are K-equivalent, where (vi , ri ) = CPeMi (ui ).
(5) K(M1 , T (M1 )p1 , p1 ) = K(M2 , T (M2 )p2 , p2 ).
(6) Qn+2 (x1 , u1 ) and Qn+2 (x2 , u2 ) are isomorphic as R-algebras.

Proof. Statements (4) and (5) are equivalent by Theorem 4.1. The
equivalence of the other statements follow from Proposition 5.2 and
Theorem 5.11. ⇤
130 Di↵erential Geometry from a Singularity Theory Viewpoint

Proposition 5.11. Let xi : (U, ui ) ! (Rn , pi ), i = 1, 2, be parametrisa-

tions of of hypersurfaces germs Mi such that their parabolic sets have no
interior points in U. If the germs of the cylindrical pedals (M1⇤ , (v1 , r1 )) and
(M2⇤ , (v2 , r2 )) are di↵eomorphic, then
K(M1 , T (M1 )p1 , p1 ) = K(M2 , T (M2 )p2 , p2 ).
In this case, (x1 1 (T (M1 )p1 ), u1 ) and (x2 1 (T (M2 )p2 ), u2 ) are di↵eomorphic.

Proof. By Proposition 5.4 and Theorem 5.14, the singular set of the
cylindrical pedal coincides with the singular set of the Gauss map, which
is the parabolic set. Thus, the corresponding Legendrian lifts LMi sat-
isfy the hypothesis of Theorem 5.10. If the germs of the cylindrical
pedals (M1⇤ , (v1 , r1 )) and M2⇤ , (v2 , r2 )) are di↵eomorphic, then LM1 and
LM2 are Legendrian equivalent, so H e1, H
e 2 are P -K-equivalent. There-
fore, e
h1,(v1 ,r1 ) , e
h1,(v2 ,r2 ) are K-equivalent, where ri = hxi (ui ), Ni (ui )i.
By Theorem 4.1, this condition is equivalent to K(M1 , T (M1 )p1 , p1 ) =
K(M2 , T (M2 )p2 , p2 ).
On the other hand, we have
i(x 1 (T (Mi )p ), ui ) = (e
i
h 1 (0), ui ).
i,(vi ,ri )
Since the K-equivalence preserves the zero level sets, the germs
(x1 1 (T (M1 )p1 ), u1 ), (x2 1 (T (M2 )p2 ), u2 ) are di↵eomorphic. ⇤
We call (x 1 (T (M )p0 ), u0 ) the germ of tangent indicatrix of M at u0
(or, p0 = x(u0 )), which is denoted by T I(M, p0 ). By Proposition 5.11, the
di↵eomorphism type of the germ of the tangent indicatrix is an invariant
of the di↵eomorphism type of the germ of the cylindrical pedal CPeM (U )
of M.
We can make use of some basic invariants of germs of functions. The
local ring of a germ of a function is a complete K-invariant for generic
germs, but it is not a numerical invariant. The Ke -codimension of the
germ is a numerical K-invariant ([Martinet (1982)]). We denote that
T-ord(x(U ), u0 ) = code (e
h(v0 ,r0 ) , K). By definition,
Cu10 (U )
T-ord(x(U ), u0 ) = dim ,
hhx(u), N (u0 )i r0 , hxui (u), N (u0 )iiCu10
where r0 = hx(u0 ), N (u0 )i. Usually T-ord(x(U ), u0 ) is called the Ke -
codimension of e
h(v0 ,r0 ) . We call it the order of contact of M with its tangent
hyperplane at p0 = x(u0 ).
We denote the corank of x at u0 by
T-corank(x(U ), u0 ) = (n 1) rank H(hv0 )(u0 ),
 Lagrangian and Legendrian Singularities 131

where v0 = N (u0 ).
By Proposition 2.5, p0 is a parabolic point if and only if
T-corank(x(U ), u0 ) 1. Moreover p0 is a flat point if and only if
T-corank(x(U ), u0 ) = n 1.
By Thom’s splitting lemma (Lemma 3.1), if T-corank(x(U ), u0 ) = 1,
then generically the height function hv0 has the Ak -type singularity at u0 .
In this case we have T-ord(x(U ), u0 ) = k. For curves in the plane (i.e.,
n = 2), this number is equal to the order of contact of the curve with the
tangent line in the classical sense (see §4.1 in Chapter 4 and [Bruce and
Giblin (1992)]). This is the reason why we call T-ord(x(U ), u0 ) the order
of contact of M with its tangent hyperplane at p0 .
We now consider the contact of hypersurfaces with families of hyper-
planes. Let xi : (U, ui ) ! (Rn , pi ), i = 1, 2, be germs of parametrisations of
hypersurfaces Mi . We consider height functions Hi : (U ⇥ S n 1 , (ui , vi )) !
R on Mi , where vi = N (ui ). We denote that hi,vi (u) = Hi (u, vi ), then we
have hi,vi (u) = hvi xi (u). We also consider that the germ of the foliation
Fhvi defined in §4.1 for each i = 1, 2. We call Fhvi a tangent family of affine
hyperplanes of Mi at pi wich is denoted by T FH(Mi , pi ).

Theorem 5.16. Let xi : (U, ui ) ! (Rn .pi ), i = 1, 2, be germs of hypersur-

faces such that the corresponding germs of Lagrangian immersions
L(Hi ) : (CHi , (ui , vi )) ! T ⇤ S n 1

are Lagrangian stable, where vi = N (ui ). Then the following statements

are equivalent:
(1) K(M1 , T FH(M1 , p1 ); p1 ) = K(M2 , T FH(M2 , p2 ); p2 ).
(2) h1,v1 and h2,v2 are R+ - equivalent.
(3) H1 and H2 are P -R+ - equivalent.
(4) L(H1 ) and L(H2 ) are Lagrangian equivalent.
(5) The ranks and the signatures of H(h1,v1 )(u1 ) and H(h2,v2 )(u2 ) are
equal and there is an isomorphism : R2 (h1,v1 ) ! R2 (h2,v2 ) such
that (h1,v1 ) = h2,v2 .

Proof. By Proposition 4.1, statement (1) is equivalent to statement (2).

Statements (2), (3), (4), (5) are equivalent by Theorem 5.6. ⇤
We remark that if L(H1 ) and L(H2 ) are Lagrangian equivalent, then
the corresponding Lagrangian map-germs ⇡ L(H1 ) and ⇡ L(H1 ) are A-
equivalent. The Gauss map of a hypersurface M = x(U ) is considered to be
the Lagrangian map-germ of L(H) (or, the catastrophe map-germ of H1 ).
132 Di↵erential Geometry from a Singularity Theory Viewpoint

Moreover, if h1.v1 and h2,v2 are R+ -equivalent then the level set germs
of function germs h1.v1 and h2,v2 are di↵eomorphic. For a hypersurface
germ x : (U, u0 ) ! (Rn .p0 ), the set germ (x 1 (Fhv0 ), u0 ) is a singular
foliation germ at u0 2 U, where v0 = N (p0 ). We call it the tangential
Dupin foliation of M = x(U) at u0 , which is denoted by DF(x(U ), u0 ).
We have the following corollary.

Corollary 5.3. If one of the statements in Theorem 5.16 is satisfied, then

(1) The Gauss map-germs G1 , G2 are A-equivalent.
(2) The germs of the Dupin foliations DF(x1 (U ), u1 ), DF(x2 (U ), u2 ) are
di↵eomorphic.
(3) The germs of Legendrian immersions LM1 and LM2 are Legednrian
equivalent. This is equivalent to the germs of the cylindrical pedals
(M1⇤ , (v1 , r1 )) and (M2⇤ , (v2 , r2 )) being di↵eomorphic.

5.6.2 Contact of hypersurfaces with hyperspheres

Let pi = xi (ui ), i = 1, 2,. We consider the following point on the evolute
of Mi :
1
ai = Evj (ui ) (ui ) = x(ui ) + N (ui ),
j (ui )
where j (ui ) is one of the principal curvatures of Mi at pi . Let Di : (U ⇥
Rn , (ui , ai )) ! R be the germ of the family of distance squared functions
and denote by di,a its restriction to a fixed a, so di,a (u) = Di (u, a).
We have the (universal) family of distance squared functions D : Rn ⇥
Rn ! R by D(y, a) = ky ak2 = da (y), so that di,ai (u) = dai xi (u).
By Proposition 2.6, we have @(dai xi )/@u` (ui ) = 0 for ` = 1, . . . , n 1.
This means that the hypersphere S n 1 (ai , 1/|j (ui )|) = dai1 (1/2j (ui )) is
tangent to Mi at pi = xi (ui ). We have the foliation Fdxi (see §4.1) given
by the family of hyperspheres with centres ai which contains the osculat-
ing hypersphere S n 1 (ai , 1/|j (ui )|). We call this foliation the osculating
family of hyperspheres of Mi at pi with respect to (ai , j (ui )). We write
Fdxi = FS(Mi , (ai , j (ui )), pi ).

Theorem 5.17. Let xi : (U, ui ) ! (Rn , pi ), i = 1, 2, be parametrisations

of two germs of hypersurfaces Mi such that the corresponding germs of
Lagrangian immersions L(Di ) : (CDi , (ui , ai )) ! T ⇤ Rn are Lagrangian
stable, where ai are centres of the osculating hyperspheres of Mi . Then the
following statements are equivalent.
 Lagrangian and Legendrian Singularities 133

(1) K(M1 , FS(M1 , (a1 , j (u1 )), p1 ); p1 ) =

K(M2 , FS(M2 , (a2 , j (u2 )), p2 ); p2 ).
(2) d1,x1 and d2,x2 are R+ -equivalent.
(3) D1 and D2 are P -R+ -equivalent.
(4) L(D1 ) and L(D2 ) are Lagrangian equivalent.
(5) The rank and signature of the H(d1,x1 )(u1 ) and H(d2,x2 )(u2 ) are equal,
and there is an isomorphism : R(2) (d1,x1 ) ! R(2) (d2,x2 ) such that
(d1,x1 ) = d2,x2 .

Here H(di,xi )(ui ) denotes the Hessian matrix of di,xi at ui .

Proof. Statements (1) and (2) are equivalent by Proposition 4.1. and
statements (2)–(5) are equivalent by Theorem 5.6. ⇤

We remark that if L(D1 ) and L(D2 ) are Lagrangian equivalent, then

the corresponding evolutes are di↵eomorphic. Since the evolute of a hyper-
surface M is the caustic of L(D), Theorem 5.17 provides an interpretation
for the contact of hypersurfaces with the family of hyperspheres from the
view point of Lagrangian singularities.
For a germ of a parametrisation x : (U, u0 ) ! (Rn .p0 ) of M , the set-
germ (x 1 (Fda0 ), u0 ) is a singular foliation-germ at u0 2 U, where a0 =
Evj (u0 ) (u0 ). We call it a spherical Dupin foliation of M = x(U ) at u0 , and
denote it by S-DF(x(U ), u0 ). We consider here the germ of the osculating
hyperspherical foliation.

Corollary 5.4. Under the hypotheses of Theorem 5.17 and if one of the
statement there is satisfied, then

(1) The germs of the images of evolutes (EvM1 (U ), a1 ) and (EvM2 (U ), a2 )

are di↵eomorphic.
(2) The germs of spherical Dupin foliations S-DF(x1 (U ), u1 ) and
S-DF(x2 (U ), u2 ) are di↵eomorphic.

Proof. Statement (1) follows from the fact that Lagrangian equivalences
among germs of Lagrangian immersions preserves their caustics. By def-
inition, the germ of the spherical Dupin foliation S-DF(xi (U ), ui ) is the
level-set foliation Fdi,xi of di,xi . Statement (2) follows from the fact that
R+ -equivalence sends the level sets of one germ to another. ⇤
134 Di↵erential Geometry from a Singularity Theory Viewpoint

5.6.3 Contact of submanifolds with hyperplanes

Let M be a patch of a submanifold of Rn of dimension s parametrised by
x : U ! Rn . We set k = n s (the codimension of M ).
We defined in §2.2.3 of Chapter 2 the canal hypersurface CM (") of M .
We choose an orthonormal frame {⌫1 (u), . . . , ⌫k (u)} of Np M at p = x(u).
Let S k 1 be the unit sphere in Rk and denote the coordinate of its points by
µ = (µ1 , . . . , µk ). Then , for " sufficiently small, the map y : U ⇥S k 1 ! Rn
given by
y(u, µ) = x(u) + "N (u, µ).
Pr
with N (u, µ) = i=1 µi ⌫i (u) 2 Np M is a parametrisation of CM (") (see
the proof of Theorem 2.7).
We consider the family of height functions H : U ⇥ S n 1 ! R on M
given, as usual, by H(u, v) = hx(u), vi. We also have the extended family of
height functions He : U ⇥ S n 1 ⇥ R ! R defined by H(u,
e v, r) = H(u, v) r.
We denote that hv (u) = H(u, v) and e e
hv,r (u) = H(u, v, r) for any
(v, r) 2 S n 1
⇥ R.
Proposition 5.12. With notation as above
(i) @hv0 /@ui (u) = 0, i = 1, . . . , s, if and only if v0 = N (u, µ) for some
µ 2 Sk 1.
(ii) e
hv0 (u) = @ e
hv0 /@ui (u) = 0, i = 1, . . . , s, if and only if
v0 = N (u, µ) and r0 = hx(u), v0 i.
Proof. (i) There exist real numbers i , i = 1, . . . , s and µj , j = 1, . . . k,
such that
Xs k
X
v0 = i xui (u) + µj ⌫j (u).
i=1 j=1
Now @hv0 /@ul (u) = hxul , v0 i = 0 if and only if
X s

i hxul , xui i = 0.
i=1
The matrix (hxul , xui i) is that of the first fundamental form of M , so
is not singular. Therefore, i = 0, i = 1, . . . s.
(ii) This follows from the fact that @ e
hv0 /@ui (u) = @hv0 /@ul (u). ⇤
Definition 5.7. Let M as above and let X ⇤ : U ⇥ S k 1
! Sn 1
⇥ R be
defined by
X ⇤ (u, µ) = (N (u, µ), hx(u), N (u, µ)i).
We call the map X ⇤ : the canal cylindrical pedal of M.
 Lagrangian and Legendrian Singularities 135

The canal cylindrical pedal X ⇤ : depends on the choice of the orthonor-

mal frame {⌫1 (u), . . . , ⌫k (u)} of Np (M ), p = x(u). However, its image does
not.

Proposition 5.13. The image of the canal cylindrical pedal of M is in-

dependent of the choice of the orthonormal frame in N M . We call it the
canal cylindrical pedal hypersurface of M .

Proof. Let {⌫ 1 (u), . . . , ⌫ k (u)} be another orthonormal frame of Np M.

Pk
Then ⌫i = j=1 ij ⌫ j , with ij = h⌫i (u), ⌫ j (u)i.
The map : U ⇥ S k 1 ! U ⇥ S k 1 given by
k
X k
X
1 k
(u, µ) = (u, ( j (u)µj , . . . , j (u)µj )),
j=1 j=1
Pk
is a di↵eomorphism. If we set N (u, µ) = i=1 µi ⌫ i (u), then N (u, µ) =
N (u, µ). Therefore,
⇤
X ⇤ (u, µ) = X (u, µ),
⇤ ⇤
where X (u, µ) = x(u) + N (u, µ). This means that the image of X is the
same as that of X ⇤ . ⇤
It follows from Proposition 5.12, the canal cylindrical pedal hypersurface
is the discriminant set of the extended family of height functions of M.
The following result follows by similar arguments to those in the proof of
Proposition 5.7.

Proposition 5.14. The germ of the family of height functions H on the

submanifold M is a Morse family of functions at each point in the domain.
Therefore, the germ family of the family of extended height functions He is
a germ of the graph-like Morse family of hypersurfaces at each point in the
domain.

Proposition 5.15. The canal cylindrical pedal hypersurface is a graph-like

wavefront.

Proof. Let y : U ⇥ S k 1 ! Rn be the parametrisation of the canal hy-

persurface CM ("). We define a mapping " : J 1 (S n 1 , R) ! J 1 (S n 1 , R)
by " (x, p, y) = (x, p, y "). Since ⇤" (dy ) = d(y ") = dy , "
is a Legendrian di↵eomorphism.
We have the Legendrian immersion LCM (") : U ⇥ S k 1 ! Rn ⇥ S n 1
corresponding to the canal hypersurface.
136 Di↵erential Geometry from a Singularity Theory Viewpoint

By Theorem 5.14, LCM (") : U ⇥ S k 1 ! J 1 (S n 1 , R) is a graph-

like Legendrian immersion and the extended family of height function of
CM (") is its graph-like generating family at each point in U ⇥ S k 1 . Here,
: Rn ⇥ S n 1 ! J 1 (S n 1 , R) is a local Lagrangian di↵eomorphism defined
in the proof of Propositon 5.8. It follows that " LCM (") : U ⇥
S k 1 1
! J (S n 1
, R) is the graph-like Legendrian immersion. The graph-
like wavefront of LCM (") (U ⇥ S k 1 ) is the cylindrial pedal hypersurface
CM (")⇤ = y⇤ (U ⇥S k 1 ) of the canal hypersurface CM ("). The Legendrian
di↵eomorphism " induces a di↵eomorphism b " : S n 1 ⇥ R ! S n 1 ⇥ R
defined by b " (x, y) = (x, y "). Then we have b " y⇤ = X ⇤ . ⇤
Following the notation in §5.6.1, we denote by T (Mi ; vi )pi the tangent
affine hyperplane of Mi at pi with respect to vi = N (ui , µi ).

Theorem 5.18. Suppose that the germs of Legendrian immersions

LCMi (") : (U ⇥ S k 1
, (ui , µi )) ! (Rn ⇥ S n 1
, (pi , vi )), i = 1, 2,
are Legendrian stable. Then the following statements are equivalent:
(1) The germs of the canal cylindrical pedals
(X1⇤ (U ⇥ S k 1
), (v1 , r1 )) and (X2⇤ (U ⇥ S k 1
), (v2 , r2 ))
are di↵eomorphic.
(2) LCM1 " and LCM2 " are Legendrian equivalent.
(3) e 1 and H
H e 2 are P -K-equivalent.
(4) e
h(v1 ,r1 ) and eh(v2 ,r2 ) are K-equivalent.
(5) K(M1 , T (M1 ; v1 )p1 , p1 ) = K(x2 (U ), T (M2 ; v2 )p2 , p2 ).
(6) The germs of cylindrical pedals of the canal hypersurfaces
(CM1 (")⇤ , p1 + "v1 ) and (CM2 (")⇤ , p2 + "v2 )
are di↵eomorphic.
(7) K(CM1 ("), T CM1 (")p1 +"v1 , p1 + "v1 ) =
K(CM2 ("), T CM2 (")p2 +"v2 , p2 + "v2 ).

Proof. We remark that the germs of the extended families of height func-
tions He i are the graph-like generating families of the Legendrian subman-
ifold covering (Xi⇤ (U ⇥ S k 1 ), (ui , µi )). By the relation b " y⇤i = Xi⇤ in
the proof of Proposition 5.15, such a Legendrian submanifold is Legendrian
equivalent to LCMi " .
Statements (1)–(4) are equivalent by Theorem 5.11 and Proposition
5.2. It also follows from the relation b " (CM (")⇤i = y⇤i (U ⇥ S k 1 ) that the
statements (1) and (6) are equivalent.
 Lagrangian and Legendrian Singularities 137

Statements (6) and (7) are equivalent by Theorem 5.15 for hypersur-
faces.
Statements (4) and (5) are equivalent by Theorem 4.4 and the fact that
1
hvi (ri ) = T (Mi ; vi )pi . ⇤
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws

This page intentionally left blank

 Chapter 6

Surfaces in the Euclidean 3-space

In the previous chapters, we considered the general framework of La-

grangian and Legendrian singularities as well as that of contact between
submanifolds in Rn . We shall apply here the results from those chapters to
study the extrinsic geometry of surfaces embedded in the Euclidean 3-space
R3 .
There are some features on a surface in R3 that determine its local
shape. The features of interest are those that can be followed if the surface
is deformed. These are given the name of robust features by Ian Porteous,
and proved to be of importance in computer vision and shape recognition
([Koenderink (1990); Porteous (2001); Siddiqi and Pizer (2008)]). An
example of a robust feature of a surface is its parabolic set. The parabolic
set is captured by the contact of the surface with planes and lines and can be
detected, on some surfaces, by the naked eye (see for example the parabolic
curves on the bust of Apollo in Figure 204 in [Hilbert and Cohen-Vossen
(1932)], and Figure 6.1).
We start by recalling some basic concepts of the di↵erential geometry
of a surface M in R3 (§6.1). We then consider the contact of M with
planes, lines and spheres and study the singularities of respectively the
height functions, orthogonal projections and distance squared functions on
M (§6.3, §6.4, §6.5). Some types of singularities of these mappings occur
on curves on the surface. These curves are robust features of M and have
geometric meanings. We deal with them in some detail in §6.6.

6.1 First and second fundamental forms

In Chapter 2 we considered hypersurfaces in Rn+1 . We deal here with the

case of surfaces in the Euclidean 3-space R3 .

139
140 Di↵erential Geometry from a Singularity Theory Viewpoint

Fig. 6.1 Visible parabolic curves (in thick) on the surface of a bell.

Let x : U ! R3 be a parametrisation of a surface patch M where U is an

open subset of R2 . The first fundamental form (or the metric) of M = x(U )
at p 2 M is given by Ip (w) = hw, wi, for any w 2 Tp M . If w = axu1 + bxu2
then
Ip (w) = a2 E + 2abF + b2 G
where
E = hxu1 , xu1 i , F = hxu1 , xu2 i , G = hxu2 , xu2 i
are the coefficients of the first fundamental form.
Observe that (EG F 2 )(u1 , u2 ) > 0 for all (u1 , u2 ) 2 U . Also, Ip is the
Riemannian metric on M induced from the Euclidean scaler product. In
the notation of Chapter 2, E = g11 , F = g12 = g21 and G = g22 .
The surface patch M is given an orientation by choosing at each point
p 2 M the unit normal vector N (p) which together with the orientation of
M gives the positive orientation of the ambient space R3 at p. By varying
p on M , we obtain the Gauss map
N : M ! S2.
The Weingarten map Wp : Tp M ! Tp M is given by Wp = dNp . The
second fundamental form of M at p is the bilinear symmetric form given
by
IIp (w, v) = hWp (w), vi = hWp (v), wi
for any v, w 2 Tp M . We still denote by IIp the quadratic form associated
to IIp and write
IIp (w) = IIp (w, w) = la2 + 2mab + nb2 ,
 Surfaces in the Euclidean 3-space 141

where
l = h Nu1 , xu1 i = hN, xu1 u1 i,
m = h Nu1 , xu2 i = hN, xu1 u2 i,
n = h Nu2 , xu2 i = hN, xu2 u2 i
are the coefficients of the second fundamental form of M . Following the
notation in Chapter 2, we have
l = h11 , m = h12 = h21 , n = h22 .
At each point on M , there are two principal curvatures 1 and 2 which
are the eigenvalues of Wp . The Gaussian curvature K and the mean cur-
vature H of M are given by
1
K = 1 2 , H = (1 + 2 ).
2
These can be expressed in terms of the coefficients of the first and second
fundamental forms as follows (see for example [do Carmo (1976)] and
Corollary 2.1)
ln m2
K=
EG F 2
and
lG 2mF + nE
H= .
2(EG F 2 )
The principal curvatures 1 and 2 are the solutions of the quadratic
equation
2 2H + K = 0,
p p
so, one is equal to H + H 2 K and the other to H H 2 K.
Recall from Chapter 1 that the Gaussian curvature K is an intrinsic
invariant of the surface M . However, the principal curvatures 1 and 2
and the mean curvature H are not intrinsic invariants of M . They provide
information about the geometry of M as a surface in the ambient space R3 ,
so they capture some extrinsic properties of M .
If p is not an umbilic point, that is, if 1 6= 2 , there are two orthogonal
principal directions at p denoted by v1 and v2 which are parallel to the
eigenvectors of Wp . At umbilic points , every tangent direction is consid-
ered a principal direction. A curve on M whose tangent direction at all
points is a principal direction is called a line of principal curvature. The
lines of principal curvature are the solution curves of the binary di↵erential
equation
(F n mG)du22 + (En lG)du1 du2 + (Em lF )du21 = 0, (6.1)
142 Di↵erential Geometry from a Singularity Theory Viewpoint

Fig. 6.2 Generic configurations of the lines of principal curvature at umbilic points:
Lemon (left), Star (centre), Monstar (right).

see [do Carmo (1976)] for a proof. Equation (6.1) can also be written in
the following determinant form
du22 du1 du2 du21
E F G = 0.
l m n
The lines of principal curvature form a pair of orthogonal foliations
Their generic topological configurations at umbilic points are determined in
[Sotomayor and Gutierrez (1982); Bruce and Fidal (1989)]; see Figure 6.2
and [Sotomayor (2004)] for historical notes. Porteous gave the colour blue
to one foliation of the lines of curvature and the colour red to the other.
In Figure 6.2, one foliation associated to the curvature i is drawn in solid
lines and the other associated to the curvature j , j 6= i, is drawn in dashed
lines.
The second fundamental form measures how the surface M bends in R3 .
Let ↵ : ( ✏, ✏) ! M be a smooth curve on M with ↵(0) = p. Suppose that
↵ is parametrised by arc length and denote by s the arc length parameter.
The acceleration vector ↵00 is orthogonal to the tangent vector t = ↵0 , so it
lies in the plane with orthonormal basis {t ⇥ N, N }, where N (s) = N (↵(s))
is the normal vector to M at ↵(s). We can write
↵00 = g t ⇥ N + n N,
for some scalars g (called the geodesic curvature) and n (called the normal
curvature). The component g t ⇥ N of ↵00 at s lies in the tangent plane to
M at ↵(s) and the component n N is parallel to the normal of the surface
at ↵(s).
If the geodesic curvature g = h↵00 , t ⇥ N i of the curve ↵ vanishes at
some point s, we say that the curve ↵ has a geodesic inflection at s. In
general, at a geodesic inflection, the curve ↵ is locally on both sides of
the plane at ↵(s) parallel to N (s) and ↵0 (s) (so the projection of ↵ to the
tangent plane T↵(s) M is a plane curve with an inflection).
 Surfaces in the Euclidean 3-space 143

As its name suggests, the geodesic curvature has something to do with

geodesic curves. There is a great amount of study of such curves as they
minimise (at least locally) the length of the distance between points on the
surface. However, we will not pursue them in this book.
If ↵ is not parametrised by arc length, it can be shown by using a
re-parametrisation by arc length that its geodesic curvature is given by
1
g = h↵
¨ , ↵˙ ⇥ N i, (6.2)
˙ 3
||↵||
where ↵˙ = d↵/dt for arbitrary parameter t. We turn now to the normal
curvature. Di↵erentiating the identity hN (s), ↵0 (s)i = 0 leads to
hN (s), ↵00 (s)i = hN 0 (s), ↵0 (s)i .
Therefore,
IIp (↵0 (0)) = h dNp (↵0 (0)), ↵0 (0)i ,
= hN 0 (0), ↵0 (0)i ,
= hN (0), ↵00 (0)i ,
= n .
Observe that the minus sign in the Weingarten map Wp = dNp is
there so that we get IIp (↵0 (0)) is equal to n and not to n . The equality
IIp (↵0 (0)) = n shows that the normal curvature at p depends only on the
unit tangent direction of the curve ↵ at p and not on the curve itself, so we
have the following result.

Proposition 6.1. All curves on a surface M with the same tangent line
at a point p on M have the same normal curvature at p.

If ↵ is not parametrised by arc length, the normal curvature at p is

defined as
↵(0)
˙ IIp (↵(0))
˙ IIp (↵(0))
˙
n = IIp ( )= = 2
.
|↵(0)|
˙ Ip (↵(0))
˙ ||↵(0)||
˙
If ↵(0)
˙ = axu1 + bxu2 , then
a2 l + 2abm + b2 n
n = .
a2 E + 2abF + b2 G
Given a tangent direction w 2 Tp M , the normal section of M at p along
w is the plane curve obtained by intersecting M with the plane through p
generated by N (p) and w. By Proposition 6.1, the normal curvature n at
p along w is the curvature of the normal section at p along w.
144 Di↵erential Geometry from a Singularity Theory Viewpoint

If p is not an umbilic point, we can write w = cos ✓v1 + sin ✓v2 for any
unit tangent vector in Tp M , where v1 and v2 are the orthonormal principal
directions at p. Then
n (✓) = IIp (w) = hWp (cos ✓v1 + sin ✓v2 ), cos ✓v1 + sin ✓v2 i
= hcos ✓1 v1 + sin ✓2 v2 , cos ✓v1 + sin ✓v2 i
= 1 cos2 ✓ + 2 sin2 ✓.
Therefore, the principal curvatures 1 and 2 are the extrema of the normal
curvature n (✓) at p when ✓ varies in [0, ⇡].
A direction along which the normal curvature is zero is called an asymp-
totic direction. Thus, a direction w 2 Tp M is asymptotic if and only if
IIp (w) = 0. We observe that the normal section along an asymptotic di-
rection has an inflection and that there are two asymptotic directions at a
hyperbolic point, one at a parabolic point and none at an elliptic point (see
§2.1.4).
A curve on M whose tangent direction at each point is an asymptotic
direction is called an asymptotic curve. The asymptotic curves are the
solutions of the binary di↵erential equation
ndu22 + 2mdu1 du2 + ldu21 = 0. (6.3)

Definition 6.1. Two directions w1 and w2 in Tp M are said to be conjugate

(with respect to the second fundamental form) if and only if
hWp (w1 ), w2 i = 0.
A conjugate direction to w is denoted by w̄.

Proposition 6.2. (i) A conjugate direction to v = axu1 + bxu2 is v̄ =

(am + bn)xu1 + (al + bm)xu2 .
(ii) IIp (v̄) = (nl m2 )IIp (v).
(iii) n (v̄) = n (v)(nl m2 )||v||2 /||v̄||2 .
(iv) ||v||2 ||v̄||2 hv̄, vi2 = (EG F 2 )IIp (v)2 .

Proof. (i) Write v̄ = ⇠1 xu1 + ⇠2 xu2 . Then hWp (v̄), vi = 0 is equivalent

to
a⇠1 l + (a⇠2 + b⇠1 )m + b⇠2 n = 0.
Rearranging the above expression gives
(al + bm)⇠1 + (am + bn)⇠2 = 0,
so we can take ⇠1 = (am + bn) and ⇠2 = (al + bm).
 Surfaces in the Euclidean 3-space 145

(ii) We have
IIp (v̄) = (am + bn)2 l 2(am + bn)(al + bm)m + (al + bm)2 n
= (nl m2 )(a2 l + 2abm + b2 n)
= (nl m2 )IIp (v).
(iii) It follows from (ii) that
1
n (v̄) = IIp (v̄)
||v̄||2
1
= (nl m2 )IIp (v)
||v̄||2
||v||2
= (nl m2 )n (v).
||v̄||2
(iv) We have
||v||2 ||v̄||2 hv̄, vi2 =
(a2 E + 2abF + b2 G)((am + bn)2 E 2(am + bn)(al + bm)F + (al + bm)2 G)
2
(a(am + bn)E + (a(al + bm) b(am + bn))F + b(al + bm)G)
= (EG F 2 )(a2 l + 2abm + b2 n)2
= (EG F 2 )IIp (v)2 . ⇤

Example 6.1. An asymptotic direction can be defined as a direction which

coincides with its conjugate direction, equivalently, as a direction which
forms a zero angle with its conjugate direction. The conjugate to the prin-
cipal direction v1 is the principal direction v2 and vice-versa. In fact, the
angle between w and w̄ is ⇡/2 if and only if w is a principal direction.
Fletcher considered in [Fletcher (1996)] directions in Tp M which form a
fixed oriented angle ↵ with their conjugate directions. It turns out that,
away from umbilic and parabolic points, there are at most two of these
directions at each point on M . Fletcher called the integral curves of these
directions the conjugate curve congruence and denoted them by C↵ . The
conjugate curve congruences are solutions of a binary di↵erential equation
and the family C↵ interpolates between the asymptotic curves (↵ = 0) and
the lines of principal curvature (↵ = ⇡/2). More work on the family C↵ and
other families can be found in [Bruce, Fletcher and Tari (2004); Bruce and
Tari (2005); Nabarro and Tari (2009); Oliver (2010, 2011); Tari (2009)].

6.2 Surfaces in Monge form

At each point p on the surface M , we can choose a coordinate system in R3

so that p is the origin, Tp M is the plane z = 0 and the surface M is locally
146 Di↵erential Geometry from a Singularity Theory Viewpoint

the graph of some function z = f (x, y), with (x, y) in a open subset U of
R2 containing the origin. Then we have the Monge form parametrisation
(x, y) = (x, y, f (x, y)), (x, y) 2 U , of M at p. We shall say that M is
parametrised locally in Monge form z = f (x, y) at the origin p. Note that
the Taylor expansion of f at the origin has no constant or linear terms.

Proposition 6.3. Let M be parametrised locally in Monge form z = f (x, y)

at the origin p. Then,
(i) E = 1 + fx2 , F = fx fy , G = 1 + fy2 .
1
(ii) N = q ( fx , fy , 1).
1 + fx2 + fy2
fxx fxy fyy
(iii) l = q ,m= q ,n= q .
2
1 + fx + fy 2 2 2
1 + fx + fy 2 2
1 + fx + fy

Proof. We have x = (1, 0, fx ) and y = (0, 1, fy ), so

E=h x, xi= 1 + fx2 ,
F =h x , y i = fx fy ,
2
G=h y , y i = 1 + fy .

The Gauss map at (x, y) is given by

⇥
x y 1
N= =q ( fx , fy , 1).
|| x ⇥ y || 2 2
1 + fx + fy
The coefficients of the second fundamental form are given by
q
l = hN, xx i = fxx / 1 + fx2 + fy2 ,
q
m = hN, xy i = fxy / 1 + fx2 + fy2 ,
q
n = hN, yy i = fyy / 1 + fx2 + fy2 .
⇤
For a surface given in Monge form, we write the homogeneous part of
degree k of the Taylor expansion of f at the origin in R2 in the form
k
X
aki xk i y i .
i=0

6.3 Contact with planes

Let x : U ! R3 be a parametrisation of a surface patch M . The contact

of M with planes is measured by the singularities of the height functions
 Surfaces in the Euclidean 3-space 147

on M . Recall that the family of height functions H : U ⇥ S 2 ! R on M is

given by
H(u, v) = hx(u), vi
and the extended family of height functions H̃ : U ⇥ S 2 ⇥ R ! R is given
by
H̃(u, v, r) = hx(u), vi r.
For v 2 S 2 fixed, the height function hv along the direction v is the
function hv (u) = H(u, v). Similarly, for v and r fixed, h̃v,r is the function
h̃v,r (u) = H̃(u, v, r).

Theorem 6.1. There is an open and dense set O1 of proper immersions

x : U ! R3 , such that for any x 2 O1 , the surface M = x(U ) has the
following properties. For any v 2 S 2 , the height function hv (resp. h̃v,hp,vi )
along the normal direction v at any point p on M has only local singularities
of type A1 , A2 or A3 which are R+ (resp. K) versally unfolded by the family
H (resp. H̃).

Proof. The result follows from Theorems 4.3, 4.5 and 4.7. ⇤

Definition 6.2. A surface is called (locally) height function generic if any

of its local parametrisations belongs to the set O1 in Theorem 6.1.

We shall derive geometric information about the surface M from the

family of height functions. We start with the A1 -singularity. Given a
direction v in S 2 , the functions hv and h̃v,hp,vi have the same R or K-type
singularity at any point p on M .

Proposition 6.4. The height functions hv is singular at p 2 M if and only

if v is a normal direction to M at p. The singularity of hv at p is of type
A+1 if and only if p is an elliptic point, A1 if and only if p is a hyperbolic
point, and A 2 if and only if p is a parabolic point.

Proof. We only need to characterise geometrically the A+ 1 and A1 sin-

gularities as the remaining part of the proposition is a particular case of
Propositions 2.4 and 2.5. These singularities are distinguished by the sign
of the determinant of the Hessian matrix of hv , positive for A+
1 and negative
for A1 .
It follows from the Weingarten formula Theorem 2.2 and Corollary 2.1
that
det H(hv )(u) = K(u)(EG F 2 )(u).
148 Di↵erential Geometry from a Singularity Theory Viewpoint

As (EG F 2 )(u) > 0, det H(hv )(u) and K(u) have the same sign, and
the result follows. ⇤
When the surface is taken in Monge form z = f (x, y), the conditions
for the height function hN0 (x, y) = f (x, y) along the normal direction N0 =
(0, 0, 1) to have a given singularity at the origin p and for the family of
height functions to be a R+ -versal unfolding of these singularities can be
expressed in terms of the coefficients of the Taylor expansion of f at p. We
write
f (x, y) = a20 x2 + a21 xy + a22 y 2 + a30 x3 + a31 x2 y + a32 xy 2 + a33 y 3 + O(4).
(6.4)
We can rotate the coordinate axes in the tangent plane Tp M and set a
chosen direction to be along the y-axis. The chosen direction could be, for
example, an asymptotic direction if p is not an elliptic point or a principal
direction.

Proposition 6.5. The following hold for M in Monge form z = f (x, y) at

the origin p with f as in (6.4) and v = (0, 1, 0) 2 Tp M .
(i) The point p is a parabolic point if and only if a221 4a20 a22 = 0.
(ii) The direction v is asymptotic at p if and only if a22 = 0.
(iii) If v is an asymptotic direction at p and p is a parabolic point but not
a flat umbilic point, the parabolic set is a smooth curve at p if and only
if a32 6= 0 or a33 6= 0.

Proof. For f as in (6.4), the 1-jets at the origin of the coefficients of the
second fundamental form for the Monge form setting in Proposition 6.3 are
given by
j 1 l = 2a20 + 6a30 x + 2a31 y,
j 1 m = a21 + 2a31 x + 2a32 y,
j 1 n = 2a22 + 2a32 x + 6a33 y.
It follows that
j 1 (ln m2 )(x, y) = 4a20 a22 a221 + 4(a20 a32 + 3a30 a22 a21 a31 )x+
4(3a20 a33 + a31 a22 a21 a32 )y.
(i) In particular, the origin is a parabolic point if and only if
(ln m2 )(0, 0) = 4a20 a22 a221 = 0.
(ii) At the origin p, l = 2a20 , m = a21 and n = 2a22 so a direction
a x (0, 0) + b y (0, 0) = (a, b, 0) is asymptotic at p if and only if
a20 a2 + a21 ab + a22 b2 = 0
 Surfaces in the Euclidean 3-space 149

(see equation (6.3)). Therefore, v = (0, 1, 0) is an asymptotic direction at

p if and only if a22 = 0.
(iii) When v is an asymptotic direction and p is a parabolic point,
j 1 (ln m2 )(x, y) = 4a20 a32 x + 12a20 a33 y.
As p is not a flat umbilic point, a20 6= 0, so ln m2 is not singular at
the origin if and only if a32 6= 0 or a33 6= 0. ⇤

Theorem 6.2. The following hold for M in Monge form z = f (x, y) at the
origin p with f as in (6.4) and v = (0, 1, 0) 2 Tp M . Assume that p is not
a flat umbilic point.
(i) For p a parabolic point and v an asymptotic direction at p, the height
function HN0 has a singularity at p of type
A2 () 6 0,
a33 =
A3 () a33 = 0 and a232 4a20 a44 6= 0.

(ii) The family of height functions on M is a R+ -versal unfolding of the

Ak -singularities, k = 1, 2, 3, of HN0 if and only if
A1 : always
A2 : always
A3 : a32 6= 0, equivalently, the parabolic set is a smooth curve.

Proof. (i) The height function along the normal direction N0 = (0, 0, 1)
at the origin p is given by hN0 (x, y) = f (x, y). By Proposition 6.5, if p is a
parabolic point and v is an asymptotic direction at p, then a22 = a21 = 0
so that
f (x, y) = a20 x2 + a30 x3 + a31 x2 y + a32 xy 2 + a33 y 3 + ⌃4i=0 a4i x4 i y i + O(5).
(We chose the unique asymptotic direction at p to be along v in order
to have j 2 f = a20 x2 . This makes the task of recognizing the singularities
of hN0 much easier.) The hypothesis on p not being a flat umbilic point is
then equivalent to a20 6= 0. The height function f has an A 2 -singularity
at the origin if and only if a33 = 0. Then the singularity is of type A3
if and only if a20 x2 + a32 xy 2 + a44 y 4 is not a perfect square, equivalently,
a232 4a20 a44 6= 0.
(ii) Observe that the family of height functions H : R2 ⇥S 2 , (0, N0 ) ! R
is P -R+ -equivalent to the modified family, that we still denote by H :
R2 ⇥ R2 , (0, 0) ! R, which is given by
H(x, y, ↵, ) = ↵x + y + f (x, y)
150 Di↵erential Geometry from a Singularity Theory Viewpoint

(all we did here is to parametrise the sphere S 2 near (0, 0, 1) by (↵, , 1)).
Recall from Theorem 3.12 that H is an R+ -versal unfolding if and only
if
n o
LRe · f + R. 1, Ḣ1 , Ḣ2 = E2 . (6.5)
We have
Ḣ1 (x, y) = H↵ (x, y, 0, 0) = x,
Ḣ2 (x, y) = H (x, y, 0, 0) = y.
For f as in (6.4),
j 3 fx = 2a20 x + a21 y + 3a30 x2 + 2a31 xy + a32 y 2
+4a40 x3 + 3a41 x2 y + 2a42 xy 2 + a43 y 3
j 3 fy = a21 x + 2a22 y + a31 x2 + 2a32 xy + 3a33 y 2
+a41 x3 + 2a42 x2 y + 3a43 xy 2 + 4a44 y 3 .
The A1 -singularity is 2-R-determined, that is M32 ⇢ LR · f , so for H
to be an R+ -versal unfolding it is enough to show that equality (6.5) holds
modulo M32 , that is
n o
j (LRe · f + R. Ḣ1 , Ḣ2 + h1iR ) = J 2 (2, 1).
2
(6.6)

We view J 2 (2, 1) as an R-vector space with basis 1, x, y, x2 , xy, y 2 . To

show that equality (6.6) holds, it is enough
n to show
o that the elements of the
basis of J 2 (2, 1) are in j 2 (LRe ·f +R. 1, Ḣ1 , Ḣ2 ). Clearly, 1, x = Ḣ1 (x, y)
and y = Ḣ2 (x, y) are in there.
Let ⇠1 = j 2 (yfx ) = 2a20 xy + a21 y 2 and ⇠2 = j 2 (yfy ) = a21 xy + 2a22 y 2
be two vectors in j 2 (LRe · f ). Then
✓ ◆ ✓ ◆
⇠1 xy
=A ,
⇠2 y2
with
✓ ◆
2a20 a21
A= .
a21 2a22
The matrix A is invertible as its determinant is 4a20 a22 a221 6= 0. Hence
✓ ◆ ✓ ◆
xy 1 ⇠1
2 =A ,
y ⇠2
which shows that xy, y 2 2 j 2 (LRe · f ). From this we get x2 = (j 2 (xfx )
a21 xy)/(2a20 ) 2 j 2 (LRe · f ). Therefore (6.6) always holds at an A1 -
singularity.
 Surfaces in the Euclidean 3-space 151

We set now a21 = a22 = 0. For the A2 -singularity which is 3-R-

determined, it is enough to show that
n o
j 3 (LRe · f + R. 1, Ḣ1 , Ḣ2 ) = J 3 (2, 1) (6.7)

for H to be an R+ -versal unfolding.

Let P be a monomial of degree 2. Then the vector j 3 (P fx ) 2
j 3 (LRe · f ), and this shows that all the monomials of degree 3 divisible
by x, i.e., x3 , x2 y, xy 2 are in j 3 (LRe · f ). Then y 3 = (j 3 (yfy ) a31 x2 y
2a32 xy 2 )/3a33 2 j 3 (LRe · f ) (we do have a33 6= 0 at an A2 -singularity of
the height function f ). We proceed similarly to show that all the degree 2
monomials are in j 3 (LRe · f ). This means that equality (6.7) always holds
at an A2 -singularity.
For the A3 -singularity, we set a21 = a22 = a33 = 0. As this singularity is
4-R-determined, we work modulo M52 . Using vectors of the form j 4 (P fx ),
we get all monomials of degree 4 divisible by x in j 4 (LRe · f ), then those
of degree 3 divisible by x2 . Working modulo these monomials we have
⌘1 = j 4 (y 2 fx ) ⌘ 2a20 xy 2 + a32 y 4
⌘2 = j 4 (yfy ) ⌘ 2a32 xy 2 + 4a44 y 4
that is
✓ ◆ ✓ ◆
⌘1 xy 2
=A ,
⌘2 y4
with
✓ ◆
2a20 a32
A= .
2a32 4a44
The determinant 2(4a20 a44 a232 ) of matrix A is distinct from zero as
the height function has an A3 -singularity. Therefore xy 2 and y 4 are in
j 4 (LRe · f ). Using now j 4 (xfx ) and j 4 (yfy ) modulo the monomials that
we have shown are in j 4 (LRe · f ), we get x2 , xy 2 j 4 (LRe · f ). We also get
y 3 and xy using ⌘1 = j 4 (yfx ) and ⌘2 = j 4 (fy ) (we get the same matrix A
as above).
The only way to get y 2 in j 4 (LRe · f ) is by using j 4 fx ⌘ a32 y 2 , modulo
the monomials that are already in j 4 (LRe · f ). Therefore, H is a P -R+ -
versal unfolding of the A3 -singularity of hN0 if and only if a32 6= 0. The
geometric interpretation of this condition is given in Proposition 6.5(iii).⇤
Several geometric properties of M can be deduced from Theorems 6.1
and 6.2.
152 Di↵erential Geometry from a Singularity Theory Viewpoint

The intersection of the surface M with its tangent plane at p is called

the tangent Dupin indicatrix of 2 M at p.

Proposition 6.6. The tangent Dupin indicatrix of M at p consists locally

of an isolated point if p is an elliptic point, a pair of transverse crossing
curves if p is a hyperbolic point, a curve with a cusp singularity if p is
an ordinary parabolic point (i.e., an A2 -singularity of the height function),
an isolated point or a tacnode (a pair of tangential curves) if p is an A3 -
singularity of the height function. See Figure 6.3.

Fig. 6.3 Intersection of a surface with its tangent plane, top figures at an elliptic,
hyperbolic and parabolic point respectively, and bottom figures at an A+ 3 and A3 -
singularities of the height function.

Proof. If we take the surface in Monge form z = f (x, y), then the height
function HN0 along the normal N0 = (0, 0, 1) is just the function f . The
intersection of the surface with its tangent plane z = 0 is the zero set of the
function f , which is di↵eomorphic to the zero set of the R-normal form of
the singularity of hN0 = f . The zero sets of the normal forms (taken up to
a sign ±) are as follows
 Surfaces in the Euclidean 3-space 153

Name R-Model Zero set

A+
1 x2 + y 2 {(0, 0)}
A−
1 x2 − y 2 {(x, y) ∈ R2 , 0 : x = y or x = −y}
A2 x2 + y 3 {(t3 , −t2 ), t ∈ R, 0}
A+
3 x2 + y 4 {(0, 0)}
A−
3 x2 − y 4 {(x, y) ∈ R2 , 0 : x = y 2 } ∪ {(x, y) ∈ R2 , 0 : x = −y 2 }

!
Proposition 6.6 can be viewed as a local classification, up to diffeo-
morphisms, of the tangent Dupin indicatrices of a height function generic
surface. We can also consider the tangential Dupin foliation of M at p,
which is obtained by intersecting the surface M with planes parallel to
Tp M .

Proposition 6.7. The tangential Dupin foliation of a surface M at a point

p is locally diffeomorphic to the foliation given by the level sets of the R-
model of the singularity of the height function on M at p along its normal
direction. See Figure 6.4.

A+
1 A−
1 A2 A+
3 A−
3

Fig. 6.4 The models, up to diffeomorphisms, of the tangential Dupin foliation of a

height function generic surface.

Proof. We consider the same setting as in the proof of Proposition 6.6,

so that hN0 = f . The result follows by observing that the diffeomorphism
in the R-equivalence between two germs maps the level sets of one germ to
the level sets of the other. !
The cylindrical pedal of the surface M is the image of the map x∗ :
U → S 2 × R, given by
x∗ (u) = (N (u), ⟨x(u), N (u)⟩),
(see Chapters 2 and 5). The cylindrical pedal can be viewed as the dual
surface of M .
154 Differential Geometry from a Singularity Theory Viewpoint

Fig. 6.5 The dual surface of M at the Ak -singularities of the height function: A1 left,
A2 centre and A3 right.

Proposition 6.8. The dual surface of a height function generic surface M ,

or its cylindrical pedal, is locally diffeomorphic to
(i) a smooth surface when hN0 has an A1 -singularity at p,
(ii) a cuspidal edge surface when hN0 has an A2 -singularity at p,
(iii) a swallowtail surface when hN0 has an A3 -singularity at p,
where N0 is the normal to M at p. See Figure 6.5.

Proof. The cylindrical pedal of M is a wavefront with H̃ its graph-like

generating family (Theorem 5.14). Thus, it has Legendrian singularities.
For a height function generic M , the family H̃ is K-versal, so the cylin-
drical surface is diffeomorphic to the discriminant of a model of a K-versal
unfolding (with three parametres) of the singularities that occur in a given
height function. These are A1 , A2 and A3 -singularities (Theorem 6.1). The
discriminants of the K-versal unfoldings with three parameters of these sin-
gularities are as stated in the proposition. !
The stable singularities of the Gauss map are given in [Looijenga (1974)]
and [Bleeker and Wilson (1978)]. These singularities are intimately related
to those of the height functions on M .

Proposition 6.9. The Gauss map N : M → S 2 of a height function

generic surface M is either a local diffeomorphism or has singularities of
type fold or cusp. The Gauss map is singular at p if and only if p is a
parabolic point. Its fold singularities occur on the parabolic set at the A2 -
singularities of the height function and its cusp singularities occur at the
A3 -singularities of the height functions.

Proof. The family H is the generating family of the Lagrangian sub-

manifold L(H)(CH ) ⊂ T ∗ S 2 and the Gauss map N is the corresponding
catastrophe map χH : CH → S 2 . For a height function generic surface M ,
 Surfaces in the Euclidean 3-space 155

the family H is an R+ -versal unfolding, so the catastrophe map N = H is

A-equivalent to the catastrophe map of a model of an R+ -versal unfolding
(with two parametres) of the singularities that occur in a given height func-
tion. The catastrophe map of an R+ -versal 2-parameter family of the A1 ,
A2 or A3 singularities are respectively the map-germs of a di↵eomorphism,
a fold or a cusp. ⇤

Corollary 6.1. For a height function generic surface M, the Gauss map
at p 2 M is A-equivalent to:

(i) a fold map-germ if and only if the dual surface of M at p is locally

di↵eomorphic to a cuspidal edge surface;
(ii) a cusp map-germ if and only if the dual surface of M at p is locally
di↵eomorphic to a swallowtail surface.

It follows from Proposition 6.9 that the image of the parabolic set by
the Gauss map is a curve with a cusp singularity at an A3 -singularity of
the height function on M . For this reason, we have the following definition.

Definition 6.3. A point on the parabolic set where the height function
along the normal direction to the surface has an A3 -singularity is called a
cusp of Gauss. A cusp of Gauss p is elliptic if the height function along the
normal direction at p has an A+3 -singularity at p and hyperbolic if it has
an A3 -singularity.

We need the following lemma for some characterisations of the cusps of

Gauss.

Lemma 6.1. Let M be a surface parametrised locally by x : U ! R3 and

let p1 = x(u1 ) and p2 = x(u2 ) be two distinct points on M . Then,
(1) N (u1 ) = N (u2 ) if and only if the tangent affine planes of M at p1
and p2 are parallel.
(2) x⇤ (u1 ) = x⇤ (u2 ) if and only if the tangent affine planes of M at p1
and p2 are the same.

Proof. The tangent affine plane of M at pi , i = 1, 2, is given by

Tpi (M ; vi )p = hvi1 (ri ), where vi = N (ui ) and ri = H(pi , vi ). The con-
dition N (u1 ) = N (u2 ) means that the normal vectors to Tp1 (M ; v1 ) and
Tp2 (M ; v2 ) are linearly dependent, so the two planes are parallel.
We have x⇤ (u1 ) = x⇤ (u2 ) if and only if N (u1 ) and N (u2 ) are parallel
and r1 = r2 . This means that Tp1 (M ; v1 ) = Tp2 (M ; v2 ). ⇤
 156 Di↵erential Geometry from a Singularity Theory Viewpoint

Cusps of Gauss can be identified geometrically in many ways ([Bancho↵,

Ga↵ney and McCrory (1982)]). We have the following result, the main
part of which is given in [Bancho↵, Ga↵ney and McCrory (1982)]. We
include some additional information from the view point of the Legendrian
singularity framework.

Theorem 6.3. Let M be a height function generic surface parametrised

locally by x : U ! R3 , and let p0 = x(u0 ) be a point on M . Then the
following statements are equivalent.

(i) The point p0 is a cusp of Gauss.

(ii) The dual surface of M is a swallowtail surface at p0 .
(iii) HN0 has an A3 -singularity at p0 .
(iv) The order of contact of M with its tangent plane at p0 is equal to 3
(i.e., T-ord(x(U ), u0 ) = 3).
(vi) For any ✏ > 0, there exist three distinct non-parabolic points pi = x(ui )
such that ||u0 ui || < ✏ for i = 1, 2, 3, and the tangent planes of M at
p1 , p2 , p3 are parallel.
(vii) For any ✏ > 0, there exist two distinct non-parabolic points pi = x(ui )
such that ||u0 ui || < ✏ for i = 1, 2, and the tangent planes to M at p1
and p2 are equal.

Proof. Statements (i) and (ii) are equivalent by Corollary 6.1.

It follows from Proposition 6.8 that statement (iii) is equivalent to state-
ment (ii).
As the Ke -codimention of the Ak -singularity is k, by definition of the
order of contact of M with its tangent plane, statements (iii) and (iv) are
equivalent.
Since M is a height function generic surface, the Gauss map has only
fold or cusp singularities. If u0 is a fold singularity, there is a neighbourhood
of u0 on which the Gauss map is 2 to 1 except on the parabolic curve (i.e,
on the curve of fold singularities of the Gauss map). Of course, if u0 is not
a parabolic point, the Gauss map is one to one locally at u0 . By Lemma
6.1, the negation of statement (i) implies the negation of statement (vi). If
u0 is a cusp singularity, the image of the discriminant of the Gauss map
has an ordinary cusp at N0 . By considering the A-normal form (x, xy + y 3 )
of the cusp singularity, it can be shown that a point has 3, 2 or 1 pre-image
points by the Gauss map. By Lemma 6.1, this means that statement (vi)
holds. Therefore, statement (i) implies statement (vi).
The dual surface of a height function generic surface M has singularities
 Surfaces in the Euclidean 3-space 157

"

!!

Fig. 6.6 Generic configurations of the asymptotic curves at a cusp of Gauss.

of type cuspidal edge or swallowtail. If p0 is a cuspidal edge point, the dual

surface has no self-intersections around the image of p0 , so by Lemma 6.1,
the negation of statement (vii) implies the negation of statement (ii).
At a swallowtail point p0 , there is a curve on M containing p0 which
corresponds to the points of self-intersections of the dual surface of M . On
this curve, there are two distinct points p1 = x(u1 ) and p1 = x(u2 ) such
that x∗ (u1 ) = x∗ (u2 ). By Lemma 6.1, this means that the tangent affine
planes to M at points p1 and p2 are equal. Since the singularities are either
cuspidal edges or swallowtails, statement (vii) characterises a swallowtail
point of the dual surface. !

Remark 6.1. Cusps of Gauss can also be captured by the singularities of

the asymptotic curves. The way these are studied is by lifting the bi-valued
direction field in the plane determined by equation (6.3) to a single vector
field ξ on a smooth surface S. The vector field ξ is singular at a point q if
and only if q is the lift of a cusp of Gauss p. The singularity of ξ is of type
saddle (has index −1) if and only if p is a hyperbolic cusp of Gauss and of
type node or focus (has index +1) if and only if p is an elliptic cusp of Gauss.
The generic topological configurations of the asymptotic curves at the cusp
of Gauss are as in Figure 6.6 ([Banchoff, Gaffney and McCrory (1982);
Banchoff and Thom (1980); Dara (1975); Davydov (1994); Kergosien and
Thom (1980); Thom (1972)]). For more results on cusps of Gauss see
[Bruce and Tari (2000); Banchoff, Gaffney and McCrory (1982); Oliver
(2011); Uribe-Vargas (2006); Izumiya and Takahashi (2011)].

Proposition 6.10. Let M be given in Monge form z = f (x, y) at p with f

as in (6.4) and let v = (0, 1, 0) ∈ Tp M .
158 Di↵erential Geometry from a Singularity Theory Viewpoint

If v is an asymptotic direction at p and p is a parabolic point but not

a cusp of Gauss, then the geodesic curvature at N (p) of the image of the
parabolic curve by the Gauss map is
3a31 a33 a232
g = .
24a320 a33

Proof. From the proof of Proposition 6.5(iii), when p is an A2 -singularity

of the height function a33 6= 0, so the zero set of ln m2 can be parametrised
by
a32
↵(x) = (x, x + ⌘x2 + g(x))
3a33
with
a42 a32 a43 2a232 a44 (a232 3a31 a33 )2
⌘= + + ,
3a33 3a233 9a333 27a333 a20
and g is a germ, at the origin, of a smooth function with a zero 2-jet. (The
constant ⌘ will not contribute to the geodesic curvature we are seeking here
and is included for completeness. It contributes to the geodesic curvature
of the parabolic set at p.) The parabolic set is the image by of the curve
↵ and we have
a32
j 2 (↵(x)) = (x, x + ⌘x2 , a20 x2 ).
3a32
The Gauss map is given by
1 @f @f
N= ( , , 1),
(1 + ( @f 2 ( @f 2 12 @x @y
@x ) + @y ) )

see Proposition 6.3. Its 2-jet at the origin is given by

( 2a20 x 3a30 x2 2a31 xy a32 y 2 , a31 x2 2a32 xy 3a33 y 2 , 1 2a220 x2 ).

Let (x) = N ↵(x) be a local parametrisation of the image of the

parabolic set by the Gauss map. Then
2a31 a32 a332 2 3a31 a33 a232 2
j 2 (x) = ( 2a20 x (3a30 + 2 )x , x , 1 2a220 x2 ).
3a33 9a33 3a33
The geodesic curvature of at N (p) is (see formula (6.2))
1 00 0 3a31 a33 a232
g = h (0), (t) ⇥ N (↵(0))i = .
0
|| (0)||3 24a320 a33
⇤
 Surfaces in the Euclidean 3-space 159

6.4 Contact with lines

As seen in §4.6 of Chapter 4, the family of orthogonal projections P :

U ⇥ S 2 ! T S 2 on M is given by
P (u, v) = (v, x(u) hx(u), viv).
We denote the second component of P by Pv and consider Pv as the
orthogonal projection of M along the fixed direction v. We have, for i =
1, 2,
@Pv
(u1 , u2 ) = xui (u) hxui (u), viv.
@ui
The map Pv is singular at p = x(u) if and only if the vectors @Pv /@u1 (u)
and @Pv /@u2 (u) are linearly dependent, and this occurs if and only if v is
tangent to M at p.
We denote by ⌃v the set of critical points of Pv (these are the points
on M where v is tangent to M ) and by v the image of ⌃v by Pv , so
v = Pv (⌃v ). The set v is called the profile or apparent contour of M
along the direction v and ⌃v is called the contour generator of M along
the direction v (see Figure 6.7).
The K-singularities of Pv measure the contact of M with lines parallel
to v. In Chapter 4, we reviewed the general framework of contact between
submanifolds. When the dimension of the target of these map-germs is
greater than 1, the group K is too large to give significant geometric infor-
mation. This is why we use its subgroup A instead. (Montaldi’s results are
still valid for the group A, see Chapter 4.)
Another justification for using the group A is the following. Our aim
is to understand the singularities of the apparent contour and how they
bifurcate as the direction of projection varies in S 2 . We can suppose that
the point p 2 M is the image of 0 2 R2 by the parametrisation x, identify
the plane of projection Tv S 2 with R2 and suppose that Pv (p) is also the
origin in R2 . We have the diagram

(M, p) ⇢ R3
6
x Pv
✏
(R2 , 0) / (Tv S 2 , Pv (p)) ' (R2 , 0)

and still denote the composite map-germ Pv x : (R2 , 0) ! (R2 , 0) by Pv .

The singularities of the apparent contour of interest are those which are
160 Differential Geometry from a Singularity Theory Viewpoint

,)&')*$ -%&%$#')*$ !""#$%&' ()&')*$

Fig. 6.7 An apparent contour of a torus: the dashed parts represent the invisible part
of the contour generator and apparent contour when the surface is not transparent.

not altered by a re-parametrisation of the surface (given by germs of diffeo-

morphisms h in the source (R2 , 0)) nor by smooth changes of coordinates in
the plane of projection (given by germs of diffeomorphisms k in the target
(R2 , 0)), as shown in the diagram below
Pv
(R2 , 0) " (R2 , 0)

h k
! k◦P ◦h−1 !
v
(R2 , 0) " (R2 , 0)
It is thus appropriate to use the group A for studying the singularities
of the germs of orthogonal projections of M .

6.4.1 Contour generators and apparent contours

Proposition 6.11. (i) The contour generator Σv is a singular curve at p

if and only if p is a parabolic point and v is the unique asymptotic direction
at p.
(ii) The apparent contour ∆v is a smooth curve at Pv (p) if and only if
v is not an asymptotic direction at p.
 Surfaces in the Euclidean 3-space 161

Proof. Let x : U ! R3 be a local parametrisation of M .

(i) A point x(u) is in ⌃v if and only if
[v, xu1 (u), xu2 (u)] = 0,
where [., ., .] denotes the determinant of the matrix formed by three vectors
in R3 . The contour generator ⌃v is singular at u if and only if

[v, xu1 u1 (u), xu2 (u)] + [v, xu1 (u), xu1 u2 (u)] = 0, (6.8)
[v, xu1 u2 (u), xu2 (u)] + [v, xu1 (u), xu2 u2 (u)] = 0. (6.9)
We drop the argument u and write v = axu1 + bxu2 . The vectors xui uj
can be written, with respect to the basis {xu1 , xu2 , N } of R3 , in the form
1 2
x u1 u1 = 11 xu1 + 11 xu2 + lN
1 2
x u1 u2 = 12 xu1 + 12 xu2 + mN
1 2
x u2 u2 = 22 xu1 + 22 xu2 + nN
where kij are the Christo↵el symbols (see for example [do Carmo (1976)]).
Then equation (6.8) becomes
[axu1 + bxu2 , 111 xu1 + 211 xu2 + lN, xu2 ]
+[axu1 + bxu2 , xu1 , 112 xu1 + 212 xu2 + mN ] = 0,
which is equivalent to
al + bm = 0
as [xu1 , xu2 , N ] 6= 0. Similarly, equation (6.9) is equivalent to am + bn = 0.
Therefore, the contour generator is singular if and only if
⇢
al + bm = 0
am + bn = 0
equivalently, nl m2 = 0 and v = nxu1 + mxu2 at u = (u1 , u2 ). These
are precisely the conditions for p to be a parabolic point and for v to be
the unique asymptotic direction at p.
(ii) The contour generator ⌃v must be a smooth curve at p otherwise
v is singular at Pv (p). We take, without loss of generality, am + bn 6= 0
at p so that ⌃v can be parametrised by u1 7! (u1 , u2 (u1 )) for some smooth
function u2 (u1 ) with
al + bm
u02 = . (6.10)
am + bn
The apparent contour v is then parametrised by
Pv (u1 ) = x(u1 , u2 (u1 )) hx(u1 , u2 (u1 )), viv.
162 Di↵erential Geometry from a Singularity Theory Viewpoint

We have
Pv0 = xu1 + u02 xu2 hxu1 + u02 xu2 , viv
= xu1 + u02 xu2 hxu1 + u02 xu2 , axu1 + bxu2 i(axu1 + bxu2 )
= (1 a(aE + (u02 a + b)F + u02 bG)) xu1
+ (u02 b(aE + (u02 a + b)F + u02 bG)) xu2
Therefore, v is singular if and only if
1 a(aE + (u02 a + b)F + u02 bG) = 0,
u02 b(aE + (u02 a + b)F + u02 bG) = 0.
The above two equations can be written, after substituting u02 by its
expression in (6.10), in the form
am + bn = a a2 (mE lF ) + ab(nE lG) + b2 (nF mG) , (6.11)
al + bm = b a2 (mE lF ) + ab(nE lG) + b2 (nF mG) . (6.12)
Observe that a2 (mE lF ) + ab(nE lG) + b2 (nF mG) = 0 means
that v is a solution of equation (6.1), that is, v is a principal direction. If
this is the case, the system of equations (6.11) and (6.12) becomes
⇢
al + bm = 0
am + bn = 0
equivalently, p is a parabolic point and v the unique asymptotic direction
at p. But this is excluded as it implies that ⌃v , and hence v , is singular.
Suppose then that a2 (mE lF ) + ab(nE lG) + b2 (nF mG) 6= 0.
Dividing side by side equation (6.11) by equation (6.12) yields
am + bn a
= .
al + bm b
Rearranging the above equality gives
a2 l + 2abm + b2 n = 0.
This implies that v is a solution of equation (6.3), that is, v is an asymptotic
direction.
Conversely, let v be a unit asymptotic direction (we project along v so
we require v 2 S 2 ). Then a2 l+2abm+b2 n = 0, so al+bm = (am+bn)b/a.
The right hand side of equation (6.11) can then be rearranged to get
a a2 (mE lF ) + ab(nE lG) + b2 (nF mG)
= a (a2 m + abn)E + (b2 n a2 l)F (abl + mb2 )G
✓ ◆
b2
= a a(am + bn)E + 2b(am + bn)F + (am + bn)G)
a
= (am + bn)(a2 E + 2abF + b2 G)
= am + bn.
 Surfaces in the Euclidean 3-space 163

Similarly, rearranging the right hand side of equation (6.12) gives

b a2 (mE lF ) + ab(nE lG) + b2 (nF mG)
b
= a (a2 m + abn)E + (b2 n a2 l)F (abl + mb2 )G
a
b
= (am + bn)
a
= al + bm.
Therefore, equations (6.11) and (6.12) are satisfied. ⇤

Proposition 6.12. Suppose that the critical set ⌃v is a smooth curve at p.

Then its tangent direction at p is the conjugate direction to v with respect
to the second fundamental form at p.

