Paulo Ventura Araújo

Differential
Geometry
Differential Geometry
Paulo Ventura Araújo

Differential Geometry
Paulo Ventura Araújo
Department of Mathematics
Faculdade de Ciências da Universidade do Porto
Porto, Portugal

Translation from the Portuguese language edition: “Geometria Diferencial” by Paulo Ventura
Araújo, © Paulo Ventura Araújo 1998. Published by IMPA. All Rights Reserved.

ISBN 978-3-031-62383-7 ISBN 978-3-031-62384-4 (eBook)

https://doi.org/10.1007/978-3-031-62384-4
Mathematics Subject Classification (2020): 53-XX, 51-XX

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland
AG 2024
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher,
whether the whole or part of the material is concerned, specifically the rights of reprinting, reuse of
illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by
similar or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with
regard to jurisdictional claims in published maps and institutional affiliations.

Cover illustration: Image created from two figured to place above the title.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

If disposing of this product, please recycle the paper.

Preface

This book is based on the lecture notes of the course Differential Geometry taught
at the Faculty of Sciences of the University of Porto in the academic years 1992–
93 and 1993–94. Students from different courses and with different mathematical
backgrounds attended the course and, consequently, its prerequisites were reduced
to Linear Algebra, Calculus (of one and several variables), and the study of curves
up to the Frenet trihedron. Furthermore, we avoided the introduction of certain
technical apparatus, such as Tensor Calculus, in order to insist instead on results
with accessible geometric content whose proofs, although possibly long, used more
elementary means.
That said, one understands why we have restricted ourselves to the study of curves
and surfaces in Euclidean space. But, in our opinion, even for students pursuing a
scientific career, this is the right approach for a first study of Differential Geometry,
grounding intuition and motivating the problems that arise in higher dimensions.
Although the idea was to reproduce, in order and content, the course notes, the
notes grew and included subjects not discussed in the lectures. There is a risk, when
teaching Differential Geometry at an introductory level, that the harvest of interesting
results will not compensate for the work spent in digesting definitions and assimilating
techniques. The digressions in this text may lead the student to discover some of the
richness of Differential Geometry which, by the imperative of bureaucratic “realism”,
is so often absent from the classroom.
The exercises included in the text are rarely routine, although few are really
difficult, and were chosen on the assumption that a good exercise, with a medium
level of difficulty, should reward the students’ effort by teaching them something.
Among the books we consulted, Manfredo do Carmo’s [6] deserves to be high-
lighted: some of the exercises and the structuring of some subjects come from there.
But several exercises are original, and the selection of themes and the composition of
the proofs reflect personal taste and work.
We now give some hints on how to use the book: Sections 1.1 – 1.3 cover subjects
probably already known to the student, and may be omitted in well-prepared classes.
Chapters 2–4 cover a basic course in Differential Geometry. Sections 3.3, 4.4 may be
omitted if time is tight. If time permits, a choice of topics from Chapter 1 (sections

v
vi Preface

1.4 to 1.8) and Chapter 5 can be made; the interdependence between the various
sections of these chapters is indicated at the beginning of each chapter.

Porto, July 1996

Paulo Ventura Araújo.
Contents

1 Differentiable Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Velocity and Arc Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Acceleration, Curvature and the Frenet Trihedron . . . . . . . . . . . . . . . . 5
1.3 Planar Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Contact of Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5 Convex Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.6 Curves of Constant Width . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7 Theorem of the Four Vertices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.8 The Isoperimetric Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2 Regular Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1 Definition and Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 Change of Parameters, Level Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3 Differentiable Functions on Surfaces, Tangent Space . . . . . . . . . . . . . 42
2.4 Orientability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.5 Areas, Lengths, and Angles: The First Fundamental Form . . . . . . . . . 51

3 The Geometry of the Gauss Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3.1 The Gauss Map and its Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2 The Second Fundamental Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.3 Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

4 The Intrinsic Geometry of Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4.1 Conformal Mappings and Isometries . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.2 Gauss’s Theorema Egregium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.3 Covariant Derivative, Parallel Transport, Geodesic Curvature . . . . . 92
4.4 The Divergence Theorem. First Variation of Area . . . . . . . . . . . . . . . . . 101
4.5 The Gauss-Bonnet Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.6 Minimizing Properties of Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Appendix: Rotation Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

vii
viii Contents

5 The Global Geometry of Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

5.1 Complete Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.2 Coverings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5.3 Complete Surfaces of Non-Positive Curvature . . . . . . . . . . . . . . . . . . . 149
5.4 Ovals (First Part): The Rigidity of the Sphere . . . . . . . . . . . . . . . . . . . . 152
5.5 Ovals: Areas and Volumes; Surfaces of Constant Width . . . . . . . . . . . 157
5.6 Abstract Surfaces. The Hyperbolic Plane . . . . . . . . . . . . . . . . . . . . . . . . 163
5.7 Complete Surfaces of Constant Curvature . . . . . . . . . . . . . . . . . . . . . . . 172

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Chapter 1
Differentiable Curves

The first three sections of this chapter contain the basics on curves, and, because of
their brevity, are rather a review of concepts and results, gathered in a form we will
use later; in the last five sections we will deal with subjects whose inclusion in the
course is optional. Section 1.5 should be read before 1.6 and 1.7, but otherwise, and
except for a few exercises, sections 1.4 – 1.8 are independent of each other.

1.1 Velocity and Arc Length

In the space R 𝑛 we will denote vectors by the symbols v, w and points by lowercase
consonants 𝑝, 𝑞. This space is equipped with a canonical Euclidean structure in which
the inner product of two vectors is the sum of the products of their components of
equal√index, i.e. ⟨v, w⟩ = 𝑣 1 𝑤 1 + ⋯ + 𝑣 𝑛 𝑤 𝑛 . The norm or length of a vector is given by
∣v∣ = ⟨v, v⟩, and the angle between the nonzero vectors v and w is the only number
𝜃 ∈ [0, 𝜋] such that
⟨v, w⟩
cos 𝜃 = .
∣v∣ ∣w∣
The distance between the points 𝑝 and 𝑞 is defined as the length of the vector 𝑝 − 𝑞.
A parametrized curve is a continuous function 𝛼∶ 𝐼 → R 𝑛 on an interval of R in
Euclidean 𝑛-dimensional space, and its trace is the image of that function. Writing
𝛼(𝑡) = (𝑥 1 (𝑡), . . . , 𝑥 𝑛 (𝑡)), the functions 𝑥𝑖 are the component functions of 𝛼. We say
that 𝛼 is of class 𝐶 ∞ if each of its component functions has continuous derivatives of
all orders (if 𝛼 is defined, for example, on a closed interval [𝑎, 𝑏], then we require the
existence of all right-hand side derivatives at 𝑎 and left-hand side derivatives at 𝑏).
The velocity vector of the curve is 𝛼′ (𝑡) = (𝑥 1′ (𝑡), . . . , 𝑥 𝑛′ (𝑡)) and, when nonzero,
points in the direction tangent to the curve at time 𝑡. Regular curves are those whose
velocity vector never vanishes and therefore have a well-defined tangent direction at
each instant.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 1

P. V. Araújo, Differential Geometry, https://doi.org/10.1007/978-3-031-62384-4_1
2 1 Differentiable Curves

From now on, unless otherwise stated, by curve we mean a regular parametrized
curve of class 𝐶 ∞ .

The simplest example of a curve is a straight line 𝑝 + 𝑡v, 𝑡 ∈ R , parametrized with

nonzero constant velocity v. Other examples are the circle (cos 𝑡, sin 𝑡) and, shown in
Fig. 1.1, the cloverleaf (cos 3𝑡 cos 𝑡, cos 3𝑡 sin 𝑡) and the helix (cos 𝑡, sin 𝑡, 𝑡).

z y

y
x

Figure 1.1

One of the first questions to ask about a curve is how to compute its length. Assume
that the curve 𝛼 is defined on a closed interval [𝑎, 𝑏]. Take an arbitrary partition
𝑎 = 𝑡0 < 𝑡1 < ⋯ < 𝑡 𝑚 = 𝑏 of that interval. The sum
𝑚
∑ ∣𝛼(𝑡𝑖 ) − 𝛼(𝑡𝑖−1 )∣
𝑖=1

gives the length of the polygonal line obtained by replacing, for each 1 ≤ 𝑖 ≤ 𝑚,
the trace of the curve in the interval [𝑡 𝑖−1 , 𝑡 𝑖 ] by the line segment joining 𝛼(𝑡𝑖−1 )
with 𝛼(𝑡𝑖 ) (see Fig. 1.2). The narrower and more numerous are the intervals of the
partition, the better the sum should approximate the length of the curve. In Exercise 3
of this section we show that the limit of these sums, as the maximum of the differences
𝑡𝑖 − 𝑡 𝑖−1 tends to zero, is given by the integral ∫𝑎 ∣𝛼′ (𝑡)∣ 𝑑𝑡, and this is how the length
𝑏

𝑙(𝛼) of the curve 𝛼 is defined.

1.1 Velocity and Arc Length 3

a (t 5 )
a (t0 ) a (t 4 )

a (t1 ) a (t 3 )

a (t 2 )

Figure 1.2

Our first result reassures us as to the correctness of the definition just given, by
showing that the line is the shortest path between two points:

Proposition 1.1.1 Let 𝛼∶ [𝑎, 𝑏] → R 𝑛 be a curve. Then 𝑙(𝛼) ≥ ∣𝛼(𝑏) − 𝛼(𝑎)∣, and
equality holds if and only if the trace of 𝛼 is a line segment.

Proof We assume that 𝛼(𝑏) ≠ 𝛼(𝑎), otherwise there is nothing to be shown. We

write 𝛼(𝑏) − 𝛼(𝑎) = ∫𝑎 𝛼′ (𝑡) 𝑑𝑡, an equality in which the right-hand side is the
𝑏

vector whose coordinates are the integrals of the components of 𝛼′ (𝑡). Forming the
inner product of both sides with the vector 𝛼(𝑏) − 𝛼(𝑎), we obtain

𝑏
∣𝛼(𝑏) − 𝛼(𝑎)∣2 = ∫ ⟨𝛼′ (𝑡), 𝛼(𝑏) − 𝛼(𝑎)⟩ 𝑑𝑡
𝑎
𝑏
≤∫ ∣𝛼′ (𝑡)∣∣𝛼(𝑏) − 𝛼(𝑎)∣ 𝑑𝑡 = 𝑙(𝛼)∣𝛼(𝑏) − 𝛼(𝑎)∣,
𝑎

from which, simplifying, we obtain the desired inequality. For equality to hold, the
Cauchy-Schwarz inequality

⟨𝛼′ (𝑡), 𝛼(𝑏) − 𝛼(𝑎)⟩ ≤ ∣𝛼′ (𝑡)∣∣𝛼(𝑏) − 𝛼(𝑎)∣

will have to reduce to an equality for all 𝑡 ∈ [𝑎, 𝑏], which happens if and only if 𝛼′ (𝑡)
is a positive multiple of 𝛼(𝑏) − 𝛼(𝑎) — i.e., if and only if the trace of 𝛼 is the line
segment joining 𝛼(𝑎) to 𝛼(𝑏). ◻

The value 𝑣(𝑡) = ∣𝛼′ (𝑡)∣ is the scalar velocity of the curve 𝛼 at time 𝑡, and
𝑡
𝑆(𝑡) = ∫ 𝑣(𝑟) 𝑑𝑟
𝑎

is the arc length function. When the scalar velocity is constant, the arc length is
proportional to time. We recall that the curve 𝛼˜ is a reparametrization of 𝛼 if 𝛼˜ = 𝛼 ○ ℎ,
for some increasing diffeomorphism ℎ∶ 𝐽 → 𝐼 between intervals of R . (If ℎ were
4 1 Differentiable Curves

decreasing, then we would obtain a curve with the same trace as 𝛼 but with reversed
orientation — that is, traversed in the opposite direction.) Let us now show, by
defining a suitable function ℎ, that it is possible to reparameterize 𝛼 so that 𝛼 ○ ℎ has
constant scalar velocity, equal to 1.
Since 𝑆 ′ (𝑡) = 𝑣(𝑡) > 0, the function 𝑆 is increasing and sends [𝑎, 𝑏] diffeomor-
phically onto the interval [0, 𝑙(𝛼)]. Defining ℎ as the inverse of 𝑆, and putting
˜
𝛼(𝑠) = 𝛼 ○ ℎ(𝑠), we have

1 ′
𝛼˜ ′ (𝑠) = ℎ′ (𝑠)𝛼′ (ℎ(𝑠)) = 𝛼 (𝑡),
𝑣(𝑡)

where 𝑡 = ℎ(𝑠), and from this we see that ∣𝛼˜ ′ (𝑠)∣ = 1 for all 𝑠 in [0, 𝑙(𝑎)].
We say that a curve is parametrized by arc length when it is traversed with constant
scalar velocity equal to 1, regardless of whether the time at which the parameterization
starts is zero.

Exercises
1. Let 𝛼∶ [𝑎, 𝑏] → R 𝑛 and 𝛽∶ [𝑐, 𝑑] → R 𝑛 be two regular, injective curves with the
same trace. Show that the function 𝛽−1 ○ 𝛼∶ [𝑎, 𝑏] → [𝑐, 𝑑] is differentiable and its
derivative never vanishes.
2. Let 𝛼(𝑡) = (𝑒 𝑏𝑡 cos 𝑡, 𝑒 𝑏𝑡 sin 𝑡), where 𝑏 is a negative constant and 𝑡 ∈ R .
(a) Sketch the trace of 𝛼.
(b) Check that 𝛼 has finite length on [𝑡0 , +∞[ and compute it.
3. Let 𝛼(𝑡) = (𝑥 1 (𝑡), . . . , 𝑥 𝑛 (𝑡)), 𝑡 ∈ [𝑎, 𝑏] be a curve of class 𝐶 1 . Show that

(a) given 𝜀 > 0, there exists 𝛿 > 0 such that for all 1 ≤ 𝑘 ≤ 𝑛,

𝑥 𝑘 (𝑡) − 𝑥 𝑘 (𝑠)
∣𝑡 − 𝑠∣ < 𝛿 ⇒ ∣ − 𝑥 ′𝑘 (𝑡)∣ < 𝜀;
𝑡−𝑠

(b) for the 𝜀 and 𝛿 just obtained, if 𝑎 = 𝑡0 < 𝑡1 < ⋯ < 𝑡 𝑚 = 𝑏 is a partition of [𝑎, 𝑏]
such that 𝑡𝑖 − 𝑡 𝑖−1 < 𝛿 for all 1 ≤ 𝑖 ≤ 𝑚, then
𝑚 𝑚 √
∣∑ ∣𝛼(𝑡𝑖 ) − 𝛼(𝑡𝑖−1 )∣ − ∑ ∣𝛼′ (𝑡𝑖 )∣(𝑡 𝑖 − 𝑡 𝑖−1 )∣ < 𝑛(𝑏 − 𝑎)𝜀;
𝑖=1 𝑖=1

(c) the limit of sums

𝑚
∑ ∣𝛼(𝑡𝑖 ) − 𝛼(𝑡𝑖−1 )∣, as max (𝑡 𝑖 − 𝑡 𝑖−1 ) → 0,
𝑖=1 1≤𝑖≤𝑚

is ∫𝑎 ∣𝛼′ (𝑡)∣ 𝑑𝑡.

𝑏

4. A curve 𝛼∶ [𝑎, 𝑏] → R 𝑛 is called rectifiable if the supremum of the sums

1.2 Acceleration, Curvature and the Frenet Trihedron 5
𝑚
∑ ∣𝛼(𝑡𝑖 ) − 𝛼(𝑡𝑖−1 )∣,
𝑖=1

where 𝑡 0 < 𝑡 1 < ⋯ < 𝑡 𝑚 is some partition of [𝑎, 𝑏], is finite. We call this supremum
the length of 𝛼.
(a) Show that, for curves of class 𝐶 1 , this new definition of length is equivalent to
the previous one.
(b) Does a rectifiable curve necessarily have derivatives at all points?
(c) Is it true that the graph of any continuous and monotone function [𝑎, 𝑏] → R
is a rectifiable curve?
(d) Consider the Weierstrass example of a continuous function [0, 1] → R that is
nowhere differentiable (see [24], chapter 23, Theorem 5), and find out whether the
graph of this function is rectifiable.

1.2 Acceleration, Curvature and the Frenet Trihedron

In this section we only consider curves in three dimensional Euclidean space. In the
previous section we learned how to compute the length of a curve with the help of its
velocity or first derivative. This calculation does not exhaust the analysis of a curve,
since it tells us nothing about the shape it can take, and does not distinguish a line
from a circle. To proceed we will also have to take into account the second derivative.
Given a curve 𝛼∶ 𝐼 → R 3 , the tangent unit vector to 𝛼 is 𝝉(𝑡) = 𝑣(𝑡) 1
𝛼′ (𝑡). If
˜
𝛼(𝑟) = 𝛼 ○ ℎ(𝑟) is a reparametrization of 𝛼 then the unit vector tangent to 𝛼˜ is given
by 𝝉˜ = 𝝉 ○ ℎ. Hence, defining 𝑣˜ (𝑟) = ∣𝛼˜ ′ (𝑟)∣, we have

𝑣˜ (𝑟) = (𝑣 ○ ℎ(𝑟))ℎ′ (𝑟),

∣𝝉˜ ′ (𝑟)∣ = ∣𝝉 ′ (ℎ(𝑟))∣ℎ′ (𝑟).

If we put
1
𝑘(𝑡) = ∣𝝉 ′ (𝑡)∣
𝑣(𝑡)
and denote by 𝑘˜ the analogous function for 𝛼,˜ then we see that 𝑘˜ = 𝑘 ○ ℎ. The quantity
𝑘(𝑡) is the curvature of 𝛼 at the point 𝛼(𝑡). The preceding calculations show that the
curvature does not depend on the way the curve is traversed but only on the point
at which it is computed, and does not change even when we reverse its orientation.
We can thus assume, whenever convenient, that the parameter of the curve is the arc
length. In this case we simply have 𝑘(𝑠) = ∣𝛼′′ (𝑠)∣.
A line, for example, has zero constant curvature: in fact, when parametrized with
constant scalar velocity, its second derivative vanishes. But the converse is also true,
since the condition 𝑘(𝑠) ≡ 0 implies that 𝛼′′ (𝑠) ≡ 0, which shows that there exist
constants 𝑝 and v such that 𝛼(𝑠) ≡ 𝑝 + 𝑠v.
Since 𝝉(𝑡) has constant norm, it is orthogonal to its derivative. In fact, by
differentiating the equality ⟨𝝉(𝑡), 𝝉(𝑡)⟩ = 1, we obtain 2⟨𝝉 ′ (𝑡), 𝝉(𝑡)⟩ = 0. Thus, when
6 1 Differentiable Curves

𝑘(𝑡) ≠ 0, the vector 𝝉 ′ (𝑡) points in a direction normal to the curve, the so-called
principal normal. In these cases we can define the unit vector
1
n(𝑡) = 𝝉 ′ (𝑡)
∣𝝉 ′ (𝑡)∣

and the center of curvature, which is the point

1
𝛼(𝑡) + n(𝑡).
𝑘(𝑡)

The value 1/𝑘(𝑡) is the curvature radius at the point 𝛼(𝑡). The osculating plane is
the plane parallel to 𝝉(𝑡) and n(𝑡) that passes through 𝛼(𝑡).
The curvature measures the variation of the direction of the curve, but it does not
determine the form of the curve: both the circumference and the helix, for instance,
have constant curvature, that of the former being equal to the inverse of the radius;
and while the circumference is a planar curve, in the helix the osculating plane varies
from point to point. What we lack then is to measure the variation of the osculating
plane — or, to put it differently, how far a curve deviates from being planar.
Continuing to assume that 𝑘(𝑡) ≠ 0, the bi-normal vector b(𝑡) is defined as the
only vector such that (𝝉(𝑡), n(𝑡), b(𝑡)) is a direct orthonormal trihedron — that is,
an ordered triplet of unit vectors, orthogonal to each other, such that the 3 × 3-matrix
whose columns are these vectors in the same order has positive determinant. Even
simpler, we have b(𝑡) = 𝝉(𝑡) × n(𝑡), where × denotes the vector product on R 3 .

n
b

b
n

Figure 1.3

The trihedron (𝝉(𝑡), n(𝑡), b(𝑡)) is the Frenet trihedron. Each of the vectors of the
trihedron is orthogonal to its derivative, so that each derivative is expressible as a
linear combination of the other two vectors. We will next study the coefficients of
this linear combination. We have
1.2 Acceleration, Curvature and the Frenet Trihedron 7

b′ (𝑡) =
𝑑
(𝝉(𝑡) × n(𝑡))
𝑑𝑡
= 𝝉 ′ (𝑡) × n(𝑡) + 𝝉(𝑡) × n′ (𝑡)
= 𝝉(𝑡) × n′ (𝑡)

(because 𝝉 ′ and n are collinear for each 𝑡). From this equality it follows that b′ , which
we already knew to be orthogonal to b, is also orthogonal to 𝝉, and is therefore a
multiple of n. The torsion of 𝛼 at the point 𝛼(𝑡) is the value 𝜈(𝑡) defined by the
equality
1 ′
b (𝑡) = 𝜈(𝑡)n(𝑡);
𝑣(𝑡)
the factor 1/𝑣(𝑡) again ensures that the torsion is independent of the parameterization.
Note that torsion can assume negative values — although, like curvature, it does not
depend on the orientation of the curve.
Differentiating the product n(𝑡) = b(𝑡) × 𝝉(𝑡), and using the formulas seen above
for b′ (𝑡) and 𝝉 ′ (𝑡), we find that

1 ′
n (𝑡) = −𝑘(𝑡)𝝉(𝑡) − 𝜈(𝑡)b(𝑡).
𝑣(𝑡)

We thus establish the equalities

1 ′
𝝉 = 𝑘n
𝑣
1 ′
n = −𝑘𝝉 − 𝜈b
𝑣
1 ′
b = 𝜈n,
𝑣
which are known as Frenet formulas. When the curve is parametrized by arc length
the factor 1/𝑣 (which is then equal to 1) is suppressed.
The next proposition gives us the geometric meaning of torsion being identically
zero.

Proposition 1.2.1 Assume that the curvature of 𝛼 never vanishes. Then 𝛼 is a planar
curve if and only if it has zero constant torsion.

Proof If 𝛼 is all contained in the plane 𝑆, the vectors 𝝉(𝑡) and n(𝑡) are parallel
to 𝑆 and linearly independent, and the product b(𝑡) = 𝝉(𝑡) × n(𝑡) is a unit vector
orthogonal to 𝑆; therefore b(𝑡) can only take two distinct values and, since it varies
continuously, it must be constant — from which it follows that 𝜈 ≡ 0.
Conversely, if 𝜈 ≡ 0 then b(𝑡) is a nonzero constant vector b and for all 𝑡 holds
⟨b, 𝛼′ (𝑡)⟩ = 0. Therefore there exists a constant 𝑐 such that ⟨b, 𝛼(𝑡)⟩ ≡ 𝑐, and this
equality shows that the curve 𝛼 is planar. ◻

Let us now consider a curve 𝛼(𝑠), where the parameter 𝑠 ∈ [𝑎, 𝑏] is the arc
length, and whose curvature 𝑘(𝑠) never vanishes. We ask ourselves to what extent the
8 1 Differentiable Curves

functions 𝑘(𝑠) and 𝜈(𝑠) determine the curve 𝛼. They do not do so completely, since
any rigid movement of R 3 transforms 𝛼 into another curve with the same curvature
and torsion functions. (By rigid movement we mean the composition of a translation
with a linear mapping that transforms the canonical basis of R 3 into some direct
orthonormal trihedron.) But this is the only freedom:

Theorem 1.2.2 Let 𝛼, 𝛼∶ ˜ [𝑎, 𝑏] → R 3 be curves parametrized by arc length and with
nonzero curvature at all points. If we have 𝑘(𝑠) = 𝑘(𝑠)
˜ and 𝜈(𝑠) = 𝜈(𝑠)
˜ for all 𝑠 on
[𝑎, 𝑏], then there exists a rigid movement 𝐿∶ R → R 3 such that 𝛼 ≡ 𝐿 ○ 𝛼.
3
˜

Proof Let 𝐿 1 be the linear mapping that transforms the trihedron (𝝉(𝑎),˜ ñ(𝑎), b̃(𝑎))
into the trihedron (𝝉(𝑎), n(𝑎), b(𝑎)), 𝐿 2 the translation that maps 𝐿 1 (𝛼(𝑎))˜ into
𝛼(𝑎), and 𝐿 = 𝐿 2 ○ 𝐿 1 . Then the curve 𝛼0 = 𝐿 ○ 𝛼˜ has at the start time 𝑠 = 𝑎 the same
Frenet trihedron as 𝛼 and satisfies 𝛼0 (𝑎) = 𝛼(𝑎). Identifying by the subindex 0 the
vectors and quantities concerning the curve 𝛼0 , we define the function
1
𝛿(𝑠) = (∣𝝉(𝑠) − 𝝉 0 (𝑠)∣2 + ∣n(𝑠) − n0 (𝑠)∣2 + ∣b(𝑠) − b0 (𝑠)∣2 ) .
2
We know that 𝛿(𝑎) = 0; furthermore, we have

𝛿′ = ⟨𝝉 ′ − 𝝉 ′0 , 𝝉 − 𝝉 0 ⟩ + ⟨n′ − n′0 , n − n0 ⟩ + ⟨b′ − b′0 , b − b0 ⟩.

From this, using Frenet’s formulas and the fact that 𝑘 0 = 𝑘 and 𝜈0 = 𝜈, we easily
obtain 𝛿′ ≡ 0, therefore also 𝛿 ≡ 0. In particular we have 𝛼0′ = 𝛼′ — and, since
𝛼0 (𝑎) = 𝛼(𝑎), we conclude that 𝛼0 = 𝛼. ◻

Note 1.2.3 To complement Theorem 1.2.2, we will now show the following result:
for any differentiable functions 𝑘, 𝜈∶ [𝑎, 𝑏] → R (with 𝑘 strictly positive), there exists
a curve 𝛼(𝑠), parametrized by arc length, whose curvature and torsion functions are
precisely 𝑘(𝑠) and 𝜈(𝑠). We make use of the Theorem of Existence and uniqueness
of solutions of differential equations that we will state in Section 3.3; it deserves
mention that this approach provides another proof of Theorem 1.2.2. We keep the
above proof though as it is more elementary.
Once functions 𝑘(𝑠) and 𝜈(𝑠) are fixed, Frénet’s formulas can be viewed as
a non-autonomous equation of the form 𝑋 = v(𝑠, 𝑋), where 𝑋 = (𝝉, n, b) ∈ R 9
and where v∶ [𝑎, 𝑏] × R 9 → R 9 is differentiable. Take any direct orthonormal
trihedron (𝝉(𝑎), n(𝑎), b(𝑎)). Then there exists 𝜀 > 0 such that the solution 𝑋(𝑠) =
(𝝉(𝑠), n(𝑠), b(𝑠)) with this initial condition is defined for [𝑎, 𝑎 + 𝜀]. But since

𝑑
{∣𝝉∣2 + ∣n∣2 + ∣b∣2 } = 2 (⟨𝝉, 𝑘n⟩ + ⟨n, −𝑘𝝉 − 𝜈b⟩ + ⟨b, 𝜈n⟩) = 0,
𝑑𝑡
√
we see that 𝑋(𝑠) stays in the compact set {𝑋 ∈ R 9 ∶ ∣𝑋∣ = 3}, and is therefore defined
on the entire interval [𝑎, 𝑏] (see Theorem 3 on p. 17 of [23]). By differentiating, we
obtain the equalities
1.3 Planar Curves 9

⟨𝝉, n⟩′ = 𝑘∣n∣2 − 𝑘∣𝝉∣2 − 𝜈⟨𝝉, b⟩

⟨n, b⟩′ = 𝜈∣n∣2 − 𝜈∣b∣2 − 𝑘⟨𝝉, b∣
⟨𝝉, b⟩′ = 𝑘⟨n, b⟩ − 𝜈⟨𝝉, n⟩
(∣𝝉∣2 )′ = 2𝑘⟨𝝉, n⟩
(∣n∣2 )′ = −2𝑘⟨𝝉, n⟩ − 2𝜈⟨n, b⟩
(∣b∣2 )′ = 2𝜈⟨n, b⟩

– from which it follows that ⟨𝝉, n⟩, ⟨n, b⟩, ⟨𝝉, b⟩, ∣𝝉∣2 , ∣n∣2 and ∣b∣2 are constant func-
tions, equal to 0, 0, 0, 1, 1 and 1 respectively, since, as it is easily seen, these constants
constitute a solution, with the same initial condition, of the same differential equation
defined in R 6 . This proves that (𝝉(𝑠), n(𝑠), b(𝑠)) is an orthonormal, necessarily
direct trihedron for all 𝑠 ∈ [𝑎, 𝑏]. To conclude we take for 𝛼(𝑠) any primitive of
𝝉(𝑠), e.g. 𝛼(𝑠) = ∫𝑎 𝝉(𝑡) 𝑑𝑡: we verify without difficulty that (𝝉(𝑠), n(𝑠), b(𝑠)) is
𝑠

the Frenet trihedron of 𝛼(𝑠), and that 𝑘(𝑠) and 𝜈(𝑠) are its curvature and torsion. ◻

Exercise
5. Show that if we permit curves whose curvature vanishes at some point then
the conclusion of Theorem 1.2.2 holds — that is, there exists a pair of curves
˜ [𝑎, 𝑏] → R 3 parametrized by arc length such that their curvature and torsion
𝛼, 𝛼∶
functions coincide whenever they are defined, but which cannot be transformed into
each other by a rigid movement.
Hint: look for an example that also shows that in Proposition 1.2.1 the assumption
that the curvature is positive is essential.

1.3 Planar Curves

We will deal in this section with planar curves, more specifically with curves in R 2 .
To simplify the calculations, we only consider curves parametrized by arc length: the
formulas we obtain are easily adapted to any other parameterizations.
Consider a curve 𝛼∶ 𝐼 → R 2 . Then there exists, for each 𝑠 ∈ 𝐼, a single vector n(𝑠)
such that the pair (𝝉(𝑠), n(𝑠)) forms a direct orthonormal or positively oriented
dihedron: if 𝝉(𝑠) = (𝑥 ′ (𝑠), 𝑦 ′ (𝑠)) then n(𝑠) = (−𝑦 ′ (𝑠), 𝑥 ′ (𝑠)). As before, we know
that the vectors 𝝉 ′ (𝑠) and 𝝉(𝑠) are orthogonal; but in this case we can conclude that
𝝉 ′ (𝑠) is a multiple of n(𝑠). The curvature of 𝛼 at the point 𝛼(𝑠) is the number 𝐾(𝑠)
such that 𝝉 ′ (𝑠) = 𝑘(𝑠)n(𝑠).
This curvature can take negative values and is therefore sometimes called signed
curvature; its absolute value is equal to the curvature defined in the previous section.
Whenever we talk about the curvature of a curve in R 2 we will be referring to the
signed curvature.
10 1 Differentiable Curves

Frenet’s formulas in this case boil down to

𝝉 ′ = 𝑘n
n′ = −𝑘𝝉.

Since 𝝉(𝑠) is a unit vector, it describes the position vector of a point on S 1 (circle
with radius 1 centered at the origin). Denoting by 𝜑(𝑠) the oriented angle that 𝝉(𝑠)
makes with the positive part of the 𝑥-axis, we have 𝝉(𝑠) = (cos 𝜑(𝑠), sin 𝜑(𝑠)) and
n(𝑠) = (− sin 𝜑(𝑠), cos 𝜑(𝑠)). From this we obtain

𝝉 ′ (𝑠) = 𝜑′ (𝑠)(− sin 𝜑(𝑠), cos 𝜑(𝑠)),

𝑘(𝑠) = ⟨𝝉 ′ (𝑠), n(𝑠)⟩ = 𝜑′ (𝑠). (1.1)

This last formula gives a geometric interpretation of curvature, showing that it

measures the variation of the angle 𝜑 that the tangent line to the curve makes with a
fixed oriented line; and that the curvature is positive or negative according to whether
the curve turns left or right (see Fig. 1.4).

Figure 1.4

Note 1.3.1 We must make one caveat: the value 𝜑(𝑠) of the angle is only determined
up to an integer multiple of 2𝜋, and it is not clear that we can make a choice for each
𝑠 such that the resulting function is differentiable or even continuous. The way to
solve this problem is to take advantage of the formula (1.1) to define 𝜑(𝑠): fixing
𝑠0 ∈ 𝐼 and a number 𝜑0 such that 𝝉(𝑠0 ) = (cos 𝜑0 , sin 𝜑0 ), we put
𝑠
𝜑(𝑠) = 𝜑0 + ∫ 𝑘(𝑡) 𝑑𝑡.
𝑠0

The function 𝜑 is obviously differentiable and all that remains to be shown is that

𝝉(𝑠) = (cos 𝜑(𝑠), sin 𝜑(𝑠))

– or, which is the same, that the function

1.3 Planar Curves 11

1
𝛿(𝑠) = ∣𝝉(𝑠) − (cos 𝜑(𝑠), sin 𝜑(𝑠))∣2
2
is identically zero. Let us put 𝝉(𝑠) = (𝑥 ′ (𝑠), 𝑦 ′ (𝑠)): by Frenet’s formulas, we have
𝑥 ′′ (𝑠) = −𝑘(𝑠)𝑦 ′ (𝑠) and 𝑦 ′′ (𝑠) = 𝑘(𝑠)𝑥 ′ (𝑠). Therefore

𝛿′ (𝑠) = ⟨𝝉 ′ (𝑠) − 𝜑′ (𝑠)(− sin 𝜑(𝑠), cos 𝜑(𝑠)), 𝝉(𝑠) − (cos 𝜑(𝑠), sin 𝜑(𝑠))⟩
= (𝑘(𝑠) − 𝜑′ (𝑠))(𝑦 ′ (𝑠) cos 𝜑(𝑠) − 𝑥 ′ (𝑠) sin 𝜑(𝑠)) = 0,

since 𝜑′ (𝑠) = 𝑘(𝑠). From this, and since 𝛿(𝑠0 ) = 0, we conclude that 𝛿 ≡ 0.
We further stress that any other continuous choice of the angle between 𝝉(𝑠)
and the 𝑥-axis is of the form 𝜑(𝑠) + 2𝑛𝜋, for some constant 𝑛 ∈ Z (since the
difference between two choices is continuous and only takes values in the discrete set
{2𝑛𝜋 ∶ 𝑛 ∈ Z }).

We end this section with a result that is the version for planar curves of Theo-
rem 1.2.2 and the note 1.2.3; its proof is kept as an exercise (it is possible to prove
directly, without using the theorem on solutions of differential equations, the existence
of 𝛼, and for uniqueness adapt the proof of Theorem 1.2.2)

Theorem 1.3.2 Given a differentiable function 𝑘∶ [𝑎, 𝑏] → R , there exists some curve
𝛼∶ [𝑎, 𝑏] → R 2 whose curvature at 𝛼(𝑠) is 𝑘(𝑠). Any other curve with the same
curvature function is the composite of 𝛼 with some rigid plane movement.

It follows in particular from Theorem 1.3.2 that the only planar curves with
nonzero constant curvature are circles or arcs of circumference. A direct proof of this
fact is certainly possible and is an exercise well worth to be done.

Exercises

6. (a) Consider a curve 𝛼(𝑡) = (𝑥(𝑡), 𝑦(𝑡)) not necessarily parametrized by arc length,
and put, as usual, 𝑣(𝑡) = ∣𝛼′ (𝑡)∣ and 𝝉(𝑡) = 𝑣(𝑡)
1
𝛼′ (𝑡). Prove each of the following
formulas for the curvature of 𝛼:
1 1 1
𝑘 = ⟨𝝉 ′ , n⟩ = 2 ⟨𝛼′′ n⟩ = 3 ⟨𝛼′′ , 𝑣n⟩
𝑣 𝑣 𝑣
𝑥 ′ 𝑦 ′′ − 𝑥 ′′ 𝑦 ′
= .
((𝑥 ′ )2 + (𝑦 ′ )2 )3/2

𝑥2 𝑦2
(b) Show that the curvature of the ellipse given by the equation 𝑎2
+ 𝑏2
= 1,
parametrized by 𝛼(𝑡) = (𝑎 cos 𝑡, 𝑏 sin 𝑡), is given by

𝑎𝑏
𝑘(𝑡) = 3/2
.
(𝑎 2 sin2 𝑡 + 𝑏 2 cos2 𝑡)
12 1 Differentiable Curves

7. Let v∶ [𝑎, 𝑏] → S 1 be a differentiable mapping. Show that there exists a differentiable

function 𝜑∶ [𝑎, 𝑏] → R such that v(𝑡) = (cos 𝜑(𝑡), sin 𝜑(𝑡)) for all 𝑡 on [𝑎, 𝑏].

8. Let 𝛼∶ ]𝑎, 𝑏[→ R 2 be a regular curve parametrized by arc length. Given 𝑠0 ∈ ]𝑎, 𝑏[
and 𝑝 ∈ R 2 ∖ {𝛼(𝑠0 )}, let 𝐶 be the circle with center 𝑝 and radius ∣𝛼(𝑠0 ) − 𝑝∣.
(a) Show that 𝐶 is tangent to 𝛼 at the point 𝛼(𝑠0 ) if and only if 𝑝 is a point of the
normal to 𝛼 at 𝛼(𝑠0 ).
(b) Assume 𝑝 = 𝛼(𝑠0 ) + 𝜆n(𝑠0 ) and consider the function 𝜌(𝑠) = ∣𝛼(𝑠) − 𝑝∣2 .
Show that if 𝜆𝑘(𝑠0 ) > 1 then 𝑠0 is a strict local maximum of 𝜌 and if 𝜆𝑘(𝑠0 ) < 1 then
𝑠0 is a strict local minimum.
(c) Let D ⊆ R 2 be a circle with radius 𝑅 such that 𝛼(]𝑎, 𝑏[) is contained in the
closed disk bounded by D. Conclude that at the instants 𝑠0 at which 𝛼(𝑠0 ) ∈ D, one
has ∣𝑘(𝑠0 )∣ ≥ 𝑅1 .

9. Let 𝛼∶ 𝐼 → R 2 be a regular curve of always nonzero curvature. The curve

𝛽(𝑡) = 𝛼(𝑡) + 𝑘(𝑡)
1
n(𝑡) (𝑡 ∈ 𝐼) traversed by the center of curvature of 𝛼 is called the
evolute of 𝛼.
(a) Show that, if it is defined, the tangent line to its evolute at time 𝑡 coincides with
the normal to 𝛼 at the same instant.
(b) Assume that the curve 𝛽(𝑡) ˜ = 𝛼(𝑡) + 𝜆(𝑡)n(𝑡) has the property described in
(a). Show that 𝜆(𝑡) = 𝑘(𝑡) 1
.
(c) Consider the normals to 𝛼 at two nearby points 𝛼(𝑡0 ) and 𝛼(𝑡 0 + ℎ). Show that
as ℎ → 0, the point of intersection of the two normals tends to 𝛽(𝑡0 ).
𝑥2 𝑦2
(d) Study the evolute of the ellipse 𝑎2
+ 𝑏2
= 1.

1.4 Contact of Curves

We continue our study of curves by considering their Taylor polynomial expansions:

given 𝑠0 inside the interval 𝐼 and 𝑛 ≥ 1, we can write
1 1
𝛼(𝑠0 + 𝑠) = 𝛼(𝑠0 ) + 𝑠𝛼′ (𝑠0 ) + 𝑠2 𝛼′ (𝑠0 ) + ⋯ + 𝑠 𝑛 𝛼(𝑛) (𝑠0 ) + 𝑜(𝑠 𝑛 ),
2 𝑛!

where the remainder 𝑜(𝑠 𝑛 ) is a vector such that lim 1

𝑛 ∣𝑜(𝑠 )∣
𝑛
= 0. This expression
𝑠→0 𝑠
of the curve can be used to detect its local properties: below we consider the contact
theory of planar curves; in the exercises we obtain the geometric meaning of the
osculating plane and the sign of torsion.
We now introduce a concept that measures the degree of closeness of two planar
curves in the neighborhood of an intersection point. We say that the two curves 𝛼 and
𝛼˜ have 𝑛-order contact at the point 𝛼(𝑠0 ) = 𝛼(𝑠
˜ 0 ) if

1
lim ∣𝛼(𝑠
˜ 0 + 𝑠) − 𝛼(𝑠0 + 𝑠)∣ = 0.
𝑠→0 𝑠 𝑛
1.4 Contact of Curves 13

Considering the Taylor polynomial expansion, it is easily seen that this condition is
equivalent to the condition that the derivatives of 𝛼 and 𝛼˜ up to order 𝑛 coincide at 𝑠0 .
In particular, since 𝛼′ (𝑠0 ) = 𝝉(𝑠0 ) and 𝛼′′ (𝑠0 ) = 𝑘(𝑠0 )n(𝑠0 ), we conclude that two
planar curves have second-order contact at a point 𝑝 if and only if they are tangent
to 𝑝 and had equal curvature there (of course when we speak of tangency here we
require that the velocity vectors of the two curves are identical — i.e., point in the
same direction); when the curvature is nonzero, this is equivalent to these two curves
having the same center of curvature at 𝑝. From this we conclude that the only circle
which has second-order contact with 𝛼 at a point of nonzero curvature is the one
with center at the center of curvature of 𝛼 at that point and radius equal to the radius
of curvature (of course there is no circle with second-order contact at a point of zero
curvature).
The definition of 𝑛-order contact we have given is perhaps not the most natural
nor the easiest to handle, since it depends on a special parameterization of the two
curves. To improve the situation, we start by defining Δ(𝑡) as the distance between
the points 𝛼(𝑠0 + 𝑠) and 𝛼(𝑠˜ 0 + 𝑠˜) given by the condition

⟨𝛼(𝑠0 + 𝑠) − 𝛼(𝑠0 ), 𝝉(𝑠0 )⟩ = 𝑡 = ⟨𝛼(𝑠

˜ 0 + 𝑠˜) − 𝛼(𝑠0 ), 𝝉(𝑠0 )⟩;

Δ(𝑡) is thus the length of the line segment bounded by the intersections with 𝛼 and 𝛼˜
of the line orthogonal to 𝝉(𝑠0 ) and the (oriented) distance 𝑡 from 𝛼(𝑠0 ) (see Fig. 1.5).

~
(t, f(t))
D (t)
n (s0)
(t, f(t))

t (s0)

Figure 1.5

Proposition 1.4.1 The curves 𝛼 and 𝛼˜ have 𝑛-order contact at the point 𝛼(𝑠0 ) =
˜ 0 ) if and only if lim 𝑡1𝑛 Δ(𝑡) = 0.
𝛼(𝑠
𝑡→0
Proof We first assume that 𝛼 and 𝛼˜ have 𝑛-order contact — i.e., that 𝛼(𝑠0 ) = 𝛼(𝑠 ˜ 0)
and 𝛼(𝑖) (𝑠0 ) = 𝛼˜ (𝑖) (𝑠0 ) for 1 ≤ 𝑖 ≤ 𝑛, — and we define functions 𝑔 and 𝑔˜ by

𝑔(𝑠) = ⟨𝛼(𝑠0 + 𝑠) − 𝛼(𝑠0 ), 𝝉(𝑠0 )⟩

˜
𝑔(𝑠) = ⟨𝛼(𝑠
˜ 0 + 𝑠) − 𝛼(𝑠0 ), 𝝉(𝑠0 )⟩.
14 1 Differentiable Curves

Note that 𝑔(0) = 𝑔(0)˜ = 0 and 𝑔 (𝑖) (0) = 𝑔˜ (𝑖) (0) for 1 ≤ 𝑖 ≤ 𝑛, and that, since
the first derivative at 0 of these functions is nonzero, both have local inverses in a
neighborhood of 0, which we will denote by 𝑔 −1 and 𝑔˜ −1 . Let us now put

𝐹(𝑠) = ⟨𝛼(𝑠0 + 𝑠) − 𝛼(𝑠0 ), n(𝑠0 )⟩,

˜
𝐹(𝑠) = ⟨𝛼(𝑠
˜ 0 + 𝑠) − 𝛼(𝑠0 ), n(𝑠0 )⟩,
𝑓 (𝑡) = 𝐹 ○ 𝑔 −1 (𝑡), 𝑓˜(𝑡) = 𝐹˜ ○ 𝑔˜ −1 (𝑡),

and note that, for 𝑡 near 0, one has Δ(𝑡) = ∣ 𝑓 (𝑡) − 𝑓˜(𝑡)∣ (see Fig. 1.5.). Moreover,
the first 𝑛 derivatives of 𝑓 and 𝑓˜ at 0 coincide, since at this point also the first 𝑛
derived from 𝐹 are identical to those from 𝐹, ˜ and those from 𝑔 −1 to those from 𝑔˜ −1 .
It follows, as desired, that lim 𝑡 𝑛 Δ(𝑡) = 0.
1
𝑡→0
We now prove the converse implication. To simplify the notation, we assume that
𝑠0 = 0, take 𝛼(0) = 𝛼(0)
˜ for the origin of the coordinates, and further assume that
the tangent and normal vectors at this point are (1, 0) and (0, 1) respectively.
With the above notation, let 𝛽(𝑡) = 𝛼 ○ 𝑔 −1 (𝑡), 𝛽(𝑡)
˜ = 𝛼˜ ○ 𝑔˜ −1 (𝑡): we then have
𝛽(𝑡) = (𝑡, 𝑓 (𝑡)) and 𝛽(𝑡)
˜ = (𝑡, 𝑓˜(𝑡)), and by hypothesis 𝑓 and 𝑓˜ are functions whose
derivatives up to the 𝑛th order coincide at 0. Since 𝛼 and 𝛼˜ are parametrized by arc
length, we have
𝑡 𝑡 √
𝑔 −1 (𝑡) = ∫ ∣𝛽′ (𝑟)∣ 𝑑𝑟 = ∫ 1 + [ 𝑓 ′ (𝑟)]2 𝑑𝑟,
0 0

𝑡 𝑡 √
𝑔˜ −1 (𝑡) = ∫ ∣ 𝛽˜′ (𝑟)∣ 𝑑𝑟 = ∫ 1 + [ 𝑓˜′ (𝑟)]2 𝑑𝑟,
0 0

and these formulas show that the derivatives of 𝑔 −1 and 𝑔˜ −1 at the point 0 coincide at
least to the same order as the derivatives of 𝑓 and 𝑓˜ at the same point — that is, to
the order 𝑛. Since 𝛼 = 𝛽 ○ 𝑔 and 𝛼˜ = 𝛽˜ ○ 𝑔, we conclude that 𝛼(𝑖) (0) = 𝛼˜ (𝑖) (0) for
1 ≤ 𝑖 ≤ 𝑛, which means that 𝛼 and 𝛼˜ have 𝑛-order contact in 𝛼(0) = 𝛼(0).˜

Exercises

10. Assume that two planar curves 𝛼 and 𝛼, ˜ not necessarily parametrized by arc
length, touch at the point 𝑝 = 𝛼(0) = 𝛼(0).
˜ Show that if lim 𝑡1𝑛 ∣𝛼(𝑡)
˜ − 𝛼(𝑡)∣ = 0, then
𝑡→0
the curves have 𝑛-order contact at 𝑝.

11. With the same terminology as in Proposition 1.4.1, but under the assumption
that 𝑠0 = 0 and 𝜑(𝑠0 ) = 0, show that 𝑓 ′ (𝑡) = 𝑡𝑔𝜑(𝑠) and 𝑓 ′′ (𝑡) = cos3 𝜑(𝑠) , where
𝑘(𝑠)

𝑠 = 𝑔 −1 (𝑡). Conclude that if 𝑘(0) > 0 then there exists 𝜀 > 0 such that, on an
appropriate coordinate system, the trace of 𝛼∣[−𝜀, 𝜀] is the graph of a convex function.
1.5 Convex Curves 15

12. Let 𝐼 be an open interval such that 𝑜 ∈ 𝐼 and let 𝛼∶ 𝐼 → R 3 be a curve parametrized
by arc length whose curvature at 0 is nonzero.
(a) Show that
1 1 1 1
𝛼(𝑠) = 𝛼(0) + (𝑠 − 𝑘 2 𝑠3 ) 𝝉 + ( 𝑘 𝑠2 + 𝑘 ′ 𝑠3 ) n − 𝑘 𝜈𝑠3 b + 𝑜(𝑠3 ),
6 2 6 6
where the quantities 𝑘, 𝜈 and the vectors 𝝉, n, b are computed at 0.
(b) Conclude that if 𝜈(0) > 0 then, when 𝑠 reaches the instant 0, the curve crosses
the osculating plane at 𝛼(0) from top to bottom (the “top part” is the one pointed to
by b(0)).
13. Using the notation and assumptions from Exercise 12, show that the plane
that contains the points 𝛼(0), 𝛼(ℎ0 ) and 𝛼(ℎ1 ), for ℎ1 < 0 < ℎ0 , converges to the
osculating plane at 𝛼(0) when ∣ℎ0 ∣ + ∣ℎ1 ∣ → 0.
Hint: Using the Taylor expansion — just up to the second order — show that

(𝛼(ℎ0 ) − 𝛼(0)) × (𝛼(ℎ1 ) − 𝛼(0))

v(ℎ0 , ℎ1 ) =
∣(𝛼(ℎ0 ) − 𝛼(0)) × (𝛼(ℎ1 ) − 𝛼(0))∣

has limit b(0) when ∣ℎ0 ∣ + ∣ℎ1 ∣ → 0.

1.5 Convex Curves

Continuing with planar curves parametrized by arc length, let us talk about simple
closed curves and characterize those that are convex. We say that a curve 𝛼∶ [𝑎, 𝑏] →
R 2 is closed if 𝛼(𝑎) = 𝛼(𝑏); if its periodic extension, defined by 𝛼(𝑠 + 𝑛(𝑏 − 𝑎)) =
𝛼(𝑠) for 𝑠 ∈ [𝑎, 𝑏] and 𝑛 ∈ Z , is differentiable (i.e., 𝐶 ∞ ), the curve is closed regular;
and if the curve has no self-intersections — that is, if its restriction on [𝑎, 𝑏[ is
injective — we say it is simple.
In this section all closed curves are regular; and, where necessary, we consider
them defined in R by periodic extension.
We recall from Section 1.3 that there is a continuous choice 𝜑(𝑠) of the angle that
the vector 𝝉(𝑠) makes with the positive part of the 𝑥-axis. Since 𝝉(𝑏) = 𝝉(𝑎), the
difference 𝜑(𝑏) − 𝜑(𝑎) is an integer multiple of 2𝜋, which, by note 1.3.1, does not
depend on the choice of 𝜑(𝑠). We call rotation index of the closed curve 𝛼 the integer
ℜ(𝛼) = 21𝜋 (𝜑(𝑏) − 𝜑(𝑎)); ℜ(𝛼) thus counts the number of turns that its tangent
vector 𝝉(𝑠) makes in the unit circle when the point 𝛼(𝑠) completes one turn around
the curve. Since 𝜑′ (𝑠) = 𝑘(𝑠), the rotation index can be given in integral form

1 𝑏
ℜ(𝛼) = ∫ 𝑘(𝑠) 𝑑𝑠.
2𝜋 𝑎
A closed curve 𝛼∶ [𝑎, 𝑏] → R 2 is called convex if, for every 𝑠0 ∈ [𝑎, 𝑏], the curve is all
on the same side of the tangent line to 𝛼 at the point 𝛼(𝑠0 ) — that is, if the function
16 1 Differentiable Curves

ℎ(𝑠) = ⟨𝛼(𝑠) − 𝛼(𝑠0 ), n(𝑠0 )⟩ does not change sign. Our next result characterizes
these curves.

non-simple closed curve non-convex curve convex curve

Figure 1.6

Theorem 1.5.1 A closed curve is convex if and only if its curvature does not change
sign and its rotation index is ±1. Any convex curve is simple.
Proof (i) We begin by showing that any closed curve with non-negative curvature
at all its points and rotation index 1 is convex. Let 𝛼∶ [𝑎, 𝑏] → R 2 be such a curve.
Then the function 𝜑(𝑠) is non-decreasing and 𝜑(𝑏) = 𝜑(𝑎) + 2𝜋.
Given 𝑠0 ∈ [𝑎, 𝑏], we want to prove that the function ℎ(𝑠) = ⟨𝛼(𝑠) − 𝛼(𝑠0 ), n(𝑠0 )⟩
does not change sign. Otherwise ℎ(𝑠) reaches a positive maximum and a negative
minimum at points 𝑠1 , 𝑠2 ∈ [𝑎, 𝑏] ∖ {𝑠0 }, and at each of these points the tangent line
is parallel to the tangent line at 𝛼(𝑠0 ). Hence, there exist 𝑖 ≠ 𝑗 ∈ {0, 1, 2} such that
𝜑(𝑠𝑖 ) = 𝜑(𝑠 𝑗 ) and, 𝜑 being non-decreasing, this is only possible if 𝜑 is constant in
the interval between 𝑠𝑖 and 𝑠 𝑗 . This means that the curve contains the line segment
from 𝛼(𝑠𝑖 ) to 𝛼(𝑠 𝑗 ), and therefore the tangent lines at these points coincide — which
is absurd given the way they were chosen. Therefore ℎ does not change sign and 𝛼 is
convex.
(ii) We assume now and until the end of the proof that 𝛼∶ [𝑎, 𝑏] → R 2 is a convex
curve. Let us first see that 𝑘(𝑠) does not change sign. Consider the function of two
variables (𝑠, 𝑡) ∈ [𝑎, 𝑏] × [𝑎, 𝑏] defined by

𝐻(𝑠, 𝑡) = ⟨𝛼(𝑠) − 𝛼(𝑡), n(𝑡)⟩.

By hypothesis, for each 𝑡 ∈ [𝑎, 𝑏] the function ℎ𝑡 given by ℎ𝑡 (𝑠) = 𝐻(𝑠, 𝑡) has constant
sign — that is, the restriction of 𝐻 to each horizontal line segment [𝑎, 𝑏]×{𝑡} is either
non-negative or non-positive. What we want for now is to prove that the function 𝐻
itself does not change sign.
Assume, instead, that there exist (𝑠0 , 𝑡 0 ), (𝑠1 , 𝑡 1 ) ∈ [𝑎, 𝑏] × [𝑎, 𝑏] such that
𝐻(𝑠0 , 𝑡 0 ) < 0 < 𝐻(𝑠1 , 𝑡 1 ); and let us agree that 𝑡0 < 𝑡 1 . Consider the set

𝐴 = {𝑡ˆ ≥ 𝑡 0 ∶ 𝐻(𝑠, 𝑡) ≤ 0 ∀(𝑠, 𝑡) ∈ [𝑎, 𝑏] × [𝑡 0 , 𝑡ˆ]};

𝐴 is a non-empty interval (because 𝑡0 ∈ 𝐴) and closed (because 𝐻 is continuous). By

continuity of 𝐻, there exists 𝛿 > 0 such that 𝐻(𝑠0 , 𝑡) < 0 whenever 𝑡0 ≤ 𝑡 ≤ 𝑡 0 + 𝛿,
1.5 Convex Curves 17

and therefore 𝐴 contains the interval [𝑡0 , 𝑡 0 + 𝛿]. Since 𝑡1 ∉ 𝐴, 𝑡2 = sup(𝐴) lies in
]𝑡 0 , 𝑡 1 [. Since 𝑡 1 ∈ 𝐴, we have 𝐻(𝑠, 𝑡 1 ) ≤ 0 for all 𝑠 ∈ [𝑎, 𝑏]. If for all 𝑠 this inequality
were strict, then the same argument as above would show that there exists 𝛿 > 0 such
that [𝑡 2 , 𝑡 2 + 𝛿] ⊂ 𝐴, which contradicts the definition of 𝑡 2 . We then conclude that
𝐻(𝑠, 𝑡 2 ) = 0 for all 𝑠 ∈ [𝑎, 𝑏] — which is absurd because it means that the curve 𝛼 is
all contained in a straight line.
So we have proved that 𝐻 does not change sign. That the curvature does not either
is now immediate: when 𝑘(𝑡0 ) > 0, the function ℎ𝑡0 (𝑠) has a strict local minimum for
𝑠 = 𝑡 0 (since ℎ′𝑡0 (𝑡0 ) = 0 and ℎ′′𝑡0 (𝑡0 ) = 𝑘(𝑡 0 )) and therefore 𝐻(𝑠, 𝑡 0 ) > 0 for 𝑠 near 𝑡 0 ;
and, conversely, when 𝑘(𝑡0 ) < 0 one has 𝐻(𝑠, 𝑡 0 ) < 0 for 𝑠 near 𝑡0 .

(iii) Let us now prove that 𝛼 is simple. Let us assume, to the contrary, that it has
some self-intersection which, changing if necessary the initial point, we suppose to
take place at 𝛼(𝑎). Then there exists 𝑥 ∈ ]𝑎, 𝑏[ such that 𝛼(𝑐) = 𝛼(𝑎).
The function 𝐻(𝑠, 𝑡) vanishes for (𝑠, 𝑡) = (𝑎, 𝑐) and, by (ii), reaches at that point
a local extremum. Thus ⟨𝝉(𝑎), n(𝑐)⟩ = 𝜕𝐻 𝜕𝑠 ∣ = 0, and therefore 𝝉(𝑎) = ±𝝉(𝑐). If
(𝑎,𝑐)
it were 𝝉(𝑎) = −𝝉(𝑐), we would have 𝐻(𝑠, 𝑎) = −𝐻(𝑠, 𝑐) for all 𝑠 ∈ [𝑎, 𝑏], which is
impossible by (ii). Therefore we have 𝝉(𝑎) = 𝝉(𝑐).
To simplify the notation, we assume that 𝛼(𝑎) = 𝛼(𝑐) = (0, 0). Let us put, (as in
the proof of 1.4.1) for 𝑠 ≥ 0,

𝑔(𝑠) = ⟨𝛼(𝑎 + 𝑠), 𝝉(𝑎)⟩, ˜

𝑔(𝑠) = ⟨𝛼(𝑐 + 𝑠), 𝝉(𝑎)⟩,
𝐹(𝑠) = ⟨𝛼(𝑎 + 𝑠), n(𝑎)⟩, 𝐹(𝑠) = ⟨𝛼(𝑐 + 𝑠), n(𝑎)⟩,
˜
𝑓 (𝑡) = 𝐹 ○ 𝑔 −1 (𝑡), 𝑓˜(𝑡) = 𝐹˜ ○ 𝑔˜ −1 (𝑡)

– where the functions 𝑓 and 𝑓˜ are defined on some interval [0, 𝜀], 𝜀 > 0. The graphs
of 𝑓 and 𝑓˜ are portions of the trace of 𝛼: in fact, putting 𝑠 = 𝑔 −1 (𝑡), 𝑠˜ = 𝑔˜ −1 (𝑡), we
can write

𝛼(𝑎 + 𝑠) = 𝑡𝝉(𝑎) + 𝑓 (𝑡)n(𝑎),

𝛼(𝑐 + 𝑠˜) = 𝑡 𝜏(𝑎)
˜ + 𝑓˜(𝑡)n(𝑎).

From these formulas we obtain

𝐻(𝑎 + 𝑠, 𝑐 + 𝑠˜) = { 𝑓 (𝑡) − 𝑓˜(𝑡)}⟨n(𝑎), n(𝑐 + 𝑠˜)⟩,

𝐻(𝑐 + 𝑠˜, 𝑎 + 𝑠) = { 𝑓˜(𝑡) − 𝑓 (𝑡)}⟨n(𝑡), n(𝑎 + 𝑠)⟩.