Proof. Following the proof of Proposition 6.11, the tangent direction to

the contour generator is parallel to w = (am + bn)xu1 + (al + bm)xu2 .
This is precisely the conjugate direction to v as hWp (w), vi = 0. ⇤
We give below an alternative proof to Koenderink’s Theorem [Koen-
derink (1984)].

Theorem 6.4 (Koenderink’s Theorem). Suppose that the apparent

contour v is a smooth curve at Pv (p). Then the Gaussian curvature of M
at p is equal to the product of the curvature of the apparent contour together
with the curvature of the normal section of M at p along the direction v.

Proof. The curvature of the normal section is just the normal curvature
of the surface M at p along the tangent direction v and is given by
IIp (v)
n (v) = .
||v||2
Let ↵(s) be an arc length parametrisation of the contour generator ⌃v .
Then, by Proposition 6.12,
v̄
↵0 = t =
||v̄||
00
↵ = g t ⇥ N + n (v̄)N.
The apparent contour v is parametrised by (s) = Pv (↵(s)). The
projection Pv for v fixed is a linear map R3 ! Tv S 2 , so
0
= dPv (↵0 ) = ↵0 h↵0 , viv,
00
= dPv (↵00 ) = ↵00 h↵00 , viv.
164 Di↵erential Geometry from a Singularity Theory Viewpoint

The vector N (s) is a unit normal vector to the curve v at (s), and
0
(s), N (s) form a positively oriented frame. If we denote 0? the vector
obtained by rotating 0 anti-clockwise by ⇡/2, then 0? = || 0 ||N , so the
curvature ( v ) of the apparent contour is given by
0?
h , 00 i h|| 0 ||N, 00
i hN, ↵00 i n (v̄)
( v) = = = = .
|| 0 ||3 || 0 ||3 || 0 ||2 || 0 ||2
We have
0 v̄ v̄ v̄ v v 1
= dPv ( )= h , i = 2
(||v||2 v̄ hv̄, viv),
||v̄|| ||v̄|| ||v̄|| ||v|| ||v|| ||v|| ||v̄||
so that
1
|| 0 ||2 = 2 2
(||v||2 ||v̄||2 hv̄, vi2 )
||v|| ||v̄||

1
= (EG F 2 )IIp (v)2 (by Proposition 6.2(iv))
||v||2 ||v̄||2
✓ ◆
||v||2 2 IIp (v)2
= (EG F )
||v̄||2 ||v||4

||v||2
= (EG F 2 )n (v)2 .
||v̄||2
Therefore,
||v̄||2 n (v̄)
( v ) = .
||v||2 (EG F 2 )n (v)2
We obtain by substituting n (v̄) by its expression from Proposition 6.2 that
✓ ◆
||v̄||2 ||v||2 K(p)
( v ) = 2 2 2 2
(nl m2 )n (v) = .
||v|| (EG F )n (v) ||v̄|| n (v)
Consequently,

K(p) = n (v)( v ). ⇤
We have the following consequences of Koenderink’s theorem.

Corollary 6.2. (i) The direction of the projection v is a principal direction

if and only if the curvature of the apparent contour is equal to the principal
curvature associated to the other principal direction.
(ii) The apparent contour v has an inflection if and only if the corre-
sponding point on the surface is a parabolic point and v is not an asymptotic
direction.
 Surfaces in the Euclidean 3-space 165

6.4.2 The generic singularities of orthogonal projections

Theorem 6.5. There is an open and dense set O2 of immersions x : U !

R3 such that for any x 2 O2 , the surface M = x(U ) has the following
properties. For any v 2 S 2 , the orthogonal projection Pv along v has only
A-singularities of Ae -codimension  2 (Table 6.1) at any point p on M .
Furthermore, these singularities are Ae -versally unfolded by the family P .

Proof. The proof is a consequence of Theorems 4.4 and 4.12. ⇤

Definition 6.4. A surface is called (locally) projection generic if any of its

local parametrisations belongs to the set O2 in Theorem 6.5.

We shall identify geometrically the conditions for a given local generic

singularity of Pv to occur. The conditions for having a given singularity
have simpler expressions when considering the following setting. We take
the surface M in Monge-form (x, y) = (x, y, f (x, y)) at a point p considered
to be the origin in R3 . We can rotate the coordinate axes if necessary and
set v = (0, 1, 0). Observe that there is no loss of generality with this setting
as the choice of a coordinate system in R3 is arbitrary. With the above
setting, we have
Pv (x, y) = (x, f (x, y)).
When considering the family of orthogonal projections P , the computa-
tions also simplify considerably if we modify P as follows. We parametrise
the directions near v = (0, 1, 0) by (↵, 1, ), with ↵ and close to zero,
and project to the fixed plane ⇡ : y = 0. The projection of the point
(x, y, f (x, y)) along (↵, 1, ) is the point (x, y, f (x, y)) + (↵, 1, ) 2 ⇡,
which implies that = y. We obtain the family of (germs of) projections
P̄ : (R2 ⇥R2 , (0, 0)) ! (R2 , 0), given by P̄ (x, y, ↵, ) = (x ↵y, f (x, y) y).
The change of coordinates (x, y) 7! (x ↵y, y) transforms the family P̄ to
the following family
P̃ (x, y, ↵, ) = (x, f (x + ↵y, y) y) (6.13)
which we call the modified family of projections. It is clear from the trans-
formations carried out above that the family P of orthogonal projections
on M is A-equivalent to the modified family of projections P̃ .
We need the following concept.

Definition 6.5. A point p on M is a flecnodal point if there is a tangent

line through p which has at least 4-point contact with M at p. Equivalently,
166 Di↵erential Geometry from a Singularity Theory Viewpoint

p is a flecnodal point if it is in the closure of the set of points where the

projection along an asymptotic direction has a swallowtail singularity, i.e.,
it is A-equivalent to (x, xy + y 4 ). The flecnodal set of M is the set flecnodal
points.

Theorem 6.6. The following holds for M in Monge form z = f (x, y) at p

with f as in (6.4) and v = (0, 1, 0).
(i) The flecnodal set (in the parameter space) consists locally of points
(x, y) for which there exist ↵ and near zero such that
(↵fx + fy )(x + ↵y, y) = 0,
(↵2 fxx + 2↵fxy + fyy )(x + ↵y, y) = 0,
(↵3 fxxx + 3↵2 fxxy + 3↵fxyy + fyyy )(x + ↵y, y) = 0.
In particular, the origin p is a flecnodal point if and only if
fy = fyy = fyyy = 0, fxy fyyyy 6= 0 at p,
that is,
a22 = a33 = 0 and a21 a44 6= 0.
(ii) If p is a flecnodal point, then the 1-jet of the equation of the flecnodal
set is given by
a232
6(a43 )x + 24a44 y.
a21
Proof. (i) The swallowtail singularities of the projection Pv are A-
invariant, so we can use the modified family of projections (6.13).
For ↵, fixed, we have a map-germ P̃(↵, ) (x, y) = (x, f (x + ↵y, y) y)
from the plane to the plane. We can now use Saji’s criteria for recognition
of the swallowtail singularities of P̃(↵, ) ([Saji (2010)]). The advantage
of using the modified family of projections is that the critical set ⌃↵, of
P̃(↵, ) is given by
@
(f (x + ↵y, y) y) = 0,
@y
and the kernel of dP̃(↵, ) at the singular points is along the constant di-
rection (0, 1). Thus, by Theorem 3 in [Saji (2010)], (x, y) is a swallowtail
singularity of P̃(↵, ) if and only if its critical set ⌃↵, is a smooth curve,
that is,
(fxy (0, 0), fyy (0, 0)) = (a21 , 2a22 ) 6= (0, 0),
 Surfaces in the Euclidean 3-space 167

and has 4-point contact at (x, y) with the kernel of dP̃(↵, ) , that is,
@
@y (f (x + ↵y, y) y) = 0,

@2
@y 2 (f (x + ↵y, y) y) = 0,

@3
@y 3 (f (x + ↵y, y) y) = 0,

@4
@y 4 (f (x + ↵y, y) y) 6= 0.

The last inequality is satisfied locally at p if and only if

@4
(f (x + ↵y, y) y)|(0,0,0,0) = fyyyy (0, 0) = 24a44 6= 0.
@y 4
The first three equations are equivalent to

(↵fx + fy )(x + ↵y, y) = 0, (6.14)

(↵2 fxx + 2↵fxy + fyy )(x + ↵y, y) = 0, (6.15)

(↵3 fxxx + 3↵2 fxxy + 3↵fxyy + fyyy )(x + ↵y, y) = 0. (6.16)

(ii) The above three equations determine the flecnodal set. Evaluating
at ↵ = = 0, we get that the origin p is a flecnodal point if and only if fy =
fyy = fyyy = 0, at (0, 0), equivalently, a22 = a33 = 0, and a21 6= 0, a44 6= 0.
(As a22 = 0, the condition for ⌃↵, to be a smooth curve becomes a21 6= 0.)
We can use equation (6.14) to obtain as a function of (x, y, ↵). Sub-
stituting in equation (6.15), we can solve for ↵ as fxy (0, 0) = a21 6= 0.
We obtain ↵ as a function of (x, y) and substituting in equation (6.16),
we obtain an equation for the flecnodal set. The 1-jet of this equation is
a2
6(a43 a32 21
)x + 24a44 y. ⇤

Theorem 6.7. Suppose that M is given in Monge form z = f (x, y) at

the origin p with f as in (6.4) and suppose that the direction of projection
is parallel to v = (0, 1, 0). Then the conditions for Pv to have a generic
singularity at p are given in Table 6.1 in terms of the coefficients of the
Taylor expansion of f at p.
The geometric characterisation of the generic singularities of Pv are
as in Table 6.2. (In Table 6.2 is given only the geometric meaning of the
conditions in Table 6.1 that are equal to zero.)
168 Di↵erential Geometry from a Singularity Theory Viewpoint

Table 6.1 Algebraic conditions for the local singularities of Pv .

Name Normal form Algebraic conditions
Fold (x, y 2 ) a22 6= 0
Cusp (x, xy + y 3 ) a22 = 0, a21 6= 0, a33 6= 0
Lips/beaks (x, y 3 ± x2 y) a22 = 0, a21 = 0, a33 6= 0, a232 3a31 a33 6= 0
Goose (x, y 3 + x3 y) a22 = 0, a21 = 0, a232 3a31 a33 = 0, a33 6= 0,
27a41 a333 18a42 a32 a233 + 9a43 a232 a33 4a44 a332 6= 0
Swallowtail (x, xy + y 4 ) a22 = 0, a33 = 0, a21 6= 0, a44 6= 0
Butterfly (x, xy + y 5 ± y 7 ) a22 = 0, a33 = 0, a44 = 0, a21 6= 0, a55 6= 0,
(8a55 a77 5a266 )a221 + 2a55 (a32 a66 20a43 a55 )a21 +
35a232 a255 6= 0
Gulls (x, xy 2 + y 4 + y 5 ) a22 = 0, a21 = 0, a33 = 0, a32 6= 0, a44 6= 0,
a55 a232 2a43 a44 a32 + 4a31 a244 6= 0

Table 6.2 Geometric characterisation of the local singularities of Pv ([Ga↵ney

(1983)]).
Name Geometric characterisation
Fold v tangent to M at p
Cusp p a hyperbolic point, v an asymptotic direction at p
Lips/beaks p a parabolic point, v an asymptotic direction at p
Goose p a parabolic point, v an asymptotic direction at p,
the Gauss image of the parabolic set has a geodesic inflection
Swallowtail p a flecnodal point, v an asymptotic direction at p
Butterfly p a flecnodal point, v an asymptotic direction at p and
tangent to the flecnodal curve at p
Gulls p a cusp of Gauss, v an asymptotic direction

Proof. If v 2 / Tp M , Pv is a local di↵eomorphism, so is locally A-

equivalent to the germ (x, y). We consider from now on the case when
v 2 Tp M and take M in Monge form z = f (x, y) at p with f as in (6.4)
and v = (0, 1, 0). Then Pv (x, y) = (x, f (x, y)).
The A-classification of map-germs (R2 , 0) ! (R2 , 0) of Ae -codimension
 2 is given in [Ga↵ney (1983)]. This is extended in [Rieger (1987)] to
cover the germs of Ae -codimension  6. In [Rieger (1987)], the germs are
also taken in the form (x, f (x, y)) and the work there is about determin-
ing the A-classes of such germs. The problem here is a recognition one:
given a germ of the form Pv (x, y) = (x, f (x, y)), find the conditions on the
coefficients of the Taylor expansion of f for Pv to be A-equivalent to one
of the germs in Table 6.1. We shall use of course the fact that when the
singularity of Pv is k-A-determined we can work with the k-jet of Pv and
ignore higher degree terms in the Taylor expansion of f .
 Surfaces in the Euclidean 3-space 169

All monomials of the form (0, ai0 xi ) i = 0, . . . , k in j k Pv can be elim-

inated by a change of coordinates in the target of the form (U, V ) 7!
(U, V ai0 U i ), so j k Pv (x, y) ⇠A (x, j k f (x, y) j k f (x, 0)). To avoid repe-
tition, we shall carry out these changes of coordinates without mentioning
them.
The singular set ⌃v of Pv is the zero set of
fy (x, y) = a21 x + 2a22 y + a31 x2 + 2a32 xy + 3a33 y 2 + O(3).
It is a smooth curve at the origin if and only a21 6= 0 or a22 6= 0.
• Fold
The fold singularity (x, y 2 ) is 2-A-determined, so consider
j 2 Pv (x, y) ⇠A(2) (x, a21 xy + a22 y 2 ).
If a22 6= 0, the change of coordinates (x, y) 7! (x, y a21 /(2a22 )x)
removes the term a21 xy in the second component of j 2 Pv . Re-scaling gives
j 2 Pv ⇠A(2) (x, y 2 ), hence Pv ⇠A (x, y 2 ).
If a22 = 0, then j 2 Pv is A(2) -equivalent to (x, xy) if a21 6= 0 and to
(x, 0) if a21 = 0. In both cases j 2 Pv cannot be A-equivalent to the fold
singularity. Therefore, the singularity of Pv at p is a fold if and only if
a22 6= 0, equivalently, v is transverse to ⌃v at p.
• Cusp
We have j 2 Pv ⇠A(2) (x, xy) if and only if a22 = 0 and a21 6= 0, equiv-
alently, v is an asymptotic direction at p and p is not a parabolic point
(so it must be a hyperbolic point). If furthermore a33 6= 0, then j 3 Pv and
hence Pv is A-equivalent to the cusp singularity (x, xy + y 3 ). The condition
a33 6= 0 means that Pv ⇠K (x, y 3 ), that is v has 2-point contact with ⌃v .
• Swallowtail
If a22 = a33 = 0 and a21 6= 0, we have
j 4 Pv ⇠A(4) (x, a21 x(y + g(x, y)) + a44 y 4 ),
where g has no constant or linear terms. The change of coordinates (x, y) 7!
(x, y + g(x, y)) gives j 4 Pv ⇠A(4) (x, a21 xy + a44 y 4 ). If a44 6= 0, then j 4 Pv
and hence is A-equivalent to the swallowtail singularity (x, xy + y 4 ). In
particular, p is a flecnodal point (Definition 6.5).
• Butterfly
Suppose now that a22 = a33 = a44 = 0 and a21 6= 0. By Theorem
6.6(ii), this means that the direction v is tangent to the flecnodal set when
this is a smooth curve (which is the case for a generic surface, see Theorem
6.8 and Table 6.3). Following the calculation for the Swallowtail singularity,
j 7 Pv is A(7) -equivalent to
F1 (x, y) = (x, a21 x(y + g(x, y)) + a55 y 5 + a66 y 6 + a77 y 7 ),
170 Di↵erential Geometry from a Singularity Theory Viewpoint

where g is a polynomial with no constant or linear terms. If a55 6= 0, then

F1 ⇠A (x, xy + y 5 ± y 7 ) or to (x, xy + y 5 ). We need to make appropriate
changes of coordinates and find the condition for F1 , and hence for Pv , to
be A-equivalent to (xy + y 5 ± y 7 ).
The calculation can be carried out with the help of a computer al-
gebra package (such as Maple or Matematica). We first eliminate the
term a21 xg(x, y) in the second component of F1 by a change of coordi-
nate of the form (x, y) 7! (x, y + h(x, y)), where h is a polynomial of degree
6 with no linear terms. To obtain h we proceed as follows. We write
h(x, y) = h2 (x, y) + . . . + h6 (x, y), with
i
X
hi (x, y) = hij xi j j
y
j=0

and consider F1 (x, y + h(x, y)). The 3-jet of F1 (x, y + h(x, y)) is given by

(x, a21 xy + a21 h20 x3 + (a21 h21 + a31 )x2 y + (a21 h22 + a32 )xy 2 )

and equating the coefficients of (0, x3 ), (0, x2 y) and (0, xy 2 ) to zero gives
h20 = 0, h21 = a31 /a21 and h22 = a32 /a21 . This determines completely
the quadratic part h2 (x, y) of h(x, y).
We proceed inductively on the jet level of F1 (x, y +h(x, y)) to determine
completely h, and this gives F1 (x, y + h(x, y)), and hence j 7 Pv (x, y), A(7) -
equivalent to

F2 (x, y) = (x, a21 xy + a55 y 5 + ↵1 y 6 + ↵2 y 7 ),

with
1
↵1 = a21 (a21 a66 5a55 a32 ),
1
1 = a221
(a77 a221 6a66 a32 a21 5a21 a55 a43 + 20a55 a232 ).

We need to eliminate the term (0, y 6 ) in F2 . For this we make the

change of coordinate (U, V ) 7! k(U, V ) = (U V, V ) in the target and set
X=x (a21 xy + a55 y + ↵1 y + 1 y ), so that x = (X + (a55 y 5 + ↵1 y 6 +
5 6 7
7
1 y ))/(1 a21 y). This gives j 7 (k F2 )(X, y) equivalent to a germ of the
form

j 7 (k F2 ) ⇠A(7) (X, a21 X(y + G(y))) + a55 y 5 + 2y

6
+ 3y
7
).

We now make a change of coordinate in the source of the form H(X, y) =

(X, y + h2 y 2 + h3 y 3 + h4 y 4 + h5 y 5 + h6 y 6 ). The coefficients of (0, Xy 2 ) and
(0, y 6 ) in j 7 (k F2 H) are, respectively, a21 ( a21 h2 ) and a55 a21 +
 Surfaces in the Euclidean 3-space 171

5a55 h2 + (a21 a66 5a55 a32 )/a21 . Setting these to be zero and solving the
linear system in and h2 gives
a21 a66 5a55 a32 a21 a66 5a55 a32
= 2 , h2 = .
4a55 a21 4a55 a21
We then equate the coefficients of (0, Xy i ) in j 7 (k F2 H) to zero to get
h3 , . . . , h6 . As a result, j 7 (k F2 H), and hence j 7 Pv , is A(7) -equivalent
to (x, a21 xy + a55 y 5 + ⇤y 7 ) with
(8a55 a77 5a266 )a221 + 2a55 (a32 a66 20a43 a55 )a21 + 35a232 a255
⇤= .
8a221 a55
The singularity of Pv at p is a Butterfly if and only if ⇤ 6= 0.
• Lips/beaks
For the lips/beaks singularity which is 3-A-determined, we consider
j 3 Pv ⇠A(3) (x, a21 xy + a22 y 2 + a31 x2 y + a32 xy 2 + a33 y 3 ).
For Pv to have a lips/beaks singularity it is necessary that a21 = a22 = 0
and a33 6= 0. From Proposition 6.5 and Theorem 6.2, these conditions mean
that v is an asymptotic direction and p is a parabolic point but not a cusp
a32
of Gauss. Then the change of coordinates (x, y) 7! (x, y 3a 33
x) in the
source and a change of coordinate in the target yields
3a31 a33 a232 2
j 3 Pv ⇠A(3) (x, x y + a33 y 3 ).
3a33
This is equivalent to a lips/beaks singularity if and only if 3a31 a33 a232 6= 0.
By Proposition 6.10, 3a31 a33 a232 = 0 if and only if the geodesic curvature
of the image of the parabolic curve by the Gauss map vanishes at N (p).
• Goose
Suppose that a21 = a22 = 3a31 a33 a232 = 0 and a33 6= 0. Thus, p is a
parabolic point, v is an asymptotic direction and the geodesic curvature of
the image of the parabolic curve by the Gauss map is zero at N (p). The
coordinate changes for the lips/beaks give
j 4 Pv ⇠A(4) F3 = (x, a33 y 3 + ↵1 x3 y + ↵2 x2 y 2 + ↵3 xy 3 + ↵4 y 4 )
with
1
↵1 = (27a41 a333 18a42 a32 a233 + 9a43 a232 a33 4a44 a332 )
27a333
and ↵2 , ↵3 , ↵4 are constants depending on the coefficients aij . The terms
(0, ↵2 x2 y 2 +↵3 xy 3 +↵4 y 4 ) in F3 can be eliminated by a change of coordinate
of the form (x, y) 7! (x, y +g(x, y)) with g a polynomial with no constant or
172 Di↵erential Geometry from a Singularity Theory Viewpoint

linear terms. This change of coordinate does not alter ↵1 , so j 4 Pv is A(4) -

equivalent (x, a33 y 3 + ↵1 x3 y), which is 4-A-determined and is the Goose
singularity if and only if ↵1 6= 0.
• Gulls
Suppose that a21 = a22 = a33 = 0 and a32 6= 0. Thus, p is a cusp
of Gauss and v is an asymptotic direction. (The condition a32 6= 0 means
that the geodesic curvature of the image of the parabolic curve by the Gauss
map is not zero at N (p), see Proposition 6.10).
The Gulls singularity is 5-A-determined, so we shall restrict to the 5-jet
space and make changes of coordinate to put the 3-jet, the 4-jet and then
the 5-jet of Pv to a simple form. Observe, as before, that the terms of the
form (0, ↵k xk ) in the Taylor expansion of Pv can be eliminated by changes
of coordinates in the target and will be ignored.
With the above conditions on the coefficients of f , j 3 Pv ⇠A(3)
(x, x(a31 xy + a32 y 2 )). The change of coordinates (x, y) 7! (x, y
a31 /(2a32 )x) in the source eliminates the term (0, a31 x2 y) in the Taylor
expansion of j 5 Pv . We get

j 5 Pv ⇠A(5) (x, a32 xy 2 + ↵1 x3 y + ↵2 x2 y 2 + ↵3 xy 3 + xg1 (x, y) + a44 y 4 + a55 y 5 )

⇠A(5) (x, x(a32 y 2 + y(↵1 x2 + ↵2 xy + ↵3 y 2 )) + xg1 (x, y) + a44 y 4 + a55 y 5 )
with ↵i , i = 1, 2, 3 depending on the coefficients of f and g1 (x, y) is a
homogeneous polynomial of degree 4. The change of coordinates (x, y) 7!
(x, y 12 (↵1 x2 + ↵2 xy + ↵3 y 2 )) shows that j 5 Pv is A(5) -equivalent to
1
(x, a32 xy 2 + a44 y 4 + 2 (a55 a232 2a43 a44 a32 + 4a31 a244 )y 5 + xg2 (x, y))
a32
with g2 (x, y) a homogeneous polynomial of degree 4. Similar change of coor-
dinates carried out to clear the 4-jet can be done to get rid of (0, xg2 (x, y))
and this change of coordinates does not alter the coefficients of (0, y 5 ).
Therefore,
1
j 5 Pv ⇠A(5) (x, a32 xy 2 + a44 y 4 + 2 (a55 a232 2a43 a44 a32 + 4a31 a244 )y 5 ),
a32
and it is 5-determined and is a Gulls singularity if and only if a44 6= 0 and
a55 a232 2a43 a44 a32 + 4a31 a244 6= 0. ⇤

Theorem 6.7 is about the singularities of the orthogonal projection Pv

along the fixed direction v. If the family of orthogonal projection P is an
Ae -versal unfolding of the singularity of Pv , one can describe completely
the bifurcations in the apparent contour v as v varies locally in S 2 .
 Surfaces in the Euclidean 3-space 173

Theorem 6.8. Let M be given in Monge form z = f (x, y) at the origin

p with f as in (6.4) and v = (0, 1, 0) 2 Tp M . Then the conditions for
the family of orthogonal projections to be an Ae -versal unfolding of the
singularities in Table 6.1 are as in Table 6.3 (where it is assumed that
the algebraic conditions in Table 6.1 are satisfied for each given singularity
type).

Table 6.3 The conditions for the family of projections to be a versal unfolding.
Name Algebraic conditions Geometric interpretation
Fold – Always
Cusp – Always
Lips/beaks – Always
Swallowtail – Always
Goose a20 6= 0 p is not a flat umbilic
Gulls a232 4a20 a44 6= 0 The image of the parabolic set by the Gauss map
is a curve with a (2, 3)-cusp at N (p)
Butterfly a232 a21 a43 6= 0 The flecnodal curve is not singular

Proof. We consider the modified family of projections (6.13) given by

P̃ (x, y, ↵, ) = (x, f (x + ↵y, y) y)
and denote by P̃0 the map-germ P̃0 (x, y) = P̃ (x, y, 0, 0) = Pv (x, y). The
conditions for P and P̃ to be Ae -versal families of a given singularity of P̃0
are the same as the two families A-equivalent.
According to Theorem 3.6, the family P̃ is an Ae -versal unfolding of the
singularity of P̃0 if and only if
n o
˙ ˙
LAe · P̃0 + R. P̃1 , P̃2 = E(2, 2),

where
P̃˙ 1 (x, y) = P̃↵ (x, y, 0, 0) = (0, yfx (x, y)),
P̃˙ 2 (x, y) = P̃ (x, y, 0, 0) = (0, y).
The fold and cusp are stable singularities so are versally unfolded
by any family. We consider the remaining singularities in Table 6.1.
As these singularities are finitely k-A-determined for some k, we have
Mk+1
2 .E(2, 2) ⇢ LA · P̃0 , so to prove that P̃ is an Ae -versal unfolding of
the singularity of P̃0 we only need to show that
⇣ n o⌘
j k LAe · P̃0 + R. P̃˙ 1 , P̃˙ 2 = J k (2, 2). (6.17)
174 Di↵erential Geometry from a Singularity Theory Viewpoint

We shall work downwards on jet levels (as in the proof of Theorem 6.2)
and start by showing that all monomials (xi y j , 0) and (0, xi y j ) of degree
k are in the left hand side of (6.17). We have @@x P̃0
(x, y) = (1, fx (x, y)),
thus j k (Q @@x
P̃0
) = (Q(x, y), 0) is in the left hand side of (6.17) for any
monomial Q(x, y) of degree k. Therefore, we only need to consider the
monomials (0, xi y j ) of degree k. Once we got these, we can work modulo
Mk2 .E(2, 2) and consider the monomials of degree k 1. Again we only need
to consider monomials of the form (0, Q(x, y)). We observe that, using the
first component of P̃ , the monomials (xi , 0) and (0, xi ) are in LLe · P̃0 and
hence in the left hand side of (6.17). We also get the monomial (0, y) from
P̃˙ .
2

• Swallowtail
The swallowtail singularity is 4-A-determined, so we need to show that
(6.17) holds for k = 4. We have
j 3 ( @@x
P̃0
) = (1, 2a20 x + a21 y + 3a30 x2 + 2a31 xy + a32 y 2
+4a40 x3 + 3a41 x2 y + 2a42 xy 2 + a43 y 3 )
j 3 ( @@y
P̃0
) = (0, a21 x + a31 x2 + 2a32 xy
+a41 x3 + 2a42 x2 y + 3a43 xy 2 + 4a44 y 3 )
We get all the monomial of degree 4 of the form (0, xQ1 (x, y)) using
j (Q1 @@y
4 P̃0
), where Q1 is of degree 3. From this we also get the degree 3
monomial (0, x2 y) using j 4 (x2 @@y
P̃0
).
Working modulo these monomials, we have
⇠1 = j 3 (y @@y
P̃0
) ⌘ (0, a21 xy + a32 xy 2 )
⇠2 = j 3 (0, f (x, y)) ⌘ (0, a21 xy + a32 xy 2 + a44 y 4 )
so that ⇠2 ⇠1 ⌘ (0, a44 y 4 ). As a44 6= 0, we get (0, y 4 ).
Now j 3 (y 2 @@y
P̃0
) ⌘ (0, a21 xy 2 ) which gives (0, xy 2 ) as a21 6= 0. Similarly,
j 3 (x @@x
P̃0
) ⌘ (0, a21 xy) which gives (0, xy).
We get (0, y 3 ) from j 3 ( @@y
P̃0
) and (0, y 2 ) from P̃˙ 1 . Therefore (6.17) holds
without additional conditions on the coefficients of f . Thus P is always an
Ae -versal unfolding of the swallowtail singularity.
• Lips/beaks
The lips/beaks singularity is 3-A-determined, so we need to show that
(6.17) holds for k = 3. In this case we have
j 2 ( @@x
P̃0
(x, y)) = (1, 2a20 x + 3a30 x2 + 2a31 xy + a32 y 2 )
j 2 ( @@y
P̃0
(x, y)) = (0, a31 x2 + 2a32 xy + 3a33 y 2 ).
 Surfaces in the Euclidean 3-space 175

We write the two vectors ⇠1 = j 3 (x @@x

P̃0
) ⌘ (0, 2a31 x2 y + a32 xy 2 ) and
⇠2 = j 3 (x @@y
P̃0
) ⌘ (0, 2a32 x2 y + 3a33 xy 2 ) in matrix form
✓ ◆ ✓ ◆✓ ◆
⇠1 2a31 a32 (0, x2 y)
= .
⇠2 2a32 3a33 (0, xy 2 )
The determinant of the 2 ⇥ 2 matrix in the above equality is 2(a232
3a31 a33 ) and does not vanish as the singularity is a lips/beaks (see Table
6.1). Therefore we get (0, x2 y) and (0, xy 2 ). We can work now modulo
these monomials to get (0, y 3 ) from j 3 (y @@y P̃0
) as a33 6= 0.
A similar argument to the above shows that we can get (0, xy) and
(0, y 2 ) from the two vectors j 3 ( @@x
P̃0
) ⌘ (0, 2a31 xy + a32 y 2 ) and j 3 ( @@y
P̃0
)⌘
2
(0, 2a32 xy + 3a33 y ). Therefore (6.17) holds without additional conditions
on the coefficients of f , and this shows that P is always an Ae -versal un-
folding of the lips/beaks singularity.
• Goose
The goose singularity is 4-A-determined, so we need to show that (6.17)
holds for k = 4. We have
j 3 ( @@x
P̃0
(x, y)) = (1, 2a20 x + 3a30 x2 + 2a31 xy + a32 y 2
+4a40 x3 + 3a41 x2 y + 2a42 xy 2 + a43 y 3 )
j 3 ( @@y
P̃0
(x, y)) = (0, a31 x2 + 2a32 xy + 3a33 y 2
+a41 x3 + 2a42 x2 y + 3a43 xy 2 + 4a44 y 3 )
with a31 = a232 /(3a33 ).
The following vectors
j 4 (x2 @@y
P̃0
) ⌘ (0, 2a32 x3 y + 3a33 x2 y 2 )
j 4 (xy @@y
P̃0
) ⌘ (0, a31 x3 y + 2a32 x2 y 2 + 3a33 xy 3 )
j 4 (a32 x @@x
P̃0
a31 x @@y
P̃0
) ⌘ (0, (3a41 a32 2a42 a31 )x3 y
+(2a42 a32 3a43 a31 )x2 y 2 + (a32 a43 4a44 a32 )xy 3 )

generate a 3 ⇥ 3 matrix with determinant

1
a32 (27a41 a333 18a42 a32 a233 + 9a43 a232 a33 4a44 a332 ).
a33
The third component of the above determinant does not vanish as the
singularity is a goose (Table 6.1). If a32 6= 0, the determinant does not
vanish and we get (0, x3 y), (0, x2 y 2 ), (0, xy 3 ), and using j 4 (y 2 @@y
P̃0
) we also
4
get (0, y ) as a33 6= 0. If a32 = 0, then a31 = 0 (condition for a Goose
singularity; see Table 6.1) and we get (0, x2 y 2 ), (0, xy 3 ) and (0, y 4 ) from
respectively j 4 (x2 @@y
P̃0
), j 4 (xy @@y
P̃0
) and j 4 (y 2 @@y
P̃0
). We can get now the
176 Di↵erential Geometry from a Singularity Theory Viewpoint

monomial (0, x3 y) from j 4 (x @@x P̃0

) as a41 6= 0 (from the Goose condition).
Therefore we get all the monomials of degree 4 without imposing any extra
condition on the coefficients of f .
The calculation for the degree 3 monomials follows similarly. Using the
vectors j 3 (x @@y
P̃0
), j 3 (y @@y
P̃0
) and j 3 (a32 @@x
P̃0
a31 @@y
P̃0
) we get the monomials
(0, x y), (0, xy ), (0, y ) if a32 6= 0. If a32 = 0, we get (0, xy 2 ) and (0, y 3 )
2 2 3

from respectively j 3 (x @@y P̃0

) and j 3 (y @@y
P̃0
). We subsequently get (0, x2 y)
from j 3 ( @@x
P̃0
). Therefore we get all the monomial of degree 3 without im-
posing any extra condition on the coefficients of f .
For the monomials of degree 2, the vectors j 2 ( @@x
P̃0
) ' (0, 2a31 xy +a32 y 2 )
and j 2 ( @@y
P̃0
) ' (0, 2a32 xy + 3a33 y 2 ) are linearly dependent and the only
other possibility to get monomials of degree 2 is from the vector j 2 (P̃˙ ) = 1
(0, 2a20 xy), which gives the monomial (0, xy) if and only if a20 6= 0. We
can then get (0, y 2 ) from j 2 ( @@y
P̃0
) as a33 6= 0. Therefore, the family P is
an Ae -versal unfolding of the Goose singularity if and only if a20 6= 0. As
a21 = a22 = 0, the condition a20 = 0 means that f has an identically zero
quadratic part, that is, the point p is a flat umbilic point.

• Gulls
Here we take k = 5 in (6.17) as the gulls singularity is 5-A-determined.
We have

j 4 ( @@x
P̃0
(x, y)) = (1, 2a20 x + 3a30 x2 + 2a31 xy + a32 y 2
+4a40 x3 + 3a41 x2 y + 2a42 xy 2 + a43 y 3
+5a50 x4 + 4a51 x3 y + 3a52 x2 y 2 + 2a43 xy 3 + a54 y 4 )
j 4 ( @@y
P̃0
(x, y)) = (0, a31 x2 + 2a32 xy
+a41 x3 + 2a42 x2 y + 3a43 xy 2 + 4a44 y 3
+a51 x4 + 2a52 x3 y + 3a43 x2 y 2 + 4a54 xy 3 + 5a55 y 4 )

The two vectors j 5 (x3 @@x

P̃0
) ⌘ (0, 2a31 x4 y + a32 x3 y 2 ) and j 5 (x2 y @@y
P̃0
)⌘
4 3 2 3 2 4
(0, a31 x y + 2a32 x y ) give (0, x y ) and (0, x y) if a31 6= 0. Then using
@ P̃0
@y we get the remaining monomials of degree 5 of the form (0, xyQ(x, y)).
If a31 = 0, we get all the monomials of degree 5 of the form (0, xyQ(x, y))
using j 5 (Q @@y
P̃0
) as a32 6= 0. We can get similarly the monomials (0, x3 y)
 Surfaces in the Euclidean 3-space 177

and (0, x2 y 2 ). The vectors

j 5 (y 2 @@y
P̃0
) ⌘ (0, 2a32 xy 3 + 4a44 y 5 )
j 5 (y @@y
P̃0
) ⌘ (0, a31 x2 y + 2a32 xy 2 + 3a43 xy 3 + 4a44 y 4 + 5a55 y 5 )
j 5 (x @@y
P̃0
) ⌘ (0, 2a32 x2 y + 4a44 xy 3 )
j 5 (x @@x
P̃0
) ⌘ (0, 2a31 x2 y + a32 xy 2 + a43 xy 3 )
j 5 (0, f (x, y)) ⌘ (0, a31 x2 y + a32 xy 2 + a43 xy 3 + a44 y 4 + a55 y 5 )
generate a 5 ⇥ 5 matrix with determinant
4a32 a44 (a55 a232 2a43 a44 a32 + 4a31 a244 )
which does not vanish at a gulls singularity. Therefore we get the monomials
(0, x2 y), (0, xy 2 ), (0, xy 3 ), (0, y 4 ) and (0, y 5 ). This shows that all monomials
of degree 5 and 4 are in the left hand side of (6.17) and the only monomial
of degree 3 still to get is (0, y 3 ). The only way to get this monomial is by
using the following vectors
j 3 ( @@y
P̃0
) ⌘ (0, 2a32 xy + 4a44 y 3 )
j 3 ( @@x
P̃0
) ⌘ (0, 2a31 xy + a32 y 2 + a43 y 3 )
j 3 (P̃˙ 1 ) ⌘ (0, 2a20 xy + a32 y 3 )
which generate a matrix with determinant 2a32 (a232 4a20 a44 ). Thus, we
require the additional condition a232 4a20 a44 6= 0 for P to be an Ae -versal
unfolding of the gulls singularity. From Theorem 6.2, this is the condition
for the height function along the normal direction to have precisely an A3 -
singularity at p, equivalently, the image of the parabolic set by the Gauss
map has a cusp singularity A-equivalent to (t2 , t3 ).
• Butterfly
This singularity is 7-A-determined, so we take k = 7 in (6.17).
For Q any monomial of degree 6, j 7 (Q @@y
P̃0
) = (0, a21 xQ(x, y)), so we
get all monomials of degree 7 of the form (0, xQ(x, y)). Using @@y
P̃0
again
and working modulo these monomials we get all monomials of the form
(0, xQ(x, y)) of degree 6 and 5 as well as the monomials (0, x3 y) and
(0, x2 y 2 ).
The following vectors
j 7 (y 3 @@y
P̃0
) ⌘ (0, a21 xy 3 + 5a55 y 7 )
j 7 (y 2 @@y
P̃0
) ⌘ (0, a21 xy 2 + 2a32 xy 3 + 5a55 y 6 + 6a66 y 7 )
j 7 (y @@y
P̃0
) ⌘ (0, a21 xy + 2a32 xy 2 + 3a43 xy 3 + 5a55 y 5 + 6a66 y 6 + 7a77 y 7 )
j 7 (x @@x
P̃0
) ⌘ (0, a21 xy + a32 xy 2 + a43 xy 3 )
j 7 (Q @@x
P̃0
(f, 0)) ⌘ (0, 2a32 a21 xy 3 + a221 xy 2 + a55 a21 y 6 + (a66 a21 + a55 a32 )y 7 )
j 7 (0, f ) ⌘ (0, a21 xy + a32 xy 2 + a43 xy 3 + a55 y 5 + a66 y 6 + a77 y 7 )
178 Di↵erential Geometry from a Singularity Theory Viewpoint

generate a matrix with determinant

a55 a221 (8a55 a77 5a266 )a221 + 2a55 (a32 a66 20a43 a55 )a21 + 35a232 a255 ,
which does not vanish as the singularity is of type butterfly. Therefore we
get (0, xy), (0, xy 2 ), (0, xy 3 ), (0, y 5 ), (0, y 6 ) and (0, y 7 ).
Now j 7 ( @@y
P̃0
) ⌘ (0, 5a55 y 4 ) so we get (0, y 4 ) as a55 6= 0. The only way
to get (0, y 2 ) and (0, y 3 ) is by using the following two vectors
j 7 (P̃˙ ) ⌘ (0, a y 2 + a y 3 ),
1 21 32
j 7 ( @@x
P̃0
P̃˙ 2 ) ⌘ (0, a32 xy 2 + a43 y 3 ).
These vectors generate a matrix with determinant a21 a43 a232 . Therefore,
the family P is an Ae -versal unfolding of the butterfly singularity if and
only if a21 a43 a232 6= 0. As a44 = 0, it follows from Theorem 6.6(ii) that
a21 a43 a232 6= 0 if and only if the flecnodal curve is not singular at p. ⇤

Remark 6.2. When the family P is a versal unfolding, the bifurcations in

the apparent contour can be described by considering the bifurcations in
the discriminants of the model families in Table 4.1. The figures for these
bifurcations can be found in [Gibson, Hawes and Hobbs (1994)].

6.5 Contact with spheres

Recall that the family of distance squared functions D : U ⇥ R3 ! R is

given by
D(u, a) = hx(u) a, x(u) ai
and the extended family of distance squared functions D̃ : U ⇥ R3 ⇥ R ! R
is given by
D̃(u, a, r) = hx(u) a, x(u) ai r2 .
The distance squared function da0 , for a0 fixed, measures the contact
of M with spheres of centre a0 . Observe that the functions da0 and d˜a0 ,r
have respectively the same R or K-singularity type.

Theorem 6.9. There is a residual set O3 of immersions x : U ! R3 such

that for any x 2 O3 , the surface M = x(U ) has the following properties.
For any a0 2 R3 , the distance squared function da0 (resp d˜a0 ) at any
point p on M has only local singularities of type A1 , A2 , A3 , A4 or D4 .
Furthermore, the singularities of da0 (resp. d˜a0 ) are R+ (resp. K) versally
unfolded by the family D (resp. D̃).
 Surfaces in the Euclidean 3-space 179

Proof. The result is a consequence of Theorems 4.4 and 4.8. ⇤

Definition 6.6. A surface is called (locally) distance squared function

generic if any of its local parametrisations belongs to the set O3 in Theorem
6.9.

When M is given in Monge form z = f (x, y), the conditions for da0 to
have one of the singularities in Proposition 6.9 and for the family D to be
a versal deformation of these singularities can be expressed in terms of the
coefficients of the Taylor expansion of f at the origin p. If p 2 M is not
an umbilic point, we can take the coordinate axes parallel to the principal
directions of M at p. We choose the x-axis in the direction of the principal
direction v1 and the y-axis in the direction of v2 . We can also rescale the
coordinates in the source if necessary, i.e., make a change of coordinates of
the form (x, y) 7! (↵x, y), so that the coefficients of the first fundamental
form at p are given by
E(0, 0) = 1, F (0, 0) = 0, G(0, 0) = 1.
A principal direction (a, b, 0) at p is a solution of the equation
b2 ab a2
E(0, 0) F (0, 0) G(0, 0) = 0,
l(0, 0) m(0, 0) n(0, 0)
that is,
m(0, 0)a2 + (n(0, 0) l(0, 0))ab + m(0, 0)b2 = 0.
As the principal directions at p are taken to be along (1, 0, 0) and
(0, 1, 0), it follows that m(0, 0) = 0, so fxy (0, 0) = 0. We have
l(0, 0) = fxx (0, 0), n(0, 0) = fyy (0, 0),
so the principal curvatures at the origin p are given by
1 (p) = fxx (0, 0), 2 (p) = fyy (0, 0).
Thus, at the origin p and with the above setting,
f (x, y) = a20 x2 + a22 y 2 + a30 x3 + a31 x2 y + a32 xy 2 + a33 y 3 + O(4) (6.18)
with 1 (p) = 2a20 and 2 (p) = 2a22 .
It follows from Proposition 2.6 that da0 is singular at the origin p if and
only if a0 = (0, 0, a3 ) with a3 some non-zero real number. That is, a is on
the normal line to M at p. The singularity is of type A1 if and only if a is
not on the focal set of M , that is, a3 6= 1/i (p), i = 1, 2.
180 Di↵erential Geometry from a Singularity Theory Viewpoint

We consider now the degenerate singularities da0 associated to the

sphere of centre a0 = (0, 0, 1/1 (p)) on the focal set F1 (we assume that
1 (p) 6= 0). The case when a0 = (0, 0, 1/2 (p)) follows similarly. (It requires
appropriate changes of indices of the coefficients of the Taylor expansion
f .)

Theorem 6.10. Let M be given in Monge form z = f (x, y) at the origin

p with f as in (6.18) and c0 = (0, 0, 1/1 (p)).
(i) The conditions for the distance squared function da0 to have one of
the degenerate singularities in Theorem 6.9 are as follows.
A2 : a30 6= 0;
A3 : a30 = 0, a231 4(a22 a20 )(a40 a320 ) 6= 0;
A4 : a30 = 0, a231 4(a22 a20 )(a40 a320 ) = 0,
a32 a231 2a41 a31 (a22 a20 ) + 4a50 (a20 a22 )2 6= 0;
D4 : a20 = a22 6= 0,
a231 a232 4a30 a332 4a33 a331 27a230 a233 + 18a30 a31 a32 a33 6= 0.
(ii) The family of distance squared functions on M is a R+ versal un-
folding of the singularities of da0 in (i) if and only if

A2 : always;
A3 : always;
A4 : 8a31 a420 8a31 a22 a320 4a51 a220 + 4(2a51 a22 a31 a42 )a20
3a231 a33 + 4a31 a42 a22 4a51 a222 6= 0;
D4 : 2a232 a231 + 3a31 a33 6= 0.

Proof. The proof follows the same steps as those of Theorem 6.2. We
consider only the singularities A4 and D4 .
We start with the A4 -singularity. We denote by dij the coefficient of
i j
x y in the Taylor expansion of da0 at the origin. Observe that
1 (p) 2 (p) a22 a20
d02 = = 6= 0
21 (p) 2a22
as p is not an umbilic point. The singularity is more degenerate than
an A3 , so a30 = 0 and a231 4(a22 a20 )(a40 a320 ) = 0. Then L =
d02 y 2 + d21 x2 y + d40 x4 is a perfect square.
We make the change of coordinates (x, Y ) = (x, y d21 /2(d02 )x2 ). Then
L(x, Y ) = d02 Y 2 and the singularity of da0 is of type A4 if and only if the
coefficient of x5 in the Taylor expansion of Da0 (x, Y ) is not zero. That
coefficient is not zero when the expression in the statement of the theorem
is not zero.
 Surfaces in the Euclidean 3-space 181

The family D, with a = (a1 , a2 , a3 ) varying near a0 , is an R+ - versal

deformation of the A4 -singularity of da0 if and only if
⇣ n o ⌘
j LRe · Da0 + R. Ḋ1 , Ḋ2 , Ḋ3 + h1iR = J 5 (2, 1),
5
(6.19)
where
Ḋ1 (x, y) = Da1 (x, y, 0, 0, 1/1 (p)) = 2x,
Ḋ2 (x, y) = Da2 (x, y, 0, 0, 1/1 (p)) = 2y,
Ḋ3 (x, y) = Da3 (x, y, 0, 0, 1/1 (p)) = 2f (x, y).
We consider the family D(x, Y, a1 , a2 , a3 ) with Y as above. Denote by
˜
dij the coefficient of xi Y j in the Taylor expansion of Da0 (x, Y ). Then
1
j 5 Da0 (x, Y ) = 1 (p) ˜ 2 + d˜32 xY 2 + d˜33 Y 3
2 + d22 Y

+d˜41 x3 Y + d˜42 x2 Y 2 + d˜43 xY 3 + d˜44 Y 4

+d˜50 x5 + d˜51 x4 Y + d˜52 x3 Y 2 + d˜53 x2 Y 3 + d˜54 xY 4 + d˜55 Y 5 .
The arguments for showing that the family D is a versal deformation
are similar to those in the proof of Theorem 6.2. We use (Da0 )x and (Da0 )Y
to show that all the monomials of degree 5 and 4 are in the 5-jet of the left
hand side of (6.19). We also get xy in there using (Da0 )Y . Now x3 appears
only in (Da0 )Y , and we need its coefficient d˜41 to be non-zero for it to be
in the 5-jet of the left hand side of (6.19). The coefficient d˜41 is a non-zero
multiple of the expression in the statement of the theorem. For degree 2
and 1 monomials, we use (Da0 )Y , Ḋ1 , Ḋ2 , and Ḋ3 to show that they are all
in the 5-jet of the left hand side of (6.19). Thus, the family D is a versal
deformation if and only if d˜41 6= 0.
We turn now to the D4 -singularity of da0 . The 2-jet of da0 vanishes
identically if and only if 1 (p) = 2 (p) = , equivalently, p is an umbilic
point. Then
1 2
j 3 Da0 (x, y) = 2 (a30 x3 + a31 x2 y + a32 xy 2 + a33 y 3 ).
 
The singularity of da0 is of type D4 if and only if the cubic form
a30 x3 + a31 x2 y + a32 xy 2 + a33 y 3 has no repeated root, equivalently, when
its discriminant is not zero. The discriminant of this cubic form is as in the
statement of the theorem.
Following the same steps as above we obtain all the monomials of degree
3 in the 3-jet of the left hand side of (6.19) using (Da0 )x and (Da0 )y . For
the monomials of degree 2 (and working modulo monomials of degree 3),
we have
0 2 1 1 0 21
j (  Ḋ3 ) x
@ j 2 (  Da0 )x A = M @ xy A ,
2
j 2 ( 2 Da0 )y y2
182 Di↵erential Geometry from a Singularity Theory Viewpoint

with 0 1
1 0 1
M = @ 3a30 2a31 a32 A .
a31 2a32 3a33
The monomial of degree 2 are in the 3-jet of the left hand side of (6.19)
if and only if the determinant of the matrix M is not zero, that is, 2a232
a231 + 3a31 a33 6= 0. We then use Ḋ1 and Ḋ2 to get x, y in the 3-jet of the
left hand side of (6.19). Therefore, the family D is a versal deformation of
the D4 -singularity of da0 if and only if 2a232 a231 + 3a31 a33 6= 0. ⇤
Remark 6.3. The geometric interpretation of the algebraic conditions in
Theorem 6.10 are given in §6.6.3.
Geometric objects can be derived from the family of distance squared
functions on the surface M , such as the focal set of M and the spheri-
cal Dupin foliation defined below. For a distance squared function generic
surface, the local structure of these objects up to di↵omorphisms is deter-
mined by the R-singularity type of the distance squared function da0 . It
turns out that the di↵eomorphism models of these objects determine the
R-singularity type of da0 . We start with the focal set.
Theorem 6.11. Away from umbilic points, the focal set F1 and F2 of M
are disjoint surfaces, and for a distance squared function generic M they
are di↵eomorphic to
(i) a smooth surface if and only if da0 has an A2 -singularity at p;
(ii) a cuspidal edge surface if and only if da0 has an A3 -singularity at p;
(iii) a swallowtail surface if and only if da0 has an A4 -singularity at p,
where a0 = p + 1/i N (p), i = 1 or 2.
At an umbilic point the focal set F1 [ F2 is di↵eomorphic to a pyramid
(resp. purse) if and only if dc has a D4 (resp. D4+ )-singularity at p.
Proof. ([Arnol’d, Guseı̆n-Zade and Varchenko (1985)]) The focal set of
M is a caustic with D the generating family of the Lagrangian submanifold
L(D)(CD ) ⇢ T ⇤ R3 . Thus, it has Lagrangian singularities. For a distance
squared function generic surface M , the family D is R+ -versal, so the focal
set is di↵eomorphic to the bifurcation set of a model R+ -versal unfolding
(with two parameters) of the singularities that occur in a given distance
squared function. These are A1 , A2 , A3 , A4 and D4 -singularities (Theo-
rem 6.9). The bifurcation set of an R+ -versal 2-parameter family of these
singularities are as stated in the proposition (see also §3.9.2) and are as in
Figure 6.8. ⇤
 Surfaces in the Euclidean 3-space 183

!! !$ !# "#% "#"

Fig. 6.8 Models of the focal set of a generic surface in R3 . The fourth and fifth figures
model the focal sets F1 and F2 joining at an umbilic point. The first three figures model
the focal sets F1 or F2 . One can have the following generic combinations for the pairs
(F1 , F2 ) or (F2 , F1 ): (A2 , A2 ), (A2 , A3 ), (A2 , A4 ), (A3 , A3 ).

Definition 6.7. The spherical Dupin foliation of M at p0 associated to the

tangential sphere of centre c0 with radius r0 is the family of curves obtained
by intersecting M with the 1-parameter family of spheres of centre c0 and
radius r varying near r0 .

Theorem 6.12. The spherical Dupin foliation of M at p0 associated to

the tangential sphere of centre c0 are diffeomorphic to the level sets of the
germs of the following functions (see also Figure 6.9):

±x21 + x22 ⇐⇒ da0 has an A1 -singularity at p0

x31 + x22 ⇐⇒ da0 has an A2 -singularity at p0
±x41 + x22 ⇐⇒ da0 has an A3 -singularity at p0
x51 + x22 ⇐⇒ da0 has an A4 -singularity at p0
x31 − x1 x22 ⇐⇒ da0 has an D4+ -singularity at p0
x31 + x32 ⇐⇒ da0 has an D4− -singularity at p0

Proof. The function da0 has A2 -singularity at p0 if and only if is R-

equivalent to f (x1 , x2 ) = x31 +x22 . Since two R-equivalent germs of functions
have diffeomorphic the level sets, the spherical Dupin foliation is diffeomor-
phic to the level set of germ of f (x1 , x2 ) = x31 + x22 (see Figure 6.9). The
other cases follow by a similar argument. !

6.6 Robust features of surfaces

We have already encountered several special curves on a surface M in R3 ,

such as its lines of principal curvatures and asymptotic curves. These form
184 Differential Geometry from a Singularity Theory Viewpoint

A+
1 A−
1 A2

A+
3 A−
3 A4

D4+ D4−

Fig. 6.9 Dupin foliations of M associated to tangential spheres.

pairs of foliations on the surface and one cannot trace in general an indi-
vidual curve when the surface is deformed. There may of course exist lines
of curvature or asymptotic curves which are homeomorphic to a circle, i.e.,
limit cycles. These are indeed robust features of the surface. Such curves
are studied in [Sotomayor and Gutierrez (1982); Garcia and Sotomayor
(1997)].
We consider below some special curves which are robust features on M ,
captured via the contact of the surface with planes, lines and spheres.

6.6.1 The parabolic curve

For a surface parametrised by x : U → R3 , the parabolic set is the image
by x of the zero set of the function ln − m2 . We start by establishing its
generic structure.

Proposition 6.13. The parabolic set of a height function generic surface is

 Surfaces in the Euclidean 3-space 185

a smooth curve (when not empty ). It consists of the A2 and A3 -singularities

of the height functions along the normal directions to the surface. The A3 -
singularities of these functions (i.e., the cusps of Gauss) occur at isolated
points on the parabolic set.

Proof. The surface M is height function generic, so the height functions

have only singularities of type A1 , A2 or A3 for any v 2 S 2 and these are
K-versally unfolded by the extended family of height functions.
The map g(u) = (u, N (u), hx(u), N (u)i) is an embedding from the sur-
face to the domain of the Legendrian map L(H̃) : ⌃(H) ! R3 . Therefore
L(H̃) g is a cuspidal edge map at an A2 -singularity of the height function
and a swallowtail map at an A3 -singularity of this function. In particular,
its singular set is di↵eomorphic to that of the A-model of the cuspidal edge
or the swallowtail map in Definition 1.4. The result now follows by direct
computation of the critical sets of these models. These are given by t = 0
for the model of the cuspidal edge map and t + 6s2 = 0 for that of the
swallowtail map. ⇤

Remark 6.4. One can also use the arguments in the proof of
Proposition 6.13 to show that there is a smooth curve on M of pairs of
points where the surface admits a bi-tangent plane meeting tangentially
the parabolic curve at a cusp of Gauss (Figure 6.10). These pairs of points
are where the height function along their common normal direction has an
A1 -singularity. The curve of bi-tangent planes corresponds to the curve
of self-intersections of the dual surface of M (a swallowtail at the cusp of
Gauss). The set is denoted the A1 A1 -set or A21 -set. For a height function
generic surface, the A1 A1 -set is a smooth curve which has ordinary tan-
gency with the parabolic curve at a cusp of Gauss. As in Proposition 6.13,
to prove this it is enough to carry out the calculations for the A-model of a
swallowtail surface in Definition 1.4. The A1 A1 -set is modelled by the pair
of (distinct) points (s1 , t1 ) and (s2 , t2 ) which have the same image by the
map (s, t) = (3s4 + s2 t, 4s3 + 2st, t). Then t1 = t2 = t and
3u41 + s21 t = 3s42 + s22 t, 4s31 + 2s1 t = 4s32 + 2s2 t.
Equivalently,
(s21 s22 )(3s21 + 3s22 + t) = 0, 2(s1 s2 )(2(s21 + s1 s2 + s22 ) + t) = 0.
The solution s1 = s2 is excluded as it gives (s1 , t1 ) = (s2 , t2 ). The
system of equations 3s21 + 3s22 + t = 0 and 2(s21 + s1 s2 + s22 ) + t = 0 also
gives s1 = s2 . Therefore the solution of the system is s2 = s1 (from the
186 Di↵erential Geometry from a Singularity Theory Viewpoint

first equation) and t = 2s21 (substituting in the second equation). Thus,

the curve of self-intersection is a smooth curve with equation t + 2s2 = 0.
It has an ordinary tangency with the singular set t + 6s2 = 0.
Remark 6.5. The parabolic curve is also detected by the singularities
of the orthogonal projections of the surface to planes. It follows from
Theorem 6.7 that p is a parabolic point if and only if the singularity of
Pv is a lips/beaks, goose or gulls, see Table 6.2.

6.6.2 The flecnodal curve

From Theorem 6.6, the flecnodal set of a generic surface is a smooth curve
when not empty. It is in fact the locus of geodesic inflections of the asymp-
totic curves. Before showing this we need the following result from [Bruce
and Tari (2000)].
Proposition 6.14. An asymptotic curve on M has a geodesic inflection at
p if and only if its projection to the tangent plane Tp M has an inflection at
p.
Proof. We take, without loss of generality, the surface M parametrised
in Monge form z = f (x, y) at the origin p. The tangent plane Tp M is the
plane z = 0. Let (t) = (x(t), y(t), f (x(t), y(t))) be a parametrisation of
the asymptotic curve with (0) = p, so that its projection to Tp M along
the normal direction (0, 0, 1) to M at p is the curve ↵(t) = (x(t), y(t)) (here
we are taking the orthogonal projection of to Tp M , but any projection
along a direction transverse to Tp M will do). As is an asymptotic curve,
we have
fxx (↵(t))x0 (t)2 + 2fxy (↵(t))x0 (t)y 0 (t) + fyy (↵(t))y 0 (t)2 = 0,
for all t near zero. The geodesic curvature of the asymptotic curve van-
ishes at t if and only if [ 0 (t), 00 (t), N (↵(t))] = 0 (see (6.2)). We have,
x0 y 0 x0 fx + y 0 fy
1
[ 0 , 00 , N ] = x00 y 00 fxx x02 + 2fxy x0 y 0 + fyy y 02 + x00 fx + y 00 fy
fx fy 1

x0 y 0 x0 fx + y 0 fy
1
= x00 y 00 x00 fx + y 00 fy
fx fy 1

= (x0 y 00 x00 y 0 ) ,
 Surfaces in the Euclidean 3-space 187

where = (1 + fx2 + fy2 )1/2 and all the partial derivatives of f are evaluated
at ↵(t). Thus [ 0 (t), 00 (t), N (↵(t))] = 0 if and only if (x00 y 0 x0 y 00 )(t) = 0,
which completes the proof. ⇤

Proposition 6.15. Let p be a hyperbolic point on M . Then p is a flecnodal

point and corresponds to the swallowtail singularity of some projection Pv
if and only if p is a geodesic inflection of the asymptotic curve through p
with tangent direction v at p.

Proof. We take M in Monge form z = f (x, y) at p with f as in (6.4)

and v = (0, 1, 0). By Theorem 6.6, p is a flecnodal point if and only if
fy = fyy = fyyy = 0 and fxy fyyyy 6= 0 at (0, 0).
Asymptotic curves satisfy the binary di↵erential equation (6.3). As p is
a hyperbolic point and v an asymptotic direction (then fyy (0, 0) = 0), the
two asymptotic curves through p are smooth and the one tangent to v can
be parametrised by (t) = (g(t), t, f (g(t), t)) for some germ of a smooth
function g with g 0 (0) = 0. We get from equation (6.3)

fxx (g(t), t)g 0 (t)2 + fxy (g(t), t)g 0 (t) + fyy (g(t), t) = 0.

Di↵erentiating the above identity and evaluating at t = 0 yields

2fxy g 00 + fxxx g 03 + 3fxxy g 02 + 3fxyy g 0 + fyyy = 2fxy g 00 + fyyy = 0

at t = 0. We have fxy (0, 0) 6= 0, so g 00 (0) = 0 if and only if fyyy (0, 0) = 0,

equivalently, p is a flecnodal point. By Proposition 6.14, the condition
g 00 (0) = 0 means that the asymptotic curve has a geodesic inflection at
(0). ⇤

Theorem 6.6 gives the equations of the flecnodal curve for M in Monge
form. Using Proposition 6.15, we can obtain its equations in terms of the
coefficients of the second fundamental form and their derivatives.

Theorem 6.13. Let x : U ! R3 be a local parametrisation of M . The

flecnodal curve is the image by x of the set of points (u1 , u2 ) for which the
following system of equations in a, b

la2 + 2mab + nb2 = 0,

lu1 a3 + (lu2 + 2mu1 )a2 b + (2mu2 + nu1 )ab2 + nu2 b3 = 0,

has a solution, where l, m, n and their partial derivatives are evaluated at

u = (u1 , u2 ).
188 Di↵erential Geometry from a Singularity Theory Viewpoint

Proof. According to Proposition 6.15, the flecnodal curve is the locus

of geodesic inflections of the asymptotic curves. Let ↵(t) = (u1 (t), u2 (t))
be a local parametrisation of an asymptotic curve and write ↵0 = (a, b).
Suppose that b 6= 0 (the case b = 0 follows similarly) and re-parametrise ↵
in the form (g(t), t), with g 0 = a/b. As ↵ is an asymptotic curve,
l(g(t), t)g 0 (t)2 + 2m(g(t), t)g 0 (t) + n(g(t), t) = 0.
Di↵erentiating with respect to t and dropping the arguments, we get
lu1 g 03 + (lu2 + 2mu1 )g 0 + (2mu2 + nu1 )g 0 + nu2 + 2(lg 0 + m)g 00 = 0.
The point p is not parabolic, so lg 0 + m 6= 0 which implies that g 00
vanishes at some t if and only if
lu1 g 03 + (lu2 + 2mu1 )g 0 + (2mu2 + nu1 )g 0 + nu2 = 0,
equivalently,
lu1 a3 + (lu2 + 2mu1 )a2 b + (2mu2 + nu1 )ab2 + nu2 b3 = 0.
⇤

Theorem 6.14. For a height function and projection generic surface, the
flecnodal set contains the cusps of Gauss, is a smooth curve at such points
and has 2-point contact with the parabolic curve there.

Proof. We take M in Monge form z = f (x, y) at the origin p which

we assume to be a cusp of Gauss. We choose (0, 1, 0) to be the unique
asymptotic direction at p, so
j 4 f = a20 x2 + a30 x3 + a31 x2 y + a32 xy 2 +
a40 x4 + a41 x3 y + a42 x2 y 2 + a43 xy 3 + a44 y 4 ,
with a20 6= 0 and a232 4a20 a44 6= 0 (Theorem 6.2). As the surface is height
function generic, a32 6= 0 (Theorem 6.2). Similar calculations to those in
the proof of Proposition 6.5 show that the parabolic set is a smooth curve
at p and can be parametrised locally in the form x = x1 (y) with
a232 6a20 a44 2
j 2 x1 (y) = y . (6.20)
a20 a32
For the flecnodal set we need to solve the system of equations (6.14),
(6.15) and (6.16). These equations are satisfied when all the parameters
are set to zero as fy = fyy = fyyy = 0 at p. Therefore, p is in the flecnodal
set.
We can use (6.14) to obtain as a function of (x, y, ↵). Substituting
in (6.16) we get ↵ as a function of (x, y) as a32 6= 0. Now (6.15) becomes
 Surfaces in the Euclidean 3-space 189

Flecnodal curve

A1A1curve

Parabolic curve

Fig. 6.10 A sketch of a configuration of the parabolic, flecnodal and A1 A1 curves of a

surface patch at a cusp of Gauss.

an equation in (x, y) and gives the flecnodal curve. This curve can be
parametrised locally in the form x = x2 (y) with
a44 2
j 2 x2 (y) = y . (6.21)
a32
It is clear from (6.20) and (6.21) that the flecnodal and parabolic curves
are tangential at p. We have j 2 (x1 x2 )(y)/y 2 = a32 /a20 6= 0, so the two
curves have 2-point contact at p. ⇤

We sketch in Figure 6.10 the three robust features of the surface at a

cusp of Gauss, captured by the contact of the surface with planes and lines.
There are several configurations of the three curve at the cusp of Gauss
([Uribe-Vargas (2006)]). These curves are used in [Uribe-Vargas (2006)]
to obtain a projective invariant of the surface at cusps of Gauss.

6.6.3 The ridge curve

Porteous introduced the following concept (see [Porteous (1987, 1983a,
2001)]).

Definition 6.8. A point on a surface M is called a ridge point if it is an

A3 -singularity of some distance squared function on M . The closure of the
set of the ridge points is called the ridge of the surface M .
190 Di↵erential Geometry from a Singularity Theory Viewpoint

The ridge inherits the colour of the principal directions. Following Por-
teous’ colouring, we have a red ridge associated, say, to the principal cur-
vature 1 and the blue ridge associated to the principal curvature 2 .
The ridge contains the points where the singularity of some distance
squared function is more degenerate than A3 . For a distance squared func-
tion generic surface, these are either A4 or D4 -singularities.
The image of the ridge associated to the principal curvature i by the
map "i (p) = p + 1/i (p)N (p), i = 1 or 2, is precisely the singular set of the
focal set, and captures the following geometric property of the surface.

Proposition 6.16. Let x : U ! R3 be a local parametrisation of M away

from umbilic points such that the coordinate curves are the lines of principal
curvature. Then the ridge is given by the set of points x(u1 , u2 ) where
@i
(u1 , u2 ) = 0,
@ui
for i = 1 or 2, that is, the ridge is the set of points where a principal
curvature is extremal along its associated lines of principal curvature.

Proof. Consider, for example, the focal set associated to the principal
curvature 1 which is parametrised by
1
"1 (u1 , u2 ) = (x + N )(u1 , u2 ).
1
The directions xu1 and xu2 are principal directions at all points in U so
Nui = i xui , i = 1, 2. Di↵erentiating "1 and dropping the argument, we
get
@"1 1 @k1
= N,
@u1 21 @u1
@"1 2 1 @k1
= (1 )xu2 N.
@u2 1 21 @u2
A point p = x(u1 , u2 ) is on the ridge if the focal set is singular at
"1 (u1 , u2 ). As p is not an umbilic point, the vectors @"1 /@u1 and @"1 /@u2
are linearly dependent at (u1 , u2 ) if and only if
@k1
(u1 , u2 ) = 0.
@u1 ⇤

Remark 6.6. It follows from the proof of Proposition 6.16 that, away from
umbilic points, the normal direction to the focal set F1 , which is given by
@"1 /@u1 ⇥ @"1 /@u2 , is the principal direction v1 associated to the principal
curvature 1 . Similarly, the normal direction to the focal set F2 is the
principal direction v2 associated to the principal curvature 2 .
 Surfaces in the Euclidean 3-space 191

Proposition 6.17. Suppose that M is a distance squared function generic

surface patch without umbilic points. Then the ridge is a smooth curve
(when not empty).