In each of these products, and since n(𝑎) = n(𝑐), the second factor is positive for 𝑠,
𝑠˜ sufficiently small; hence, if it were 𝑓 (𝑡) ≠ 𝑓˜(𝑡) for some 𝑡 ∈ [0, 𝜀], 𝐻(𝑎 + 𝑠, 𝑐 + 𝑠˜)
and 𝐻(𝑐 + 𝑠˜, 𝑎 + 𝑠) would have opposite signs, in contradiction to (ii). We thus have
𝑓 (𝑡) = 𝑓˜(𝑡) for all 𝑡 ∈ [0, 𝜀] — and from this, since 𝛼 is parametrized by arc length,
we conclude that there exists 𝛿 > 0 such that 𝛼(𝑎 + 𝑠) = 𝛼(𝑐 + 𝑠) for all 𝑠 ∈ [0, 𝛿]. A
18 1 Differentiable Curves

trivial argument now proves that for all 𝑠 ≥ 0 one has 𝛼(𝑎 + 𝑠) = 𝛼(𝑐 + 𝑠), and this
says that when the curve returns to the starting point, it repeats the same path from
then on. The given hypothesis thus leads us to conclude that 𝛼∣[𝑎,𝑏] gives more than
one turn to the same closed curve. Assuming this does not happen, such a 𝑐 ∈ ]𝑎, 𝑏[
with 𝛼(𝑐) = 𝛼(𝑎) does not exist and the curve is simple.
(iv) Let us now show that ℜ(𝛼) = ±1. Assuming that 𝛼 is positively oriented,
and putting 𝜑(𝑏) = 𝜑(𝑎) + 2𝑛𝜋, we have 𝑛 = ℜ(𝛼) ≥ 1, and we want to see that
𝑛 = 1. We can assume, without loss of generality, that 𝑘(𝑎) > 0. Take 𝑐 ∈ [𝑎, 𝑏]
such that 𝜑(𝑐) = 𝜑(𝑎) + 2𝜋: then 𝐻(𝑎, 𝑐) = −𝐻(𝑐, 𝑎), and it follows by (ii) that
𝐻(𝑎, 𝑐) = 0, which means that the tangent lines at 𝛼(𝑎) and 𝛼(𝑐) coincide. The
function 𝜆(𝑠) = 𝐻(𝑐, 𝑎 + 𝑠) reaches a minimum at 0, and so

0 = 𝜆′ (0) = −𝑘(𝑎)⟨𝛼(𝑐) − 𝛼(𝑎), 𝝉(𝑎)⟩,

whence it follows, since 𝑘(𝑎) > 0 and the points 𝛼(𝑐) and 𝛼(𝑎) lie on a straight line
parallel to 𝝉(𝑎), that 𝛼(𝑐) = 𝛼(𝑎). Since 𝛼 is simple, we must have be 𝑐 = 𝑏 and
therefore ℜ(𝛼) = 1. ◻

It is important to note that any simple curve, whether convex or not, has rotation
index ±1: this is what the rotation index theorem says, the proof of which we give in
the Appendix to Chap. 4, but of which a special case is given in Exercise 16 below. If
we already had this result, the proof of 1.5.1 would be somewhat simplified; another
simplification would be to suppress step (iii) if, as some authors do, we already
required in the definition that a convex curve be simple.
One result we will not prove, but of which we will make important use, not always
explicit, is the Jordan curve theorem. This theorem states that any simple closed
curve divides the plane into two disjoint connected open subsets of which it is a
common boundary. (For a proof of the theorem in the differentiable case, and its
generalization to higher dimensions, see [15]; for the topological version we suggest
[17], which also includes a proof of Schönflies’ theorem: the region bounded by a
simple closed curve is homeomorphic to an open disc).
To finish this section we mention that a convex curve of nonzero curvature at all
its points is usually called strictly convex. In this case 𝜑(𝑠) is strictly monotone and
therefore every tangent line touches the curve at a single point.

Exercises
14. Show that if a line intersects a closed convex curve then one and only one of the
following cases occurs: either the line is tangent to the curve, or it intersects the curve
at exactly two points.
15. Let 𝛼 be a closed, simple, regular curve, Ω be the open set bounded by 𝛼, and
Ω = Ω ∪ 𝛼 be the closure of 𝛼. Show that the following conditions are equivalent:

(i) Ω is a convex set (i.e., 𝑝, 𝑞 ∈ Ω ⇒ the line segment [𝑝, 𝑞] ⊆ Ω);

1.6 Curves of Constant Width 19

(ii) 𝛼 is a convex curve.

(Suggestion for (i) ⇒ (ii): Show that for each 𝑠0 , the image of the function 𝑠
(𝑠 ≠ 𝑠0 ) ↦ (𝛼(𝑠) − 𝛼(𝑠0 ))/∣𝛼(𝑠) − 𝛼(𝑠0 )∣ is contained in a semicircle.)

16. (a) Let H∶ [0, 1] × [𝑎, 𝑏] → R 2 be a differentiable mapping such that every
𝛼𝑠 = H(𝑠, ⋅ ) is a regular closed curve. Show that ℜ(𝛼𝑠 ) is constant. (Exercise 7 may
be helpful.)
(b) Let 𝛼∶ [𝑎, 𝑏]→R 2 be a regular closed curve and 𝑝 ∈ R 2 ∖ 𝛼([𝑎, 𝑏]) such that
each half-line with origin at 𝑝 intersects 𝛼 exactly once, and this intersection is
transverse (i.e., the half-line is not tangent to 𝛼 at the point of intersection). Prove
that ℜ(𝛼) = ±1.
(c) Now assume only that all intersections of 𝛼 with half-lines 𝑟 of origin 𝑝 are
transversal. Show that the cardinal of the set 𝑇(𝑟) = {𝑡 ∈ [𝑎, 𝑏[∶ 𝛼(𝑡) ∈ 𝑟} is the same
for all such half-lines.

1.6 Curves of Constant Width

In this section, we explore the varying width of a planar curve. The width of the curve
in any given direction is the narrowest distance between two lines perpendicular to
that direction that can contain the curve. This means that the width of a curve is not
necessarily the same in all directions. Remarkably, besides the circle, there are other
convex curves that have a constant width regardless of direction; and the perimeter of
such curves is equal to that of the circle of the same width.
We will deal in this section with the width of a planar curve. The width of a curve
in a given direction is the minimal width among the strips that contain the curve and
are bounded by lines orthogonal to that direction.
Given a closed curve 𝛼∶ [𝑎, 𝑏] → R 𝑛 we define, for v ∈ S1 , ℎ(v) = max ⟨𝛼(𝑠), v⟩.
𝑎≤𝑠≤𝑏
Since the maximum of ⟨𝛼(𝑠), v⟩ is only reached at points 𝑠 such that ⟨𝝉(𝑠), v⟩ = 0,
ℎ(v) is the maximum among the (oriented) distances from the origin to the tangent
lines to 𝛼 that are orthogonal to v. The width of 𝛼 in the direction of v is L(v) =
ℎ(v) + ℎ(−v).
If 𝛼 is a convex curve, then there are exactly two tangent lines to 𝛼 that are
orthogonal to v (although each of them may be tangent to 𝛼 at more than one point),
and L(v) is the distance between these lines (see Fig. 1.7). For example, the width of
a circle is, in all directions, equal to its diameter.

Proposition 1.6.1 In any closed curve the diameter and maximum width are equal.
20 1 Differentiable Curves

L( v ) v

Figure 1.7
1.7
Proof We denote the diameter of 𝛼 by 𝐷 = max{∣𝛼(𝑠) − 𝛼(𝑡)∣ ∶ 𝑠, 𝑡 ∈ [𝑎, 𝑏], and by
𝔏 = max{L(v) ∶ v ∈ S1 } the maximum width of 𝛼.
Let us check that 𝐷 ≤ 𝔏. Consider the function D(𝑠, 𝑡) = ∣𝛼(𝑠) − 𝛼(𝑡)∣. This
function reaches its maximum 𝐷 on a pair of points (𝑠0 , 𝑡 0 ) such that

𝜕D 𝜕D
∣ = ∣ = 0,
𝜕𝑠 (𝑠0 ,𝑡0 ) 𝜕𝑡 (𝑠0 ,𝑡0 )

conditions which are translated by the equalities

⟨𝝉(𝑠0 ), 𝛼(𝑠0 ) − 𝛼(𝑡0 )⟩ = ⟨𝝉(𝑡0 ), 𝛼(𝑠0 ) − 𝛼(𝑡0 )⟩ = 0.

This shows that the tangent lines at 𝛼(𝑠0 ) and 𝛼(𝑡 0 ) are parallel, both being orthogonal
to the line segment joining 𝛼(𝑠0 ) to 𝛼(𝑡 0 ). Moreover the curve is completely contained
in the strip between these tangent lines, otherwise the maximum distance between
distinct points of 𝛼 would exceed 𝐷. Thus the width of the strip, which is 𝐷, is also
equal to L(n(𝑠0 )), and thus 𝐷 ≤ 𝔏.
Let us now deal with the opposite inequality. Given v ∈ S1 , let 𝑠0 , 𝑡 0 ∈ [𝑎, 𝑏] be
such that ℎ(v) = ⟨𝛼(𝑠0 ), v)⟩ and ℎ(−v) = ⟨𝛼(𝑡0 ), −v⟩. Then

∣𝛼(𝑠0 ) − 𝛼(𝑡0 )∣ ≥ ⟨𝛼(𝑠0 ) − 𝛼(𝑡 0 ), v⟩ = L(v),

and therefore 𝐷 ≥ L(v). Since this inequality holds for all v, it follows that 𝐷 ≥ 𝔏. ◻

Let us now assume that the curve 𝛼 is convex and has constant width 𝔏. By 1.6.1,
also its diameter is equal to 𝔏. Let us now see that this diameter is realized by many
pairs of points on the curve.
Fixing 𝑠0 ∈ [𝑎, 𝑏], let 𝛼(𝑠1 ) be a point such that 𝝉(𝑠1 ) = −𝝉(𝑠0 ). We know from
the analysis done in Section 1.5 that the curve is contained in the strip bounded by
tangent lines at 𝛼(𝑠0 ) and 𝛼(𝑠1 ). Hence these tangent lines are at a distance 𝔏 from
each other, and therefore ∣𝛼(𝑠0 ) − 𝛼(𝑠1 )∣ ≥ 𝔏. Since the diameter of 𝛼 is 𝔏, it must
be ∣𝛼(𝑠0 ) − 𝛼(𝑠1 )∣ = 𝔏, an equality that is only possible if the line segment between
1.6 Curves of Constant Width 21

𝛼(𝑠0 ) and 𝛼(𝑥 1 ) is orthogonal to the tangent lines to 𝛼 at these points. Furthermore,
there is no other point 𝛼(̃ 𝑠1 ) such that ∣𝛼(𝑠0 ) − 𝛼(𝑠˜1 )∣ = 𝔏, for the proof of 1.6.1
shows that 𝛼(̃ 𝑠1 ) would also be on the normal to 𝛼 at 𝛼(𝑠0 ).
We conclude that for every point 𝑝 of a convex curve of constant width 𝔏, there is
a single point 𝑝̃ of the curve at the maximum distance 𝔏 from 𝑝, and 𝑝̃ is situated
on the normal to 𝛼 at 𝑝. We call this point 𝑝̃ the antipode of 𝑝. Assuming that the
curve is positively oriented (and therefore has non-negative curvature at all points),
our conclusion translates into 𝑝̃ = 𝑝 + 𝔏n(𝑝). We stress that “being antipodal to” is a
reflexive relation, and that two tangent lines to 𝛼 that are parallel and distinct meet 𝛼
at points that are antipodes of each other.
Consider now the circle C with center 𝑝̃ and radius 𝔏. Such a circle is tangent to
𝛼 at the point 𝑝; and all other points of 𝛼 are contained in the interior of the disk
bounded by C. Exercise 8 then says that the absolute value of the curvature of 𝛼 at 𝑝
is greater than or equal to 1/𝔏. We thus conclude that any convex curve of constant
width is strictly convex.

Example 1.6.2 At this point it is good to wonder about the existence of constant-width
convex curves that are not circles. The above discussion suggests that such a curve is
determined by knowing the arc between two antipodal points 𝑝 and 𝑝̃: the remaining
segment is found by marking, from each point of this arc, a distance of 𝔏 along the
normal.

To construct an example where 𝔏 = 2, we consider a curve 𝛼∶ [0, 𝑐] → R 2 with

the following properties:

(i) 𝛼 is parametrized by arc length;

(ii) 𝛼(0) = (1, 0), 𝛼(𝑐) = (−1, 0), and there exists 𝜀 > 0 such that 𝛼([0, 𝜀] ∪ [𝑐 −
𝜀, 𝑐]) ⊆ S1+ = {(𝑥, 𝑦) ∈ S1 ∶ 𝑦 ≥ 0};
(iii) the trace of 𝛼 is not contained in S+1 ;
(iv) the tangent vector 𝝉(𝑠), 𝑠 ∈ [0, 𝑐], describes a semicircle;
1
(v) 𝑘(𝑠) > for all 𝑠 ∈ [0, 𝑐].
2
Such a curve can be obtained by considering S1+ as the graph of a function
[−1, 1] → R and adding to that function a non-constant function of class 𝐶 ∞ that
is identically zero in the intervals [−1, −1 + 𝛿] and [1 − 𝛿, 1], for some 𝛿 > 0. (See
Exercise 17 for the existence of functions with these properties.) Reparameterizing
such that the graph of the resulting function starts at (1, 0), we obtain a curve 𝛼 that
verifies conditions (i)-(iv). If the added function and its first and second derivatives
are close to zero then the curvature of 𝛼 is close to that of S1+ , which guarantees (v).
We define 𝛽∶ [0, 2𝑐] → R 𝑛 by 𝛽(𝑡) = 𝛼(𝑡) for 0 ≤ 𝑡 ≤ 𝑐, and 𝛽(𝑡) = 𝛼(𝑡 − 𝑐) +
2n(𝑡 − 𝑐) for 𝑐 ≤ 𝑡 ≤ 2𝑐. Condition (ii) guarantees that 𝛽 is well-defined at 𝑡 = 𝑐
and that 𝛽(2𝑐) = 𝛽(0). For 𝑐 ≤ 𝑡 ≤ 2𝑐 we have 𝛽′ (𝑡) = {1 − 2𝑘(𝑡 − 𝑐)}𝝉(𝑡 − 𝑐) and,
by (v), this vector never vanishes. Furthermore conditions (ii) and (i) imply that, in
neighborhoods of the “gluing points” (1, 0) and (−1, 0), the curve 𝛽 runs through
arcs of S1 and is parametrized by arc length. Therefore 𝛽 is a regular closed curve.
22 1 Differentiable Curves

a(t-c)

(-1,0) (1,0)

b (t)

Figure 1.8

Noting that we have

𝑘(𝑡 − 𝑐)
𝑘(𝑡) = ,
2𝑘(𝑡 − 𝑐) − 1
for 𝑐 ≤ 𝑡 ≤ 2𝑐, if 𝑣(𝑡) = 2𝑘(𝑡 − 𝑐) − 1 and 𝝉(𝑡) = −𝝉(𝑡 − 𝑐) and this equality shows
that the curvature of 𝛽 is positive at all points. Finally, by (iv), the vector 𝝉(𝑡) turns
exactly once around the circumference when 𝑡 runs through [0, 2𝑐], and therefore
ℜ(𝛽) = 1. By Theorem 1.5.1, 𝛽 is convex. That 𝛽 has constant width equal to 2 is
now an immediate exercise. ◻
We now state the most important result of this section, originally proved by E.
Barbier in the 19th century using probabilistic methods (see [1]; Barbier’s proof is
also reproduced in [5], pp. 161-163).
Theorem 1.6.3 The perimeter of any constant-width curve 𝔏 is equal to 𝜋𝔏.
Proof We fix a curve 𝛼∶ [0, 𝐿] → R 2 that is convex and positively oriented and has
constant width 𝔏. We assume that the parameter of the curve is given by arc length,
so that its perimeter is 𝐿, and we consider 𝛼 defined on R through its periodic
extension. We further denote by 𝜑(𝑠) a differentiable choice of the angle of 𝝉(𝑠)
with the positive 𝑥-half-axis. The rotation index of 𝛼 is 1, so 𝜑(𝑠 + 𝐿) = 𝜑(𝑠) + 2𝜋
for all 𝑠 ∈ R ; and, since 𝛼 is strictly convex, 𝜑∶ R → R is strictly increasing and has
differentiable inverse.
For each 𝑠 ∈ R , we denote the antipode of 𝛼(𝑠) by 𝛼 ̃(𝑠). This function 𝑠 ↦ 𝛼 ̃(𝑠)
is also periodic with period equal to 𝐿; and is differentiable, since we can write
̃(𝑠) = 𝛼(𝑠) + 𝔏n(𝑠).
𝛼
Lemma 1.6.4 There exists a differentiable function 𝑓 ∶ R → R such that 𝛼 ̃(𝑠) =
𝛼( 𝑓 (𝑠)) for all 𝑠 ∈ R . This function satisfies 𝑓 (𝑠 + 𝐿) = 𝑓 (𝑠) + 𝐿, and its derivative
is strictly positive at all points.
1.6 Curves of Constant Width 23

Proof Assume that a certain function 𝑓 satisfies the equation 𝜑 ○ 𝑓 (𝑠) = 𝜑(𝑠) + 𝜋.
Then 𝝉( 𝑓 (𝑠)) = (cos(𝜑 ○ 𝑓 (𝑠)), sin((𝜑 ○ 𝑓 (𝑠))) = −𝝉(𝑠), and therefore the points
𝛼(𝑠) and 𝛼( 𝑓 (𝑠)) are antipodes of each other — that is, 𝛼 ̃(𝑠) = 𝛼( 𝑓 (𝑠)) just as we
intend.
This means that we just have to find 𝑓 such that 𝜑○ 𝑓 (𝑠) = 𝜑(𝑠)+𝜋. Such a function
is given by 𝑓 (𝑠) = 𝜑−1 (𝜑(𝑠) + 𝜋), which is differentiable and has positive derivative.
Furthermore, we have 𝑓 (𝑠+𝐿) = 𝜑−1 (𝜑(𝑠+𝐿)+𝜋) = 𝜑−1 ({𝜑(𝑠)+𝜋}+2𝜋) = 𝑓 (𝑠)+𝐿,
as we want. ◻

We now finish the proof of 1.6.3. Differentiating the equality 𝛼(𝑠) + 𝔏n(𝑠) =
𝛼( 𝑓 (𝑠)), we obtain {1−𝔏𝑘(𝑠)} 𝝉(𝑠) = 𝑓 ′ (𝑠)𝝉( 𝑓 (𝑠)) — and from this, as 𝝉( 𝑓 (𝑠)) =
−𝝉(𝑠), yields 𝑓 ′ (𝑠) = −1 + 𝔏𝑘(𝑠). Finally, using 1.6.4, and since the rotation index
of 𝛼 is 1, we have
𝐿 𝐿
𝐿 = 𝑓 (𝐿) − 𝑓 (0) = ∫ 𝑓 ′ (𝑠) 𝑑𝑠 = −𝐿 + 𝔏 ∫ 𝑘(𝑠) 𝑑𝑠 = −𝐿 + 2𝜋𝔏.
0 0

Note 1.6.5 We cannot omit the simplest example of a non-circular curve of constant
width: Reuleaux’s triangle, which is formed by three arcs, each centered at one of the
vertices of an equilateral triangle ABC and radius equal to the side of the triangle. Its
perimeter 𝐿 and width 𝔏 are also related by 𝐿 = 𝜋𝔏, but the proof of 1.6.3 does not
⌢ ⌢ ⌢
cover this case: the antipode of each point of the arc 𝐵𝐶 (resp. 𝐶 𝐴, 𝐴𝐵) is the point
𝐴 (resp. 𝐵, 𝐶).

C A

Figure 1.9

The Reuleaux triangle is a piecewise regular curve; we say that 𝛼∶ [𝑎, 𝑏] → R 2

is such a curve if there exists a partition 𝑎 = 𝑡0 < 𝑡1 ⋯ < 𝑡 𝑘 = 𝑏 of [𝑎, 𝑏] such that
each restriction 𝛼∣(𝑡 −𝑡 ] is regular. We now sketch how to extend 1.6.3 to piecewise
𝑖−1 𝑖
regular convex closed curves, provided that each regular segment is at least of
class 𝐶 2 .
Given 𝑑 > 0, let 𝛼𝑑 be the curve which surrounds 𝛼 at a constant distance from it
equal to 𝑑. The curve 𝛼𝑑 is called parallel to 𝛼, and if 𝛼 has constant width 𝔏, 𝛼𝑑 has
24 1 Differentiable Curves

constant width 𝔏 + 2𝑑. In Fig. 1.9 we show a curve parallel to the Reuleaux triangle.
Each “corner” of 𝛼 is replaced by an arc of a circle at 𝛼𝑑 . Therefore 𝛼𝑑 already has a
well-defined tangent vector 𝝉 𝑑 (𝑠) at each point; the angle 𝜑 𝑑 (𝑠) that 𝝉 𝑑 (𝑠) makes
with the 𝑥-axis is a continuous, strictly increasing function, which is piecewise 𝐶 1 ;
and the antipode mapping is already a bijection of the curve onto itself. The proof of
1.6.3 can easily be adapted to show that the perimeter of 𝛼𝑑 is

𝑙(𝛼𝑑 ) = 𝜋(𝔏 + 2𝑑)

– and, letting 𝑑 → 0, we obtain 𝑙(𝛼) = 𝜋𝔏, as desired. ◻

There are numerous results on constant width curves: [7] contains a careful
discussion of the topic (and its generalization to higher dimensions) and an extensive
bibliography. In Section 5.5 we give some results on constant width surfaces.

Exercises

17. (a) Check that the function 𝑓 (𝑥) = 𝑒 −1/𝑥(1−𝑥) if 0 < 𝑥 < 1, 𝑓 (𝑥) = 0 otherwise it
𝑥 1
is 𝐶 ∞ , and that 𝐹(𝑥) = ∫ 𝑓 (𝑡) 𝑑𝑡/ ∫ 𝑓 (𝑡) 𝑑𝑡 satisfies the conditions: 𝐹(𝑥) = 0
0 0
for 𝑥 < 0, 𝐹 is strictly increasing on [0, 1], 𝐹(𝑥) = 1 for 𝑥 ≥ 1.
(b) Given 𝑎 < 𝑏 and 𝑦 1 , 𝑦 2 ∈ R , show that there exists a nonconstant 𝑔∶ R → R of
class 𝐶 ∞ such that 𝑔(𝑥) = 𝑦 1 ∀ 𝑥 ∈ [−∞, 𝑎], and 𝑔(𝑥) = 𝑦 2 ∀ 𝑥 ∈ [𝑏, +∞[.

18. Let 𝛼∶ [𝑎, 𝑏] → R 2 be a closed curve and L(v) be the corresponding “width
function”. Show that:
(a) L is continuous and therefore there exists max L(v);
v∈S
(b) if 𝛼 is regular and strictly convex then L is differentiable (i.e., the function
𝜃 → L(cos 𝜃, sin 𝜃) is differentiable).

19. Show that the function of Lemma 1.6.4 is unique but for the addition of a constant.

20. Let 𝛼 be a regular convex curve of constant width 𝔏. Show that:

(a) 𝑘(𝑝) > 1/𝔏 for any point 𝑝 of 𝛼 (assume that 𝛼 has positive orientation);
(b) if 𝑝 and 𝑝̃ are antipodal points then
1 1
+ = 𝔏;
𝑘(𝑝) 𝑘( 𝑝̃)

(c) if each pair of antipodal points divide 𝛼 into two arcs of equal length then 𝛼 is a
circle.

21. Convex curves of constant width 𝔏 are characterized by the fact that all the
rectangles that circumscribe them are squares of side 𝔏. In this exercise we prove a
generalization of Barbier’s Theorem: if 𝛼 is a regular curve strictly convex such that
all rectangles that surround it have perimeter 4𝔏, then the perimeter of 𝛼 is 𝜋𝔏.
1.7 Theorem of the Four Vertices 25

Using the same notation as in the proof of 1.6.3, denote by 𝐿 the perimeter of the
curve and by 𝛼 ̃(𝑠) the only point of 𝛼 at which the tangent vector is −𝝉(𝑠). Show
that:
̃(𝑠) = 𝛼( 𝑓 (𝑠)) and
(a) there exists a differentiable function 𝑓 ∶ R → R such that 𝛼
𝑓 (𝑠 + 𝐿) = 𝑓 (𝑠) + 𝐿;
(b) there exist differentiable functions 𝜆, 𝜂 such that

̃(𝑠) = 𝛼(𝑠) + 𝜆(𝑠)𝝉(𝑠) + 𝜂(𝑠)n(𝑠);

𝛼

(c) 𝜆 and 𝜂 are periodic functions of period 𝐿;

(d) 𝑓 ′ (𝑠) = −1 − 𝜆′ (𝑠) + 𝜂(𝑠)𝑘(𝑠);
𝐿
(e) 2𝐿 = ∫ 𝜂(𝑠)𝑘(𝑠) 𝑑𝑠.
0
Now make the change of variable 𝜃 = 𝜑(𝑠) to prove that
𝐿 2𝜋
∫ 𝜂(𝑠)𝑘(𝑠) 𝑑𝑠 = ∫ 𝜂(𝜃) 𝑑𝜃,
0 0

where 𝜂(𝜃) denotes 𝜂(𝜑−1 (𝜃)). Note that 𝜂(𝜃) is periodic of period 2𝜋 and that the
𝜋
assumption about 𝛼 translates to 𝜂(𝜃) + 𝜂 (𝜃 + ) = 2𝔏. Finally, we can write
2
2𝜋 2𝜋 𝜋
4𝐿 = ∫ 𝜂(𝜃) 𝑑𝜃 + ∫ 𝜂 (𝜃 + ) 𝑑𝜃
0 0 2
2𝜋 𝜋
=∫ {𝜂(𝜃) + 𝜂 (𝜃 + )} 𝑑𝜃 = 4𝜋𝔏.
0 2

22. Modify 1.6.2 to give examples of curves that satisfy the hypothesis of Exercise 21
but do not have constant width.

1.7 Theorem of the Four Vertices

We now give the four-vertex theorem, which states that the curvature of any closed
convex curve has at least four critical points (this result is also valid for nonconvex
closed planar curves, but we will not prove it in such generality). This result is best
possible: a non-circular ellipse has exactly four vertices, which are its points of
intersection with the axes (see Exercise 6).
Let 𝛼∶ [𝑎, 𝑏] → R 2 be a regular closed curve, and 𝑘(𝑠) its curvature function. A
vertex of 𝛼 is a point 𝛼(𝑠0 ) such that 𝑘 ′ (𝑠0 ) = 0. This definition does not depend on
the parameterization and so we assume that 𝑠 is the arc length.

Theorem 1.7.1 Any closed convex curve has at least four vertices.

Proof We can assume that 𝑘(𝑠) has a finite number of critical points, because
otherwise there is nothing to be shown. The function 𝑘(𝑠) attains some maximum
26 1 Differentiable Curves

and some minimum — which, changing the starting point if necessary, we suppose
happen at the points 𝑠 = 𝑎 and 𝑠 = 𝑠0 ∈]𝑎, 𝑏[. By applying a rotation or translation we
can ensure that both points 𝛼(𝑎) and 𝛼(𝑠0 ) are on the 𝑥-axis.
Let us check that there are no other points of 𝛼 on the 𝑥-axis: for if there were
another one — let it be 𝑝 — then the tangent line to 𝛼 at the one of the three points
𝛼(𝑎), 𝛼(𝑠0 ) and 𝑝 which lies between the other two is the 𝑥-axis; otherwise there
would be points of 𝛼 on opposite sides of this tangent line, in contradiction to the
convexity of 𝛼. It follows that the tangent line at 𝛼(𝑎) and 𝛼(𝑠0 ) is also the horizontal
axis, and that (as in the proof of 1.5.1) the trace of 𝛼∣[𝑎,𝑠 ] is a line segment. This
0
however contradicts our assumption that 𝛼 has a finite number of vertices.
Putting 𝛼(𝑠) = (𝑥(𝑠), 𝑦(𝑠)), we then have that 𝑦(𝑠) never vanishes on the
intervals ]𝑎, 𝑠0 [ and ]𝑠0 , 𝑏[, taking on the second interval a sign opposite to the
one it takes on the first; and the same is true of 𝑘 ′ (𝑠) if we assume that 𝛼 has at
most two vertices. Under this assumption the function 𝑘 ′ (𝑠)𝑦(𝑠) then has constant
𝑏
sign, vanishing only at 𝑎, 𝑥0 and 𝑏, and so ∫ 𝑘 ′ (𝑠)𝑦(𝑠) 𝑑𝑠 ≠ 0. But, writing
𝑎
(𝑥 ′ (𝑠), 𝑦 ′ (𝑠)) = (cos 𝜑(𝑠), sin 𝜑(𝑠)), and using integration by parts and the equality
𝜑′ (𝑠) = 𝑘(𝑠), we have
𝑏 𝑏 𝑏
𝑘 ′ (𝑠)𝑦(𝑠) 𝑑𝑠 = 𝑘(𝑠)𝑦(𝑠)∣𝑎 − ∫ 𝑘(𝑠)𝑦 ′ (𝑠) 𝑑𝑠 = − ∫ 𝑘(𝑠)𝑦 ′ (𝑠) 𝑑𝑠
𝑏
∫
𝑎 𝑎 𝑎
𝑏
𝜑′ (𝑠) sin 𝜑(𝑠) 𝑑𝑠 = cos 𝜑(𝑠)∣𝑎 = 0.
𝑏
= −∫
𝑎

This contradiction shows that 𝑘 ′ (𝑠) changes sign on some intervals ]𝑎, 𝑠0 [ and ]𝑠0 , 𝑏[.
Since in each of them 𝑘 ′ (𝑠) has the same sign near the endpoints, we conclude that
𝑘 ′ (𝑠) changes sign at least twice in such an interval, which proves the theorem. ◻

The four-vertex theorem is still valid for non-convex curves. For a very elegant
geometric proof that also covers this generalization, we suggest [21].

1.8 The Isoperimetric Inequality

The isoperimetric inequality states that, among all planar curves with a given perimeter,
the circumference encompasses the largest area. The proof we give (by A. Hurwitz,
1902) makes essential use of the theory of Fourier series (see [11] for an introduction
to this theory). An elementary proof appears in [18], and [9] contains a generalization
of the isoperimetric inequality for convex bodies in dimensions greater than two.

Lemma 1.8.1 (Wirtinger) Let 𝑓 be a function of class 𝐶 1 , periodic of period 2𝜋,

2𝜋
such that ∫0 𝑓 (𝑡) 𝑑𝑡 = 0. Then
2𝜋 2𝜋
𝑓 ′ (𝑡) 𝑑𝑡 ≥ ∫
2 2
∫ 𝑓 (𝑡) 𝑑𝑡,
0 0
1.8 The Isoperimetric Inequality 27

and equality holds if and only if there exist 𝑎 and 𝑏 such that 𝑓 (𝑡) = 𝑎 cos 𝑡 + 𝑏 sin 𝑡.
Proof Be
𝑎0 ∞
𝑓 (𝑡) ∼ + ∑ (𝑎 𝑛 cos 𝑛𝑡 + 𝑏 𝑛 sin 𝑛𝑡)
2 𝑛=1
the Fourier series expansion of 𝑓 . Since 𝑓 ′ (𝑡) is continuous, its expansion is obtained
from that of 𝑓 (𝑡) by term-by-term differentiation, thus
∞
𝑓 ′ (𝑡) ∼ ∑ (𝑛 𝑏 𝑛 cos 𝑛𝑡 − 𝑛 𝑎 𝑛 sin 𝑛𝑡).
𝑛=1

2𝜋
Since ∫0 𝑓 (𝑡) 𝑑𝑡 = 𝜋 𝑎 0 , our hypothesis yields 𝑎 0 = 0. Using Parseval’s formula,
we have
2𝜋 ∞
1 2
∫ 𝑓 (𝑡) 𝑑𝑡 = ∑ (𝑎 2𝑛 + 𝑏 2𝑛 ),
𝜋 0 𝑛=1
2𝜋 ∞
1
𝑓 ′ (𝑡) 𝑑𝑡 = ∑ 𝑛2 (𝑎 2𝑛 + 𝑏 2𝑛 ).
2
∫
𝜋 0 𝑛=1

It follows that
2𝜋 2𝜋 ∞
𝑓 ′ (𝑡) 𝑑𝑡 − ∫
2 2
∫ 𝑓 (𝑡) 𝑑𝑡 = ∑ 𝜋(𝑛2 − 1)(𝑎 2𝑛 + 𝑏 2𝑛 ) ≥ 0,
0 0 𝑛=1

and equality only holds if 𝑎 𝑛 = 𝑏 𝑛 = 0 for all 𝑛 > 1. Since continuous functions are
determined by their Fourier expansion, this is equivalent to 𝑓 (𝑡) = 𝑎 1 cos 𝑡 + 𝑏 1 sin 𝑡.
◻
Theorem 1.8.2 (Isoperimetric inequality) Let 𝛼 be a simple regular closed curve of
perimeter 𝐿, bounding a region Ω of area 𝐴. Then

𝐿2
𝐴≤ ,
4𝜋
and equality holds only when 𝛼 is a circle.
We can rescale the figure using a homothety, so there is no loss of generality
if we suppose that 𝐿 = 2𝜋, and therefore 𝛼(𝑠) = (𝑥(𝑠), 𝑦(𝑠)), 𝑠 ∈ [0, 2𝜋]. With a
2𝜋
translation, we can achieve ∫0 𝑥(𝑠) 𝑑𝑠 = 0. Furthermore, we assume that 𝛼(𝑠) runs
through the boundary of Ω in the counterclockwise direction. Applying Green’s
theorem
𝜕𝑄 𝜕𝑃
(∫ 𝑃 𝑑𝑥 + 𝑄 𝑑𝑦 = ∬ ( − ) 𝑑𝑥𝑑𝑦)
𝜕Ω Ω 𝜕𝑥 𝜕𝑦
to the vector field (𝑃, 𝑄) = (0, 𝑥), we obtain
2𝜋
𝐴=∫ 𝑥𝑦 ′ 𝑑𝑠,
0
28 1 Differentiable Curves

and on the other hand

2𝜋
𝐿 = 2𝜋 = ∫ (𝑥 ′2 + 𝑦 ′2 ) 𝑑𝑠.
0
We can then write
2𝜋 2𝜋
(𝑥 ′2 − 𝑥 2 ) 𝑑𝑠 + ∫ (𝑥 − 𝑦 ′ ) 𝑑𝑠.
2
2(𝜋 − 𝐴) = ∫
0 0

The second integral of this sum is non-negative and, by Wirtinger’s lemma, so is the
first. We thus conclude, as desired, that 𝐴 ≤ 𝜋. To achieve equality both integrals
have to be zero, so 𝑦 ′ (𝑠) = 𝑥(𝑠) and 𝑥(𝑠) = 𝑎 cos 𝑠 + 𝑏 sin 𝑠. We thus have

𝑥(𝑠) = 𝑎 cos 𝑠 + 𝑏 sin 𝑥,

𝑦(𝑠) = 𝑎 sin 𝑠 − 𝑏 cos 𝑠 + 𝑐,

and 𝛼 is therefore the unit circle with center at (0, 𝑐). ◻

Exercises

23. The coefficients of the Fourier series

𝑎0 ∞
𝑓 (𝑡) ∼ + ∑ (𝑎 𝑛 cos 𝑛𝑡 + 𝑏 𝑛 sin 𝑛𝑡)
2 𝑛=1

of a periodic function of period 2𝜋, integrable on [0, 2𝜋] (a class that includes
bounded functions with a finite number of discontinuities), are defined by

1 2𝜋
𝑎0 = ∫ 𝑓 (𝑡) 𝑑𝑡,
𝜋 0
1 2𝜋
𝑎𝑛 = ∫ 𝑓 (𝑡) cos 𝑛𝑡 𝑑𝑡,
𝜋 0
1 2𝜋
𝑏𝑛 = ∫ 𝑓 (𝑡) sin 𝑛𝑡 𝑑𝑡.
𝜋 0

Show that:

+∞ +∞
(a) if 𝑓 is of class 𝐶 1 and (𝑎 ′𝑛 )𝑛=0 and (𝑏 ′𝑛 )𝑛=1 are the Fourier coefficients of 𝑓 ′ ,
then 𝑎 ′0 = 0 and 𝑎 ′𝑛 = 𝑛 𝑏 𝑛 and 𝑏 ′𝑛 = −𝑛 𝑎 𝑛 for 𝑛 ≥ 1 (use integration by parts);

(b) the result of (a) is still true if 𝑓 is only piecewise 𝐶 1 ;

(c) the isoperimetric inequality is valid for piecewise 𝐶 1 curves.

1.8 The Isoperimetric Inequality 29

Figure 1.10

24. Consider a straight line 𝑟 in the plane and a flexible string C of length 𝐿. By
placing C in the plane so that its ends are on 𝑟, we obtain a figure bounded by 𝑟 and by
C and whose area depends on the shape we give the string (see figure above). Show
that the figure of maximum area among all those so obtained is a semicircle based on
𝑟.
25. Given two points 𝑝 and 𝑞 in the plane and a flexible string C of length 𝐿 > ∣𝑝 − 𝑞∣,
determine the figure of largest area among those bounded by C and by the line segment
𝑝𝑞.
26. Let 𝛼 be a convex closed curve, piecewise 𝐶 1 , of perimeter 𝐿, bounding a region
Ω of area 𝐴. Let 𝑟 1 and 𝑟 2 be two parallel lines at a distance 𝑑 from each other such
that both touch 𝛼, and 𝛼 is contained in the strip bounded by the them. Consider an
orthonormal Cartesian coordinate system whose vertical axis is 𝑟 1 and whose origin
is the midpoint of the line segment 𝑟 1 ∩ 𝛼 (a line segment which may contain a single
point). There thus exist functions piecewise 𝐶 1 𝑓 , 𝑔∶ [0, 𝑑] → R such that:

r1 r2

(t,f(t))

0
W

(t,g(t))

Figure 1.11
30 1 Differentiable Curves

1. 𝑓 (𝑡) ≥ 𝑔(𝑡) ∀ 𝑡 ∈ [0, 𝑑], 𝑓 (0) = −𝑔(0) ≥ 0;

2. the trace of 𝛼 consists of the graphs of 𝑓 and 𝑔 and the vertical segments
{0} × [𝑔(0), 𝑓 (0)] and {𝑑} × [𝑔(𝑑), 𝑓 (𝑑)];
3. 𝑓 is a concave function and 𝑔 is a convex function.
1
(a) Define ℎ = ( 𝑓 − 𝑔). Show that ℎ is a concave function piecewise 𝐶 1 , and
2
conclude that the region Ω ̃ bounded by 𝑟 1 and 𝑟 2 and by the graphs of the functions ℎ
and −ℎ is convex.
(b) Let ̃𝐿 and 𝐴 ̃ Show that 𝐴
̃ be the perimeter and area of Ω. ̃ = 𝐴 and ̃ 𝐿 ≤ 𝐿, and
that the inequality is an equality only in the case where 𝑓 = −𝑔.
(c) Assume that 𝛼 has minimal perimeter among all convex piecewise 𝐶 1 curves
encompassing a fixed area 𝐴. Conclude, without using the isoperimetric inequality,
that 𝛼 is a circle.
Hint: 𝛼 has an axis of symmetry in each direction. Show that all these axes pass
through the same point.
27. Let 𝛼 be a regular convex curve of constant width 𝔏 that bounds a region of area
𝜋𝔏2
𝐴. Show that 𝐴 ≤ , with equality only if 𝛼 is a circle.
4
Chapter 2
Regular Surfaces

In this chapter we introduce regular surfaces, the object of all our further study,
defining them as those subsets of R 3 that can be described locally by two independent
parameters. We introduce notions such as tangent space, differentiable function and
diffeomorphism, and consider the problems of orientability and the measurement of
quantities (areas, lengths . . .) on surfaces.

2.1 Definition and Examples

We all have an intuitive notion of what a surface is, and any attempt to describe that
notion would inevitably fall into redundancy. We accept, however, that the plane is
the simplest surface of all, and that a good way to construct models of others is by
gluing together various pieces of paper. Our definition of surface is the mathematical
elaboration of this idea.
A subset 𝑆 of R 3 is called a regular surface if, for each 𝑝 ∈ 𝑆, there exist an open
neighborhood 𝑉 ⊆ R 3 of 𝑝, an open subset 𝑈 ⊆ R 2 , and a bijection Φ∶ 𝑈 → 𝑉 ∩ 𝑆
with the following properties:

i. Φ is of class 𝐶 ∞ ;
ii. Φ is a homeomorphism (i.e., its inverse Φ−1 ∶ 𝑉 ∩ 𝑆 → 𝑈 is continuous);
iii. for all 𝑞 ∈ 𝑈 the Jacobian matrix 𝐽Φ(𝑞) has rank two.

A mapping Φ with these three properties is named parameterization or system

of (local) coordinates of 𝑆. We usually denote the points of 𝑈 by (𝑢, 𝑣), so that 𝑢
and 𝑣 are local parameters of 𝑆, and the partial derivatives of Φ are denoted by Φ𝑢
and Φ𝑣 . These vectors describe the velocities of the coordinate curves, which are the
curves obtained by fixing one of the parameters and varying the other. Moreover the
columns of the matrix 𝐽Φ(𝑢, 𝑣) are precisely Φ𝑢 and Φ𝑣 , so that condition iii. above
expresses that, for each (𝑢, 𝑣) ∈ 𝑈, Φ𝑢 and Φ𝑣 are linearly independent.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 31

P. V. Araújo, Differential Geometry, https://doi.org/10.1007/978-3-031-62384-4_2
32 2 Regular Surfaces

F
F -1 (p) Fv
p

Fu
u S

Figure 2.1

A subset of the surface 𝑆 is called open if it is of the form 𝑉 ∩ 𝑆, where 𝑉 is

an open subset of R 3 . An important observation is that any point of 𝑆 has an open
neighborhood (in 𝑆) homeomorphic to a disk, and this neighborhood can be taken as
small as we wish: indeed, if (𝑈, Φ) is a parameterization in the neighborhood of 𝑝,
there exists some open disk 𝐷 (of arbitrarily small radius) containing Φ−1 (𝑝) and
contained in 𝑈, and Φ(𝐷) is the sought neighborhood.

Examples 2.1.1 A. Any plane Π in R 3 is a surface: in fact Π admits a description of

the form 𝜙(𝑢, 𝑣) = 𝑝 0 + 𝑢w1 + 𝑣w2 , where (𝑢, 𝑣) ∈ R 2 , 𝑝 0 ∈ Π and w1 , w2 are linearly
independent vectors. The conditions i. and iii. are trivially verified, and ii. follows
from the fact that the solution of the equations 𝜙(𝑢, 𝑣) = 𝑝, for 𝑝 = (𝑎, 𝑏, 𝑐) ∈ Π, is a
first degree function at 𝑎, 𝑏, 𝑐, hence continuous. This means that Π is all covered by
a single (called global) parameterization.
B. If 𝑓 ∶ 𝑈 → R 2 is a differentiable function defined on an open subset R 2 , its graph
{(𝑢, 𝑣, 𝑓 (𝑢, 𝑣))∶ (𝑢, 𝑣) ∈ 𝑈} ⊆ R 3 is a surface admitting the global parameterization
Φ(𝑢, 𝑣) = (𝑢, 𝑣, 𝑓 (𝑢, 𝑣)), (𝑢, 𝑣) ∈ 𝑈.
C. The union 𝑆 = Π1 ∪ Π2 of two non-parallel planes is not a surface, since the points
of Π1 ∩ Π2 have on 𝑆 no neighborhood homeomorphic to a disk.
D. A parameterization of the sphere S2 = {(𝑥, 𝑦, 𝑧) ∈ R 3 ∶ 𝑥 2 + 𝑦 2 + 𝑧2 = 1} covering
the northern hemisphere is
√
Φ(𝑢, 𝑣) = (𝑢, 𝑣, 1 − (𝑢 2 + 𝑣 2 )) ,

defined on the disk {(𝑢, 𝑣)∶ 𝑢 2 +𝑣 2 < 1}. With a few more analogous parameterizations
(how many are needed?) we can cover the whole sphere, which therefore is a surface.
2.1 Definition and Examples 33

p
q

j
y

Figure 2.2

Another type of parameterization of S2 , which excludes only one meridian, is given

by the spherical coordinates, which are the colatitude 𝜃 ∈ ]0, 𝜋[ and the longitude
𝜑 ∈ ] −𝜋, 𝜋[ (Fig. 2.2), defining the point

Ψ(𝜑, 𝜃) = (sin 𝜃 cos 𝜑, sin 𝜃 sin 𝜑, cos 𝜃).

The tangent vectors to the coordinate curves are

Ψ𝜑 (𝜑, 𝜃) = (− sin 𝜃 sin 𝜑, sin 𝜃 cos 𝜑, 0),

Ψ 𝜃 (𝜑, 𝜃) = (cos 𝜃 cos 𝜑, cos 𝜃 sin 𝜑, − sin 𝜃),

whose vector product

Ψ𝜑 × Ψ 𝜃 = −sin 𝜃 (sin 𝜃 cos 𝜑, sin 𝜃 sin 𝜑, cos 𝜃)

has length sin 𝜃, and so is nonzero: therefore Ψ𝜑 and Ψ 𝜃 are linearly independent.
We mention that some authors use, in spherical coordinates, the latitude 𝜃̃ =
− 𝜃 (𝜃̃ ∈ ]− , [) instead of the colatitude, thus obtaining
𝜋 𝜋 𝜋
2 2 2
̃
Ψ(𝜑, ̃ = (cos 𝜃̃cos 𝜑, cos 𝜃̃sin 𝜑, sin 𝜃).
𝜃) ̃

E. The sphere is a special case of a surface of revolution, which is obtained by rotating

a planar curve around an axis contained in the plane of the curve. Assuming that the
curve 𝛼(𝑣) = (𝜌(𝑣), 0, 𝑧(𝑣)) is defined on an open interval 𝐼, is a homeomorphism
onto its image, and that 𝜌(𝑣) > 0 for all 𝑣 ∈ 𝐼, the mapping

Φ(𝑢, 𝑣) = (𝜌(𝑣) cos 𝑢, 𝜌(𝑣) sin 𝑢, 𝑧(𝑣)),

34 2 Regular Surfaces

where (𝑢, 𝑣) ∈ ]−𝜋, 𝜋[ × 𝐼, is a parameterization of the surface obtained by rotating

𝛼 around the 𝑧-axis. To show that Φ−1 is continuous, we make use of the formula
𝑢 sin 𝑢
tg =
2 1 + cos 𝑢
𝑢
and of the fact that 𝑓 (𝑢) = tg is a diffeomorphism of ]−𝜋, 𝜋[ and R . The equality
2
Φ(𝑢, 𝑣) = (𝑥, 𝑦, 𝑧) is then equivalent to the combination of the two equalities

⎛ ⎞ √
𝑢 = 𝑓 −1 𝑣 = 𝛼−1 ( 𝑥 2 + 𝑦 2 , 𝑂, 𝑧),
𝑦
√ ,
⎝ 𝑥 + 𝑥2 + 𝑦2 ⎠

which proves that Φ−1 is continuous.

F. By analogy with the definition of a parametrized curve, we define parametrized
surface as a differentiable mapping Φ of a connected open subset of R 2 into R 3
whose Jacobian has rank two at all points. This definition does not require that Φ be
injective, and so its image, being allowed to have self-intersections, is not necessarily
a regular surface; but it may not be so even if Φ is injective, as the following example
shows:

1/2p 1/p

Figure 2.3

Let 𝛼(𝑢) = (𝑥(𝑢), 𝑦(𝑢)), 𝑢>0, be a simple regular curve of class 𝐶 ∞ that includes
the vertical line segment {0} × ]1, 1] and the graph of the function sin 1𝑥 , 𝑥 ∈ ]0, 𝜋1 ];
this curve accumulates in the neighborhood of {0} × ] −1, 1[ (see Fig. 2.3). The
trace 𝑆 of the parametrized surface Φ(𝑢, 𝑣) = (𝑥(𝑢), 𝑦(𝑢), 𝑣) is not a surface: if it
were, each 𝑝 ∈ 𝑆 would have arbitrarily small neighborhoods 𝑉 in R 3 with 𝑉 ∩ 𝑆
homeomorphic to disks; but that does not happen if 𝑝 ∈ {0} × ] −1, 1[ × R , because
𝑉 ∩ 𝑆 has infinitely many connected components for all sufficiently small 𝑉.
This means that not all parametrized surfaces define regular surfaces. But it is
also not easy, and in some cases not even possible, to describe regular surfaces as
2.1 Definition and Examples 35

parametrized surfaces (i.e., as the image of one single function Φ). There is in general
no reason to privilege a particular (even global) parameterization on a given surface.

In conclusion: surface for us means regular surface, and only in the exercises we
will mention parametrized surfaces.

Exercises

28. For each 𝑎 ∈ R , the polar coordinates Φ(𝜌, 𝜑) = (𝜌 cos 𝜑, 𝜌 sin 𝜑), with 𝜌 > 0
and 𝜑 ∈ ]𝑎 − 𝜋, 𝑎 + 𝜋[, define a parameterization of R 2 that excludes a half-line. (We
consider R 2 as a surface by identifying it with the plane R 2 × {0} ⊆ R 3 .)

29. Consider a helix parametrized by (cos 𝑡, sin 𝑡, 𝑡) (𝑡 ∈ R ). The helicoid is the set
formed by all (horizontal) lines connecting each point of the 𝑧-axis with the point of
the helix at the same height (see Fig. 2.4). Show that the helicoid is a regular surface.

Figure 2.4
36 2 Regular Surfaces

p (p)

Figure 2.5

30. Consider the sphere S2 = {(𝑥, 𝑦, 𝑧)∶ 𝑥 2 + 𝑦 2 + 𝑧 2 = 1}. The stereographic

projection is the mapping 𝜋∶ S2 /{(0, 0, 1)} → R 2 defined as follows: (𝜋(𝑝), −1)
is the intersection point of the plane 𝑧 = −1 with the line that contains the points
(0, 0, 1) and 𝑝 (see Fig. 2.5).
(a) Obtain an explicit formula for 𝜋. Show that 𝜋 is a bijection and that (R 2 , 𝜋 −1 )
is a parameterization of S2 .
(b) Conclude that there are two parameterizations of S2 whose union covers the
sphere.
(c) Is there any global parameterization of S2 (i.e., whose image is S2 )?
2
31. Consider in R 2 the circle 𝐶 = {(𝑥, 𝑦, 𝑧)∶ (𝑦 − 2) + 𝑧2 = 1, 𝑥 = 0}. Show that the
set T 2 which is obtained by rotating 𝐶 around the 𝑧-axis is a regular surface (the torus)
— a parameterization is given by Φ(𝑢, 𝑣) = ((2 + cos 𝑣) cos 𝑢, (2 + cos 𝑣) sin 𝑢, sin 𝑣),
where (𝑢, 𝑣) ∈ ]−𝜋, 𝜋[ × ]−𝜋, 𝜋[ (see Fig. 2.6).

Figure 2.6
2.2 Change of Parameters, Level Surfaces 37

32. The tractrix is the planar curve obtained as follows: let us fix a line (let it be
the 𝑧-axis); the distance from any point 𝑝 on the curve to the point of intersection
𝑝 ′ of the tangent line to the curve at 𝑝 with the fixed line is constant, equal to
𝐶 > 0. Parametrize the tractrix using the angle 𝑡 in Fig. 2.7 as a parameter (note that
𝑡 ∈ ] 𝜋2 , 𝜋[). [The surface of revolution obtained from the tractrix around the 𝑧-axis is
the pseudosphere].

2.2 Change of Parameters, Level Surfaces

In this section we gather a number of technical results, such as the change of

parameters on a surface, level surfaces, and the fact that any surface is locally the
graph of a function. We make systematic use of the inverse function theorem, and the
proofs are largely routine — so it seems more instructive (and less monotonous) if,
instead of reading all these proofs, the reader tries to reconstruct some of them by
herself.
In our further study we will make use of local coordinates to express certain
concepts, and our first caution is that such concepts should not depend on the
̃ Ψ)
coordinate system used, but only on the surface. Assume then that (𝑈, Φ) and (𝑈,
are two parameterizations of the surface 𝑆, and that the open set 𝑊 = Φ(𝑈) ∩ Ψ(𝑈) ̃
is non-empty. Under these assumptions (see Fig. 2.8) we have the following result.

p´

Figure 2.7
38 2 Regular Surfaces

Proposition 2.2.1 The coordinate change Φ−1 ○ Ψ∶ Φ−1 (𝑊) → Ψ−1 (𝑊) is a diffeo-
morphism.
Proof It suffices to show that Φ−1 ○Ψ is differentiable, because the same argument
proves the differentiability of its inverse Ψ−1 ○Φ. We write

Φ(𝑢, 𝑣) = (𝑥(𝑢, 𝑣), 𝑦(𝑢, 𝑣), 𝑧(𝑢, 𝑣)),

𝑢, ̃
Ψ(̃ 𝑢, ̃
𝑣 ) = (𝑥(̃ 𝑢, ̃
𝑣 ), 𝑦(̃ 𝑢, ̃
𝑣 ), 𝑧(̃ 𝑣 ));

and, given (̃ 𝑣 0 ) ∈ Ψ−1 (𝑊), let us show that Φ−1 ○ Ψ is differentiable at (̃

𝑢0 , ̃ 𝑢0 , ̃
𝑣 0 ).
Putting (𝑢 0 , 𝑣 0 ) = Φ−1 ○ Ψ(̃ 𝑢0 , ̃
𝑣 0 ), some 2 × 2 submatrix of 𝐽Φ(𝑢 0 , 𝑣 0 ) has
nonzero determinant, and we assume that it is the one formed by the first two rows
𝜕(𝑥, 𝑦)
(whose determinant is usually denoted by ). By the inverse mapping theorem,
𝜕(𝑢, 𝑣)
−1
there exists some open neighborhood 𝐷 ⊆ Φ (𝑊) of (𝑢 0 , 𝑣 0 ) such that the restriction
of 𝑓 (𝑢, 𝑣) = (𝑥(𝑢, 𝑣), 𝑦(𝑢, 𝑣)) to 𝐷 is a diffeomorphism onto the image. It follows
that Φ−1 ○ Ψ∣Ψ−1 ○Φ(𝐷) is differentiable because it is a composition of differentiable
mappings:

𝑓 −1
𝑢, ̃
(̃ 𝑣 ) ↦ Ψ(̃ 𝑣 ) = (𝑥, 𝑦, 𝑧) ↦ (𝑥, 𝑦) z→ (𝑢, 𝑣) = Φ−1 ○ Ψ(̃
𝑢, ̃ 𝑢, ̃
𝑣 ).◻

F
Y

Y -1(W)

F -1(W)
U ~
F -1 Y U

Figure 2.8

Let us point out that the foregoing proof establishes a more general fact than 2.2.1:
if 𝛼 is a differentiable function defined on an open subset of R 𝑛 (a curve, for example)
2.2 Change of Parameters, Level Surfaces 39

whose image is contained in a surface 𝑆, then, for any parameterization Φ of 𝑆,

Φ−1 ○ 𝛼 is differentiable at all points where it is defined.
In the previous section (example D) we noted that the graph of any differentiable
function is a surface; the next proposition says that locally this example is as general
as possible.

Proposition 2.2.2 Any point 𝑝 of a regular surface 𝑆 has in 𝑆 a neighborhood 𝑊 of

one of the following three forms:

𝑊 = {(𝑥, 𝑦, ℎ(𝑥, 𝑦))∶ (𝑥, 𝑦) ∈ 𝑅},

𝑊 = {(𝑥, ℎ(𝑥, 𝑧), 𝑧)∶ (𝑥, 𝑧) ∈ 𝑅},
𝑊 = {(ℎ(𝑦, 𝑧), 𝑦, 𝑧)∶ (𝑦, 𝑧) ∈ 𝑅}
— with, in all three cases, 𝑅 is an open subset of R 2 and ℎ is a differentiable function.

Proof Let (𝑈, Φ) be a parameterization in the neighborhood of 𝑝. One of the three

determinants
𝜕(𝑥, 𝑦) 𝜕(𝑥, 𝑧) 𝜕(𝑦, 𝑧)
, and ,
𝜕(𝑢, 𝑣) 𝜕(𝑢, 𝑣) 𝜕(𝑢, 𝑣)
−1
say the first one, is nonzero when computed at Φ (𝑝). The inverse mapping theorem
then guarantees that there exists an open neighborhood 𝐷 ⊆ 𝑈 of Φ−1 (𝑝) restricted
to which 𝑓 (𝑢, 𝑣) = (𝑥(𝑢, 𝑣), 𝑦(𝑢, 𝑣)) is a diffeomorphism onto the image. Now 𝑊 =
Φ(𝐷) is the sought neighborhood, since 𝑅 = 𝑓 (𝐷) is an open subset of R 2 , ℎ(𝑥, 𝑦) =
𝑧 ○ 𝑓 −1 (𝑥, 𝑦) is differentiable, and 𝑊 = Φ ○ 𝑓 −1 (𝑅) = {(𝑥, 𝑦, ℎ(𝑥, 𝑦))∶ (𝑥, 𝑦) ∈ 𝑅}. ◻

Example 2.2.3 Proposition 2.2.2 gives us a criterion to show that certain sets are not
surfaces, which we illustrate with the cone
√
C = {(𝑥, 𝑦, 𝑧) ∈ R 3 ∶ 𝑧 = 𝑥 2 + 𝑦 2 }.
If C is a surface, there exists an open subset 𝑉 of R 3 containing the point (0, 0, 0)
such that 𝑉 ∩ 𝑆 is the graph of a differentiable function ℎ. But ℎ can only be a
function of (𝑥, 𝑦), because none of the projections of C√on the other coordinate planes
contains a neighborhood of (0, 0). Thus ℎ(𝑥, 𝑦) = 𝑥 2 + 𝑦 2 and this function is
not differentiable at (0, 0). Therefore, C is not a surface. [But the origin is the only
problematic point: C/{(0, 0, 0)} is a surface]. ◻

A ready way to define a surface is by an equation of the form 𝑓 (𝑥, 𝑦, 𝑧) = 𝑎, where

𝑓 ∶ 𝑉 ⊆ R 3 → R is a differentiable function. Not always, of course, does such an
equation define a surface: we have to impose on 𝑓 a certain degree of non-degeneracy,
which we are going to describe.
A point 𝑝 ∈ 𝑉 is called regular (for the function 𝑓 ) if the gradient vector

𝜕𝑓 𝜕𝑓 𝜕𝑓
∇𝑓 = ( , , ),
𝜕𝑥 𝜕𝑦 𝜕𝑧

computed at 𝑝, is a nonzero vector; and 𝑎 ∈ R is a regular value of 𝑓 if 𝑓 −1 ({𝑎}) is

non-empty and contains only regular points.
40 2 Regular Surfaces

𝜕𝑓
The condition (𝑝) ≠ 0 guarantees that in a neighborhood 𝑈 of 𝑝 ∈ 𝑓 −1 ({𝑎}),
𝜕𝑧
𝑓 is strictly monotone along vertical segments, and therefore each of these segments
intersects 𝑓 −1 ({𝑎}) at most one point. In fact, the proof of the next proposition
consists essentially in showing that under these circumstances 𝑓 −1 ({𝑎}) ∩ 𝑈 is the
graph of a differentiable function of (𝑥, 𝑦).

Proposition 2.2.4 If 𝑎 is a regular value of 𝑓 ∶ 𝑉 → R then 𝑓 −1 ({𝑎}) is a regular

surface.

𝜕𝑓
Proof Assuming that (𝑝) ≠ 0, we define
𝜕𝑧
𝐹(𝑥, 𝑦, 𝑧) = (𝑥, 𝑦, 𝑓 (𝑥, 𝑦, 𝑧)).

𝜕𝑓
Since det 𝐽𝐹(𝑝) = (𝑝) ≠ 0, the function 𝐹 is invertible in a neighborhood of
𝜕𝑧
𝑝: so there exist open subsets 𝑈, 𝑊 ⊆ R 3 such that 𝑝 ∈ 𝑈 ⊆ 𝑉, and 𝐹 sends 𝑈
diffeomorphically onto 𝑊. Let us now note that the inverse 𝐺∶ 𝑊 → 𝑈 of 𝐹∣𝑈 has
the form 𝐺(𝑥, 𝑦, 𝑧) = (𝑥, 𝑦, 𝑔(𝑥, 𝑦, 𝑧)) and that, for (𝑥, 𝑦, 𝑧) ∈ 𝑈, all the following
equalities are equivalent:

𝑓 (𝑥, 𝑦, 𝑧) = 𝑎,
𝐹(𝑥, 𝑦, 𝑧) = (𝑥, 𝑦, 𝑎),
(𝑥, 𝑦, 𝑧) = 𝐺(𝑥, 𝑦, 𝑎),
𝑧 = 𝑔(𝑥, 𝑦, 𝑎).