Proof. The surface M is distance squared function generic, so the dis-

tance squared function Da has only singularities of R-type A1 , A2 , A3 , A4
or D4 for any a 2 R3 . The D4 -singularities occur at umbilic points and the
surface patch is supposed to be umbilic free. Therefore, the ridge contains
only A3 or A4 singular points of the distance squared function.
The map hi (u) = (u, "i (u)), i = 1 or 2, is an embedding from the
surface to the critical set of the Lagrangian map L(D) : C(D) ! R3 .
Thus, "i = L(D) hi is a cuspidal edge map at an A3 -singularity of the
distance squared function and a swallowtail map at an A4 -singularity of
this function. The result now follows by direct computation of the critical
sets of the A-models of the cuspidal edge and swallowtail maps in Definition
1.4. These are given by t = 0 for the model of the cuspidal edge map and
t + 6s2 = 0 for that of the swallowtail map. ⇤

Corollary 6.3. The distance squared function has an A3 -singularity if and

only if the ridge is transverse to its associated principal direction. The
singularity is of type A4 if and only if the ridge is tangent to its associated
principal direction.

Proof. According to [Saji, Umehara and Yamada (2009)], a map-germ

h : R2 , 0 ! R3 is a cuspidal edge map if and only if its critical set is a
smooth curve and is transverse to the kernel of dh. It is a swallowtail map
if and only if its critical set is a smooth curve and has an ordinary tangency
with the kernel of dh at the swallowtail point.
The proof is a direct consequence of the above criteria applied to the
map "i , i = 1, 2. The singular set of "i is the ridge curve and it follows from
the proof of Proposition 6.16 that the kernel of d"i is along the principal
direction associated to the ridge. ⇤

We can now characterise geometrically the generic singularities of the

distance squared function on M .

Theorem 6.15. Suppose that M is a distance squared function generic

surface. Then, the singularities of the distance squared function dc at p 2 M
are characterised as follows.
192 Di↵erential Geometry from a Singularity Theory Viewpoint

A1 : c is on the normal line to M at p but is not a focal point.

A2 : c is a focal point and p is not a ridge point
A3 : c is a focal point, p is a ridge point and the ridge curve is
transverse to its associated principal direction
A4 : c is a focal point, p is a ridge point and the ridge is tangent
to its associated principal direction
D4 : c is a focal point and p is an umbilic point.
One expects, by looking at the focal sets in Figure 6.8, one or three ridge
curves arriving at an umbilic point. The structure of the closure of these
curves is established in [Bruce (1984)]. At an umbilic point the normal
curvatures in all tangent directions to the surface coincide and are equal
to some constant . At such a point the surface can be taken locally in
Monge-form z = f (x, y) with

j 3 f (x, y) = (x2 + y 2 ) + C(x, y),
2
where C is a cubic form in x, y (i.e., a homogeneous polynomial of degree
3 in x, y). The distance-squared function dc with c = (0, 0, 1/) is given
locally by C(x, y) + O(4). It has a D4 -singularity if and only if the cubic
form C has three distinct roots. If the three roots are real, dc has a D4 -
singularity (elliptic umbilic). If two of the roots are complex (conjugate),
dc has a D4+ -singularity (hyperbolic umbilic).

Theorem 6.16 ([Bruce (1984)]). The ridge set through an umbilic

point of a generic surface has local model, up to di↵eomorphism, xy(x y) =
0 at a D4 and x = 0 at a D4+ .

Proof. The idea of the proof is as follows. At every point q on the

surface, one can write the surface locally in Monge form z = fq (x, y).
The 1-jet of fq at the origin is identically zero. Let Vk denote the set of
polynomials in (x, y) of degree d with 2  d  k. Define the Monge-Taylor
map j k : M ! Vk , by j k (q) = j k fq (0, 0).
Recall that an A3 -singularity is 4-R-determined. The conditions for
a distance squared function to have an A3 -singularity at p are expressed
as algebraic conditions on the coefficients of j 3 fp (expressing the failure
to have an A2 -singularity) together with an open semi-algebraic condition
involving degree 4 terms in j 4 fp (expressing the condition to have an A3 -
singularity and not a more degenerate one). The algebraic condition defines
a variety in Vk (k 4) labeled the A3 -stratum. The structure of the closure
of the A3 -stratum is determined in [Bruce (1984)] and it is also shown there
 Surfaces in the Euclidean 3-space 193

that the Monge-Taylor map is transverse to the D4 -stratum at an umbilic

point. The ridge is the pre-image by j k of the intersection of j k (M ) with
the A3 -stratum and is di↵eomorphic to the models in the statement of the
theorem. See [Bruce (1984)] for more details. ⇤

It is also known that ridges change colour at an umbilic [Porteous

(2001)]. A consequence of Theorem 6.16 and of this remark is the following
characterisation of the D4 -singularities of the distance squared function.

Corollary 6.4. Let M be a distance squared function generic surface

parametrised by x : U ! R3 . Suppose that p0 = x(u0 ) is an umbilic
point. Then
(i) da0 has a D4 -singularity if and only if for any ✏ > 0, there exist six
distinct points pi = x(ui ) such that ||u0 ui || < ✏ for i = 1, . . . , 6, and all
the points pi are ridge points.
(ii) da0 has a D4+ -singularity if and only if for any ✏ > 0, there exist two
distinct points pi = x(ui ) such that ||u0 ui || < ✏ for i = 1, 2, and p1 and
p2 are ridge points of di↵erent colours.

At a hyperbolic umbilic there are two possible configurations of the

ridges (Figure 6.11, left). The configuration depends only on the cubic form
C (Proposition 6.18 below). Any cubic form can be written as Re(↵z 3 +
z 2 z̄), where ↵ and are complex numbers, z = x + iy and Re(w) denotes
the real part of the complex number w. Any such cubic form is SO(2)-
equivalent to one of the form

Re(z 3 + z 2 z̄)

or is SO(2)-equivalent to Re(z 2 z̄) = x(x2 + y 2 ). (The SO(2)-equivalence

corresponds to a change of an orthonormal basis in Tp M and does not af-
fect the 2-jet of f (x, y).) Therefore, the set of cubic forms can be viewed as
the set of points in the -plane. There are the following three exceptional
curves in this plane:

= 2ei✓ + e 2i✓
: this is a hypocycloid in Figure 6.11 (right) and con-
sists of the umbilics which are more degenerate than
D4 , i.e., those cubic forms with repeated roots.
194 Di↵erential Geometry from a Singularity Theory Viewpoint

4
3
2

3 1 2

2
3
Regions 3 and 4 Region 1 Region 2

Fig. 6.11 Generic configurations of the ridge curves at umbilic points left and partition
of the -plane right. The ridge associated to one focal sheet is drawn in continuous line
and the one associated to the other focal sheet is drawn in dashed line.

| |=3: this is the outer circle in Figure 6.11. It consists of

the umbilics which are not versally unfolded by the
family of distance-squared functions (this is equiv-
alent to the algebraic condition given in Theorem
6.10(ii)).
| | = 1: this is the inner circle in Figure 6.11. It consists of
umbilics at which two of the ridge lines (or more) are
tangential.

Proposition 6.18 ([Bruce (1984)]). The configuration ridges at an

umbilic point is completely determined by the cubic C. For not on one of
the three exceptional curves above, the configuration is as in Figure 6.11.

Proof. See [Bruce (1984)]. ⇤

6.6.4 The sub-parabolic curve

In [Bruce and Wilkinson (1991); Wilkinson (1991)], the authors considered
the singularities of the folding maps on M . Let ⇡ be a plane in R3 and let v
be a unit vector orthogonal to ⇡. Denote by d(p, ⇡) the distance of a point
p 2 R3 to ⇡. Then the folding map along ⇡ is the map F⇡ : R3 ! R3 given
by F⇡ (p) = Pv (p) + d(p, ⇡)2 v, where Pv is the orthogonal projection along
v to ⇡. For instance, the folding map along the plane y = 0 is the map
Fy=0 (x, y, z) = (x, y 2 , z). By varying ⇡, one obtains a 3-parameter family
of folding maps in the ambient space R3 ([Bruce and Wilkinson (1991)]).
The restriction of this family to M is called the family of folding maps on
M.
 Surfaces in the Euclidean 3-space 195

The authors in [Bruce and Wilkinson (1991)] proved that the bifurca-
tion set of the family of folding maps is dual to the bifurcation set of the
family of distance squared function (which is the focal set). Thus, the fold-
ing maps capture the geometry of the focal set obtained via its contact with
planes. In particular, it captures its parabolic set. The set of points on the
surface M which correspond to the parabolic set of its focal set is defined in
[Bruce and Wilkinson (1991); Wilkinson (1991)] as the sub-parabolic curve
of M . Before giving other geometric characterisations of the sub-parabolic
curve, we require the following result.

Proposition 6.19. Let x : U ! R3 be a local parametrisation of M at a

non-umbilic point p such that the coordinate curves are the lines of principal
curvature. Suppose that p is not a ridge point. Then the coefficients of the
first fundamental form of the focal set F1 are
1 @k1 2 1 @k1 @k1 2 2 1 @k1 2
EF 1 = ( ) , FF1 = , GF1 = (1 ) + 4( )
41 @u1 41 @u1 @u2 1 1 @u2
and those of its second fundamental form are
1 @k1 p 1 2
lF1 = E, mF1 = 0, nF1 = p (1 )hxu1 , xu2 u2 i.
1 @u1 E 1
For the focal set F2 , the coefficients of its first fundamental form are
1 @k2 2 1 @k2 @k2 1 2 1 @k2 2
EF 2 = ( ) , FF2 = , GF2 = (1 ) + 4( )
42 @u2 42 @u1 @u2 2 2 @u1
and those of its second fundamental form are
1 @k2 p 1 1
l F2 = G, mF2 = 0, nF2 = p (1 )hxu2 , xu1 u1 i.
2 @u2 G 2
Proof. The focal set F1 is parametrised by "1 (u) = x(u) + 11(u) N (u).
The coefficients of the first fundamental form of F1 are computed using the
partial derivatives of the map "1 which are given in the proof of Proposition
6.16. p
The Gauss map of the focal set F1 is NF1 = (1/ E)xu1 , so the coeffi-
cients of the second fundamental form of F1 are
@ 2 "1 1 @k1 p
lF1 = hNF1 , i = E
@u21 1 @u1

@ 2 "1
mF1 = hNF1 , i=0
@u1 u2
196 Di↵erential Geometry from a Singularity Theory Viewpoint

@ 2 "1 2
nF1 = hNF1 , i = (1 p )hxu1 , xu2 u2 i
@u22 1 E
The coefficients of the first and second fundamental forms of F2 are
computed in a similar way. ⇤

Theorem 6.17 ([Morris (1996)]). Let x : U ! R3 be a local parametri-

sation of M at a non-umbilic point p such that the coordinate curves are
the lines of principal curvature. Suppose that p is not a ridge point asso-
ciated to F1 . Then the sub-parabolic curve of M which corresponds to the
parabolic set of the focal set F1 is the set of points x(u1 , u2 ) where
@2
(u1 , u2 ) = 0.
@u1
Similarly, the sub-parabolic curve which corresponds to the parabolic curve
of the focal set F2 is given by
@1
(u1 , u2 ) = 0.
@u2
That is, the sub-parabolic curve is the set of points where a principal
curvature is extremal along the other line of principal curvature.

Proof. We shall give a slightly di↵erent proof to that of Morris in [Morris

(1996)] and consider only the case of the focal set F1 . The case of the focal
set F2 follows in a similar way.
The sub-parabolic curve associated to the focal set F1 is the set of points
on M where the Gaussian curvature
lF1 nF1 m2F1
KF1 =
EF1 GF1 FF2 1
of F1 vanishes. It follows from Proposition 6.19 that KF1 = 0 if and only
if
hxu1 , xu1 u2 i = 0 (6.22)
The coordinate curves are the lines of principal curvature, so
hxu1 , xu2 i = 0 and we get by di↵erentiating this identity with respect to u2
hxu2 , xu1 u2 i = hxu1 , xu2 u2 i. (6.23)
Again, by the choice of the coordinate system, we have Nu1 = 1 xu1
and Nu2 = 2 xu2 , so
hNu1 , xu2 i = h1 xu1 , xu2 i = 0.
 Surfaces in the Euclidean 3-space 197

Di↵erentiating the above identity with respect to u2 yields

hNu1 u2 , xu2 i + hNu1 , xu1 u2 i = 0.
Hence,
hNu1 u2 , xu2 i = hNu1 , xu1 u2 i = 1 hxu1 , xu1 u2 i. (6.24)
Di↵erentiating Nu2 = 2 xu2 with respect to u1 gives
@2
N u1 u 2 = x u + 2 x u 1 u 2 .
@u1 2
so that
@2
hNu1 u2 , xu2 i = hxu2 , xu2 i + 2 hxu2 , xu1 u2 i.
@u1
Equivalently,
@2
hNu1 u2 , xu2 i 2 hxu2 , xu1 u2 i = G. (6.25)
@u1
Using (6.24) and (6.23), equation (6.25) becomes

@2
(2 1 )hxu1 , xu2 u2 i = G. (6.26)
@u1
As the surface patch is umbilic free, it follows from (6.26) that equation
(6.22) is satisfied if and only if
@2
= 0.
@u1 ⇤

Proposition 6.20 ([Morris (1996)]). The sub-parabolic curve corre-

sponding to the focal set F1 (resp. F2 ) is the locus of geodesic inflections
of the lines of curvature corresponding to the principal curvature 2 (resp.
1 ).

Proof. We consider the sub-parabolic curve corresponding to the focal

set F1 , the other case follows similarly. With the parametrisation as in
Theorem 6.17, a line of principal curvature corresponding to the principal
curvature 2 is parametrised by ↵(u2 ) = x(a, u2 ), for some constant a. By
(6.2), its geodesic curvature is given by
1 00 0 1 1
g = h↵ , ↵ ⇥ N i = 3 hx u u , x u ⇥ N i = 3 hxu1 , xu2 u2 i.
||↵0 ||3 G2
2 2 2
G2
Therefore, g = 0 if and only if hxu1 , xu2 u2 i = 0 and the result follows
by (6.22) and (6.23). ⇤
198 Di↵erential Geometry from a Singularity Theory Viewpoint

L=Lemon
M=Monstar
S=Star M

M S

M
L
Lemon Star Monstar

Fig. 6.12 The partition of the -plane (right) and the generic configurations of the sub-
parabolic curves at an umbilic point (first three left). The sub-parabolic curve associated
to one focal sheet is drawn in continuous line and the one associated to the other focal
sheet is drawn in dashed line.

The sub-parabolic curve of a generic surface is a smooth curve away from

umbilic points. At an umbilic point it consists of three transverse curves or
a single curve. The configurations of the sub-parabolic curves are closely
related to those of the lines of principal curvature. If we write the surface in
Monge-form 2 (x2 + y 2 ) + C(x, y) + O(4) and take C(x, y) = Re(z 3 + z 2 z̄),
then the configuration of the sub-parabolic curves depends only on the
cubic C. There are two exceptional curves in the plane:
= 3(2ei✓ + e 2i✓
): this is the hypocycloid in Figure 6.12 (right),
separating the lemon from the star and mon-
star umbilics (Figure 6.2).
| | = 3: this is the circle in Figure 6.12 (right) separat-
ing the star and monstar umbilics.
The configurations of the sub-parabolic curves in the open regions of
the -plane delimited by the above two exceptional curves are as in Figure
6.12 ([Wilkinson (1991)]).

6.7 Notes

The contact of surfaces with planes and lines is an affine and projective
property of the surface, so does not depend on the metric in the ambient
space (see for example [Bruce, Giblin and Tari (1995); Shcherbak (1986)]).
The results in this chapter are, in particular, valid for surfaces embedded
in Minkowski 3-space. For such surfaces, additional geometric information
can be obtained when considering projections along the lightlike directions
in R31 , see [Izumiya and Tari (2013)].
 Surfaces in the Euclidean 3-space 199

Extensive work was carry out on the singularities of orthogonal projec-

tions of surfaces to planes. We refer to the following (incomplete) list for
related works: [Arnol’d (1983); Ga↵ney and Ruas (1979); Ga↵ney (1983);
Kergosien (1981); Koenderink (1990); Koenderink and van Doorn (1976);
Koenderink (1984); Platonova (1984); Rieger (1987); Landis (1981);
Lyashko (1979); Goryunov (1981a,b)].
We considered here only the local singularities of the relevant germs of
functions and mappings. Their multi-local singularities are also of impor-
tance and yield geometric information about the surface. For instance, the
results in Theorems 6.1, 6.5, 6.9 can be extended, using multi-transversality
arguments, to cover the multi-local singularities. See [Bancho↵, Ga↵ney
and McCrory (1982); Uribe-Vargas (2006)] for some geometric conse-
quences of these at a cusp of Gauss and [Bruce, Giblin and Gibson (1985);
Damon (2003, 2004, 2006); Giblin and Holtom (1999); Giblin and Janeczko
(2012); Giblin and Zakalyukin (2005)] for work on symmetry sets and me-
dial axis. Other works on multi-local singularities for other dimensions of
the target can be found in [Atique (2000); Dreibelbis (2001, 2006)].
Orthogonal projections of non-smooth surfaces are also studied, see
[Arnol’d (1979); Bruce and Giblin (1990); Goryunov (1990)] for pro-
jections of surfaces with boundary and [Tari (1991)] for those with creases
and corners.
Suppose the direction of the orthogonal projection of a surface is chang-
ing along a curve. One question of great interest is how to reconstruct the
surface from the resulting family of apparent contours. One can in fact re-
construct part of the surface (the visible part) from this family of apparent
contours [Giblin and Weiss (1987); Siddiqi and Pizer (2008)].
We highlighted some properties of the parabolic, ridge and sub-parabolic
curves. The complete catalogues of the bifurcations of these curves as the
surface is deformed in generic 1-parameter families of surfaces are given in
[Bruce, Giblin and Tari (1995, 1998, 1999)], see also [Uribe-Vargas (2001)].
We considered only smooth and regular surfaces. The techniques de-
scribed in this book are also currently been used to study the geometry
of singular submanifolds in Rn and Rn1 , see for example [Bruce and West
(1998); Dias and Nuño (2008); Oset Sinha and Tari (2010); Saji, Umehara
and Yamada (2009); West (1995)]. See also Chapter 1 for a brief outline of
the techniques and for references for the study of special singular surfaces
such as rules and developable surfaces.
The families of orthogonal projections and height functions on a surface
are related via a duality result between their bifurcation sets [Bruce and
200 Di↵erential Geometry from a Singularity Theory Viewpoint

Romero-Fuster (1991); Bruce (1994a); Bruce, Giblin and Tari (1998);

Shcherbak (1986)]. The families of distance squared functions and of the
folding maps are also related via a duality result between their bifurcation
sets [Bruce and Wilkinson (1991); Bruce, Giblin and Tari (1999); Wilkin-
son (1991)]. Similar duality results in other dimensions can be found for
instance in [Bruce and Nogueira (1998); Nabarro (2003)].
For results on the contact of the surface with circles see [Bruce (1994b);
Montaldi (1986b)].
A line of research that we did not explore in detail in this book is on
the pairs of foliations determined by the asymptotic and principal directions
(Figure 6.2 and Figure 6.6). Their study also reveals a great deal of the
geometry of the surface; see for example [Tari (2010)] for a survey article
and references.
 Chapter 7

Surfaces in the Euclidean 4-space

We consider in this chapter the extrinsic di↵erential geometry of a surface

M immersed in R4 . The study of the second order geometry of M is
considered in depth in [Little (1969)]. The work of Little inspired much of
recent work on the subject.
Geometric properties of which are related to its second fundamental
form depend only on the 2-jet of the immersion. These properties can be
derived, at each point p on M , from an ellipse in the normal plane of M at
p, called the curvature ellipse. Isometric invariants of the curvature ellipse,
such as its area, are isometric invariants of the surface. Also, the position of
the point p with respect to the curvature ellipse (outside, on, inside) gives
an isometric invariant partition of the surface.
We consider the contact of the surface with flat objects (hyperplanes,
planes and lines) and derive from it extrinsic properties of the surface.
This contact is affine invariant, so the derived properties from it are also
affine invariant. In fact, various properties of the surface derived from
the isometric properties of the curvature ellipse can also be derived from
the affine properties of this ellipse. A key observation is that the second
fundamental form at each point p on the surface defines a pair of quadratic
forms (Q1 (x, y), Q2 (x, y)). The action of GL(2, R) ⇥ GL(2, R) on the set of
these quadratic forms determines much of the second order affine properties
of the surface at p.
As in the case of surfaces in R3 , the contact viewpoint reveals new
extrinsic geometric properties of the surface that depend on higher order
jets of the immersion. We consider in this chapter in detail the contact of
M with hyperplanes and lines and touch briefly on its contact with planes.
Certain types of singularities of the map-germs that define the contact
of the surface with a given model object occur on curves on the surface.

201
202 Di↵erential Geometry from a Singularity Theory Viewpoint

These curves are robust features of the surface and part of this chapter is
devoted to determining and characterising them. Certain robust curves are
related to the contact of the surface with two model objects. This suggest
a (duality) relationship between the mappings that define the contact of
the surface with the model objects. We explore briefly this relationship.
We recall in §7.1 some properties of the second fundamental form of
M and consider some of its local intrinsic and affine invariants. We also
define the asymptotic directions at a given point and give the expression
of the binary di↵erential equation of their integral curves. The coefficients
of this equation define at each point on the surface a point in the projec-
tive plane. It turns out that its polar line with respect to the conic of
degenerate quadratic forms represent the ⌫-shape operators. Exploring the
polarity further gives us a way of choosing one special pair of ⌫-lines of
curvature which we call the lines of curvature of M . We study the con-
tact of the surface with hyperplanes (§7.6), lines (§7.7) and planes (§7.8),
and derive robust features of the surface. The contact of a surface with
hyperplanes is intimately related to that of its canal hypersurface with hy-
perplanes (§7.6.1). The contact of the surface with lines is captured by the
singularities of orthogonal projections to 3-spaces. The image of the surface
by a projection can be singular and the geometry of the singular projected
surface gives geometric information about the surface itself. We discuss
in the last section (§7.9) the contact of the surface with hyperspheres and
give geometric characterisations of the generic singularities of the family of
distance squared functions on the surface.

7.1 The curvature ellipse

Let x : U ! R4 be a local parametrisation of M , where U is an open subset

of R2 . Let {e1 , e2 , e3 , e4 } be a positively oriented orthonormal frame in R4
such that at any u 2 U, {e1 (u), e2 (u)} is a basis for the tangent plane
Tp M and {e3 (u), e3 (u)} is a basis for the normal plane Np M at p = x(u).
Associated to this frame is the dual basis of 1-forms {!1 , !2 , !3 , !4 } in the
dual space of R4 .
In [Little (1969)] are given some properties of the surface which are
invariant under rotations in Tp M and Np M , that is, invariant under the
action of SO(2) ⇥ SO(2) on Tp M ⇥ Np M . The expressions for the invariants
are given with respect to an orthonormal frame {e1 , e2 , e3 , e4 }. We recall
here some of them.
 Surfaces in the Euclidean 4-space 203

The vectors ei and the 1-forms !1 can be extended to an open subset in

R , and we keep the same notation for their extensions. Define the 1-forms
4

by !i and !ij , i, j = 1, . . . , 4 by
!i = hdx, ei i and !ij = hdei , ej i,
where d is the exterior di↵erential. It is worth observing that !3 = !4 = 0
and !ij = !ji . The Maurer-Cartan structural equations (see [do Carmo
(1976)]) are
d!i = ⌃4j=1 !ij ^ !j and d!ij = ⌃4k=1 !ik ^ !kj . (7.1)
We have
0 = d!3 = !31 ^ !1 + !32 ^ !2 ,
0 = d!4 = !41 ^ !1 + !42 ^ !2 .
It follows by a lemma of Cartan (see [do Carmo (1976)]) that there
exist a, b, c, e, f, g, such that
!13 = a!1 + b!2 , !14 = e!1 + f !2 ,
!23 = b!1 + c!2 , !24 = f !1 + g!2 .
The second fundamental form of M is the vector valued quadratic form
associated to the normal component of the second derivative d2 x of x at p
(see Chapter 2), that is,
IIp = hd2 x, e3 ie3 + hd2 x, e4 ie4 .
We have
hd2 x, e3 i = hdx, de3 i
= h!1 e1 + !2 e2 , de3 i
= !1 he1 , de3 i !2 he2 , de3 i
= !1 !31 !2 !32
= !1 !13 + !2 !23
= a!12 + 2b!1 !2 + c!22 .
Similarly, hd2 x, e4 i = e!12 + 2f !1 !2 + g!22 , so that
IIp = (a!12 + 2b!1 !2 + c!22 )e3 + (e!12 + 2f !1 !2 + g!22 )e4 .
✓ ◆
ab c
The matrix ↵ = is called the matrix of the second fundamental
ef g
form with respect to the orthonormal frame {e1 , e2 , e3 , e4 }. Its entries are
called the coefficients of the second fundamental form with respect to that
frame.
204 Di↵erential Geometry from a Singularity Theory Viewpoint

The 1-form !12 is the connection form in the tangent bundle of M and
!34 is the connection form in the normal bundle of M , and d!12 and d!34
are the respective curvature forms in those bundles. Following Little, the
forms !1 , !2 and !12 can be regarded as forms in the tangent bundle and
depend only on the metric !12 + !22 on the surface. The Gaussian curvature
of the surface can be found using
d!12 = K!1 ^ !2 .
Little defines another scalar invariant N , called the normal curvature, by
d!34 = N !1 ^ !2 . The Gaussian and normal curvatures K and N can be
expressed as follows in terms of the coefficients of the second fundamental
form:
K = (ac b2 ) + (eg f 2) and N = (a c)f (e g)b.
Let ↵1 and ↵2 be the symmetric matrices associated to the quadratic
forms hd2 x, e3 i and hd2 x, e4 i respectively, so that
✓ ◆ ✓ ◆
ab ef
↵1 = and ↵2 = .
bc f g
Then
K = K1 + K 2 with K1 = det ↵1 , K2 = det ↵2 .
In fact, there is a more general result explaining the above relation (see [do
Carmo (1976); Basto-Gonçalves (2013)]).

Theorem 7.1. The Gaussian curvature K of M at a point p is the sum of

the curvatures K1 and K2 of the images M1 and M2 of the surface M by
orthogonal projections along any two orthogonal normal directions n1 and
n2 in Np M respectively.

Let S 1 be the unit circle in Tp M parametrised by ✓, and denote by

✓ the normal section of M in the direction u = cos ✓e1 + sin ✓e2 . We
parametrise ✓ by arc length and denote by ⌘(✓) its curvature vector at p.
Thus, ⌘(✓) is the projection of ✓00 to Np M , equivalently, ⌘(✓) = IIp (u). We
have then, at each point p on M , a map ⌘ : S 1 ⇢ Tp M ! Np M , given by
⌘(✓) = (a cos2 ✓ + 2b cos ✓ sin ✓ + c sin2 ✓)e3
+(e cos2 ✓ + 2f cos ✓ sin ✓ + g sin2 ✓)e4 . (7.2)
Using the trigonometric identities for double angles, expression (7.2)
can be written in the form
⌘(✓) = H + cos(2✓)B + sin(2✓)C,
 Surfaces in the Euclidean 4-space 205

with
H = 12 (a + c)e3 + 12 (e + g)e4 ,
B = 21 (a c)e3 + 12 (e g)e4 ,
C = be3 + f e4 .
The normal field H is called the mean curvature vector of M at p. The
map ⌘ can also be written in matrix form
✓ ◆
cos 2✓
⌘(✓) H = A , (7.3)
sin 2✓
with
✓1 ◆
A= 2 (a c) b
.
1
2 (e g) f
It follows from (7.3) that the image of the map ⌘ : S 1 ⇢ Tp M ! Np M
is an ellipse in the normal plane Np M with centre H and principal axes
along the vectors B and C.

Definition 7.1. The curvature ellipse of M at p is defined as the image of

the map ⌘ : S 1 ⇢ Tp M ! Np M.

Rotations in the tangent plane Tp M leave invariant the curvature ellipse

and rotations in the normal plane result in rotating the curvature ellipse
(see Propositions 7.2 and 7.3). Thus any isometric scalar invariant of the
curvature ellipse is an isometric scalar invariant of the surface (under the
action of SO(2) ⇥ SO(2) on Tp M ⇥ Np M ). For instance, the area of the
curvature ellipse (which is ⇡||N ||/2) and the length ||H|| of the mean cur-
vature vector are scalar invariants of the surface. Little defined another
scalar invariant of the surface, namely the resultant
a b c 0
1 e f g 0
= (7.4)
4 0 a b c
0 ef g
1
= 4(af eb)(bg f c) (ag ec)2
4
of the two polynomials ax2 +2bxy +cy 2 and ex2 +2f xy +gy 2 . The function
has the following property. The curvature ellipse passes through the point
p if and only if ⌘(✓) = 0 for some ✓. It follows from (7.2) that this happens
if and only if the above two polynomials have a common root. Therefore,
p is a point on the curvature ellipse if and only if (p) = 0. The point p
206 Differential Geometry from a Singularity Theory Viewpoint

$%&'()*+,% -,+%*.,&./(0'.*1-( $%&'()*+,% -,+%*.,&.+506+%0/1.*1-(

23!#4
23!"4 23!#4
! ! 23!"4

"!# "!#

:(5+;59+'+).-,+%* 80/09,'+).-,+%*

23!#4
23!"4

!
"!# "!#

<1-(/9,'+).-,+%* 7''+-*+).-,+%*

23!#4

!
23!"4
!
" !# "!#

!# !"

! $!#

Fig. 7.1 Classification of points on a surface in R4 in terms of the curvature ellipse.

being inside or outside the curvature ellipse can also be determined by the
sign of ∆(p), positive for inside and negative for outside.
Points on the surface are classified in terms of the curvature ellipse.

Definition 7.2. A point p ∈ M is called semiumbilic if the curvature ellipse

degenerates to a line segment that contains p. If the curvature ellipse is a
radial segment, the point p is called an inflection point. An inflection point
is of real type, (resp. imaginary type, flat) if p is an interior point of the
radial segment, (resp. does not belong to it, is one of its end points).
 Surfaces in the Euclidean 4-space 207

When the curvature ellipse reduces to the point p, then p is said to be a flat
umbilic. A non inflection point p 2 M is called elliptic (resp. hyperbolic,
parabolic) when it lies inside (resp. outside, on) the curvature ellipse. See
Figure 7.1.
The notation hyperbolic, elliptic and parabolic point in Definition 7.2
has nothing to do with the sign of the Gaussian curvature. They are intro-
duced in [Mond (1982)] in an analogy to the behaviour of the asymptotic
directions and curves on surfaces in R3 (see §7.3).

7.2 Second order affine properties

The invariants in §7.1 are computed using an orthonormal frame

{e1 , e2 , e3 , e4 } with {e1 , e2 } a basis of Tp M and {e3 , e4 } is a basis of Np M
at each point p on the surface. Those invariants are under the action of
SO(2) ⇥ SO(2) on Tp M ⇥ Np M .
Given a local parametrisation x : U ! R4 of the surface, a practical
basis of the tangent plane Tp M to work with is {f1 , f2 } = {xu1 , xu2 }. Of
course, in general it is not an orthonormal basis of Tp M . One can com-
plete it at each point p on M to obtain a basis {f1 , f2 , f3 , f4 } of R4 varying
smoothly with p, with {f3 , f4 } a basis of Np M (which is also not necessar-
ily orthonormal). The vector fields fi , i = 1, . . . , 4 defined on M can be
extended locally to an open subset of R4 .
We are interested in the contact of the surface with hyperplanes, planes
and lines and this contact is affine invariant ([Bruce, Giblin and Tari
(1995)]). Thus, it is natural to consider (affine) invariants properties of
the surface under the action of GL(2, R) ⇥ GL(2, R) on Tp M ⇥ Np M , where
GL(2, R) denotes the general linear group. Given an orthonormal basis
{e1 , e2 , e3 , e4 } in Tp M ⇥ Np M , the above action can be viewed as a change
of basis in Tp M and Np M .
We analyse the e↵ect of this change of basis on the coefficients of the
second fundamental form and on the curvature ellipse. We assume that we
are given a frame {e1 , e2 , e✓3 , e4 } of◆M near p. We start with the GL(2, R)
↵1 ↵2
action on Tp M . Let ⌦ = represent an element in GL(2, R) and
1 2
write
f 1 = ↵ 1 e1 + 1 e2 ,
f 2 = ↵ 2 e1 + 2 e2 .
Then we have the following.
208 Di↵erential Geometry from a Singularity Theory Viewpoint

Proposition 7.1. Denote by a, b, c, e, f, g the coefficients of the sec-

ond fundamental form with respect to the basis {e1 , e2 , e3 , e4 } and by
l1 , m1 , n1 , l2 , m2 , n2 its coefficients with respect to the basis {f1 , f2 , e3 , e4 }.
Then
0 1 0 1 0 1 0 1
l1 a l2 e
@ m1 A = ⇤ @ b A and @ m2 A = ⇤ @ f A
n1 c n2 g
with
0 2 2
1
↵1 2↵1 1 1
⇤ = @ ↵1 ↵2 ↵1 2 + ↵2 1 1 2 A and det ⇤ = (det ⌦)3 .
↵22 2↵2 2 2
2

Proof. The proof is an elementary linear algebra exercise . Let w 2 Tp M

and denote by (w1 , w2 ) (resp. (w̄1 , w̄2 )) its coordinates with respect to the
basis {e1 , e2 } (resp. {f1 , f2 }) of Tp M . We have w1 = ↵1 w̄1 + ↵2 w̄2 and
w2 = 1 w̄1 + 2 w̄2 and IIp (w1 , w2 ) = (aw12 + 2bw1 w2 + cw2 )e3 + (ew12 +
2ew1 w2 + gw2 )e4 so that
hIIp (w1 , w2 ), e3 i = a(↵1 w̄1 + ↵2 w̄2 )2 + 2b(↵1 w̄1 + ↵2 w̄2 )( 1 w̄1 + 2 w̄2 )
+c( 1 w̄1 + 2 w̄2 )2
= (↵12 a + 2↵1 1 b + 12 c)w̄12
+2(↵1 ↵2 a + (↵1 2 + ↵2 1 )b + 1 2 c)w̄1 w̄2
+(↵22 a + 2↵2 2 b + 22 c)w̄22 .
It follows that
l1 = ↵12 a + 2↵1 1 b + 12 c,
m1 = ↵1 ↵2 a + (↵1 2 + ↵2 1 )b + 1 2 c,
n1 = ↵22 a + 2↵2 2 b + 22 c.
The coefficients l2 , m2 , n2 can be obtained similarly using the scalar
product or IIp (w1 , w2 ) with e4 . ⇤
Remark 7.1. (1). The change of basis in Proposition 7.1 leaves the cur-
vature ellipse unchanged as a point set. It is parametrised as in (7.3)
where a, b, c, e, f, g in H and A are replaced by their expressions in terms of
l1 , m1 , n1 , l2 , m2 , n2 obtained by inverting the matrix ⇤ in Proposition 7.1.
(2). If we take f1 = xu1 and f2 = xu2 , then we can use the orthonormal
basis {e1 , e2 } in Tp M , with f1 = ↵1 e1 , f2 = ↵2 e1 + 2 e2 and
✓ ◆ p !
F
↵1 ↵2 E p E
= EG 2 ,
0 2 0 p F
E
with E, F, G the coefficients of the first fundamental form with respect to
the parametrisation x.
 Surfaces in the Euclidean 4-space 209

We turn now to the action of GL(2, R) on Np M and still represent by

✓ ◆
↵1 ↵2
⌦= and element in GL(2, R). We write
1 2

f 3 = ↵ 1 e3 + 1 e4 ,
f 4 = ↵ 2 e3 + 2 e4 .

and denote by
a = hxu1 u1 , e3 i, b = hxu1 u2 , e3 i, c = hxu2 u2 , e3 i,
e = hxu1 u1 , e4 i, f = hxu1 u2 , e4 i, g = hxu2 u2 , e4 i
and
l1 = hxu1 u1 , f3 i, m1 = hxu1 u2 , f3 i, n1 = hxu2 u2 , f3 i,
l2 = hxu1 u1 , f4 i, m2 = hxu1 u2 , f4 i, n2 = hxu2 u2 , f4 i
the coefficients of the second fundamental form with respect to the basis
{e3 , e4 } and {f3 , f4 } respectively. Also denote by
En = hf3 , f3 i, Fn = hf3 , f4 i, Gn = hf4 , f4 i.

Proposition 7.2. Denote by a, b, c, e, f, g the coefficients of the sec-

ond fundamental form with respect to the basis {e1 , e2 , e3 , e4 } and by
l1 , m1 , n1 , l2 , m2 , n2 its coefficients with respect to the basis {e1 , e2 , f3 , f4 }.
Then
✓ ◆ ✓ ◆✓ ◆ 1✓ ◆
ab c ↵1 ↵2 E n Fn l1 m1 n1
= .
ef g 1 2 Fn G n l2 m2 n2
Proof. We have
IIp (w1 , w2 ) = xu1 u1 w12 + 2xu1 u2 w1 w2 + xu2 u2 w22 = µ1 f3 + µ2 f4 .
The scalar product with f3 and f4 gives
hIIp (w1 , w2 ), f3 i = l1 w12 + 2m1 w1 w2 + n1 w22 = µ1 En + µ2 Fn ,
hIIp (w1 , w2 ), f4 i = l2 w12 + 2m2 w1 w2 + n2 w22 = µ1 Fn + µ2 Gn
so that
✓ ◆ ✓ ◆ 1 ✓ ◆
µ1 E n Fn l1 w12 + 2m1 w1 w2 + n1 w22
= .
µ2 Fn G n l2 w12 + 2m2 w1 w2 + n2 w22
We also have
IIp (w1 , w2 ) = µ 1 f3 + µ 2 f4
= µ1 (↵1 e3 + 1 e4 ) + µ2 (↵2 e3 + 2 e4 )
= (↵1 µ1 + ↵2 µ2 )e3 + ( 1 µ1 + 2 µ2 )e4
= (aw12 + 2bw1 w2 + cw22 )e3 + (ew12 + 2f w1 w2 + gw22 )e4
210 Di↵erential Geometry from a Singularity Theory Viewpoint

and from it we get

✓ ◆ ✓ ◆✓ ◆ 1 ✓ ◆
aw12 + 2bw1 w2 + cw22 ↵1 ↵2 En Fn l1 w12 + 2m1 w1 w2 + n1 w22
=
ew12 + 2f w1 w2 + gw22 1 2 Fn G n l2 w12 + 2m2 w1 w2 + n2 w22
which gives
✓ ◆ ✓ ◆✓ ◆ 1 ✓ ◆
ab c ↵1 ↵2 E n Fn l1 m1 n1
= .
ef g 1 2 Fn G n l2 m2 n2 ⇤

Proposition 7.3. With notation as in Proposition 7.2, the image of the

curvature ellipse by the affine transformation ⌦ is an ellipse parametrised
by
✓ ◆ ✓ ◆ 1 ✓ ◆ 1 ✓ ◆
µ1 ↵1 ↵2 ↵1 ↵2 cos 2✓
⌘(✓) = = H+ A (7.5)
µ2 1 2 1 2 sin 2✓
with respect to the basis {f3 , f4 } of Np M .

Proof. The curvature ellipse is parametrised by

⌘(✓) = xu1 u1 cos2 ✓ + 2xu1 u2 cos ✓ sin ✓ + xu2 u2 sin2 ✓
= µ 1 f3 + µ 2 f4 .
It follows from the proof of Proposition 7.2 that
✓ ◆✓ ◆ ✓ ◆
E n Fn µ1 l1 cos2 ✓ + 2m1 cos ✓ sin ✓ + n1 sin2 ✓
=
Fn G n µ2 l2 cos2 ✓ + 2m2 cos ✓ sin ✓ + n2 sin2 ✓
✓ ◆
cos 2✓
= H̄ + Ā ,
sin 2✓
with
✓ ◆✓ ◆ 1 ✓ ◆✓ ◆ 1
E n Fn ↵1 ↵2 E n Fn ↵1 ↵2
H̄ = H and Ā = A.
Fn G n 1 2 Fn G n 1 2
◆ 1 ✓
E n Fn
Multiplying both sides with gives the parametrisation
Fn G n
(7.5) of the image of curvature ellipse by the transformation ⌦ with re-
spect to the basis {f3 , f4 }. ⇤

Remark 7.2. It is clear from (7.5) that some properties of the curvature
ellipse (Definition 7.2) remain invariant under the action of GL(2, R) on
Np M . For instance, the position of the point p with respect to the curvature
ellipse is an affine invariant property. The concept of an inflection point is
also affine invariant. However, semiumbilicity is not affine invariant.
 Surfaces in the Euclidean 4-space 211

Table 7.1 The GL(2, R) ⇥ GL(2, R)-classes of pairs of

quadratic forms.
GL(2, R) ⇥ GL(2, R)-class Name
2
(x , y )2 hyperbolic point
(xy, x2 y 2 ) elliptic point
(x2 , xy) parabolic point
(x2 ± y 2 , 0) inflection point
(x2 , 0) degenerate inflection
(0, 0) degenerate inflection

The position of p with respect to the curvature ellipse is determined by

the sign of (p). Given a parametrisation x, denote by l1 , m1 , n1 , l2 , m2 , n2
the coefficients of the second fundamental form with respect to any basis
{xu1 , xu2 , f3 , f4 } of Tp M ⇥ Np M , and define ˜ by
l1 2m1 n1 0
˜ = 1 l2 2m2 n2 0 (7.6)
4 0 l1 2m1 n1
0 l2 2m2 n2
1
= 4(l1 m2 l2 m1 )(m1 n2 m2 n1 ) (l1 n2 l2 n1 )2 .
4
It follows from Propositions 7.1 and 7.2 that ˜ < 0 (resp = 0, > 0) if
and only if < 0 (resp = 0, > 0), as the two determinant di↵er by an even
power of the determinant of ⌦.

7.2.1 Pencils of quadratic forms

The second order affine invariants of M at p are described by the class of
the second fundamental form
IIp (w1 , w2 ) = (l1 w12 + 2m1 w1 w2 + n1 w22 , l2 w12 + 2m2 w1 w2 + n2 w22 )
under the action of GL(2, R)⇥GL(2, R). The second fundamental form at p
is a pair of quadratic forms (Q1 , Q2 ) and the action of GL(2, R) ⇥ GL(2, R)
on these pairs is as given in Propositions 7.1 and 7.2.

Theorem 7.2 ([Gibson (1979)]). The orbits of the action of GL(2, R)⇥
GL(2, R) on the set of pairs of quadratic forms are as in Table 7.1.

We have the following characterisation of the points of M in terms of

the sign of the invariant .
Proposition 7.4. Let M be a smooth surface in R4 and p a point on M .
Then
212 Di↵erential Geometry from a Singularity Theory Viewpoint

(i) (p) < 0 if and only if p is a hyperbolic point.

(ii) (p) > 0 if and only if p is a parabolic point.
(iii) (p) = 0 and rank ↵(p) = 2 if and only if p is an elliptic point.
(iv) (p) = 0 and rank ↵(p) < 2 if and only if p is an inflection point.

Proof. All the properties in the statement of the proposition are invariant
under the action of GL(2, R)⇥GL(2, R) on Tp M ⇥Np M . Thus, it is enough
to take (Q1 , Q2 ) as one of the normal forms in Table 7.1. After that, the
proof follows by straightforward calculations. ⇤

Corollary 7.1. The inflection points are the common solutions of the 2 ⇥ 2
minors of the matrix
✓ ◆
l1 m1 n1
↵(u) = (u),
l2 m2 n2
of the second fundamental form. That is, p = x(u) is an inflection point if
and only if
(l1 m2 m1 l2 )(u) = 0,
(l1 n2 n1 l2 )(u) = 0,
(m1 n2 n1 m2 )(u) = 0.

A quadratic form Aw12 + 2Bw1 w2 + Cw22 can be represented by the

point Q = (A : B : C) in the real projective plane RP2 . Denote by
= {Q | B 2 AC = 0} the conic of degenerate quadratic forms in RP2 .
The second order affine invariant of M at p can also be described via
pencils of quadratic forms determined by (Q1 , Q2 ) ([Bruce and Nogueira
(1998)] and also [Mochida (1993)]).

Proposition 7.5. With notation as above,

(i) If Q1 , Q2 are distinct, they determine a line in RP2 which meets the
conic in 0 (resp. 1, 2) points according to (p) < 0 (resp. = 0, > 0)
with

(p) = (l1 n2 l 2 n1 ) 2 4(l1 m2 l2 m1 )(m1 n2 n1 m2 ) = 4 ˜ (p).

In particular, p is a hyperbolic (resp. parabolic, elliptic) point if and

only if the pencil intersect the conic in 2 (resp. 1, 0) points.
(ii) If the quadratic forms Q1 and Q2 are linearly dependent but not iden-
tically zero, the pencil determines a point in RP2 which may lie inside,
on or outside the conic .
 Surfaces in the Euclidean 4-space 213

Proof. (i) If Q1 and Q2 are distinct, they determine a line in RP2

parametrized by ↵Q1 + Q2 , with (↵ : ) 2 RP1 . The intersection points
of this line with are given by the solutions of the quadratic equation

(↵l1 + l2 )(↵n1 + n2 ) (↵m1 + m2 )2 = 0,

equivalently,

(l1 n1 m21 )↵2 (l1 n2 + l2 n1 2m1 m2 )↵ + (l2 n2 m22 ) 2

= 0.

The number of the real solutions of the above equation is determined

by its discriminant
(p) = (l1 n2 + l2 n1 2m1 m2 )2 4(l1 n1 m21 )(l2 n2 m22 )
= (l1 n2 l2 n1 )2 4(l1 m2 l2 m1 )(m1 n2 n1 m2 )
= 4 ˜ (p).
The rest follows by Proposition 7.4.
(ii) Here Q2 and Q1 determine the same point in the projective plane.
The point is outside (resp. on, inside) if the quadratic forms have two
(resp. one, no) real roots. ⇤

7.3 Asymptotic directions

Definition 7.3. Let ⌘(✓) be the parametrisation (7.2) of the curvature

ellipse. A tangent direction ✓ at a point p 2 M is an asymptotic direction
at p if ⌘(✓) and d⌘
d✓ (✓) are linearly dependent vectors in Np M .
A curve on M whose tangent direction at each point is an asymptotic
direction is called an asymptotic curve.

By definition, a direction ✓ is asymptotic if the line joining p and ⌘(✓)

is tangent to the curvature ellipse. This is invariant under the action of
GL(2, R) in the normal plane, so the concept of asymptotic direction is
affine invariant (we will show this too using the contact of the surface with
hyperplanes, planes and lines). This means that there are 2 asymptotic
directions at a hyperbolic point, one at a parabolic point and none at an
elliptic point.

Theorem 7.3. Let x : U ! R4 be a local parametrisation of a surface M

and denote by l1 , m1 , n1 , l2 , m2 , n2 the coefficients of its second fundamen-
tal form with respect to any frame {xu1 , xu2 , f3 , f4 } of Tp M ⇥ Np M which
214 Di↵erential Geometry from a Singularity Theory Viewpoint

depends smoothly on p = x(u1 , u2 ). Then the asymptotic curves of M are

the solution curves of the binary di↵erential equation
(l1 m2 l2 m1 )du21 + (l1 n2 l2 n1 )du1 du2 + (m1 n2 m2 n1 )du22 = 0, (7.7)
which can also be written in the following determinant form
du22 du1 du2 du21
l1 m1 n1 = 0.
l2 m2 n2
Proof. We first write the second fundamental form with respect to an
orthonormal frame {e1 , e2 , e3 , e4 } of Tp M ⇥ Np M at p. By definition, a
direction ✓ at p 2 M is asymptotic if and only if
@⌘
⌘(✓) =
(✓),
@✓
with ⌘(✓) as in (7.2). Di↵erentiating the above equality gives
l1 cos2 ✓ + 2m1 cos ✓ sin ✓ + n1 sin2 ✓ = 2 ((n1 l1 ) cos ✓ sin ✓ + m1 (cos2 ✓ sin2 ✓))
l2 cos2 ✓ + 2m2 cos ✓ sin ✓ + n2 sin2 ✓ = 2 ((n2 l2 ) cos ✓ sin ✓ + m2 (cos2 ✓ sin2 ✓))

Eliminating gives
(af be) cos2 ✓ + (ag ce) cos ✓ sin ✓ + (bg cf ) sin2 ✓ = 0. (7.8)
Consider now a frame {xu1 , xu2 , e3 , e4 } of Tp M ⇥ Np M , write
✓ ◆ ✓ ◆ ✓ ◆✓ ◆
cos ✓ du1 ↵1 ↵2 du1
=⌦ =
sin ✓ du2 1 2 du2
and use the relations in Proposition 7.1. Then equation (7.8) becomes
1
(l1 m2 l2 m1 )du21 + (l1 n2 l2 n1 )du1 du2 + (m1 n2 m2 n1 )du22 = 0,
det ⌦
that is
(l1 m2 l2 m1 )du21 + (l1 n2 l2 n1 )du1 du2 + (m1 n2 m2 n1 )du22 = 0.
Similarly, using the relations in Proposition 7.2, equation (7.8) becomes
equation (7.7) with respect to a frame {e1 , e2 , f3 , f4 } in Tp M ⇥ Np M . ⇤
An immediate consequence of Theorem 7.3 is the following.

Corollary 7.2. Let M be a smooth surface in R4 .

(i) The coefficients of the binary di↵erential equation of the asymptotic
curves (7.7) all vanish at a point p if and only if p is an inflection
point. At such points all tangent directions are asymptotic directions.
 Surfaces in the Euclidean 4-space 215

˜ 1 , u2 ). Consequently, there
(ii) The discriminant of equation (7.7) is −4∆(u
are 2 (resp. 1, 0) asymptotic directions at hyperbolic (resp. not inflec-
tion parabolic, elliptic) points.

Remark 7.3. Asymptotic curves determine a pair of regular foliations on

the hyperbolic region of the surface. At generic points on the parabolic
curve, they form a family of cusps with the cusps tracing the parabolic
set. At isolated parabolic points (see Remark 7.7), the configuration of
the asymptotic curves is as in Figure 7.2 ([Bruce and Tari (2002)]). At
inflection points, the parabolic set generically has a Morse singularity. The
singularity is of type A+1 at an inflection point of imaginary type and of
−
type A1 at an inflection point of real type. The generic configurations of
the asymptotic curves at an inflection point of imaginary type are as in
Figure 7.3 ([Garcia, Mochida, Romero Fuster and Ruas (2000)]) and those
at an inflection point of real type are as in Figure 7.4 ([Bruce and Tari
(2002)]).

Fig. 7.2 Topological configurations of the asymptotic curves at special parabolic points.

Fig. 7.3 Generic topological configurations of the asymptotic curves at an inflection

point of imaginary type.

Equation (7.7) is, at each point on the surface, a quadratic form in

du1 , du2 . As multiplying the equation by a nowhere vanishing function gives
an equation with the same solutions as the original one, we can represent
(7.7), at each point on the surface, by a point A = (α : β : γ) in the
real projective plane RP2 , with α = l1 m2 − l2 m1 , β = l1 n2 − l2 n1 , γ =
m1 n2 − m2 n1 the coefficients of (7.7).
216 Differential Geometry from a Singularity Theory Viewpoint

Fig. 7.4 Generic topological configurations of the asymptotic curves at an inflection

point of real type.

The polar line Q ! of a point Q in RP2 (with respect to the conic Γ of

degenerate quadratic forms) is the line that contains all points Q such that
Q and Q′ are harmonic conjugate points with respect to the intersection
points R1 and R2 of the conic Γ and a variable line through Q. Geometri-
cally, if the polar line Q ! meets Γ, then the tangents to Γ at the points of
intersection meet at Q (Figure 7.5). Furthermore, if Q = ω1 ω2 , then ω12
and ω22 are the points of intersection of Q ! with Γ. A point (a′ : 2b′ : c′ ) is
on the polar line of a point Q = (a : b : c), if and only if ac′ − bb′ + ca′ = 0.
Three points in the projective plane are said to form a self-polar triangle
if the polar of any vertex of the triangle is the line through the remaining
two points.
Let {xu1 , xu2 , e3 , e4 } be a frame in R4 with {e3 , e4 } and orthonormal
basis of Np M . Denote by li = ⟨xu1 u1 , e2+i ⟩, mi = ⟨xu1 u2 , e2+i ⟩, ni =
⟨xu2 u2 , e2+i ⟩, i = 1, 2, the coefficient of the second fundamental form with
respect to the above frame.
Given a vector v = λ1 e3 + λ2 e4 in Np M , consider the shape operator
ν
Wp : Tp M → Tp M along any normal vector field with ν(p) = v. Then the
matrix of Wpν with respect to the basis {xu1 , xu2 } is the symmetric matrix
" #
λ 1 l 1 + λ 2 l 2 λ 1 m1 + λ 2 m2
.
λ1 m1 + λ2 m2 λ1 n1 + λ2 n2
We can represent Wpν by the point (λ1 l1 + λ2 l2 : 2(λ1 m1 + λ2 m2 ) :
λ1 n1 + λ2 n2 ) in R4 . Varying (λ1 : λ2 ) ∈ RP1 gives a pencil in RP1 , which
we call the pencil of the ν-shape operators at p.
Theorem 7.4. At each point p on a surface M , the polar line of the asymp-
totic curves BDE (7.7) is the pencil of the ν-shape operators at p; see
Figure 7.5.

Proof. A point Q = (a′ : 2b′ : c′ ) is on the pencil of the ν-shape operators

at p if and only if there exists λ1 and λ2 such that
λ1 l1 + λ2 l2 = a′ , λ1 m1 + λ2 m2 = b′ , λ1 n1 + λ2 n2 = c′ .
 Surfaces in the Euclidean 4-space 217

%
$
$&!"!#

!##

*+)

% !"# (
(
'
)

%
*+)

Fig. 7.5 Polar line of the asymptotic BDE tracing the pencil of ν-shape operators, and
the determination of a special shape operator.

The above system of three equations in λ1 and λ2 has a solution if and only
if

(l1 m2 − l2 m1 )c′ − (l1 n2 − l2 n1 )b′ + (m1 n2 − m2 n1 )c′ = 0

which is precisely the condition for the point Q to be on the polar line of
asymptotic BDE A. !

Remark 7.4. (1) The induced metric on M can also be represented at each
point by the point L = (E : 2F : G) in RP2 . The point L lies inside the
conic Γ as the metric is Riemannian. Its polar line L̂ represents BDE with
orthogonal solutions at p and intersect the polar line  of the asymptotic
BDE (7.7) at a unique point P (Figure 7.5). There is a unique point WpνP
on  such that A, C, WpνP form a self-polar triangle (see Figure 7.5). In fact
the solutions of P are the νP -principal directions at p. This construction
gives a way of choosing a unique pair of orthogonal foliation on M coming
from a ν-shape operator, which can be called the lines of principal curvature
of M .
(2) The above construction is in fact inspired from and is analogous to
that for surfaces in R3 . In that case too the asymptotic curves are given by
a BDE A and the lines of principal curvatures by a BDE P (see Chapter
6). There is a unique BDE C such that A, P, C form a self-polar triangle,
which is the BDE of characteristic curves ([Bruce and Tari (2005)]). In fact
using the metric L as above, the triple A, P, C is completely determined by
A. The construction runs into difficulties for surfaces immersed in higher
218 Di↵erential Geometry from a Singularity Theory Viewpoint

dimensional Euclidean spaces. For surfaces in R5 the asymptotic curves

are given a binary quintic di↵erential equation ([Mochida, Romero Fuster
and Ruas (1999); Romero Fuster, Ruas and Tari (2008)]). An attempt to
choose special ⌫-lines of curvature is made in [Ruas and Tari (2012)] using
invariants and covariants of binary forms (see for example [Olver (1999)]).
(3) More on binary di↵erential equations and polarity in the projective
plane with respect to can be found in [Bruce and Tari (2005); Oliver
(2010, 2011); Tari (2009)]. The case when the metric is Lorentzian (L is
outside ) is treated in [Nabarro and Tari (2011)].

7.4 Surfaces in Monge form

At each point p on the surface M , we can choose a coordinate system

q = (x, y, z, w) in R4 so that p is the origin, Tp M is the plane z = w = 0
and the surface M is locally the graph of some smooth function f : U ! R2 ,
with (x, y) in a open subset U of R2 containing the origin. Let (f1 , f2 ) be the
coordinate functions of f. Then we have the Monge form parametrisation
(x, y) = (x, y, f1 (x, y), f2 (x, y)),
of M at p. The Taylor expansions of f1 and f2 at the origin have no constant
or linear terms. We use the following notation for the Taylor polynomials
of order k at the origin of f1 and f2
Pi=k Pj=i
j k f1 (x, y) = a20 x2 + a21 xy + a22 y 2 + i=3 j=0 aij xi j y j ,
Pi=k Pj=i
j k f2 (x, y) = b20 x2 + b21 xy + b22 y 2 + i=3 j=0 bij xi j y j .
The following is straightforward.

Lemma 7.1. Let M be parametrised locally in Monge form (x, y) =

(x, y, f1 (x, y), f2 (x, y)) at the origin p. Then,

(i) E = 1 + f1 2x + f2 2x , F = f1 x f1 y + f2 x f2 y , G = 1 + f1 2y + f2 2y .
(ii) We can choose the frame F = {f1 , f2 , f3 , f4 } in U with
f1 = x = (1, 0, f1 x , f2 x )
f2 = y = (0, 1, f1 y , f2 y )
f3 = ( f1 x , f1 y , 1, 0)
f4 = ( f2 x + f1 y (f1 x f2 y f1 y f2 x ), f2 y f1 x (f1 x f2 y f1 y f2 x ),
f1 x f2 x f1 y f2 y , 1 + f1 x 2 + f2 x 2 )
The frame F at the origin is the standard basis of R4 .
 Surfaces in the Euclidean 4-space 219

Proposition 7.6. Let M be parametrised locally in Monge form (x, y) =

(x, y, f1 (x, y), f2 (x, y)) at the origin p. The second fundamental form at p
with respect to the frame F in Lemma 7.1 is given by
IIp (w) = 2(a20 w12 + a21 w1 w2 + a22 w22 , b20 w12 + b21 w1 w2 + b22 w22 ), (7.9)
with w = (w1 , w2 , 0, 0) in Tp M , and
(p) = 4 (a20 b21 b20 a21 )(a21 b22 b21 a22 ) (a20 b22 b20 a22 )2 .
Proof. The frame F at the origin is the standard basis of R4 , so the
coefficients of the second fundamental form of M at the origin are
l1 = f1 xx (0) = 2a20 , m1 = f1 xy (0) = a21 , n1 = f1 yy (0) = 2a22 ,
l2 = f2 xx (0) = 2b20 , m2 = f2 xy (0) = b21 , n2 = f2 yy (0) = 2b22 . ⇤

7.5 Examples of surfaces in R4

We present below three examples of orientable surfaces in 4-space, namely

surfaces in S 3 , complex curves and S 1 -bundles, taken from [Garcia,
Mochida, Romero Fuster and Ruas (2000)].

Example 7.1 (Surfaces in S 3 ). Examples of surfaces in the unit sphere

S 3 ⇢ R4 can be obtained as images of surfaces in 3-space by the inverse of
the stereographic projection ' : R3 ! S 3 ⇢ R4 .
Let (x, y, z) and (X, Y, Z, W ) denote the coordinates of R3 and R4 ,
respectively. Then, '(x, y, z) = (X, Y, Z, W ), with
2x 2y 2z 1+⇢
X= , Y = , Z= , W =
1+⇢ 1+⇢ 1+⇢ 1+⇢
2 2 2
and ⇢ = x + y + z .
If M is a surface in R3 parametrised in Monge form by = (x, y, f (x, y))
with (x, y) in an open set U of R2 . The = ' : U ! R4 is a parametri-
sation of '(M ). The (X, Y )-plane and the (Z, W )-plane are respectively
the tangent and normal planes to '(M ) at the origin. We have by di↵er-
entiating
xx (0) = (0, 0, 2fxx (0), 2),
xy (0) = (0, 0, 2fxy (0), 0),
yy (0) = (0, 0, 2fyy (0), 2).
The matrix of the second fundamental form of '(M ) at the origin p
(with respect to the standard basis of R4 ) is
✓ ◆
2fxx (0) 2fxy (0) 2fyy (0)
↵(p) = (0).
2 0 2
220 Di↵erential Geometry from a Singularity Theory Viewpoint

Since one of the rows of ↵(0) is a definite positive quadratic form, it

follows that (0)  0. Moreover, rank ↵(0) = 1 if and only fxy (0) = 0 and
fxx (0) = fyy (0). It follows that '(M ) has no elliptic points. It also follows
by Theorem 4.11 that a point p 2 M is an umbilic point if and only if
'(p) is an inflection point of imaginary type of '(M ). In particular, the
image by ' of any ellipsoid in R3 with three distinct axes is an example of
a 2-sphere in R4 with 4 inflection points.
Sufaces in S 3 are also considered in [Nagai (2012)] using Legendrian
dualities.

Example 7.2 (Complex curves). An example of a surface with bound-

ary in R4 is given by an algebraic regular complex curve in C2 defined as a
level set of a polynomial function f : C2 ! C.
Let c be a regular value of f and let p = (z0 , w0 ) be a point on the level
set M = {(z, w)|f (z, w) = c}. By the inverse function theorem f (z, w) = c
if and only if z = g(w) in some neighbourhood of p for some holomorphic
function g. Hence, (g(w), w) parametrises locally the surface M in a neigh-
bourhood U of w0 . Writing w = x+iy, g(w) = g1 (w)+ig2 (w) and using the
Cauchy-Riemann equations @g @g2
@x = @y , @y =
1 @g1 @g2
@x , we get the following
matrix of the second fundamental form of M at w0
!
@ 2 g1 @ 2 g2 @ 2 g1
↵(w0 ) = @x2 (w0 ) @x2 (w0 ) @x2 (w0 ) .
@ 2 g2 @ 2 g1 @ 2 g2
@x 2 (w 0 ) @x 2 (w 0 ) @x 2 (w 0 )

Hence,
✓ ◆2 ✓ ◆2 ! 2
˜ (w0 ) = @ 2 g1 @ 2 g1
+ (w0 ) 0.
@x2 @x2

We have ˜ (w0 ) = 0 if and only if all second order derivatives of g1 and

g2 at w0 are zero. That is, the inflection points of M are degenerate and
the non inflection points are all elliptic points. Moreover, one can show
that H(w) = 0 for all w 2 U . Thus, M is a minimal surface in R4 .

Example 7.3 (Torus embedded in R4 ). An example of a surface in R4

is an S 1 -bundle over a closed curve in R4 , with the fibres lying in the normal
space of the curve at each point.
Let fs be the one-parameter family of tori, parametrised by

fs (x, y) = (f1 (x, y, s), f2 (x, y), f3 (x, y), f4 (x, y)),
 Surfaces in the Euclidean 4-space 221

with
1 1
f1 (x, y, s) = (1 cos(y)) cos(x) + sin(x) sin(y) + s cos(y),
10 10
1 1
f2 (x, y) = (1 cos(y)) sin(x) cos(x) sin(y),
10 10
2 4
f3 (x, y) = (1 cos(y)) cos(2x) + sin(2x) sin(y),
5 5
2 4
f4 (x, y) = (1 cos(y)) sin(2x) cos(2x) sin(y).
5 5
The inflection points on the image of fs can be computed numerically
for a given value of the parameter s. These are the solutions of any two of
the following three equations
↵ = l1 m2 m1 l2 = 0,
= l 1 n2 n1 l 2 = 0,
= m1 n2 n1 m2 = 0,
where li , mi , ni , i = 1, 2 are the coefficients of the second fundamental form
with respect to any frame.
Figure 7.6 left (s = 1/100) and Figure 7.6 right (s = 1/20) are computer
plots of the curves = 0, ↵ = 0, = 0 and = 0. The surface for
s = 1/100 has no inflection points. When s = 1/20 the surface has 4
inflection points which can be depicted as the triple points formed by the
intersection of the three curves ↵ = 0, = 0 and = 0. In fact, one can
see from Figure 7.6 right that the parabolic curve does not pass through
these points (it has a Morse singularity of type A+ 1 at such points), so the
inflection points are of imaginary type.

7.6 Contact with hyperplanes

The contact of a smooth surface M in R4 with hyperplanes is measured by

the singularities of the height functions on M. Given a local parametrisation
x : U ! R4 of M , the family of height functions H : U ⇥ S 3 ! R on M is
given by
H(u, v) = hx(u), vi.
For v fixed, we have the height function hv on M given by hv (u) =
H(u, v). A point p = x(u) is a singular point of hv if and only if v is a
normal vector to M at p.
222 Differential Geometry from a Singularity Theory Viewpoint

Fig. 7.6 No inflection points for s = 1/100 (left), and four inflection points for s = 1/20
(right) depicted as the points of intersection of the curves α = 0, β = 0 and γ = 0.

Theorem 7.5. There is an open and dense set OH in Imm(U, R4 ) such

that for any x ∈ OH , the surface M = x(U ) has the following properties.
For any v ∈ S 3 , the height function hv along the normal direction v at any
point p on M has only local singularities of R type A1 , A2 , A3 , A4 or D4 .
Furthermore, the singularities of hv are R+ -versally unfolded by the family
H.

Proof. The proof follows from Theorem 4.4. !

Definition 7.4. A hyperplane with orthogonal direction v is an osculating

hyperplane of M at p = x(u) if it is tangent to M at p and hv has a
degenerate (i.e., non Morse) singularity at u. We call the direction v a
binormal direction of M at p.