The equivalence between the first and last subset of these equalities shows that
𝑈 ∩ 𝑓 −1 ({𝑎}) is the graph of the differentiable function ℎ(𝑥, 𝑦) = 𝑔(𝑥, 𝑦, 𝑎), whose
domain is the open subset 𝑅 = {(𝑥, 𝑦) ∈ R 2 ∶ (𝑥, 𝑦, 𝑎) ∈ 𝑊} — and this concludes the
proof that 𝑓 −1 ({𝑎}) is a surface. ◻

The sets 𝑓 −1 ({𝑎}) are the level sets of 𝑓 and, when 𝑎 is a regular value, they are
also called level surfaces. Taking for example

𝑥 2 𝑦2 𝑧2
𝑓 (𝑥, 𝑦, 𝑧) = + + ,
𝑎2 𝑏2 𝑐2
we see that the ellipsoid
𝑥 2 𝑦2 𝑧2
+ + =1
𝑎2 𝑏2 𝑐2
is a regular surface, since ∇ 𝑓 (𝑥, 𝑦, 𝑧) is nonzero for all (𝑥, 𝑦, 𝑧) ≠ (0, 0, 0). In general,
any non-degenerate quadric in R 3 is a regular surface, since, for an appropriate
orthonormal basis, it has equation 𝑥 + 𝜀2 𝑦 2 + 𝜀3 𝑧2 = 0 if it is a paraboloid, or
𝜀1 𝑥 2 + 𝜀2 𝑦 2 + 𝜀3 𝑧 2 = 1, with (𝜀1 , 𝜀 2 , 𝜀 3 ) ≠ (0, 0, 0), if it is a hyperboloid or an
ellipsoid.
2.2 Change of Parameters, Level Surfaces 41

An important caveat is that the condition that a be a regular value is by no means

necessary for 𝑓 −1 ({𝑎}) to be a surface. A simple example is given by 𝑓 (𝑥, 𝑦, 𝑧) = 𝑥 2 :
the set 𝑓 −1 ({0}) is a surface, even though it consists only of singular points of 𝑓 .

Note 2.2.5 A level surface is not necessarily connected, as shown by the two-leaf
hyperboloid defined by the equation 𝑧 2 − 𝑥 2 − 𝑦 2 = 1. For the benefit of the reader
unfamiliar with the concept, we give here the definition of connectedness and a brief
discussion: a set 𝐴 ⊆ R 𝑛 is called connected if it cannot be split, i.e., if there are no
disjoint open subsets 𝑉 and 𝑊 of R 𝑛 such that 𝐴 ∩ 𝑉 ≠ ∅ ≠ 𝐴 ∩ 𝑊 and 𝐴 ⊆ 𝑉 ∪ 𝑊.
The hyperboloid above, for example, is not connected because it admits the splitting
𝑉 = {(𝑥, 𝑦, 𝑧) ∈ R 3 ∶ 𝑧 > 0} and 𝑊 = {(𝑥, 𝑦, 𝑧) ∈ R 3 ∶ 𝑧 < 0}. The connected subsets of
R are the intervals; the balls in R 𝑛 (disks in R 2 ) are connected. Connectedness is a
topological property, in the sense that the image of a connected set under a continuous
function is still connected.
To put it suggestively, a surface is connected when it is made up of a single
chunk. A useful connectivity criterion for surfaces (and for open subsets of R 𝑛 )
is the following: 𝑆 is connected if and only if, for every 𝑝 and 𝑞 on 𝑆, there exists
a piecewise differentiable curve 𝛼∶ [𝑎, 𝑏] → 𝑆 such that 𝛼(𝑎) = 𝑝 and 𝛼(𝑏) = 𝑞.
[Proof: if 𝑉 and 𝑊 split 𝑆, then there is no curve in 𝑆 that joins 𝑝 ∈ 𝑆 ∩𝑉 to 𝑞 ∈ 𝑆 ∩𝑊,
because the trace of a curve, being a continuous image of an interval, is connected.
On the other hand, if 𝑆 is connected and 𝑝 ∈ 𝑆, consider the set 𝑅 = {𝑞 ∈ 𝑆: there
exists a curve in 𝑆 from 𝑝 to 𝑞}. Given 𝑞 ∈ 𝑆, let (𝐷, Φ) be a parameterization in the
neighborhood of 𝑞, where 𝐷 ⊆ R 2 is an open disk. Any 𝑟 ∈ Φ(𝐷) can be joined to 𝑞
by a curve in 𝑆: the image under Φ of the line segment [Φ−1 (𝑞), Φ−1 (𝑟)]. Thus, if
𝑞 ∈ 𝑅 (resp. 𝑞 ∈ 𝑆/𝑅) then Φ(𝐷) ⊆ 𝑅 (resp. Φ(𝐷) ⊆ 𝑆/𝑅). Therefore 𝑅 and 𝑆/𝑅 are
open subsets of 𝑆 and, since 𝑆 is connected, one of them, necessarily 𝑆/𝑅, is empty.
Therefore 𝑆 = 𝑅, which proves what we wanted. ] ◻

Now that we have a method for establishing that a set is a surface without using
any parameterization, our next proposition states that if 𝑆 is a surface, anything that
appears to be a parameterization of 𝑆 is indeed so.

Proposition 2.2.6 Let 𝑆 be a surface, 𝑈 be an open subset of R 2 , and Φ∶ 𝑈 → 𝑆 be a

differentiable mapping. If Φ is injective and 𝐽Φ(𝑢, 𝑣) has rank two for all (𝑢, 𝑣) on
𝑈, then Φ is a parameterization of 𝑆.

Proof One just has to check the continuity of the inverse Φ−1 ∶ Φ(𝑈) → 𝑈. Given
(𝑢 0 , 𝑣 0 ) ∈ 𝑈, the point Φ(𝑢 0 , 𝑣 0 ) has, by 2.2.2, an open neighborhood 𝑉 in R 3 such
that 𝑉 ∩ 𝑆 is the graph of a function that we assume to depend on (𝑥, 𝑦). Thus,
𝑉 ∩ 𝑆 = {(𝑥, 𝑦, ℎ(𝑥, 𝑦))∶ (𝑥, 𝑦) ∈ 𝑅}, where 𝑅 is an open subset of R 2 ; and, taking an
open disk 𝐷 ⊆ 𝑈 centered at (𝑢 0 , 𝑣 0 ) and such that Φ(𝐷) ⊆ 𝑉, the restriction of Φ to
𝐷 can be written as Φ(𝑢, 𝑣) = (𝑥(𝑢, 𝑣), 𝑦(𝑢, 𝑣), ℎ(𝑥(𝑢, 𝑣), 𝑦(𝑢, 𝑣))). We then have
42 2 Regular Surfaces

𝜕𝑥 𝜕ℎ 𝜕𝑦 𝜕ℎ
Φ𝑢 = (1, 0, )+ (0, 1, ),
𝜕𝑢 𝜕𝑥 𝜕𝑢 𝜕𝑦
𝜕𝑥 𝜕ℎ 𝜕𝑦 𝜕ℎ
Φ𝑣 = (1, 0, )+ (0, 1, ),
𝜕𝑣 𝜕𝑥 𝜕𝑣 𝜕𝑦
𝜕𝑥 𝜕𝑦 𝜕𝑥 𝜕𝑦 𝜕ℎ 𝜕ℎ
Φ𝑢 × Φ𝑣 = { − } (1, 0, ) × (0, 1, )
𝜕𝑢 𝜕𝑣 𝜕𝑣 𝜕𝑢 𝜕𝑥 𝜕𝑦
𝜕(𝑥, 𝑦) 𝜕ℎ 𝜕ℎ
= (− ,− , 1)
𝜕(𝑢, 𝑣) 𝜕𝑥 𝜕𝑦

𝜕(𝑥, 𝑦)
– and from this, since Φ𝑢 ×Φ𝑣 is nonzero, it follows that ≠ 0. We can therefore
𝜕(𝑢, 𝑣)
assume, shrinking 𝐷 if necessary, that 𝜋 ○ Φ∣𝐷 [where 𝜋∶ R 3 → R 2 is the projection
on the first two coordinates] is a diffeomorphism onto its image, which is then an open
subset 𝐸 of R 2 . Thus, Φ(𝐷) = {(𝑥, 𝑦, ℎ(𝑥, 𝑦))∶ (𝑥, 𝑦) ∈ 𝐸} is an open neighborhood
of Φ(𝑢 0 , 𝑣 0 ) in 𝑆, and the restriction Φ−1 ∣Φ(𝐷) is continuous, because it is given by
−1
the composite (𝜋 ○ Φ∣𝐷 ) ○ 𝜋 of continuous functions. Thus Φ−1 is continuous on
Φ(𝑢 0 , 𝑣 0 ). ◻

Exercises
33. Show that any surface is locally a level surface. Given 𝑝 ∈ 𝑆, there exist an
open neighborhood 𝑉 of 𝑝 in R 3 and a differentiable function 𝑓 ∶ 𝑉 → R such that
𝑆 ∩ 𝑉 = 𝑓 −1 ({0}) and 0 is a regular value of 𝑓 .
34. Show that if two surfaces 𝑆1 and 𝑆2 intersect transversely at 𝑝 then there exists
an open neighborhood 𝑉 of 𝑝 (in R 3 ) such that 𝑆1 ∩ 𝑆2 ∩ 𝑉 is the trace of a regular
curve. (We say that 𝑆1 and 𝑆2 intersect transversely at 𝑝 if 𝑇 𝑝 𝑆1 ≠ 𝑇 𝑝 𝑆2 .)

2.3 Differentiable Functions on Surfaces, Tangent Space

The results of the previous section prepared the setting to do Differential Calculus on
surfaces; and we can now, in this section, explain what is a differentiable function in
such a context. The derivatives of such functions are defined not on the surface but
on its tangent spaces, a concept that we also introduce here.
Let 𝑆1 and 𝑆2 be two surfaces. A mapping 𝑓 ∶ 𝑆1 → 𝑆2 is called differentiable if its
expression in local coordinates is differentiable: more precisely, if there exist, for each
𝑝 ∈ 𝑆1 , parameterizations (𝑈, Φ) of 𝑆1 and (𝑉, Ψ) of 𝑆2 in the neighborhoods of 𝑝
and 𝑓 (𝑝), respectively, such that Ψ−1 ○ 𝑓 ○ Φ is differentiable. Similarly, a function
𝑓 ∶ 𝑆1 → R is called differentiable if every point of 𝑆1 has a parametrized neighborhood
(𝑈, Φ) such that 𝑓 ○ Φ is differentiable. A diffeomorphism is a differentiable bijection
𝑓 ∶ 𝑆1 → 𝑆2 whose inverse is also differentiable.
2.3 Differentiable Functions on Surfaces, Tangent Space 43

Observations and Examples 2.3.1 A. Proposition 2.2.1 guarantees that if 𝑓 ∶ 𝑆1 → 𝑆2

is differentiable, then, for all parameterizations Φ and Ψ, the mapping Ψ−1 ○ 𝑓 ○ Φ
is differentiable. This means that our definition does not depend on any choice of
parameterization.
B. If 𝑓 ∶ 𝑆1 → 𝑆2 is differentiable and (𝑈, Φ) is a parameterization of 𝑆1 then
𝑓 ○Φ∶ 𝑈 → R 3 is differentiable, because locally we can write 𝑓 ○Φ = Ψ○(Ψ−1 ○ 𝑓 ○Φ),
using appropriate local coordinates (𝑉, Ψ) in 𝑆2 . But the converse is also true: if
𝑓 ○ Φ is differentiable for any parameterization (𝑈, Φ) of 𝑆1 then 𝑓 ∶ 𝑆1 → 𝑆2 is
differentiable. This is a consequence of the observation we make following the proof
of 2.2.1.
C. Let 𝑓 ∶ R 3 → R 3 be a differentiable mapping such that 𝑓 (𝑆1 ) ⊆ 𝑆2 . Then
𝑓 ∣𝑆1 ∶ 𝑆1 → 𝑆2 is differentiable, and an analogous observation can be made for
functions 𝑔∶ R 3 → R . As examples, we have 𝑔1 (𝑝) = ⟨𝑝, v⟩, where v is a unit vector
of R 3 [𝑔1 measures the “oriented height”, in the direction of v, of 𝑝 relative to the
origin (0, 0, 0)], and 𝑔2 (𝑝) = ∣𝑝 − 𝑝 0 ∣2 , which measures the square of the distance
from 𝑝 to a fixed point 𝑝 0 . Both these functions, restricted to any surface 𝑆, are
differentiable.
D. Let 𝑆 be a surface of revolution around an axis 𝑟 and let 𝑅 𝜃 be the rotation of
angle 𝜃 around 𝑟. The mapping 𝑅 𝜃 ∣𝑆 ∶ 𝑆 → 𝑆 is a diffeomorphism: its inverse is 𝑅−𝜃 ∣𝑆 .
For a more interesting example, consider the torus of revolution T 2 parametrized
by Φ(𝑢, 𝑣) = ((2 + cos 𝑣) cos 𝑢, (2 + cos 𝑣) sin 𝑢, sin 𝑣). The restriction of Φ to any
square of the form ]𝑎 − 𝜋, 𝑎 + 𝜋[×]𝑏 − 𝜋, 𝑏 + 𝜋[ is injective, and is therefore a
parameterization of T 2 . Fixing (𝑢 0 , 𝑣 0 ) ∈ R 2 , we define a mapping 𝑓 ∶ T 2 → T 2 by
the condition:
● if 𝑝 = Φ(𝑢, 𝑣) then 𝑓 (𝑝) = Φ(𝑢 + 𝑢 0 , 𝑣 + 𝑣 0 ).
Let us show that 𝑓 is differentiable: in fact, given 𝑞 ∈ T, 𝑞 = Φ(𝑢 1 , 𝑣 1 ), the mappings
̃ = Φ∣
Φ ]𝑢 1 − 𝜋,𝑢1 + 𝜋[×]𝑣1 − 𝜋,𝑣1 + 𝜋[

̃ = Φ∣
Ψ ]𝑢 1 +𝑢0 − 𝜋,𝑢1 +𝑢0 + 𝜋[×]𝑣1 +𝑣0 − 𝜋,𝑣1 +𝑣0 + 𝜋[

are parameterizations of T 2 in the neighborhoods of 𝑞 and 𝑓 (𝑞), respectively; and

̃ −1 ○ 𝑓 ○ Φ(𝑢,
Ψ ̃ 𝑣) = (𝑢 + 𝑢 0 , 𝑣 + 𝑣 0 ) is obviously differentiable, which proves that 𝑓
is differentiable in a neighborhood of 𝑞. The same argument proves that the inverse
is differentiable, and therefore 𝑓 is a diffeomorphism. It follows in particular that,
given any two points 𝑝, 𝑞 ∈ T 2 , there exists some diffeomorphism 𝑓 ∶ T 2 → T 2 such
that 𝑓 (𝑝) = 𝑞. (See also Exercise 40 in this section.) ◻

We now deal with the tangent space to a surface 𝑆 at a point 𝑝, which we denote
by 𝑇 𝑝 𝑆. We define 𝑇 𝑝 𝑆 to be the set of velocity vectors, at the point 𝑝, of the curves
whose graph is in 𝑆:

𝑇 𝑝 𝑆 = {𝛼′ (0) ∣ 𝛼∶ ]−𝜀, 𝜀[ → 𝑆 is 𝐶 ∞ and 𝛼(0) = 𝑝}.

44 2 Regular Surfaces

Our next proposition shows that 𝑇 𝑝 𝑆 is a vector subspace of R 3 of dimension two,

which justifies calling 𝑝 + 𝑇 𝑝 𝑆 the tangent plane to 𝑆 at 𝑝.

Proposition 2.3.2 If (𝑈, Φ) is a parameterization of 𝑆 in the neighborhood of 𝑝 then

𝑇 𝑝 𝑆 = 𝐷ΦΦ−1 ( 𝑝) (R 2 ).

Proof Let us prove the inclusion 𝑇 𝑝 𝑆 ⊆ 𝐷ΦΦ−1 ( 𝑝) (R 2 ). Given 𝛼′ (0) ∈ 𝑇 𝑝 𝑆, we

can assume that 𝛼( ] −𝜀, 𝜀[ ) ⊆ Φ(𝑈). Then the curve 𝛽 = Φ−1 ○ 𝛼∶ ] −𝜀, 𝜀[ →
𝑈 is differentiable; writing 𝛼 = Φ ○ 𝛽, we obtain, by the chain rule, 𝛼′ (0) =
𝐷ΦΦ−1 ( 𝑝) (𝛽′ (0)).
Let us treat the opposite inclusion. Given a vector w ∈ R 2 , let us take 𝜀 > 0 so that
the line segment 𝛽(𝑡) = Φ−1 (𝑝) + 𝑡w, ∣𝑡∣ < 𝜀, is contained in 𝑈. Putting 𝛼 = Φ ○ 𝛽,
we have 𝐷ΦΦ−1 ( 𝑝) (w) = 𝐷ΦΦ−1 ( 𝑝) (𝛽′ (0)) = 𝛼′ (0) ∈ 𝑇 𝑝 𝑆. ◻

In practice, what we did was to write the curve 𝛼 in local coordinates: if 𝛼(𝑡) =
Φ(𝑢(𝑡), 𝑣(𝑡)) then 𝛼 is differentiable if and only if so are both functions 𝑢(𝑡) and
𝑣(𝑡); and the chain rule provides the equality 𝛼′ (𝑡) = 𝑢 ′ (𝑡)Φ𝑢 + 𝑣 ′ (𝑡)Φ𝑣 , which
shows that at each point of Φ(𝑈) the tangent space is generated by the vectors Φ𝑢
and Φ𝑣 .

Example 2.3.3 The tangent space to the level surface 𝑆 𝑎 = 𝑓 −1 ({𝑎}) at point 𝑝 is the
orthogonal complement of the line generated by ∇ 𝑓 (𝑝). In fact, if 𝛼∶ ]−𝜀, 𝜀[ → 𝑆 𝑎
satisfies 𝛼(0) = 𝑝 then 𝑓 ○ 𝛼(𝑡) = 𝑎 for all 𝑡 ∈ ]−𝜀, 𝜀[, so that ⟨∇ 𝑓 (𝑝), 𝛼′ (0)⟩ =
( 𝑓 ○ 𝛼)′ (0) = 0 — which shows that ∇ 𝑓 (𝑝) is orthogonal to 𝑇 𝑝 𝑆 𝑎 . ◻

Let 𝑓 ∶ 𝑆1 → 𝑆2 be a differentiable mapping at 𝑝 ∈ 𝑆1 . The derivative of 𝑓 at

𝑝 is the mapping 𝐷 𝑓 𝑝 ∶ 𝑇 𝑝 𝑆1 → 𝑇 𝑓 ( 𝑝) 𝑆2 defined as follows: if 𝛼′ (0) ∈ 𝑇 𝑝 𝑆1 then
𝐷 𝑓 𝑝 (𝛼′ (0)) = ( 𝑓 ○ 𝛼)′ (0). That is, 𝐷 𝑓 𝑝 sends the velocity vector at 𝑝 of a given
curve to the velocity vector at 𝑓 (𝑝) of the transform of that curve by 𝑓 . Of course,
the same vector represents the velocity vector at 𝑝 of many different curves, but we
will see below that 𝐷 𝑓 𝑝 is well-defined. Let us take local coordinates Φ(𝑢, 𝑣) and
Ψ(̃𝑢, ̃𝑣 ) at 𝑝 and 𝑓 (𝑝), and let us put 𝑓̃ = Ψ−1 ○ 𝑓 ○ Φ: with this notation we have
the following result.

Proposition 2.3.4 𝐷 𝑓 𝑝 ∶ 𝑇 𝑝 𝑆1 → 𝑇 𝑓 ( 𝑝) 𝑆2 is a linear mapping whose matrix with

respect to the bases (Φ𝑢 , Φ𝑣 ) of 𝑇 𝑝 𝑆1 and (Ψ𝑢̃ , Ψ̃𝑣 ) of 𝑇 𝑓 ( 𝑝) 𝑆2 is the Jacobian of 𝑓̃
on Φ−1 (𝑝).

Proof Writing 𝛼(𝑡) = Φ(𝑢(𝑡), 𝑣(𝑡)), the curve 𝛽 = 𝑓 ○ 𝛼 is given by 𝛽(𝑡) =

𝑢 (𝑡), ̃
Ψ(̃ 𝑣 (𝑡)), where (̃ 𝑣 (𝑡)) = 𝑓̃(𝑢(𝑡), 𝑣(𝑡)). Differentiating the last equality,
𝑢 (𝑡), ̃
we obtain
𝑢̃′ (0) 𝑢 ′ (0) 𝑢 ′ (0)
( ′ ) = 𝐽 𝑓̃(𝑢(0),𝑣(0)) ( ′ ) = 𝐽 𝑓̃Φ−1 ( 𝑝) ( ′ ) (*)
̃𝑣 (0) 𝑣 (0) 𝑣 (0)
The equality 𝐷 𝑓 𝑝 (𝛼′ (0)) = ( 𝑓 ○ 𝛼)′ (0), which defines 𝐷 𝑓 𝑝 , can be rewritten as

𝐷 𝑓 𝑝 (𝑢 ′ (0)Φ𝑢 + 𝑣 ′ (0)Φ𝑣 ) = 𝑢̃′ (0)Ψ𝑢̃ + ̃

𝑣 ′ (0)Ψ̃𝑣 . (**)
2.3 Differentiable Functions on Surfaces, Tangent Space 45

From (*) and (**) it follows that 𝐷 𝑓 𝑝 (𝛼′ (0)) is well-defined, not depending on 𝛼
but only on 𝛼′ (0); and that furthermore 𝐷 𝑓 𝑝 is linear and has matrix 𝐽 𝑓̃Φ−1 ( 𝑝) with
respect to the given bases. ◻

We end this section with two results which are the transpositions of the inverse
mapping theorem and the chain rule to the context of surfaces. The proof of the first
one is left as an (easy) exercise.

Proposition 2.3.5 Let 𝑓 ∶ 𝑆1 → 𝑆2 be a differentiable mapping and 𝑝 ∈ 𝑆1 such

that 𝐷 𝑓 𝑝 ∶ 𝑇 𝑝 𝑆1 → 𝑇 𝑓 ( 𝑝) 𝑆2 is a linear isomorphism. Then there exists an open
neighborhood 𝑈 of 𝑝 in 𝑆1 and an open subset 𝑉 of 𝑆2 such that 𝑓 ∣𝑈 ∶ 𝑈 → 𝑉 is a
diffeomorphism.

Proposition 2.3.6 If 𝑓 ∶ 𝑆1 → 𝑆2 and 𝑔∶ 𝑆2 → 𝑆3 are differentiable then 𝑔 ○ 𝑓 is

differentiable and, for all 𝑝 ∈ 𝑆1 , we have 𝐷(𝑔 ○ 𝑓 ) 𝑝 = 𝐷𝑔 𝑓 ( 𝑝) ○ 𝐷 𝑓 𝑝 .

Proof The verification that 𝑔 ○ 𝑓 is differentiable is left to the reader. As for the
second statement, let us take u ∈ 𝑇 𝑝 𝑆1 and a curve 𝛼 such that 𝛼′ (0) = u, and let
us put 𝛽 = 𝑓 ○ 𝛼, 𝛾 = 𝑔 ○ 𝑓 ○ 𝛼, v = 𝛽′ (0), w = 𝛾 ′ (0). We then have 𝐷 𝑓 𝑝 (u) = v
(because 𝛽 = 𝑓 ○ 𝛼), 𝐷𝑔 𝑓 ( 𝑝) (v) = w (because 𝛾 = 𝑔 ○ 𝛽), 𝐷(𝑔 ○ 𝑓 ) 𝑝 (u) = w (because
𝛾 = (𝑔 ○ 𝑓 ) ○ 𝛼), and therefore 𝐷(𝑔 ○ 𝑓 ) 𝑝 (u) = (𝐷𝑔 𝑓 ( 𝑝) ○ 𝐷 𝑓 𝑝 )(u). ◻

Exercises
35. Consider the function 𝑓 ∶ R 3 → R given by 𝑓 (𝑥, 𝑦, 𝑧) = 2𝑥 2 − 𝑦 2 − 𝑧2 . Determine
the equations of the planes which √are tangent to the surface 𝑓 −1 ({1}) and parallel to
the plane given by the equation 2 2𝑥 + 𝑦 + 𝑧 = 0.
36. (a) Show that the paraboloid 𝑧 = 𝑥 2 + 𝑦 2 is diffeomorphic to the plane.
(b) Show that the sphere S2 and the ellipsoid

𝑥 2 𝑦2 𝑧2
+ + =1
𝑎2 𝑏2 𝑐2
are diffeomorphic.
37. Let 𝑉 be a neighborhood of the origin in R 2 and let Φ∶ 𝑉 → R 3 given by
Φ(𝑢, 𝑣) = 𝑓 (𝑢) + 𝑔(𝑣) be a parameterization of a regular surface 𝑆. Show that the
tangent planes to 𝑆 along the curve Φ(𝑢, 0) are all parallel to the same line.
38. A differentiable mapping 𝑓 ∶ 𝑆1 → 𝑆2 is called a local diffeomorphism if each point
𝑝 ∈ 𝑆1 has a neighborhood 𝑊 in 𝑆1 such that 𝑓 ∣𝑊 ∶ 𝑊 → 𝑓 (𝑊) is a diffeomorphism.
Show that if 𝑓 is a local diffeomorphism then 𝐷 𝑓 𝑝 is a linear isomorphism for all
𝑝 ∈ 𝑆1 .
39. Show that if all normal lines to a connected surface pass through the same point,
then that surface is contained in a sphere.
46 2 Regular Surfaces

40. (a) Given 0 < 𝑟 1 < 𝑟 2 and 𝜑0 , consider a function 𝑔∶ R → R that is 𝐶 ∞ , monotone
and such that 𝑔(𝑥) = 𝜑0 for 𝑥 ≤ 𝑟 1 , and 𝑔(𝑥) = 0 for 𝑥 ≥ 𝑟 2 (see ex. 17). Let
ℎ∶ R 2 → R 2 be the mapping that sends the point with polar coordinates (𝜌, 𝜑) to the
point with coordinates (𝜌, 𝜑 + 𝑔(𝜌)). Show that ℎ is a 𝐶 ∞ diffeomorphism. How
does ℎ behave in {𝑝 ∈ R 2 ∶ ∣𝑝∣ ≤ 𝑟 1 } and {𝑝 ∈ R 2 ∶ ∣𝑝∣ ≥ 𝑟 2 }?
(b) Let (𝑈, Φ) be a parameterization of 𝑆 such that 𝑈 contains the closed disk
with radius 𝑟 2 centered at the origin. Show that Φ ○ ℎ ○ Φ−1 ∶ Φ(𝑈) → Φ(𝑈) extends
to a diffeomorphism of 𝑆.
(c) Show that if 𝑆 is connected then for any two points of 𝑆 there exists a
diffeomorphism of 𝑆 that sends one of these points to the other one.
41. Define explicitly a differentiable mapping T 2 → S2 that is surjective.
42. Let 𝑆 = {(𝑥, 𝑦, 𝑧) ∈ R 3 ∶ 𝑥 ≠ 0, 𝑧 = 𝑥 𝑓 (𝑦/𝑥)}, where 𝑓 ∶ R → R is a 𝐶 ∞ function.
Show that 𝑆 is a regular surface, and that all tangent planes to 𝑆 pass through the
origin.
43. Consider the mapping

𝑎(𝑢𝑣 + 1) 𝑏(𝑢 − 𝑣) 𝑐(𝑢𝑣 − 1)

Φ(𝑢, 𝑣) = ( , , ),
𝑢+𝑣 𝑢+𝑣 𝑢+𝑣

where 𝑎, 𝑏, 𝑐 ≠ 0 and 𝑢 + 𝑣 ≠ 0. Find an equation that implicitly defines the image of

Φ, and conclude that it is a surface. Compute the normal vector and the tangent plane
at each point.

2.4 Orientability

A surface is orientable when it is possible to distinguish its top from the bottom, so
that an observer placed on it can distinguish left from right. This approach works
when the observer is three-dimensional and has an idea of the position of the surface
in space; it is more intricate to explain how two-dimensional beings whose universe
is the surface will know whether it is orientable or not.
Given two linearly independent vectors v and w in R 3 , the trihedron (v, w, 𝑁),
where
1
𝑁= (v × w),
∣v × w∣
forms a positively oriented basis of R 3 , meaning that the matrix whose columns
are (in the same order) these vectors has positive determinant. The unit vector 𝑁 is
orthogonal to the plane Π generated by v and w, introducing an orientation in Π as
follows: a basis (v1 , w1 ) of Π is called positively oriented if the triplet (v1 , w1 , 𝑁) is
a positively oriented basis of R 3 ; in other words, if
1
𝑁= (v1 × w1 ).
∣v1 × w1 ∣
2.4 Orientability 47

We thus recognize that Π has exactly two orientations, one induced by 𝑁 and the
other one by −𝑁.
We say that the surface 𝑆 is orientable if it is possible to choose, for each 𝑝 ∈ 𝑆, an
orientation on 𝑇 𝑝 𝑆 that varies continuously with 𝑝 — more precisely, if there exists a
continuous function 𝑁∶ 𝑆 → S2 such that, for each 𝑝, 𝑁(𝑝) is orthogonal to 𝑇 𝑝 𝑆. We
call such a field of normal vectors 𝑁 an orientation of 𝑆.
For example, level surfaces are orientable, because the vector field
1
𝑁(𝑝) = ∇ 𝑓 (𝑝)
∣∇ 𝑓 (𝑝)∣

is an orientation of 𝑓 −1 ({𝑎}) (see example 2.3.3).

Surfaces that admit a global parameterization Φ(𝑢, 𝑣) are also orientable, because
on them we can define
1
𝑁(𝑝) = Φ𝑢 × Φ𝑣 ∣Φ−1 ( 𝑝) .
∣Φ𝑢 × Φ𝑣 ∣

More generally, each parameterization (𝑈, Φ) of 𝑆 induces, by the preceding formula,

an orientation in the open Φ(𝑈) ⊆ 𝑆; the problem is to “glue” together the various
local orientations to obtain an orientation of the whole surface.
Proposition 2.4.1 Any orientable connected surface has exactly two distinct orienta-
tions.
Proof Given two orientations 𝑁 and 𝑁 ̃ of 𝑆, we have, for each 𝑝 in 𝑆, 𝑁(𝑝) = 𝑁(𝑝)
̃
̃
or 𝑁(𝑝)) = − 𝑁(𝑝), since these two unit vectors are orthogonal to the same plane
𝑇 𝑝 𝑆. Thus, the function 𝜎∶ 𝑆 → R defined by 𝜎(𝑝) = ⟨𝑁(𝑝), 𝑁(𝑝)⟩̃ is continuous
and only takes the values 1 or −1. Since 𝑆 is connected, its image 𝜎(𝑆) ⊆ {−1, 1}
is also connected, and is therefore reduced to only one element. Hence, we have
̃
𝑁(𝑝) = 𝑁(𝑝) ̃
or 𝑁(𝑝) = − 𝑁(𝑝) for all 𝑝 ∈ 𝑆. ◻
Example 2.4.2 The Möbius strip M is the surface obtained by gluing the two ends of
a paper strip so that their opposite vertices coincide. We will now see that this surface
is non-orientable:
To obtain a parameterization of M , we fix a circumference and consider a line
segment that intersects, at its midpoint, orthogonally the circumference. We let
the line segment travel around the whole circumference, letting it rotate around its
midpoint and return to the starting point with reversed endpoints. We put
𝜃 𝜃 𝜃
Φ(𝜃, 𝑡) = ((2 − 𝑡 sin ) cos 𝜃, (2 − 𝑡 sin ) sin 𝜃, 𝑡 cos ) ,
2 2 2
where (𝜃, 𝑡) ∈ R × ]−1, 1[. We note that every restriction of 𝜃 to some interval of
length 2𝜋 yields a different parameterization of M . The curves 𝜃 = 𝑐te represent the
various positions of the generating line segment of M along the circumference 𝑡 = 0.
We also note that Φ(𝜃 + 2𝜋, 𝑡) = Φ(𝜃, −𝑡). To simplify the calculations, we introduce
the moving orthonormal trihedron
48 2 Regular Surfaces

Figure 2.9

given by the three vectors

e1 (𝜃) = (cos 𝜃, sin 𝜃, 0)

e2 (𝜃) = (−sin 𝜃, cos 𝜃, 0)
e3 (𝜃) = (0, 0, 1),

which satisfies the relations e1 ′ (𝜃) = e2 (𝜃), e2 ′ (𝜃) = −e1 (𝜃), e1 (𝜃) × e2 (𝜃) = e3 (𝜃),
e2 (𝜃) × e3 (𝜃) = e1 (𝜃), e3 (𝜃) × e1 (𝜃) = e2 (𝜃). Writing
𝜃 𝜃
Φ(𝜃, 𝑡) = (2 − 𝑡 sin ) e1 (𝜃) + 𝑡 cos e3 (𝜃),
2 2
we easily conclude that
𝜃 𝜃 𝜃 𝑡
Φ 𝜃 × Φ𝑡 = (2 − 𝑡 sin ) [cos e1 (𝜃) + sin e3 (𝜃)] + e2 (𝜃).
2 2 2 2

Let us now assume M orientable with orientation given by 𝑁∶ M → S2 . The

̃= 1
parameterization Φ∣]0,2 𝜋[×]−1,1[ induces the orientation 𝑁 Φ 𝜃 × Φ𝑡 on
∣Φ 𝜃 × Φ𝑡 ∣
𝑊 = M /{(2, 0, 𝑡)∶ 𝑡 ∈ ] − 1, 1[ }; and, by 2.4.1, we have 𝑁 ̃ = 𝑁∣ or 𝑁 ̃ = −𝑁∣ .
𝑊 𝑊
We shall now see that 𝑁 ̃ has no continuous extension to M , which proves the
non-existence of 𝑁.
In fact, if there was such an extension, then there would exist lim 𝑁(𝑝), ̃ but
𝑝→(2,0,0)
2.4 Orientability 49

lim 𝑁 ̃ ○ Φ(𝜃, 0) = lim [cos 𝜃 e1 (𝜃) + sin 𝜃 e2 (𝜃)] = (1, 0, 0) ≠

𝜃→0+ 𝜃→0+ 2 2
lim 𝑁 ̃ ○ Φ(𝜃, 0) = lim [cos e1 (𝜃) + sin 𝜃 e2 (𝜃)] = (−1, 0, 0).
𝜃
𝜃→2 𝜋 − 𝜃→2 𝜋 − 2 2
To sum up, what we have done shows that if a normal vector makes one complete
turn around M , then it returns to the starting position pointing in the opposite direction.
It is thus possible to go “from up to down” by walking on the Möbius strip, which
justifies the statement that it has only one side. ◻
Our definition of orientability has the disadvantage of obscuring that this concept
is invariant under diffeomorphisms. For example, we saw that a Möbius strip,
corresponding to a certain M of R 3 , is non-orientable; but can we from this draw the
same conclusion for all Möbius strips (i.e., for all surfaces that are diffeomorphic to
M )? There is another orientability criterion that allows one to more easily justify the
(affirmative) answer to this question.
An oriented atlas of a surface 𝑆 is a collection (𝑈 𝛼 , Φ 𝛼 ) 𝛼∈I of parameterizations
of 𝑆 such that:
(i) the parameterizations cover 𝑆, i.e., 𝑆 = ⋃ Φ 𝛼 (𝑈 𝛼 );
𝛼∈I
(ii) for all 𝛼, 𝛽 ∈ I, the Jacobian of Φ−1
𝛽 ○ Φ 𝛼 has positive determinant at all points
where it is defined.
Before we move on, consider two parameterizations Φ(𝑢, 𝑣) and Ψ(̃ 𝑢, ̃
𝑣 ) that
intersect on an open 𝑊 of 𝑆. We thus have

𝑢, ̃
Φ(𝑢, 𝑣) = Ψ(̃ 𝑣) (*)

for (𝑢, 𝑣) ∈ Φ−1 (𝑊) and (̃ 𝑣 ) = Ψ−1 ○ Φ(𝑢, 𝑣). By differentiation of (*) we obtain
𝑢, ̃
the two equalities

𝜕̃
𝑢 𝜕̃
𝑣 𝜕̃
𝑢 𝜕̃
𝑣
Φ𝑢 = Ψ𝑢̃ + Ψ̃𝑣 , Φ𝑣 = Ψ𝑢̃ + Ψ̃𝑣 ,
𝜕𝑢 𝜕𝑢 𝜕𝑣 𝜕𝑣
from which

) Ψ𝑢̃ × Ψ̃𝑣 = (det 𝐽(Ψ−1 ○ Φ)(𝑢,𝑣) )Ψ𝑢̃ × Ψ̃𝑣 .

𝜕̃
𝑢 𝜕̃
𝑣 𝜕̃𝑢 𝜕̃
𝑣
Φ𝑢 × Φ 𝑣 = ( −
𝜕𝑢 𝜕𝑣 𝜕𝑣 𝜕𝑢
This latter formula allows us to conclude that the two conditions
1 1
a) Φ𝑢 × Φ𝑣 = Ψ𝑢̃ × Ψ̃𝑣 ,
∣Φ𝑢 × Φ𝑣 ∣ ∣Ψ𝑢̃ × Ψ̃𝑣 ∣ (**)
b) det 𝐽(Ψ−1 ○ Φ)(𝑢,𝑣) > 0,

are equivalent. We can now state the alternative orientability criterion.

Proposition 2.4.3 𝑆 is orientable if and only if it has an orientable atlas.
Proof Suppose that 𝑁 is an orientation of 𝑆. Given 𝑝 𝛼 ∈ 𝑆, let (𝑈, Φ), with 𝑈
connected, be a parameterization in the neighborhood of 𝑝 𝛼 . Then, by 2.4.1, we
50 2 Regular Surfaces

1 1
have Φ𝑢 × Φ𝑣 = 𝑁∣Φ(𝑈) or Φ𝑢 × Φ𝑣 = −𝑁∣Φ(𝑈) . In the first
∣Φ𝑢 × Φ𝑣 ∣ ∣Φ𝑢 × Φ𝑣 ∣
hypothesis, we take (𝑈 𝛼 , Φ 𝛼 ) = (𝑈, Φ); in the second one, we take 𝑈 𝛼 = {(𝑢, 𝑣) ∈
R 2 ∶ (𝑢, −𝑣) ∈ 𝑈} and Φ 𝛼 (𝑢, 𝑣) = Φ(𝑢, −𝑣); in each case Φ 𝛼 induces on Φ 𝛼 (𝑈 𝛼 )
the same orientation as 𝑁. The atlas (𝑈 𝛼 , Φ 𝛼 ) 𝛼∈I obtained in this way covers 𝑆 and,
by the equivalence of conditions a) and b) above, is oriented.
Given now an oriented atlas (𝑈 𝛼 , Φ 𝛼 ) 𝛼∈I , we define an orientation 𝑁∶ 𝑆 → S2 by
requiring that its restriction to each open subset Φ 𝛼 (𝑈 𝛼 ) is the orientation induced
by Φ 𝛼 . By (**), there is no ambiguity in the definition of 𝑁; and, since 𝑁∣Φ (𝑈 ) is
𝛼 𝛼
continuous and (Φ 𝛼 (𝑈 𝛼 )) 𝛼∈I is a covering of 𝑆 by open sets, 𝑁 is continuous. ◻

It follows from this proof that any orientation 𝑁∶ 𝑆 → S2 is a differentiable mapping,

because 𝑁 is expressed as a differentiable function of the local parameters.
Let now 𝑓 ∶ 𝑆1 → 𝑆2 be a diffeomorphism between surfaces. If 𝑆1 is oriented then
𝑓 induces an orientation on 𝑆2 as follows: given an oriented atlas (𝑈 𝛼 , Φ 𝛼 ) 𝛼∈I of 𝑆1
(which is compatible with the orientation of 𝑆1 ), then Ψ 𝛼 = 𝑓 ○ Φ 𝛼 , (𝑈 𝛼 , Ψ 𝛼 ) 𝛼∈I is
an oriented atlas of 𝑆2 — since, by Ψ𝛽−1 ○ Ψ 𝛼 = Φ−1 𝛼 ○ Φ 𝛼 , the coordinate changes
in either atlas are precisely the same. We call the orientation defined by the atlas
(𝑈 𝛼 , Ψ 𝛼 ) 𝛼∈I on 𝑆2 the orientation induced by the diffeomorphism 𝑓 (from the given
orientation of 𝑆1 ). In particular, any two diffeomorphic surfaces are either both
orientable or both non-orientable.
In the case of a diffeomorphism 𝑓 ∶ 𝑆 → 𝑆 of an orientable connected surface onto
itself, we say that 𝑓 preserves orientation if the orientation induced by 𝑓 on 𝑆 from a
given orientation of 𝑆 is equal to the original one; if it is the opposite one, then we
say that 𝑓 reverses orientation.

Proposition 2.4.4 Let 𝑓 ∶ 𝑆 → 𝑆 be a diffeomorphism of a connected orientable

surface. Then:
(i) Whether or not 𝑓 preserves orientation only depends on 𝑓 , not on the orientation
of 𝑆.
(ii) For each oriented atlas A of S, one and only one of the following statements
holds:
det 𝐽(Ψ−1 ○ 𝑓 ○ Φ) > 0 for any (𝑈, Φ), (𝑉, 𝜓) ∈ A;
or
det 𝐽(Ψ−1 ○ 𝑓 ○ Φ) < 0 for any (𝑈, Φ), (𝑣, 𝜓) ∈ A.
In the first case 𝑓 preserves orientation, while in the second case it inverts it.

Proof Let 𝑁 and −𝑁 be the two orientations of 𝑆. We define two oriented atlases A1
and A2 of 𝑆 as follows: A1 (resp. A2 ) includes all parameterizations (𝑈, Φ) of 𝑆 such
that Φ induces in Φ(𝑈) the orientation 𝑁∣Φ(𝑈) (resp. −𝑁∣Φ(𝑈) ). Thus, any oriented
atlas is included either in A1 or in A2 , so that we can assume A = A1 . Furthermore,
det 𝐽(Ψ−1 ○ Φ) < 0 whenever (𝑈, Φ) ∈ A1 and (𝑉, Ψ) ∈ A2 .
The set 𝑓 (A1 ) of the parameterizations (𝑈, 𝑓 ○ Φ) such that (𝑈, Φ) ∈ A1
is an oriented atlas of 𝑆, so that it defines on 𝑆 one of the orientations 𝑁 or
2.5 Areas, Lengths, and Angles: The First Fundamental Form 51

−𝑁. In the first hypothesis, 𝑓 (A1 ) ⊆ A1 and therefore det 𝐽(Ψ−1 ○ 𝑓 ○ Φ) > 0
whenever (𝑈, Φ), (𝑉, Ψ) ∈ A1 ; in the second hypothesis, 𝑓 (A1 ) ⊆ A2 and thence
det 𝐽(Ψ−1 ○ 𝑓 ○ Φ) < 0 for (𝑈, Φ), (𝑉, Ψ) ∈ A1 . This proves (ii).
To prove (i), we take (𝑈, Φ) in A1 and set 𝑉 = {(𝑢, 𝑣) ∈ R 2 ∶ (𝑢, −𝑣) ∈ 𝑈},
Ψ(𝑢, 𝑣) = Φ(𝑢, −𝑣). Then (𝑉, Ψ) belongs to A2 , since the Jacobian of Ψ−1 ○Φ(𝑢, 𝑣) =
(𝑢, −𝑣) is negative; and, for the same reason, the parameterizations (𝑈, 𝑓 ○ Φ) and
(𝑉, 𝑓 ○ Ψ) cannot belong both to A1 or both to A2 . Therefore 𝑓 (A1 ) and 𝑓 (A2 )
define distinct orientations, which proves (i). ◻

Example 2.4.5 Consider the diffeomorphism 𝑓 ∶ S2 → S2 given by 𝑓 (𝑥, 𝑦, 𝑧) =

(−𝑥, 𝑦, 𝑧). The parameterization
√
Φ(𝑢, 𝑣) = (𝑢, 𝑣, 1 − (𝑢 2 + 𝑣 2 ))

belongs to some oriented atlas of S2 , since its domain is connected; and, since the
Jacobian of Φ−1 ○ 𝑓 ○ Φ(𝑢, 𝑣) = (−𝑢, 𝑣) is negative, we conclude that 𝑓 reverses
orientation.

Exercises
44. Find out whether the antipodal mapping ℎ∶ S2 → S2 given by ℎ(𝑥, 𝑦, 𝑧) =
(−𝑥, −𝑦, −𝑧) preserves orientation or not.
45. Consider the Möbius strip M parametrized by Φ(𝜃, 𝑡) = ((2 − 𝑡 sin 2𝜃 ) cos 𝜃, (2 −
𝑡 sin 2𝜃 ) sin 𝜃, 𝑡 cos 2𝜃 ). Show that if the circumference 𝑡 = 0 is removed from M , then
the resulting surface is still connected but is then orientable.
46. Let 𝑓 ∶ 𝑆1 → 𝑆2 be a local diffeomorphism. Check whether the following statements
are true:
(a) if 𝑆2 is orientable then 𝑆1 is orientable;
(b) if 𝑆1 is orientable and 𝑓 is surjective then 𝑆2 is orientable.
47. Let 𝑆 be a connected orientable surface and let 𝑓 ∶ 𝑆 → 𝑆 be a diffeomorphism. Is
it true that 𝑓 ○ 𝑓 preserves orientation?

2.5 Areas, Lengths, and Angles: The First Fundamental Form

Any surface 𝑆 ⊆ R 3 inherits from the ambient space a notion of size that can be
used to measure the area of regions and the length of curves in 𝑆. This metric
structure, which we now introduce, enriches the concept of surface and enables a
finer classification than that by diffeomorphisms.
The first fundamental form of 𝑆 at 𝑝 ∈ 𝑆 is the quadratic form 𝐼 𝑝 ∶ 𝑇 𝑝 𝑆 → R +
defined by 𝐼 𝑝 (v) = ⟨v, v⟩ 𝑝 , where ⟨⋅ , ⋅⟩ 𝑝 is the restriction to 𝑇 𝑝 𝑆 of the usual inner
product on R 3 .
If Φ(𝑢, 𝑣) is a parameterization of 𝑆 and 𝛼(𝑡) = Φ(𝑢(𝑡), 𝑣(𝑡)) is a differentiable
curve, we have
52 2 Regular Surfaces

𝐼 𝛼(𝑡) (𝛼′ (𝑡)) = ⟨𝑢 ′ (𝑡)Φ𝑢 + 𝑣 ′ (𝑡)Φ𝑣 , 𝑢 ′ (𝑡)Φ𝑢 + 𝑣 ′ (𝑡)Φ𝑣 ⟩ 𝛼(𝑡)

= 𝐼 𝛼(𝑡) (Φ𝑢 )𝑢 ′ (𝑡) + 2⟨Φ𝑢 , Φ𝑣 ⟩ 𝛼(𝑡) 𝑢 ′ (𝑡)𝑣 ′ (𝑡) + 𝐼 𝛼(𝑡) (Φ𝑣 )𝑣 ′ (𝑡)
2 2

= 𝐸 𝑢 ′ (𝑡) + 2𝐹 𝑢 ′ (𝑡)𝑣 ′ (𝑡) + 𝐺𝑣 ′ (𝑡) ,

2 2

where 𝐸, 𝐹 and 𝐺 are the so-called coefficients of the first fundamental form
for the parameterization Φ(𝑢, 𝑣), defined by 𝐸(𝑢, 𝑣) = 𝐼Φ(𝑢,𝑣) (Φ𝑢 ), 𝐹(𝑢, 𝑣) =
⟨Φ𝑢 , Φ𝑣 ⟩Φ(𝑢,𝑣) , 𝐺(𝑢, 𝑣) = 𝐼Φ(𝑢,𝑣) (Φ𝑣 ). The above calculations show that the length
of 𝛼(𝑡), 𝑡 ∈ [𝑎, 𝑏], is given by
𝑏√
2 2
𝑙(𝛼) = ∫ 𝐸 𝑢 ′ (𝑡) + 2𝐹 𝑢 ′ (𝑡)𝑣 ′ (𝑡) + 𝐺 𝑣 ′ (𝑡) 𝑑𝑡.
𝑎

Therefore it is possible to compute the length of any curve in 𝑆 knowing only the first
fundamental form (and hence its coefficients in any parameterization) without further
reference to the ambient space.
We point out that the matrix of the quadratic form 𝐿 Φ(𝑢,𝑣) relative to the basis
𝐸 𝐹
(Φ𝑢 , Φ𝑣 ) of 𝑇Φ(𝑢,𝑣) 𝑆 is 𝑀 = [ ]: so if v = 𝑎Φ𝑢 + 𝑏Φ𝑣 and w = 𝑐Φ𝑢 + 𝑑Φ𝑣 , the
𝐹𝐺
𝑐
inner product of v and w is given by the matrix product [𝑎, 𝑏] 𝑀 [ ] = 𝐸 𝑎𝑐 + 𝐹(𝑎𝑑 +
𝑑
𝑏𝑐) + 𝐺 𝑏𝑑.
Examples 2.5.1 If v and w are orthonormal vectors and 𝑝 ∈ R 3 then the parameteri-
zation Ψ(𝑢, 𝑣) = 𝑝 + 𝑢v + 𝑣w of the plane parallel to v and w which passes through
𝑝 has coefficients 𝐸 = 1, 𝐹 = 0, 𝐺 = 1. On the other hand, the coefficients of the
parameterization Φ(𝑢, 𝑣) = ((2 + cos 𝑣) cos 𝑢, (2 + cos 𝑣) sin 𝑢, sin 𝑣) of T 2 are 𝐸 = 1,
2
𝐹 = 0 and 𝐺 = (2 + cos 𝑣) . ◻
The first fundamental form also allows one to compute the angle between two
nonzero vectors v, w ∈ 𝑇 𝑝 𝑆: this (non-oriented) angle is the only 𝜃 ∈ [0, 𝜋] such that
1
cos 𝜃 = √ ⟨v, w⟩ 𝑝 ;
𝐼 𝑝 (v)𝐼 𝑝 (w)
in local coordinates, writing v = 𝑎Φ𝑢 + 𝑏Φ𝑣 and w = 𝑐Φ𝑢 + 𝑑Φ𝑣 , we have

𝐸𝑎𝑐 + 𝐹(𝑎𝑑 + 𝑏𝑐) + 𝐺𝑏𝑑

cos 𝜃 = √ ⋅ (*)
(𝐸𝑎 + 2𝐹𝑎𝑏 + 𝐺𝑏 2 )(𝐸𝑐2 + 2𝐹𝑐𝑑 + 𝐺𝑑 2 )
2

If the surface 𝑆 is oriented and Φ(𝑢, 𝑣) is compatible with the orientation, we can
assign a sign to the angles: the oriented angle ∠(v, w) (from v to w) is the only
𝜃 ∈ ] − 𝜋, 𝜋] such that equality (*) is satisfied and such that it is negative when
𝑎𝑑 − 𝑏𝑐 < 0, non-negative when 𝑎𝑑 − 𝑏𝑐 ≥ 0. [Since the oriented angles are defined
up to integer multiples of 2𝜋, the representatives of the angle ∠(v, w) are thus all
numbers of the form 𝜃 + 2𝑘 𝜋, 𝑘 ∈ Z .]
The angle between two curves 𝛼(𝑡) and 𝛽(𝑠) in 𝑆 at an intersection point
𝛼(𝑡 0 ) = 𝛽(𝑠0 ) is, by definition, the angle between the velocity vectors 𝛼′ (𝑡 0 ) and
2.5 Areas, Lengths, and Angles: The First Fundamental Form 53

𝛽′ (𝑠0 ). For instance, it follows from the formulas deduced above that the
√angle between
the coordinate curves of the parameterization Φ(𝑢, 𝑣) is arccos (𝐹/ 𝐸𝐺) ∈ ]0, 𝜋[.
When 𝐹 = 0 the coordinate curves intersect each other orthogonally; in this case
we say that Φ(𝑢, 𝑣) is an orthogonal parameterization. The above examples 2.5.1
are orthogonal parameterizations; in fact, as we showed in Section 3.3, any surface
admits orthogonal parameterizations.
We finally deal with the measurement of areas. If Δ ⊆ 𝑆 is a region contained in a
single coordinate system (𝑈, Φ), its area is defined by the integral

∬ ∣Φ𝑢 × Φ𝑣 ∣ 𝑑𝑢 𝑑𝑣
Φ−1 (Δ)

— if such an integral exists (and it certainly exists when Δ is open or closed and the
closure of Φ−1 (Δ) is a compact set contained in 𝑈). If Δ is not contained in a single
parametrized neighborhood, we can write it as a disjoint, finite or countable union of
regions Δ𝑛 whose areas we can compute, and add up the results.
In Section 2.4 we deduced the formula Φ𝑢 × Φ𝑣 = (det 𝐽(𝜓 −1 ○ Φ)(𝑢,𝑣) )Ψ𝑢̃ × Ψ̃𝑣 .
Therefore, if we have Δ ⊆ Φ(𝑈) ∩ Ψ(𝑈), the equality

∬ ∣Φ𝑢 × Φ𝑣 ∣ 𝑑𝑢 𝑑𝑣 = ∬ ∣Ψ𝑢̃ × Ψ̃𝑣 ∣𝑑̃

𝑢 𝑑̃
𝑣
Φ−1 (Δ) Ψ−1 (Δ)

is a consequence of the change of variables theorem for multiple integrals, and it

follows that the area is well-defined, being independent of any parameterizations
used to compute it.
To motivate the formula

∬ ∣Φ𝑢 × Φ𝑣 ∣ 𝑑𝑢 𝑑𝑣
Φ−1 (Δ)

for the calculation of the area of Δ, let us cover Φ−1 (Δ) with a fine lattice of horizontal
and vertical lines, and let 𝑅𝑖, 𝑗 = [𝑢 𝑖 , 𝑢 𝑖+1 ] × [𝑣 𝑗 , 𝑣 𝑗+1 ] be any rectangle of this lattice,
whose intersection with Φ−1 (Δ) be non-empty. Then the integral in question is the
limit, as the maximum diameter of the 𝑅𝑖, 𝑗 tends to zero, of the sums

∑(𝑢 𝑖+1 − 𝑢 𝑖 )(𝑣 𝑗+1 − 𝑣 𝑗 )∣Φ𝑢 × Φ𝑣 ∣∣ =

𝑖, 𝑗 (𝑢𝑖 ,𝑣 𝑗 )

= ∑ ∣(𝑢 𝑖+1 − 𝑢 𝑖 )Φ𝑢 × (𝑣 𝑗+1 − 𝑣 𝑗 )Φ𝑣 ∣

𝑖, 𝑗

— where each summand gives the area of the parallelogram of sides (𝑢 𝑖+1 − 𝑢 𝑖 )Φ𝑢
and (𝑣 𝑗+1 − 𝑣 𝑗 )Φ𝑣 . The sides of this parallelogram are tangent to Φ(𝑢 𝑖 , 𝑣 𝑗 ) and have
lengths approximating the sides of the “rectangle” Φ(𝑅𝑖, 𝑗 ).
We now want to express the area using the coefficients 𝐸, 𝐹 and 𝐺. From the
identity
∣Φ𝑢 × Φ𝑣 ∣2 + ⟨Φ𝑢 , Φ𝑣 ⟩2 = ∣Φ𝑢 ∣2 ∣Φ𝑣 ∣2 ,
we obtain
54 2 Regular Surfaces

√
∣Φ𝑢 × Φ𝑣 ∣ = 𝐸𝐺 − 𝐹 2 ,
and therefore the area of Δ is given by the integral
√
∬ 𝐸𝐺 − 𝐹 2 𝑑𝑢 𝑑𝑣.
Φ−1 (Δ)

We again point out that it follows from this formula that the notion of area depends
only on the knowledge of the first fundamental form.

Example 2.5.2 The coefficients of the first fundamental form of the spherical coordi-
nates (𝜑, 𝜃) in S2 (example 2.1.1 D) are 𝐸 = sin2 𝜃, 𝐹 = 0 and 𝐺 = 1. The area of S2
is then given by the integral
𝜋 𝜋
∫ (∫ sin 𝜃 𝑑𝜃) 𝑑𝜑 = 4𝜋.
−𝜋 0

It is also interesting to note that the area of the spindle between the meridians 𝜑 = 0
and 𝜑 = 𝜑0 is equal to 2𝜑0 ; more generally, any spindle of amplitude 𝜑0 (bounded
by two maximal semicircles of S2 that intersect at an angle 𝜑0 ∈ ]0, 𝜋[) has area
2𝜑0 . This allows us to deduce Girard’s formula, which gives the area of a spherical
triangle (which is the figure inside S2 bounded by three maximal circular arcs) as a
function of its interior angles.
Assume that such a triangle T has vertices 𝐴, 𝐵, 𝐶 and interior angles 𝜑1 , 𝜑2 , 𝜑3
and denote by 𝛼, 𝛽, 𝛾 the maximal circles containing respectively the pairs of points
𝐵 and 𝐶, 𝐴 and 𝐶, 𝐴 and 𝐵. The antipodes 𝐴, ̃ 𝐵,
̃ and 𝐶̃ of the vertices of T form a
triangle 𝐶̃ bounded by arcs of the same maximal circles 𝛼, 𝛽 and 𝛾 (see Fig. 2.10).
Because they are antipodes of each other, T and T̃ have the same area (see Exercise
51 in this section).
The two maximal circles 𝛽 and 𝛾 define two spindles of amplitude 𝜑1 in S2 ; one
of them contains T and the other one T̃ . We denote by Δ1 the union of these two
spindles, and define analogously (using the pairs 𝛼 and 𝛾, 𝛼 and 𝛽) the regions Δ2
and Δ3 . The union of the Δ𝑖 covers S2 , but each point of 𝑇 ∪ T̃ is counted three times,
since, for 𝑖 ≠ 𝑗, we have Δ𝑖 ∩ Δ 𝑗 = T ∪ T̃ . Thus,
3
∑ area(Δ𝑖 ) = area(S ) + 2[area(T ) + area(T̃ )]
2
𝑖=1

and therefore

1 3
area(T ) = {∑ area(Δ𝑖 ) − area(S2 )}
4 𝑖=1
= 𝜑1 + 𝜑2 + 𝜑3 − 𝜋.

We can describe our conclusion by saying that the area of a spherical triangle is
proportional to its spherical excess (with proportionality constant equal to the square
2.5 Areas, Lengths, and Angles: The First Fundamental Form 55

of the radius). This formula is generalized by the remarkable Gauss-Bonnet theorem,

which we discuss later. ◻

We end this section by defining what is meant by the integral of a real function
defined on a surface: given a function 𝑓 ∶ 𝑆 → R , a parameterization (𝑈, Φ) of 𝑆, and
a region Δ ⊆ Φ(𝑈), the integral of 𝑓 along Δ is
√
∫ 𝑓 𝑑𝜎 = ∬ 𝑓 ○ Φ(𝑢, 𝑣) 𝐸𝐺 − 𝐹 2 𝑑𝑢 𝑑𝑣.
Δ Φ−1 (Δ)

For regions not contained in a single parametrized neighborhood, we partition them,

as before, into smaller regions and add up the results.

~ j1 ~
C B

j2 j3
a
B C

g ~
A

D1

j1
A
j1
D3 D2

j2 j3
B
j2 C j3
D2 D1 D3

Figure 2.10
56 2 Regular Surfaces

In particular, we have just defined what is meant by ∫𝑆 𝑓 𝑑𝜎. Again it can be shown
that these definitions do not depend on the parameterizations
√ we use. The “quantity”
𝑑𝜎, which in local coordinates is written 𝐸𝐺 − 𝐹 2 𝑑𝑢 𝑑𝑣, is usually called area
element of the surface.