We parametrise M locally at p in Monge form φ(x, y) =

(x, y, f1 (x, y), f2 (x, y)), with f1 and f2 as in §7.4. For v = (v1 , v2 , v3 , v4 ) ∈
S 3 , we have
hv (x, y) = v1 x + v2 y + v3 f1 (x, y) + v4 f2 (x, y)
so that hv is singular at the origin p if and only if v1 = v2 = 0, that is
v ∈ Np M . Denote by
Q1 (x, y) = j 2 f1 (x, y) = a20 x1 + a21 xy + a22 y 2 ,
Q2 (x, y) = j 2 f2 (x, y) = b20 x1 + b21 xy + c22 y 2 .
 Surfaces in the Euclidean 4-space 223

Then j 2 hv (x, y) is the pencil v3 Q1 (x, y) + v4 Q2 (x, y), and its R-singularity
type is completely determined by the GL(2, R) ⇥ GL(2, R)-class of the pair
(Q1 , Q2 ). In particular, we can take (Q1 , Q2 ) as in Table 7.1.
If p is a hyperbolic point, we take (Q1 , Q2 ) = (x2 , y 2 ) so that
j 2 hv (x, y) = v3 x2 + v4 y 2 . There are exactly two binormal directions at
p along (0, 0, 1, 0) and (0, 0, 0, 1).
If p is an elliptic point, we take (Q1 , Q2 ) = (xy, x2 y 2 ) so that
j 2 hv (x, y) = v3 xy + v4 (x2 y 2 ). The discriminant of this quadratic form
is 4v32 + v42 > 0, so the singularity of hv at the origin is always of type A1 .
Consequently, there are no binormal directions at p.
If p is a parabolic point, we take (Q1 , Q2 ) = (x2 , xy) so that j 2 hv (x, y) =
v3 x2 + v4 xy. There is one binormal direction at p along (0, 0, 1, 0).
Finally, if p is a non-degenerate inflection point and (Q1 , Q2 ) = (x2 ±
y 2 , 0), we get j 2 hv (x, y) = v3 (x2 ±y 2 ). The direction (0, 0, 0, 1) is the unique
binormal direction at p and the 2-jet of j 2 hv is identically zero along this
direction. We have thus the following proposition.

Proposition 7.7. Let M be a smooth surface immersed in R4 and let p be

a point on M.
(i) If p is an elliptic point, then hv has a non-degenerate singularity at p
of type A1 for all v in Np M.
(ii) If p is a hyperbolic point, then there are exactly two distinct binormal
directions v1 , v2 in Np M.
(iii) If p is a parabolic point but not an inflection point, then there is a
unique binormal direction v in Np M.
(iv) If p is a non-degenerate inflection point, then there is a unique binormal
direction v in Np M and the 2-jet of hv at p is identically zero.

Proposition 7.8. A normal direction v at p = x(u) is a binormal direction

if and only if any tangent direction lying in the kernel of the Hessian of hv
at u is an asymptotic direction of M at p.

Proof. Take the case p a hyperbolic point and j 2 hv (x, y) = v3 x2 + v4 y 2 .

The Hessian matrix of H(hv ) at the origin is the diagonal matrix with
entries 2v3 and 2v4 . When v3 = 0 (resp. v4 = 0), the kernel of H(hv )(0, 0)
is along (1, 0, 0, 0) (resp. (0, 1, 0, 0)). The directions (1, 0, 0, 0) and (0, 1, 0, 0)
are the asymptotic directions at p (see (7.7)). ⇤

Remark 7.5. It is not difficult to check that when v is a binormal direction

at p, its associated asymptotic direction u (i.e., a direction in the kernel
224 Di↵erential Geometry from a Singularity Theory Viewpoint

of H(hv )(p)) is an eigenvector of the shape operator Wpv along v and its
associated eigenvalue is zero. That is, u is v-principal direction.

Next, we characterise the degenerate singularity of hv at the origin in

terms of the coefficients in the Taylor expansion of the functions f1 and f2
giving M in Monge form. By a rotation at the origin in the normal plane,
we can assume that v0 = (0, 0, 0, 1) is a binormal direction and take nearby
directions in the form v = (v1 , v2 , v3 , 1), with vi , i = 1, 2, 3 small enough.
Then, the (modified) height function in the direction v is given by
hv (x, y) = v1 x + v2 y + v3 f1 (x, y) + f2 (x, y).

Proposition 7.9. With notation as in §7.4, the conditions for hv0 to have
one of the generic singularities at the origin are as follows
A2 : b20 6= 0, b33 6= 0
A3 : b20 6= 0, b33 = 0, 4b20 b44 b232 6= 0
A4 : b20 6= 0, b33 = 0, 4b20 b44 b232 = 0, 4b20 2 b55 2b20 b32 b43 + b31 b232 6= 0
D4 : b20 = 0, b30 x3 + b31 x2 y + b32 xy 2 + b33 y 3 is non-degenerate.
The above singularities are R+ -versally unfolded by the family of height
functions if and only if
A2 : always
A3 : a22 6= 0 or b32 6= 0
A4 : a22 (b20 b43 b31 b32 ) b32 (b20 a33 21 a21 b32 ) 6= 0
D4 : 3b30 (a22 b32 32 a21 b33 ) b31 (a22 b31 12 a21 b32 )
+a20 (3b31 b33 b232 ) 6= 0.
Proof. The height function on M along the normal direction v0 =
(0, 0, 0, 1) is given by hv0 (x, y) = f2 (x, y).
We first assume that b20 6= 0. Then, the singularity of hv0 at the origin
is of type Ak . In this case, hv0 has an A2 -singularity at the origin if and
only if b33 6= 0. If b33 = 0, we can make changes of coordinates in the
source to reduce the 4-jet of f2 to j 4 f2 (x, y) = b20 x2 + b32 xy 2 + b44 y 4 .
Then, the singularity is of type A3 if and only if b33 = 0 and b20 x2 +
b32 xy 2 + b44 y 4 is not a perfect square, equivalently, b232 4b20 b44 6= 0.
If b33 = 0 and b20 x2 + b32 xy 2 + b44 y 4 is a perfect square, we can make
changes of coordinates in the source and show that j 5 f2 ⇠R(5) b20 x2 + Dy 5 ,
with D = a22 (b20 b43 b31 b32 ) b32 (b20 a33 12 a21 b32 ). Hence, hv0 has A4
singularity if and only if D 6= 0.
If b20 = 0, the result follows from the characterisation of a D4
singularity.
 Surfaces in the Euclidean 4-space 225

The modified family of height functions H : R2 ⇥ R3 , (0, v0 ) ! R is

given by
H(x, y, v1 , v2 , v3 ) = v1 x + v2 y + v3 f1 (x, y) + f2 (x, y).
The family H is an R+ -versal deformation if and only if
n o
LRe · f2 + R. Ḣ1 , Ḣ2 , Ḣ3 + h1iR = E2 (7.10)
(see Theorem 3.12). We have
Ḣ1 (x, y) = Hv1 (x, y, 0, 0, 0) = x,
Ḣ2 (x, y) = Hv2 (x, y, 0, 0, 0) = y,
Ḣ3 (x, y) = Hv3 (x, y, 0, 0, 0) = f1 (x, y).
We carry out the calculations for the A3 -singularity. We have
j 3 f2 x = 2b20 x + 3b30 x2 + 2b31 xy + b32 y 2 +
4b40 x3 + 3b41 x2 y + 2b42 xy 2 + b43 y 3 .
j 3 f2 y = b31 x2 + 2b32 xy + 3b33 y 2 +
b41 x3 + 2b42 x2 y + 3b43 xy 2 + 4b44 y 3 .
Since f2 is 4-R-determined, it is enough to verify that (7.10) holds in
J (2, 1). As b20 6= 0, for any d 2 and k 1, we have (xk 1 y d k )j 3 f2 x 2
4

LRe · f2 , hence
xk y d k
2 J d (LRe · f2 ). (7.11)
Then, when d = 2, if b32 6= 0 or a22 6= 0, we can solve (7.10) in J 2 (2, 1).
When d = 3 and 4, writing y i j 3 f2 x , and y i 1 j 3 f2 y i = 1, 2 and using
(7.11) to eliminate the monomials of degree 2 + i, i = 1, 2 divisible by x,
we can write the equations
2b20 xy i + b32 y 2+i 2 J 2+i (LRe · f2 ), i = 1, 2,
b31 x2 y i 1 + 2b32 xy i + 4b44 y 3+i 1 2 J 2+i (LRe · f2 ), i = 1, 2,
b31 x2 y i 1 2 J 2+i (LRe · f2 ).
Since 4b20 b44 b232 6= 0, it follows that all monomials of degree 3 and 4 are
in J 4 (LRe · f2 ). ⇤

7.6.1 The canal hypersurface

The canal hypersurface of the surface M = x(U ) ⇢ R4 is the 3-manifold
CM (") = {p + "v 2 R4 | p 2 M and v 2 (Np M )1 }
where (Np M )1 denotes the unit sphere in Np M and " is a sufficiently small
positive real number chosen so that CM (") is a regular hypersurface (see
226 Di↵erential Geometry from a Singularity Theory Viewpoint

Theorem 2.7). We can consider (Np M )1 a subset of S 3 and identify (p, v)

and p + "v.
Let H̄ : CM (") ⇥ S 3 ! R denote the family of height functions on
CM ("). For w 2 S 3 , let h̄w : CM (") ! R, be given by h̄w (p, v) =
H̄((p, v), w). Since CM (") is a hypersurface in R4 , we have the Gauss
map G : CM (") ! S 3 . By Proposition 2.12, it is given by G(p, v) = v. We
call it the generalised Gauss map of M .

Proposition 7.10. Let M be a smooth surface in R4 and let p be a point

on M.

(i) The point p is a singular point of hv if and only if (p, v) 2 CM (") is a

singular point of h̄v . Furthermore, the R-singularity type of hv at p is
the same as that of h̄v at (p, v).
(ii) A point (p, v) is a degenerate singular point of h̄v if and only if (p, v)
is a singular point of the generalised Gauss map G.

Proof. The proof follows by Proposition 2.12 and Corollary 2.4. ⇤

Theorem 7.6. There is an open and dense set OH̄ in Imm(U, R4 ) such
that for any x 2 OH̄ , the surface M = x(U ) satisfies the conditions in
Theorem 7.5. Moreover, the following properties hold.

(i) For any v 2 S 3 , the height function h̄v on CM (") has only local sin-
gularities of R type A1 , A2 , A3 , A4 or D4 .
(ii) The singularities of h̄v are R+ -versally unfolded by the family H̄.
(iii) The generalised Gauss map G is stable as a Lagrangian map. The
singularities of G are given in Table 7.2.

Proof. By Proposition 5.14, H̄ is a Morse family of functions and the gen-

eralised Gauss map is the Lagrangian map of L(H̄). We apply Theorem 4.4
0
to obtain an open and dense set OH̄ of immersions for which H̄ is an R+ -
versal unfolding of h̄v . For any immersion x in this set, the associated
generalised Gauss mapping G : CM (") ! S 3 is Lagrangian stable. We
0
take OH̄ as the intersection of OH̄ and the set OH in Theorem 7.5. The
singularities of the generalised Gauss maps appearing in Table 7.2 are the
Lagrangian stable singularities when n  3 from Table 5.1. ⇤

Definition 7.5. A surface is called height function generic if any of its

local parametrisation belongs to the set OH̄ in Theorem 7.6.
 Surfaces in the Euclidean 4-space 227

Table 7.2 Lagrangian stable singularities of generalised Gauss maps of sur-

faces in R4 .
h̄v singularity type G singularity type Normal form
A1 Immersion (x, y, v3 )
A2 Fold (x, y 2 , v3 )
A3 Cuspidal edge (x, xy + y 3 , v3 )
A4 Swallowtail (x, y 4 + xy + v3 y 2 , v3 )
D4+ Hyperbolic umbilic (x2 + v3 y, y 2 + v3 x, v3 )
D4 Elliptic umbilic (x2 y 2 + v3 x, xy + v3 y, v3 )

Proposition 7.11. Let M be a height function generic surface in R4 . Let

Kc : CM (") ! R be the Gauss-Kronecker curvature function of CM (").
Then the following hold.
(i) The singular set of G is the parabolic set Kc 1 (0) of CM (").
(ii) Kc 1 (0) := {p + "v 2 CM (") | hv has a degenerate singularity at p}.
(iii) Kc 1 (0) is a regular surface except at a finite number of singular points
corresponding to the D4± -singularities of h̄v .

Proof. The proof of part (i) is analogous to the proof of Proposition

6.9, part (ii) follows from Proposition 7.10, and part (iii) follows from the
normal forms (1), (2), (3), (5) and (6) in Theorem 5.5. The D4± are the
corank 2 singularities of G, so they correspond to the singular points of
Kc 1 (0). ⇤

Remark 7.6. Let M be a height function generic surface, so that the

generalised Gauss map G is a Lagrangian stable map. We use the Thom-
Boardman symbols (see §3.4) to distinguish the singular points of G. Let
Kc 1 (0) = S1 (G) [ S2 (G), where S1 (G) are the singular points of corank 1
and S2 (G) are the singular points of corank 2 of G. Since G is a Lagrangian
stable map, the set S1 (G) is a smooth surface which can be decomposed
as S1,0 (G) [ S1,1,0 (G) [ S1,1,1,0 (G). The points in S1,0 (G) are the regular
points of the restriction of the map G to S1 (G); they are the fold points of
G. The singular set of G|S1 (G) is the smooth curve S1,1 (G) = S1,1,0 (G) [
S1,1,1,0 (G). The restriction of G to this set is regular on S1,1,0 (G) and
singular on S1,1,1,0 (G); they are respectively the cuspidal edge points and
the swallowtail points of G. These geometric conditions are verified for the
normal forms in Table 7.2.
At each point (p, v) 2 Kc 1 (0) \ S2 (G), the unique kernel direction of
dGp is the principal direction of zero curvature of CM (") at p. This direc-
tion coincides with the kernel of the Hessian quadratic form of the height
function in the normal direction v on CM (") (see the proof of Proposition
228 Di↵erential Geometry from a Singularity Theory Viewpoint

2.13). It follows from the definition of the sets S1,0 (G), S1,1,0 (G), S1,1,1,0 (G)
that this direction is transverse to the surface S1,0 (G) and is tangent to the
surface Kc 1 (0) on the curve of points of type S1,1 (G). It is transverse to
this curve at general points and tangent to it at points of type S1,1,1 (G).
Let ⇠ : CM (") ! M be the projection of CM (") to M , given by
⇠(p, v) = p. Then, it follows from part (ii) in Proposition 7.11 that the
image of the parabolic set Kc 1 (0) by ⇠ is the set  0. More precisely,
let ⇠¯ be the restriction of ⇠ to the regular surface S1 (G) = Kc 1 (0) \ S2 (G).
We denote by M = {p 2 M : (p) < 0} the set of hyperbolic points
on M and by B = {(p, v) 2 Kc 1 (0) : p 2 M }.
Proposition 7.12. With notation as above,
(i) the restriction ⇠¯|B : B ! M is a local di↵eomorphism, more precisely,
it is a double cover;
(ii) a point p 2 is not an inflection point if and only if there exists v 2 S 3
such that (p, v) is a fold singularity of ⇠¯ .
Proof. We take M locally in Monge form with f1 and f2 as in Section
7.4 and choose the setting in the proof of Proposition 7.9. The modified
height function on M along v = (v1 , v2 , v3 , 1) near v0 = (0, 0, 0, 1) is given
by
hv (x, y) = v1 x + v2 y + v3 f1 (x, y) + f2 (x, y).
The function hv has a singularity at a point (x, y) if and only if
@f1 @f2 @f1 @f2
v1 = v3 (x, y) (x, y) and v2 = v3 (x, y) (x, y).
@x @x @y @y
The singularity p = (x, y) is degenerate if furthermore det H(hv )(x, y) =
0, equivalently Kc (x, y, v3 ) = 0 (see Proposition 7.11). We have
detH(hv )(x, y) = A0 (x, y)v32 + A1 (x, y)v3 + A2 (x, y) = 0,
where
2 2
@ 2 f1 2
A0 (x, y) = @@xf21 (x, y) @@yf21 (x, y) ( @x@y ) (x, y)
@ 2 f1 @ 2 f2 @ 2 f1 @ 2 f2
A1 (x, y) = @x 2 (x, y) @y 2 (x, y) + @y 2 (x, y) @x2 (x, y)
@ 2 f1 @ 2 f2
(7.12)
2 @x@y (x, y) @x@y (x, y)
@ 2 f2 @ 2 f2 @ 2 f2 2
A2 (x, y) = @x 2 (x, y) @y 2 (x, y) ( @x@y ) (x, y).
Write K̃c = det(H(hv )) so that Kc = K̃c for some function with
(x, y, v3 ) 6= 0. Then, denoting by z any of the variables x, y, v3 , one can
show inductively that
@Kc @ m Kc @ m+1 Kc
Kc = = ... = = 0 and 6= 0
@z @z m @z m+1
 Surfaces in the Euclidean 4-space 229

at (x, y, v3 ) if and only if

@ K̃c @ m K̃c @ m+1 K̃c
K̃c = = ... = = 0 and 6= 0
@z @z m @z m+1
at (x, y, v3 ). For (i), we have v3 = 0 is a simple root of K̃c (x, y, v3 ) = 0
if and only if @@v K̃c
3
(x, y, 0) 6= 0. Equivalently, Kc (x, y, v3 ) = 0 if and only if
@Kc ¯
@v3 (x, y, 0) 6= 0. That is, ⇠ is a local di↵eomorphism.
For (ii), v3 = 0 is a double root of K̃c (x, y, v3 ) = 0 if and only if
@ K̃c @ 2 K̃c
@v3 (x, y, 0) = 0 and @v 2
(x, y, 0) 6= 0. Equivalently Kc (x, y, v3 ) = 0, if and
3
@Kc @ 2 Kc
only if @v3 (x, y, 0) = 0 and @v32
(x, y, 0) 6= 0. That is (p, v) is a fold point
¯
for ⇠. ⇤

Proposition 7.13. The zero curvature principal directions on the parabolic

set Kc 1 (0) \ S2 (G) are mapped by d⇠¯ to the asymptotic directions of M .

Proof. We choose the setting as in the proof of Proposition 7.12 and

take coordinates for M and CM (") such that p = (0, 0) and v = (0, 0, 0, 1).
Hence K̃c (p, v3 ) = A0 (x, y)v32 + A1 (x, y)v3 + A2 (x, y).
At p = (0, 0), v = (0, 0, 0, 1), the singularity of the Gauss map G :
CM (") ! S 3 is the singularity at the origin of the map-germ g : (R3 , 0) !
(R3 , 0), given by
@f1 @f2 @f1 @f2
g(x, y, v3 ) = ( v3 (x, y) (x, y), v3 (x, y) (x, y), v3 ).
@x @x @y @y
The eigenvectors of the matrices dG0 and dg0 are the same. Thus, if e2 =
(0, 1, 0, 0) 2 T(p,v) CM (") is the zero curvature direction, it follows that
b11 = b02 = 0. Then T ⇠(p, v)e2 = e2 2 Tp M is an asymptotic direction at
p. ⇤

7.6.2 Characterisation of the singularities of the height

function
We characterise geometrically the degenerate singularities of generic height
functions.
Denote by the normal section of M tangent to the asymptotic direction
✓ at p associated to the binormal direction v.
The characterisation of the singularities of the height functions at hy-
perbolic points is given below.

Theorem 7.7. Let p be a hyperbolic point on a height function generic

surface M. Then p is an A2 singularity of hv if and only if has non
230 Di↵erential Geometry from a Singularity Theory Viewpoint

vanishing normal torsion at p. If has a vanishing normal torsion at p,

then:

(i) p is an A3 singularity of hv if and only if the direction ✓ is transversal

¯ 1,1 (G)).
to the curve ⇠(S
(ii) p is an A4 singularity of hv if and only if the direction ✓ is tangent to
¯ 1,1 (G)) with first order contact at p.
the curve ⇠(S

Proof. We choose local coordinates as in Proposition 7.9. Let K̃c : R2 ⇥

R, 0 given by

K̃c (x, y, v3 ) = A0 (x, y)v32 + A1 (x, y)v3 + A2 (x, y),

where A0 , A1 , A2 are as in Equation (7.12).

The zero curvature direction of CM at (p, v) in these coordinates is the
y-axis. Then d⇠¯(p,v) e2 = e2 = ✓ (Proposition 7.13).
The normal section can be parametrised by

(s) = (0, s, a20 s2 + ..., b33 s3 + ...)

and it follows that has non-zero torsion if and only if b33 6= 0. This is the
case if and only if p is an A2 singularity of hv .
Now, it follows from Proposition 7.10 and Theorem 7.6 that a hyperbolic
point p 2 M (✏) is a singularity of type A2 , A3 , A4 of hv if and only if
(p, v) 2 CM is respectively a fold, cusp or swallowtail singularity of G.
Hence, if b33 = 0, the following follows from Remark 7.6.

(i) p is an A3 singularity of hv if and only if (p, v) 2 S1,1,0 (G), if and only

if the zero curvature direction is tangent to Kc 1 (0) and transverse to
the curve S1,1,0 (G).
(ii) p is an A4 singularity of hv if and only if (p, v) 2 S1,1,1,0 (G), and
this happens if and only if the zero principal direction is tangent to
S1,1,0 (G), with first order contact.

Since ⇠¯ : B ! M is a local di↵eomorphism, ✓ is transversal to

¯ 1,1,0 (G)) in (i) and tangent to ⇠(S
⇠(S ¯ 1,1,0 (G)) with first order contact in
(ii). ⇤

Definition 7.6. By analogy with the case of surfaces in R3 , we define the

flat rib set of a surface M in R4 as the subset of points (p, v) 2 CM (✏)
such that the height function hv on M has a singularity of type Ak , k 3
at p. We call its projection to M by ⇠ the flat ridge set of M.
 Surfaces in the Euclidean 4-space 231

We give now the characterisation of the singularities of the height func-

tions at parabolic points.
Theorem 7.8. Let M be a height function generic surface in R4 and p 2
M.
i) Suppose p is a parabolic point, but not an inflection point. Then
(a) p is an A2 -singularity of hv if and only if ✓ is transversal to the
parabolic curve .
(b) p is an A3 -singularity of hv if and only if ✓ is tangent to the parabolic
curve with first order contact.
ii) The point p is a D4 -singularity of hv if and only if p is an inflection
point of M. Moreover,
(c) p is a normal crossing point of if and only if p is an inflection
point of real type;
(d) p is an isolated point of if and only if p is an inflection point of
imaginary type.
Proof. (i) Observe first that the curves of points p in M = x(U ) such
that hv has a singularity of type A3 or more degenerate for some v 2
Np M, cannot meet the curve at a point of type A4 . This follows from
Thom transversality theorem by observing that for a point (x, y) 2 U to
be an A4 point of hv , the 4-jet extension j 4 h : U ⇥ S 3 ! J 4 (U, R), must
meet transversally the set of jets of type A4 in J 4 (U, R). This set is an
algebraic variety of codimension 5 in J 4 (U, R). The conditions on the 4-
jet j 4 to intercept this set transversally at the origin, where (x, y) =
(x, y, f1 (x, y), f2 (x, y)) is the Monge form embedding of M, are given in
Proposition 7.9. Now, as the point also belongs to , the image of j 4 h meets
another algebraic variety of codimension 1. Thus to have both conditions
at the same time, the map j 4 h meets the intersection of both varieties,
which has codimension 6 in J 4 (M, R). But this can be avoided by a generic
embedding of the surface.
We choose local coordinates as in Proposition 7.12. Let K̃c : R2 ⇥ R, 0
given by
K̃c (x, y, v3 ) = A0 (x, y)v32 + A1 (x, y)v3 + A2 (x, y),
where A0 , A1 , A2 are defined in (7.12).
Since p 2 , a22 = 0 and there is a unique binormal direction v at p.
The discriminant set
@ K̃c
{(x, y)| 9 v3 : K̃c (x, y, v3 ) = 0 and (x, y, v3 ) = 0}
@v3
232 Di↵erential Geometry from a Singularity Theory Viewpoint

1 2
is A0 (x, y)A2 (x, y) 4 A1 (x, y), which is exactly the set . Then:
(a) p is an A2 singularity of hv if and only if @A
@y (0, 0) = 12b20 b33 6= 0 if
2

and only if the asymptotic direction e2 is transversal to the curve .

(b) If p is an A3 singularity of hv then 4b20 b33 = 0, 12b20 b32 6= 0. Then,
the asymptotic direction e2 is tangent to the curve .
(ii) If the point p is a D4 -singularity of hv , then b20 = 0, and it follows
that rank↵(p) = 1, which implies that p is an inflection point. We can
verify conditions (c) and (d) by direct calculations. ⇤
The following result is a corollary of the proof of the above theorem.

Corollary 7.3. For a height function generic immersion, the flat ridge set
1
is a regular curve in M which is tangent to the curve (0).

7.7 Contact with lines

The family of orthogonal projections P : M ⇥ S 3 ! T S 3 of M to 3-spaces

is given by
P (p, v) = (p, Pv (p))
where Pv (p) = p hp, viv. For a given v 2 S 3 , the map Pv is singular at
p if and only if v is in Tp M .
The map Pv can be considered locally as a smooth map-germ R2 , 0 !
R3 , 0. Since P is a 3-parameter family, we expect the map Pv to have only
simple singularities of Ae -codimension  3 or non-simple singularities with
the Ae -codimension of the stratum  3. (The stratum is formed by the jets
in some k-jet space which are A(k) -equivalent to a member of the family
parametrised by the moduli). The expected singularities are extracted from
the classification in [Mond (1985)] and are as in Table 7.3. The following
result follows from Theorem 4.12.

Theorem 7.9. There is an open and dense set OP in Imm(U, R4 ) such

that for any x 2 OP , the surface M = x(U ) has the following properties.
(i) Given any point p 2 M , the projection Pv has local singularities of
A-type Sk , k = 0, 1, 2, 3, B2 , B3 , C3 , H2 , H3 or P3 (c) at p.
(ii) The simple singularities of Pv are Ae -versally unfolded by the family
P. The singularity P3 (c0 ), c0 6= 0, 12 , 1, 32 , is Ae -versally unfolded by the
family P with c near c0 included as a parameter in the family.
 Surfaces in the Euclidean 4-space 233

Table 7.3 Generic local singularities of Pv .

Name Normal form Ae -codimension
Immersion (x, y, 0) 0
Cross-cap (x, xy, y 2 ) 0
S1 =B1 (x, y 2 , y 3 ± x2 y) 1
S2 (x, y 2 , y 3 + x3 y) 2
S3 (x, y 2 , y 3 ± x4 y) 3
B2 (x, y 2 , x2 y ± y 5 ) 2
B3 (x, y 2 , x2 y ± y 7 ) 3
C3 (x, y 2 , xy 3 ± x3 y) 3
H2 (x, xy + y 5 , y 3 ) 2
H3 (x, xy + y 8 , y 3 ) 3
P3 (c) (x, xy + y 3 , xy 2 + cy 4 ), c 6= 0, 12 , 1, 3
2
4(3⇤ )
⇤ The codimension of stratum P3 is 3.

Definition 7.7. An immersion x : M ! R4 is projection P -generic if x

belongs to the set OP .

In §7.6, the second order geometry of a surface derived from its contact
with hyperplanes is determined by a pencil of quadratic forms. This pencil
also determines some of the second order geometry of the surface associated
to its contact with lines.
We take the surface in Monge form (x, y) = (x, y, f1 (x, y), f2 (x, y)),
at the origin p with f1 and f2 as in §7.4 and project along a unit tangent
direction v = (v1 , v2 , 0, 0) 2 Tp M (so that Pv is singular at p). Then
Pv (x, y) = (x, y) h (x, y), viv
= (x, y, f1 (x, y), f2 (x, y)) (xv1 + yv2 )(v1 , v2 , 0, 0)
= (x (xv1 + yv2 )v1 , y (xv1 + yv2 )v2 , f1 (x, y), f2 (x, y))
= (v2 (v2 x v1 y), v1 (v2 x v1 y), f1 (x, y), f2 (x, y)).
The image of Pv is in v? , the linear 3-space orthogonal to v. If we
denote by (X, Y, Z, W ) the coordinates in R4 , and if v2 6= 0, we can compose
Pv with the projection (X, Y, Z, W ) 7! (X, Z, W ) and obtain a map-germ
A-equivalent to Pv (the projection is a di↵eomorphism restricted to v? ).
If v2 = 0, we compose Pv with the projection (X, Y, Z, W ) 7! (Y, Z, W )
instead. In both cases we still denote by Pv the composite map. Rescaling
the first coordinate (or the second if v2 = 0) gives
Pv (x, y) = (v2 x v1 y, f1 (x, y), f2 (x, y)). (7.13)
We have j 2 Pv (x, y) = (v2 x v1 y, Q1 (x, y), Q2 (x, y)), with Q1 and Q2
representing j 2 f1 and j 2 f2 at the origin, respectively. We are interested in
the A-singularities of Pv . Then one of the conditions that determines the
234 Di↵erential Geometry from a Singularity Theory Viewpoint

A-type of Pv at the origin is whether or not v2 x v1 y is a common factor

of Q1 and Q2 . For this reason, following [Bruce and Nogueira (1998)],
we associate to each tangent direction v a line in RP2 which consists of
quadratic forms having (v2 x v1 y) as a factor. This line is the tangent line
to the conic at (v2 x v1 y)2 .

Proposition 7.14 ([Bruce and Nogueira (1998)]). (i) Suppose that

p is not an inflection point. The direction of projection yields a cross-
cap unless the line it determines passes through one of the points of
intersection of with the pencil (Q1 , Q2 ).
(ii) At an inflection point, where we have an associated single quadratic
form Q = Q1 = Q2 , most directions of projection yield an Sk or Bk
singularity, and we have a direction of projection of type Hk provided
the pencil it determines contains Q. If Q is inside (resp. outside) the
conic there are 0 (resp. 2) such projections.

Proof. (i) We prove the statement for the case p a hyperbolic or parabolic
point; the argument for the elliptic points is similar. We take (Q1 , Q2 ) =
(x2 , y 2 ) at a hyperbolic point so that j 2 Pv (x, y) = (v2 x v1 y, x2 , y 2 ). If v2 6=
0, we can make the change of coordinates X = v2 x v1 y, Y = y and obtain
a map-germ A(2) -equivalent to (X, 2v v2
1
XY, Y 2 ). We get a cross-cap if and
2
only if v1 6= 0. When v1 = 0, j 2 Pv is A(2) -equivalent to (X, 0, Y 2 ). If v2 = 0,
then j 2 Pv is A(2) -equivalent (Y, X 2 , 0). Therefore, the singularity is a cross-
cap unless v1 = 0 or v2 = 0. Observe that the degenerate singularities of
Pv are Sk , Bk or Ck .
At a parabolic point and with (Q1 , Q2 ) = (x2 , xy), we get j 2 Pv (x, y) =
(v2 x v1 y, x2 , xy). If v2 = 0, the singularity is a cross-cap. If v2 6= 0,
the same change of coordinates as above reduces the 2-jet to (X, 2v1 XY +
v12 Y 2 , XY +v1 Y 2 ) which is a cross-cap unless v1 = 0. If v1 = 0, j 2 Pv (x, y) is
A(2) -equivalent to (X, XY, 0), which leads to a projection P -generic surface
to H2 , H3 or P3 (c) singularity.
As before, the point (A : B : C) 2 RP2 represents the quadratic form
Ax2 +2Bxy+cy 2 . The conic has equation AC B 2 = 0 and any point on it
represents perfect squared (ax+by)2 , so has the form (a2 : ab : b2 ). It follows
that the tangent line to at (a2 : ab : b2 ) has equation Ca2 2Bab+Ab2 = 0.
Now the pencil determined by v consists of quadratic forms having
(v2 x v1 y) as a factor, so is the line tangent to at (v22 , v1 v2 , v12 ). There-
fore it has equation Cv22 + 2Bv1 v2 + Av12 = 0. The pencil determined by
(Q1 , Q2 ) intersects at (1 : 0 : 0) and (0 : 0 : 1). The point (1 : 0 : 0) (resp.
 Surfaces in the Euclidean 4-space 235

(0 : 0 : 1)) is on the line determined by v if and only if v2 = 0 (resp. v1 = 0)

which are precisely the conditions for the projection to have a singularity
more degenerate than a cross-cap.
(ii) At a non-degenerate inflection point, we take (Q1 , Q2 ) = (x2 ±y 2 , 0),
so that j 2 Pv (x, y) = (v2 x v1 y, x2 ± y 2 , 0). Following the same calculations
as in (i) above, if v2 6= 0, we can reduce the j 2 Pv to (X, 2v1 XY + (v12 ±
v22 )Y 2 , 0). Completing the square in the second component, we can reduce
further the 2-jet to (X, Y 2 , 0) when v12 ± v22 6= 0 or to (X, XY, 0) when
v12 ± v22 = 0 (but v2 6= 0). This means that most directions of projections
yield singularities of type Sk and Bk , and there are 2 directions (when the
inflection point is of real type) or 0 direction (when the inflection point is
of imaginary type) yielding singularities of Hk -type. ⇤
Asymptotic directions can also be described via the singularities of pro-
jections to hyperplanes.

Proposition 7.15 ([Mond (1982); Bruce and Nogueira (1998)]).

A tangent direction v at p on M is an asymptotic direction if and only if
the projection in the direction v yields a singularity more degenerate than
a cross-cap.

Proof. We take the surface as before in Monge form. If p is not an

inflection point, a tangent direction v = (v1 , v2 , 0, 0) at p is an asymptotic
direction if and only if there exists b1 , b2 such that b1 Q1 +b2 Q2 is degenerate
at p and v is in the kernel of its Hessian at p, that is, (v2 x v1 y)2 =
b1 Q1 +b2 Q2 . But this means that the pencil generated by Q1 and Q2 meets
the conic at (v2 x v1 y)2 , which by Proposition 7.14 is equivalent to the
projection along v having a singularity more degenerate than a cross-cap.
If p is an inflection point then every tangent direction at p is an asymp-
totic at p, and the projection along these directions is more degenerate than
a cross-cap. ⇤

Proposition 7.16. Let M be a projection P -generic surface in R4 . The Ae -

codimension 2 singularities of the family P occur on curves on the surface
and the Ae -codimension 3 singularities are special points on these curves.

Proof. This follows from the assumption that P is a versal unfolding,

hence it is transversal to the A-orbits in jet space. ⇤

Definition 7.8. We call S2 -curve (resp. B2 -curve and H2 -curve) the clo-
sure of the set of points p on M for which there exists a projection Pv
236 Di↵erential Geometry from a Singularity Theory Viewpoint

having an S2 (resp. B2 , H2 )-singularity at p.

Proposition 7.17. Let M be a projection P -generic surface in R4 . Then

the H2 -curve coincides with the set = 0 and the B2 -curve coincides
with the closure of the set of point where the height functions has an A3 -
singularity.

Proof. The statement on the H2 -curve follows from the proof of

Proposition 7.14.
Again, from Proposition 7.14, we can take the point p to be a hy-
perbolic point. We work with the surface in Monge form (x, y) =
(x, y, f1 (x, y), f2 (x, y)) with j 3 (x, y) given by
(x, y, x2 +a30 x3 +a31 x2 y+a32 xy 2 +a33 y 3 , y 2 +b30 x3 +b31 x2 y+b32 xy 2 +b33 y 3 ).
Consider the projection along the asymptotic direction v = (0, 1, 0, 0)
so that Pv (x, y) = (x, f1 (x, y), f2 (x, y)). Then j 3 P (x, y) is A(3) -equivalent
to (x, a31 x2 y + a33 y 3 , y 2 ). The singularity is of type Bk , k 1, if and only
if a31 6= 0 and of type B 2 if further more a33 = 0.
The binormal direction associated to v is w = (0, 0, 1, 0) and the height
function along w is given by hw = x2 + a30 x3 + a31 x2 y + a32 xy 2 + a33 y 3 .
It has an A 3 singularity if and only if a33 = 0. Thus the B2 -curve (of
Pv with v an asymptotic direction) is also the locus of points where the
height function hw (along the binormal direction w associated to v) has an
A 3 -singularity. ⇤

Remark 7.7. (1). The results in Proposition 7.17 follow in fact from
the duality in [Bruce and Nogueira (1998)] between certain strata of the
bifurcation set of the family of height functions with that of the family of
projections.
(2) The A4 -singularities hw are in general not related to the B3 -
singularities of Pv .
(3) The C3 -singularity of Pv occurs at a point of intersection of the
S2 -curve and B2 -curve. The S2 -curve is in general not related to the sin-
gularities of the height functions.
(4) For a projection generic surface, the robust curves captured by the
singularities of Pv are as in Figure 7.7. It is shown in [Bruce and Tari
(2002)] that the B2 -curve and the -set (i.e., the parabolic curve) meet
tangentially. The point of tangency is a P3 (c)-singularity of a Pv . The two
asymptotic direction fields in the hyperbolic region can be coloured, so we
have two di↵erent coloured B2 -curves. The B2 -curve changes colour at a
 Surfaces in the Euclidean 4-space 237

P3 (c)-singularity. The P3 (c)-singularity is also a point where the asymptotic

configurations have a folded singularity, i.e., they are as in Figure 7.2.

!"#$%&'()*+%$,)'- !"#$%&'(
)"#$%&'(

)*+!*
/, !,
),
-,
.,
3-4($*/)'-+#')-/2 -"#$%&'(
.(()#/)*+%$,)'-
012$/

Fig. 7.7 Robust curves on the surface depicted by its contact with lines.

7.7.1 The geometry of the projections

The surface Pv (M ) lives in v⊥ = Tv S 3 , and we are interested in its geom-
etry as a surface in a 3-dimensional Euclidean space. For this reason, we
consider its contact with planes in Tv S 3 which is measured by the height
functions on Pv (M ) in Tv S 3 . These planes can be parametrised by unit
vectors w in Tv S 3 , i.e., w · v = 0 and w · w = 1. We denote by

D = {(v, w) ∈ S 3 × S 3 : ⟨v, w⟩ = 0}.

Given (v, w) ∈ D, the height function on the projected surface Pv (M )

along the vector w is given by

H(v,w) (u) = ⟨Pv (u), w⟩ = ⟨x(u) − ⟨x(u), v⟩v, w⟩ = ⟨x(u), w⟩.

This is precisely the height function on M along the direction w. In par-

ticular, it follows that

Proposition 7.18. The height function H(v,w) on Pv (M ) along the direc-

tion w has the same singularity type at Pv (p) as the height function Hw
on M along w at p.

The family H : U ×D → R has parameters in D which is a 5-dimensional

manifold. However, it is trivial along the parameter v. Thus, the generic
singularities that can appear in H(v,w) are those of Re -codimension ≤ 3.
238 Di↵erential Geometry from a Singularity Theory Viewpoint

For v fixed, w varies in a 2-dimensional sphere, so for a generic M

and for most directions v, the height function on Pv (M ) has singularities
of type A± ±
1 , A2 and A3 , and these are versally unfolded by varying w.
(Recall from Chapter 6 that the closure of the A2 -singularities form the
parabolic set on Pv (M ), the A3 -singularities are its cusps of Gauss.) For
isolated directions v, we expect the following singularities: A4 , D4± and an
A3 -singularity which is not versally unfolded by the family Hv . We denote
the latter by N V A3 .

Definition 7.9. We call the pre-image on M by Pv of the parabolic set

of Pv (M ) as a surface in the 3-space Tv v S 3 the v-pre-parabolic set and
denote it by v-P P S.

We consider how the geometric data of Pv (M ) at Pv (p) can provide

geometric information about M at p. (More details on this can be found
in [Nuño-Ballesteros and Tari (2007); Oset Sinha and Tari (2010)].)
We start with the case when v is not a tangent direction at p. We write
v = vT + vN where vT is the orthogonal projection of v to the tangent
space Tp M and vN is its orthogonal projection to the normal space Np M .
Since vN 6= 0, the surface Pv (M ) is a smooth surface at Pv (p).

Proposition 7.19. The height function H(v,w) on Pv (M ) is singular at

Pv (p) if and only if w 2 Np M . For a generic surface, the singularity of
H(v,w) at Pv (p) is of type

?
A2 : if p is a hyperbolic or parabolic point, w = vN and is a binor-
mal direction.
?
A3 : w = vN is a binormal direction, p is on the B2 -curve and v
is away from a circle of directions C in the sphere w? 2 D.
Then the v-P P S is a regular curve.
?
N V A3 : w = vN is a binormal direction, p is on the B2 -curve and
v 2 C. For generic v 2 C the singularity of the v-P P S is
an A1 . For isolated directions in C the singularity becomes
an A2 , and for special points on the B2 -curve it becomes an
A3 -singularity.
?
A4 w = vN is a binormal direction, p is an A4 -point on the B2 -
curve.
?
D4 : w = vN is a binormal direction, p is an inflection point.
 Surfaces in the Euclidean 4-space 239

Proof. The proof can be found in [Oset Sinha and Tari (2010)]; see also
[Nuño-Ballesteros and Tari (2007)]. ⇤
Suppose now that v 2 S 3 is a tangent direction at p 2 M . Then Pv (M )
is a singular surface Pv (p). The map-germ Pv is of corank 1 and these can
be written, in some coordinate system, in the form
(x, y) = (x, p(x, y), q(x, y))
with p, q 2 M2 (x, y). We denote by Q1 (x, y) = j 2 p(x, y) = p20 x2 + p21 xy +
p22 y 2 and Q2 (x, y) = j 2 q(x, y) = q20 x2 + q21 xy + q22 y 2 . We can consider
the action of GL(2, R)⇥GL(2, R) on (Q1 , Q2 ) as in §7.2.1 to define an affine
property of the singular surface S, image of , at its singular point.
Definition 7.10. The singular point of the surface S is called hyperbolic
(resp. elliptic, parabolic) or (a generic) inflection point if the GL(2, R) ⇥
GL(2, R)-class of (Q1 , Q2 ) can be represented by (x2 , y 2 ) (resp. (x2 , x2 y 2 ),
(x2 , xy)) or (x2 ± y 2 , 0), as in Table 7.1.
The surface S = Pv (M ) has a cross-cap singularity if and only if v 2
Tp M is not an asymptotic direction at p. It is shown in [Bruce and West
(1998); West (1995)] that a parametrisation of a cross-cap can be taken,
by a suitable choice of a coordinate system in the source and affine changes
of coordinates in the target, in the form
(x, y) = (x, xy + f1 (y), y 2 + ax2 + f2 (x, y)), (7.14)
4 3
where f1 2 M (y) and f2 2 M (x, y). The following is also shown in [West
(1995)]. When a < 0, the height function along any normal direction at
the cross-cap point has an A1 -singularity. Such cross-caps are labelled hy-
perbolic cross-caps as the surface has negative Gaussian curvature at all
its regular points (Figure 7.8, left). When a > 0, there are two normal
p
directions (0, ±2 a, 1) at the cross-cap point along which the height func-
tion has a singularity more degenerate than A1 (i.e., of type A 2 ). Such a
cross-cap is labelled elliptic cross-cap (Figure 7.8, right). The singularity
of the height function along the degenerate normal direction is precisely of
type A2 if and only if j 3 f2 (⌥ p1a , 1) 6= 0. When a = 0, there is a unique
normal direction at the cross-cap point where the height function has a
singularity more degenerate than A1 . The singularity of its corresponding
3
height function is of type A2 if and only if @@xf32 (0, 0) 6= 0. Such a cross-cap
is labelled parabolic cross-cap (Figure 7.8, centre).
Theorem 7.10. A cross-cap is hyperbolic (resp. elliptic, parabolic) if
and only if its singular point is elliptic (resp. hyperbolic, parabolic) as in
Definition 7.10.
240 Di↵erential Geometry from a Singularity Theory Viewpoint

Fig. 7.8 Hyperbolic cross-cap (left) and elliptic cross-cap (right) separated by a
parabolic cross-cap (centre).

Proof. The pair of quadratic forms associated to in (7.14) is (xy, y 2 +

ax2 ). This is GL(2, R) ⇥ GL(2, R)-equivalent to (xy, x2 y 2 ), (x2 , y 2 ) or
(x2 , xy) in Table 7.1 if and only if a < 0, a > 0 or a = 0, and the result
follows from the discussion above. ⇤
The surface S = Pv (M ) has a cross-cap singularity when v 2 Tp M is
not an asymptotic direction at p. We can say more about Pv (M ) and M .
Theorem 7.11. Suppose that v 2 Tp M but is not an asymptotic direction
(in particular, p is not an inflection point).
(i) The v-PPS has a Morse singularity if p 2 / -set. Furthermore, the
singularity is of type A1 (i.e. Pv (M ) is an elliptic cross-cap) if p is a
hyperbolic point and of type A+ 1 (i.e. Pv (M ) is a hyperbolic cross-cap) if p
is an elliptic point.
(ii) At points on -set that are not on the B2 -curve, the v-PPS has an
A2 -singularity (i.e. Pv (M ) is a parabolic cross-cap). At the point of tan-
gency of the B2 -curve with -set, the singularity of the v-PPS is generically
of type A3 .
(iii) The tangent directions to the v-PPS are along the asymptotic di-
rections to M at p.
Proof. We take M in Monge form (x, y) = (x, y, f1 (x, y), f2 (x, y))
at the origin p. Suppose that p is a hyperbolic point, so we can take
(Q1 , Q2 ) = (x2 , y 2 ). The asymptotic directions at p are (1, 0, 0, 0) and
(0, 1, 0, 0). We consider a tangent vector v = (↵, , 0, 0) which is not an
asymptotic direction, that is, ↵ 6= 0.
Calculating the Gaussian curvature of Pv (M ) away from the singular
point, one can get the v-PPS as the zero set of the function
K̃v = det(v, x , y , xx ) det(v, x , y , yy ) det(v, x , y , xy )2 .
 Surfaces in the Euclidean 4-space 241

We have j 2 K̃v (x, y) = 16↵ xy, which has an A1 -singularity at the

origin. It is clear that its tangent directions coincide with the asymptotic
directions of M at the origin.
Analogously, if p is an elliptic point, we take (Q1 , Q2 ) = (xy, x2 y 2 )
and v = (↵, , 0, 0), with ↵2 + 2 = 1. (Recall that there are no asymptotic
directions at an elliptic point.) Then j 2 K̃v (x, y) = 4(x2 + y 2 ) and this
has an A+1 -singularity at the origin, that is the v-PPS is locally an isolated
point.
Suppose now that p 2 . We take (Q1 , Q2 ) = (x2 , xy) and let
v = (↵, , 0, 0) be a tangent vector with ↵ 6= 0 (↵ = 0 gives the unique
asymptotic direction at p). We have j 2 K̃v (x, y) = 4↵2 x2 , so the v-PPS
has an Ak -singularity, with k 2. Note that we have one tangent direc-
tion to the v-PPS which is exactly the unique asymptotic direction at the
origin. The coefficients of y 3 in j 3 K̃v (x, y) is 12a33 ↵2 . Hence, for a generic
point on the -set the v-PPS has an A2 -singularity. The points where
a33 = 0 correspond to points where the height function along the unique
binormal direction (0, 0, 1, 0) has an A 3 -singularity. That is p is also on
the B2 -curve, so p is the point of tangency of -set and the B2 -curve. For
a generic surface, K̃v has an A3 -singularity at such a point. ⇤

Definition 7.11. We call an elliptic cross-cap where the height function

has an Ai -singularity along one degenerate direction and an Aj -singularity
along the other degenerate direction an elliptic cross-cap of type Ai Aj or
an Ai Aj -elliptic cross-cap. Likewise, we label an Ak -parabolic cross-cap one
where the height function has a degenerate singularity (of type Ak ) along
the unique degenerate normal direction.

Proposition 7.20. Suppose that v 2 Tp M but is not an asymptotic direc-

tion at p.
(i) If p is a hyperbolic point, then Pv (M ) is a surface with an elliptic
cross-cap of type A2 A2 if p is not on the B2 -curve. If it is, the elliptic
cross-cap becomes of type A2 A3 and at isolated points on this curve it can
be of type A2 A4 or A3 A3 .
(ii) If p is a parabolic point, then Pv (M ) is in general an A2 -parabolic
cross-cap and becomes an A3 -parabolic cross-cap if p is the point of tangency
of the B2 -curve with the -set.

Proof. The type of the cross-cap is determined by the singularities of the

height function H(v,wi ) on Pv (M ) at Pv (p) along the binormal directions
wi , i = 1, 2. The result follows from Proposition 7.18 that these are the
242 Di↵erential Geometry from a Singularity Theory Viewpoint

same as the singularities of the height function Hwi on M at p and from

Proposition 7.17.
In (i), the A2 A4 cross-cap occurs at special points on the B2 -curve where
the height function has an A4 -singularity and these are distinct in general
from the B3 and C3 -points. The A3 A3 cross-cap occurs at the point of
intersection of two B2 -curves associated to the two binormal directions. ⇤

When the direction of projection is asymptotic at p, Pv has a singularity

more degenerate than a cross-cap. The v-PPS has also singularities more
degenerate than Morse. For a projection P -generic surface, these are as in
Table 7.4.
Table 7.4 The singularities of Pv and of the v-P P S.
Pv B1± B2 B3 S2 S3 C3 H2 H3 P3
⌥
v-P P S D4 D5 D5 E7 J10 X1,0 D5 D5 J10

Recall that the parabolic curve on a surface in 3-space is the locus

of degenerate singularities of the height function. From Proposition 7.18,
the degenerate singularities of the height function on Pv (M ) are along
the binormal directions of the surface M . Thus, at a hyperbolic point
where are two distinct binormal directions, we can split the v-PPS into two
components with each component corresponding to one of the binormal
direction. We denote L1 and L2 these components. Analysing in Table
7.5 the singularities of L1 and L2 give a better understanding why the
singularities of the v-PPS (which is that of L1 [L2 ) is highly degenerate (see
Table 7.4). In Table 7.5, “tg” is for tangency and “t” is for transversality
between the components L1 and L2 .

7.8 Contact with planes

We discuss briefly in this section orthogonal projections of a surface to a

plane. Some of their aspects are studied in [Nogueira (1998)].
Consider the orthogonal projection from R4 to a 2-dimensional vector
space ⇡ (which we refer to as a plane). Its kernel is the plane ⇡ ? , the
orthogonal complement of ⇡. The set of all planes in R4 is the Grassmanian
Gr(2, 4), which can be used to parametrise all the orthogonal projections
from R4 to planes. Clearly, Gr(2, 4) can be used to parametrise either the
planes of projections or their orthogonal complements.
Let {u, v} be an orthonormal basis of ⇡ ? and denote by ⇧(u,v) the
 Surfaces in the Euclidean 4-space 243

Table 7.5 The generic structure of the v-P P S and of its two components.
S B1 B2 B3

L1 A±
1 A2 A2

L2 A0 (!) A0 (!) A0 (!)

v-P P S D4± D5 D5

S S2 S3 C3

L1 A2 A±
3 D4±

L2 A0 (tg) A0 (tg) A0 (tg)

v-P P S E7 J10 X1,0

orthogonal projection from R4 to π. Then,

Π(u,v) (p) = p − ⟨p, u⟩u − ⟨p, v⟩v. (7.15)
3 3
Now (u, v) is in D = {(u, v) ∈ S × S : ⟨u, v⟩ = 0} which is a 5-
dimensional manifold. However, the choice of an orthonormal basis in π ⊥
is determined up-to rotations, so we have an action of SO(2) on each plane.
This gives the quotient space D/SO(2) which is a 4-dimensional manifold
and can be identified locally with Gr(2, 4).
Given a surface M in R4 , we still denote by Π(u,v) the restriction of
Π(u,v) to M . This map can be considered locally at a point p ∈ M as a map-
germ Π(u,v) : R2 , 0 → R2 , 0. Since D/SO(2) is a 4-dimensional manifold,
using the transversality theorem, we expect the map Π(u,v) to have only
simple singularities of Ae -codimension ≤ 4 or non-simple singularities with
the Ae -codimension of the stratum ≤ 4.
The corank 1 singularities of Ae -codimension ≤ 4 are given in Table 7.6.
They are taken from the classification in [Rieger (1987)].
The A-simple germs of corank 2 are classified in [Rieger and Ruas
(1991)]. There are two such families:

l,m
I2,2 : (x2 + y 2l+1 , y 2 + x2m+1 ), l ≥ m ≥ 1
l
II2,2 : (x2 − y 2 + x2l+1 , xy), l ≥ 1.
244 Di↵erential Geometry from a Singularity Theory Viewpoint

Table 7.6 Corank 1 singularities of projections of surfaces in R4 to

2-planes.
Name Normal form Ae -codimension
1 : Immersion (x, y) 0
2 : Fold (x, y 2 ) 0
3 : Cusp (x, xy + y 3 ) 0
4k (x, y 3 ± xk y), k 2 2k5
5 (x, xy + y 4 ) 1
6 (x, xy + y 5 ± y 7 ) 2
7 (x, xy + y 5 ) 3
8 (x, xy + y 6 ± y 8 + ay 9 ) 5(4(⇤) )
9 (x, xy + y 6 + y 9 ) 4
10 (x, xy + y 7 ± y 9 + ay 10 + by 11 ) 6(4(⇤) )
112k+1 (x, xy 2 + y 4 + y 2k+1 ), k 2 k = 2, 3, 4
12 (x, xy 2 + y 5 + y 6 ) 3
13 (x, xy 2 + y 5 ± y 9 ) 4
15 (x, xy 2 + y 6 + y 7 + ay 4 ) 5(4(⇤) )
16 (x, x2 y + y 4 ± y 5 ) 3
17 (x, x2 y + y 4 ) 4
18 (x, x2 y + xy 3 + ay 5 + y 6 + by 7 ) 6(4(⇤) )
(⇤) The codimension is that of the stratum.

The classification of corank 2 non-simple singularities with Ae -

codimension of the stratum  4 so far is not complete. They are K-
equivalent to (x2 + y 3 , xy) when the Ae -codimension of the stratum is 3
and to (x2 + y 4 , xy) when this codimension is 4.
An orthogonal projection of M to a plane ⇡ is singular at p 2 M if
and only if dim(Tp M \ ⇡ ? ) 1. If the dimension is 1, the singularity is of
corank 1, otherwise it is of corank 2.
Suppose that dim(Tp M \⇡ ? ) = 1 and let u be a unit vector in Tp M \⇡ ? .
Let v be a unit vector in ⇡ ? orthogonal to u. (Then v belongs to the
2-sphere {(u, v) 2 S 3 ⇥ S 3 : hu, vi = 0} with the poles ±u? removed,
where u? is orthogonal to u in Tp M .) To simplify notation, we take M
in Monge form at the origin and suppose, without loss of generality, that
u = (1, 0, 0, 0). Then v = (0, v2 , v3 , v4 ) with either v3 or v4 not zero. It
follows from (7.15) that
⇧(u,v) (x, y) = (x, y, f1 (x, y)), f2 (x, y)) (x, 0, 0, 0)
(v2 y + v3 f1 (x, y) + v4 f2 (x, y))(0, v2 , v3 , v4 )
= (0, y (v2 y + v3 f1 (x, y) + v4 f2 (x, y))v2 ,
f1 (x, y) (v2 y + v3 f1 (x, y) + v4 f2 (x, y))v3 ,
f2 (x, y) (v2 y + v3 f1 (x, y) + v4 f2 (x, y))v4 ).
This is A-equivalent to
((1 v22 )y (v3 f1 (x, y) + v4 f2 (x, y))v2 , v4 f1 (x, y) v3 f2 (x, y)).
 Surfaces in the Euclidean 4-space 245

Observe that (1 v22 ) 6= 0 so at the 2-jet level we have

j 2 ⇧(u,v) (x, y) 'A(2) (y, v4 Q1 (x, y) v3 Q2 (x, y)),

with the second component tracing the pencil (Q1 , Q2 ) in RP2 . As we fixed
u, we cannot take (Q1 , Q2 ) as one of the normal forms in Table 7.1. We
have

j 2 ⇧(u,v) 'A(2) (y, (v4 l1 v3 l2 )x2 + 2(v4 m1 v3 m2 )xy + (v4 n1 v3 n2 )y 2 ),

where li , mi , ni , i = 1, 2 denote the coefficients of the second fundamental

form at the origin.
If v4 l1 v3 l2 6= 0, the singularity is a fold, and if v4 l1 v3 l2 = 0 but
v4 m1 v3 m2 6= 0, then j 2 ⇧(u,v) 'A(2) (y, xy). This 2-jet leads to the
singularities of type 2, 3 and 5, · · · , 10 in Table 7.6. These singularities are
characterised by the property that the singular set ⌃(u,v) of ⇧(u,v) in M is
a smooth curve at p.
Now v4 l1 v3 l2 = v4 m1 v3 m2 = 0 if and only if l1 m2 l2 m1 = 0,
equivalently, u is an asymptotic direction at p (see Theorem 7.3). The
condition v4 l1 v3 l2 = v4 m1 v3 m2 = 0 also means that the direction
w = (0, 0, v4 , v3 ) is the binormal direction associated to u. Observe that
the vector w is in Np M \ ⇡. We conclude that at an elliptic point only
the singularities of type 2, 3 and 5, · · · , 10 can occur. At a non-elliptic
point, we get these singularities too but also those with a 2-jet equivalent
to (y, 0). The latter singularities occur when the plane of projections is in
the pencil of planes in the 3-space u? that contain the binormal direction
associated to the asymptotic direction u. At a hyperbolic point there are
two such pencils associated to the two distinct asymptotic directions and
at a parabolic point which is not an inflection point there is only one such
pencil. At an inflection point, there is a unique binormal direction but any
tangent direction is asymptotic. Thus any plane of projection that contains
the binormal direction gives a singularity with a singular critical set.
We consider now the corank 2 singularities of ⇧(u,v) and take
M in Monge form as above. The vectors u and v generate Tp M ,
so we can take them to be u = (1, 0, 0, 0) and v = (0, 1, 0, 0).
Then ⇧(u,v) (x, y) = (f1 (x, y), f2 (x, y)). At the 2-jet level, we have
j 2 ⇧(u,v) (x, y) = (Q1 (x, y), Q2 (x, y)) and we have the classification in the
2-jet space given by that of pairs of quadratic forms in Theorem 7.1 (see
also Table 7.2). Of course that classification is related to the partition of
the surface into elliptic, hyperbolic, parabolic and inflection points.
246 Di↵erential Geometry from a Singularity Theory Viewpoint

7.9 Contact with hyperspheres

We give in this section geometric characterisations of the generic singulari-

ties of the distance-squared functions on the surface. We refer to [Montaldi
(1983)] for the study of the contact of the surface with k-spheres, k = 1, 2, 3.
The family of distance squared functions D : U ⇥ R4 ! R on M is given
by

D(u, a) = ||x(u) a||2 ,

and the extended family of distance squared functions D̃ : U ⇥ R4 ⇥ R ! R

is defined by

D̃(x, a, r) = ||x(u) a||2 r2 .

The functions Da and D̃a,r have a singularity at u if and only if the

direction a x(u) is normal to M at p = x(u).
The catastrophe set CD of the family D coincides with the normal bun-
dle

N M = {(u, a) 2 U ⇥ R4 | a = x(u) + v, v 2 (Np M )1 , 2 R}

and its bifurcation set BD is the focal set of M. It follows that the catastro-
phe map of the family D coincides with the map Ge : N M ! R4 defined
by Ge (u, a) = a, which we call the normal exponential map of M.

Theorem 7.12. There is an open and dense set OD in Imm(U, R4 ) such

that for any x 2 OD , the surface M = x(U ) has the following properties:

i) The distance squared function Da has only singularities of R-type

Ak , 1  k  5, D4 and D5 at any point p in M. Similarly, the function
D̃a,r has only singularities of K-type Ak , 1  k  5, D4 and D5 at p.
ii) The singularities of Da are R+ -versally unfolded by the family D, and
the singularities of D̃a,r are K-versally unfolded by the family D̃.
iii) The normal exponential map Ge is stable as a Lagrangian map.

Proof. The proofs of (i) and (ii) follow from Theorem 4.8. Since the
catastrophe map of a Morse family of functions can be identified with the
Lagrangian map (cf. §5.4), Ge is a Lagrangian map. The proof of (iii)
follows from Theorem 5.4. ⇤

Definition 7.12. An immersion x : M ! R4 is called distance squared

function generic if it belongs to the set OD in Theorem 7.12.
 Surfaces in the Euclidean 4-space 247

Definition 7.13. The points of BD are called focal points of M. If p 2 M

is a degenerate singularity of Da , with a 2 BD , we say that the hypersphere
centred at a and tangent to M at p = x(u) is an osculating hypersphere of
M at p. In such case, the kernel of the Hessian of Da at u, Hess(Da )(u),
is non-zero and the directions in this kernel are called spherical contact
directions of M at p.

We prove next that the spherical contact directions are principal direc-
tions associated to normal fields pointing towards the focal points of the
surface.

Lemma 7.2. For p = x(u) 2 M , a = p + v, with v 2 (Np M )1 ,

Hess(Da (u)) = 2 Id Hess(hv )(u) .

Proof. The proof follows from the definitions of Da and hv . ⇤

Lemma 7.3. Let a be a focal centre of M at p = x(u) and let w 2 Tp M

be a spherical contact direction associated to the distance squared function
Da , with a = p + v. Then w is an eigenvector of the shape operator Wpv
in the normal direction v at p with eigenvalue 1/ .

Proof. The matrices of Wpv and Hess(hv )(u) coincide at p. Then w is a

spherical contact direction associated to the distance squared function da ,
with a = p v if and only if det Hess(da )(u) = 0. It follows from Lemma
7.2 that Wpv (w) = 1/ w. Clearly, 1/ is a v-principal curvature at p. ⇤

Definition 7.14. The strong principal directions at p 2 M are the spheri-

cal contact directions corresponding to singularities at p of type Ak , k 3,
of the distance squared functions.

The following result on the number of strong principal directions at the

points of a surface generically immersed in 4-space was proved by Montaldi
in [Montaldi (1983)].

Theorem 7.13. For a distance squared function generic immersion of a

surface in R4 there are at least 1 and at most 5 strong principal directions
at each point.

Definition 7.15. We define the rib of order k of M as the set of points of

R4 that determine distance squared functions with (corank 1) singularities
of type Ak , k 3.
248 Di↵erential Geometry from a Singularity Theory Viewpoint

Definition 7.16. A point p 2 M which is a singular point of type Ak , k 4

for some distance-squared function Da is said to be a ridge point of order
k. We call ridge of order k the set of all k-order ridge points for a given
k 4.

Observe that the ridge of order k is the projection of the (regular)

curve S1,1,1 (Ge ) in the normal bundle through the canonical projection
⇡N : N M ! M .

Proposition 7.21. On a distance squared generic immersed surface, the

set of 4-order ridge points is either empty or is a smooth curve called ridge.
The ridge points of order 5, if they exist, are isolated points on the ridge.

Proof. The ridge of order 4 is the projection of the (regular) curve

S1,1,1 (G) in the normal bundle of M by the canonical projection ⇡M :
N M ! M . For a generically immersed surface, the kernel of this projec-
tion is transversal to this curve at every point. Therefore, its image in M
is a regular curve too.
The ridge points of order 5 are the images of singular points of type A4
of G, which are generically isolated in S1,1,1 (G). ⇤
Another characterisation of 5-order ridge points is given in [Romero
Fuster and Sanabria Codesal (2002)].
We consider now the focal hyperspheres whose centres define distance-
squared functions with corank 2 singularities on M . In this case, Hess(Da )
vanishes at the given point.

Definition 7.17. An umbilical focus is a point a 2 R4 for which the dis-

tance squared function Da has a corank 2 singularity. We refer to the focal
3-spheres centred at umbilical foci as umbilical focal hyperspheres.

Theorem 7.14 ([Montaldi (1983)]). A point p = x(u) 2 M is a semi-

umbilic and not an inflection point if and only if it is a singularity of corank
2 of some distance squared function on M.

Proof. Let

a 3 b3 c 3
↵(p) = ,
a 4 b4 c 4
be the matrix of the second fundamental form of x at p.
The curvature ellipse is given by
⌘(✓) = H + cos(2✓)B + sin(2✓)C.
 Surfaces in the Euclidean 4-space 249

Therefore, p = x(u) is a semiumbilic point if and only if rank (B, C) = 1.

For v = v3 e3 + v4 e4 2 Np M , we have
"P P4 #
4
a v
i i b v
Hess(hv )(u) = Pi=3 4 P4i=3 i i .
i=3 bi vi i=3 ci vi
1
Then, if we denote = ka pk ,
we get
" P4 P4 #
1 1 i=3 ai vi 1
b v
Hess(Da )(u) = 1 P4 1
P4i=3 i i ,
i=3 bi vi 1 i=3 ci vi

so the rank of Hess(Da )(u) vanishes if and only if

P4 P4 P4
i=3 ai vi = i=3 ci vi = i=3 bi vi = 0.
This is equivalent to
4
X 4
X 4
X
ai vi = ci vi , and bi vi = 0,
i=3 i=3 i=3
which imply that v is orthogonal to both B and C in Np M. Equivalently,
rank (B, C) = 1, that is, p is a semiumbilic point.
Conversely, if p = x(u) is a semiumbilic point and v is a unit normal
vector orthogonal to the curvature segment at p in Np M , then the distance
squared function Da , with a = p + 1/hH, viv, has a corank 2 singularity at
u. ⇤
Observe that |hH, vi| is the distance of p to the affine line determined by
the curvature segment. When this distance vanishes, the point p belongs to
the curvature segment, so p is an inflection point and the focal hypersphere
becomes the (unique) osculating hyperplane.

Proposition 7.22 ([Montaldi (1983)]). The semiumbilics of a surface

generically immersed in 4-space form regular curves.

Corollary 7.4. If p is a semiumbilic point of M with umbilical focus

at a 2 R4 , then p is a v-umbilic point for the unit normal direction
v = (a p)/ka pk and the corresponding v-curvature is hH, vi.

Proof. The result follows from Theorem 7.14 and Lemma 7.3. ⇤

7.10 Notes

Montaldi studied in [Montaldi (1983)] the extrinsic geometry of surfaces in

R4 based on their generic contacts with 2-spheres and hyperspheres. The
250 Di↵erential Geometry from a Singularity Theory Viewpoint

contact of the surface with hyperspheres is studied in [Romero Fuster and

Sanabria Codesal (2002)]. Mond applied in [Mond (1982)] his classification
of simple singularities of map-germs R2 , 0 ! R3 , 0 to study generic central
projections of surfaces in R4 to hyperplanes.
We touched briefly on the asymptotic curves. There are various other
pairs of geometrical foliations on the surface of interest. These can be
found, for example, in [Garcia, Mello and Sotomayor (2005); Garcia and So-
tomayor (2000); Gutierrez, Guadalupe, Tribuzy and Guı́ñez (1997); Gutier-
rez, Guadalupe, Tribuzy and Guı́ñez (2001); Gutierrez and Guı́ñez (2003);
Little (1969); Mello (2003); Ramı́rez-Galarza and Sánchez-Bringas (1995);
Tari (2009)].
For a generic immersion of a surface in R4 , the height function has
isolated double points, which correspond to bi-tangencies of the immersion.
Bi-tangency properties of generic immersions of surfaces in R4 are studied
by Dreibelbis in [Dreibelbis (2001, 2004, 2006, 2007)].
Conformal properties of surfaces in R4 are studied in [Romero Fuster
and Sanabria Codesal (2004, 2008, 2013)].
 Chapter 8

Surfaces in the Euclidean 5-space

With the aim of illustrating how the singularity techniques can be applied
to analyse the extrinsic geometry of surfaces in higher codimensions we
discuss in this chapter the geometrical properties associated to the contacts
of surfaces with hyperplanes and hyperspheres in R5 . In this, as well as in
higher codimensional cases, the curvature ellipse at each point of the surface
determines a proper subspace of the normal space and both, its dimension
and relative positions with respect to the considered point are relevant in
the description of the second order geometry of the surface at this point.
Moreover, di↵erently from the case of surfaces in R4 , the directions leading
to degenerate singularities of height functions at each point are not finite.
These directions determine a cone in the normal space at the considered
point. Attending to the behaviour of this cone we can distinguish among
di↵erent types of points. The corresponding properties are described in
§8.1. The generic singularities of the height functions on surfaces in R5 are
analysed in §8.2. Analogously to the case of surfaces in R4 , this setting
leads naturally to the introduction of binormal and asymptotic directions
at each point of the surface. Since the number of normal directions leading
to degenerate singularities of height functions at each point is not finite,
we use the concept of binormal direction for those leading to singularities
of height functions on the surface with codimension higher than one, which
only occur in a finite number of normal directions at every point. The
corresponding contact directions are the asymptotic directions at the point.
Alternative characterisations of asymptotic directions in terms of normal
sections and of the geometrical behaviour of orthogonal projections of the
surface into 3- and 4-spaces are also given. We see in Proposition 8.9 that
the number of asymptotic directions at each point is at least one and at
most five. The possible local configurations, described in Proposition 8.10,

251
252 Di↵erential Geometry from a Singularity Theory Viewpoint

were determined in [Romero Fuster, Ruas and Tari (2008)].