Exercises
48. Obtain the coefficients of the first fundamental form for: (i) the helicoid (choose a
parameterization); (ii) the sphere (parametrized by the inverse of the stereographic
projection); (iii) a surface of revolution (parametrized as in example 2.1.1 E).
49. The coordinate curves of the parameterization Φ(𝑢, 𝑣) constitute a Chebyshev
net if the opposite sides of any quadrangle formed by them have the same lengths.
𝜕𝐸 𝜕𝐺
Show that this happens if and only if ≡ ≡ 0.
𝜕𝑣 𝜕𝑢
50. Find all curves of the cylinder {(𝑥, 𝑦, 𝑧) ∈ R 3 ∶ 𝑥 2 + 𝑦 2 = 1} that intersect the
generatrices (vertical lines) at a constant angle.
51. We say that a diffeomorphism of one surface onto another (or of an open subset
of one surface onto an open subset of another surface) preserves area if the area of
each open set is equal to that of its image.
(a) Let Ψ∶ 𝑈 → 𝑆1 and Φ∶ 𝑈 → 𝑆2 be parameterizations of two surfaces, 𝐸, 𝐹, 𝐺
̃ 𝐹,
and 𝐸, ̃ 𝐺̃ the coefficients of the first fundamental form for Ψ and Φ, respectively.
̃ − 𝐹̃2
Show that Ψ ○ Φ−1 preserves areas if and only if the functions 𝐸𝐺 − 𝐹 2 and 𝐸̃𝐺
are identical.
(b) Show that the antipodal mapping S2 → S2 , (𝑥, 𝑦, 𝑧) ↦ (−𝑥, −𝑦, −𝑧), preserves
areas.
(c) The Archimedes projection sends each point 𝑝 of S2 (except the north and
south poles) to the point of intersection of the circumscribing vertical cylinder with
the half-line 𝑞 𝑝 whose origin is the point 𝑞 on the 𝑧-axis at the same height as 𝑝.
Show that this mapping preserves areas.
(d) Find a mapping of an open subset of T 2 into the plane that preserves areas.
52. Let 𝑈 be a connected open subset of R 2 and let ℎ∶ 𝑈 → R be a differentiable
function. Consider the surface 𝑆 = {(𝑥, 𝑦, 𝑧) ∈ R 3 ∶ (𝑥, 𝑦) ∈ 𝑈, 𝑧 = ℎ(𝑥, 𝑦)}. Show
that:
(a) the mapping 𝜋∶ 𝑆 → 𝑈 given by 𝜋(𝑥, 𝑦, 𝑧) = (𝑥, 𝑦) is a diffeomorphism;
(b) 𝜋 decreases area (for any open subset 𝑊 ⊆ 𝑆 the area of 𝜋(𝑊) is ≤ to the area
of 𝑊);
(c) if 𝜋 preserves areas then 𝑆 is contained in a horizontal plane.
53. The gnomonic projection PΠ ∶ S2 ∖𝛾Π → Π sends each point on the sphere (except
the points on a certain maximal circle 𝛾Π ) into a tangent plane Π by projecting it
from the center of the sphere. Show that:
2.5 Areas, Lengths, and Angles: The First Fundamental Form 57

(a) two points have the same image for PΠ if and only if they are antipodes, and
the restriction PΠ to each of the hemispheres of S2 ∖𝛾Π is a diffeomorphism;
(b) the maximal circles of S2 are transformed into the lines of Π;
(c) the two drawings of Fig. 2.10 are related by a gnomonic projection.
Chapter 3
The Geometry of the Gauss Map

In this chapter we deal with the extrinsic geometry of the surface, by defining
quantities (curvatures) that express how the surface is embedded in its ambient space.
The main tool for this study is the normal vector field to the surface; hence, we shall
deal only with oriented surfaces.

3.1 The Gauss Map and its Derivative

Given an oriented surface 𝑆, the Gauss map is the field of normal vectors 𝑁∶ 𝑆 → S2
that defines the orientation of 𝑆. We noted in the previous chapter (following
Proposition 2.4.3) that 𝑁 is a differentiable mapping. By analogy with planar curves,
it is to be expected that the study of the variation of 𝑁 (i.e., of its derivative) will
shed light on the local shape of 𝑆.
For 𝑝 ∈ 𝑆 the tangent spaces 𝑇 𝑝 𝑆 and 𝑇𝑁 ( 𝑝) S2 are the same subspace of R 3 ,
since both are the orthogonal complement of the line generated by 𝑁(𝑝). This
means that the derivative 𝐷𝑁 𝑝 is an endomorphism 𝑇 𝑝 𝑆 → 𝑇 𝑝 𝑆. Let us now take a
parameterization Φ(𝑢, 𝑣) of 𝑆 and put 𝑁(𝑢, 𝑣) = 𝑁 ○ Φ(𝑢, 𝑣). By definition of the
derivative, we have
𝑁𝑢 = 𝐷𝑁Φ(𝑢,𝑣) (Φ𝑢 ), 𝑁 𝑣 = 𝐷𝑁Φ(𝑢,𝑣) (Φ𝑣 ).
Differentiating the equalities ⟨Φ𝑢 , 𝑁⟩ = 0 = ⟨Φ𝑣 , 𝑁⟩ with respect to 𝑣 and 𝑢 respec-
tively, we obtain
⟨Φ𝑢𝑣 , 𝑁⟩ + ⟨Φ𝑢 , 𝑁 𝑣 ⟩ = 0, ⟨Φ𝑣𝑢 , 𝑁) + ⟨Φ𝑣 , 𝑁𝑢 ⟩ = 0,
and from this, subtracting term by term, and given that Φ𝑢𝑣 = Φ𝑣𝑢 , we get ⟨Φ𝑢 , 𝑁 𝑣 ⟩ =
⟨Φ𝑣 , 𝑁𝑢 ⟩. The latter equality can be rewritten in the form ⟨Φ𝑢 , 𝐷𝑁Φ(𝑢,𝑣) (Φ𝑣 )⟩ =
⟨Φ𝑣 , 𝐷𝑁Φ(𝑢,𝑣) (Φ𝑢 )⟩. It follows that for all vectors w1 , w2 ∈ 𝑇Φ(𝑢,𝑣) 𝑆, we have

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 59

P. V. Araújo, Differential Geometry, https://doi.org/10.1007/978-3-031-62384-4_3
60 3 The Geometry of the Gauss Map

⟨w1 , 𝐷𝑁Φ(𝑢,𝑣) (w2 )⟩ = ⟨w2 , 𝐷𝑁Φ(𝑢,𝑣) (w1 )⟩, (3.1)

To verify the equality it suffices to express w1 and w2 as linear combinations of Φ𝑢

and Φ𝑣 .
Equality (3.1) means that 𝐷𝑁 𝑝 is a symmetric linear mapping of 𝑇 𝑝 𝑆 (with respect
to the inner product ⟨⋅ , ⋅⟩ 𝑝 on 𝑇 𝑝 𝑆). In general, a linear endomorphism 𝐿∶ 𝐸 → 𝐸 of
a finite-dimensional real vector space equipped with an inner product ⟪ ⋅, ⋅ ⟫, is called
symmetric (or self-adjoint) if, for all w1 , w2 ∈ 𝐸, we have

⟪w1 , 𝐿(w2 )⟫ = ⟪𝐿(w1 ), w2 ⟫. (3.2)

The next proposition, whose complete proof can be found in numerous Linear Algebra
texts, gathers the essentials about symmetric endomorphisms.
Proposition 3.1.1 Let 𝐸 be a space with inner product ⟪ ⋅, ⋅ ⟫, B = (e1 , . . . , e𝑛 ) an
orthonormal basis of 𝐸, and 𝐿∶ 𝐸 → 𝐸 an endomorphism. Then:
(i) 𝐿 is symmetric if and only if its matrix with respect to the basis B is symmetric;
(ii) if 𝐿 is symmetric, 𝐸 has an orthonormal basis formed by eigenvectors of 𝐿.
Proof (i) Since B is orthonormal, the matrix of 𝐿 in this basis is 𝑀 = (𝑎 𝑖 𝑗 )1≤𝑖, 𝑗≤𝑛
given by 𝑎 𝑖 𝑗 = ⟪e𝑖 , 𝐿(e 𝑗 )⟫. We thus observe that if 𝐿 is symmetric then 𝑎 𝑖 𝑗 = 𝑎 𝑗𝑖 for
all 𝑖, 𝑗, a condition, that expresses the symmetry of 𝑀. Conversely, if 𝑀 is symmetric
then ⟪e𝑖 , 𝐿(e 𝑗 )⟫ = ⟪𝐿(e𝑖 ), e 𝑗 ⟫ for all 1 ≤ 𝑖, 𝑗 ≤ 𝑛, and it follows that the symmetry
condition (3.2) is verified for any two vectors that are linear combinations of the e𝑖 ;
but every vector of 𝐸 is such a linear combination, and therefore 𝐿 is symmetric.
(ii) We give the proof only in the case 𝑛 = 2, the only one we will need. Let

𝑎𝑐
𝑀 =[ ]
𝑐𝑏

be the (symmetric) matrix of 𝐿 with respect to the orthonormal basis B = (e1 , e2 ).

The eigenvalues of 𝐿 are the roots of the characteristic polynomial 𝑃(𝜆) = 𝜆2 − (𝑎 +
2
𝑏)𝜆 + 𝑎𝑏 − 𝑐2 , whose discriminant is Δ = (𝑎 − 𝑏) + 4𝑐2 ≥ 0. If Δ = 0 then 𝑐 = 0
and 𝑎 = 𝑏, which shows that 𝐿 is a homothety, and so any orthonormal basis of 𝐸
is formed by eigenvectors of 𝐿. If Δ > 0, then 𝐿 has two real eigenvalues 𝜆1 < 𝜆2 ,
and we let v1 and v2 be the associated unit eigenvectors . These vectors constitute the
promised basis, since

𝜆1 ⟪v1 , v2 ⟫ = ⟪𝐿(v1 ), v2 ⟫ = ⟪v1 , 𝐿(v2 )⟫ = 𝜆2 ⟪v1 , v2 ⟫,

and therefore ⟪v1 , v2 ⟫ = 0. ◻

Corollary 3.1.2 Let 𝜁∶ 𝐸 × 𝐸 → R be a symmetric bilinear form on the Euclidean
space 𝐸. Then there exists an orthonormal basis C = (v1 , . . . , v𝑛 ) of 𝐸 such that
𝜁(v𝑖 , v 𝑗 ) = 0 for all 1 ≤ 𝑖 < 𝑗 < 𝑛.
Proof Consider the matrix 𝑀 = (𝑎 𝑖 𝑗 )𝑖≤𝑖, 𝑗≤𝑛 of the bilinear form 𝜁 relative to an
orthonormal basis B = (e1 , . . . , e𝑛 ) of 𝐸. This matrix, defined by 𝑎 𝑖 𝑗 = 𝜁(e𝑖 , e 𝑗 ), is
3.1 The Gauss Map and its Derivative 61

symmetric because of the symmetry of 𝜁. Let 𝐿 be the linear mapping whose matrix
with respect to B is 𝑀: by 3.1.1 (i), 𝐿 is symmetric. A simple calculation shows that
𝜁(v, w) = ⟪v, 𝐿(w)⟫. By 3.1.1 there exists an orthonormal basis C of 𝐸 formed by
eigenvectors of 𝐿, and C is the sought basis. ◻

The eigenvalues 𝑘 1 (𝑝) ≤ 𝑘 2 (𝑝) of the symmetric endomorphism −𝐷𝑁 𝑝 (beware

of the minus sign!) are called principal curvatures of 𝑆 at the point 𝑝. If 𝑘 1 (𝑝) < 𝑘 2 (𝑝),
we call principal directions the two orthogonal directions defined in 𝑇 𝑝 𝑆 by the
eigenvectors of −𝐷𝑁 𝑝 .
To justify this terminology we introduce yet another definition. Let 𝛼∶ ]𝑎, 𝑏[→ 𝑆
be a curve parametrized by arc length. The normal curvature of 𝛼 at 𝛼(𝑠) is the
component of 𝛼′′ (𝑠) in the direction of the normal to 𝑆 at that point, and is given
by 𝑘 𝑛 (𝛼, 𝑠) = ⟨𝛼′′ (𝑠), 𝑁 ○ 𝛼(𝑠)⟩. Note that this quantity does not depend on the
orientation of the curve and that if the curve is not parametrized by arc length, the
1
formula for computing the normal curvature is 𝑘 𝑛 (𝛼, 𝑡) = 2
⟨𝛼′′ (𝑡), 𝑁 ○ 𝛼(𝑡)⟩,
𝑣(𝑡)
where 𝑣(𝑡) = ∣𝛼′ (𝑡)∣.

Proposition 3.1.3 (i) The normal curvature 𝑘 𝑛 (𝛼, 𝑠) at 𝛼(𝑠) depends only on the
tangent direction to the curve at instant 𝑠: more precisely, if 𝛼 and 𝛽 are curves in 𝑆
tangent to each other at 𝛼(𝑠0 ) = 𝛽(𝑡0 ) = 𝑝 0 then 𝑘 𝑛 (𝛼, 𝑠0 ) = 𝑘 𝑛 (𝛽, 𝑡0 ).
(ii) The set of normal curvatures at 𝑝 0 is the interval [𝑘 1 (𝑝 0 ), 𝑘 2 (𝑝 0 )]. If
𝑘 1 (𝑝 0 ) < 𝑘 2 (𝑝 0 ), then the minimum and maximum of these normal curvatures are
the principal curvatures at 𝑝 0 , which occur precisely in the principal directions
associated with 𝑘 1 (𝑝 0 ) and 𝑘 2 (𝑝 0 ).

Proof Let us put 𝑁(𝑠) = 𝑁 ○ 𝛼(𝑠). Differentiating the equality ⟨𝛼′ (𝑠), 𝑁(𝑠)⟩ = 0,
we obtain ⟨𝛼′′ (𝑠), 𝑁(𝑠)⟩ + ⟨𝛼′ (𝑠), 𝑁 ′ (𝑠)⟩ = 0, and from this we get 𝑘 𝑛 (𝛼, 𝑠) =
⟨𝛼′ (𝑠), −𝑁 ′ (𝑠)⟩ = ⟨𝛼′ (𝑠), −𝐷𝑁 𝑝 (𝛼′ (𝑠))⟩, where we let 𝑝 = 𝛼(𝑠). This equality
shows that 𝑘 𝑛 (𝛼, 𝑠) only depends on 𝛼′ (𝑠) ∈ 𝑇 𝑝 𝑆 and proves statement (i).
Let us now fix 𝑝 0 = 𝛼(𝑠0 ) and let (v1 , v2 ) be an orthonormal basis of 𝑇 𝑝0 𝑆
consisting of eigenvectors of −𝐷𝑁 𝑝0 . Putting 𝛼′ (𝑠0 ) = 𝑎 v1 +𝑏 v2 , we have 𝑎 2 +𝑏 2 = 1;
furthermore,

𝑘 𝑛 (𝛼, 𝑠0 ) = ⟨𝑎v1 + 𝑏v2 , −𝐷𝑁 𝑝0 (𝑎v1 + 𝑏v2 )⟩

= 𝑎 2 ⟨v1 , −𝐷𝑁 𝑝0 (v1 )⟩ + 𝑏 2 ⟨v2 , −𝐷𝑁 𝑝0 (v2 )⟩
= 𝑘 1 𝑎2 + 𝑘 2 𝑏2 .

Thus we obtain the inequalities

𝑘 1 = 𝑘 1 (𝑎 2 + 𝑏 2 ) ≤ 𝑘 𝑛 (𝛼, 𝑠0 ) ≤ 𝑘 2 (𝑎 2 + 𝑏 2 ) = 𝑘 2 ,

From which it follows that the normal curvatures cover the entire interval [𝑘 1 , 𝑘 2 ] and
that if 𝑘 1 < 𝑘 2 then the minimum is only reached for 𝛼′ (𝑠0 ) = ±v1 and the maximum
for 𝛼′ (𝑠0 ) = ±v2 . ◻
62 3 The Geometry of the Gauss Map

As we have said, the normal curvature 𝑘 𝑛 (𝛼, 𝑠) gives the component of the
curvature vector 𝛼′′ (𝑠0 ) of 𝛼 in the direction of the normal 𝑁 ○ 𝛼(𝑠) to the surface.
If these vectors are collinear, i.e., if the principal normal to the curve 𝛼 at instant 𝑠
points in the direction of the normal to the surface at 𝛼(𝑠), then the absolute value of
𝑘 𝑛 (𝛼, 𝑠) is equal to the curvature of 𝛼 at that point.
Given an arbitrary direction v ∈ 𝑇 𝑝 𝑆 with ∣v∣ = 1, there exists at least one curve
that passes through 𝑝 with velocity v, and whose principal normal at 𝑝 points in the
direction of 𝑁(𝑝): the intersection of 𝑆 with the plane that passes through 𝑝 and is
parallel to the vectors 𝑁(𝑝) and v (see Fig. 3.1 and Exercise 34 in Section 2.2). A
curve obtained this way is called a normal section of 𝑆 at 𝑝.

Figure 3.1

This means that to analyze the normal curvatures at 𝑝 ∈ 𝑆 it suffices to study

the curvatures of the normal sections. The sign of 𝑘 𝑛 will depend on whether the
concavity at 𝑝 of such a curve points in the direction of 𝑁(𝑝) or in the opposite
direction.
The Gaussian curvature of 𝑆 at 𝑝 is defined by 𝐾(𝑝) = 𝑘 1 (𝑝)𝑘 2 (𝑝), and the mean
curvature is 𝐻(𝑝) = 12 (𝑘 1 (𝑝) + 𝑘 2 (𝑝)); equivalently, 𝐾(𝑝) and 𝐻(𝑝) are equal,
respectively, to the determinant and the semi-trace of the linear mapping −𝐷𝑁 𝑝 .
According to the value of these curvatures, a point 𝑝 ∈ 𝑆 is called
• elliptic if 𝐾(𝑝) > 0 (i.e., if 𝑘 1 (𝑝) and 𝑘 2 (𝑝) are both positive or both negative);
• hyperbolic if 𝐾(𝑝) < 0 (the principal curvatures have opposite signs);
• parabolic if 𝐾(𝑝) = 0 and 𝐻(𝑝) ≠ 0 (one of the principal curvatures is zero, the
other one is nonzero);
• planar if 𝐾(𝑝) = 0 = 𝐻(𝑝) (both principal curvatures are zero);
2
• umbilical if 𝑘 1 (𝑝) = 𝑘 2 (𝑝) (this condition is equivalent to the equality 𝐻(𝑝) −
𝐾(𝑝) = 0).
3.1 The Gauss Map and its Derivative 63

Note that any point on the surface belongs to one and only one of the first four
classes, but that the umbilical points can be either elliptical or planar. We further note
that although the sign of the principal curvatures depends on the orientation of 𝑆,
the Gaussian curvature does not change when one changes the orientation and the
classification of the points we gave above does not depend on the orientation chosen.
Hence, since any surface is locally orientable (any parametrized neighborhood has
an orientation induced by the parameterization), the classification given extends also
to points on non-orientable surfaces.
In the next section we study the meaning of these definitions with the help of local
coordinates. Now we give some examples.

Examples 3.1.4 A. All points in a plane are planar: in fact, the normal sections are
straight lines, and therefore have zero curvature. This means that all normal curvatures
(and hence both principal curvatures) of the plane are zero.
B. The normal sections of a sphere with radius 𝑟 are maximal circles whose curvature
is 1/𝑟. This means that at each point, the absolute values of all normal curvatures
are equal to 1/𝑟, and therefore the two principal curvatures are equal (if they were
−1/𝑟 and 1/𝑟, some normal curvature would be zero, which is not the case) and
have absolute value 1/𝑟. All points on the sphere are therefore umbilical, and their
Gaussian curvature is constant and positive, equal to 1/𝑟 2 .
Of course, the analysis of the signs of the normal curvatures could be replaced, in
this example, by a simple calculation. But with this analysis we illustrate a useful
principle: at the point 𝑝 ∈ 𝑆 there is some direction in which the normal curvature
vanishes (called an asymptotic direction) if and only if 𝐾(𝑝) ≤ 0.
C. Consider the hyperbolic paraboloid {(𝑥, 𝑦, 𝑧)∶ 𝑧 = 𝑥 2 − 𝑦 2 } with the orientation
induced by the parameterization Φ(𝑢, 𝑣) = (𝑢, 𝑣, 𝑢 2 − 𝑣 2 ), i.e.

1
𝑁(𝑢, 𝑣) = √ (−2𝑢, 2𝑣, 1).
1 + 4𝑢 2 + 4𝑣 2
At the point 𝑂 = Φ(0, 0) we have

−𝑁𝑢 = (2, 0, 0) = 2Φ𝑢 ,

−𝑁 𝑣 = (0, −2, 0) = −2Φ𝑣 .

The principal curvatures at 𝑂 are thus 2 and −2, the principal directions are those of
the vectors Φ𝑢 = (1, 0, 0) and Φ𝑣 = (0, 1, 0), and the Gaussian curvature is negative,
equal to −4.
D. Let 𝐸 be the ellipsoid
𝑥 2 𝑦2 𝑧2
+ + = 1,
𝑎2 𝑏2 𝑐2
where 𝑎 ≥ 𝑏 ≥ 𝑐 > 0. We use the symmetric bilinear form 𝜁∶ R 3 × R 3 → R given by

𝑥𝑥 ′ 𝑦𝑦 ′ 𝑧𝑧′
𝜁((𝑥, 𝑦, 𝑧), (𝑥 ′ , 𝑦 ′ , 𝑧 ′ )) = + 2 + 2
𝑎2 𝑏 𝑐
64 3 The Geometry of the Gauss Map

and we denote by Q the associated quadratic form:

𝑥 2 𝑦2 𝑧2
Q(𝑥, 𝑦, 𝑧) = + + ⋅
𝑎2 𝑏2 𝑐2
Given 𝑝 0 ∈ E and a plane Π passing through the origin (0, 0, 0), let us study the
intersection of E with the plane Π0 = 𝑝 0 + Π.
Note that, for every vector w ∈ R 3 , we have
1 1
2
∣w∣2 ≤ Q(w) ≤ 2 ∣w∣2 . (3.3)
𝑎 𝑐
Given now w ∈ Π (so that 𝑝 0 + w ∈ Π0 ), we have

Q(𝑝 0 + w) = Q(w) + 2𝜁(𝑝 0 , w) + Q(𝑝 0 )

= Q(w) + 𝐿(w) + 1,

where 𝐿∶ Π → R is a linear form. Thus, the intersection F = E ∩ Π0 is the set of sums

𝑝 0 + w, where w ∈ Π satisfies Q(w) + 𝐿(w) = 0.
Consider the bilinear form 𝜁̃ = 𝜁∣Π×Π and the associated quadratic form Q ̃=
Q∣Π . By corollary 3.1.2, there exists an orthonormal basis (v1 , v2 ) of Π such that
̃ 1 ) and Q(v
̃ 1 , v2 ) = 0. By (3.3), Q(v
𝜁(v ̃ 2 ) are positive, belonging to the interval
1 1
[ 2 , 2 ]. We then have
𝑎 𝑐

̃ 𝑥 v1 + ̃ ̃𝑥2 ̃𝑦2
Q(̃ 𝑦 v2 ) = 2 + ,
𝑎̃ ̃
𝑏2

where we can assume that 𝑎 ≥ 𝑎̃ ≥ ̃

𝑏 ≥ 𝑐. Using coordinates ̃
𝑥, ̃
𝑦 the equation of
−𝑝 0 + F is written
̃𝑥2 ̃ 𝑦2
+ + 𝜆̃
𝑥 + 𝜇̃
𝑦 = 0,
𝑎̃2 ̃
𝑏2
or, “completing squares”,
2 2
𝑥 −̃
(̃ 𝑥0 ) 𝑦−̃
(̃ 𝑦0)
+ = 𝑟 2,
𝑎̃2 ̃
𝑏2
where 𝜆, 𝜇, ̃𝑥0 , ̃
𝑦 0 and 𝑟 are certain constants. Assuming that 𝑟 > 0 (otherwise
𝜆 = 𝜇 = 0 and Π0 ∩ F = {𝑝 0 }), the last equation is equivalent to
2 2
𝑥 −̃
(̃ 𝑥0 ) 𝑦−̃
(̃ 𝑦0)
+ = 1,
(̃
𝑎 𝑟)
2
̃
(𝑏 𝑟)
2

which describes an ellipse of semi-major axis 𝑎̃𝑟 and semi-minor axis ̃

𝑏 𝑟. Thus, the
maximum and minimum curvatures of F are, respectively,
3.1 The Gauss Map and its Derivative 65

𝑎̃𝑟 𝑎̃ ̃
𝑏𝑟 ̃
𝑏
= and =
(̃
𝑏 𝑟)
2 ̃
𝑏2𝑟 (̃
𝑎 𝑟)
2 𝑎̃2 𝑟

(see Exercise 6, in Section 1.3). Since F ⊆ E and the diameter of E is 2𝑎, we have
𝑎̃𝑟 ≤ 𝑎. Hence, the curvature of F is, at all points, greater than or equal to
̃
𝑏 ̃
𝑏 𝑐
≥ ≥ 2⋅
𝑎̃ 𝑟 𝑎 𝑎̃ 𝑎
2

𝑐
We thus prove that the normal curvatures of E are not less than in absolute value.
𝑎2
Hence (see example B) the Gaussian curvature of E is positive at all points, being
bounded by 𝑐2 /𝑎 4 .
We now want to obtain an upper bound for the principal curvatures of E, and for
this we need a lower bound for the diameter of its normal sections. The normal line
to E at the point (𝑥 0 , 𝑦 0 , 𝑧 0 ) ∈ E, given by
𝑥0 𝑦 0 𝑧0
𝑡 ↦ (𝑥 0 , 𝑦 0 , 𝑧 0 ) + 𝑡 ( , , ),
𝑎2 𝑏2 𝑐2
intersects E for 𝑡 = 0 and for 𝑡 = 𝑡 0 , where

𝑥2 𝑦02 𝑧02
−2 ( 𝑎04 + 𝑏4
+ 𝑐4
)
𝑡0 = ⋅
𝑥02 𝑦02 𝑧02
𝑎6
+ 𝑏6
+ 𝑐6

The length of the line segment between the two intersections is then
3/2
√ 𝑥2 𝑦02 𝑧02
𝑥02 𝑦 20 𝑧20 2( 𝑎04 + 𝑏4
+ 𝑐4
)
∣𝑡0 ∣ + + =
𝑎4 𝑏4 𝑐4 𝑥02 𝑦02 𝑧02
𝑎6
+ 𝑏6
+ 𝑐6
√
𝑥02 𝑦 20 𝑧 20
≥ 2𝑐 2
+ +
𝑎4 𝑏4 𝑐4
√
2𝑐2 𝑥02 𝑦 20 𝑧 20 2𝑐2
≥ + + = ⋅
𝑎 𝑎2 𝑏2 𝑐2 𝑎
Thus, under the assumption that F = E ∩ Π0 is a normal section of E, we have
𝑐2 1 𝑎̃ 𝑎
𝑎̃𝑟 ≥ , or ≤ 2 . The curvature of F is then not greater than
𝑎 𝑟 𝑐
𝑎̃ 𝑎̃2 𝑎 𝑎3
≤ ≤ 4,
̃
𝑏2𝑟 ̃ 𝑏2 𝑐2 𝑐
and so the absolute values of the normal curvatures of E are also not greater than
𝑎 3 /𝑐4 . In conclusion: at any point 𝑝 ∈ E we have
𝑐2 𝑎6
≤ 𝐾(𝑝) ≤ ⋅
𝑎4 𝑐8
66 3 The Geometry of the Gauss Map

Note that when E is a sphere then 𝑎 = 𝑏 = 𝑐, and both these inequalities become
the equality already seen in example B. It should however be made clear that it is
possible, by more ingenious methods than ours, to obtain an explicit expression for
the curvature of the points of E. This example was intended to show that we can
estimate (in this case obtain lower and upper bounds) the Gaussian curvature without
computing it explicitly. ◻

In examples 3.1.4 A, B we observed that all points on the plane and the sphere are
umbilical; we end the section by showing that these are the only surfaces with such a
property.

Proposition 3.1.5 Let 𝑆 be a connected surface whose points are all umbilical. Then
𝑆 is contained in a sphere or in a plane.

Proof Our hypothesis implies the existence of a function 𝜆∶ 𝑆→R such that, for every
𝑝∈𝑆, 𝐷𝑁 𝑝 is a homothety of ratio 𝜆(𝑝). Let us take a parameterization (𝑈, Φ) of 𝑆,
with 𝑈 connected, and put 𝑁(𝑢, 𝑣) = 𝑁 ○ Φ(𝑢, 𝑣) and 𝜆(𝑢, 𝑣) = 𝜆 ○ Φ(𝑢, 𝑣). We then
have
𝑁𝑢 = 𝜆(𝑢, 𝑣)Φ𝑢
(3.4)
𝑁 𝑣 = 𝜆(𝑢, 𝑣)Φ𝑣
and these equalities ensure that 𝜆(𝑢, 𝑣) is differentiable, for we deduce from them
that
⟨𝑁𝑢 , Φ𝑢 ⟩ ⟨𝑁 𝑣 , Φ𝑣 ⟩
𝜆(𝑢, 𝑣) = = ⋅
∣Φ𝑢 ∣2 ∣Φ𝑣 ∣2
By differentiation of (3.4) we obtain

𝑁𝑢𝑣 = 𝜆 𝑣 Φ𝑢 + 𝜆Φ𝑢𝑣
𝑁 𝑣𝑢 = 𝜆𝑢 Φ𝑣 + 𝜆Φ𝑣𝑢

Whence, subtracting term by term,

𝜆 𝑣 Φ𝑢 − 𝜆𝑢 Φ𝑣 = 0,
an equality which is only possible when 𝜆 𝑣 and 𝜆𝑢 are identically zero. Thus, the
function 𝜆(𝑢, 𝑣) is constant on 𝑈, and therefore 𝜆∶ 𝑆 → R is locally constant, hence
(since 𝑆 is connected) constant, equal to 𝜆 ∈ R .
If 𝜆 = 0 then by (3.4) the normal vector 𝑁 is constant, and the function 𝑝 ↦ ⟨𝑁, 𝑝⟩
is locally constant, hence constant, on 𝑆, which means that 𝑆 is contained in a plane
{𝑝 ∈ R 3 ∶ ⟨𝑁, 𝑝⟩ = 𝑎}, for some 𝑎 ∈ R .
1
If 𝜆 ≠ 0 then, again by (3.4), the mapping 𝑆 → R 3 given by 𝑝 ↦ 𝑝 − 𝑁(𝑝) is
𝜆
locally constant, hence constant. Denoting by 𝑞 0 this constant, we have, for all 𝑝 in 𝑆,
∣𝑝 − 𝑞 0 ∣ = ∣ 𝜆1 𝑁(𝑝)∣ = ∣𝜆∣
1
, and therefore 𝑆 is contained in the sphere with center 𝑞 0
and radius 1
∣𝜆∣ . ◻

Exercises
3.2 The Second Fundamental Form 67

54. What is the region on the sphere covered by the image of the Gauss map of the
surface given by the equation: (i) 𝑧 = 𝑥 2 +𝑦 2 ; (ii) 𝑥 2 +𝑦 2 −𝑧 2 = 1; (iii) 𝑥 2 +𝑦 2 = cosh2 𝑧.
55.(a) Compute the principal curvatures at (0, 0, 0) of each of the following surfaces:
(i) 𝑧 = 𝑥 2 +𝑦 3 ; (ii) 𝑧 = 𝑥 2 +𝑦 4 ; (iii) 𝑧 = 𝑥 3 −3𝑥𝑦 2 . Sketch the surface (iii), indicating
the region that lies above the plane 𝑧 = 0.
(b) Conclude that when 𝑝 0 is a parabolic(or planar)pointof 𝑆 the following two
cases are possible: (i) there exists a neighborhood of 𝑝 0 in 𝑆 that lies entirely on the
same side of the tangent plane to 𝑆 at 𝑝 0 ; (ii) any neighborhood of 𝑝 0 in 𝑆 contains
points on both sides of the tangent plane.
56. (a) Show that at a point on a surface, the arithmetic mean of the normal curvatures
in two orthogonal directions is equal to the mean curvature at that point.
(b) Show that the mean curvature at 𝑝 ∈ 𝑆 is given by 𝜋1 ∫0 𝑘 𝑛 (𝜃) 𝑑𝜃, where
𝜋

𝑘 𝑛 (𝜃) is the normal curvature at 𝑝 in the direction that makes an angle 𝜃 with a fixed
principal direction.
57. Let 𝑆 be an oriented regular surface, and suppose that 𝑝 0 ∈ 𝑆 is a maximum of
the function 𝑓 ∶ 𝑆 → R , 𝑓 (𝑝) = ∣𝑝∣2 . Show that:
(a) the line segment [𝑂, 𝑝 0 ] is orthogonal to 𝑆 at 𝑝 0 ;
(b) the Gaussian curvature of 𝑆 at 𝑝 0 is greater than or equal to 1/ 𝑓 (𝑝 0 ) (use
Exercise 8, in Section 1.3);
(c) if 𝑆 is compact, then 𝑆 has some point with positive Gaussian curvature.
58. Show that if a surface is tangent to a plane along a regular curve then the points
on that curve are parabolic or planar.
59. Let 𝑝 be a hyperbolic point of 𝑆, and assume there exists a neighborhood 𝑈 of 𝑝
in 𝑆 such that (𝑝 + 𝑇 𝑝 𝑆) ∩ 𝑈 is the union of two regular curves that intersect at 𝑝.
Show that the tangent line at 𝑝 to each of these curves defines an asymptotic direction
in 𝑇 𝑝 𝑆.