Another relevant subset in this context is the flat ridges curve. Here we
shall understand by flat ridge points as those at which there is some height
function having a singularity of type Ak 4 or worse.
We use the stereographical projection to transport these results on con-
tacts of surfaces with hyperplanes in 5-space into results on contacts of
surfaces with hyperspheres in 4-space.
We include in 8.3 a description of the generic behaviour of orthogonal
projections onto hyperplanes, 3-spaces and planes. Finally, in §8.3 we anal-
yse the generic singularities of distance squared functions on surfaces in R5 .
This setting leads to the introduction of geometrical concepts such as rib
and ridge points, as well as umbilical foci and umbilical curvature.
The basic references for this chapter are [Mochida (1993); Mochida,
Romero Fuster and Ruas (2003); Romero Fuster, Ruas and Tari (2008);
Costa, Moraes and Romero Fuster (2009)].

8.1 The second order geometry of surfaces in R5

Let M be a closed surface in R5 , and x : U ! R5 a local parametrisation of

M. Let {e1 , e2 , e3 , e4 , e5 } be a positively oriented orthonormal frame in R5
such that at any u 2 U, {e1 (u), e2 (u)} is a basis for the tangent plane Tp M
and {e3 (u), e4 (u), e5 (u)} is a basis for the normal plane Np M at p = x(u).
By using the notation of Chapter 7, we can represent the matrix of the
second fundamental form associated to the embedding as
2 3
a 1 b1 c 1
↵ p = 4 a 2 b2 c 2 5
a 3 b3 c 3
where ai = hxu1 u1 , ei+2 i , bi = hxu1 u2 , ei+2 i and ci = hxu2 u2 , ei+2 i, i =
1, 2, 3. We can naturally extend the concept of curvature ellipse to surfaces
immersed with higher codimension:
Definition 8.1. The curvature ellipse at a point p of the surface M is the
image ⌘ : S 1 ! Np M , obtained by assigning to each tangent direction ✓
the curvature vector of the normal section ✓ .
We can apply the same arguments to those applied to surfaces in 4-space
to obtain the following expression for the curvature ellipse,

⌘(✓) = H + cos(2✓)B + sin(2✓)C, (8.1)

 Surfaces in the Euclidean 5-space 253

where
H = 12 (a1 + c1 )e3 + 12 (a2 + c2 )e4 + 12 (a3 + c3 )e5 ,
B = 12 (a1 c1 )e3 + 12 (a2 c2 )e4 + 12 (a3 c3 )e5 ,
C = b 1 e3 + b 2 e4 + b 3 e5 .
Definition 8.2. The normal field
1 1 1
H = (a1 + c1 )e3 + (a2 + c2 )e4 + (a3 + c3 )e5
2 2 2
evaluated at a point p is the mean curvature vector of M at p.

As in the previous chapter, we consider (affine) invariant properties of

the surface under the action of GL(2, R) ⇥ GL(3, R) on Tp M ⇥ Np M , where
GL(j, R), j = 2, 3 denotes the general linear group. This action can be
viewed as a change of basis in Tp M and Np M .
Let {f1 , f2 } = {xu1 , xu2 } be the basis of Tp M given by the parametri-
sation x : U ! R5 of the surface . One can complete it at each point p on
M to obtain a basis {f1 , f2 , f3 , f4 , f5 } of R5 varying smoothly with p, with
{f3 , f4 , f5 } a basis of Np M (which is also not necessarily orthonormal).
The analysis of the e↵ects of the change of basis on the coefficients of the
second fundamental form and on the curvature ellipse follows similarly to
Theorems 7.1, 7.2 and 7.3. We state the corresponding results for surfaces
in R5 for completeness.
Let
f 1 = ↵ 1 e1 + 1 e2 ,
f 2 = ↵ 2 e1 + 2 e2 .

Then we have the following.

Proposition 8.1. Denote by ai , bi , ci , i = 1, 2, 3 the coefficients of

the second fundamental form with respect to the basis {e1 , e2 , e3 , e4 , e5 }
and by li , mi , ni , i = 1, 2, 3 its coefficients with respect to the basis
{f1 , f2 , e3 , e4 , e5 }. Then
0 1 0 1
li ai
@ mi A = ⇤ @ bi A i = 1, 2, 3,
ni ci
with
0 1
↵12 2↵1 1 2
1
⇤ = @ ↵1 ↵2 ↵1 2 + ↵2 1 1 2
A and det ⇤ = (det ⌦)3 .
↵22 2↵2 2 2
2
254 Di↵erential Geometry from a Singularity Theory Viewpoint

We write
f 3 = ↵ 1 e3 + 1 e4 + 1 e5 ,
f 4 = ↵ 2 e3 + 2 e4 + 2 e5 ,
f 5 = ↵ 3 e3 + 3 e4 + 3 e5

and denote by
ai = hxu1 u1 , ei+2 i, bi = hxu1 u2 , ei+2 i, ci = hxu2 u2 , ei+3 , i , i = 1, 2, 3
and
li = hxu1 u1 , fi+2 i, mi = hxu1 u2 , fi+2 i, ni = hxu2 u2 , fi+2 i , i = 1, 2, 3
the coefficients of the second fundamental form with respect to the basis
{e3 , e4 } and {f3 , f4 , f5 } respectively. Also denote by
En = hf3 , f3 i, Fn = hf3 , f4 i Gn = hf3 , f5 i,
Hn = hf4 , f4 i, In = hf4 , f5 i, Jn = hf5 , f5 i.
Proposition 8.2. Denote by ai , bi , ci , i = 1, 2, 3 the coefficients of the sec-
ond fundamental form with respect to the basis {e1 , e2 , e3 , e4 , e5 } and by
li , mi , ni , i = 1, 2, 3 its coefficients with respect to the basis {e1 , e2 , f3 , f4 , f5 }.
Then
0 1 0 10 1 10 1
a 1 b1 c 1 ↵1 ↵2 ↵3 E n Fn G n l 1 m1 n 1
@ a 2 b 2 c 2 A = @ 1 2 3 A @ Fn H n I n A @ l 2 m 2 n 2 A .
a 3 b3 c 3 1 2 3 Gn I n J n l 3 m3 n 3
Definition 8.3. We define the subsets
Mi = {p 2 M | rank ↵p = i}, i  3.

The property that rank ↵p = i is invariant by the action of GL(2, R) ⇥

GL(3, R) on Tp M ⇥ Np M.
Given p 2 M , the second fundamental form induces a linear map, Ap :
Np M ! Q2 , from the normal 3-space of M at p to the space Q2 of quadratic
forms on Tp M. It is defined by Ap (v) = IIvp , where IIvp : Tp M ! R is
the second fundamental form at p with respect to any normal direction
v 2 Np M . With respect to the basis B(x) = {xu1 , xu2 } of Tp M, we have
w = w1 xu1 + w2 xu1 , and we can write
Q2 = {aw12 + 2bw1 w2 + cw22 | a, b, c 2 R}.
If v 2 Np M is represented by its coordinates (v1 , v2 , v3 ) with respect to
the basis {e3 , e4 , e5 }, then
3
X
Ap (v3 , v4 , v5 ) = vi (li w12 + 2mi w1 w2 + ni w22 ).
i=1
 Surfaces in the Euclidean 5-space 255

We can identify Q2 with the space of real symmetric matrices

⇢✓ ◆
ab
Q2 = a, b, c 2 R .
bc

Then we have the canonical basis

⇢✓ ◆ ✓ ◆ ✓ ◆
10 01 00
, ,
00 10 01

of Q2 . We remark that the matrix ↵pT is the representation matrix of the

linear mapping Ap with respect to the basis {e3 , e4 , e5 } of Np M and the
above basis of Q2 .
Let C be the cone of degenerate quadratic forms in Q2 , which is defined
by ac b2 = 0 in the above representation and denote by Cp the subset
Ap 1 (C) ⇢ Np M . The following properties are an immediate consequence
of the definition of Ap and do not depend on the choices of basis on Tp M
and Np M.

1) If p 2 M3 , the linear map Ap has maximal rank, so Cp is a cone in

Np M .
2) If p 2 M2 , the image of Ap is a plane through the origin in Q2 . Now,
according to the relative position of the image of Ap with respect to
the cone C, we have the following three possible cases for the set Cp .

(2a) Hyperbolic type (denoted by M2h ): ImAp \ C consists of two lines

and then Cp is the union of two planes intersecting along the line
ker ↵(p).
(2b) Elliptic type (denoted by M2e ): ImAp \ C = {0} and thus Cp is the
line = ker ↵(p).
(2c) Parabolic type (denoted by M2p ): ImAp is tangent to C along a line,
in which case Cp is a plane containing the line ker ↵(p).

3) If p 2 M1 , the image of Ap is a line through the origin in Q2 and Cp

can be either a plane coinciding with KerAp , or the whole Np M .

Proposition 8.3 ([Mochida, Romero Fuster and Ruas (2003)]).

Given a closed surface M , there exists a residual set O in Emb(M, R5 )
such that for any x 2 OH , we have that M = M3 [ M2 . Moreover,

a) M3 is an open subset of M .
b) M2 is a regularly embedded curve.
256 Di↵erential Geometry from a Singularity Theory Viewpoint

Proof. We show first that generically the rank of the second fundamental
form at any point is at least 2. Let x : U ! R5 a local parametrisation of M,
and {xu1 , xu1 , e3 , e4 , e5 } a frame in U. With respect to these coordinates,
for each p 2 U, the second fundamental form at p is represented by the 3⇥3
matrix
0 1
l1 m1 n1
↵p = @ l2 m2 n2 A
l3 m3 n3
where

li = hxu1 u1 , ei+2 i, mi = hxu1 u2 , ei+2 i, ni = hxu2 u2 , ei+2 i, i = 1, 2, 3.

Let M(3) be the set of real 3 ⇥ 3 matrices and Mi (3) = {↵ 2

M(3) |rank ↵  i, i = 0, 1, 2, 3}. Then Mi (3) is a singular variety of codi-
mension (3 i)2 , whose singular set is Mi 1 (3) ([Arbarello, Cornalba,
Griffiths and Harris (1985)]). The decomposition M(3) = [3i=0 Si , where
Si = (Mi (3) \ Mi 1 (3)) is a stratification of M(3).
We consider now the following diagram

j2x ⇧⇤
U ! J 2 (U, R5 ) ! M(3),

where ⇧⇤ (j 2 x(p)) = ↵p .
The mapping ⇧⇤ : J 2 (U, R5 ) ! M(3) is a submersion, as we can
take the variables li , mi and ni as coordinates in the jet space J 2 (U, R5 ).
Then, the stratification (Si ) in M(3) pulls back to a stratification (Ti ) in
J 2 (U, R5 ), such that cod Ti = cod Si .
Then, it follows from Thom’s Transversality Theorem that for a generic
embedding x : U ! R5 , the map j 2 x : U ! J 2 (U, R5 ) does not intersect
the strata Ti , i  1, since they have codimension greater than or equal to
4. Hence, for a generic embedding x, rank ↵p 2, 8p 2 U.
Let (p) = det(↵p ), with p = x(u) 2 M . We have that M M3 =
1
(0) and since is a continuous function on M , it follows that M3 must
be an open region in M .
We can consider a local representation of M in its Monge form at a
point p 2 M ,
: R2 , 0 ! R5
(x, y) 7 ! (x, y, f1 (x, y), f2 (x, y), f3 (x, y)).
In these coordinates
 Surfaces in the Euclidean 5-space 257

(p) = f1xx f2xy f3yy f1xy f2xx f3yy f1xx f2yy f3xy + f1yy f2xx f3xy +
f1xy f2yy f3xx f1yy f2xy f3xx .
Now, it follows from this expression that, under appropriate transver-
sality conditions on the 3-jet of , the set = 0 represents a curve possibly
with isolated singular points determined by the vanishing of the deriva-
tives of the function . Since the orthogonality property of the frame
{e1 , e2 , e3 , e4 , e5 } is irrelevant for our study, we can take {e3 , e4 , e5 } such
that e5 generates Ker(Ap ).
If p 2 M2h , we choose {e3 , e4 } as the two degenerate directions in Np M .
Furthermore, we can also make linear changes of coordinates in source and
target, such that the two degenerate directions correspond to the quadratic
forms x2 and y 2 in C. Thus f can be locally written as

(x, y) = (x, y, x2 + R1 (x, y), y 2 + R2 (x, y), R3 (x, y)),

where Ri 2 m3 , i.e., all the derivatives of the Ri vanish up to order

3, i = 1, 2, 3.
If p 2 M2e , then Im(Ap ) \ C = {0} and we take e5 as the generator of
kerAp = Ap 1 (Im(Ap ) \ C). With additional changes of coordinates, f can
be written as

f (x, y) = (x, y, x2 y 2 + R1 (x, y), xy + R2 (x, y), R3 (x, y)).

If p 2 M2p then analogously, f can be written as

f (x, y) = (x, y, x2 + R1 (x, y), xy + R2 (x, y), R3 (x, y)).

In each of the above cases it is a simple (but tedious) calculation to verify

that under generic conditions on the 3-jet of at (0, 0), the point p is a
regular point of 1
(0). ⇤

Definition 8.4. We denote by Ep and Af fp the vector subspace and the

affine subspace determined by the curvature ellipse respectively in Np M .
That is, Af fp is the affine plane spanned by the vectors C H and B H
and Ep is the parallel plane through the origin (spanned by C and B). The
orthogonal complement of Ep in Np M will be denoted by Ep? .

The points of M are characterised as follows in terms of the relative

positions of these subspaces:

1) If p 2 M3 , then Af fp is a plane that does not contain the origin of

Np M .
258 Differential Geometry from a Singularity Theory Viewpoint

2) If p ∈ M2 , then Af fp is either a plane through the origin of Np M (and

thus coincides with Ep ), or a line that does not contain the origin of
Np M .
3) If p ∈ M1 , then Af fp is a line through the origin p of Np M , that is
Af fp = Ep .

We thus have that the M2 points are either semiumbilics, or points at

which Af fp passes through the origin p of Np M . In the last case, it is not
difficult to see that, similarly to the case of surfaces in R4 , a point p is of
type M2e , M2h or M2p according to the origin p of Np M lies inside, outside,
or on the curvature ellipse at p. The semiumbilic points can be considered
as points of type M2h .

curvature ellipse Affp

Cp

Fig. 8.1 The curvature ellipse and the cone of degenerate directions at a point on a
surface in R5 .

Proposition 8.4. The cone whose basis is the curvature ellipse is perpen-
dicular to the cone Cp of degenerate directions at p.

Proof. Given any normal direction v, the v-principal curvatures at p

are the extremal values of the projection of the normal curvature vectors
at p, that is the extremal values of the projection of the curvature ellipse
 Surfaces in the Euclidean 5-space 259

on the vector line determined by v in Np M . Then it follows that v is

perpendicular to some direction on the cone whose basis is the curvature
ellipse if and only if one of these v-principal curvatures vanishes, which
implies that v lies in the cone of degenerate directions at p. ⇤

8.2 Contacts with hyperplanes

Consider the family of height functions,

H : M ⇥ S4 ! R
(p, v) 7! hv (p).

We observe that, analogously to the case of surfaces in R4 if we denote

by H̄ the height functions family on the canal hypersurface CM , it follows
that p 2 M is a (degenerate) singular point of hv if and only if (p, v) 2
CM is a (degenerate) singular point (of the same K-type) of h̄v and, as a
consequence of Theorem 4.4, we can state:

Theorem 8.1. There is an open and dense set OH in Imm(U, R5 ) such

that for any x 2 OH , the surface M = x(U ) has the following properties:

i) For any v 2 S 4 , the height function hv along the normal direction

v at a point p 2 M (and hence h̄v at (p, v) 2 CM ) has only local
singularities of K- type A1 , A2 , A3 , A4 , A5 , D4 and D5 .
ii) The singularities of hv (resp. h̄v ) are R-versally unfolded by the family
H (resp. H̄).
iii) The generalised Gauss map G : CM ! S 4 is stable as a Lagrangian
map.

Definition 8.5. A surface in R5 is called (locally) height function generic

if any of its local parametrisations belongs to the set OH in Theorem 8.1.

Definition 8.6. Given v 2 Cp ⇢ Np M , the point p is a non-stable singu-

larity of hv . We say that v is a degenerate normal direction at p. In such
case, ker(H(hv )(p)) 6= {0} and any direction u 2 ker H(hv )(p) is called flat
contact direction associated to v. Here, H(hv )(p) is the Hessian matrix of
hv at p.

Definition 8.7. A degenerate direction v 2 Cp for which hv has a singu-

larity A3 or worse (i.e., the K-codimension of hv is at least 2) is said to
be a binormal direction at p. The tangent hyperplane orthogonal to the
260 Di↵erential Geometry from a Singularity Theory Viewpoint

binormal direction has higher order contact with M at p and we call it os-
culating hyperplane at p. The corresponding contact directions are called
asymptotic directions on M . In the special case of the corank 2 singularities
D4± and D5 we refer to them as umbilic binormals.

We say that a point p 2 M is 2-regular if for some coordinate chart

at p, we have that rank {xu1 , xu2 , xu1 u1 , xu1 u2 , xu2 u2 } = 5. If p fails to be
2-regular, we say that it is 2-singular.
We have the following geometrical characterisation for the corank 2
singularities of the height functions on M :

Proposition 8.5. For a local parametrisation x : M ! R5 , the following

conditions are equivalent:
a) p 2 M2 .
b) p 2 M is a singularity of corank 2 for some height function on M .
c) p is 2-singular.

Proof. This follows easily by taking the embedding in Monge form and
observing that the matrix of the second fundamental form in a normal
direction v at p coincides with the Hessian matrix of the height function
hv at p. ⇤

Consequently the generic singularities of the height functions at points

of type M3 are of type Ak , k = 2, 3, 4, 5. At a point of type M2 , there is
some normal direction defining a height function with a singularity of type
D4± or D5 .
The binormal and asymptotic directions at the M3 points can be char-
acterised in terms of normal sections as follows: let v be a degenerate
direction at p 2 M3 and let ✓ be a tangent direction in ker(H(hv )(p)). De-
note by ✓ the normal section of the surface M in the tangent direction ✓,
that is, ✓ is the curve obtained by the intersection of M with the 4-space
V✓ = Np M h✓i.

Proposition 8.6. The direction ✓ 2 Tp M is an asymptotic direction at

p 2 M corresponding to the binormal v if and only if v is the binormal
direction at q of the curve ✓ in the 4-space V✓ .

Proof. Observe that the restriction of the family of height functions H to

(some parametrisation of) the curve ✓ gives the family of height functions
on this curve. Now, the result follows from the fact that a normal vector
 Surfaces in the Euclidean 5-space 261

v is the binormal of a curve in 4-space if and only if the height function in

the direction v over this curve has a singularity of type Ak , k 3. ⇤
We can also characterise the asymptotic directions on the M3 points in
terms of the geometry of the orthogonal projections of the surface into 3-
spaces as follows ([Romero Fuster, Ruas and Tari (2008)]): Given v 2 Nq M
denote by Mv the surface patch obtained by projecting M orthogonally to
the 3-space Tq M hvi (considered as an affine space through q).

Proposition 8.7. Given a point q 2 M3 , we have

(1) A direction v 2 Nq M is degenerate if and only if q is a parabolic
point of Mv . In this case, the unique principal asymptotic direction of Mv
at q coincides with the contact direction associated to v.
(2) A direction ✓ 2 Tq M is asymptotic if and only if there exists v 2
Nq M such that q is a cusp of Gauss of Mv and ✓ is its unique asymptotic
direction there.

Proof. Both assertions follow from observing that in appropriate coordi-

nates, the matrix of the Gauss map of the surface patch Mv at p coincides
with the Hessian matrix of the height function hv at this point. ⇤
Let Kc : CM ! R be the Gaussian curvature function on CM . The
parabolic set of the hypersurface CM is the singular set of the Gauss map
G : CM ! S 4 , given by Kc 1 (0). This is a stratified subset that can be
decomposed in terms of the Thom-Boardman symbols of the singularities
of the generalised Gauss map G as Kc 1 (0) = S1 (G) [ S2 (G), where
S1 (G) = S1,0 (G) [ S1,1,0 (G) [ S1,1,1,0 (G) [ S1,1,1,1,0 (G)
and
S2 (G) = S2,0 (G) [ S2,1,0 (G).
We recall that
S1,1 (G) = {(p, v) 2 S1 (G) | Ker d(G) ⇢ T(p,v) S1 (G)}.
S1,1,1 (G) = {(p, v) 2 S1,1 (G) | Ker d(G) ⇢ T(p,v) S1,1 (G)}.
S2,1 (G) = {(p, v) 2 S2 (G) | dimKer (d(G) \ T(p,v) S2 (G)) = 1}.
Then S1,0 (G) = S1 (G) S1,1 (G), S1,1,0 (G) = S1,1 (G) S1,1,1 (G) and
so on. Moreover, from the properties of Thom-Boardman symbols ([Golu-
bitsky and Guillemin (1973)]) we get that for a generic immersion

a) S1 (G) is a 3-dimensional submanifold of CM .

b) S1,1 (G) is a regular surface.
c) S1,1,1 (G) and S2 (G) are regular curves in CM .
262 Di↵erential Geometry from a Singularity Theory Viewpoint

d) S1,1,1,1 (G) and S2,1 (G) are made of isolated points.

Taking into account the relation between the singularities of the gener-
alised Gauss map G, as a catastrophe map of the height functions family,
and the singularities of the height functions on the surface M (see Remark
3.2), we observe:

a) (p, v) 2 S1,0 (G) if and only if p is a singularity of type A2 of hv .

b) (p, v) 2 S1,1,0 (G) if and only if p is a singularity of type A3 of hv .
c) (p, v) 2 S1,1,1,0 (G) if and only if p is a singularity of type A4 of hv .
d) (p, v) 2 S1,1,1,1,0 (G) if and only if p is a singularity of type A5 of hv .
e) (p, v) 2 S2,0 (G) if and only if p is a singularity of type Dk± , k = 4, 5 of
hv .

Then we can characterise the flat contact directions and the asymptotic
directions as follows:

1) (p, v) 2 S1 (G) if and only if v is a degenerate direction at p.

2) (p, v) 2 S1,1 (G) if and only if v is an asymptotic direction at p.

Denote by ⇠ : CM ! M the natural projection and by ⇠1 : S1 (G) ! M

and ⇠2 : S1,1 (G) ! M its restrictions to the 3-manifold S1 (G) and to
the surface S1,1 (G) respectively. Observe that there is a unique principal
asymptotic direction (in the sense that it has zero principal curvature) at
each point of S1 (G). The following result is a consequence of the above
considerations relating the singularities of the generalised Gauss map G on
CM and the singularities of height functions on M .

Proposition 8.8. The flat contact directions at p 2 M are the projections

of the principal asymptotic directions of CM at the points (p, v) 2 ⇠ 1 (p)
through the derivative d⇠1 (p, v) : T(p,v) S1 (G) ! Tp M .
The (unique) principal asymptotic direction at any point (p, v) 2 S1,1 (G)
is projected through d⇠2 onto an asymptotic direction at p 2 M .

We can study now the possible numbers and the distribution of asymp-
totic directions over the points of a generic surface M .

Proposition 8.9 ([Mochida, Romero Fuster and Ruas (2003)]).

There are at least one and at most 5 asymptotic directions at each point of
M3 .
 Surfaces in the Euclidean 5-space 263

Proof. Let M be given in the Monge form

(x, y) = (x, y, f1 (x, y), f2 (x, y), f3 (x, y))
in a neighbourhood of p = (0, 0) 2 R2 , with fi (x, y) = Qi (x, y) +
Ki (x, y) + Ri (x, y), where Qi are quadratic forms, Ki are cubic forms and
Ri 2 M42 , i = 1, 2, 3. Since p 2 M3 , the 3 quadratic forms Q1 (x, y), Q2 (x, y)
and Q3 (x, y) are linearly independent and without loss of generality we can
take local coordinates at p in such a way that Q3 (x, y) = (x2 + y 2 ) and
K3 (x, y) = 0. To simplify the notation we denote as Qi (w, z), i = 1, 2, 3 the
bilinear form d2 fi (x, y)(w, z), where w, z 2 Tq M and q = (x, y) varies in
a small enough neighbourhood of p in R2 . Analogously Ki (w3 ), i = 1, 2, 3
denotes the cubic form associated to x at the point q acting on a vector
w 2 Tq M .
Let v 2 Nq M be a solution of the equation Aq (v) = 0. Then
H(hv (q)) is a degenerate quadratic form and so there is w 2 Tq M such
that H(hv (q))(w, z) = 0, for any z 2 Tq M . By writing v = v1 e1 +
v2 e2 + v3 e3 2 Nq M in terms of the normal frame {e1 , e2 , e3 } we have
v1 Q1 (w, z) + v2 Q2 (w, z) v3 hw, zi = 0. This expression must be true in
particular for the vector w and a vector z 2 Tq M orthogonal to w, so we
have the equations,

v1 Q1 (w, w) + v2 Q2 (w, w) v3 (w12 + w22 ) = 0 (1)

v1 Q1 (w, z) + v2 Q2 (w, z) = 0 (2)

On the other hand, q is a singular point of cusp type or worse if the
vector w satisfies v1 K1 (w3 ) + v2 K2 (w3 ) + v3 K3 (w3 ) = 0 (see [Mochida,
Romero Fuster and Ruas (2003)]). And since in the chosen local coordi-
nates K3 (w1 , w2 ) = 0, this gives
v1 K1 (w3 ) + v2 K2 (w3 ) = 0 (3)
Given v1 and v2 we can obtain v3 from (1). On the other hand, elimi-
nating v1 and v2 in (2) and (3) gives
K1 (w3 )Q2 (w, z) K2 (w3 )Q1 (w, z) = 0.
Since the coordinates of the vector z (orthogonal to w in Tq M ) can be
given as a linear combination of those of w, we obtain that for each q in
a neighbourhood of p = (0, 0), the above equation is defined by a quintic
form in two variables. This gives the di↵erential equation for the asymptotic
curves on M . We observe that this equation cannot be identically zero on
regular points of order 2 of a generic surface. ⇤
264 Di↵erential Geometry from a Singularity Theory Viewpoint

A di↵erent proof of the above result appears in [Romero Fuster, Ruas

and Tari (2008)]. The integral lines of the fields of asymptotic directions are
the asymptotic curves on M . They are the solutions of a quintic equation
of the form:
A0 dy 5 + A1 dxdy 4 + A2 dx2 dy 3 + A3 dx3 dy 2 + A4 dx4 dy + A5 dx5 = 0
whose coefficients Ai (i = 0, ..., 5) depend on those of the second fundamen-
tal form, (l1 , m1 , n1 ), (l2 , m2 , n2 ), (l3 , m3 , n3 ) and their first order partial
derivatives, and are given by
A0 = [ @n
@y , m, n]

A1 = [ @n @m @n
@x , m, n] + 2[ @y , m, n] + [ @y , l, n]

A2 = [ @n @m @l @m @n
@x , l, n] + 2[ @x , m, n] + [ @y , m, n] + 2[ @y , l, n] + [ @y , l, m]

@l
A3 = [ @x , m, n] + 2[ @m @n @m @l
@x , l, n] + [ @x , l, m] + 2[ @y , l, m] + [ @y , l, n]

@l
A4 = [ @x , l, n] + 2[ @m @l
@x , l, m] + [ @y , l, m]

@l
A5 = [ @x , l, m]
where l = (l1 , l2 , l3 ), m = (m1 , m2 , m3 ), n = (n1 , n2 , n3 ).
Definition 8.8. The singular set of ⇠2 : S1,1 (G) ! M is known as the
criminant set and its image by ⇠2 , denoted by , is the discriminant set.
The discriminant set is a (non necessarily connected) curve separating
M into regions made of points with a constant number of asymptotic di-
rections. The possible local configurations of the asymptotic curves on a
surface generically immersed in R5 are described in the following proposi-
tion whose proof can be found in [Romero Fuster, Ruas and Tari (2008)].
Proposition 8.10.
(1) The local configurations of the asymptotic curves of a height function
generic surface in R5 are modeled by super-imposing in each quadrant
in Figure 8.2 1 and 2 one figure from the left column with one from
the right column.
(2) Let q 2 M3 be a point on the discriminant of the asymptotic IDE and
u the double asymptotic direction there. Then q is a folded-singularity
of the asymptotic IDE at (q, u) if and only if q is an A4 -singularity of
the height function along u⇤ (Figure 8.2 1 (a or b)+(g, h or i)).
 Surfaces in the Euclidean 5-space 265

(3) The discriminant ∆ intersects transversally the M2 -curve at M2p and

D5 -points (it may also intersect it at other points). The D5 -points are
generically not folded singularities, so the configuration of the asymp-
totic curves at such points is as in Figure 8.2 ⃝ 1 (a or b)+(f). An
p
M2 -point is (at the appropriate direction) a folded singularity of the
IDE of the asymptotic curves and the configurations there are as in
Figure 8.2 ⃝1 (a or b)+(g, h or i).

! %

&

+ "

' +
$

( #
#

)
$

Fig. 8.2 Configurations of the asymptotic curves on surfaces in R5 .

Definition 8.9. The flat ridge FR of M is the set of points for which there
exists a height function (in some binormal direction), having a singularity
of type Ak , k ≥ 4.

It is easy to see that

FR = ξ(S1,1,1 (G)).

Proposition 8.11. Let q ∈ M3 and v ∈ Nq M a binormal direction. Let θ

be its corresponding asymptotic direction and γθ the corresponding normal
section of M . Then
266 Di↵erential Geometry from a Singularity Theory Viewpoint

(1) q = ✓ (0) is a flat ridge point of M if and only if q is a flattening

of ✓ (as a curve in the 4-space V✓ ).
(2) q = ✓ (0) is a higher order flat ridge point of M if and only if q is
a degenerate flattening of ✓ .

Proof. A point q is a singularity of type Ak of the height function hv on

M if and only if it is a singularity of type Ak of hv | ✓ . Therefore it is a
flattening (resp. degenerate flattening) of ✓ if and only if it is a flat ridge
point (resp. higher order flat ridge point) of M . ⇤
Let P 2 S 4 be the north pole, it follows from Theorem 4.11 that the
stereographic projection : S 4 {P } ! R4 takes the singularities of type
Ak and Dk of height functions on a surface M ⇢ S 4 (considered as a surface
in R5 ) into the singularities of the same type for distance squared functions
on (M ) ⇢ R4 . Moreover, the di↵erential of takes contact directions
of M with hyperplanes to contact directions of (M ) with hyperspheres.
Therefore, we have the following:

Corollary 8.1. The stereographic projection maps di↵eomorphically:

1) The asymptotic curves of M into the strong principal lines of (M ).
2) The flat ridges of M to the ridges of (M ).
3) The M2 points of M to the semiumbilic points of (M ).

As a consequence of this, together with the previous results regarding

the generic contacts of surfaces with hyperplanes in R5 , we get the following
results concerning generic contacts of surfaces with hyperspheres in R4 as
announced in §7.9. We point out that from Theorem 4.11 and Remark 4.4
we get that a surface immersed in S 4 is generic from the viewpoint of its
contacts with hyperplanes of R5 if and only if the surface (M ) is generic
from the viewpoint of its contacts with hyperspheres in R4 . We thus have
that the assertion 2) of the following Corollary provides an alternative proof
of Theorem 7.14.

Corollary 8.2. Any distance squared function generic immersion of a sur-

face in R4 satisfies the following properties:

1) There are at least one and at most five strong principal lines through
any point of M . The generic local qualitative behaviour of the strong
principal foliations of M corresponds to the one described above for the
asymptotic curves of surfaces generically immersed in R5 (see Figure
8.2).
 Surfaces in the Euclidean 5-space 267

2) The semiumbilic points of M form a smooth (non necessarily con-

nected) curve.
3) The discriminant curve ⌥ of the strong principal directions IDE of a
surface M immersed in R4 is the image of the discriminant of a sur-
face immersed in S 4 ⇢ R5 by the stereographic projection &. Moreover,
a) The intersection of the ridge curve with ⌥ are folded singularities of
the strong principal directions IDE.
b) The semiumbilic points of type D5 lie at transversal intersections of
⌥ with the curve of semiumbilics, but are not folded singularity of
the strong principal directions IDE.
c) If q is a non semiumbilic point of ⌥, then q is a folded singularity
of the strong principal directions IDE , q is a ridge point of type
A4 .

8.3 Orthogonal projections onto hyperplanes, 3-spaces and

planes

We describe now how the asymptotic directions can be characterised in

terms of singularities of projections onto hyperplanes, 3-spaces and planes.
We analyse the di↵erent possibilities for these three cases.

8.3.1 Contact with lines

If T S 4 denotes the tangent bundle of the 4-sphere S 4 , the family of projec-
tions to 4-planes is given by
P : M ⇥ S4 ! T S4
(q, v) ! (q, Pv (q))
where Pv (q) = q hq, viv. Observe that for a given v 2 S 4 , the map
Pv can be considered locally as a germ of a smooth map R2 , 0 ! R4 , 0 .
As in the case of surfaces in 4-space, it follows from Theorem 4.12 that
there is an open and dense subset OP in Imm(U, R5 ) for which the family
P is a generic family of mappings. So the singularities of Pv that occur
in an irremovable way in the family P are those of Ae -codimension  4,
and these are versally unfolded by the family P . For a generic surface, the
singularities of Pv are simple and are given in Table 4.3.

Definition 8.10. A surface in R5 is called projection P -generic if any of

its local parametrisations belongs to the set OP .
268 Di↵erential Geometry from a Singularity Theory Viewpoint

The bifurcation set of the family of projections P is the set of parameters

v 2 S 4 where Pv has a non-stable singularity at some point q 2 M , i.e.
has a singularity of Ae -codimension 1. The following result provides
a geometric characterisation of the singularities of projections of surfaces
immersed in R5 into hyperplanes.

Proposition 8.12 ([Romero Fuster, Ruas and Tari (2008)]). For

a projection P -generic surface parametrised by x : M ! R5 , we have

1) Given a point p 2 M3 , a direction u 2 Tp M is asymptotic if and only

if the orthogonal projection of M to the 4-space u? has a singularity of
type I2 or worse.
2) For p 2 M2 there are at most 3 and at least 1 directions u 2 Tp M ,
where pu has a singularity of type I2 . Moreover,
2a) If p 2 M2h , there are two asymptotic directions where pu has a sin-
gularity of type II2 .
2b) If p 2 M2e , there are no asymptotic directions where pu has a singu-
larity of type II2 or more degenerate.
2c) If p 2 M2p , there is one asymptotic direction where pu has a singu-
larity of type VII1 at p.
3) There may be a curve in M3 where projecting along one of the asymp-
totic directions yields a singularity of type I3 and isolated points of this
curve where the singularity is of type I4 .
4) There may be isolated points where projecting along one of the asymp-
totic directions yields a singularity of type II3 . There may also be iso-
lated M2h -points where projecting along one of the directions yields a
singularity of type III2,3 . The above points are in general distinct from
the D5 -points.

The proof follows the standard procedure of making successive changes of

coordinates in order to reduce the appropriate jet of Pv to a normal form.

8.3.2 Contact with planes

An orthogonal projection from R5 to a 3-dimensional subspace is deter-
mined by its kernel, so we can parametrise all these projections by the
Grassmanian G(2, 5) of 2-planes in R5 . If w1 , w2 are two linearly indepen-
dent vectors in R5 , we denote by {w1 , w2 } the plane they generate and by
⇡(w1 ,w2 ) the orthogonal projection from R5 to the orthogonal complement
 Surfaces in the Euclidean 5-space 269

of hw1 , w2 i. The restriction of ⇡(w1 ,w2 ) to M , ⇡(w1 ,w2 ) |M , can be consid-

ered locally at a point q 2 M as a map-germ ⇡(w1 ,w2 ) |M : R2 , 0 ! R3 , 0.
We distinguish between the two following cases:

1) The plane {w1 , w2 } is transverse to Tq M , so the projection ⇡(w1 ,w2 ) |M

is locally an immersion.
2) The projection ⇡(w1 ,w2 ) |M is singular.

In the first case, given any v 2 Nq M , let Mv be the surface patch

obtained by projecting M orthogonally to the 3-space Tq M hvi (considered
as an affine space through q). Now, we can take the Gauss map on Mv
considered as a surface in the Euclidean 3-space Tq M hvi. The local
stable singularities of this map are folds and cusps. We have the following
result characterising the degenerate normal directions and the asymptotic
directions at a point q 2 M in terms of the geometry of the surface patch
Mv at the point q.

Proposition 8.13 ([Romero Fuster, Ruas and Tari (2008)]). (1)

Suppose that q 2 M3 and let v 2 Nq M .
(a) The direction v is degenerate if and only if q is a parabolic point of
Mv . In this case, the unique principal asymptotic direction of Mv
at q coincides with the contact direction associated to v.
(b) A direction u 2 Tp M is asymptotic if and only if there exists v 2
Np M such that q is a cusp of Gauss of Mv and u is its unique
asymptotic direction there.
(2) Suppose that q 2 M2 . Then there are two distinct directions v 2 Nq M
if q 2 M2h , none if q 2 M2e , and a unique direction if q 2 M2p , where q
is a cusp of Gauss of Mv and v⇤ is the unique asymptotic direction of
Mv at q. In addition, there is a unique direction v̄ 2 Nq M where Mv̄
has a flat umbilic at q. The asymptotic directions of M at q associated
to v̄ are the tangent directions to the separatrices of the asymptotic
curves of Mv̄ at q (see Figure 8.3).

We deal now with the case when ⇡(w1 ,w2 ) |M is singular. This means
that the kernel of the projection ⇡(w1 ,w2 ) contains a tangent direction at
q. When {w1 , w2 } = Tq M , the map-germ ⇡(w1 ,w2 ) |M has rank zero at the
origin and does not identify the asymptotic directions. We shall assume
that {w1 , w2 } is distinct from Tq M . Then ⇡(w1 ,w2 ) |M has rank 1 at the
origin.
270 Di↵erential Geometry from a Singularity Theory Viewpoint

Fig. 8.3 Configurations of the asymptotic curves on surfaces in R3 at a flat umbilic point
(elliptic left and hyperbolic right). There are three separatrices at an elliptic umbilic
and one separatrix at a hyperbolic umbilic ([Bruce and Tari (1995)]).

The map ⇡(w1 ,w2 ) can be considered locally as a corank 1 smooth map-
germ R2 , 0 ! R3 , 0. Since ⇡(w1 ,w2 ) is a 6-parameter family, we expect the
map ⇡(w1 ,w2 ) to have only simple singularities of Ae -codimension  6 or
non-simple singularities with the Ae -codimension of the stratum  6 (as
dim G(2, 5) = 6). The A-simple singularities of map-germs R2 , 0 ! R3 , 0 of
Ae -codimension  6 are given in Table 4.2. The complete classification of
non-simple singularities of Ae -codimension of the stratum  6 is not known
so far.

We now state the following result for a generic immersion of a surface

in R5 whose detailed proof can be found in [Romero Fuster, Ruas and Tari
(2008)].

Proposition 8.14. Let u 2 Tq M and v in the unit sphere S 2 ⇢ Nq M .

(1) The projection ⇡(u,v) |M has a cross-cap singularity for almost all
v 2 S2.
(2) On a circle of directions v in S 2 minus a point, ⇡(u,v) |M has a
singularity with 2-jet A-equivalent to (x, y 2 , 0).
(3) There is a unique direction v 2 S 2 where ⇡(u,v) |M has a singularity
of type Hk provided u is not an asymptotic direction. If u is asymptotic,
then the singularity becomes non-simple with 2-jet A-equivalent to (x, xy, 0).

8.3.3 Contact with 3-spaces

An orthogonal projection from R5 to a 2-dimensional subspace is also de-
termined by its kernel and hence we can parametrise all these projections
by the Grassmanian G(3, 5) of 3-planes in R5 . However, G(3, 5) can be
identified with G(2, 5), so the projections can also be parametrised by
{w1 , w2 } 2 G(2, 5), where {w1 , w2 } is the orthogonal complement of the
 Surfaces in the Euclidean 5-space 271

kernel of the projection. We denote the associated projection by ⇧(w1 ,w2 ) .

The restriction ⇧(w1 ,w2 ) |M of ⇧(w1 ,w2 ) to M can be considered, locally at
a point q 2 M , as a map-germ

⇧(w1 ,w2 ) |M : R2 , 0 ! R2 , 0.

As in the previous section, we expect that simple singularities of Ae -

codimension  6 or non-simple singularities with the Ae -codimension of
the stratum  6 to occur for generic surfaces. The list of corank 1 singular-
ities of map-germs R2 , 0 ! R2 , 0, of Ae -codimension  6 is given by Rieger
([Rieger (1987)]). The corank 1 simple singularities are given in Table 7.6.
The A-simple germs of corank 2 are classified in [Rieger and Ruas (1991)].
The complete classification of non-simple singularities from the plane to
the plane of Ae -codimension of the stratum  6 is not known so far.
We start with the corank 1 singularities. Let u 2 Tq M , u? an orthog-
onal vector to u in Tq M and take v = (v1 , v2 , v3 ) 2 Nq M . We consider
the projection ⇧(u? ,v) |M . We take M in Monge form at the origin and
suppose, without loss of generality, that the intersection of the kernel of
⇧(u? ,v) with Tq M occurs along u = (1, 0). Then we have

⇧(u? ,v) |M (x, y) = (y, v1 (x2 +f1 (x, y))+v2 (xy+f2 (x, y))+v3 (cy 2 +f3 (x, y))),

where c is equal to 0 or 1 according to q being an M3 or an M2 point.

Observe that the A-type of the singularities of the above map-germ is in-
dependent of c. Therefore, the corank 1 singularities of the projections to
planes do not distinguish between the M3 and M2 points.
If v1 6= 0, then ⇧(u? ,v) |M is A-equivalent to a fold map-germ.
If v1 = 0 and v2 6= 0, then j 2 ⇧(u? ,v) |M ⇠A (y, xy). The A-singularities
of ⇧(u? ,v) |M are given by the normal forms 5, 6 and 7 in Table 7.6.
Non-simple singularities such that the codimension of the stratum is Ae -
codimension  6 may also occur.
If v1 = v2 = 0, ⇧(u? ,v) |M (x, y) = (y, f3 (x, y)), and the singularities
are of type 4k unless f3 xxx (0, 0) = 0. In this case, the singularities are of
type 112k+1 (Table 7.6) or more degenerate. The condition f3 xxx (0, 0) = 0
is precisely the condition for u = (1, 0) to be an asymptotic direction at
the origin. Therefore we can characterise asymptotic directions in terms of
corank 1 singularities of projections to 2-planes.

Proposition 8.15. Let u 2 Tq M and v in the unit sphere S 2 ⇢ Nq M .

(1) The projection ⇧(u? ,v) |M has a fold singularity for almost all v 2 S 2 .
272 Di↵erential Geometry from a Singularity Theory Viewpoint

(2) On a circle of directions v in S 2 minus a point ⇧(u? ,v) |M has a sin-

gularity with 2-jet A-equivalent to (x, xy) (equivalently, it is not a fold
and has a smooth critical set).
(3) There is a unique direction v 2 S 2 where ⇧(u? ,v) |M has a singularity
of type 4k provided u is not an asymptotic direction. If u is asymptotic,
then the singularity is A-equivalent to 112k+1 or is more degenerate.

We analyse now the corank 2 singularities of the projection. Let

{w1 , w2 } be a plane in Nq M and denote by M(w1 ,w2 ) the surface patch
obtained by projecting M orthogonally to the 4-space Tq M {w1 , w2 } (con-
sidered as an affine space through the point q). The map-germ ⇧(w1 ,w2 ) |M
has then a corank 2 singularity at the origin, and this singularity can be
characterised as follows in terms of the geometry of M(w1 ,w2 ) .

Proposition 8.16 ([Romero Fuster, Ruas and Tari (2008)]). The

assertions hold for a generic immersed surface M in R5 .

(1) The 2-jet of the projection ⇧(w1 ,w2 ) |M is A-equivalent to (x2 , y 2 ), (x2
y 2 , xy) or (x2 , xy) if and only if q is, respectively, a hyperbolic, elliptic
or parabolic point of M(w1 ,w2 ) .
(2) The 2-jet of the projection ⇧(w1 ,w2 ) |M is A-equivalent to (x2 + y 2 , 0),
(x2 y 2 , 0), or (x2 , 0) if and only if q is, respectively, an inflection point
of real type, of imaginary type or of flat type of M(w1 ,w2 ) .
Moreover, if q 2 M3 then ⇧(w1 ,w2 ) |M satisfies (1) for every plane
{w1 , w2 } ⇢ Nq M . The point q 2 M2 if and only if there exists a
direction w2 2 Nq M such that q is an inflection point of M(w1 ,w2 ) , for
any w1 2 Nq M.

The proof of these assertions follows directly from the classification of

points of surfaces in R4 described in Chapter 7 and the A-classification of
map-germs from the plane to the plane ([Rieger (1987); Rieger and Ruas
(1991)]).

8.4 Contacts with hyperspheres

Consider the family of distance squared functions on M = x(R2 ) ⇢ R5 ,

D : R2 ⇥ R 4 ! R
(u, a) 7 ! kx(u) ak2 = da (u)
 Surfaces in the Euclidean 5-space 273

The catastrophe manifold and map are respectively given by the normal
bundle of M in R5 , N M = {(u, a) 2 U ⇥ R5 | a = x(u) + v, v 2 (Np M )1 }
and the normal exponential map,

Ge : N M ! R5
(u, a) 7 ! a.

Then we have the following description of the generic singularities of

the distance squared functions as a consequence of Theorem 4.4.

Theorem 8.2. There is an open and dense set OD of proper immersions

x : U ! R5 such that for any x 2 OD , the surface M = x(U ) has the
following properties:

(i) Given any point p 2 M , the distance squared function da from a point
a 2 R5 such that a p 2 Np M has only singularities of K- type
A1 , A2 , A3 , A4 , A5 , A6 , D4 , D5 and D6 .
(ii) The singularities of da are R-versally unfolded by the family D.
(iii) The normal exponential map Ge is stable as a Lagrangian map.

Definition 8.11. A surface in R5 is called distance squared function generic

if any of its local parametrisations belongs OD in Theorem 8.2.

Definition 8.12. A focal centre at p 2 M is a point a at which Da has a

degenerate singularity. The directions lying in the kernel of the correspond-
ing Hessian quadratic form are said to be spherical contact directions at p.
A focal centre a is said to be an umbilical focus at p if Da has a singularity
of corank 2 at p. The corresponding focal hypersphere is called umbilical
focal hypersphere.
The set of the focal centres of the points of M is said to be the focal set
of M . We denote it by F.

We observe that the proofs of Lemmas 7.2 and 7.3 can be easily adapted
to surfaces in R5 , so we can state the following:

Proposition 8.17. For a point a = p + µv lying in the normal plane of

M at p (i. e. v 2 Np M ), we have that H(da )(u) = 2(I µH(hv )(u)) and
moreover, if a = p + µv is a focal centre of M at p, the spherical contact
directions associated to the distance squared functions from focal centres at
p are eigenvectors of the shape operators Sv in the normal direction v at p
with eigenvalues 1/µ.
274 Di↵erential Geometry from a Singularity Theory Viewpoint

The focal set of M clearly coincides with the bifurcation set of the family
D, given by

F = {a 2 R5 : 9p = x(u) 2 M |da has a degenerate singularity at u}.

The focal set F of a distance squared generic embedding is an algebraic

variety of dimension 4 in R5 .

Definition 8.13. The rib set of M is the singular set of the focal set F.

As in the case of height functions, we can relate the singularity types

of the normal exponential map Ne , as a catastrophe map of the family of
distance squared functions, with those of the distance squared functions on
the surface M (see Remark 3.2):

a) (u, a) 2 S1,0 (Ne ) if and only if u is a singularity of type A2 of da .

b) (u, a) 2 S1,1,0 (Ne ) if and only if u is a singularity of type A3 of da .
c) (u, a) 2 S1,1,1,0 (Ne ) if and only if u is a singularity of type A4 of da .
d) (u, a) 2 S1,1,1,1,0 (Ne ) if and only if u is a singularity of type A5 of da .
e) (u, a) 2 S2 (Ne ) if and only if u is a singularity of type Dk± , K = 4, 5 of
da .

So the focal set F is a stratified set that can be decomposed as

S1,1 (Ne ) [ S2 (Ne ), where S1,1 (Ne ) can be decomposed in turn as a union
S1,1,0 (Ne ) [ S1,1,1,0 (Ne ) [ S1,1,1,1,0 (Ne ) [ S1,1,1,1,1,0 (Ne ) and S2 (Ne ) =
S2,0 (Ne )[S2,1,0 (Ne ). These are respectively stratified 3- and 2-dimensional
subsets of F in R5 .

Definition 8.14. The ridge curves on M are the connected components

of the image of S1,1,1,1 (Ne ) by the natural projection ⇡N : N M ! M .

The ridge curves are made of singularities of type Aj 5 of the distance

squared functions on M . For a distance squared function generic surface,
the ridge points of type A6 are isolated points on curves of points of type
A5 . On the other hand, the projection of S2,1,0 (Ne ) by ⇡N on M is a curve
made of singularities of type D5 of distance squared functions on M with
isolated points of type D6± .
Given an immersion x of M in R5 for p = x(0) 2 M and v = a p 2
Np M we have

hxu1 u1 (0, 0), vi 1 hxu1 u2 (0, 0), vi
H(da )(0, 0) = 2 .
hxu1 u2 (0, 0), vi hxu2 u2 (0, 0), vi 1
 Surfaces in the Euclidean 5-space 275

So the focal set of M at p is given by

n o
Fp = a 2 R5 |a = p + v, with v 2 Wp ,

where Wp is the subset of Np M made of vectors v 2 Np M satisfying,

2
(hxu1 u1 (0, 0), vi 1)(hxu2 u2 (0, 0), vi 1) hxu1 u2 (0, 0), vi = 0.
A straightforward calculation shows that
n o
2 2 2
Wp = v 2 Np M |(hH, vi 1) = hB, vi + hC, vi ,

where H, B and C are the vectors that determine the curvature ellipse of
M at p. We can express the focal set at p, in terms of these vectors:
n o
2 2
Fp = a = p + v 2 Np M ; (hH, vi 1)2 = hB, vi + hC, vi .

The analysis of the above expression for Wp at the di↵erent points of a

surface generically immersed in R5 leads to the following result.

Proposition 8.18 ([Costa, Moraes and Romero Fuster (2009)]).

Given a surface M immersed in R5 , a point p 2 M , let H? be the pro-
jection of the mean curvature vector H on the orthogonal complement Ep?
of Ep in Np M .
?
H
(1) If p 2 M3 then Fp is a cone with vertex at kH? k2 .

(2) For p 2 M2 and dim(Ep ) = 2 all focal points at p have corank 1 and
we may have:
(a) If p 2 M2e then Wp = ?.
(b) If p 2 M2h then Wp is a hyperbola in Ep .
(c) If p 2 M2p then Wp is a parabola in Ep .

We can now analyse the distribution of the umbilical foci of M on its

di↵erent types of points.

Theorem 8.3.
(a) Given a point p 2 M3 , there exists a unique umbilical focus at p, given
by
1
ap = p + vp ,
d(p, Ep )
with vp = unit normal vector orthogonal to the plane Ep , such that
hvp , Hi > 0.
276 Di↵erential Geometry from a Singularity Theory Viewpoint

(b) If p 2 M2 is not semiumbilic, the umbilical focus lies at infinity and

the umbilical focal hypersphere becomes an osculating hyperplane.
(c) If p 2 M2 is semiumbilic then there is a line of umbilical foci at p
contained in the vector plane orthogonal to the affine line Ep .

Proof. With analogous argument to that of Theorem 7.14 we can see

that if a point a = p + v is an umbilical focus then the (unit) vector v
must be orthogonal to the vectors B and C, i.e. v 2 Ep? .
Then, provided Ep is a plane, we get from the condition that rank Hess
1
Da (p) = 0 that = hH,vi . And the statement a) follows from observing
that hH, vi = ±d(p, Ep ), with sign + or according hvp , Hi > 0 or < 0.
For statement b) we just observe that H, B, C are linearly dependent and
d(p, Ep ) = 0. In this case, there is no proper umbilical focal hypersphere,
but the height function in the direction vp has a corank 2 singularity at
p and hence, the hyperplane with orthogonal direction vp is an osculating
hyperplane (with corank 2 contact) that can be considered as a degenerate
umbilical focal hypersphere.
In the case that p is a semiumbilic point, Ep? is a plane and the condition
rank Hess Da (p) = 0 is fulfilled by a whole line of points in Ep? . ⇤

Definition 8.15. Given M R5 and p 2 M , we define the umbilical curvature

of M at p as
u (p) = d(p, Ep ).

Provided M has no semiumbilic points, we can give a sign to the function

u (p) by defining
B⇥C
(u) = hH, i,
kB ⇥ Ck
where B, C, H are the 3 vectors that determine the curvature ellipse at the
point p.

Corollary 8.3.
(a) If p 2 M3 , then u (p) is the curvature of the unique umbilic focal
hypersphere of M at p.
(b) If p 2 M2 is a non semiumbilic point, then u (p) = 0.
(c) If p 2 M2 is a semiumbilic point, u (p) is the maximum curvature
among those of all the tangent hyperspheres centred at umbilical foci of
M at p.
(d) If p is 2-regular then u (p) 6= 0.
 Surfaces in the Euclidean 5-space 277

Proof. The proofs of all these assertions follow easily from the definition
of u . ⇤

8.5 Notes

Relative mean curvature foliations: The study of the mean directional con-
figurations carried out by Mello ([Mello (2003)]) on surfaces in R4 has
been generalised to surfaces in Rn , n 5 in [Gonçalves, Martı́nez Alfaro,
Montesinos Amilibia and Romero Fuster (2007)].
We point out first that surfaces immersed in codimension higher than
2 do not admit mean curvature directions in the way defined for surfaces
in R4 . Nevertheless it is possible to introduce certain foliations on surfaces
immersed in Rn , that in the particular case of surfaces in 4-space coincide
with the mean curvature foliations. The procedure is based on the fact
that, from a qualitative viewpoint, all the principal configurations on S
arise from normal vector fields parallel to the subspace determined by the
curvature ellipse at every point. In fact, any normal vector v 2 Np S can
be decomposed into a sum v > + v ? , with v > and v ? respectively parallel
and orthogonal to the plane determined by the curvature ellipse. Since the
shape operator associated to v ? is a multiple of the identity ([Moraes and
Romero Fuster (2005)]), then the eigenvectors of the shape operator Wv
and Wv> must coincide. This idea lead in [Gonçalves, Martı́nez Alfaro,
Montesinos Amilibia and Romero Fuster (2007)] to the definition of the
relative mean curvature directions at a point p of a surface immersed in Rn
with n > 4 as those inducing normal sections with curvature vector par-
allel to H(p)> . This gives rise to two orthogonal foliations whose critical
points are the semiumbilics and the pseudo-umbilics (with inflection points
and minimal points considered as non generic particular cases). Interest-
ing global consequences of this setting are the following facts ([Gonçalves,
Martı́nez Alfaro, Montesinos Amilibia and Romero Fuster (2007)]):

a) The mean curvature vector of a 2-sphere generically immersed in

Rn , n 5 becomes orthogonal to the normal subspace determined by
the curvature ellipse in at least 4 points.
b) Closed oriented surfaces with non-vanishing Euler number immersed
into Rn , n 5 always have either semiumbilic, pseudo-umbilic, inflec-
tion, or minimal points.
c) Closed oriented 2-regular surfaces with non-vanishing Euler number in
278 Di↵erential Geometry from a Singularity Theory Viewpoint

Rn , n 5, always have pseudo-umbilic points.

Immersions in higher codimension: The curvature locus at a point p of

an m-manifold M immersed in Rn is a projection of the Veronese submani-
fold of dimension m 1 into the normal space of the submanifold at p. The
possible shapes and topological types of this projection for 3-manifolds im-
mersed in Rn are analysed in [Binotto (2008)] (see also [Binotto, Costa and
Romero Fuster (2015)] ). Figure 8.4 illustrates the curvature locus of the
immersion f : R3 ! R6 ; f (x1 , x2 , x3 ) = (x1 , x2 , x3 , x21 + x22 x33 , x1 x3 , x1 x2 )
at a point near the origin.

Fig. 8.4 Curvature locus at a generic point of a 3-manifold in R6 .

The generic behaviour of the ridges and flat ridges of 3-manifolds in

R , n 4 has been analysed in [Nabarro and Romero Fuster (2009)] show-
n

ing that they can be seen as the image of a stable map from a convenient
surface (non necessarily connected) into the 3-manifold. On the other hand,
Dreibelbis ([Dreibelbis (2012)]) generalises to n-manifolds in R2n the no-
tions of asymptotic directions and parabolic and inflection points of sur-
faces in R4 , analysing in detail the case of 3-manifolds in R6 . An interesting
fact, described in this work is that the parabolic subset of a 3-manifold M
generically immersed in R6 is a surface with normal crossings and possible
isolated cross-caps. Moreover, the cross-caps (resp. triple points) are the
points of M for which the origin of the normal space is a cross-cap (resp.a
triple) point of the curvature locus.
 Surfaces in the Euclidean 5-space 279

kth-regular immersions: The concept of kth-regular immersion of a sub-

manifold in Euclidean space was introduced by Pohl ([Pohl (1962)]) and
Feldman ([Feldman (1965)]) in terms of maps between osculating bundles.
In the case of curves, the k-regularity condition means that the first k
derivatives are linearly independent. An interesting question arises when
we consider the problem from a global viewpoint: Under what conditions
can we ensure the existence of 2-regular immersions from a given class
of closed submanifold? For surfaces in R4 , the 2-singular points coincide
with the inflection points and local convexity is an obstruction for the 2-
regularity of closed surfaces with non-vanishing Euler characteristic (see
Chapter 7). In the case of surfaces in R5 , a classically known example of 2-
regular immersion of S 2 into R5 is given by the Veronese surface. Examples
of 2-regular immersions of closed orientable surfaces with non-zero genus
are not known so far. The existence of 2-regular embeddings of orientable
closed surfaces in S 4 is investigated in [Romero Fuster (2004)]. Some ob-
structions to the 2-regularity of surfaces in R5 are given in terms of the
umbilical curvature ([Romero Fuster (2007)]).
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws

This page intentionally left blank

 Chapter 9

Spacelike surfaces in the Minkowski

space-time

The Lorentz-Minkowski space, also called Minkowski space-time, provides

a mathematical setting in which Einstein’s theory of special relativity is
most conveniently formulated. In this setting, the three ordinary dimen-
sions of space are combined with a single dimension of time to form a
four-dimensional manifold for representing space-time. The geometry on
this space is defined by the Poincaré’s group or the group of Lorentz trans-
formations. The geometrical properties of submanifolds of Minkowski space
can be studied, in a similar way to those of Euclidean space, through the
analysis of their contact with model submanifolds invariant by the Lorentz
transformations group. The situation is richer in this case than in the
Euclidean one due to the di↵erent possibilities for the type of the subman-
ifolds and of the models. The submanifolds can be spacelike, timelike or
lightlike and the models can be taken to be the lightlike hyperplanes or
hyperspheres.
The geometrical properties associated to the contact of spacelike sub-
manifolds with spacelike models do not di↵er much from those of submani-
folds of Euclidean space. A new interesting and geometrically rich situation
arises when considering the contact of submanifolds with lightlike hyper-
planes or with lightcones. We call the geometric properties obtained from
such contact Lightlike Geometry and devote this chapter to introducing it.
We point out that Lightlike Geometry can be seen as a generalisation of
Horospherical Geometry, which is concerned with the study of geometric
properties derived from the contact of submanifolds with hyperhorospheres
in hyperbolic space.
We shall restrict in this chapter to spacelike surfaces in 4-dimensional
Minkowski space-time (surfaces in Euclidean 3-space and surfaces in hy-
perbolic space being special cases). However, the results are valid for any

281
282 Di↵erential Geometry from a Singularity Theory Viewpoint

codimension 2 spacelike submanifold of a higher dimensional Minkowski

space-time.
We highlight the following important features of Lightlike Geometry of
surfaces in Minkowski 4-space.

(i) The study of contact of spacelike surfaces with lightlike hyperplanes

in Minkowski 4-space has interesting similarities to that of surfaces
in Euclidean 3-space with planes. For instance, as we shall see in
Chapter 10, some global results in Euclidean Geometry, such as the
Gauss-Bonnet Theorem, can be formulated in a natural way in Lightlike
Geometry.
(ii) Lightlike 3-manifolds, whose study is of special interest in Cosmology,
present a challenge for the classical di↵erential geometry techniques.
However, they can be viewed (at least locally) as line bundles with
lightlike fibres over spacelike surfaces and their contact with lightlike
hyperplanes provides a powerful tool for studying their geometry. In
particular, it allows us to describe their generic singularities as well
as to introduce properties, such as flatness, which are invariant under
Lorentz transformations (see [Izumiya and Romero Fuster (2007)]).

The structure of this chapter is as follows. In the first three sections we

discuss the basic tools, leading to the concepts of Lightcone Gauss map,
normalised lightcone Gauss map and their corresponding Gauss-Kronecker
curvatures on a spacelike surfaces. In §9.5 we define the family of lightcone
height functions and study their properties. This allows us to describe in
§9.6 the normalised lightcone Gauss map as a Lagrangian map whose gen-
erating family is the family of lightcone height functions on the considered
surface. As a consequence, we define the lightcone pedal in §9.7 and study
its properties from the Legendrian viewpoint. In §9.8 we study some special
cases of spacelike surfaces. These are surfaces in spacelike hyperplanes (i.e.
in Euclidean 3-space), and spacelike surfaces contained in either timelike
hyperplanes (i.e. in Minkowski 3-space), de Sitter 3-space or the hyperbolic
3-space. In §9.9 we study the contact of spacelike surfaces with lightcones
and its link with lightlike 3-dimensional submanifolds. In §9.10 we intro-
duce several Legendrian dualities between pseudo-spheres generalising the
well known classical projective (and spherical) duality. Finally in §9.11 we
use these dualities to study spacelike surfaces in the lightcone and prove an
analogous Theorema Egregium of Gauss for these surfaces, showing that
the mean curvature (as defined in §9.11) is an intrinsic invariant and is
 Spacelike surfaces in the Minkowski space-time 283

equal to the section curvature.

9.1 Minkowski space-time

We give a brief introduction to Minkowski space-time and refer to [O’Neill

(1983)] for more details.
The 4-dimensional Minkowski space-time (R4 , h, i1 ), denoted by R41 , is
the vector space R4 endowed with the pseudo scalar product
hu, vi1 = u0 v 0 + u1 v 1 + u2 v 2 + u3 v 3 ,
with u = (u0 , u1 , u2 , u3 ) and v = (v0 , v1 , v2 , v3 ) any two vectors in R4 . A
non-zero vector u is said to be
spacelike if hu, ui1 > 0,
timelike if hu, ui1 < 0,
lightlike if hu, ui1 = 0.

Two vectors u, v are said to be pseudo-orthogonal p if hu, vi1 = 0.

The norm of a vector u is defined to be kuk1 = |hu, ui1 |.
A linear plane (i.e., a 2-dimensional vector subspace of R41 ) is said to be
spacelike if all its non-zero vectors are spacelike. It is timelike if it contains
a timelike and a spacelike vector. It is lightlike if it contains a single lightlike
direction.
Any linear hyperplane (i.e., a 3-dimensional vector subspace of R41 ) can
be defined as the set of vectors u 2 R41 which are pseudo-orthogonal to
a given non-zero vector v. The hyperplane is said to be spacelike (resp.
timelike, lightlike) if v is timelike (resp. spacelike, lightlike).
We choose a system of coordinates (O, {e0 , e1 , e2 , e3 }) in R41 , where O
is referred to as the origin and where {e0 , e1 , e2 , e3 } denotes the canonical
basis of the vector space R4 . We can now view R41 , in the usual way, as a
set of points by identifying a point p with the vector Op.
For any non-zero vector v and c 2 R, the hyperplane with the pseudo-
normal v is the set of points
HP (v, c) = {p 2 R41 | hp, vi1 = c}.
We say that HP (v, c) is spacelike (resp. timelike, lightlike) if v is time-
like (resp. spacelike, lightlike). p
A hypersphere of centre q and radius |r| is the set of points
{p 2 R41 | hp q, p qi1 = r}.
284 Di↵erential Geometry from a Singularity Theory Viewpoint

There are three types of hyperspheres determined by r < 0, r = 0,

r > 0. The hyperspheres of centre q are the translations to the centre q of
the following hyperspheres of centre the origin.
The hyperbolic 3-space:
3
H+ ( 1) = {p 2 R41 | hp, pi1 = 1, p0 > 0}.
The de Sitter 3-space:
S13 = {p 2 R41 | hp, pi1 = 1}.
The open lightcone:
LC ⇤ = {p 2 R41 | p 6= O, hp, pi1 = 0}.
We sometimes view H+ 3
( 1), S13 and LC ⇤ as sets of vectors and denote
their elements by u.
Since R41 is time-orientable ([O’Neill (1983)]), we choose e0 as the future
direction. We say that a timelike vector u is future directed (resp. past
directed) if hu, e0 i1 < 0 (resp. hu, e0 i1 > 0)
The Minkowski wedge product of three vectors ui = (ui0 , ui1 , ui2 , ui3 ),
i = 1, 2, 3, is defined as the vector
e0 e1 e2 e3
u10 u11 u12 u13
u1 ^ u2 ^ u3 = .
u20 u21 u22 u23
u30 u31 u32 u33
Clearly, hu, u1 ^ u2 ^ u3 i1 = det(u, u1 , u2 , u3 ). Therefore, u1 ^ u2 ^ u3
is pseudo orthogonal to the three vectors ui , i = 1, 2, 3.

9.1.1 The hyperbolic space and the Poincaré ball model

3
The hyperbolic space H+ ( 1) with the induced metric provides the model
for the Gauss-Bolyai-Lobachevski non-Euclidean geometry. Another model
for the hyperbolic space is the Poincaré ball model which is constructed as
follows. Let D = {(x1 , x2 , x3 ) 2 R3 | x21 + x22 + x23 < 1 } be the open ball
in R3 . We can consider R3 as a subset of R41 by identifying it with the
hyperplane x0 = 0. The Poincaré ball model of the hyperbolic space is the
set D endowed with the hyperbolic metric
4(dx21 + dx22 + dx23 )
ds2 = .
1 x21 x22 x23
3
Let x = (x0 , x1 , x2 , x3 ) be a point in H+ ( 1) and consider the line
joining x and the fixed point ( 1, 0, 0, 0). This line intersects the unit disc
 Spacelike surfaces in the Minkowski space-time 285

D at a unique point whose coordinates are ( x0x+1 1

, x0x+1
2
, x0x+1
3
). The map
3
: H+ ( 1) ! D, given by
✓ ◆
x1 x2 x3
(x0 , x1 , x2 , x3 ) = , ,
x0 + 1 x0 + 1 x0 + 1
is an isometry. The hyperbolic metric ds2 on D is conformally equivalent
to the Euclidean space.
3
The hyperbolic space H+ ( 1) can be considered as a homogeneous
space. Let SO0 (1, 3) be the identity component of the matrix group
SO(1, 3) = {g 2 GL(4, R) | g · I1,3 · g t = I1,3 },
where
0 1
10 0 0
B 0 1 0 0C
I1,3 =B
@ 0 0
C 2 GL(4, R).
1 0A
0 0 0 1
3
The group SO0 (1, 3) acts transitively on H+ ( 1) and the isotropic group
at p = (1, 0, 0, 0) is SO(3), which is naturally embedded in SO0 (1, 3). More-
3
over, the action induces isometries on H+ ( 1). Therefore, we have an
3
isometry between SO0 (1, 3)/SO(3) and H+ ( 1). By the uniqueness of the
model of the hyperbolic space, any model of the hyperbolic space is iso-
3
metric to the homogeneous space SO0 (1, 3)/SO(3). Therefore, H+ ( 1) is
isometric to the Poincaré disc model of the hyperbolic 3-space.

9.2 The lightcone Gauss maps

Consider the orientation of R41 given by the volume form e⇤0 ^ e⇤1 ^ e⇤2 ^ e⇤3 ,
where {e⇤0 , e⇤1 , e⇤2 , e⇤3 } is the dual basis of the canonical basis of R41 . The
Minkowski space R41 has also a time-orientation given by the choice of e0
as a future timelike vector field.
As in the previous chapters, we are interested in the local geometric
properties of smooth surfaces in R41 . For this reason we consider a surface
patch M which is the image of an embedding x : U ! R41 , where U is an
open set of R2 (so M = x(U )).
We say that M is a spacelike surface if its tangent plane Tp M is spacelike
at all points p 2 M . Then the pseudo-normal space Np M of M at p is a
timelike plane. We denote by N (M ) the pseudo-normal bundle over M.
Since this is a trivial bundle, there are two well defined transverse lightlike
286 Di↵erential Geometry from a Singularity Theory Viewpoint

smooth direction fields on M . These direction fields can be defined by two

smooth vector fields Li , i = 1, 2 on M . Of course, the two vector fields are
not unique.

Lemma 9.1. For any given smooth lightlike vector field L on M , there is a
unique unit timelike normal vector field u on M and a unique unit spacelike
normal vector field v on M such that L = u + v. Observe that u and v are
necessarily pseudo-orthogonal on M .

Proof. Since the normal bundle N (M ) is trivial we can work on Np M

at any given point p on M . Choose a local coordinate system in Np M ,
identify Np M with R21 and represent L(p) by a vector w = (a, a), with
a 6= 0. We are seeking two vectors u = (u0 , u1 ) and v = (v0 , v1 ) with the
following properties
u20 + u21 = 1,
v02 + v12 = 1,
u0 + v 0 = a,
u1 + v 1 = a.
Straightforward calculations show that
a2 + 1 a2 1 a2 1 a2 + 1
u=( , ), v = ( , ).
2a 2a 2a 2a ⇤

Remark 9.1. The result in Lemma 9.1 shows that any normal lightlike
vector field on M can be obtained in the following way. Suppose that M is
parametrised by x : U ! R41 and choose a unit timelike normal vector field
nT on M (this can always be done as N (M ) is a trivial bundle). Then
nT (u) ^ xu1 (u) ^ xu2 (u)
nS (u) = (9.1)
knT (u) ^ xu1 (u) ^ xu2 (u)k1
is a unit spacelike normal vector field which is pseudo orthogonal to nT (u).
The vector fields nT + nS and nT nS are lightlike normal vector fields
on M and determine at each point p 2 M the two lightlike directions in
Np M . Any smooth lightlike vector field L on M coincides with nT + nS or
nT nS for some chosen nT (take nT to be u in Lemma 9.1 so v in Lemma
9.1 coincides with nS or nS ).
Since M is spacelike, e0 is a future directed timelike vector field along
M which is transverse to M at all points. We can write any v 2 Tp R41 |M in
the form v = v1 +v2 , where v1 2 Tp M and v2 2 Np M. If v is timelike, then
v2 is timelike. Let ⇡Np M : Tp R41 |M ! Np M be the canonical projection,
 Spacelike surfaces in the Minkowski space-time 287

where p is any point on M . Then ⇡Np M (e0 ) is a future directed timelike

normal vector of M at the point p. Therefore, we can always define a future
directed unit timelike normal vector field along M , i.e., globally. It follows
that nT can be defined on a closed spacelike surface in R41 (i.e., in a global
sense) even when the surface is not orientable. However, nS depends on the
orientation of the surface M and can be defined only for orientable surfaces
in R41 .

Remark 9.2. There is nothing special about choosing a future directed

normal frame. We could choose nT to be a past unit normal vector field
and construct a past directed normal frame (nT , nS ) along M . The results
in this chapter are the same for past directed normal frames. (We choose
a future directed frame because we like the future!)

Definition 9.1. Let nT be a future unit normal timelike vector field on

M and let nS be its associated unit normal spacelike vector field on M
as in (9.1). We call (nT , nS ) a future directed normal frame along M and
set L+ = nT + nS and L = nT nS the two lightlike normal vector
fields defined by nT . (Of course L+ and L are uniquely determined by nT
and depend on the choice of nT .) The vector fields L± determine the two
lightlike normal direction fields on M . In rest of the chapter, L indicates
one of the lightlike vector fields L+ or L .

Remark 9.3. In Chapter 7 and for a surface M in the Euclidean space

R4 , we associated to a unit normal vector field ⌫ on M a ⌫-shape operator.
For a spacelike surface in R41 , and following Definition 9.1, we have two
distinguished normal direction fields on M parallel to L = L± . The vector
fields L are lightlike, so have zero length at each point. However, we can
still define an L-shape operator, and the rest of the chapter is devoted to
deriving geometric properties of M from L.

The vector field L is a map M ! LC ⇤ which we call the lightcone Gauss

map. We consider its derivative map at a point p 2 M which is linear map
dLp : Tp M ! Tp R41 = Tp M Np M.
We have the canonical projections
⇡1 : T p M N p M ! Tp M and ⇡2 : T p M Np M ! Np M.
We define the L-shape operator of M at p 2 M , and denote it by W (L)p ,
as the linear map Tp M ! Tp M , given by
W (L)p = ⇡1 dLp .
288 Di↵erential Geometry from a Singularity Theory Viewpoint

We also consider the shape operators W (nT )p = ⇡1 d(nT )p and

W (nS )p = ⇡1 d(nS )p which we call, respectively, the nT -shape operator
and the nS -shape operator of M at p. We have, by definition

W (L± )p = W (nT )p ± W (nS )p .

We call the eigenvalues of W (L)p the L-lightcone principal curvatures

and denote them by 1 and 2 . (We shall use the notation ± ±
1 and 2 when
needed.)
A point p on M is said to be an L-umbilic point if 1 (p) = 2 (p),
equivalently, W (L)p is a multiple of the identity map 1Tp M .
We also define the nT -principal curvature i (nT ), i = 1, 2, at p (resp.
nS -principal curvature i (nS ), i = 1, 2) as the eigenvalues of W (nT )p (resp.
W (nS )p ).

Proposition 9.1. With notation as above and for i = 1, 2,

± T S
i = i (n ) ± i (n ).

Proof. This follows from the fact that W (L± )p = W (nT )p ± W (nS )p .⇤

Definition 9.2. The L-lightcone Gauss-Kronecker curvature of M at p is

defined to be

K` (p) = det(W (L)p ).

The L-lightcone mean curvature of M at p is defined to be

1
H` (p) = Trace(W (L)p ).
2
A point p on M is said to be an L-parabolic point if K` (p) = 0.

Since M is a spacelike surface, we have a Riemannian metric (the first

fundamental form) on M defined by

ds2 = Edu21 + 2F du1 du2 + Gdu22 = g11 du21 + 2g12 du1 du2 + g22 du22 ,

where
E = g11 = hxu1 , xu1 i1 ,
F = g12 = g21 = hxu1 , xu2 i1 ,
G = g22 = hxu2 , xu2 i1 .
The second fundamental form associated to the normal vector field L,
which we refer to as the lightcone second fundamental form, is given by
 Spacelike surfaces in the Minkowski space-time 289

IIp (u) = hW (L)p (u), ui, with u 2 Tp M . The coefficients of the lightcone
second fundamental form are given by
l = h11 = hLu1 , xu1 i1 = hL, xu1 u1 i1 ,
m = h12 = h21 = hLu1 , xu2 i1 = hLu2 , xu1 i1 = hL, xu1 u2 i1 ,
n = h22 = hLu2 , xu2 i1 = hL, xu2 u2 i1 .

We shall use the notation l± , m± , n± (or, h± ± ±

11 , h12 , h22 ) when necessary.
We denote by
✓ ◆
a11 a12
Ap =
a21 a22
the matrix of the shape operator W (L)p with respect to the basis {xu1 , xu2 }
of Tp M at p = x(u).

Proposition 9.2 (The lightcone Weingarten formula). We have the

following formulae:
L u1 = (a11 xu1 + a21 xu2 ) ± hnS , nTu1 i1 L,
L u2 = (a12 xu1 + a22 xu2 ) ± hnS , nTu2 i1 L
and the following expression for the matrix of the shape operator W (L)p
✓ ◆✓ ◆ ✓ 11 12 ◆ ✓ ◆
1 G F l m g g h11 h12
Ap = = , (9.2)
EG F 2 F E m n g 12 g 22 h12 h22
where
✓ ◆ ✓ ◆ 1 ✓ ◆
g 11 g 12 g11 g12 1 g22 g12
= = .
g 12 g 22 g12 g22 g11 g22 2
g12 g12 g11

Proof. The vectors xu1 , xu2 , nT , nS form a basis of R41 , so there exist real
numbers , µ such that
L u1 = (a11 xu1 + a21 xu2 ) + nT + µnS .
Since L = nT ± nS , we have hLu1 , nT i1 = . On the other hand
Lu1 = nu1 ±nu1 , so hLu1 , n i1 = hnu1 ±nu1 , n i1 . But hnT , nT i1 = 1 im-
T S T T S T

plies hnTu1 , nT i1 = 0, and hnT , nS i1 = 0 implies hnT , nSu1 i1 = hnTu1 , nS i1 .

Therefore,
= hnTu1 ± nSu1 , nT i1 = ±hnTu1 , nS i1 .
Similarly,
µ = hLu1 , nS i1 = hnTu1 ± nSu1 , nS i1 = hnTu1 , nS i1 .
290 Di↵erential Geometry from a Singularity Theory Viewpoint

It follows that

L u1 = (a11 xu1 + a21 xu2 ) ± hnTu1 , nS i1 nT + hnTu1 , nS i1 nS

= (a11 xu1 + a21 xu2 ) ± hnTu1 , nS i1 (nT ± nS )
= (a11 xu1 + a21 xu2 ) ± hnTu1 , nS i1 L.

The formula for Lu2 follows similarly. It follows now from the expres-
sions of Lu1 and Lu2 that

h11 = hLu1 , xu1 i1 = a11 g11 + a21 g12 ,

h12 = hLu1 , xu2 i1 = a11 g12 + a21 g22 ,
h12 = hLu2 , xu1 i1 = a12 g11 + a22 g12 ,
h22 = hLu2 , xu2 i1 = a12 g12 + a22 g22 .

Writting the above equations in matrix form

✓ ◆ ✓ ◆✓ ◆
h11 h12 a11 a21 g11 g12
=
h12 h22 a12 a22 g12 g22

gives
✓ ◆ ✓ ◆✓ ◆ 1
a11 a21 h11 h12 g11 g12
Atp = =
a12 a22 h12 h22 g12 g22

so that
✓ ◆ 1 ✓ ◆ ✓ ◆✓ ◆
g11 g12 h11 h12 1 g22 g12 h11 h12
Ap = = 2 .
g12 g22 h12 h22 g11 g22 g12 g12 g11 h12 h22
⇤

Corollary 9.1. It follows from the Weingarten formulae that:

(i) The shape operators W (L)p , W (nT )p and W (nS )p are self-adjoint op-
erators on Tp M .
(ii) The L-lightcone Gauss-Kronecker curvature is given by

ln m2 h11 h22 h212

K` = = 2 .
EG F 2 g11 g22 g12

(iii) The L-lightcone mean curvature is given by

lG 2mF + nE h11 g22 2h12 g12 + h22 g11

H` = 2
= 2 ) .
2(EG F ) 2(g11 g22 g12
 Spacelike surfaces in the Minkowski space-time 291

9.3 The normalised lightcone Gauss map

Define the lightcone unit 2-sphere as the set

2
S+ = {u = (u0 , u1 , u2 , u3 ) 2 LC ⇤ | u0 = 1}
= {u = (1, u1 , u2 , u3 ) 2 LC ⇤ | u21 + u22 + u23 = 1}.
If u is a lightlike vector, then u0 6= 0. Denote
✓ ◆
u1 u2 u3
e = 1, , ,
u
u0 u0 u0
2
the unique vector in S+ parallel to u.
We define the projection ⇡SL : LC ⇤ ! S+ 2
by ⇡SL (u) = u
e.
Given a spacelike surface M in R1 , the normal lightlike vector field
4

L depends on the choice of the unit timelike normal vector field nT (see
Definition 9.1). However, the normal lightlike vector field L e does not depend
on the choice of nT . Indeed, given any two unit timelike normal vector fields
nT1 and nT2 , we have ⇡SL (nT1 ± nS1 ) = ⇡SL (nT2 ± nS2 ). We can thus introduce
the following map which is independent of the choice of nT .

Definition 9.3.
Let M be a spacelike surface patch parametrised by x : U ! R41 . The
normalised lightcone Gauss map of M is the map Le : U ! S 2 given by
+
e
L(u) = ⇡SL (L(u)).

We define the normalised lightcone shape operator of M at p as the

linear map W (L) e p = ⇡1 dL e p from the tangent plane Tp M to itself.
The eigenvalues  e1 (p) and e2 (p) of W (L)e p are called the normalised
lightcone principal curvatures.
The normalised lightcone Gauss-Kronecker curvature and normalised
lightcone mean curvature of M are defined, respectively, to be

K e p ) and H
e ` (p) = det(W (L) e ` (p) = 1 Trace(W (L)e p ).
2
A point p is said to be a lightcone parabolic point if K e ` (p) = 0.
Let Aep denote the matrix of the normalised lightcone shape operator
with respect to the basis {xu1 , xu2 } of Tp M at p = x(u). We write
L(u) = (`0 (u), `1 (u), `2 (u), `3 (u)).

Proposition 9.3. We have, at p = x(u),

Aep = 1 Ap .
`0 (u)
292 Di↵erential Geometry from a Singularity Theory Viewpoint

Proof. We have, by definition, `0 L e = L, so `0 L

e u = Lu e for
(`0 )ui L
i i
e
i = 1, 2. Since L(u) e u = 1 Lu , i = 1, 2,.
2 Np M , we have ⇡1 L ⇤
i `0 i

Corollary 9.2. It follows from Proposition 9.3 that, for i = 1, 2,

✓ ◆2
1 e` = 1 e ` = 1 H` .
e i = i , K
 K` , H
`0 `0 `0
The following also follows from Proposition 9.3.

Proposition 9.4. For any future directed normal frame (nT , nS ) defining
L, Ke ` (p) = 0 if and only if K` (p) = 0, and H
e ` (p) = 0 if and only if
H` (p) = 0.

Remark 9.4. Observe that the eigenvectors of W (L)p coincide with those
e p . Therefore, the L-principal directions and L-umbilic points are
of W (L)
e
concepts that depend only on L.

9.4 Marginally trapped surfaces

We define the mean curvature vector of M at p as the vector

H(p) = Trace(W (nT )p )nT (p) + Trace(W (nS )p )nS (p).
The mean curvature vector is independent of the choice of (nT , nS ).
The concept of trapped surfaces in space-time, introduced by Penrose in
[Penrose (1965)], plays an important role in cosmology and general relativ-
ity. (In particular, it plays a principal role in the proofs of the theorems of
space-time singularities, gravitational collapse, the cosmic censorship and
Penrose inequality.)
A spacelike surface in a space-time is said to be marginally trapped if
its mean curvature vector is lightlike at each point.

Proposition 9.5. The following assertions are equivalent:

(i) H`+ (p) = 0 or H` (p) = 0.
(ii) The mean curvature vector H(p) is isotropic (i.e., it is lightlike or zero).
(iii) e + (p) = 0 or H
H e (p) = 0.
` `

Proof. We have W (L)p = W (nT )p ± W (nS )p , so

Trace(W (L)p ) = Trace(W (nT )p ) ± Trace(W (nS )p ).
On the other hand,
hH(p), H(p)i1 = Trace(W (nT )p )2 + Trace(W (nS )p )2 .
 Spacelike surfaces in the Minkowski space-time 293

Thus, hH(p), H(p)i1 = 0 if and only if Trace(W (nT )p ) = ±Trace(W (nS )p ),

e ± (p) = (1/`0 (p)) = H ± (p),
that is, if and only if condition (i) holds. Since H ` `
the conditions (i) and (iii) are equivalent. ⇤
It follows from Proposition 9.5 that a spacelike surface M in R41 is
marginally trapped if and only if H`+ ⌘ 0 or H` ⌘ 0 for any future directed
normal frame (nT , nS ). Surfaces satisfying H`+ ⌘ 0 and H` ⌘ 0 are called
minimal surfaces. However, this class of spacelike surfaces contains the class
of maximal spacelike surfaces in R41 ([Cheng and Yau (1973); Kobayashi
(1983)]), where a spacelike surface in R31 is maximal if its mean curvature
vanishes identically. For this reason, we say that M is strongly marginally
trapped if H`+ ⌘ 0 and H` ⌘ 0.

9.5 The family of lightcone height functions

The family of lightcone height functions H : R41 ⇥ S+

2
! R is defined by
H(p, v) = hp, vi1 .
The family of lightcone height functions H : U ⇥ S+2
! R on M
parametrised by x : U ! R1 is the restriction of the family H to M .
4

Thus,
H(u, v) = hx(u), vi1 .
For v fixed, we write hv (u) = H(u, v) for the lightcone height function
on M along v.
We have the following characterisation of the lightlike parabolic points
and lightlike flat points in terms of the lightcone height function.
2
Proposition 9.6. Let p0 = x(u0 ) be a point on M and v0 a vector in S+ .
e 0 ).
(i) The function hv0 is singular at u0 if and only if v0 = L(u
e 0 ). Then,
(ii) Suppose that v0 = L(u
(a) The point p0 is a lightcone parabolic point if and only if
det Hess(hv0 )(u0 ) = 0.
(b) The point p0 is a lightcone flat point if and only if
rank Hess(hv0 )(u0 ) = 0.

Proof. (i) Since {xu1 (u), xu2 (u), nT (u), nS (u)} is a basis of Tp R41 for all
p = x(u) on M , there exist real numbers ⇠1 , ⇠2 , 1 , 2 such that
v0 = 1 xu1 (u0 ) + 2 xu2 (u0 ) + µ1 nT (u0 ) + µ2 nS (u0 ).
294 Di↵erential Geometry from a Singularity Theory Viewpoint

We have
@hv0
h , v0 i1 = hxu1 , v0 i1 = 1E + 2 F,
@u1
@hv0
h , v0 i1 = hxu2 , v0 i1 = 1F + 2 G.
@u2
Therefore @hv0 /@u1 (u0 ) = @hv0 /@u2 (u0 ) = 0 if and only if 1 = 2 = 0
(EG F 2 > 0 as M is spacelike). Thus, when hv0 is singular at u0 , we
have v0 = µ1 nT (u0 ) + µ2 nS (u0 ). Because v0 lightlike, µ21 + µ22 = 0. This
means that v0 = µ1 L(u0 ), and as v0 is in S+ 2 e 0 ).
, we have v0 = L(u
(ii)-(a) We have
!
e e
hxu1 u1 (u0 ), L (u0 )i1 hxu1 u2 (u0 ), L (u0 )i1
± ±
Hess(hv0 )(u0 ) = e ± (u0 )i1 hxu u (u0 ), L
e ± (u0 )i1
hxu1 u2 (u0 ), L
✓ ◆ 2 2
1 h11 (u0 ) h12 (u0 )
= `0 (u ) .
0 h12 (u0 ) h22 (u0 )
It follows from Propositions 9.2 and 9.3 that
✓ ◆
ep (u0 ) g 11 (u 0 ) g 12 (u 0 )
Hess(hv0 )(u0 ) = A ,
g12 (u0 ) g22 (u0 )
so
ep (u0 )) = det(Hess(hv0 )(u0 )) .
e ` (u0 ) = det(A
K
g11 g22 g122

(ii)-(b) The point p is a lightcone umbilic point if and only

ep = I,
A
where I is the identity matrix and  one of the equal normalised lightcone
principal curvatures at p. Equivalently,
✓ ◆
g11 g12
Hess(hv0 ) =  .
g12 g22
The point p is a lightcone flat point if and only if  = 0, if and only if
rank Hess(hv0 )(u0 ) = 0. ⇤

Corollary 9.3. The following assertions are equivalent at any p = x(u) on

M.
(i) The point p is a L-parabolic point for any future directed normal frame
(nT , nS ).
(ii) The point p is a singular point of the lightcone Gauss map L.e
(iii) Ke ` (p) = 0.
e
(iv) det Hess(hv )(u) = 0 for v = L(u).
 Spacelike surfaces in the Minkowski space-time 295

Corollary 9.4. The following assertions are equivalent at any p = x(u) on

M.
(i) The point p is a L-flat point for any future directed normal frame
(nT , nS ).
(ii) There exists a future directed normal frame (nT , nS ) such that p is a
L-flat point.
e
(iii) The point p is a lightcone flat point with respect to L.

We have the following characterisation of lightcone flatness of spacelike

surfaces.

Proposition 9.7. The following assertions are equivalent.

(i) M is totally L-flat.
(ii) The normalised lightcone Gauss map L e is a constant map.
(iii) There exists a lightlike vector v and a real number c such that M is
contained in the lightlike hyperplane HP (v, c).

Proof. Suppose that M is totally L-flat. This means that the matrix Ap
in Proposition 9.2 is identically zero for any future directed frame (nT , nS ).
Therefore, by Proposition 9.2, have
Lui (u) = ±hnS (u), nTui (u)i1 L(u)
at all points u 2 U and for i = 1, 2. Then

e u (u) = ( (`0 )ui (u) ± hnS (u), nT (u)i1 )L(u), i = 1, 2,

L ui
i
`0 (u)2
and this implies that L e u (u) is the zero vector for i = 1, 2, as being a
i

spacelike vector it cannot be a non-zero multiple of a lightlike vector. It

follows that Le is a constant map, so (i) implies (ii). For the converse if L
e
is constant then A ep is identically zero and so is A
ep by Proposition 9.3, so
all points are L-flat umbilic points by Proposition 9.6.
Suppose now that the normalised lightcone Gauss map L(u) e = v is
constant. We consider a function F : U ! R defined by F (u) = hx(u), vi1 .
By definition, we have
@F e
(u) = hxui (u), vi1 = hxui (u), L(u)i 1 = 0,
@ui
for i = 1, 2. Therefore, F (u) = hx(u), vi1 = c is constant. Since v is
lightlike, M is a subset of the lightlike hyperplane HP (v, c), (ii) implies
(iii).
296 Di↵erential Geometry from a Singularity Theory Viewpoint

Suppose that M is a subset of a lightlike hyperplane H(v, c). For any

point p 2 M, the tangent space of HP (v, c) can be identified with HP (v, 0).
Since M ⇢ HP (v, c), Tp M ⇢ HP (v, 0), so that Np (M ) \ HP (v, 0) is the
line generated by v. For any future directed timelike unit normal vector
field nT along M, there exists a lightlike vector v such that v is parallel
to v and v nT is a spacelike unit normal vector field along M. We write
nS = v nT , so that we have a future directed normal frame (nT , nS )
along M with
e
L(u) = ⇡SL (nT (u) + nS (u)) = ⇡SL (v).
e is constant. This
This means that the corresponding lightcone Gauss map L
completes the proof. ⇤

Corollary 9.5. The following assertions are equivalent.

(i) M is totally L+ -flat and L -flat.
(ii) M is a part of a spacelike plane.

Proof. By Proposition 9.7, condition (i) is equivalent to the condi-

tion that there exist two linearly independent lightlike vectors v± and
real numbers c± such that M ⇢ HP (v+ , c+ ) \ HP (v , c ). (The set
HP (v+ , c+ ) \ HP (v , c ) is a spacelike plane.) ⇤

9.6 The Lagrangian viewpoint

The space R30 = {(x0 , x1 , x2 , x3 ) 2 R41 | x0 = 0} can be identified with the

Euclidean 3-space. We consider the canonical projection ⇡ : R41 ! R30
defined by
⇡(x0 , x1 , x2 , x3 ) = (0, x1 , x2 , x3 ).

Lemma 9.2. Let M be a spacelike surface patch parametrised by x : U !

e is transversal to ⇡(M ) in
R41 . Then the direction of the vector field ⇡ L
R30 , that is,
e
R.{⇡ L(u)} + d(⇡ x)u (Tu U ) = Tp R30
at any p = x(u) 2 M .
e
Proof. Since L(u) is lightlike and Ker(d⇡p ) is a timelike one-dimensional
4 e
subspace of R1 , L(u) 2/ Ker(d⇡p ). The fact that dxu (Tu U ) is spacelike
implies that
e
R.{L(u), Ker(d⇡p )} + dxu (Tu U ) = Tp R41 (9.3)
 Spacelike surfaces in the Minkowski space-time 297

at any point p = x(u) 2 M.

Suppose that there exists a point u 2 U such that the direction of the
vector field ⇡ Le ± (u) is not transversal to ⇡ x(U ) in R3 at p = x(u).
0
Since ⇡ x(U ) is a smooth surface in R30 , our assumption implies that
e
⇡ L(u) 2 d(⇡ x)u (Tu U ). This means that
e
L(u) 2 dxu (Tu U ) + Ker(d⇡p ).
But this implies in turn that the dimension of R.{L(u), e Ker(d⇡p )} +
dxu (Tu U ) is at most 3, which contradicts equality (9.3). ⇤

Proposition 9.8. Let M be a spacelike surface patch in R41 . Then family

of lightcone height functions H is a Morse family of functions.
2 2
Proof. Let v = (1, v1 , v2 , v3 ) be in S+ and consider the chart of S+
with
p v1 > 0 (the result follows similarly for the other charts). Then v1 =
1 v22 v32 and
q
H(u, v) = x0 (u) + x1 (u) 1 v22 v32 + x2 (u)v2 + x3 (u)v3 ,
where x(u) = (x0 (u), x1 (u), x2 (u), x3 (u)). We show that the mapping
@H @H
H= , = (hxu1 , vi1 , hxu2 , vi1 )
@u1 @u2
is not singular at any point in CH = ( H) 1 (0). The Jacobian matrix of
H is given by
0 v2 v3 1
hxu1 u1 , vi1 hxu1 u2 , vi1 x1u1 + x2u1 x1u1 + x3u1
B v1 v1 C
@ v2 v3 A.
hxu2 u1 , vi1 hxu2 u2 , vi1 x1u2 + x2u2 x1u2 + x3u2
v1 v1
Consider the 2 ⇥ 2-matrix A given by the last two columns of the Jaco-
bian matrix of H, that is,
0 v2 v3 1
x1u1 + x2u1 x1u1 + x3u1
B v1 v1 C
A=@ v2 v3 A,
x1u2 + x2u2 x1u2 + x3u2
v1 v1
✓ ◆
xiu1
and let ai = for i = 1, 2, 3. Then,
xiu2
v2 v3
A= a1 + a2 , a1 + a3
v1 v1
and
v2 v3
detA = det(a2 , a3 ) det(a1 , a3 ) + det(a1 , a2 ).
v1 v1
298 Di↵erential Geometry from a Singularity Theory Viewpoint

Since x is a spacelike embedding, x = ⇡ x : U ! R30 is an immersion, so

that xu1 ⇥ xu2 6= 0. (Here “⇥” denotes the vector product in R30 , identified
with the Euclidean 3-space.) Then,
v1 v2 v3 1
detA = h( , , ), xu1 ⇥ xu2 i1 = hv, xu1 ⇥ xu2 i1 .
v1 v1 v1 v1
Since (u, v) 2 CH = ( H) 1 (0), by Proposition 9.6 we may take v =
e
L(u). e
By Lemma 9.2, the direction of ⇡(v) = ⇡ L(u) is transverse to the
tangent space of ⇡(M ) at ⇡ x(u). Therefore, detA 6= 0 and so H has
maximal rank. ⇤
We have shown in Chapter 5 how to construct a germ of a Lagrangian
immersion from a Morse family of functions. We use below that construc-
tion and the result in Proposition 9.8 to define a germ of a Lagrangian
immersion whose generating family is the family of lightcone height func-
tions H on M .
2 2
Consider the local chart U1 = {v = (1, v1 , v2 , v3 ) 2 S+ | v1 > 0 } of S+
(we proceed similarly for the other local charts). By Proposition 9.6, we
have
e
CH = {(u, L(u)) | u 2 U }.
Since T ⇤ S+2
|U1 is a trivial bundle, we can define the map L(H) : CH !
T ⇤ S+
2
|U1 by

e ± e ± `±
2 (u) `± (u)
L(H)(u, L (u)) = (L (u), x2 (u) ⌥ x1 (u) ± , x3 (u) ⌥ x1 (u) 3± ),
`1 (u) `1 (u)
where L(u) = L± (u) = (`± ± ± ± ⇤
0 (u), `1 (u), `2 (u), `3 (u)) 2 LC . The following
result follows from Proposition 9.8.

Corollary 9.6. The map-germ L(H) is a Lagrangian immersion whose

generating family is the family of the lightcone height functions H on M .

Consider the lightcone height function hv : R41 ! R, given by hv (p) =

e 0 ). By Propo-
H(p, v) = hp, vi1 . Clearly, hv is a submersion. Let v0 = L(u
sition 9.6, we have for i = 1, 2,
@hv0 x @H
(u0 ) = (u0 , v0 ) = 0.
@ui @ui
This means that the lightlike hyperplane hv01 (c) = HP (v0 , c) is tangent
to M at p0 = x(u0 ), where c = hx(u0 ), v0 i1 . We say that HP (v0 , c) is
a lightlike tangent hyperplane of M with pseudo-normal direction v0 and
denote it by T HP (x, u0 ).
 Spacelike surfaces in the Minkowski space-time 299

Let ✏ be a sufficiently small positive real number. For any t 2 (c ✏, c+✏),

we have a lightlike hyperplane HP (v0 , t) = hv01 (t). Clearly,
Fhv0 = {HP (v0 , t) | t 2 (c ✏, c + ✏)}
is a family of parallel lightlike hyperplanes around p0 .
Let M and M̄ be two germs spacelike surfaces parametrised by x :
(U, u) ! (R41 , p) and x̄ : (Ū , ū) ! (R41 , p̄), respectively. Set h1,v (u) =
hv x(u) and h2,v̄ (u) = hv̄ x̄(u), where v = L(u) e e
and v̄ = L(ū).
We have the following result.

Theorem 9.1. Let M and M̄ as above and suppose that h1,v and h2,v̄
satisfy the Milnor condition. Then the following assertions are equivalent.

(i) K(M, Fh1,v ; x(u)) = K(M̄ , Fh2,v̄ ; x̄(ū))

(ii) h1,v1 and h2,v2 are R+ -equivalent.
(iii) H1 and H2 are P -R+ - equivalent.
(iv) L(H1 ) and L(H2 ) are Lagrangian equivalent.
(v) The rank and signature of Hess(h1,v )(u) and Hess(h2,v̄ )(ū) are equal,
and there is an isomorphism : R2 (h1,v ) ! R2 (h2,v̄ ) such that
(h1,v ) = h2,v̄ .

Proof. The statements (i) and (ii) are equivalent by Proposition 4.1.
Since the germs L(H1 ) and L(H2 ) are Lagrangian stable, the germs H1
and H2 are R+ -versal unfoldings of h1,v and h2,v̄ , respectively. Then (ii)
is equivalent to (iii) by the uniqueness of R+ -versal unfoldings.
By Theorem 5.4, (iii) is equivalent to (iv) and it also follows from that
theorem that h1,v and h2,v̄ satisfy the Milnor condition. Then we can apply
Proposition 5.1 to show that (ii) is equivalent to (v). ⇤

Corollary 9.7. Let Embs (U, R41 ) be the set of spacelike embeddings U !
R41 endowed with the Whitney C 1 -topology. Then there exists a residual
subset O ⇢ Embs (U, R41 ) such that for any x 2 O, the following properties
hold.

(i) The lightcone parabolic set is a regular curve.

(ii) The lightcone parabolic curve consists of fold singularities of the nor-
malised lightcone Gauss map except possibly at isolated points where it
has a cusp singularity.

Proof. We remark that Embs (U, R41 ) is an open subset of C 1 (U, R41 ).
The proof is similar to those of Theorems 4.7 and 6.1. We use the lightcone
300 Di↵erential Geometry from a Singularity Theory Viewpoint

height function H : U ⇥ S+ 2
! R here instead of the height function in
Theorem 4.7. For a generic spacelike embedding x : U ! R41 , the R+ -
singularities of hv are type Ak , k = 1, 2, 3 and H is an R+ -versal unfolding
2
of hv at any point (u, v) 2 U ⇥ S+ (Theorems 3.13 and 4.4). In this case
the catastrophe map-germ ⇡CH is A-equivalent to the fold singularity or
the cusp singularity by the calculation of §3.9.2. Moreover, the singular set
of the catastrophe map is a regular curve. Since the catastrophe map of H
is identified with the normalised lightcone Gauss map L, e the singular set of
the catastrophe map is the lightcone parabolic curve. This completes the
proof. ⇤

Remark 9.5. Let O be the set given in Corollary 9.7 endowed with the
Whitney C 1 -topology. For any x 2 O, we can show that the corresponding
Lagrangian submanifold L(H) is stable under the perturbation of x, and
it follows from this that the set O is open. The full details of the proof of
this observation requires the use of more tools from singularity theory and
is omitted.

9.7 The lightcone pedal and the extended lightcone height

function: the Legendrian viewpoint

We shall associate to the surface M a mapping (or its image) M ! LC ⇤

whose singularities correspond to the those of the lightcone Gauss map of
M.

Definition 9.4. We define the lightcone pedal of M as the mapping LPM :

U ! LC ⇤ given by
e
LPM (u) = hx(u), L(u)i e
1 L(u).

We call the image of LPM the lightcone pedal surface (associated to M ).

We call the family of functions H̄ : R41 ⇥ LC ⇤ ! R given by

H̄(p, v) = h̄v (p) = hp, vi1 v0 ,
where v = (v0 , v1 , v2 , v3 ), the extended family of lightcone height functions.
Given a spacelike surface patch M parametrised by x : U ! R41 , we call H̄,
restriction of H̄ to M , the extended family of lightcone height functions on
M . The family H̄ : U ⇥ LC ⇤ ! R is given by
H̄(u, v) = h̄v (u) = hx(u), vi1 v0 .
 Spacelike surfaces in the Minkowski space-time 301

Since @ H̄/@ui = @H/@ui for i = 1, 2 and Hess(h̄v ) = Hess(hv ), we have

the following result as a corollary of Proposition 9.6.

Proposition 9.9. Let M be a spacelike surface patch parametrised by x :

U ! R41 and p0 = x(u0 ). Then,

@ H̄ @ H̄
(i) H̄(u0 , v) = (u0 , v) = (u0 , v) = 0 if and only if v =
@u1 @u2
e 0 )i1 L(u
hx(u0 ), L(u e 0 ).
(ii) Suppose that v0 = hx(u0 ), L(u e 0 )i1 L(u
e 0 ). Then

(a) p0 is a lightcone parabolic point if and only if det Hess(h̄v0 )(u0 ) = 0.

(b) p0 is a lightcone flat point if and only if rank Hess(h̄v0 )(u0 ) = 0.

Proposition 9.9(i) implies that the discriminant set of H̄ is the lightcone

pedal surface, that is, DH̄ = LPM (U ). Moreover, the set of singular points
of LPM is the lightcone parabolic set of M .

Proposition 9.10. The extended family of lightcone height functions H̄ is

a Morse family of hypersurfaces.

Proof. Consider the canonical di↵eomorphism 2

: S+ ⇥ (R \ {0}) ! LC ⇤
given by

(v, r) = rv.

e = H̄
and denote by H 2
: U ⇥ (S+ ⇥ (R \ {0})) ! R the composite map
given by

e
H(u, v, r) = hx(u), vi1 r.

We only need to show that H e is a Morse family of hypersurfaces.

2
Let (v, r) p= ((1, v1 , v2 , v3 ), r) 2 S+ ⇥ (R \ {0}), and assume that v1 > 0,
2 2
so that v1 = 1 v2 v3 (the proof follows similarly for the other charts
2
of S+ ). It follows that
q
e
H(u, v, r) = x0 (u) + x1 (u) 1 v22 v32 + x2 (u)v2 + x3 (u)v3 r,

where x(u) = (x0 (u), x1 (u), x2 , x3 (u)). We show that the mapping

e e e e
⇤ e @ H , @ H = H,
e = H,
H e @H , @H
@u1 @u2 @u1 @u2
302 Di↵erential Geometry from a Singularity Theory Viewpoint

is not singular at any point in ⌃He = ( ⇤ H)e 1 (0). The Jacobian matrix of
⇤ e
H is given by
0 v2 v3 1
hxu1 , vi1 hxu2 , vi1 1 x1 + x2 x1 + x3
B v1 v1 C
B v2 v3 C
B hx C
B u1 u1 , vi1 hxu1 u2 , vi1 0 x1u1 + x2u1 x1u1 + x3u1 C .
B v1 v1 C
@ v2 v3 A
hxu2 u1 , vi1 hxu2 u2 , vi1 0 x1u2 + x2u2 x1u2 + x3u2
v1 v1
Consider the matrix
0 v2 v3 1
x1u1
+ x2u1 x1u1 + x3u1
B v1 v1 C
A=@ v2 v3 A
x1u2 + x2u2 x1u2 + x3u3
v1 v1
which is the same matrix in the proof of Proposition 9.8, so detA 6= 0 at
e 1 (0). Therefore rank ⇤ H
any point of ⌃He = ( ⇤ H) e = 3 on ⌃ e . ⇤
H

We define now a Legendrian immersion whose generating family is the

family of extended lightcone height functions on M .
Consider the projective cotangent bundle ⇡ : P T ⇤ (LC ⇤ ) ! LC ⇤ with
the canonical contact structure (observe that P T ⇤ (LC ⇤ ) ⇠
= LC ⇤ ⇥ P (R4 )⇤
⇤
p bundle). Given any point v = (v0 , v1 , v2 , v3 ) 2 LC , we have
is a trivial
v0 = ± v12 + v22 + v32 , so we can adopt the coordinates (v1 , v2 , v3 ) for the
manifold LC ⇤ .
We define the map L (H̄) : ⌃H̄ ! P T ⇤ (LC ⇤ ) as follows. Given
(u, v) 2 ⌃H̄ , with v = (v0 , v1 , v2 , v3 ) = LPM (u) and x(u) =
(x0 (u), x1 (u), x2 (u), x3 (u)), we set
L (H̄)(u, v) = (v, r, z),
with
z = [v0 x1 (u) ⌥ v1 x0 (u) : v0 x2 (u) ⌥ v2 x0 (u) : v0 x3 (u) ⌥ v3 x0 (u)]
for L = L± . An immediate consequence of the definition of L (H̄) is the
following result.

Proposition 9.11. The map L (H̄) is a Legendrian immersion and the

extended family of lightcone height functions H̄ on M is a generating family
of L (H̄) at any point of the domain U of the parametrisation of M .

Observe that the corresponding wavefront of the Legendrian immersion

L (H̄) is the lightcone pedal surface LPM (U ). Therefore, the singularities
of the lightcone pedal are Legendrian singularities.
 Spacelike surfaces in the Minkowski space-time 303

Recall that T HP (x, u) denotes the tangent lightlike hyperplane to M

at x(u).
Lemma 9.3. Let M be a spacelike surface patch parametrised by x : U !
R41 . For any two points u, ū 2 U , LPM (u) = LPM (ū) if and only if
T HP (x, u) = T HP (x, ū).
Let LPM : (U, u) ! (LC ⇤ , v) and LPM̄ : (U, ū) ! (LC ⇤ , v̄) be two
germs of lightlike pedal mappings of spacelike surface patches parametrised
by x : (U, u) ! R41 and x̄ : (Ū , ū) ! R41 , respectively. If the regular sets of
LPM and LPM̄ are dense in their domain, then by Theorem 5.10, LPM and
LPM̄ are A-equivalent if and only if their corresponding germs of Legendrian
immersions LH̄1 : (U, u) ! P T ⇤ (LC ⇤ ) and LH̄2 : (Ū , ū) ! P T ⇤ (LC ⇤ ) are
Legendrian equivalent. Equivalently, by Theorem 5.11, the two generating
families H̄1 and H̄2 are P -K-equivalent.
On the other hand, if we denote h̄1,v (u) = H̄1 (u, v), and
h̄2,v̄ (u) = H̄2 (u, v̄), then by Theorem 4.1, K(M, T HP (x, u), v) =
K(M̄ , T HP (x̄, ū), v̄) if and only if h̄1,v and h̄2,v̄ are K-equivalent.
Let Qr (x, u) (resp. Qr (x̄, ū)) denote the local ring of the function germ
h̄1,v (resp. h̄2,v̄ ) with v = LPM (u) (resp. v̄ = LPM̄ (ū)) (see Chapter 5 for
the explicit expression of Qr (x, u)).
Theorem 9.2. Let x : (U, u) ! R41 and x̄ : (U, ū) ! R41 be parametri-
sations of spacelike surface patches such that the germs of the Legendrian
germs ⇡ L (H̄1 ) : (U, u) ! (LC ⇤ , v) and ⇡ L (H̄2 ) : (U, ū) ! (LC ⇤ , v̄)
are Legendrian stable. Then the following statements are equivalent.

(i) The lightcone pedal germs LPM and LPM̄ are A-equivalent.
(ii) H̄1 and H̄2 are P -K-equivalent.
(iii) h̄1,v and h̄2,v̄ are K-equivalent.
(iv) K(M, T HP (x, u), v) = K(M̄ , T HP (x̄, ū), v̄).
(v) Q4 (x, u) and Q4 (x̄, ū) are isomorphic as R-algebras.

Proof. Assertion (iii) is equivalent to (iv) by the arguments in the proof

of Lemma 9.3. The equivalence between the other assertions follows from
Theorem 5.11 and Proposition 5.2. ⇤
Corollary 9.8. Let M be a surface patch parametrised by x 2 O with O
as in Corollary 9.7. Then the following hold.

(i) The lightcone pedal surface is a cuspidal edge at each point of the light-
like parabolic curve except possibly at isolated points. At such points it
304 Di↵erential Geometry from a Singularity Theory Viewpoint

is a swallowtail surface.
(ii) A lightcone parabolic point is a fold of the normalised lightcone Gauss
map L e if and only if it is a cuspidal edge point of the lightcone pedal
map LPM .
(iii) A lightcone parabolic point is a cusp of the normalised lightcone Gauss
map L e if and only if it is a swallowtail point of the lightcone pedal map
LPM .

Proof. Corollary 9.7, the family of lightcone height function germ H at

2
any point (u, v) 2 U ⇥ S+ is the R+ -versal unfolding of hv . Therefore, it
is P -R+ -equivalent to one of the following germs:

(A1 ) F (q, x1 , x2 ) = ±q 2 ,
(A2 ) F (q, x1 , x2 ) = q 3 + x1 q,
(A3 ) F (q, x1 , x2 ) = ±q 4 + x1 q + x2 q 2 .

For each F, we have F (q, x1 , x2 , y) = F (q, x1 , x2 ) y. On the other

hand, H is the family of extended lightcone height functions such that
L (H)(⌃⇤ (H) = LH (CH ) is a germ of a graph-like Legendrian submanifold
at any point and its corresponding Lagrangian map-germ is ⇡ e LH |CH = L. e
Thus, if H is P -R+ -equivalent to F , then H is P -K-equivalent to F . If H is
e LH |CH is Lagrangian non-singular. If H is of A2 -type,
of A1 -type, then ⇡
then ⇡e LH |CH is fold map-germ. In this case, the discriminant DF is a
cuspidal edge. Moreover, the discriminant DH is the image of the lightcone
pedal map LPM . This means that we have assertion (ii). If H is of A3 -type,
then we have assertion (iii). The set of singular points of Le is the lightcone
parabolic curve, which is equal to the set of singular points of LPM . This
completes the proof. ⇤

9.8 Special cases of spacelike surfaces

The Euclidean space R30 and any spacelike hyperplanes in R41 are Rieman-
nian manifolds so any surface embedded in these spaces is a spacelike sur-
face. We illustrate below some properties concerning the lightcone Gauss
map of surfaces in these spaces. We also deal with the case of spacelike
surfaces contained in de Sitter 3-space and treat with more details the case
of surfaces in the hyperbolic 3-space.
 Spacelike surfaces in the Minkowski space-time 305

9.8.1 Surfaces in Euclidean 3-space

Let M be a surface patch contained in Euclidean space R30 ⇢ R41 and
suppose that it is parametrised by x : U ! R30 . Let S 2 denote the unit
sphere in R30 .
Here, we take nT = e0 , so that
e0 ^ xu1 (u) ^ xu2 (u)
nS (u) = 2 S2
ke0 ^ xu1 (u) ^ xu2 (u)k1
is the Euclidean unit normal of M ⇢ R30 at p = x(u).
In this case the lightcone height functions measure the contact of M
with planes in R30 , and hence coincides with the usual family of height
functions in Euclidean 3-space.
The lightcone Gauss map, given by
L± (u) = e0 ± nS (u),
coincides with the classical Gauss map on surfaces in Euclidean 3-space.
The normalised lightcone shape operator is given by
W (Le ± )p = dL± = ⌥dnS ,
p p
and coincides with the Weingarten map of M considered as a surface in
Euclidean space R30 . It follows that Ke ± (u) = ±K(u) and H̄ ± (u) = ±H(u),
` `
where K and H are, respectively, the Gauss curvature and mean curvature
of M as a surface in Euclidean space R30 . Consequently, in R30 , the lightcone
flat spacelike surfaces are developable surfaces and the marginally trapped
surfaces are minimal surfaces.
In general, when nT (u) = v is a constant timelike unit vector, the
spacelike surface M is contained in a spacelike hyperplane HP (v, c). Since
HP (v, c) is isometric to R30 , the results for surfaces in R30 hold for surfaces
in HP (v, c).

9.8.2 Spacelike surfaces in de Sitter 3-space

Let M be a spacelike surface in de Sitter 3-space parametrised by x : U !
S13 . Then x(u) is a unit spacelike vector pseudo-orthogonal to M at x(u),
so we can take nS (u) = x(u) for all u 2 U and we get
xu1 (u) ^ xu2 (u) ^ x(u)
nT (u) = 2 H 3 ( 1).
kxu1 (u) ^ xu2 (u) ^ x(u)k1
Given a lightlike vector v, the set HP (v, c) \ S13 is called a de Sitter
horosphere. Thus, the lighcone height functions measure the contact of M
with de Sitter horospheres.
306 Di↵erential Geometry from a Singularity Theory Viewpoint

The lightcone Gauss map is given by

L± (u) = x(u) ± nT (u).

The following result follows from Proposition 9.7.

Proposition 9.12. Let M be a spacelike surface patch in de Sitter 3-space.

Then one of the lightcone Gauss maps (L+ or L ) is a constant vector if
and only if M is part of a de Sitter horosphere.

We refer to [Kasedou (2009)] for further results on contact of surfaces

with de Sitter horospheres.

9.8.3 Spacelike surfaces in Minkowski 3-space

The 3-dimensional Minkowski space R31 can be identified isometrically with
the set

{(x0 , x1 , x2 , x3 ) 2 R41 | x3 = 0}.

Let M be a spacelike surface patch in R31 ⇢ R41 , parametrised by x :

U ! R31 and denote by H 2 ( 1) the pseudo sphere {u 2 R31 : hu, ui = 1}.
Here we set nS (u) = e2 , so that

xu1 (u) ^ xu2 (u) ^ e2

nT (u) = 2 H 2 ( 1)
kxu1 (u) ^ xu2 (u) ^ e2 k1

is the timelike unit normal vector of M as a surface in R31 at p = x(u).

The lightcone Gauss map is given by

L± (u) = nT (u) ± e3 ,

and the normalised lightcone shape operator is given by

e ± )p =
W (L dnT (u).

It is the spacelike shape operator of M as a surface in Minkowski 3-space.

We thus have K e ± (u) = KS (u) and H̄ ± (u) = HS (u), where KS (resp. HS )
` `
is the spacelike Gauss-Kronecker curvature (resp. mean curvature) defined
in [Cheng and Yau (1973); Kobayashi (1983)]. It follows that the lightcone
flat spacelike surfaces in R31 ⇢ R41 are spacelike developable surfaces and
the marginally trapped surface are maximal surfaces.
 Spacelike surfaces in the Minkowski space-time 307

9.8.4 Surfaces in hyperbolic 3-space

The hyperbolic 3-space H+ 3
( 1) ⇢ R41 , with the induced metric, is a Rie-
3
mannian 3-manifold. Therefore, any embedded surface in H+ ( 1) is space-
3
like. The intersection of H+ ( 1) with a spacelike (resp. timelike, lightlike)
hyperplane is called a sphere (resp. an equidistant surface, a horosphere).
An equidistant surface is said to be a plane if it is given by the intersection
3
of H+ ( 1) with a timelike hyperplane through the origin. The study of
3
contact of a surface M in H+ ( 1) with lightlike hyperplanes is the study of
3
the contact of M with the horospheres of H+ ( 1). We refer to the geomet-
ric properties of M derived from this contact as the horospherical properties
of M .
Let M be a surface patch in the hyperbolic 3-space parametrised by
3
x : U ! H+ ( 1). Here we take nT (u) = x(u) so that
x(u) ^ xu1 (u) ^ xu2 (u)
nS (u) = 2 S13 .
kx(u) ^ xu1 (u) ^ xu2 (u)k1
We define the de Sitter Gauss map E : U ! S13 , as the map given by
E(u) = nS (u).
The lightcone Gauss map L± : U ! LC ⇤ is given by
L± (u) = x(u) ± E(u).

Remark 9.6. The normalised lightcone Gauss map is the same as the
hyperbolic Gauss map defined [Epstein (1986); Bryant (1987); Kobayashi
(1986)] in the Poincaré disk model of the hyperbolic space.
3
Remark 9.7. The intersection of lightlike hyperplanes with H+ ( 1) are
the horospheres. Therefore, the family of lightcone height functions on
3
M ⇢ H+ ( 1) measures the contact of M with horospheres.

Proposition 9.13. Let M be a surface patch in the hyperbolic 3-space

3
parametrised by x : U ! H+ ( 1). If one of the hyperbolic Gauss maps
e + or L
L e is constant, then the surface M is contained in a horosphere.

e + or L
Proof. It follows from Proposition 9.7 that L e is a constant light-
3
like vector v if and only if M ⇢ H(v, c) \ H+ ( 1), equivalently, M is a
subset of a horosphere. ⇤
For a general spacelike surface M in R41 , the image of the di↵erential
map dLp is not a subset of Tp M . In order to obtain a shape operator, we
308 Di↵erential Geometry from a Singularity Theory Viewpoint

considered the composite of dLp with the canonical projection ⇡1 : Tp R41 !

Tp M , i.e., W (L)p = ⇡1 dLp . However, for surfaces in hyperbolic space
the situation is rather di↵erent as shown by the following result, where
Dv denotes the covariant derivative with respect to the tangent vector
v 2 Tp M.
3
Lemma 9.4. Let M be a surface patch in H+ ( 1). For any p 2 M and
v 2 Tp M, we have Dv E 2 Tp M and Dv L 2 Tp M.
3
Proof. Let x : U ! H+ ( 1) be a parametrisation of M . Since
{xu1 , xu2 , x, E} is a basis of R1 , we have at p = x(u),
4

Dv E = 1 x u1 + 2 x u2 + µ1 x + µ2 E
for some real numbers 1 , 2 , µ1 , µ2 .
It follows from the fact that hE, Ei1 = 1 that hDv E, Ei1 = 0, so µ2 = 0.
From the identity hE, xi1 = 0 we get hDv E, xi1 = hE, Dv xi. But E is
a normal vector and Dv x is a tangent vector to M , so hE, Dv xi = 0, which
implies hDv E, xi1 = 0. Therefore µ1 = 0. Consequently,
Dv E = 1 x u1 + 2 x u2

is a vector in Tp M. As Dv x is in Tp M , Dv L± = Dv x ± Dv E is also a vector

in Tp M. ⇤
As a consequence of Lemma 9.4, the di↵erential maps dEp and dL± p =
1Tp M ± dE(u) are linear transformations Tp M ! Tp M , at p = x(u), and
they are both self-adjoint operators.

Remark 9.8. Let W (E)p = dEp denote the shape operator associated to
the de Sitter Gauss map and let Hd be the de Sitter mean curvature, so
Hd (u) = 12 Trace(W (E)p ), with p = x(u). Then
H`± (u) = ±Hd (u) 1.
It follows that surfaces in hyperbolic space with Hd ⌘ ±1 correspond
to surfaces with H`± ⌘ 0 which are marginally trapped in R41 .

Proposition 9.14. Suppose that M is a totally L-umbilic surface patch in

3
hyperbolic 3-space parametrised by x : U ! H+ ( 1). Then the lightcone
principal curvatures 1 and 2 are equal to a constant , and we have the
following classification of the surface M .
(i) Suppose that  6= 0.
(a) If  6= 1 and |+1| < 1, then M is a part of an equidistant surface.
 Spacelike surfaces in the Minkowski space-time 309

(b) If  6= 1 and | + 1| > 1, then M is a part of a sphere.

(c) If  = 1, then M is a part of a plane.
(ii) If  = 0, then M is a part of a horosphere.

Proof. We deal with the case for L+ , the case L follows similarly. By
definition, we have L+
ui = xui for i = 1, 2. Therefore, for i, j = 1, 2,

L+
ui uj = uj xui + xui uj .

Since L+
ui u j = L+
uj ui and xui uj = xuj ui , we have

u j x u i ui xuj = 0.

As xu1 , xu2 are linearly independent, ui = 0, i = 1, 2, and hence  is

constant.
We have dL+ = 1Tp M + dE, so L+ ui = xui is equivalent to E ui =
d xui , with  = 1 + d .
Suppose now that  6= 0. If  6= 1, then d 6= 0, so xui = (1/d )Eui .
Therefore, there exists a constant vector a such that x = a (1/d )E. But
since hx a, x ai1 = (1/d )2 ,
✓ ◆ ✓ ◆
1 1 1 1 1
ha, xi1 = + 1 ha, ai1 = +1+1 = 1,
2 2d 2 2d 2d
3
which means that M =⇢ HP (a, 1)\H+ ( 1). If |d | < 1, a = x+(1/d )E
is spacelike and we have assertion (i)-(a), and in case |d | > 1 we get
assertion (i)-(b).
If  = 1, then we have Eui = 0, so E is a constant vector a. Since a is
3
a spacelike vector and we have hx, ai1 = 0, M ⇢ HP (a, 0) \ H+ ( 1). This
means that M is a part of a plane.
Finally, we assume that  = 0. In this case, we have L+ ui = 0, so that L
+

is a constant lightlike vector a. This means that the L+ -lightcone Gauss

map is constant, and assertion (ii) follows from Proposition 9.13. ⇤

9.9 Lorentzian distance squared functions

The family of Lorentzian distance squared functions in R41 is the map G :

R41 ⇥ R41 ! R given by

G(p, a) = hp a, p ai1 .

The fibres of the map G are the pseudo spheres in R41 of centre a.
310 Di↵erential Geometry from a Singularity Theory Viewpoint

Let M be a spacelike surface patch parametrised by x : U ! R41 . The

family of Lorentzian distance squared functions G : U ⇥ R41 ! R on M is
the restriction of G to M and is given by
G(u, a) = hx(u) a, x(u) ai1 .
For a fixed a0 2 R41 , we write ga0 (u) = G(u, a0 ).

Proposition 9.15. Let M be a spacelike surface patch parametrised by

x : U ! R41 . Suppose that a0 6= p0 = x(u0 ). Then

(i) u0 is a singular point of ga0 with ga0 (u0 ) = 0 if and only if

a0 = x(u0 ) e 0)
L(u
for real non-zero scalar .
(ii) u0 is a degenerate singular point of ga0 with ga0 (u0 ) = 0 if and only if
1 e
a0 = x(u0 ) + L(u0 ),
ei (u0 )

for i = 1 or i = 2.

Proof. (i) The condition ga0 (u0 ) = 0 means that p0 a0 2 LC ⇤ . We

observe that dga0 (u0 ) = hdx(u0 ), x(u0 ) a0 i1 = 0 if and only if p0 a0 2
Np M. Hence ga0 (u0 ) = dga0 (u0 ) = 0 if and only if p0 a0 2 Np0 M \ LC ⇤ .
This is equivalent to p0 a0 = L(u e 0 ) for some real non-zero scalar .
(ii) We have, for i = 1, 2,
@g @2g
= 2hxui , x a0 i 1 and = 2(hxui uj , x a0 i1 + hxui , xuj i1 ).
@ui @ui @uj
When p0 e 0 ), hxu u (u0 ), x(u0 ) a0 i1 = hxu u (u0 ), L(u
a0 = L(u e 0 )i1
i j i j

so
✓ ◆ ✓ ◆
1 h11 h11 g11 g12
Hess(g)(u0 ) = 2 +2
`0 h12 h22 g12 g22
where all the entries of the right hand side of the above equality are evalu-
ated at u0 and `0 is the first coordinate of L. Therefore,
✓ ◆ 1
g11 g12 ep + I)
Hess(g)(u0 ) = 2( A 0
g12 g22
with I denoting the 2 ⇥ 2 identity matrix. Thus, det(Hess(ga0 ))(u0 ) = 0
if and only if 1/ is an eigenvalue of A ep , that is, = 1/e 1 (u0 ) or
0

= 1/e2 (u0 ). ⇤
 Spacelike surfaces in the Minkowski space-time 311

It can be shown by straightforward calculations that G is a Morse family

in a neighbourhood of each point (u, a) 2 G 1 (0), so we have the Legendrian
immersion L (G) : ⌃G ! P T ⇤ (R41 ), with
L (G)(u, a) = (a, [(x0 (u) a0 ) : (a1 x1 (u)) : (a2 x2 (u)) : (a3 x3 (u))]),
where a = (a0 , a1 , a2 , a3 ) and x(u) = (x0 (u), x1 (u), x2 (u), x3 (u)).
By Proposition 9.15, we have
⌃G = ( ⇤ G) 1 (0) = {(u, a) 2 U ⇥ R4 | a = x(u) + L(u), e 2 R \ {0}}.
1
This means that G is a generating family of the Legendrian immersion
L (G) whose discriminant set is given by
e
DG = {a = x(u) + L(u), u 2 U, 2 R \ {0}}. (9.4)
Theorem 9.3. There exists a residual subset O ⇢ Embs (U, R41 ) such that
for any x 2 O, the germ of the Legendrian immersion L (G) at each point
is Legendrian stable.
Proof. This is a consequence of Theorems 4.3, 5.11 and 5.12. ⇤

9.9.1 Lightlike hypersurfaces

Definition 9.5. A hypersurface S in R41 is said to be lightlike if its tangent
space at all of its points is a lightlike hyperplane.
Let M be a spacelike surface patch in R4 parametrised by x : U ! R41 .
We can construct a lightlike hypersurface S from M , as the image of the
map LHM : U ⇥ R ! R41 with
e
LHM (u, ) = x(u) + L(u).
The hypersurface LHM (U ⇥ R) is a ruled hypersurface with M its
base surface and L e its ruling. In fact, any lightlike hypersurface is,
at least locally, a ruled hypersurface with base surface some spacelike
surface patch and ruling one of the lightlike normal directions of the
base surface ([Kossowski (1989)]). It is clear that the lightlike hyper-
surface LHM (U ⇥ (R \ {0})) is the discriminant DG of the family of
Lorentzian distance squared functions G on M ; see (9.4). In particular,
LHM (U ⇥ (R \ {0})) is the wavefront of the Legendrian immersion L (G).
Since LHM is non-singular at (u, 0) 2 U ⇥ R, LHM (U ⇥ R) is a wavefront
set. As a consequence (see Proposition 9.15), a point (u, ) is a singular
point of LHM (U ⇥ R) if and only if
1 e
a = x(u) + L(u)
ei (u)

312 Di↵erential Geometry from a Singularity Theory Viewpoint

for i = 1 or i = 2.
We define
2 ⇢
[ 1 e
FM = x(u) + L(u) | i (u) 6= 0, u 2 U ,
i=1
ei (u)

and call it the lightlike focal set of M.
From Theorem 9.3 and from the classification of stable Legendrian map-
pings (Theorem 5.12), we obtain the following classification of germs of
generic lightlike hypersurface.

Corollary 9.9. For x 2 O, with O as in Theorem 9.3, the germ at any

point (u, ) of LHM of the parametrisation of the lightlike hypersurface
determined by M = x(U ) is A-equivalent to one of the following map-germs
(R3 , 0) ! (R4 , 0) :
(A1 ) f (x, y, z) = (x, y, z, 0),
(A2 ) f (x, y, z) = (3x2 , 2x3 , y, z),
(A3 ) f (x, y, z) = (4x3 + 2xy, 3x4 + yx2 , y, z),
(A4 ) f (x, y, z) = (5x4 + 3zx2 + 2xz, 4x5 + 2yx3 + zx2 , x, y),
(D4+ ) f (x, y, z) = (2(x2 + y 2 ) + xyz, 3x2 + yz, 3y 2 + xz, z),
(D4 ) f (x, y, z) = (2(x3 xy 2 ) + (x2 + y 2 )z, y 2 3x2 2xz, xy yz, z).

Proof. By Theorems 5.11 and 9.3, the Lorentzian distance squared func-
tion G is a K-versal deformation of ga0 at each point (u, a0 ) 2 U ⇥ R.
Therefore, we can apply the classification of Legendrian stable map-germs
(Theorem 5.12). Then the generating family is P -K-equivalent to one for
the following normal forms:
(Ak ) F (q1 , q2 , x) = xk+1 ± q22 + x1 + x2 q1 + · · · + xk q1k 1
, 1  k  4,
(D4+ ) F (q1 , q2 , x) = q13 + q23 + x1 + x2 q1 + x3 q2 + x4 q1 q2 ,
(D4 ) F (q1 , q2 , x) = q13 q1 q22 + x1 + x2 q1 + x3 q2 + x4 (q12 + q22 )
with x = (x1 , x2 , x3 , x4 ). We consider the D4+ -case and take F as above
(the other cases follow by similar calculations). Then ⌃⇤ (F ) is given by
{(q1 , q2 , 2(q12 +q22 )+x4 q1 q2 , 3q12 x4 q2 , 3q22 x4 q1 , x4 ) | (q1 , q2 , x4 ) 2 R3 }.
If we change the parameters into (q1 , q2 , x4 ) = (x, y, z) and apply the
linear transformation (X, Y, Z, W ) = (X, Y, Z, W ), the corresponding
Legendrian map-germ is
f (x, y, z) = (2(x2 + y 2 ) + xyz, 3x2 + yz, 3y 2 + xz, z). ⇤
 Spacelike surfaces in the Minkowski space-time 313

9.9.2 Contact of spacelike surfaces with lightcones

We have the following characterisation of spacelike surfaces contained in a
given lightcone LC ⇤ (a) = {p 2 R41 : hp a, p ai1 = 0}.

Proposition 9.16. Let M be a spacelike surface patch without lightcone

parabolic points and suppose that it is parametrised by x : U ! R41 . Then M
is contained in a lightcone LC ⇤ (a0 ) if and only if the lightlike hypersurface
LHM (U ⇥ R) is contained in LC ⇤ (a0 ) and a0 is an isolated singular value
of the map LHM .

Proof. In the first place, we remark that K e ` (u) 6= 0 if and only if the
e
three vectors L(u), e u (u), L
L e u (u) are linearly independent.
1 2

By definition, M is contained in a lightcone LC ⇤ (a0 ) if and only if

ga0 (u) = 0 for all u 2 U, where ga0 is the Lorentzian distance squared
function on M. It follows from Proposition 9.15 that there exists a smooth
function µ : U ! R \ {0} such that
e
x(u) = a0 + µ(u)L(u).
We have then
e
LHM (u, ) = a0 + ( + µ(u))L(u),
which implies that LHM (U ⇥ R) is contained in LC ⇤ (a0 ). Moreover, it
follows that
@LHM e
(u, ) = L(u),
@
@LHM e e u (u), i = 1, 2,
(u, ) = µui (u)L(u) + ( + µ(u))L i
@ui
from which we get
@LHM @LHM @LHM e e u (u) ^ Le u (u).
( ^ ^ )(u, ) = ( + µ(u))2 L(u) ^L 1 2
@µ @u1 @u2
Under the assumption K e
e ` (u) 6= 0, L(u) e u (u) ^ L
^L e u (u) 6= 0, so that
1 2

( @LH @LHM @LHM

@µ ^ @u1 ^ @u2 )(u, ) = 0 if and only if
M
+ µ(u) = 0. This means
that a0 is an isolated singularity value of LHM . Since M ⇢ LHM (U ⇥ R),
the converse assertion holds. ⇤
By Proposition 9.16, the lightlike hypersurface has the most degenerate
singular point if and only if the spacelike surface M is a subset of a light-
cone. In this case, the lightlike hypersurface itself is a part of a lightcone.
Therefore, we can consider lightcones as model lightlike hypersurfaces and
study the contact of spacelike surfaces with lightcones.
314 Di↵erential Geometry from a Singularity Theory Viewpoint

From Proposition 9.15, the lightcone LC ⇤ (a0 ) is tangent to the surface

M at p0 = x(u0 ) if and only if u0 is a singularity of ga0 and ga0 (u0 ) = 0.
We call LC ⇤ (a0 ) a tangent lightcone to M at p0 .
Let M and M̄ be two germs spacelike surfaces parametrised by x :
(U, u) ! (R41 , p) and x̄ : (Ū , ū) ! (R41 , p̄), respectively. Let LHM : (U, u) !
(R41 , p) and LHM̄ : (Ū , ū) ! (R41 , p̄), be parametrisations of the germs of
the lightlike hypersurfaces associated to M and M̄ respectively.
If the regular sets of LHM and LHM̄ are dense in their respective do-
maines, then by Theorem 5.10, LHM and LHM̄ are A-equivalent if and only
if the germs of their corresponding Legendrian lift are Legendrian equiva-
lent. Equivalently, by Theorem 5.11, the generating families G1 and G2 of
the Legendrian lifts are P -K-equivalent. Here G1 : (U ⇥ R41 , (u, a)) ! R
(resp. G2 : (Ū ⇥ R41 , (ū, ā)) ! R) denotes the germ of the family of the
Lorentzian distance squared function on M (resp. M̄ ).
Let g1,a (u) = G1 (u, a), and g2,ā (u) = G2 (u, ā). By Proposition 5.2,
K(M, LC ⇤ (a), p) = K(M̄ , LC ⇤ (ā), p̄) if and only if g1,a and g2,ā are K-
equivalent, so we can apply the results in Proposition 5.2.
We denote by Qr (x, u) (resp. Qr (x̄, ū)) the local ring of the germ of
function g1,a (resp. g2,ā ), which is defined by

E2
Qr (x, u) = ⇤ (M )E + Mr+1
.
g1,a 1 2 2

Theorem 9.4. Let M and M̄ be two germs of spacelike surfaces

parametrised by x : (U, u) ! (R41 , p) and x̄ : (Ū , ū) ! (R41 , p̄), respec-
tively. Suppose that their corresponding Legendrian lifts are Legendrian
stable. Then the following statements are equivalent.

(i) The germs LHM and LHM̄ are A-equivalent.

(ii) G1 and G2 are P -K-equivalent.
(iii) g1,a and g2,ā are K-equivalent.
(iv) K(M, LC ⇤ (a), p) = K(M̄ , LC ⇤ (ā), p̄)
(v) Q5 (x, u) and Q5 (x̄, ū) are isomorphic as R-algebras.

Proof. The discussion proceeding the proposition shows that (iii) and
(iv) are equivalent. The equivalence of the other statements follow by
Proposition 5.2. ⇤
 Spacelike surfaces in the Minkowski space-time 315

9.10 Legendrian dualities between pseudo-spheres

Given a Legendrian double fibration ⇡1 : E ! B1 and ⇡2 : E ! B2 and

L a Legendrian submanifold of E, we say that the projections ⇡1 (L) and
⇡2 (L) are Legendrian dual to each other.
The Legendrian duality is a generalisation of the classical projective
duality and the spherical duality [Shcherbak (1986); Nagai (2012)].
We present here a theorem on Legendrian dualities for pseudo-spheres
in Lorentz-Minkowski space. It plays a fundamental role in the study of the
extrinsic di↵erential geometry of submanifolds in these pseudo-spheres from
the view point of singularity theory ([Izumiya (2009); Chen and Izumiya
(2009); Izumiya and Tari (2008, 2010b,a); Izumiya and Yıldırım (2011);
Izumiya and Saji (2010)]).
Although the duality theorem holds for pseudo-spheres in general di-
mensions, we shall state it here for the pseudo-spheres in R41 .
Let v = (v0 , v1 , v2 , v3 ) and w = (w0 , w1 , w2 , w3 ) be two non-zero vectors
in R41 . The 1-forms hdv, wi1 and hv, dwi1 are defined as follows:
hdv, wi1 = w0 dv0 + w1 dv1 + w2 dv2 + w3 dv3 ,
hv, dwi1 = v0 dw0 + v1 dw1 + v2 dw2 + v3 dw3 .
Consider the following double fibrations and 1-forms:

(1) In H 3 ( 1) ⇥ S13 1 = {(v, w) | hv, wi1 = 0 },

(a) ⇡11 : 1 ! H 3 ( 1), ⇡12 : 1 ! S13 ,
(b) ✓11 = hdv, wi1 | 1 , ✓12 = hv, dwi1 | 1 .
±
(2) In H 3 ( 1) ⇥ LC ⇤ 2 = {(v, w) | hv, wi1 = ±1 },
±
(a) ⇡21 : ± n
2 ! H ( 1), ⇡22 :
± ±
2 ! LC ,
⇤
±
(b) ✓21 = hdv, wi1 | ± ± ±
2 , ✓22 = hv, dwi1 | 2 .
±
(3) In LC ⇤ ⇥ S13 3 = {(v, w) | hv, wi1 = ±1 },
±
(a) ⇡31 : ± ⇤
3 ! LC , ⇡32 :
± ± n
3 ! S1 ,
±
(b) ✓31 = hdv, wi1 | ± ±
3 , ✓32 = hv, dwi1 |
±
3.
±
(4) In LC ⇤ ⇥ LC ⇤ 4 = {(v, w) | hv, wi1 = ±2 },
±
(a) ⇡41 : ± ⇤
4 ! LC , ⇡42 :
± ±
4 ! LC ,
⇤
±
(b) ✓41 = hdv, wi1 | ± ±
4 , ✓42 = hv, dwi1 |
±
4.

If we consider a surface x : U ! H 3 ( 1), then we have the de Sitter

Gauss map E : U ! S13 . It follows that hx(u), E(u)i1 = 0. Moreover,
316 Di↵erential Geometry from a Singularity Theory Viewpoint

we have L± (u) = x(u) ± E(u), so that hx(u), x(u) E(u)i1 = 1 and

hx(u) + E(u), E(u)i1 = 1. We also have hx(u) + E(u), x(u) E(u)i1 = 2.
Thereofore, if we start with a suface in the hyperbolic space, 1 , 2 , + 3
and 4 are natural sets to consider. However, there is no reason to exclude
the other cases from the mathematical view point.

Remark 9.9. Observe that ✓11 1 (0) and ✓12 1 (0) define the same field of
± 1 ± 1
tangent hyperplanes over 1 , denoted by K1 . Also ✓i1 (0) and ✓i2 (0)
±
define the same field of tangent hyperplanes over i , i = 2, 3, 4, denoted
by Ki± .

We have the following duality theorem.

Theorem 9.5 ([Izumiya (2009)]). The contact manifolds ( 1 , K1 ) and

( ± ± ±
i , Ki ), i = 2, 3, 4, are contact manifolds such that ⇡1j and ⇡ij , j = 1, 2,
are Legendrian fibrations. Moreover, these contact manifolds are contact
di↵eomorphic to each other.

Proof. It follows from the definition of 1 and ± i , i = 2, 3, 4, that these

±
sets are smooth submanifolds of R1 ⇥R1 and that ⇡1j and ⇡ij
4 4
, i = 2, 3, 4; j =
1, 2, are smooth fibrations.
It is shown in [Izumiya (2009)] that ( 1 , K1 ) is a contact manifold.
We give here an outline of the proof. The restriction of the pseudo-scalar
product h, i1 to H 3 ( 1) is a Riemannian metric. Let ⇡ : S(T H 3 ( 1)) !
H 3 ( 1) be the unit tangent sphere bundle of H 3 ( 1).
For v 2 H 3 ( 1), a vector w 2 R41 is in Tv H 3 ( 1) if and only if
hv, wi = 0. Therefore, w 2 S(Tv H 3 ( 1)) if and only if hw, wi1 = 1
and hv, wi1 = 0. The last two conditions are equivalent to (v, w) 2 1 .
This means that we can identify canonically S(T H 3 ( 1)) with 1 . The
canonical contact structure on S(T H 3 ( 1)) is given by the 1-form ✓(V ) =
hd⇡(V ), ⌧ (V )i1 , where ⌧ : T S(T H 3 ( 1)) ! S(T H 3 ( 1)) is the tangent
bundle of S(T H 3 ( 1)) (see [Blair (1976); Cecil (1980)]). This form can
be represented by ✓11 = hdv, wi1 | 1 in the above identification. Thus,
( 1 , K1 ) is a contact manifold.
For ± i , i = 2, 3, 4, we define the smooth mappings
±
1i : 1 !
±
i by
±
12 (v, w) = (v, ⌥v + w),
±
13 (v, w) = (v ± w, w),
±
14 (v, w) = (v ± w, ⌥v + w).
The above mappings have smooth inverses, so they are di↵eomorphisms.
 Spacelike surfaces in the Minkowski space-time 317

±
Consider, for example, the map 12 . We have
± ⇤ ±
( 12 ) ✓21 = hdv, ⌥v + wi1 | 1

= hdv, ⌥vi1 | 1 + hdv, wi1 | 1

= hdv, wi1 | 1 = ✓11 .

This means that ( ± ±
2 , K2 ) is a contact manifold and
±
12 is a contacto-
± ±
morphism. Similar argument shows that ( i , Ki ), i = 3, 4, are contact
manifolds and ±1i , i = 3, 4, are contactomorphisms. ⇤
For a surface x : U ! H 3 ( 1), we have an embedding L1 : U ! 1
defined by L1 (u) = (x(u), E(u)). By the definition of the contact structure
K1 , L (U ) is a Legendrian submanifold such that the image of the de Sitter
Gauss map E(U ) is the wavefront set of L1 (U ) with respect to the Legen-
drian fibration ⇡12 : 1 ! S13 . Moreover, L (U ) is the wavefront set of a
certain Legendrian submanifold in 2 . Even if we start with a spacelike
surface in other pseudo spheres, we can apply the theory of Legendrian
singularities to obtain geometric invariants.
We can also consider the contactomorphisms ± ij :
±
i !
±
j for the
± ± ± ± ± 1
other pairs (i, j) by ij = i1 1j , where i1 = ( 1i ) . All these
contactomorphisms give rise to the following commutative diagram

F O X1

± ±
± 41 14
±
12 13
±
21
✏ ±
± 31
± 4 ±
42 : d 43

⌃ z ± ± ±
$ ⇠
24 34
± 23
/ ±
2 o 3
±
32

The above commutative diagram has a similar structure to a religious pic-

ture in Buddhism called “mandala”. For this reason, it is called mandala
of Legendrian dualities in [Chen and Izumiya (2009)].

9.11 Spacelike surfaces in the lightcone

The induced metric on the lightcone is degenerate, so one cannot apply

directly the results on di↵erential geometry of Riemannian and Lorentzian
318 Di↵erential Geometry from a Singularity Theory Viewpoint

surfaces to surfaces in the lightcone. We construct here the basic tools for
the study of the extrinsic di↵erential geometry of spacelike surfaces in the
lightcone LC ⇤ as an application of the mandala of Legendrian dualities.
Let M be a spacelike surface patch in the lightcone parametrised by
x : U ! LC ⇤ . We shall show the existence and uniqueness of the lightcone
normal vector to M as a consequence of Theorem 9.5.
Consider the double Legendrian fibration ⇡41 : 4 ! LC ⇤ , ⇡42 : 4 !
LC ⇤ and let v 2 LC ⇤ , where we set 4 = 4 , ⇡41 = ⇡41 , ⇡42 = ⇡42 . The
fibre of ⇡411 (v) can be identified with
{w 2 LC ⇤ | hv, wi1 = 2},
which is the intersection of LC ⇤ with a lightlike hyperplane, so it is a two
dimensional spacelike submanifold. For any p = x(u) 2 M, the normal
space Np M is a timelike plane, so there exists two lightlike lines on Np M.
One of the lines is generated by p = x(u). We remark that a lightlike
plane consists of lightlike vectors and spacelike vectors only. Moreover,
all lightlike vectors are linearly dependent. Therefore, if another lightlike
line is generated by w, then we have hw, x(u)i1 = c 6= 0. If necessary,
we consider w e = cw/2. Then we have hw, e x(u)i1 = 2. Therefore, the
1
intersection of ⇡41 (v) with the normal plane of M (a timelike plane) in R41
consists of only one point at each point on M . Since ⇡41 : 4 ! LC ⇤ is
a Legendrian fibration, there is a Legendrian submanifold parametrised by
L4 : U ! 4 such that ⇡41 L4 (u) = x(u). It follows that we have a smooth
map x` : U ! LC ⇤ such that L4 (u) = (x(u), x` (u)), i.e., ⇡42 L4 = x` .
Since L4 is a Legendrian embedding, we have hdx(u), x` (u)i1 = 0, so x` (u)
belongs to the normal plane of M at x(u).
Given another Legendrian embedding L14 (u) = (x(u), x`1 (u)), we have
that x` (u) and x`1 (u) are parallel. However, the relation hx(u), x` (u)i1 =
hx(u), x`1 (u)i1 = 2 holds, so x` (u) = x`1 (u). This means that L4 is the
unique Legendrian lift of x.

Definition 9.6. We call the vector x` (u) = ⇡42 L4 (u) the lightcone normal
vector to M at p = x(u). The map x` : U ! LC ⇤ (or, its image) is called
the lightcone dual of M. The map x e ` : U ! S+2 e` (u) = ⇡SL (x` (u)),
, with x
is called the lightcone Gauss map of M .

We can construct directly the lightcone normal vector x` (u) as follows.

Let ⇡ : R41 ! R30 be the canonical projection given by ⇡(x0 , x1 , x2 , x3 ) =
(x1 , x2 , x3 ) and denote r(u) = ⇡ x(u). Since ⇡|LC ⇤ : LC ⇤ ! R30 is an
embedding, r : U ! R30 is an embedding too. Therefore, we have the
 Spacelike surfaces in the Minkowski space-time 319

ordinary Euclidean unit normal N (u) of r(U ) = ⇡(M ) and the Euclidean
Gauss map N : U ! S 2 ⇢ R30 . We can now define a transversal vector field
r` (u) to M along ⇡(M ) in R30 by
r(u) 2(r(u) · N (u))N (u)
r` (u) = ,
(r(u) · N (u))2
where “·” is the usual Euclidean scalar product. It follows by the uniqueness
of the vector x` that
kr(u)k
x` (u) = ( 2
, r` (u)).
(r(u) · N (u))
We can study the extrinsic di↵erential geometry of M using the normal
vector field x` . Here too, as in the case of surfaces in the hyperbolic space,
the di↵erential of x` at each point is a linear transformation of Tp M . Indeed,

Lemma 9.5. For any p = x(u) 2 M and v 2 Tp M, we have Dv x` (u) 2

Tp M.

Proof. The proof is similar to that of Lemma 9.4. We have

Dv x` = 1 x u1 + 2 x u2 + µ1 x + µ2 x `
for some real numbers 1 , 2 , µ1 , µ2 .
Since hx` , xi1 = 2, Dv hx` , xi1 = 0, so that hDv x` , xi1 = hx` , Dv xi1 .
As Dv x is a tangent vector and x` is a normal vector we get hDv x` , xi1 =
hx` , Dv xi1 = 0, so µ2 = 0.
Similarly, hx` , x` i1 = 0 so hDv x` , x` i1 = 0, and from that we get µ1 =
0. (We have used in both calculations above the fact that the vectors
xui , i = 1, 2, are pseudo orthogonal to x and x` .) It follows then that
Dv x` = 1 xu1 + 2 xu2 2 Tp M. ⇤
We define the lightcone shape operator Wp` = dx` (u) : Tp M ! Tp M .
Following the same steps in the previous sections, we can define its asso-
ciated principal curvature and the Gauss-Kronecker and mean curvatures.
We also obtain a similar Weingarten formula as well as the formulae for the
Gauss-Kronecker and mean curvatures.
Taking into account the fact that the defining equation for 4 is
hv, wi1 = 2, we define the family of height functions H : U ⇥ LC ⇤ ! R
on M by
H(u, v) = hv (u) = hx(u), vi1 + 2
and call it the family of lightcone height functions on M .
320 Di↵erential Geometry from a Singularity Theory Viewpoint

One can obtain similar results to those in the previous sections about the
family H and derive accordingly information about the extrinsic geometry
of M in LC ⇤ .

Remark 9.10. The study of surfaces in hyperbolic space in §9.8.4 can

also be carried out using the duality Theorem 9.5. For any regular surface
patch parametrised by x : U ! H 3 ( 1), we have the lightcone Gauss map
L : U ! LC ⇤ . By definition, we have a Legendrian embedding L2 : U ! 2
given by L2 (u) = (x(u), L(u)). Since ⇡22 : 2 ! LC ⇤ is a Legendrian
fibration, L = ⇡22 L2 is a Legendrian map. All the results in §9.8.4 (and in
[Izumiya, Pei and Sano (2003)]) can be interpreted using this construction.

9.11.1 The Lightcone Theorema Egregium

The duality Theorem 9.5 can be used to obtain other curvatures of spacelike
surfaces in the lightcone. Let M be a spacelike surface patch in the lightcone
parametrised by x : U ! LC ⇤ . The map 41 : 4 ! 1 , given by
✓ ◆
v+w v w
41 (v, w) = ,
2 2
is a di↵eomorphism and
⇤ v+w v w 1 1 1
( 41 ) ✓12 = hd , i1 | 4 = hdv, wi1 + hdw, vi1 | 4 = ✓42 .
2 2 4 4 2
Therefore, 41 is a contactomorphism. It follows that the map L1 : U !
1 , given by

L1 (u) = 41 L4 (u),
is a Legendrian embedding, where L4 is as in §9.11. If we write L1 (u) =
(xh (u), xd (u)), then
x(u) + x` (u) x(u) x` (u)
xh (u) = , xd (u) = .
2 2
Definition 9.7. We call xh (u) (resp. xd (u)) the hyperbolic normal vector
(resp. de Sitter normal vector) of M at p = x(u).

As a consequence of Lemma 9.5, we have the hyperbolic shape operator

Wph = dxh (u) : Tp M ! Tp M
and the de Sitter shape operator
Wpd = dxd (u) : Tp M ! Tp M
 Spacelike surfaces in the Minkowski space-time 321

at p = x(u). From these, we define the following curvatures of M at p:

Hyperbolic Gauss-Kronecker curvature : Kh (u) = det(Wph )
de Sitter Gauss-Kronecker curvature : Kd (u) = det(Wpd )
Hyperbolic mean curvature : Hh (u) = 12 Trace(Wph )
de Sitter mean curvature : Hd (u) = 12 Trace(Wpd ).

Let l, m, n denote the coefficients of the second fundamental form asso-

ciated to x` , i.e., l = hx`u1 , x`u1 i1 , m = hx`u1 , x`u2 i1 , n = hx`u2 , x`u2 i1 .

Proposition 9.17. We have the following formulae:

(E l)(G n) (F m)2
Kh = ,
4(EG F 2 )

(E + l)(G + n) (F + m)2
Kd = .
4(EG F 2 )
Proof. We denote by
✓ ◆
a`11 a`12
A`p =
a`21 a`22
the matrix of the shape operator Wp` = dx` (u) with respect to the basis
{xu1 , xu2 } of Tp M . Following similar arguments in the proof of Proposition
9.2, we get a similar formula to (9.2), namely,
✓ ◆ ✓ ◆
EF ` l m
Ap = .
F G m n
Since xh = (x + x` )/2, we have
1
xhu1 = 2 (1 a`11 )xu1 a`21 xu2 ,
1
xhu2 = 2 a`12 xu1 + (1 a`22 )xu2
so the matrix of dxh with respect to the basis {xu1 , xu2 } of Tp M is
✓ ◆ ✓ ◆ ✓ ` ` ◆
1 1 a`11 a`12 1 10 a11 a12
` ` = 2( )
2 a21 1 a22 01 a`21 a`22
✓ ◆ 1 ✓ ◆ ✓ ◆✓ ` ` ◆
1 E F EF EF a11 a12
= 2 ( )
F G F G F G a`21 a`22
✓ ◆ 1 ✓ ◆ ✓ ◆
1 E F EF l m
= 2 ( )
F G F G m n
✓ ◆ 1✓ ◆
E F E l F m
= 12 .
F G F m G n
322 Di↵erential Geometry from a Singularity Theory Viewpoint

The curvature Kh is the determinant of the matrix in the left hand side
of the equality above, and the result follows by taking the determinant of
the right hand side of that equality.
The formula for Kd follows the same steps as above with an appropriate
change of signs as xd = (x x` )/2. ⇤
We denote, as usual, by kij the Christo↵el symbols of M (which is a
Riemannian manifold), where
X2 ⇢
k 1 @gjm @gim @gij
ij = g km + .
2 m=1 @ui @uj @um

Above, (gij ) is the matrix of the first fundamental form and (g km ) is its
inverse matrix. Using the notation E, F, G for the coefficients of the first
fundamental form, the Christo↵el symbols are given by the following six
functions:
1 GEu1 2F Fu1 + F Eu2 2 2EFu1 EEu2 F Eu1
11 = , 11 = ,
2(EG F 2 ) 2(EG F 2 )
1 1 GEu2 F Gu1 2 2 EGu1 F Eu2
12 = 21 = , 12 = 21 = ,
2(EG F 2 ) 2(EG F 2 )
1 2GFu2 GGu1 F Gu2 2 EGu2 2F Fu2 + F Gu1
22 = , 22 = .
2(EG F 2 ) 2(EG F 2 )
Proposition 9.18. Let M be a spacelike surface patch in the lightcone
parametrised by x : U ! LC ⇤ and let l, m, n denote the coefficients of
the second fundamental form associated to x` . Then the lightcone Gauss
equations are given by
1 2 1 `
x u1 u 1 = 11 xu1 + 11 xu2 2 (lx + Ex ),
1 2 1 `
x u1 u 2 = 12 xu1 + 12 xu2 2 (mx + F x ),
1 2 1 `
x u2 u 2 = 22 xu1 + 22 xu2 2 (nx + Gx ).

Proof. Since {xu1 , xu2 , x, x` } is a basis of R41 , we can write

xui uj = ⇤1ij xu1 + ⇤2ij xu2 + ij x + µij x` ,
for some scalars ⇤kij , ij and µij . From hx, xi1 = 0 we get hxu1 , xi1 = 0
and di↵erentiating again we get hxu1 u1 , xi1 = hxu1 , xu1 i1 = E. But
hxu1 u1 , xi1 = µ11 hx` , xi1 = 2µ11 , so µ11 = (1/2)E.
By definition hxu1 u1 , x` i1 = l, hence 2 11 = l, so that 11 = (1/2)l.
It follows that
1
xu1 u1 = ⇤111 xu1 + ⇤211 xu2 (lx + Ex` ).
2
 Spacelike surfaces in the Minkowski space-time 323

The expressions for xu1 u2 and xu2 u2 follow similarly. Moreover, we have
1
Eu1 = hxu1 u1 , xu1 i1 = ⇤111 E + ⇤211 F.
2
By the similar arguments to the above, we have
1 1
Eu2 = ⇤112 E + ⇤212 F, Gu1 = ⇤112 F + ⇤212 G,
2 2
1 1
Fu 2 Gu1 = ⇤122 E + ⇤222 F, Gu2 = ⇤122 F + ⇤222 G.
2 2
It follows that
✓ ◆✓ 1 1 1 ◆ ✓ ◆
EF ⇤11 ⇤12 ⇤22 1 Eu1 Eu2 2Fu2 Gu1
= ,
F G ⇤211 ⇤212 ⇤222 2 2Fu1 Eu2 Gu1 Gu 2

which gives ⇤kij = k

ij . ⇤

Now from xh = (x + x` )/2 and xd = (x x` )/2 we get x = xh + xd and

x` = xh xd . Substituting these expressions in Propostion 9.18 yields the
following corollary.

Corollary 9.10. Let M be a spacelike surface patch in the lightcone

parametrised by x : U ! LC ⇤ . Then,
1 2 1 h 1
x u 1 u1 = 11 xu1 + 11 xu2 2 (l + E)x 2 (l E)xd ,
1 2 1 h 1
x u 1 u2 = 12 xu1 + 12 xu2 2 (m + F )x 2 (m F )xd ,
1 2 1 h 1
x u 2 u2 = 22 xu1 + 22 xu2 2 (n + G)x 2 (n G)xd .

Let Ks denote the sectional curvature of M . It is known that

E(Eu2 Gu2 2Fu1 Gu2 + G2u1 )
Ks =
4(EG F 2 )2
F (Eu1 Gu2 Eu2 Gu1 2Eu2 Fu2 2Fu1 Gu1 + 4Fu1 Fu2 )
+
4(EG F 2 )2
G(Eu1 Gu1 2Eu1 Fu2 + Eu22 ) Eu2 u2 2Fu1 u2 + Gu1 u1
+ .
4(EG F 2 )2 2(EG F 2 )
It follows that Ks depends only on the first fundamental form, which means
that is an intrinsic property of the surface M .

Proposition 9.19. Let M be a spacelike surface in the lightcone. Then,

Kh Kd = Ks .
324 Di↵erential Geometry from a Singularity Theory Viewpoint

Proof. We have (hxu1 u1 , xu2 i1 )u2 = hxu1 u1 u2 , xu2 i1 + hxu1 u1 , xu2 u2 i1 . By

the relations ⇤kij = kij in the proof of Proposition 9.18, we have
✓ ◆ ✓ ◆
1 1 1 1 2
hxu1 u1 u2 , xu2 i1 = Fu1 Eu 2 Eu1 22 Fu 1 Eu 2 22
2 u2 2 2
1 1
+ (` + E)(n + G) (` E)(n G).
4 4
We also have
1 1 1
hxu1 u2 u1 , xu2 i1 = Gu1 u1 Eu2 112 Gu 2
2 2 2 1 12
1 1
+ (m + F )2 (m F )2 .
4 4
Since xu1 u1 u2 = xu1 u2 u1 , we have
✓ ◆ ✓ ◆
1 1 1 1 2
Fu 1 Eu 2 Eu1 22 Fu 1 Eu 2 22
2 u2 2 2
1 1 1
Gu1 u1 + Eu2 112 + Gu1 212
2 2 2
1 1
= (` + E)(n + G) + (` E)(n G)
4 4
1 1
+ (m + F )2 (m F )2 .
4 4
If we substitute the Christo↵el symbols kij by their expressions in terms
of the coefficients of the first fundamental form in the above equality, we
get the left hand side of the equality equal to (EG F 2 )Ks . Therefore we
have
((` E)(n G) (m F )2 (` + E)(n + G) + (m + F )2 )
Ks =
4(EG F 2 )
= Kh Kd .
⇤
The classical “Theorema Egregium of Gauss”asserts that the Gaussian
curvature of a surface in Euclidean space is equal to its sectional curvature.
For a spacelike surface in the lightcone, we have the following result which
asserts that the lightcone mean curvature is equal to the sectional curvature
of M . In particular, the lightcone mean curvature is an intrinsic property
of M .

Theorem 9.6 (The Lightcone Theorema Egregium). For a space-

like surface M in the lightcone LC ⇤ , we have
Ks = Kd K h = H` = Hh Hd .
 Spacelike surfaces in the Minkowski space-time 325

Proof. Let p = x(u) be a point on M . We denote the eigenvalues of Wp`

by `1 (u) and `2 (u), those of Wph by h1 (u) and h2 (u), and those of Wpd by
d1 (u) and d2 (u).
From the equality Wp` = Wph Wpd , we get `i (u) = hi (u) di (u), for
i = 1, 2. It follows that H` (u) = Hh (u) Hd (u).
On the other hand, we have
h x(u) + x` (u) d x(u) x` (u)
x (u) = , x (u) = ,
2 2
so that
1 1
Wph = ( 1Tp M + Wp` ), Wpd = ( 1Tp M Wp` ).
2 2
Therefore, we have for i = 1, 2,
1 1
hi (u) = ( 1 + `i (u)), di (u) = ( 1 `i (u)),
2 2
and from these equalities we get
1
Kh (u) = h1 (u)h2 (u) = (1 2H` (u) + K` (u)),
4
1
Kd (u) = d1 (u)d2 (u) = (1 + 2H` (u) + K` (u)).
4
It follows that
Kd (u) Kh (u) = H` (u).
The relation Ks (u) = Kd (u) Kh (u) is the result in Proposition 9.19. ⇤

9.12 Notes

The lightlike geometry in the hyperbolic space is called Horospherical Ge-

ometry and describes the contact of submanifolds in Hyperbolic space with
hyperhorospheres. The basic study of the horospherical geometry for hy-
persurfaces in Hyperbolic space with general codimension is carried out in
[Izumiya, Pei and Sano (2003)]. A detailed study of the contact of surfaces
in Hyperbolic 4-space with hyperhorospheres is carried out in [Izumiya, Pei
and Romero-Fuster (2006)] via the singularities of the lightcone height
functions. That led to the introduction of the concepts of osculating hyper-
horospheres, horobinormals, horoasymptotic directions and horospherical
points and to certain conditions ensuring their existence.
Kasedou [Kasedou (2010a,b)] studied spacelike submanifolds immersed
in codimension higher than one in de Sitter space. The results are analogous
326 Di↵erential Geometry from a Singularity Theory Viewpoint

to those of submanifolds immersed in Hyperbolic space studied in [Izumiya,

Pei, Romero Fuster and Takahashi (2005)]. Kasedou introduced the notion
of horospherical hypersurface and spacelike canal hypersurface by using
timelike unit normal vector fields, and showed that the horospherical hy-
persurface of a spacelike submanifold is the wavefront set of horospherical
height functions. The singularities of the lightcone Gauss map and the
lightcone Gauss image of hypersurfaces in de Sitter space are investigated
in [Kasedou (2009)]
The study of the lightlike geometry of codimension two submanifolds of
a Lorentz-Minkowski space ([Izumiya and Romero Fuster (2007)] for the
general dimension case) is generalised in [Izumiya and Kasedou (2014)] to
higher codimension submanifolds. This is done by introducing the notion
of codimension two spacelike canal submanifold and applying the results
presented in this chapter or [Izumiya and Romero Fuster (2007)].
The behaviour of lightlike hypersurfaces in Lorentz-Minkowski space is
investigated in [Izumiya, Romero Fuster and Saji (2009)] where a con-
cept of flatness for lightlike hypersurfaces is introduced. The flat lightlike
hypersurfaces are characterised as envelopes of certain families of lightlike
hyperplanes. Their generic singularities are the suspended cuspidal edge,
the suspended swallowtail, the suspended cuspidal cross-cap and the A4 -
type hypersurface singularity. It is interesting to notice that the D4± -type
singularities do not appear generically as singularities of flat lightlike hy-
persurfaces, whereas the suspended cuspidal cross-cap does not appear as a
generic singularity in the general case of lightlike hypersurfaces (Corollary
9.9).
Surfaces in the Minkowski 3-space are also studied using the singularitry
theory approach exposed in this book. The induced pseudo-metric on any
closed surface in R31 degenerates at some point on the surface. We call
the locus of such points the locus of degeneracy of the metric (LD). In
general, the LD is a smooth curve and separates locally the surface into
a spacelike and a timelike region. Very interesting problems arrise when
trying to understand how the Riemannian geometry on one side and the
Lorentzian geometry on the other side meet on the LD (and in general,
on surfaces with degenerate metrics, see for example, [Genin, Khesin and
Tabachnikov (2007); Kossowski (1987); Pelletier (1995); Tabachnikov
(1997); Remizov (2009); Steller (2006)].) Here, the contact theory can
be very usefull. Contact with lines and planes does not depend on the
metric, so the results in Chapter 6 apply. When considering the contact
with lines (i.e., projections of the surface to planes), we can restrict to
 Spacelike surfaces in the Minkowski space-time 327

projecting along the lightlike directions. These can be parametrised by a

circle on the lightcone. Varying the lightlike direction of projections gives
a 1-parameter family of contour generators and apparent contours. It is
shown in [Izumiya and Tari (2013)], amongst other things, that the families
of contour generators and apparent contours are solutions of certain first
order ordinary di↵erential equations. The caustic of a surface in R31 can
also be defined without the use of the metric. In [Tari (2012)] is studied
the behaviour of the induced metric on the caustic, including at the LD of
the surface. (For caustics of plane curves, see [Saloom and Tari (2012)].)
The lines of principal curvatures and the way they extend across the LD
are studied in [Izumiya and Tari (2010c)].
May 2, 2013 14:6 BC: 8831 - Probability and Statistical Theory PST˙ws

This page intentionally left blank

 Chapter 10

Global viewpoint

In the previous chapters we used singularity theory techniques to study a

submanifold M of Euclidean space and of Minkowski space-time locally at
a given point on M . In this chapter, we illustrate how those techniques can
be used to obtain global results on M (which we shall assume to be closed,
i.e., compact without boundary). The approaches we adopt here are the
following.
The first uses a stratification of the parameters space of the family of
functions and mappings defined on M . The parameters space can be strat-
ified according to the singularity types of the di↵erent functions associated
to each parameter. In fact, these stratifications can be regarded as pull-
backs of convenient stratifications on the space of smooth functions (or
mappings) on M [Gibson, Wirthmüller, du Plessis and Looijenga (1976)].
The multi-transversality conditions imposed by the genericity theorems on
these families provide all the information we need on the behaviour of strata
in the parameter spaces. Then, appropriate global topological considera-
tions on the induced strata lead to the desired results. This allows us to
obtain, for instance, relations between di↵erent geometric invariants asso-
ciated to M .
The second approach is a topological one. We compute the Euler char-
acteristic of a surface (or a given geometric set such as the wavefront) either
in terms of the total curvatures defined in the previous chapters or in terms
of the number of certain stable singularity types of the members of the
families of functions and maps on the surface.
The third approach gravitates around the Poincaré-Hopf formula. We
have on the surface (or part of it) direction fields defined by the contact
directions associated to the families of functions and mappings defined on
M . The integral curves of these fields define foliations whose singular points

329
330 Di↵erential Geometry from a Singularity Theory Viewpoint

are of interest. (For example, the umbilic points on a surface in R3 are the
singular points of the integral curves of the principal directions, which are
the contact directions of the distance squared functions on the surface.)
One can use topological arguments, such as the Poincaré-Hopf formula, to
obtain lower bounds for the number of such points.
We outline several applications of the above approaches on surfaces M
in R3 and R4 (§10.1). We also consider the case of spacelike surfaces in
Minkowski space-time R41 (§10.2). We comment in §10.3 on some other
global results on submanifolds of Euclidean and Minkowski spaces.
We emphasise that our aim in this chapter is to give some applications of
singularity theory to the study of global properties of submanifolds, related
mainly to the work of the authors on the subject.

10.1 Submanifolds of Euclidean space

10.1.1 Surfaces in R3

Definition 10.1. The convex hull H(S) of a subset S in Rn is the inter-

section of all the convex subsets of Rn containing S, that is the minimal
convex subset of Rn that contains S.

The convex hull H(M ) of a surface in R3 is homeomorphic to a closed

3-disc and its boundary H(M ) is a C 1 -surface which is C 1 -di↵eomorphic
to the standard 2-sphere S 2 (see for example [Romero Fuster (1981)]).

Definition 10.2. A closed surface M in R3 is said to be convex if it coin-

cides with the boundary H(M ) of its convex hull. Points on the surface M
that lie on H(M ) are called external points. Other points on M are called
internal points.

It is a well established geometrical property that a surface M is convex

if and only if its Gaussian curvature is non-negative.
A point p 2 M is an external point if and only if it is an absolute
minimum of some height function hv , v 2 S 2 , on M . Then v is, up to a
sign, the unit normal vector to M at p.

Definition 10.3. A stratification of a subset S of a manifold M is a locally

finite partition S of S into locally closed submanifolds of M , called strata.
The pair (S, S) is said to be a stratified subset of M .
 Global viewpoint 331

We can view a finite plane graph as a stratified set. The vertices and
edges being respectively the 0- and 1-dimensional strata. Moreover, if this
graph lies on a surface, we can view the whole surface as a 2-dimensional
stratified set whose 2-dimensional strata are the connected components of
the complement of the graph.
Any embedding f of a surface M in R3 induces a stratification of the
2-sphere. This stratification is constructed according to the type of the
absolute minima of the family of height functions on the embedded sur-
face M . Provided that the embedding f is height functions generic, the
multi-transversality conditions on the height functions family imply that we
only have the following possibilities for the absolute minima of the height
functions on the surface M (a detailed definition for the general case of
hypersurfaces in Rn can be found in [Romero Fuster (1983)]):

(1) Morse strata: The absolute minimum of hv is attained at a unique

point of type A1 . We say that v 2 S 2 is of type A1 . All points of type
A1 form an open region in S 2 whose complement is a graph, called the
Maxwell subset associated to the embedding f .
(2) Conflict strata: There are k = 2 or k = 3 absolute minima of hv , each
one of type A1 . Then hv lies in a codimension k 1 stratum of the
Maxwell subset in S 2 .
(3) Bifurcation strata: The absolute minimum of hv is attained at a unique
point of type A3 . The function hv lies in a codimension 2 stratum of
the Maxwell subset of f in S 2 .

We call the above the Maxwell stratification of S 2 associated to the

embedding f . It is shown in [Romero Fuster (1983)] that the strata of
the Maxwell stratification is a Whitney regular stratification satisfying the
frontier condition. Moreover, we have the following relation
3
X
2
(S M) + ( 1)j ( (Bj ) + (Mj ) + (Cj )) = 2 (10.1)
j=0

where M is the union of strata of codimension 1 in the Maxwell stratifi-

cation and Bj , Mj and Cj are, respectively, the union of bifurcation, mixed
and conflict strata of codimension j, and (X) denotes the Euler charac-
teristic of the set X.
From the characterisation of the cusp points of the Gauss map in terms
of height functions, it follows that the extremal points of the Maxwell graph
(bifurcation strata of type A2 ) correspond to external cusps of the Gauss
332 Di↵erential Geometry from a Singularity Theory Viewpoint

map (i.e., cusps of the Gauss map lying on the boundary of the convex
hull of the surface). The other vertices of the graph (conflict strata of type
A1 A1 A1 ) are of degree 3 (i.e., they are the end points of exactly 3 edges)
and correspond to the (isolated) tri-tangent support planes of the surface.
The edges of the graph (conflict strata of type A1 A1 ) correspond to the
normal directions to the 1-parameter family of support bi-tangent planes
of the surface.
Applying equality (10.1) to surfaces in R3 leads to the following result.

Corollary 10.1 ([Romero Fuster (1988)]). Given a height function

generic surface M in R3 , the numbers C of external cusps of Gauss and T
of tri-tangent support planes of M satisfy
C T =4 2 (M \ H(M )).

It follows from Corollary 10.1 that the existence of support tri-tangent

planes implies the existence of external cusps of the Gauss map.
The above considerations can be applied to the canal surface of a closed
space curve . When the radius of the circle of intersection of the canal
surface with the normal plane to the curve is small enough, the canal surface
is a torus embedded in R3 . One can show that the Maxwell graph of the
family of height functions on coincides with the Maxwell graph of the
canal surface. Comparing the singularities of the height function on the
curve and on its canal surface leads to the following result.

Theorem 10.1 ([Romero Fuster (1988)]). Given a closed curve

generically immersed in R3 , denote respectively by C, T and ⇢ the numbers
of external torsion zero points, support tri-tangent planes and connected
components of \ H( ). Then the following relation holds:
C T =4 2⇢.

A consequence of the above theorem is the following result for convex

curves, i.e. closed curves lying on the boundary of their convex hull. This
is known as the 4-vertex theorem for closed curves in R3 .

Corollary 10.2. Any convex closed curve generically immersed in R3 has

at least 4 torsion zero points.

An extension of 4-vertex theorem for closed curves in R3 to (non nec-

essarily generic) convex closed curves in R3 with no zero curvature points
is given in [Sedykh (1992)]. Other generalisations to curves with isolated
 Global viewpoint 333

points of zero curvature and to singular curves are obtained in [Romero

Fuster and Sedykh (1995); Costa and Romero Fuster (1997); Romero
Fuster and Sedykh (1997)].

10.1.2 Wavefronts
For a Legendrian fibration ⇡ : E ! N , where dim N = 3, we consider a
Legendrian immersion i : L ! E. The wavefront W (L) of a generic Legen-
drian immersion has singularitites of type cuspidal edges (A2 ), swallowtails
(A3 ) and points of transversal self-intersection (A1 A1 , A1 A2 , A1 A1 A1 ); see
[Arnol’d (1990)]. (In Theorem 10.2 and Corollary 10.3 bellow, generic
means that the wavefront has only the above local and multi-local singular-
ities.) If L is a closed surface, then the numbers of swallowtails and triple
points are finite and satisfy the following relation.

Theorem 10.2 ([Izumiya and Marar (1993)]). Let i : L ! E be a

generic Legendrian immersion of a closed surface. Then,
1
(W (L)) = (L) + T (L) + S(L),
2
where T (L) is the number of triple points on W (L) and S(L) is the number
of swallowtail points.

Proof. We define the following sets:

D2 (L) = {x 2 L | ](⇡|L ) 1 (x) 2},
D3 (L) = {x 2 D2 (L) | ](⇡|L ) 1
(⇡(x)) = 3},
D2 (L, (2)) = {x 2 D2 (L) | ](⇡|L ) 1
(⇡(x)) = 1},
where X is the topological closure of X. Since W (L) is a generic wavefront,
D2 (L) is a union of circles and curves on L with self-intersection, D3 (L) is
the inverse image of triple points, and D2 (L, (2)) is the set of swallowtails of
⇡|L . It follows that these are immersed submanifolds of L with dim D2 (L) =
1 and dim D3 (L) = dim D2 (L, (2)) = 0.
We consider the following equation
(W (L)) = ↵ (L) + (D2 (L)) + (D2 (L, (2)) + (D3 (L)), (10.2)
in the unknown variables ↵, , , . We solve the equation by a purely
combinatorial method. We consider a triangulation TL of the stratified
set W (L) as follows. We start to triangulate W (L) by including the
image of D2 (L, (2)) and the image of D3 (L) among the vertices of KL .
334 Di↵erential Geometry from a Singularity Theory Viewpoint

(1)
After this, we build up the one-skeleton KL of KL so that the im-
(1)
age of D2 (L) is a sub-complex of KL . We complete our procedure by
(2)
constructing the two-skeleton KL . Since ⇡|D2 (L) , ⇡|D2 (L,(2)) , ⇡|D3 (L) are
proper finite to one maps, we can pull back KL to obtain a triangulation
of L, D2 (L), D2 (L, (2)) and D3 (L). Let CjX be the number of j-cells in X,
where X = W (L), L, D2 (L), D2 (L, (2)) or D3 (L). Then equation (10.2) can
be written as
X W (L)
X X D 2 (L)
( 1)j Cj =↵ ( 1)j C Lj + ( 1)j Cj
j j j
X 2
D (L,(2))
X D 3 (L)
+ ( 1)j Cj + ( 1)j Cj ,
j j

where CjX = 0 for i > dim X. Therefore, if we can find real numbers ↵, ,
and such that
W (L) D 2 (L) D 2 (L,(2)) D 3 (L)
Cj = ↵CjL + Cj + Cj + Cj ,
for any j, then we have solutions of equation (10.2). We deal with the case
j = 0. We remark that ⇡|L is three-to-one over the points of ⇡(D3 (L)),
one-to-one over the points of ⇡(D2 (L, (2))), two-to-one over the points of
⇡(D2 (L) \ (D2 (L, (2)) [ D3 (L))), and one-to-one over the points of ⇡(N \
D2 (L)). It follows that equation
W (L) D 2 (L) D 2 (L,(2)) D 3 (L)
C0 = ↵C0L + C0 + C0 + C0
is equivalent to the system of linear equations
0 1 0 10 1
1 1000 ↵
B1C B2 2 0 0C B C
B C=B CB C
@1A @1 1 1 0A @ A .
1 3303
Solving the above linear system gives ↵ = 1, = 1/2, = 1/2 and
= 1/6. Thus, we have
1 1 1
(W (L)) = (L) (D2 (L)) + (D2 (L, (2))) (D3 (L)). (10.3)
2 2 6
We have, by definition, (D2 (L, (2)) = S(L) and (D3 (L)) = 3T (L).
Since D2 (L) is the union of closed curves on the surface L with 3T (L)
crossings, we can triangulate it with (3T (L) + n) 0-cells and (6T (L) + n)
1-cells, where n is the number of circles in D2 (L). We get the desired result
by substituting these in (10.3). ⇤
 Global viewpoint 335

Remark 10.1. We have a similar formula to that in Theorem 10.2 when

considering cross-caps or cuspidal cross-caps instead of swallowtails [Izu-
miya and Marar (1993)]. The reason is that these points are locally home-
omorphic to a swallowtail. Further generalisations of Theorem 10.2 can
be found in [Sedykh (2012); Nuño Ballesteros and Saeki (2001); Houston
(1999); Kossowski (2007)].

We can apply Theorem 10.2 to closed surfaces in R3 . For an orientable

surface M ⇢ R3 , we have the global Gauss map N : M ! S 2 and the
corresponding cylindrical pedal surface given by the image of the map
CPe : M ! S 2 ⇥ R, with CPe (p) = (N (p), hp, N (p)i). In Chapter 5,
we have shown that the cylindrical pedal CPe (M ) is the wavefront of a
certain Legendrian immersion. We have also shown in Chapter 6 that the
swallowtail singularities of CPe are exactly the cusps of the Gauss map. We
have the geometric characterisation of the cusps of the Gauss map (The-
orem 6.3). The triple points of CPe (M ) are the tri-tangent planes of M .
From the Gauss-Bonnet theorem we get the following corollary of Theorem
10.2.

Corollary 10.3. For a generic closed and orientable surface M in R3 , we

have
Z
1 1
(CPe (M )) = K dvM + T t (M ) + C g (M ),
2⇡ M 2
where T t (M ) is the number of tri-tangent planes of M and C g (M ) is the
number of the cusps of the Gauss.

As with the classical Gauss-Bonnet formula, all invariants of the right

hand side of the formula in Corollary 10.3 are di↵erential geometric invari-
ants and the left hand side is a topological invariant of the surface. Other
applications of Theorem 10.2 can be found in [Izumiya and Marar (1993);
Izumiya and Romero Fuster (2006); Izumiya (2009); Houston and van
Manen (2009)].

10.1.3 Surfaces in R4
Inflection points of a surface M in R4 are the singular points of the asymp-
totic curves on M (see Chapter 7). For a generic surface, the fields of
asymptotic directions have index ± 12 at inflection points of imaginary type.
The inverse of the stereographic projection from R3 to S 3 maps the lines
of principal curvature of a surface in R3 to the asymptotic curves of the
336 Di↵erential Geometry from a Singularity Theory Viewpoint

spherical image of the surface (which we consider as a surface in R4 ). It

follows that a closed surface in R4 which lies in the unit sphere S 3 and is
height function generic has two orthogonal asymptotic directions at each
point except maybe at a finite number of inflection points of imaginary
type. This property holds for a more general class of surfaces in R4 .

Definition 10.4. We say that a surface M in R4 is locally convex if it has

locally a support hyperplane at each point.

In fact, any surface in S 3 is locally convex. The orthogonal direction to

a support hyperplane at a point p 2 M determines a height function that
has either a (local) minimum or a (local) maximum at p.
A height function generic surface M in R4 is locally convex if and only if
it consists of hyperbolic and (isolated) inflection points ([Mochida, Romero
Fuster and Ruas (1995)]). It follows that generic locally convex surfaces
have two globally defined asymptotic directions whose critical points are
isolated imaginary inflection points. We can now apply the generalised
Poincaré-Hopf formula ([Pugh (1968)] for surfaces with boundary to get
the following result.

Theorem 10.3. For a height function generic closed and locally convex
surface M in R4 we have
2| (M )|  ]{inflection points}.

Remark 10.2. A similar inequality to that in Theorem 10.3 is given in

[Garcia, Mochida, Romero Fuster and Ruas (2000)] for the Euler charac-
teristic of the closure of the hyperbolic region cl(Mh ) of a generic non-locally
convex surface M. Also in [Bruce and Tari (2002)] are obtained formulae
for (M ) in the locally convex case and for (cl(Mh )) in the general case
in terms of the number and type of the inflection points of M .

A consequence of Theorem 10.3 is the following.

Corollary 10.4. Any height function generic closed and locally convex sur-
face in R4 with non-vanishing Euler characteristic has at least 4 inflection
points.

Using the stereographic projection we recover the following result from

[Feldman (1967)] for the number of umbilics of a generic closed surface in
R3 .
 Global viewpoint 337

Corollary 10.5. Any 2-sphere generically immersed into R3 has at least

4 umbilic points. Here generic means that the surface is distance-squared
function generic.

The result in Corollary 10.5 is the generic version of the following con-
jecture.
Carathéodory conjecture: any closed, convex and sufficiently smooth
surface in three dimensional Euclidean space has at least two umbilic points.
A possible approach for solving the conjecture is by using Poincaré-Hopf
formula and investigating the possible values for the index of an umbilic
point. It is known, for instance, how to construct examples of local immer-
sions of surfaces with umbilics of any index  1. This leads to the following
conjecture which can be considered as a local version of the Carathéodory
conjecture.
Loewner conjecture: the index of the principal directions field at an
umbilic point of a sufficiently smooth surface in three dimensional Euclidean
space is at most 1.
A review on the state of Carathéodory’s conjecture can be found in
[Gutierrez and Sotomayor (1998)]. In view of Corollary 10.4, it is reason-
able to propose the following conjecture for surfaces in R4 .
Carathéodory conjecture for surfaces in R4 : every closed, convex
and sufficiently smooth surface in four dimensional Euclidean space which
is homeomorphic to a 2-sphere has at least two inflection points.
The above conjecture is shown to be true in some particular cases in
[Gutiérrez and Ruas (2003); Nuño Ballestero (2006)].

Remark 10.3. It is shown in [Gutierrez and Sánchez Bringas (1998)] that,

given any n 2 Z, there is an analytic immersion f : R2 ! R4 having a nor-
mal field ⌫ and a ⌫-umbilic point p with index n2 . This implies that the
Loewner conjecture does not hold for principal configurations associated to
arbitrary normal fields on submanifolds in 4-space. The Carathéodory con-
jecture for surfaces in R4 is only about the singular points of the asymptotic
curves.

10.1.4 Semiumbilicity
Recall from Definition 7.2 that a point p on M ⇢ R4 is semiumbilic if the
curvature ellipse at p is a line segment that contains p. It can be shown
that the asymptotic directions of M are orthogonal at semiumbilic points.
Also, a point p is semiumbilic if and only if it is a ⌫-umbilic point of some
 338 Di↵erential Geometry from a Singularity Theory Viewpoint

normal field on M defined locally at p.

Let N (p) be the curvature of the normal bundle of M at p (see §7.1). In
fact, the area of the curvature ellipse of M at p is equal to 12 |N (p)| ([Little
(1969)]). We have the following characterisation of semiumbilic points.

Theorem 10.4. Let M be a smooth surface in R4 and let p be a point on

M . Then the following assertions are equivalent

(i) p is semiumbilic.
(ii) N (p) = 0.
(iii) p is ⌫-umbilic for some normal field ⌫ on M .
(iv) There are two orthogonal asymptotic directions at p.

It is shown in [Bancho↵ and Farris (1993)] that a surface in R4 is

orientable if and only if it admits some globally defined normal field. As a
consequence, and using the Poincaré-Hopf formula, we obtain the following
corollary.

Corollary 10.6. Any embedded orientable closed surface in R4 with non-

vanishing Euler characteristic has semiumbilic points.

Little proved in [Little (1969)] that any embedding of the torus in R4

has semiumbilic points. We have thus the following.

Corollary 10.7. Any embedded orientable closed surface in R4 has semi-

umbilic points.

By Theorem 10.4, Corollary 10.7 is equivalent to stating that there does

not exist any closed orientable surface embedded in R4 with never vanishing
normal curvature.
A surface whose points are all semiumbilic is said to be a totally semi-
umbilic surface. An example of a totally semiumbilic surface is a surface
lying in a 3-sphere in R4 . It is natural to search for sufficient conditions on
a totally semiumbilic surface to be hyperspherical, i.e., to lie in a 3-sphere.
A totally semiumbilic surface M has two binormal fields b1 and b2
which are globally defined. Their singular points are the inflection points
and the umbilic points considered as a particular case of inflection points
of M . The binormal fields are characterised by the fact that one of their
two principal curvatures vanishes identically on M ; the other principal
curvature we call binormal curvature of M . We denote by 1 and 2 the
two binormal curvatures associated to b1 and b2 respectively of a totally
 Global viewpoint 339

semiumbilical surface M . The following result gives necessary and sufficient

conditions on a totally semiumbilical surface to lie in a 3-sphere.

Theorem 10.5 ([Romero Fuster and Sánchez-Bringas (2002)]).

Suppose that M is a surface with isolated inflection points in R4 . Then
M is hyperspherical if and only if its asymptotic curves are globally defined
and orthogonal and its binormal curvatures 1 , 2 satisfy the following re-
lation
1 2
( + + 2 cos ⌦)E = constant,
2 1
where ⌦ is the angle between the two binormals at each point and E is the
coefficient of the first fundamental form of M in isothermal coordinates.

10.2 Spacelike submanifolds of Minkowski space-time

We introduced new invariants in Chapter 9 as applications of singularity

theory to di↵erential geometry of submanifolds of Minkowski space-time.
We prove here a Gauss-Bonnet and Chern-Lashof type theorems for closed
spacelike surfaces M in Minkowski space-time R41 using those invariants.
Consider the canonical projection ⇡ from R41 to the Euclidean space R30
given by
⇡(x0 , x1 , x2 , x3 ) = (0, x1 , x2 , x3 ),
and denote by M the image of M by ⇡. Since M is spacelike, ⇡|M : M ! R30
is an immersion. If M is orientable, then so is M . In this case we have a
globally defined Gauss map N : M ! S 2 on M .
Recall from Definition 9.3, that we have the normalised lightcone Gauss
map L e : U ! S 2 and the normalised lightcone Gauss-Kronecker curvature
+
e e
K` associated to L.
e and N are
Lemma 10.1. Under a suitable choice of direction of N , ⇡ L
homotopic.

Proof. By Lemma 9.2, we have ⇡ L(p) e / d(⇡|M )p (Tp M ) ⇢ R30 . There-

2
e
fore h⇡ L(p), 6 0 at any p 2 M. We choose the direction of N that
N (p)i =
e
makes h⇡ L(p), N (p)i > 0 and construct a homotopy F : M ⇥ [0, 1] ! S 2
between ⇡ Le and N as follows:
e
tN (p) + (1 t)⇡ L(p)
F (p, t) = ,
e
ktN (p) + (1 t)⇡ L(p)k
340 Di↵erential Geometry from a Singularity Theory Viewpoint

where k·k is the Euclidean norm in R30 . Suppose that there exists t0 2 (0, 1)
and p0 2 M such that

t0 N (p0 ) + (1 e 0 ) = 0.
t0 )⇡ L(p

e 0 ). But this contradicts

Then we would have N (p0 ) = ((1 t0 )/t0 )⇡ L(p
the assumption that h⇡ L(p), N (p)i > 0. Therefore, F is a continuous
e
mapping satisfying F (p, 0) = ⇡ L(p) and F (p, 1) = N (p) at any point p
on M . ⇤

The degree deg N of the map N is a homotopy invariant and satisfies

deg N = (1/2) (M ) (see for example [Guillemin and Pollack (1974)]). A
consequence of this and of Lemma 10.1 is the following result.

Corollary 10.8. Let M be a closed orientable spacelike surface in R41 .

Then

e= 1
deg L (M ).
2

Proposition 10.1. Denote by dvM (respectively, dvS+2 ) the volume form

2
of M (respectively, S+ ). Then the following relation holds:

e ⇤ dvS 2 ,
e ` dvM = L
K +

e ⇤ dvS 2 is the pull back by L

where L e of the di↵erential form dvS 2 .
+ +

Proof. Take a local parametrisation x : U ! R41 of M , where U is

an open subset of R2 with coordinates (u1 , u2 ). Assume first that the
normalised lightcone Gauss map L e is not singular at p = x(u). In this
case, there exists an open neighbourhood W ⇢ U containing p such that
e : W ! S 2 is an embedding. Therefore, {L
L eu , L
e u } is a basis of Tz S 2
+ 1 2 +
e
at any point z 2 V = L(W ). We denote by geij the Riemannian metric
on V and by g↵ the Riemannian metric on W given by the restriction
of the Minkowski metric. Since L = `0 L e (see Chapter 9 for notation),
e u = Lu
`0 L e By Proposition 9.2, we have
`0ui L.
i i
 Global viewpoint 341

eu , L
geij = hL e u i1
i j

1
= 2 hLui , Luj i1
`0
X2 2
X
1
= 2h h↵
i x u↵ , hi x u i 1
`0 ↵=1 =1
2
1 X ↵
= 2 hi hj hxu↵ , xu i1
`0
↵, =1
2
1 X ↵
= 2 hi hj g ↵ .
`0
↵, =1

From the definition of K e ` and the proof of Corollary 9.2 we get K

e` =
(1/`0 )2 det(hij ) and therefore
det(e e `2 det(g↵ ).
gij ) = K
Denote by (e e2 ) the local coordinates on V via the embedding L.
u1 , u e
This means that the pull-back L e ⇤ (de
u1 ^ de u1 ^ de
u2 ) of the volume form de u2
e
in V by L satisfies
(
du1 ^ du2 if K e ` (u) > 0,
e ⇤ (de
L u1 ^ de u2 ) =
du1 ^ du2 if K e ` (u) < 0.

Here (u1 , u2 ) is a local parameter of W ⇢ M and (e e2 ) is a local

u1 , u
parameter of V ⇢ S+ via L
2 e : W ! V , that is, L(u
e 1 , u2 ) = (e e is
e2 ). If L
u1 , u
orientation preserving, L e ⇤ (de
u1 ^de e ⇤ (de
u2 ) = du1 ^du2 , otherwise L u1 ^de u2 ) =
du1 ^ du2 . By definition of the volume form, we have
q q
dvV = det(e gij )deu1 ^ de
u2 , dvW = det(g↵ )du1 ^ du2 ,
so that
K e ⇤ dvV .
e ` dvW = L
e then both sides of the equality are zero.
If p is a singular point of L,
This completes the proof. ⇤
We have now all the ingredients to prove the following theorem.

Theorem 10.6 (The Lightcone Gauss-Bonnet Theorem). Let M be

a closed orientable spacelike surface in Minkowski space-time R41 . Then
Z
e ` dvM = 2⇡ (M ).
K
M
342 Di↵erential Geometry from a Singularity Theory Viewpoint

Proof. By Proposition 10.1 and Corollary 10.8 we have

Z Z Z
Ke ` dvM = e ⇤ dvS 2 = deg (L)
L e e
dvS+2 = deg (L)4⇡ = 2⇡ (M ).
⇤
+ 2
M M S+

An immediate consequence of the Lightcone Gauss-Bonnet Theorem is

that the total lightcone Gauss-Kronecker curvature is a topological invari-
ant. However, the lightcone Gauss-Kronecker curvature is not invariant
under the Lorentz group. In fact, it is an SO(3)-invariant where SO(3) acts
canonically on R41 as a subgroup of SO0 (1, 3).
If we consider the total absolute lightcone curvature, we obtain the
following Chern-Lashof type inequality.

Theorem 10.7 (The Lightcone Chern-Lashof Type Theorem).

The following inequality holds for any closed orientable spacelike surface
M in Minkowski space-time R41 :
Z
e ` |dvM 2⇡(4
|K (M )).
M

Proof. We define the two subsets M + = {p 2 M | K e ` (p) > 0} and

M = {p 2 M | Ke ` (p) < 0} of M . Then we can write
Z Z Z
e
|K` |dvM = e
K` dvM e ` dvM
K
M M+ M
and Z Z Z
e ` dvM =
K e ` dvM +
K e ` dvM .
K
M M+ M
By Theorem 10.6 and the above equalities, we get
Z Z
e ` |dvM = 2
|K e ` dvM 2⇡ (M ).
K
M M+
Thus, it is enough to show that
Z
e ` dvM
K 4⇡.
M+

Let M0 , M1 , M2 , M2+ be the subsets of M defined by M0 = (K e ` ) 1 (0),

M1 = {p 2 M \ M0 | 9q 2 M0 with L(q) e e
= L(p) }, M2 = M \ (M0 [ M1 )
+
and M2 = M \ M2 . Since M0 is the singular set of L,
+ e we have by Sard’s
e 0 ) and L(M
Theorem that L(M e 0 ) [ L(M
e 1 ) are measure zero sets in S 2 . For
+
e e
any v 2 S \ (L(M0 ) [ L(M1 )), the lightcone height function hv has at least
2
+
two critical points: a maximum and a minimum. We know that
e ` (p) = det Hess (hv (p)) ,
K
det (gij (p))
 Global viewpoint 343

e
for v = L(p). e hv has a Morse-type singular
Since v is a regular value of L,
point of index 0 or 2 at the minimum and maximum points. The light-
cone Gauss-Kronecker curvarture K e +
e ` is positive at such points, so L|M
is surjective. Since the area of the unit sphere is 4⇡, we have the desired
inequality by Proposition 10.1. ⇤

Remark 10.4. The classical Gauss-Bonnet Theorem (see for example

[do Carmo (1976)]) and the Chern-Lashof Theorem ([Chern and Lashof
(1957)]) for surfaces in Euclidean space can be obtained as particular cases
of Theorems 10.6 and 10.7 by taking nT defining L as a constant timelike
unit vector.

10.3 Notes

Curves. We have not touched on the subject of plane curves in this book.
However, there are some global properties of plane and space curves that
are worth mentioning here. We start with plane curves and the celebrated
4-Vertex Theorem which states that any smooth closed simple curve in
Euclidean plane has at least 4 vertices. An analogous result for curves in the
Minkowski plane R21 is proved in [Tabachnikov (1997)]. Using stereographic
projections H+ 2
( 1) ! R2 and S12 ! R2 , one can show that an analogous
2
result of the 4-vertex Theorem is true for curves in H+ ( 1) and in S12 .
Results of global nature on space curves obtained by similar methods to
those explored in this book concerning the number of tri-tangent planes, bi-
tangent osculating planes and torsion zero points can be found in [Bancho↵,
Ga↵ney and McCrory (1985); Freedman (1980); Nuño Ballesteros and
Romero Fuster (1992); Sedykh (1989); Ozawa (1985)].
It is worth observing that the case of spacelike embeddings of a circle
(i.e., spacelike knots) in R31 is di↵erent from ordinary knots R3 . There are,
for instance, many cases of closed spacelike curves which are un-knotted in
the ordinary sense ([Izumiya, Kikuchi and Takahashi (2006)]).
Umbilics on surfaces in R31 and H+ 3
( 1). It is shown in [Tari (2013)]
that any closed and convex surface in Minkowski 3-space R31 of class C 3
has at least two umbilic points. For ovaloids (i.e., strictly convex surfaces)
of class C 3 , the umbilic points all lie in the Riemannian part of the surface
and there are at least two of them.
Using the stereographic projection and Feldman result for generic sur-
faces in R3 ([Feldman (1967)]), it can be shown that the number of light-
344 Di↵erential Geometry from a Singularity Theory Viewpoint

cone umbilic points (singularities of the lightcone lines of curvature) of a

3
closed surface M generically immersed in H+ ( 1) is greater or equal than
2| (M )|. Furthermore, if M is homeomorphic to a 2-sphere, then it has at
least 4 lightcone umbilic points.
Characterisation of metric spheres in Minkowski space-time. Morse the-
ory played a principal role in di↵erential topology. A Morse function on
f : M ! R induces a decomposition of M as a CW-complex whose cell
types are determined by the distribution and the indices of critical points
(see for instance [Hirsch (1973); Milnor (1963); Matsumoto (1997)]). A well
known consequence of this decomposition is the theorem of Reeb ([Reeb
(1946)]), which says that any closed (compact without boundary) manifold
that admits a Morse function with only two critical points is homeomorphic
to a sphere. A natural question related to Reeb’s theorem is to characterise
the submanifolds in Euclidean or Lorentz-Minkowski spaces which are met-
ric spheres. This problem is studied by Nomizu and Rodrı́guez in the case
of submanifolds of Euclidean space. Let f : M n ! Rn be an immersion.
They proved in [Nomizu and Rodrı́guez (1972)] that if M n is connected
and complete in the induced Riemannian metric, and almost every distance
squared function on M n has index 0 or n at each of its critical points, then
f embeds M n either as an Euclidean n-sphere or as a flat affine subspace.
The analogous problem for submanifolds of Hyperbolic 3-space was
treated by Cecil and Ryan. Given a surface M in R31 , Cecil ([Cecil (1974)])
3
proved that a connected compact surface M immersed in H+ ( 1) is a met-
ric sphere if and only if every Morse function of type timelike has exactly two
singularities. Later, Cecil and Ryan ([Cecil and Ryan (1979a)]) considered
the three families of height functions describing respectively the contact of a
surface with spacelike, timelike and lightlike hyperplanes in R31 and proved
3
that a connected complete surface M of H+ is either a sphere, a horosphere,
or an equidistant plane if and only if every Morse function of one of the
above 3 types on M has index 0 or 2 at all its critical points. These results
are extended in [Izumiya, Nuño Ballesteros and Romero Fuster (2010)] for
closed surfaces in R41 that admit a globally defined non-degenerate parallel
normal field.
Gauss-Bonnet and Chern-Lashof Theorems. For surfaces in the hy-
perbolic space, we have the Horospherical Gauss-Bonnet Theorem ([Izu-
miya and Romero Fuster (2006)]) and the Horospherical Chern-Lashof
Type Theorem ([Buosi, Izumiya and Ruas (2010)]). The Lightcone Gauss-
Bonnet Theorem holds for codimension two closed and orientable spacelike
 Global viewpoint 345

submanifolds in Minkowski space-time Rn1 , n 4 ([Izumiya and Romero

Fuster (2007)]). The Lightcone Chern-Lashof Type Theorem for spacelike
submanifolds of higher codimension in Rn1 , n 4, holds ([Izumiya (2014)]).
We point out that this is a generalisation of both the Chern-Lashof type
theorem for horospherical curvatures in hyperbolic space ([Buosi, Izumiya
and Ruas (2010)]) and the original Chern-Lashof theorem for Euclidean
space ([Chern and Lashof (1957)]). An interesting consequence is the intro-
duction of the notion of lightcone tightness which generalises the concepts
of horotightness in hyperbolic space ([Buosi, Izumiya and Ruas (2011); Ce-
cil and Ryan (1979b)]) and tightness in Euclidean space ([Cecil and Ryan
(1985)]).
kth-regular immersions. The concept of kth-regular immersions of a
submanifold in Euclidean space was introduced in terms of maps between
osculating bundles in [Pohl (1962)] and [Feldman (1965)]. In the case
of curves, the k-regularity condition means that the first k derivatives are
linearly independent. The existence of 3-regular embedded closed space
curves was investigated in [Costa (1990)]. Corollary 10.2 and its gen-
eralisations show that convexity is an obstruction for the 3-regularity of
embedded closed space curves.
A point p of a surface M in Rn is 2-regular if and only if, in some
coordinates near p, the subset of vectors determined by the first and second
derivatives of the immersion at p has maximal rank. Otherwise, p is 2-
singular. Feldman ([Feldman (1965)]) proved that the set of 2-regular
immersions of any closed surface M in Rn is open and dense for n = 3 and
n 7. The 2-singular points of surfaces in R4 coincide with the inflection
points.
The 2-regular points of a surface in R5 coincide with the M2 points
(Proposition 8.5). Examples of 2-regular immersions of closed orientable
surfaces with non-zero genus are not known so far. On the other hand, it
was shown in [Romero Fuster (2007)] that there are no 2-regular embed-
dings of orientable closed surfaces immersed in S 4 .
Thom polynomials and Chern classes. We pointed out in the Notes
of Chapter 3 that there is a rich branch of research on global topological
invariants using Thom polynomials, see for example Ohmoto lecture notes
[Ohmoto (2013)] and [Ando (1996); Kazarian (2006, 2001)].
Topological approach. Since Morse functions are stable mappings whose
target manifold is R, it was natural to seek to extend Morse theory to stable
mappings with higher dimensional target manifolds, see for example [Saeki
346 Di↵erential Geometry from a Singularity Theory Viewpoint

(1996, 2004); Lippner and Szucs (2010); Szabó, Szucs and Terpai, (2010)].
Other approaches to global problems using Vassiliev type invariants and
h-principal can be found, for example, in [Ando (1985, 2007a,b); Eliashberg
and Mishachev (2002); du Plessis (1976a,b); Gromov (1986); Goryunov
(1998); Yamamoto (2006); Ohmoto and Aicardi (2006)].
 Bibliography

Abraham, R. (1963), Transversality in manifolds of mappings. Bull. Amer. Math.

Soc. 69, pp. 470-474.
Ando, Y. (1985), On the elimination of Morin singularities. J. Math. Soc. Japan
37, pp. 471–487.
Ando, Y. (1996), On Thom polynomials of the singularities Dk and Ek . J. Math.
Soc. Japan 48, pp. 593–606.
Ando, Y. (2007a), A homotopy principle for maps with prescribed Thom-
Boardman singularities. Trans. Amer. Math. Soc. 359, pp. 489–515.
Ando, Y. (2007b), The homotopy principle for maps with singularities of given
K-invariant class. J. Math. Soc. Japan 59, pp. 557–582.
Arbarello, E., Cornalba, M., Griffiths, P. A. and Harris, J. (1985), Geometry of
algebraic curves. Vol. I., Grundlehren der Mathematischen Wissenschaften
[Fundamental Principles of Mathematical Sciences], 267.
Arnol’d, V. I. (1979), Indices of singular points of 1-forms on a manifold with
boundary, convolution of invariants of reflection groups, and singularities of
projections of smooth surfaces. Uspekhi Mat. Nauk 34:2, pp. 3-38 (Russian
Math. Surveys 34:2, pp. 1-42).
Arnol’d, V. I. (1981), Lagrangian manifolds with singularities, asymptotic rays,
and the open swallowtail. Funct. Anal. Appl. 15-4, pp. 235–246.
Arnol’d, V. I. (1983), Singularities of systems of rays. Russian Math. Surveys 38,
pp. 87–176.
Arnol’d, V. I. (1990), Singularities of caustics and wave fronts. Mathematics and
its Applications (Soviet Series), 62. Kluwer Academic Publishers Group.
Arnol’d,V. I., Guseı̆n-Zade S. M. and Varchenko A. N. (1985), Singularities of
di↵erentiable maps. Vol. I. The classification of critical points, caustics and
wave fronts. Monographs in Mathematics, 82. Birkhäuser.
Atique, R. G. W. (2000), On the classification of multi-germs of maps from C2
to C3 under A-equivalence. Chapman & Hall/CRC Res. Notes Math. 412,
pp. 119–133. Chapman & Hall/CRC, Boca Raton, FL.
Bancho↵, T. and Farris, F. (1993), Tangential and normal Euler numbers, com-
plex points, and singularities of projections for oriented surfaces in four-
space. Pacific J. Math. 161, pp. 1-24.

347
348 Di↵erential Geometry from a Singularity Theory Viewpoint

Bancho↵, T., Ga↵ney, T. and McCrory, C. (1982), Cusps of Gauss mappings. Re-
search Notes in Mathematics 55. Pitman (Advanced Publishing Program),
Boston, Mass.-London.
Bancho↵, T., Ga↵ney, T. and McCrory, C. (1985), Counting tritangent planes of
space curves. Topology 24, pp. 15–23.
Bancho↵, T. and Thom, R. (1980), Erratum et compléments: “Sur les points
paraboliques des surfaces” by Y. L. Kergosien and Thom. C. R. Acad. Sci.
Paris Sér. A-B 291, A503–A505.
Basto-Gonçalves, J. (2013), Local geometry of surfaces in R4 . Preprint, arXiv :
1304.2242.
Binotto, R. (2008), Projetivos de curvatura. Doctoral Thesis, Instituto de
Matemática, Estatı́stica e Computação Cientı́fica, UNICAMP (Brasil).
Binotto, R, Costa, S. I. R. and Romero Fuster, M. C. (2015), Geometry of 3-
manifolds in Euclidean space. To appear in RIMS Kôkyûroku Bessatsu.
Birbrair, L. (2007), Metric theory of singularities. Lipschitz geometry of singular
spaces. Singularities in geometry and topology, pp. 223–233. (World Sci.
Publ., Hackensack, NJ).
Blair, D. E. (1976), Contact manifolds in Riemannian geometry, Lecture Notes
in Mathematics, 509, Springer-Verlag, Berlin.
Bleeker, D. and Wilson, L. (1978), Stablility of Gauss maps, Illinois J. Math. 22,
pp. 279–289.
Brieskorn, E. and Knörrer, H (1986), Plane algebraic curves. Birkhäuser.
Bröcker, Th. (1975), Di↵erentiable germs and catastrophes. London Mathematical
Societry Lecture Notes 17 (Cambridge University Press).
Bruce, J. W. (1981), The duals of generic hypersurfaces; Math. Scand. 49, pp.
36–60.
Bruce, J. W. (1984), Generic reflections and projections; Math. Scand. 54, pp.
262–278.
Bruce, J. W. (1986), On transversality. Proc. Edinburgh Math. Soc. 29, pp. 115-
123.
Bruce, J. W. (1994a), Generic geometry, transversality and projections. J. London
Math. Soc. 49, pp. 183-194
Bruce, J. W. (1994b), Lines, circles, focal and symmetry sets. Math. Proc. Cam-
bridge Philos. Soc. 118, pp. 411–436.
Bruce, J. W. and Fidal, D. (1989), On binary di↵erential equations and umbilics.
Proc. Royal Soc. Edinburgh,111A, pp. 147–168.
Bruce, J. W., Fletcher, G. J. and Tari, F. (2004), Zero curves of families of curve
congruences. Contemp. Math.,354, Amer. Math. Soc. Amer. Providence,
RI, pp. 1–18.
Bruce, J. W. and Giblin, P. J. (1990), Projections of surfaces with boundary.
Proc. London Math. Soc. 60, pp. 392–416.
Bruce, J. W. and Giblin, P. J. (1992), Curves and Singularities. Cambridge Uni-
versity Press.
Bruce, J. W., Giblin, P. J. and Gibson, C. G. (1985), Symmetry sets. Proc. Roy.
Soc. Edinburgh Sect. A 101, pp. 163–186.
Bruce, J. W., Giblin, P. J. and Tari, F. (1995), Families of surfaces: height func-
 Bibliography 349

tions, Gauss maps and duals. Real and complex singularities (São Carlos,
1994). Pitman Res. Notes Math. Ser. (Longman, Harlow) 333, pp. 148–178.
Bruce, J. W., Giblin, P. J. and Tari, F. (1998), Families of surfaces: height
functions and projections to plane. Math. Scand. 82, pp. 165–185.
Bruce, J. W., Giblin, P. J. and Tari, F. (1999), Families of surfaces: focal sets,
ridges and umbilics. Math. Proc. Camb. Phil. Soc 125, pp. 243–268.
Bruce, J. W., Kirk N. P. and du Plessis A. A. (1997), Complete transversals and
the classification of singularities. Nonlinearity 10, pp. 253-275.
Bruce, J. W. and Nogueira, A. C. (1998), Surfaces in R4 and duality. Quart. J.
Math. Oxford Ser. 49, pp. 433–443.
Bruce, J. W., du Plessis, A. A. and Wall, C. T. C. (1987), Determinacy and
unipotency. Invent. Math. 88, pp. 521–554.
Bruce, J. W. and Romero Fuster, M. C. (1991), Duality and orthogonal projec-
tions of curves and surfaces in Euclidean 3-space. Quart. J. Math. 42, pp.
433–441.
Bruce, J. W. and Tari, F. (1995), On binary di↵erential equations. Nonlinearity
8, pp. 255–271.
Bruce, J. W. and Tari, F. (2000), Duality and implicit di↵erential equations.
Nonlinearity 13, pp. 791–811.
Bruce, J. W. and Tari, F. (2002), Families of surfaces in R4 . Proc. Edinb. Math.
Soc. 45, pp. 181–203.
Bruce, J. W. and Tari, F. (2005), Dupin indicatrices and families of curve con-
gruences. Trans. Amer. Math. Soc. 357, pp. 267–285.
Bruce, J. W. and West, J. M. (1998), Functions on a crosscap. Math. Proc.
Cambridge Philos. Soc. 123, pp. 19–39.
Bruce, J. W. and Wilkinson, T. C. (1991), Folding maps and focal sets. Pro-
ceedings of Warwick Symposium on Singularities, Springer Lecture Notes
in Math., Vol. 1462, pp. 63–72, Springer-Verlag, Berlin and New York.
Bryant, R. L. (1987), Surfaces of mean curvature one in hyperbolic space. in
Théorie des variétés minimales et applications (Palaiseau, 1983–1984),
Astérisque No. 154–155, 12, pp. 321–347, 353 (1988).
Buchner, M.A. (1974), Stability of the cut locus. Doctoral thesis, Harvard Uni-
versity.
Buosi, M., Izumiya, S. and Ruas, M. A. S. (2010), Total absolute horospherical
curvature of submanifolds in hyperbolic space. Adv. Geom. 10, pp. 603–620.
Buosi, M., Izumiya, S. and Ruas, M. A. S. (2011), Horo-tight spheres in hyperbolic
space. Geom. Dedicata 154, pp. 9–26.
do Carmo, M. P. (1976), Di↵erential geometry of curves and surfaces. Prentice-
Hall, 1976.
do Carmo, M. P. (1992), Riemannian geometry. Translated from the second Por-
tuguese edition by Francis Flaherty Mathematics: Theory & Applications.
Birkhuser Boston, Inc., Boston, MA.
Cecil, Th. E. (1974), A characterization of metric spheres in hyperbolic space by
Morse theory. Tohoku Math. J. 26, pp. 341–351.
Cecil, Th. E. (1980), Lie Sphere Geometry with Applications to Submanifolds.
Universitext, Springer-Verlag, New York Berlin.
350 Di↵erential Geometry from a Singularity Theory Viewpoint

Cecil, Th. E. and Ryan, P. J. (1979a), Distance functions and umbilic submani-
folds of hyperbolic space. Nagoya Math. J. 74, pp. 67–75.
Cecil, Th. E. and Ryan, P. J. (1979b), Tight ant taut immersions into hyperbolic
space. J. Lond. Math. Soc. 19, pp. 561–572.
Cecil, Th. E. and Ryan, P. J. (1985), Tight and taut immersions of manifolds.
Research Notes in Mathematics, 107. Pitman (Advanced Publishing Pro-
gram), Boston, MA.
Chen, L. and Izumiya, S. (2009), A mandala of Legendrian dualities for pseudo-
spheres in semi-Euclidean space Proc. Japan Acad. 85, pp. 49–54.
Cheng, S. -Y. and Yau, S.-T. (1973), Maximal space-like hypersurfaces in the
Lonrentz-Minkowski spaces. Ann. of Math. 104, pp. 407–419.
Chern, S. and Lashof, R. K. (1957), On the total curvature of immersed manifolds.
Amer. J. Math. 79, pp. 306–318.
Cleave, J. P. (1980), The form of the tangent developable at points of zero torsion
on space curves. Math. Proc. Camb. Phil. 88, pp. 403–407.
Costa, S. I. R. (1990), On closed twisted curves. Proc. Amer. Math. Soc. 109,
pp. 205–214.
Costa, S. I. R. and Romero Fuster, M. C. (1997), Nowhere vanishing torsion
closed curves always hide twice. Geom. Dedicata 66, pp. 1–17.
Costa, S. I. R., Moraes, S. M. and Romero Fuster, M. C. (2009), Geometric
contacts of surfaces immersed in Rn , n 5. Di↵erential Geom. Appl. 27,
pp. 442–454.
Damon J. N. (1984), The unfolding and determinacy theorems for subgroups of
A and K. Mem. Amer. Math. Soc. 50, No. 306.
Damon J. N. (1988), Topological triviality and versality for subgroups of A and
K. Mem. Amer. Math. Soc. 75, No. 389.
Damon J. N. (1992), Topological triviality and versality for subgroups of A and
K. II. Sufficient conditions and applications. Nonlinearity, 5, pp. 373–412.
Damon J. N. (2003), Smoothness and geometry of boundaries associated to skele-
tal structures. I. Sufficient conditions for smoothness. Ann. Inst. Fourier
(Grenoble) 53, pp. 1941–1985.
Damon J. N. (2004), Smoothness and geometry of boundaries associated to skele-
tal structures. II. Geometry in the Blum case. Compos. Math. 140, pp.
1657–1674.
Damon J. N. (2006), The global medial structure of regions in R3 . Geom. Topol.
10, pp. 2385–2429.
Dara, L. (1975), Singularités génériques des équations di↵erentielles multiformes.
Bol. Soc. Brasil Math. 6, pp. 95–128.
David, J. M. S. (1983), Projection-generic curves. J. London Math. Soc. 27, pp.
552–562.
Davydov, A. A. (1994), Qualitative control theory. Translations of Mathematical
Monographs, 142. AMS, Providence, RI.
Dias, F. S. and Nuño Ballesteros, J. J. (2008), Plane curve diagrams and geomet-
rical applications. Q. J. Math. 59, pp. 287–310.
Dreibelbis, D. (2001), Bitangencies on surfaces in four dimensions. Quart. J.
Math. 52, pp. 137–160
 Bibliography 351

Dreibelbis, D. (2004), Invariance of the diagonal contribution in a bitangency

formula. Real and complex singularities, pp. 45–56, Contemp. Math., 354,
Amer. Math. Soc., Providence, RI.
Dreibelbis, D. (2006), Birth of bitangencies in a family of surfaces in R4 . Di↵er-
ential Geom. Appl. 24, pp. 321–331.
Dreibelbis, D. (2007), The geometry of flecnodal pairs. Real and complex singu-
larities, pp. 113-126, Trends Math., Birkhäuser, Basel.
Dreibelbis, D. (2012), Self-conjugate vectors of immersed 3-manifolds in R6 .
Topology Appl. 159 (2012), pp. 450–456.
Eliashberg, Y. and Mishachev, N. (2002), Introduction to the h-principle. Grad-
uate Studies in Mathematics, 48. American Mathematical Society, Provi-
dence,RI.
Epstein, C. L. (1986), The hyperbolic Gauss map and quasiconformal reflections.
J. Reine Angew. Math. 372, pp. 96–135
Feldman, E. A. (1965), Geometry of immersions I. Trans. AMS 120, pp. 185–224.
Feldman, E. A. (1967), On parabolic and umbilic points of immersed hypersur-
faces. Trans. Amer. Math. Soc. 127, pp. 1–28.
Fletcher, G. J. (1996), Geometrical problems in computer vision. Ph.D thesis,
Liverpool University.
Freedman, M. H. (1980), Planes triply tangent to curves with non-vanishing tor-
sion. Topology 19, pp. 1–8.
Ga↵ney, T. (1983), The structure of the T A(f ), classification and application
to di↵erential geometry. Singularities, Part 1 (Arcata, Calif., 1981), pp.
409–427, Proc. Sympos. Pure Math. 40, Amer. Math. Soc., Providence,
RI.
Ga↵ney, T. and Massey, D. (1999), Trends in equisingularity theory. Singularity
theory (Liverpool, 1996), xix–xx, pp. 207–248, London Math. Soc. Lecture
Note Ser., 263. (Cambridge Univ. Press, Cambridge).
Ga↵ney, T. and Ruas, M. A. S. (1979), Singularities of mappings and orthogonal
projections of spaces. Notices Am. Math. Soc. 26, no.1, January 1979, 763-
53-23.
Garcia, R., Mello, L. F. and Sotomayor, J. (2005), Principal mean curvature
foliations on surfaces immersed in R4 . EQUADIFF 2003, pp. 939-950, World
Sci. Publ., Hackensack, NJ.
Garcia, R., Mochida, D. K. H., Romero Fuster, M. C. and Ruas, M. A. S. (2000),
Inflection points and topology of surfaces in 4-space. Trans. Amer. Math.
Soc. 352, pp. 3029-3043.
Garcia, R. and Sotomayor, J. (1997), Structural stability of parabolic points and
periodic asymptotic lines. Mat. Contemp. 12, pp. 83–102.
Garcia, R. and Sotomayor, J. (2000), Lines of axial curvature on surfaces im-
mersed in R4 . Di↵erential Geom. Appl. 12, pp. 253–269.
Genin, D., Khesin, B. and Tabachnikov, S. (2007), Geodesics on an ellipsoid in
Minkowski space. Enseign. Math. 53 , pp. 307–331.
Giblin, P. J. and Brassett, S. A. (1985), Local symmetry of plane curves. Amer.
Math. Monthly 92, pp. 689–707.
Giblin, P. J. and Holtom, P. (1999), The centre symmetry set. Banach Center
352 Di↵erential Geometry from a Singularity Theory Viewpoint

Publ., Polish Acad. Sci. Inst. Math., Warsaw 50, pp. 91–105.
Giblin, P. J. and Janeczko, S. (2012), Geometry of curves and surfaces through
the contact map. Topology Appl. 159, pp. 466–475.
Giblin, P. J. and Weiss, R. (1987), Reconstruction of surfaces from profiles. Proc.
First Internat. Conf. on Computer Vision; Computer Society of the IEEE,
pp. 136–144.
Giblin, P. J. and Zakalyukin, V. M. (2005), Singularities of centre symmetry sets.
Proc. London Math. Soc. 90, pp. 132–166.
Gibson, C. G. (1979), Singular points of smooth mappings. Research Notes in
Mathematics, 25. Pitman (Advanced Publishing Program), Boston, Mass.-
London.
Gibson, C. G., Hawes, W. and Hobbs, C. A. (1994), Local pictures for general
two-parameter planar motions. Advances in robot kinematics and compu-
tational geometry (Ljubljana), Kluwer Acad. Publ., Dordrecht, pp. 49–58.
Gibson, C. G., Wirthmüller, K., du Plessis, A. A. and Looijenga, E. J. N. (1976),
Topological stability of smooth mappings. Lecture Notes in Mathematics,
552. Springer-Verlag, Berlin-New York,
Golubitsky, M. and Guillemin, V. (1973), Stable mappings and their singularities
GTM, 14. Springer-Verlag, New York.
Golubitsky, M. and Guillemin, V. (1975), Contact equivalence for Lagrangian
manifold, Adv. Math. 15, pp. 375–387.
Golubitsky, M. and Schae↵er, D. G. (1985), Singularities and groups in bifurcation
theory. Vol. I. Applied Mathematical Sciences, 51. Springer-Verlag, New
York.
Gonçalves, R. A., Martı́nez Alfaro, J. A., Montesinos Amilibia, A. and Romero
Fuster, M. C. (2007), Relative mean curvature configurations for surfaces
in Rn , n 5. Bull. Braz. Math. Soc. New Series 38(2), pp. 1–22.
Goryunov, V. V. (1981a), Surface projection singularities. Ph. D. thesis, Moscow’s
Lomonosov State University.
Goryunov, V. V. (1981b), Geometry of bifurcation diagrams of simple projections
on a line. Funktsional. Anal. i Prilozhen 15, pp. 1–8, 96.
Goryunov, V. V. (1990), Projections of Generic Surfaces with Boundaries. Adv.
Soviet Math. 1, pp. 157–200.
Goryunov, V. V. (1998), Vassiliev type invariants in Arnold’s J + -theory of plane
curves without direct self-tangencies. Topology 37, pp. 603-620
Gromov, M. (1986), Partial di↵erential relations. Ergebnisse der Mathematik und
ihrer Grenzgebiete (3) [Results in Mathematics and Related Areas (3)], 9.
Springer-Verlag, Berlin.
Guillemin, V. and Pollack, A. (1974), Di↵erential topology. Prentice-Hall, Inc.,
Englewood Cli↵s, N.J.
Gutierrez, C., Guadalupe, I., Tribuzy, R. and Guı́ñez V. (1997), Lines of curvature
on surfaces immersed in R4 . Bol. Soc. Brasil. Mat. (N.S.) 28, pp. 233–251.
Gutierrez, C., Guadalupe, I., Tribuzy, R. and Guı́ñez, V. (2001), A di↵erential
equation for lines of curvature on surfaces immersed in R4 . Qual. Theory
Dyn. Syst., pp. 207–220.
Gutierrez, C. and Guı́ñez, V. (2003), Simple umbilic points on surfaces immersed
 Bibliography 353

in R4 . Discrete Contin. Dyn. Syst. 9, pp. 877–900.

Gutiérrez, C. and Ruas, M. A. S. (2003), Indices of Newton non-degenerate vector
fields and a conjecture of Loewner for surfaces in R4 . Real and complex
singularities, pp. 245–253, Lecture Notes in Pure and Appl. Math., 232,
Dekker, New York.
Gutierrez, C. and Sánchez-Bringas, F. (1998), On a Loewner umbilic-index con-
jecture for surfaces immersed in R4 . J. Dynam. Control Systems 4, pp.
127–136.
Gutierrez, C. and Sotomayor, J. (1998), Lines of curvature, umbilic points and
Carathéodory conjecture. Resenhas 3, pp. 291–322.
Hilbert, D. and Cohen-Vossen, S. (1932), Anschauliche Geometrie. Julius
Springer, Berlin. Geometry and the Imagination. Chelsea Publishing Com-
pany, NY, 1952.
Hirsch, M. W. (1976), Di↵erential topology. Graduate Texts in Mathematics, 33.
Springer-Verlag, New York-Heidelberg.
Hörmander, L. (1971), Fourier integral operaators I. Acta. Math. 127, pp. 71–183.
Houston, K. (1999), Images of finite maps with one-dimensional double point set.
Topology Appl. 91, pp. 197–219.
Houston, K. and van Manen, M. A. (2009), Bose type formula for the internal
medial axis of an embedded manifold. Di↵erential Geom. Appl. 27, pp.
320–328.
Ishikawa, G. (1993), Determinacy of envelope of the osculating hyperplanes to a
curve. Bull. London Math.Soc., 25, pp. 787–798.
Ishikawa, G. (1995), Developable of a curve and determinacy relative to
osculation-type. Quart. J. Math. 46, pp. 437–451.
Izumiya, S. (1993), Perestroikas of optical wave fronts and graphlike Legendrian
unfoldings. J. Di↵erential Geom. 38, pp. 485–500.
Izumiya, S. (1995), Completely integrable holonomic systems of first-order di↵er-
ential equations. Proc. Royal Soc. Edinburgh 125A, pp. 567–586.
Izumiya, S. (2009), Legendrian dualities and spacelike hypersurfaces in the light-
cone Moscow Math. J. 9, pp. 325–357.
Izumiya, S. (2014), Total lightcone curvatures of spacelike submanifolds in
Lorentz-Minkowski space. Di↵erential Geom. Appl. 34, pp. 103–127.
Izumiya, S. and Janeczko, S. (2003), A sympletic framework for multiplane grav-
itational lensing. J. Math. Physics 44, pp. 2077–2093.
Izumiya, S. and Kasedou, M. (2014) Lightlike flat geometry of spacelike subman-
ifolds in Lorentz-Minkowski space. Int. J. Geom. Methods Mod. Phys. 11,
pp. 1450049–1–35.
Izumiya, S., Kikuchi, M. and Takahashi, M. (2006), Global properties of spacelike
curves in Minkowski 3-space. J. Knot Theory Ramifications 15, pp. 869–
881.
Izumiya, S. and Kossioris, G. (1995), Semi-local Classification of Geometric Sin-
gularities for Hamilton-Jacobi Equations. J. Di↵erential Equations 118,
pp. 166–193.
Izumiya, S. and Kossioris, G. (1997a), Bifurcations of shock waves for viscosity
354 Di↵erential Geometry from a Singularity Theory Viewpoint

solutions of Hamilton-Jacobi equations of one space variable. Bull. Sci.

math. 121, pp. 166–193.
Izumiya, S. and Kossioris, G. (1997b), Geometric Singularities for Solutions of
Single Coservation Laws. Arch. Rational Mech. Anal. 139, pp. 255–290.
Izumiya, S., Kossioris, G. and Makrakis, G. (2001), Multivalued solutions to the
eikonal equation in srtratified media. Quart. Applied Math. LIX, pp. 365–
390.
Izumiya, S. and Marar, W. L. (1993), The Euler characteristic of a generic wave-
front in a 3-manifold. Proc. Amer. Math. Soc. 118, pp. 1347–1350.
Izumiya, S. and Marar, W. L. (1995), On topologically stable singular surfaces in
a 3-manifold. J. Geom. 52, pp. 108–119.
Izumiya, S., Nuño Ballesteros, J. J. and Romero Fuster, M. C. (2010), Global
properties of codimension two spacelike submanifolds in Minkowski space.
Advances in Geometry 10, pp. 51–75.
Izumiya, S. and Otani, S. (2015) Flat approximations of surfaces along curves.
Demonstratio Math. XLVIII, pp. 217–241.
Izumiya, S., Pei, D. and Romero Fuster, M. C. (2006), Horospherical geometry
of surfaces in Hyperbolic 4-space. Israel J. Math. 154, pp. 361–379.
Izumiya, S., Pei, D., Romero Fuster, M. C. and Takahashi, M. (2005), The horo-
spherical geometry of submanifolds in hyperbolic space. J. London Math.
Soc. 71, pp. 779–800.
Izumiya, S., Pei, D. and Sano, T. (2003), Singularities of Hyperbolic Gauss maps.
Proc. London Math. Soc. 86, pp. 485–512.
Izumiya, S. and Romero Fuster, M. C. (2006), The horospherical Gauss-Bonnet
type theorem in hyperbolic space. J. Math. Soc. Japan 58, pp. 965-984.
Izumiya, S. and Romero Fuster, M. C. (2007), The lightlike flat geometry on
spacelike submanifolds of codimension two in Minkowski space. Sel. Math.
New ser. 13, pp. 23–55.
Izumiya, S, Romero Fuster, M.C. and Saji, K(2009), Flat lightlike hypersurfaces
in Lorentz-Minkowski 4-space. J. Geom. Phys. 59, pp. 1528–1546.
Izumiya, S. and Saji, K (2010), The mandala of Legendrian dualities for pseudo-
sphers in Lorentz-Minkowski space and “flat”spacelike surfaces. J. Sing. 2,
pp. 92–127.
Izumiya, S. and Takahashi, M. (2011), Pedal foliations and Gauss maps of hyper-
surfaces in Euclidean space. J. Sing. 6, pp. 84–97.
Izumiya, S. and Takeuchi, N. (2001), Singularities of ruled surfaces in R3 . Math.
Proc. Cambridge Philos. Soc. 130, pp. 1–11.
Izumiya, S. and Takeuchi, N. (2003), Geometry of Ruled Surfaces, in Applicable
Mathematics in the Golden Age (Edited by J. C. Misra), Narosa Publishing
House, New Delhi, India, pp. 305–338.
Izumiya, S. and Tari, F. (2008), Projections of hypersurfaces in the hyperbolic
space to hyperhorospheres and hyperplanes. Rev. Mat. Iberoam. 24, pp.
895–920.
Izumiya, S. and Tari, F. (2010a), Projections of timelike surfaces in the de Sitter
space. Real and complex singularities, pp. 190–210, London Math. Soc.
Lecture Note Ser. 380, Cambridge Univ. Press, Cambridge.
 Bibliography 355

Izumiya, S. and Tari, F. (2010b), Projections of surfaces in the hyperbolic space

along horocycles. Proc. Roy. Soc. Edinburgh Sect. A 140, pp. 399–418.
Izumiya, S. and Tari, F. (2010c), Self-adjoint operators on surfaces with a singular
metric. J. Dyn. Control Syst. 16, pp. 329–353.
Izumiya, S. and Tari, F. (2013), Projections of surfaces in the Minkowski 3-space
along lightlike directions. Nonlinearity 26, pp. 911–932.
Izumiya, S. and Yıldırım, H. (2011), Slant geometry of spacelike hypersurfaces in
lightcone. J. Math. Soc. Japan 63, pp. 715–752
Kasedou, M. (2009), Singularities of lightcone Gauss images of spacelike hyper-
surfaces in de Sitter space. J. Geometry 94, pp. 107–121.
Kasedou, M (2010a), Spacelike submanifolds of codimension at most two in de
Sitter space. in Real and complex singularities. London Math. Soc. Lecture
Note Ser. 380, pp. 211–228.
Kasedou, M (2010b), Spacelike submanifolds in de Sitter space. Demonstratio
Math. 43, pp. 401–418.
Kazarian, M. (2001), Classifying spaces of singularities and Thom polynomials.
New Developments in Singularity Theory (Cambridge 2000), NATO Sci.
Ser. II Math. Phys. Chem, 21, Kluwer Acad. Publ., Dordrecht, pp. 117–
134.
Kazarian, M. (2006), Thom polynomials. Singularity theory and its applications,
Adv. Stud. Pure Math. 43, Math. Soc. Japan, Tokyo, pp. 85–135.
Kergosien, Y. L. (1981), La famille des projections orthogonales d’une surface et
ses singularités. C. R. Acad. Sci. Paris Ser. I Math. 292, pp. 929–932.
Kergosien, Y. L. and Thom, R. (1980), Sur les points paraboliques des surfaces.
C. R. Acad. Sci. Paris Sér. A-B 290, A705–A710.
Kirk, N. P. (2000), Computational aspects of classifying singularities. LMS J.
Comput. Math. 3, pp. 207–228.
Klein, F. (1974), Le programme d’Erlangen. Considérations comparatives sur les
recherches géométriques modernes. Collection ”Discours de la Méthode”.
Gauthier-Villars Éditeur, Paris-Brussels-Montreal, Que., 1974.
Klotz, C., Pop, O. and Rieger, J. H. (2007), Real double-points of deformations
of A-simple map-germs from Rn to R2n . Math. Proc. Camb. Phil. Soc. 142,
pp. 341–363.
Kobayashi, O. (1983), Maximal surfaces in the 3-Dimensional Minkowski space.
Tokyo J. Math. 6, pp. 297–309.
Kobayashi, T. (1986), Asymptotic behaviour of the null variety for a convex
domain in a non-positively curved space form. J. Fac. Sci. Univ. Tokyo
36, pp. 389–478.
Koenderink, J. J. (1984), What does the occluding contour tell us about solid
shape? Perception 13, pp. 321–330.
Koenderink, J. J. (1990), Solide Shape. (MIT Press, Cambridge, MA).
Koenderink, J. and van Doorn, A. J. (1976), The singularities of the visual map-
ping. Biological Cybernetics 24, pp. 51–59.
Kokubu, M., Rossman, W., Saji, K., Umehara, M. and Yamada, K. (2005), Singu-
larities of flat fronts in Hyperbolic space. Pacific J. Math. 221, pp. 303–351.
Kommerell, K. (1905), Riemannsche Flächen im ebenen Raum von vier Dimen-
356 Di↵erential Geometry from a Singularity Theory Viewpoint

sionen. Math. Ann.60 no. 4, pp. 548–596.

Kossowski, M. (1987), Pseudo-Riemannian metrics singularities and the extend-
ability of parallel transport. Proc. Amer. Math. Soc. 99, pp. 147–154.
Kossowski, M. (1989), The intrinsic conformal structure and Gauss map of light-
like hypersurface in Minkowski space. Trans. Amer. Math. Soc. 316, pp.
369–383.
Kossowski, M. (2007), Extrinsic Euler characteristic of a non-immersed hypersur-
face (focal surfaces, pedal images, and extrinsic Morse indices). Di↵erential
Geom. Appl. 25, pp. 44–55.
Landis, E. E. (1981), Tangential singularities. Funktsional. Anal. i Prilozhen 15
(1981), pp. 36–49.
Lê Dung Tráng and Ramanujam, C. P. (1976), The invariance of Milnor’s number
implies the invariance of the topological type. Amer. J. Math. 98, pp. 67–78.
Lippner, G. and Szucs, A. (2010), Multiplicative properties of Morin maps. Al-
gebr. Geom. Topol. 10 , pp. 1437–1454.
Little, J. A. (1969), On singularities of submanifolds of higher dimensional Eu-
clidean spaces. Ann. Mat. Pura Appl. 83 (4), pp. 261–335.
Looijenga, E. J. N. (1974), Structural stability of smooth families of C 1 functions.
Ph. D. thesis, University of Amsterdan.
Lyashko, O. V. (1979), Geometry of bifurcation diagrams. Uspekhi Mat. Nauk
34:3 (1979), 205–206. (Russian Math. Surveys 34:3 (1979), pp. 209–210.)
Mac Lane, S. and Birkho↵, G. (1967), Algebra. The Macmillan Co., New York;
Collier-Macmillan Ltd., London.
Martinet, J. N. (1982), Singularities of smooth functions and maps. London Math-
ematical Society Lecture Notes 58. Cambridge University Press.
Martins, R. and Nuño-Ballesteros, J. J. (2009), Finitely determined singularities
of ruled surfaces in R3 . Math. Proc. Cambridge Philos. Soc. 3, pp. 701–733.
Massey, D. B. (1995), L cycles and hypersurface singularities. Lecture Notes in
Mathematics, 1615. Springer-Verlag, Berlin.
Mather, J. N. (1968), Stability of C 1 mappings, I. The division theorem. Ann.
of Math. 87, pp. 89–104.
Mather, J. N. (1969a), C 1 mappings, II. Infinitesimal stability implies stability.
Ann. of Math. 89, pp. 254–291.
Mather, J. N. (1969b), Stability of C 1 mappings, III. Finitely determined map-
germs. Publ. Math., IHES 35, pp. 279–308.
Mather, J. N. (1969c), Stability of C 1 mappings, IV. Classification of stable
germs by R-algebras. Publ. Math., IHES 37, pp. 223–248.
Mather, J. N. (1970), Stability of C 1 mappings, V. Transversality. Advances in
Math. 4, pp. 301–336.
Mather, J. N. (1971), Stability of C 1 mappings, VI. The nice dimensions. Proc.
Liverpool Singularities Symp. I, LNM 192, Springer, pp. 207–253.
Matsumoto, Y. (1997), An introduction to Morse theory. Translated from the
1997 Japanese original by Kiki Hudson and Masahico Saito. Translations of
Mathematical Monographs, 208. Iwanami Series in Modern Mathematics.
American Mathematical Society, Providence, RI, 2002. xiv+219 pp.
McDu↵, D. and Salamon, D. (1995), Introduction to symplectic topology. (Oxford
 Bibliography 357

University Press, New York.)

Mello, L. F. (2003), Mean directionally curved lines on surfaces immersed in R4 .
Publ. Mat. 47, pp. 415–440.
Milnor, J. (1963), Morse Theory. Annals of Mathematics Studies, No. 51 Prince-
ton University Press, Princeton, N.J.
Milnor, J. (1968), Singular points of complex hypersurfaces. Annals of Mathemat-
ics Studies, No. 61 Princeton University Press, Princeton, N.J.
Mochida, D.K.H. (1993), Geometria generica de subvariedades em codimensao
maior que 1. Ph.D. thesis, University of São Paulo.
Mochida, D.K.H., Romero Fuster, M.C. and Ruas, M. A. S. (1995), The geometry
of surfaces in 4-space from a contact viewpoint. Geom. Dedicata 54, pp.
323–332.
Mochida, D. K. H., Romero Fuster, M. C. and Ruas, M. A. S. (1999), Osculating
hyperplanes and asymptotic directions of codimension two submanifolds of
Euclidean spaces. Geom. Dedicata 77, pp. 305–315.
Mochida, D. K. H., Romero Fuster, M. C. and Ruas, M. A. S. (2001), Singular-
ities and duality in the flat geometry of submanifolds of Euclidean spaces.
Beiträge Algebra Geom. 42, pp. 137–148.
Mochida, D. K. H., Romero Fuster, M. C. and Ruas, M. A. S. (2003), Inflection
points and nonsingular embeddings of surfaces in R5 . Rocky Mountain J.
Math. 33, pp. 995–1009.
Mond, D. M. Q. (1982), The Classification of Germs of Maps from Surfaces to 3-
space, with Applications to the Di↵erential Geometry of Immersions. Ph.D.
thesis, University of Liverpool.
Mond, D. M. Q. (1985), On the classification of germs of maps from R2 to R3 .
Proc. London Math. Soc. 50, pp. 333–369.
Mond, D. M. Q. (1989), Singularities of the tangent developable surface of a space
curve. Quart. J. Math. 40, pp. 79–91.
Montaldi, J. A. (1983), Contact with applications to submanifolds. Ph.D. thesis,
University of Liverpool.
Montaldi, J. A. (1986a), On contact between submanifolds. Michigan Math. J.
33, pp. 81–85.
Montaldi, J. A. (1986b), Surfaces in 3-space and their contact with circles. J.
Di↵erential. Geom. 23, pp. 109–126.
Montaldi, J. A. (1991), On generic composites of mappins. Bull. London Math.
Soc. 23, pp. 81–85.
Moraes, S. and Romero Fuster, M.C. (2005), Semiumbilic and 2-regular immer-
sions of surfaces in Euclidean spaces. Rocky Mountain J. Math. 35, pp.
1327–1345.
Morris, R. J. (1996), The sub-parabolic lines of a surface. The mathematics of
surfaces, VI (Uxbridge, 1994), pp. 79–102, Inst. Math. Appl. Conf. Ser.
New Ser., 58, Oxford Univ. Press.
Mostowski, T. (1985), Lipschitz equisingularity. Dissertationes Math. (Rozprawy
Mat.) 243, 46 pp.
Nabarro, A. C. (2003), Duality and contact of hypersurfaces in R4 with hyper-
planes and lines. Proc. Edinb. Math. Soc. 46, pp. 637–648.
358 Di↵erential Geometry from a Singularity Theory Viewpoint

Nabarro, A. C. and Romero Fuster, M. C. (2009), 3-manifolds in Euclidean space

from a contact viewpoint. Comm. Anal. Geom. 17, pp. 755–776.
Nabarro, A. C. and Tari, F. (2009), Families of surfaces and conjugate curve
congruences. Adv. Geom. 9, pp. 279–309.
Nabarro, A. C. and Tari, F. (2011), Families of curve congruences on Lorentzian
surfaces and pencils of quadratic forms. Proc. Roy. Soc. Edinburgh Sect. A
141, pp. 655–672.
Nagai, T. (2012), The Gauss map of a hypersurface in Euclidean sphere and the
spherical Legendrian duality. Tology Appl. 159, pp. 545–554.
Nogueira, C. A. (1998), Superfı́cies em R4 e dualidade. Ph.D. thesis, Univesity of
São Paulo.
Nomizu, K. and Rodrı́guez, L. (1972), Umbilical submanifolds and Morse func-
tions. Nagoya Math. J. 48, pp. 197–201.
Nuño Ballesteros, J. J. (2006), Submanifolds with a non-degenerate parallel nor-
mal vector field in euclidean spaces. Singularity Theory and Its Applica-
tions. Adv. Stud. Pure Math., 43, pp. 311–332.
Nuño Ballesteros, J. J. and Romero Fuster, M. C. (1992), Global bitangency
properties of generic closed space curves. Math. Proc. Cambridge Philos.
Soc. 112, pp. 519–526.
Nuño Ballesteros, J. J. and Romero Fuster, M. C. (1993), A four vertex theorem
for strictly convex space curves. J. Geom. 46, pp. 119–126.
Nuño Ballesteros, J. J. and Saeki, O. (2001), Euler characteristic formulas for
simplicial maps. Math. Proc. Cambridge Philos. Soc. 130, pp. 307–331.
Nuño Ballesteros, J. J. and Tari, F. (2007), Surfaces in R4 and their projections
to 3-spaces. Proc. Roy. Soc. Edinburgh Sect. A, 137, pp. 1313-1328.
Ohmoto, T. (2013), Singularities and characteristic classes for di↵erentiable
maps. Lecture Notes of a mini-course in the International Workshop on
Real and Complex Singularities, ICMC-USP (São Carlos, Brazil), 2012.
arXiv:1309.0661v3
Ohmoto, T. and Aicardi, F. (2006), First order local invariants of apparent con-
tours. Topology 45, pp. 27–45.
Oliver, J. M. (2010), Pairs of geometric foliations of regular and singular surfaces.
Ph.D. thesis, Durham University. http://etheses.dur.ac.uk/280/.
Oliver, J. M. (2011), On the characteristic curves on a smooth surface. J. London.
Math. Soc. 83, pp. 755–767.
Olver, P. J. (1999), Classical Invariant Theory. London Math. Soc. Stud. Texts,
vol. 44, Cambridge University Press,.
O’Neill, B. (1983), Semi-Riemannian Geometry. With applications to relativity.
Pure and Applied Mathematics, 103, Academic Press.
Oset Sinha, R., Ruas, M. A. S. and Atique, R. G. W. (2015), On the simplicity
of multigerms. To appear in Math. Scand.
Oset Sinha, R. and Tari, F. (2010), Projections of space curves and duality. Quart.
J. Math. 64, pp. 281–302.
Oset Sinha, R. and Tari, F. (2015), Projections of surfaces in R4 to R3 and the
geometry of their singular images. Rev. Mat. Iberoamericana 31, pp. 33–51.
Ozawa, T. (1985), The numbers of triple tangencies of smooth space curves.
 Bibliography 359

Topology 24, pp. 1–13.

Pelletier, F. (1995), Quelques proprietes geomtriques des varietes pseudo-
riemanniennes singulieres. Ann. Fac. Sci. Toulouse Math. 4, pp. 87–199.
Penrose, R. (1965), Gravitational collapse and space-time singularities. Phys. Rev.
Lett. 14, pp. 57–59.
Platonova, O. A. (1984), Projection of smooth surfaces. Trudy Sem. Petrovsk,
238, pp. 135–149.
du Plessis, A. (1976a), Homotopy classification of regular sections. Compositio
Math. 32, pp. 301–333.
du Plessis, A. (1976b), Contact-Invariant regularity conditions. Singularités
d’applications di↵érentiables, pp. 205–236 . Lecture Notes in Math. Vol.
535, Springer, Berlin.
du Plessis, A. A. and Wall, C. T. C. (1995), The geometry of topological stability.
London Mathematical Society Monographs. New Series, 9. Oxford Science
Publications. Oxford University Press.
Pohl, W. F. (1962), Di↵erential geometry of higher order. Topology 1, pp. 169–
211.
Porteous, I. R. (1983a), The normal singularities of surfaces in R3 . Singularities,
Part 2 (Arcata, Calif., 1981), pp. 379–393, Proc. Sympos. Pure Math. 40,
Amer. Math. Soc., Providence, RI, 1983.
Porteous, I. R. (1983b), Probing singularities. Singularities, Part 2 (Arcata, Calif.,
1981), pp. 395–406, Proc. Sympos. Pure Math. 40, Amer. Math. Soc. Prov-
idence, R.I., 1983.
Porteous, I. R. (1987), Ridges and umbilics of surfaces. The mathematics of sur-
faces, II (Cardi↵, 1986), pp. 447–458, Inst. Math. Appl. Conf. Ser. New
Ser. 11, Oxford Univ. Press, New York, 1987.
Porteous, I. R. (2001), Geometric di↵erentiation. For the intelligence of curves
and surfaces. Cambridge University Press, Cambridge.
Poston, T. and Stewart, I. (1996), Catastrophy theory and its applications. New
Yerk: Dover.
Pugh, C. (1968), A generalized Poincaré index formula. Topology 7, pp. 217-226.
Ramı́rez-Galarza, A. I. and Sánchez-Bringas, F. (1995), Lines of curvature near
umbilical points on surfaces immersed in R4 . Ann. Global Anal. Geom. 13,
pp. 129–140.
Reeb, G. (1946), Sur les points singuliers d’une forme de Pfa↵ completement
integrable ou d’une fonction numérique. C. R. Acad. Sci. Paris 222, pp.
847–849.
Remizov, A. O. (2009), Geodesics on 2-surfaces with pseudo-Riemannian metric:
singularities of changes of signature. Mat. Sb. 200:3, pp. 75–94.
Rieger, J. H. (1987), Families of maps from the plane to the plane. J. London
Math. Soc. 2, pp. 351–369.
Rieger, J. H. and Ruas, M. A. S. (1991), Classification of A-simple germs from
Kn to K2 . Compositio Math. 79, pp. 99–108.
Rieger, J. H. and Ruas, M. A. S. (2005), M -deformations of A-simple ⌃n p+1 -
germs from Rn to Rp , n p. Math. Proc. Camb. Phil. Soc. 139, pp. 333–349.
Romero Fuster, M. C. (1981), The convex hull of an immersion. Ph.D. thesis,
360 Di↵erential Geometry from a Singularity Theory Viewpoint

University of Southampton.
Romero Fuster, M. C. (1983), Sphere stratifications and the Gauss map. Proc.
Edinburgh Math. Soc. 95, pp. 115–136.
Romero Fuster, M. C. (1988), Convexly generic curves in R3 . Geom. Dedicata 28,
pp. 7–29.
Romero Fuster, M. C. (1997), Stereographic projections and geometric singulari-
ties. Workshop on Real and Complex Singularities (São Carlos, 1996). Mat.
Contemp. 12, pp. 167–182.
Romero Fuster, M.C. (2004), Semiumbilics and geometrical dynamics on surfaces
in 4-spaces. Real and complex singularities, pp. 259–276, Contemp. Math.,
354, Amer. Math. Soc., Providence, RI.
Romero Fuster, M. C. (2007), Geometric contacts and 2-regularity of surfaces in
Euclidean space. Singularity theory, pp. 307–325, World Sci. Publ., Hack-
ensack, NJ, 2007.
Romero Fuster, M. C., Ruas, M. A. S. and Tari, F. (2008), Asymptotic curves on
surfaces in R5 , Communications in Contemporary Maths. 10, pp. 1–27.
Romero Fuster, M. C. and Sanabria Codesal, E. (2002), On the flat ridges of
submanifolds of codimension 2 in Rn . Proc. Roy. Soc. Edinburgh Sect. A
132, pp. 975–984.
Romero Fuster, M. C. and Sanabria-Codesal, E. (2004), Lines of curvature, ridges
and conformal invariants of hypersurfaces. Biträge Algebra Geom. 45, pp.
615–635.
Romero Fuster, M. C. and Sanabria Codesal, E. (2008), Conformal invariants
interpreted in de Sitter space. Mat. Contemp. 35, pp. 205–220.
Romero Fuster, M. C. and Sanabria Codesal, E. (2013), Conformal invariants and
spherical contacts of surfaces in R4 . Rev. Mat. Complut. 26, pp. 215–240.
Romero Fuster, M. C. and Sánchez-Bringas F. (2002), Umbilicity of surfaces
with orthogonal asymptotic lines in R4 . Di↵erential Geom. and Appl. 16,
pp. 213–224.
Romero Fuster, M. C. and Sedykh, V. D.(1995), On the number of singularities,
zero curvature points and vertices of a simple convex space curve. J. Geom.
52, pp. 168–172.
Romero Fuster, M. C. and Sedykh, V. D. (1997), A lower estimate for the number
of zero-torsion points of a space curve. Biträge Algebra Geom. 38, pp. 183–
192.
Ruas, M. A. S. and Tari, F. (2012), A note on binary quintic forms and lines of
principal curvature on surfaces in R5 . Topology Appl. 159, pp. 562–567.
Saeki, O. (1996), Simple stable maps of 3-manifolds into surfaces. Topology 35,
pp. 671–698.
Saeki, O. (2004), Topology of singular fibers of di↵erentiable maps. Lecture Notes
in Mathematics, 1854. (Springer-Verlag, Berlin).
Saji, K. (2010), Criteria for singularities of smooth maps from the plane into the
plane and their applications. Hiroshima Math. J. 40, pp. 229–239.
Saji, K., Umehara, M. and Yamada, K. (2009), The geometry of fronts. Ann. of
Math. 169, pp. 491–529.
Saloom, A. and Tari, F. (2012), Curves in the Minkowski plane and their contact
 Bibliography 361

with pseudo-circles. Geometria Dedicata 159, pp. 109–124.

Seade, J. (2007), On Milnor’s fibration theorem for real and complex singulari-
ties. Singularities in geometry and topology, pp. 127–158, World Sci. Publ.,
Hackensack, NJ, 2007.
Sedykh, V. D. (1989), Double tangent planes to a space curve. (Russian) Sibirsk.
Mat. Zh. 30, pp. 209–211; translation in Siberian Math. J. 30, pp. 161–162.
Sedykh, V. D. (1992), The four-vertex theorem of a convex space curve. Funct.
Anal. Appl. 26, pp. 28–32.
Sedykh, V. D. (2012), On Euler characteristics of manifolds of singularities of wave
fronts. (Russian) Funktsional. Anal. i Prilozhen. 46, pp. 92–96; translation
in Funct. Anal. Appl. 46, pp. 77–80.
Shcherbak, O. P. (1986), Projectively dual space curve and Legendre singularities.
Sel. Math. Sov. 5, pp. 391–421.
Siddiqi, K. and Pizer, S. (Eds.) (2008), Medial Representations. Mathematics,
Algorithms and Applications. Computational Imaging and Vision, Vol. 37.
(Springer).
Sotomayor, J. (2004), Historical Comments on Monge’s Ellipsoid and the Con-
figuration of Lines of Curvature on Surfaces Immersed in R3 . ArXiv :
math/0411403v1.
Sotomayor, J. and Gutierrez, C. (1982), Structurally stable configurations of lines
of principal curvature. Bifurcation, ergodic theory and applications (Dijon,
1981), pp. 195–215, Astérisque, 98-99, Soc. Math. France, Paris.
Steller, M. (2006), A Gauss-Bonnet formula for metrics with varying signature.
Z. Anal. Anwend. 25, pp. 143–162.
Szabó, E., Szucs, A. and Terpai, T. (2010), On bordism and cobordism groups of
Morin maps. J. Singul. 1, pp. 134–145.
Tabachnikov, S. (1997), Parametrized plane curves, Minkowski caustics,
Minkowski vertices and conservative line fields. Enseign. Math. 43, pp.
3–26.
Tari, F. (1991), Projections of piecewise-smooth surfaces. J. London Math. Soc.
44, pp. 155–172.
Tari, F. (2009), Self-adjoint operators on surfaces in Rn . Di↵erential Geom. Appl.
27, pp. 296-306.
Tari, F. (2010), Pairs of foliations on surfaces. Real and complex singularities,
pp. 305–337, London Math. Soc. Lecture Note Ser., 380, Cambridge Univ.
Press, Cambridge.
Tari, F. (2012), Caustics of surfaces in the Minkowski 3-space. Quart. J. Math.
63, pp. 189–209.
Tari, F. (2013), Umbilics of surfaces in the Minkowski 3-space. J. Math. Soc.
Japan, 65, pp. 723–731.
Thom, R. (1956), Les singularités des applications di↵érentiables. Ann. Inst.
Fourier, Grenoble 6, pp. 43–87.
Thom R. (1972), Sur les équations di↵érentielles multiformes et leurs intégrales
singulières. Bol. Soc. Brasil. Mat. 3 , pp. 1–11.
Thom, R. (1976) Structural stability and morphogenesis. An outline of a general
theory of models. Translated from the French by D. H. Fowler. With a
362 Di↵erential Geometry from a Singularity Theory Viewpoint

foreword by C. H. Waddington. Second printing. W. A. Benjamin, Inc.,

Reading, Mass.-London-Amsterdam.
Thom, R. (1983), Mathematical models of morphogenesis. Halsted Press [John
Wiley & Sons, Inc.], New York.
Uribe-Vargas, R. (2001), Singularités symplectiques et de contact en géomtrie
di↵érentielle des courbes et des surfaces. Ph. D. thesis, Université Paris 7.
Uribe-Vargas, R. (2006), A projective invariant for swallowtails and godrons and
global theorems on the flecnodal curve. Mosc. Math. J. 6, pp. 731–768.
Vaisman, I. (1984) A first course in Di↵erential Geometry. Marcel Dekker.
Wall, C. T. C. (1977), Geometric properties of generic di↵erentiable manifolds.
Geometry and Topology, Rio de Janeiro, 1976, Springer Lecture Notes in
Math. 597, pp. 707–774.
Wall, C. T. C. (1981), Finite determinacy of smooth map-germs. Bull. London
Math. Soc. 13, pp. 481–539.
Wall, C. T. C. (2004), Singular points of plane curves. (London Mathematical
Society Student Texts 63, Cambridge University Press, Cambridge).
West, J. (1995), The di↵erential geometry of the cross-cap. Ph.D. thesis, Univer-
sity of Liverpool.
Whitney, H. (1944), The singularities of a smooth n-manifold in (2n 1)-space.
Ann. of Math. 45, pp. 247–293.
Whitney, H. (1955), On singularities of mappings of euclidean spaces. I. Mappings
of the plane into the plane. Ann. of Math. 62, pp. 374–410.
Wilkinson, T. C. (1991), The geometry of folding maps. Ph.D. thesis, University
of Newcastle-upon-Tyne.
Wong, Y. C. (1952), A new curvature theory for surfaces in a Euclidean 4-space.
Comment. Math. Helv. 26, pp. 152–170.
Yamamoto, M. (2006), First order semi-local invariants of stable maps of 3-
manifolds into the plane. Proc. London Math. Soc. 92, pp. 471–504.
Zakalyukin, V. M. (1976), Lagrangian and Legendrian singularities. Funct. Anal.
Appl. 10, pp. 23–31.
Zakalyukin, V. M. (1984), Reconstructions of fronts and caustics depending on a
parameter and versality of mappings. J. Soviet Math. 27, pp. 2713–2735.
Zakalyukin, V. M. (1995), Envelopes of families of wave fronts and Control theory.
Proc. Steklov Institute of Math. 209, pp. 114–123.
 Index

P -R+ -equivalence, 62, 64 binormal direction

P -K-equivalent, 63 surfaces in R4 , 222, 223, 236, 238,
⌫-umbilic point, 40 241
⌫-flat umbilic point, 41 surfaces in R5 , 259, 260, 265
⌫-principal curvature, 38 binormal vector, 14
⌫-shape operator
submanifolds in Rn+r , 37–39 canal
surfaces in R4 , 216 cylindrical pedal, 134
h-principal, 346 hypersurface, 42, 43, 134, 225
k-jet space, 47 surface, 43
⌫-parabolic point, 41, 43 canonical
⌫-principal direction, 38 1-form, 98
4-Vertex Theorem, 332, 343 contact structure, 106
symplectic structure, 99
kth-regular immersion, 345 Carathéodory conjecture, 337
catastrophe
apparent contour, 159, 160, 163, 164 map, 9, 102
asymptotic curve map-germ, 64
surfaces in R3 , 144, 157, 186, 187 set, 9, 101
surfaces in R4 , 213–217, 339 caustic, 9
surfaces in R5 , 264, 266, 269 Christo↵el symbols, 322
asymptotic direction classification, 59
surfaces in R3 , 144, 148, 158, 160 codimension of an extended orbit, 54
surfaces in R4 , 213, 214, 223, 235, codimension of an orbit, 54
240, 241 complex curve, 219
surfaces in R5 , 260–262, 265, computer vision, 139
268–270, 272 conjugate curve congruence, 145
conjugate direction, 144, 163
bi-Lipschitz equivalence, 71 contact, 7
bifurcation set, 9, 64, 101 between submanifolds, 75
binormal curvature (surfaces in R4 ), form, 105
338, 339 group, 75

363
364 Di↵erential Geometry from a Singularity Theory Viewpoint

manifold, 105 de Sitter 3-space, 284

map-germ, 75 de Sitter horosphere, 305
structure, 105 degree of a map, 340
type, 76 developable surface, 19
with a foliation, 80 discriminant, 10, 48, 64, 68
contactomorphic, 107 distance squared function, 7
contactomorphism, 107, 108 distance squared function generic
contour generator, 159, 160 surface in R3 , 179, 191
convex, 330, 336 surface in R4 , 247
convex hull, 330 surfaces in R5 , 273
cosmic censorship, 292 dual
cotangent bundle, 9, 98 coordinates, 99
criminant, 48 hypersurface, 32
critical set, 48 dual surface, 154
cross-cap, 13, 18, 50, 234, 235, Dupin foliation
239–241 spherical, 133, 183
elliptic, 239 tangential, 132, 153
hyperbolic, 239 Dupin indicatix
parabolic, 239 tangent, 152
curvature
de Sitter Gauss-Kronecker, 321 elliptic point
de Sitter mean, 321 surfaces in R3 , 12, 33
Gauss-Kronecker, 26–28, 33, 34, 43 surfaces in R4 , 207, 212, 213, 215,
geodesic, 142, 158 223, 239
hyperbolic Gauss-Kronecker, 321 equidistant surface, 307
hyperbolic mean, 321 equisingularity, 71
Lipschitz-Killing, 38, 39, 43 Euler characteristic of
mean, 26 a cylindrical pedal, 335
normal, 142, 143 a locally convex surface in R4 , 336
plane curve, 2, 3 a wavefront, 333
principal, 26, 27, 29–31, 164 evolute, 3, 4, 30
space curve, 14 extended family of
curvature ellipse distance squared functions, 90, 178
surfaces in R4 , 205, 208, 210, 213 height functions, 88, 119, 147
surfaces in R5 , 252, 257, 258 lightcone height functions, 300
cusp, 4, 6, 49 extended tangent space Le G · f , 54
cusp of Gauss, 155–157, 185, 188, external point, 330
332, 335 extremal of a principal curvature,
cuspidal cross-cap, 20 190, 196
cuspidal edge, 20, 154 extrinsic properties, 141
cylindrical pedal, 32, 153
singularities of, 154 family of
distance squared functions, 8, 9,
Damon, 69 34, 90, 120, 178, 246
Darboux Theorem, 99 height functions, 33, 88, 119, 129,
147, 221, 224, 226, 236
 Index 365

lightcone height functions, 293, 319 surfaces in R3 , 147, 155, 156, 184
Lorentzian distance squared surfaces in R4 , 226, 227, 229, 231,
functions, 309 232
orthogonal projections, 93, 159 surfaces in R5 , 259
finite determinacy, 55 Hessian matrix, 34, 41
first fundamental form, 24, 25, 36, horosphere, 307
140 horospherical
flat rib, 230 Chern-Lashof Type Theorem, 344
flat ridge, 230, 232, 265 Gauss-Bonnet Theorem, 344
flecnodal Horospherical Geometry, 281, 325
curve, 187 hyperbolic 3-space, 284
point, 165, 166, 187 hyperbolic Gauss map, 307
set, 166, 188 hyperbolic point
focal set, 13, 30, 182, 195 surfaces in R3 , 12, 33
fold, 49 surfaces in R4 , 207, 212, 215, 223,
future 229, 238, 239
directed, 284 hypocycloid, 193, 198
direction, 284
inflection, 3, 164, 186
Gauss map, 12, 25–27, 32, 43, 44, 140 inflection point
Gauss map with respect to a normal surfaces in R4 , 206, 210, 212, 214,
vector field, 37 215, 220, 223, 228, 231, 234,
Gauss-Bolyai-Lobachevski, 284 238, 335, 336
Gauss-Bonnet Theorem, 335 internal points, 330
Gaussian curvature, 12, 141, 163, 204,
239 Koenderink Theorem, 163
generalised Gauss map, 226, 227
generating family, 10, 11, 102, 109, Lagrangian
116 di↵eomorphism, 100
generic equivalent, 101
embedding, 81 fibration, 100
immersion, 81, 85–87 immersion, 9
property, 81 map, 9, 101
geodesic inflection, 142, 186, 187, 197 stable, 102
geometric subgroups, 69 submanifolds, 100
germ, 46 surface, 9
graph-like left group, 51
Legendrian left-right group, 51
immersion, 113 Legendrian
submanifold, 113 di↵eomorphism, 108
Morse family of hypersurfaces, 115 equivalent, 108, 109
wavefront, 116 fibration, 107
gravitational collapse, 292 immersion, 10
lift, 108
height function generic map, 10, 108
stable, 109
366 Di↵erential Geometry from a Singularity Theory Viewpoint

submanifold, 107 mean curvature, 141

surface, 10 mean curvature vector
Lie group, 52 surfaces in R4 , 205
lightcone surfaces in R41 , 292
Chern-Lashof Type Theorem, 342 surfaces in R5 , 253
dual surface, 318 Metric properties, 71
Gauss map, 287 Milnor, 70
Gauss-Bonnet Theorem, 341 Milnor Fibration Theorem, 70
Gauss-Kronecker curvature, 288 Milnor number, 70
mean curvature, 288 minimal surface, 293
normal vector, 318 miniversal unfolding, 57, 58, 62, 63,
parabolic point, 288, 291 65
pedal, 300 Minkowski space-time, 281, 283
pedal surface, 300 Monge form, 146
principal curvature, 288 Morse family of
second fundamental form, 288 functions, 102
shape operator, 319 hypersurfaces, 108
Theorema Egregium, 320 multi-germ, 47
tightness, 345 multi-jet space, 47
umbilic, 288
Weingarten formula, 289 nice dimensions, 86
lightlike normal curvature, 204
focal set, 312 normal form, 59
hyperplane, 283 normal frame
hypersurface, 311 future directed, 287
plane, 283 past directed, 287
tangent hyperplane, 298 normal section, 143, 163
vector, 283 normal vector, 14
Lightlike Geometry, 281 normalised lightcone
line of principal curvature, 141 Gauss map, 291
Liouville form, 98 Gauss-Kronecker curvature, 291
Little, 201, 204, 205, 338 mean curvature, 291
local ring, 112 principal curvatures, 291
Loewner conjecture, 337 shape operator, 291
Lorentz-Minkowski space, 281
open lightcone, 284
mandala, 317 osculating
map-germ, 46 circle, 73
marginally trapped, 292 family of hyperspheres, 132
strongly, 293 hyperplane, 222, 249, 260
Mather, 45 hypersphere, 73, 247
Mather’s groups, 51
Maurer-Cartan structural equations, parabolic point
203 hypersurfaces in Rn+1 , 33, 34
maximal spacelike surface, 293 surfaces in R3 , 12, 148
 Index 367

surfaces in R4 , 207, 212, 215, 223, shape recognition, 139

231, 238 simple germ, 60
parabolic set singular fibre, 70
hypersurfaces in Rn+1 , 33 singular map-germ, 48
surfaces in R3 , 44, 184 singular point, 47
surfaces in R4 , 215 space-time singularities, 292
parallel, 5, 6, 10 spacelike
past directed, 284 hyperplane, 283
pencils of quadratic forms, 212, 216 plane, 283
Penrose inequality, 292 surface, 285
Plücker conoid, 18 vector, 283
pleat, 49 spacelike knot, 343
3
Poincaré, 284 sphere (in H+ ( 1)), 307
Poincaré-Hopf formula, 329, 336 standard contact structure, 105
principal stereographic projection, 336
curvature, 12, 141 strata, 330
direction, 12, 26 bifurcation, 331
Principal Axis Theorem, 25 conflict, 331
profile, 159 Morse, 331
projection P -generic stratification, 329, 330
surfaces in R4 , 233, 235, 236 Maxwell, 331
surfaces in R5 , 267 stratified set, 330
projection generic surface in R3 , 165 sub-parabolic curve, 195–197
projective cotangent bundle, 10, 106 swallowtail, 11, 20, 154
swallowtail point, 333
regular point, 48 Symmetry Lemma, 75
rib, 274 Symmetry Set, 21
rib of order k, 247 symplectic
ridge, 189–192 form, 98
ridge of order k (surfaces in R4 ), 248 manifold, 98
ridge point, 189 structure, 98
right-group, 51 symplectomorphism, 99
robust features, 139, 184
ruled surface, 15, 18 tangent
affine hyperplane, 129
second fundamental form indicatrix, 130
hypersurfaces in Rn+1 , 27 tangent developable, 19
submanifold in Rn+1 , 36 tangent space LG · f , 54
surfaces in R3 , 140 The contact Darboux Theorem, 107
surfaces in R4 , 203, 204, 209, 211, The Morse Lemma, 60
219 Theorema Eugregium of Gauss, 12
surfaces in R5 , 252, 254 Thom, 95
semiumbilic point, 206, 248, 249, 337, Thom’s seven elementary
338 catastrophes, 63
Serret-Frenet, 15 Thom’s splitting lemma, 61
shape operator, 12, 25, 28, 29, 43
368 Di↵erential Geometry from a Singularity Theory Viewpoint

Thom’s transversality theorem, 81 umbilical curvature (surfaces in R5 ),

Thom-Bordman symbols, 51, 227 276
time-orientable, 284 umbilical focal hypersurface (surfaces
time-orientation, 285 in R5 ), 273
timelike umbilical focus (surfaces in R5 ), 273
hyperplane, 283
plane, 283 Vassiliev type invariant, 346
vector, 283 versal deformation, 56, 62
torsion, 15 versal unfolding, 56–58
torsion zero point, 332 vertex, 3
totally ⌫-umbilic submanifold, 40
totally semiumbilic surface, 338 wavefront, 5, 108
totally umbilic hypersurface, 29 Weingarten
transversality, 81 formula, 28
tri-tangent plane, 335 map, 25, 140
tri-tangent support planes, 332 Weingarten map, 12
triple point, 333 Whitney, 45
Whitney C 1 -topology, 80
umbilic point, 12, 29, 141, 193, 194, Whitney umbrella, 50
336