3.2 The Second Fundamental Form

Using local coordinates, we will in this section continue the study of Gaussian
curvature, obtaining explicit formulas to compute it and a better understanding of its
geometric meaning. The tool is again a quadratic form, now related to the Gaussian
normal mapping.
We observed in Section 3.1 that 𝐷𝑁 𝑝 ∶ 𝑇 𝑝 𝑆 → 𝑇 𝑝 𝑆 is a symmetric linear mapping
with respect to the inner product ⟨⋅, ⋅⟩ 𝑝 on 𝑇 𝑝 𝑆, which means that the bilinear
form (v, w) ↦ ⟨v, −𝐷𝑁 𝑝 (w)⟩ is symmetric. The second fundamental form at
𝑝 ∈ 𝑆 is the quadratic form associated with this symmetric bilinear form, i.e.
Π 𝑝 (v) = ⟨v, −𝐷𝑁 𝑝 (v)⟩.
From the proof of 3.1.3 it follows that the normal curvature at 𝑝 in the direction
of v is precisely Π 𝑝 (v) when v∣ = 1 [if v ≠ 0 is not a unit vector, then that normal
curvature is given by Π 𝑝 (v)/∣v∣2 = Π 𝑝 (v)/𝐼 𝑝 (v)] and that the principal curvatures
68 3 The Geometry of the Gauss Map

are the maximum and minimum of the set {Π 𝑝 (v)∶ v ∈ 𝑇 𝑝 𝑆, ∣v∣ = 1}. This means that
at each point of the surface the second fundamental form gathers all information
about normal curvatures, principal curvatures and Gaussian curvature.

𝑒 𝑓
Given a parameterization Φ(𝑢, 𝑣), we want to determine the matrix ( ) of
𝑓 𝑔
ΠΦ(𝑢,𝑣) relative to the base (Φ𝑢 , Φ𝑣 ) of 𝑇Φ(𝑢,𝑣) 𝑆. The entries 𝑒, 𝑓 , 𝑔 of this matrix,
which are functions of (𝑢, 𝑣), are called the coefficients of the second fundamental
form in the coordinates (𝑢, 𝑣), and are computed by the formulas

𝑒 = ⟨Φ𝑢 , −𝐷𝑁Φ(𝑢,𝑣) (Φ𝑢 )⟩ = ⟨Φ𝑢 , −𝑁𝑢 ⟩

= ⟨Φ𝑢𝑢 , 𝑁⟩

𝑓 = ⟨Φ𝑢 , −𝐷𝑁Φ(𝑢,𝑣) (Φ𝑣 )⟩ = ⟨Φ𝑣 , −𝐷𝑁Φ(𝑢,𝑣) (Φ𝑢 )⟩

= ⟨Φ𝑢 , −𝑁 𝑣 ⟩ = ⟨Φ𝑣 , −𝑁𝑢 ⟩
= ⟨Φ𝑢𝑣 , 𝑁⟩)

𝑔 = ⟨Φ𝑣 , −𝐷𝑁Φ(𝑢,𝑣) (Φ𝑣 )⟩ = ⟨Φ𝑣 , −𝑁 𝑣 ⟩

= ⟨Φ𝑣𝑣 , 𝑁⟩.

For computational purposes, formulas that do not involve the derivatives of 𝑁 are
in general easier to handle. Once we have computed 𝑒, 𝑓 , 𝑔 it is easy to compute
the normal curvature of a curve 𝛼(𝑡) = Φ(𝑢(𝑡), 𝑣(𝑡)) expressed in local coordinates:
writing 𝛼′ (𝑡) = 𝑢 ′ (𝑡)Φ𝑢 + 𝑣 ′ (𝑡)Φ𝑣 , we have

Π 𝛼(𝑡) (𝛼′ (𝑡)) = ⟨𝑢 ′ Φ𝑢 + 𝑣 ′ Φ𝑣 , −𝐷𝑁 𝛼(𝑡) (𝑢 ′ Φ𝑢 + 𝑣 ′ Φ𝑣 )⟩

= 𝑢 ′2 ⟨Φ𝑢 , −𝐷𝑁 𝛼(𝑡) (Φ𝑢 )⟩ + 𝑢 ′ 𝑣 ′ (⟨Φ𝑢 , −𝐷𝑁 𝛼(𝑡) (Φ𝑣 )⟩+
+ ⟨Φ𝑣 , −𝐷𝑁 𝛼(𝑡) (Φ𝑢 )⟩) + 𝑣 ′2 ⟨Φ𝑣 , −𝐷𝑁 𝛼(𝑡) (Φ𝑣 )⟩
= 𝑒𝑢 ′2 + 2 𝑓 𝑢 ′ 𝑣 ′ + 𝑔𝑣 ′2 ,

where 𝑒, 𝑓 , 𝑔 are computed at (𝑢(𝑡), 𝑣(𝑡)); the normal curvature is then

𝑒𝑢 ′2 + 2 𝑓 𝑢 ′ 𝑣 ′ + 𝑔𝑣 ′2
𝑘 𝑛 (𝑡) =
∣𝛼′ (𝑡)∣2
(3.5)
𝑒𝑢 ′2 + 2 𝑓 𝑢 ′ 𝑣 ′ + 𝑔𝑣 ′2
= ⋅
𝐸𝑢 ′2 + 2𝐹𝑢 ′ 𝑣 ′ + 𝐺𝑣 ′2

𝑎 11 𝑎 12
Let us now determine the matrix [ ] of −𝐷𝑁Φ(𝑢,𝑣) relative to the basis
𝑎 21 𝑎 22
(Φ𝑢 , Φ𝑣 ): the entries of the matrix are determined by the equalities
3.2 The Second Fundamental Form 69

−𝑁𝑢 = 𝑎 11 Φ𝑢 + 𝑎 21 Φ𝑣
−𝑁 𝑣 = 𝑎 12 Φ𝑢 + 𝑎 22 Φ𝑣 .

Forming the inner product of each of these equalities with Φ𝑢 and with Φ𝑣 we obtain

𝑒 = 𝑎 11 𝐸 + 𝑎 21 𝐹 𝑓 = 𝑎 11 𝐹 + 𝑎 21 𝐺
𝑓 = 𝑎 12 𝐸 + 𝑎 22 𝐹 𝑔 = 𝑎 12 𝐹 + 𝑎 22 𝐺,

and these equalities can be written in matrix form as

𝑒 𝑓 𝐸 𝐹 𝑎 11 𝑎 12
[ ]=[ ][ ],
𝑓 𝑔 𝐹 𝐺 𝑎 21 𝑎 22

i.e.
−1
𝑎 11 𝑎 12 𝐸 𝐹 𝑒 𝑓
[ ]=[ ] [ ]
𝑎 21 𝑎 22 𝐹𝐺 𝑓 𝑔
(3.6)
1 𝐺 −𝐹 𝑒 𝑓
= [ ][ ]⋅
𝐸𝐺 − 𝐹 2 −𝐹 𝐸 𝑓 𝑔
From this we derive explicit formulas for the Gaussian curvature and mean curvature:

𝑒𝑔 − 𝑓 2
𝐾 ○ Φ(𝑢, 𝑣) = det ( −𝐷𝑁Φ(𝑢,𝑣) ) = (3.7)
𝐸𝐺 − 𝐹 2

1 1
𝐻 ○ Φ(𝑢, 𝑣) = tr ( −𝐷𝑁Φ(𝑢,𝑣) ) = (𝑎 11 + 𝑎 22 )
2 2
𝐺𝑒 − 2𝐹 𝑓 + 𝐸𝑔 (3.8)
= ;
2(𝐸𝐺 − 𝐹 2 )
and also for the principal curvatures 𝑘 1 and 𝑘 2 , which are the eigenvalues of the
matrix [𝑎 𝑖 𝑗 ]: √ √
𝑘 1 = 𝐻 − 𝐻 2 − 𝐾, 𝑘 2 = 𝐻 + 𝐻 2 − 𝐾. (3.9)

Example 3.2.1 The surface of revolution

Φ(𝑢, 𝑣) = (𝜌(𝑣) cos 𝑢, 𝜌(𝑣) sin 𝑢, 𝑧(𝑣))

has coefficients
−𝑧 𝜌 𝜌𝑧 − 𝜌𝑧
𝑒 =√ , 𝑓 = 0, 𝑔= √ ,
𝜌2 + 𝑧2 𝜌2 + 𝑧2
𝐸=𝜌 ,
2
𝐹 = 0, 𝐺 = 𝜌2 + 𝑧2 .

It then follows from (3.6) that the matrix [𝑎 𝑖 𝑗 ] is diagonal. The principal directions
at the non-umbilical points are thus those of Φ𝑢 and Φ𝑣 , the tangent lines to the
meridians and the parallels. The principal curvatures are
70 3 The Geometry of the Gauss Map
𝑒 −𝑧 𝑔 𝜌𝑧 − 𝜌𝑧
= √ and =
𝐸 𝜌 𝜌2 + 𝑧2 𝐺 (𝜌 2 + 𝑧2 )3/2

and the Gaussian curvature is

𝑧(𝜌𝑧 − 𝜌𝑧)
𝐾= 2
.
𝜌(𝜌 2 + 𝑧2 )

The expression of the curvature is simplified by assuming that the generating

curve 𝛼(𝑣) = (𝜌(𝑣), 0, 𝑧(𝑣)) is parametrized by arc length — that is, 𝜌 2 + 𝑧2 = 1.
Differentiating this equality we obtain 𝑧𝑧 = −𝜌𝜌, and replacing the left-hand side
with the right-hand side in the expression of 𝐾 yields

−𝜌(𝜌 2 + 𝑧2 ) −𝜌
𝐾= 2
= .
𝜌(𝜌 2 + 𝑧2 ) 𝜌

Using this formula we will now determine the surfaces of revolution of constant
curvature; to do this we simply solve the differential equation 𝜌 + 𝐾 𝜌 = 0, with 𝐾
constant. Leaving the cases 𝐾 = 0 and 𝐾 < 0 as an exercise, let us deal with the case
𝐾 > 0; to simplify the formulas we let 𝐾 = 1.

The general solution of the equation 𝜌 + 𝜌 = 0 is of the form 𝜌(𝑣) = 𝐶 cos(𝑣 + 𝐵),
where 𝐶 and 𝐵 are constants. We can choose 𝐵 = 0, since the solutions we obtain
with 𝐵 ≠ 0 correspond only to a translation of the domain; so we are reduced to
the solutions 𝜌𝐶 (𝑣) = 𝐶 cos 𝑣. Since we want 𝜌 > 0, we take 𝐶 > 0 and restrict 𝑣
to an interval of the form ] − 𝑣 0 , 𝑣 0 [. Putting 𝑧𝐶 (0) = 0 and integrating the equality
𝜌𝐶2
+ 𝑧𝐶
2
= 1, we obtain
𝑣 √
𝑧𝐶 (𝑣) = ∫ 1 − 𝐶 2 sin2 𝑡 𝑑𝑡
0

(the opposite solution gives another parameterization of the same surface).

3.2 The Second Fundamental Form 71

Figure 3.2

So necessarily 𝑣 0 = 𝜋2 when 0 < 𝐶 ≤ 1, and 𝑣 0 = arcsin(1/𝐶) ∈ ]0, 𝜋2 [ when 𝐶 > 1.

The resulting surfaces 𝑆𝐶 are symmetric with respect to the equator 𝑧 = 0, whose
radius is precisely 𝐶. When 𝐶 < 1, then 𝑆𝐶 has a cusp; when 𝐶 = 1 it is the sphere
with radius 1. When 𝐶 > 1 it is a kind of bead of a rosary (flattened sphere pierced
from one end to the other one). [See Figure 3.2]. ◻
We can now specify the meaning of the sign of the Gaussian curvature:
Proposition 3.2.2 Let 𝑝 0 be a point on the surface 𝑆. Then:
(i) if 𝐾(𝑝 0 ) > 0 then there exists a neighborhood 𝑈 of 𝑝 0 in 𝑆 which is entirely on
the same side of the plane tangent to 𝑆 at 𝑝 0 ;
(ii) if 𝐾(𝑝 0 ) < 0 then any neighborhood of 𝑝 0 in 𝑆 contains points on either side
of the plane tangent to 𝑆 at 𝑝 0 .
Proof It suffices to analyze the sign of the function ̃ ℎ∶ 𝑆 → R given by ̃ ℎ(𝑝) =
⟨𝑁(𝑝 0 ), 𝑝 − 𝑝 0 ⟩ at points 𝑝 near 𝑝 0 . Let us consider a parameterization Φ(𝑢, 𝑣) such
that Φ(0, 0) = 𝑝 0 , and let ℎ = ̃ℎ ○ Φ. The expansion of ℎ into a Taylor polynomial at
the point (0, 0) gives

𝜕ℎ 𝜕ℎ 1 𝜕2 ℎ 𝜕2 ℎ 𝜕2 ℎ
ℎ(𝑢, 𝑣) = ℎ(0, 0) + 𝑢+ 𝑣 + ( 2 𝑢 2 +2 𝑢𝑣 + 2 𝑣 2 ) + 𝑅(𝑢, 𝑣),
𝜕𝑢 𝜕𝑣 2 𝜕𝑢 𝜕𝑢𝜕𝑣 𝜕𝑣

where all partial derivatives are computed at (0, 0) and

𝑅(𝑢, 𝑣)
lim = 0.
(𝑢,𝑣)→(00) 𝑢2 + 𝑣 2

𝜕ℎ 𝜕ℎ
A quick calculation shows that ℎ(0, 0) = = = 0 and that the second-order
𝜕𝑢 𝜕𝑣
derivatives are nothing else but the coefficients at 𝑝 0 of the second fundamental form:
i.e.,
72 3 The Geometry of the Gauss Map

𝜕2 ℎ 𝜕2 ℎ 𝜕2 ℎ
= 𝑒, = 𝑓, = 𝑔.
𝜕𝑢 2 𝜕𝑢𝜕𝑣 𝜕𝑣 2
This leaves us with
1
ℎ(𝑢, 𝑣) = (𝑒𝑢 2 + 2 𝑓 𝑢𝑣 + 𝑔𝑣 2 ) + 𝑅(𝑢, 𝑣). (3.10)
2
Let us put Q(𝑢, 𝑣) = 𝑒𝑢 2 + 2 𝑓 𝑢𝑣 + 𝑔𝑣 2 . By (3.7), 𝐾(𝑝 0 ) and 𝑒𝑔 − 𝑓 2 have the same
sign. If 𝑒𝑔 − 𝑓 2 > 0 then 𝑒, 𝑔 are both positive or both negative: let us now show
that in this case there exists 𝛿 > 0 such that ℎ(𝑢, 𝑣) is nonzero and has constant sign
(positive if 𝑒, 𝑔 > 0, negative otherwise) whenever 0 < ∣(𝑢, 𝑣)∣ < 𝛿.
Assuming then that 𝑒𝑔 − 𝑓 2 > 0 and that 𝑒, 𝑔 are positive, we can write

𝑓 2 𝑒𝑔 − 𝑓 2 2 𝑒𝑔 − 𝑓 2 2
Q(𝑢, 𝑣) = 𝑒(𝑢 + 𝑣) + 𝑣 ≥ 𝑣 ,
𝑒 𝑒 𝑒
and analogously,
𝑒𝑔 − 𝑓 2 2
Q(𝑢, 𝑣) ≥ 𝑢 .
𝑔
From these inequalities we obtain Q(𝑢, 𝑣) ≥ 2𝑀(𝑢 2 + 𝑣 2 ), where

𝑒𝑔 − 𝑓 2 1 1
𝑀= min { , } > 0.
4 𝑔 𝑒

Finally we have, by (3.10), that

ℎ(𝑢, 𝑣) 𝑅(𝑢, 𝑣)
≥𝑀− 2 >0
𝑢2 + 𝑣 2 𝑢 + 𝑣2
whenever 0 < ∣(𝑢, 𝑣)∣ < 𝛿, provided that 𝛿 > 0 is chosen sufficiently small. The case
where 𝑒, 𝑔 are negative is treated analogously, and the first statement is thus proved.
If 𝑒𝑔 − 𝑓 2 < 0 then we can write Q(𝑢, 𝑣) as a product 𝔏1 (𝑢, 𝑣)𝔏2 (𝑢, 𝑣) of linearly
independent linear forms 𝔏𝑖 ∶ R 2 → R . The two lines 𝔏𝑖 (𝑢, 𝑣) = 0 divide the plane
into four sectors: in two of these sectors 𝔏1 and 𝔏2 have equal signs, and in the other
two they have opposite signs. There are therefore unit vectors (𝑢 0 , 𝑣 0 ), (𝑢 1 , 𝑣 1 ) ∈ R 2
such that 𝑎 0 = Q(𝑢 0 , 𝑣 0 ) < 0 < Q(𝑢 1 , 𝑣 1 ) = 𝑎 1 . Now observe that

ℎ(𝑡𝑢 𝑖 , 𝑡𝑣 𝑖 ) Q(𝑡𝑢 𝑖 , 𝑡𝑣 𝑖 ) 𝑅(𝑡𝑢 𝑖 , 𝑡𝑣 𝑖 )

lim = lim { + }
𝑡→0 𝑡 2 𝑡→0 2𝑡 2 ∣(𝑡𝑢 𝑖 , 𝑡𝑣 𝑖 )∣2
𝑎𝑖 𝑅(𝑡𝑢 𝑖 , 𝑡𝑣 𝑖 ) 𝑎 𝑖
= + lim 𝑡 → 0 =
2 ∣(𝑡𝑢 𝑖 , 𝑡𝑣 𝑖 )∣2 2

for 𝑖 = 0.1. This shows that ℎ(𝑢, 𝑣) takes positive and negative values for (𝑢, 𝑣)
arbitrarily close to the origin and proves (ii). ◻
Let us now study the contact between two surfaces. Assume that 𝑆1 and 𝑆2 are
tangent at 𝑝 0 , and denote 𝜋∶ R 2 → 𝑝 0 + 𝑇 𝑝0 𝑆1 the orthogonal projection on the
3.2 The Second Fundamental Form 73

tangent plane to 𝑆 at 𝑝 0 : thus, the vector 𝜋(𝑝) − 𝑝 0 belongs to 𝑇 𝑝0 𝑆1 , and 𝜋(𝑝) − 𝑝

is orthogonal to the same plane. Note that the derivative at 𝑝 0 of the restriction of 𝜋 to
each of the surfaces is the identity. Hence, by 2.3.5, there exist open neighborhoods 𝑈1
and 𝑈2 of 𝑝 0 in 𝑆1 and 𝑆2 , respectively, such that 𝜋∣𝑈 and 𝜋∣𝑈 are diffeomorphisms
2 2
onto their image. Let us fix 𝛿 > 0 such that

{𝑝 ∈ 𝑝 0 + 𝑇 𝑝0 𝑆1 ∶ ∣𝑝 − 𝑝 0 ∣ < 𝛿} ⊆ 𝜋(𝑈1 ) ∩ 𝜋(𝑈2 );

and, given an orthonormal basis (w1 , w2 ) of 𝑇 𝑝0 𝑆1 and putting 𝑁0 = w1 × w2 , define

−1
Φ(𝑢, 𝑣) = 𝜋∣𝑈 (𝑝 0 + 𝑢w1 + 𝑣w2 )
1

= 𝑝 0 + 𝑢w1 + 𝑣w2 + ℎ1 (𝑢, 𝑣)𝑁0 ,

−1
Ψ(𝑢, 𝑣) = 𝜋∣𝑈 (𝑝 0 + 𝑢w1 + 𝑣w2 )
2

= 𝑝 0 + 𝑢w1 + 𝑣w2 + ℎ2 (𝑢, 𝑣)𝑁0 ,

whenever 𝑢 2 + 𝑣 2 < 𝛿2 . These parameterizations correspond to regarding 𝑆1 and 𝑆2

locally as the graphs of the functions ℎ1 and ℎ2 defined at 𝑝 0 + 𝑇 𝑝0 𝑆1 .
We say that 𝑆1 and 𝑆2 have contact order ≥ 2 at 𝑝 0 if

ℎ1 (𝑢, 𝑣) − ℎ2 (𝑢, 𝑣)
lim = 0.
(𝑢,𝑣)→(0,0) 𝑢2 + 𝑣 2

Expanding the functions ℎ𝑖 into Taylor polynomials, we see that this happens if and
only ℎ1 and ℎ2 and their partial derivatives up to second order are equal at (0, 0);
but, since we have ℎ𝑖 (0, 0) = 0 and

𝜕ℎ𝑖 𝜕ℎ𝑖
∣ = ∣ =0
𝜕𝑢 (0,0) 𝜕𝑣 (0,0)
for 𝑖 = 1, 2, this condition boils down to

𝜕 2 ℎ1 𝜕 2 ℎ2
∣(0,0) = ∣ ,
𝜕𝑢 2 𝜕𝑢 2 (0,0)
𝜕 2 ℎ1 𝜕 2 ℎ2
∣(0,0) = ∣ , (3.11)
𝜕𝑢𝜕𝑣 𝜕𝑢𝜕𝑣 (0,0)
𝜕 2 ℎ1 𝜕 2 ℎ2
∣ = ∣ ⋅
𝜕𝑣 2 (0,0) 𝜕𝑣 2 0,0)
Now if 𝑆1 and 𝑆2 are oriented such that their normal vectors both coincide with 𝑁0
at the point 𝑝 0 , then the second-order derivatives of ℎ1 and ℎ2 are the coefficients
of the second fundamental form at 𝑝 0 of the parameterizations Φ(𝑢, 𝑣) and Ψ(𝑢, 𝑣).
Since Φ𝑢 ∣ 𝑝 = Ψ𝑢 ∣ 𝑝 = w1 and Φ𝑣 ∣ 𝑝 = Ψ𝑣 ∣ 𝑝 = w2 , the equality of the coefficients
0 0 0 0
implies that the second fundamental forms of the two surfaces coincide at the point
𝑝 0 . We thus proved half of the following proposition.
74 3 The Geometry of the Gauss Map

Proposition 3.2.3 Let 𝑆1 and 𝑆2 be two oriented surfaces tangent at 𝑝 0 whose normal
vectors coincide at that point. Then the following two conditions are equivalent:
(i) 𝑆1 and 𝑆2 have contact of order ≥ 2 at 𝑝 0 ;
(ii) the restrictions on 𝑇 𝑝0 𝑆1 = 𝑇 𝑝0 𝑆2 of the second fundamental forms of 𝑆1 and
𝑆2 coincide.

It is left to the reader to prove that (ii) ⇒ (i).

Given a point 𝑝 0 on a surface 𝑆1 , the equalities 3.1.1, plus the fact that the function
and its first derivatives vanish at (0, 0), completely determine a polynomial ℎ2 (𝑢, 𝑣)
1
of degree at most two, given by ℎ2 (𝑢, 𝑣) = (𝑒𝑢 2 + 2 𝑓 𝑢𝑣 + 𝑔𝑣 2 ) — where 𝑒, 𝑓 , 𝑔
2
are the coefficients of the second fundamental form of Φ(𝑢, 𝑣) at 𝑝 0 . This means
that there is exactly one paraboloid which has contact of order ≥ 2 with 𝑆1 on 𝑝 0 ; it
is called the osculating paraboloid. It is an elliptic paraboloid when 𝑝 0 is an elliptic
point, hyperbolic when 𝑝 0 is hyperbolic; it is a plane when 𝑝 0 is planar, and when 𝑝 0
is parabolic it is a straight paraboloid (a figure generated by a line perpendicular to a
plane when its base point runs through a parabola).
We end this section by considering two special types of curves on surfaces; the
question of the existence of such curves will be dealt with in the next section.
We say that a regular curve 𝛼(𝑡) in 𝑆 is a line of curvature if, for every 𝑡, the
vector 𝛼′ (𝑡) defines one of the principal directions in 𝛼(𝑡) — that is, if 𝛼′ (𝑡) is
an eigenvector of 𝐷𝑁 𝛼(𝑡) . Using this formulation, we recognize that 𝛼 is a line of
curvature if and only if there exists a function 𝜆(𝑡) such that (𝑁 ○ 𝛼)′ (𝑡) = 𝜆(𝑡)𝛼′ (𝑡)
and, if such a function exists, it is differentiable because it is defined by

⟨(𝑁 ○ 𝛼)′ (𝑡), 𝛼′ (𝑡)⟩

𝜆(𝑡) = ⋅
∣𝛼′ (𝑡)∣2

An asymptotic line is a regular curve whose velocity vector defines at each point
an asymptotic direction — that is, it is a curve whose normal curvature is constant
and equal to zero. It follows from the argument in example 3.1.4 B that at all points
on an asymptotic line the Gaussian curvature is nonpositive. In local coordinates,
formula (3.5) implies that a regular curve Ψ(𝑢(𝑡), 𝑣(𝑡)) is asymptotic if and only if
𝑒(𝑢 ′ ) + 2 𝑓 𝑢 ′ 𝑣 ′ + 𝑔(𝑣 ′ ) ≡ 0.
2 2

Exercises
60. Define orientations for 𝑆1 = {(𝑥, 𝑦, 𝑧)∶ 𝑧 = 𝑥 2 − 𝑦 2 } and 𝑆2 = {(𝑥, 𝑦, 𝑧)∶ 𝑧 =
𝑥 3 − 3𝑥𝑦 2 }, and then determine for each of these surfaces:
(a) the Gaussian curvature and the mean curvature at each point;
(b) the points where the mean curvature vanishes.
(c) Do any of the answers to the above questions depend on the chosen orientations?
61. (a) Show that the Gaussian curvature of the Möbius strip M (with the parameteri-
zation given in 2.4.2) is given by
3.2 The Second Fundamental Form 75
−2
1 𝜃 2
𝐾(𝜃, 𝑡) = −( 𝑡 2 + (2 − 𝑡 sin ) ) .
4 2

(b) Show that if 𝑆 is a surface of strictly positive curvature at all its points, then
𝑆 is orientable. Show further that it is possible to choose this orientation such that
all principal curvatures are positive. Hint: For every point 𝑝 0 of 𝑆, let 𝑈 be the
neighborhood given by 3.2.2 (i) and 𝑁(𝑝 0 ) the normal vector pointing to the side
where 𝑈 lies; then 𝑁∶ 𝑆 → S2 is an orientation of 𝑆.

62. Show that the pseudosphere (Exercise 32) has constant negative curvature.

63. (a) Show that the only surfaces of revolution with zero constant curvature are the
cylinder, cone and plane.
(b) Show that any surface of revolution of constant curvature 𝐾 = −1 is, up to
reparametrization, of the form Φ(𝑢, 𝑣) = (𝜌(𝑣) cos 𝑢, 𝜌(𝑣) sin 𝑢, 𝑧(𝑣)), where the
generating curve 𝑣 ↦ (𝜌(𝑣), 0, 𝑧(𝑣)) is of one of the following three types:
√
(i) 𝜌(𝑣) = 𝐴 cosh 𝑣, 𝑧(𝑣) = ∫0 1 − 𝐴2 sinh2 𝑡 𝑑𝑡, 𝐴 > 0;
𝑣

𝑣√
(ii) 𝜌(𝑣) = 𝐴𝑒 −𝑣 , 𝑧(𝑣) = ∫0 1 − 𝐴2 𝑒 −2𝑡 𝑑𝑡, 𝐴 > 0;
√
(iii) 𝜌(𝑣) = 𝐴 sinh 𝑣, 𝑧(𝑣) = ∫0 1 − 𝐴2 cosh2 𝑡 𝑑𝑡, 0 < 𝐴 < 1;
𝑣

In each of the cases find the domain of 𝑣 and sketch the generating curve.
(c) Show that the surface of type (ii) is the pseudosphere.

64. Let 𝑝 0 ∈ 𝑆 be such that 𝐾(𝑝 0 ) ≠ 0, and let (𝑈, Φ) be a parametrized neighborhood
of 𝑝 0 where 𝐾 has constant sign. Show that:
(a) 𝑁𝑢 × 𝑁 𝑣 = 𝐾(𝑢, 𝑣)(Φ𝑢 × Φ𝑣 );
(b) if 𝑉 ⊆ Φ(𝑈), then the area of 𝑁(𝑉) ⊆ S2 is given by ∫ ∣𝐾∣ 𝑑𝜎;
𝑉
area of𝑁 (𝑉)
(c) ∣𝐾(𝑝 0 )∣ is the limit of the ratio area of𝑉 as the diameter of its neighborhood
𝑉 tends to zero.

65. Complete the proof of Proposition 3.2.3.

66. Let 𝑝 0 be a point shared by the surfaces 𝑆1 and 𝑆2 . Show that the following
assertions are equivalent:
(i) 𝑆1 and 𝑆2 have contact of order ≥ 2 at 𝑝 0 .
(ii) There exist parameterizations Φ(𝑢, 𝑣) and Ψ(𝑢, 𝑣) of 𝑆1 and 𝑆2 such that
𝑝 0 = Φ(0, 0) = 𝜓(0, 0) and

∣Φ(𝑢, 𝑣) − Ψ(𝑢, 𝑣)∣

lim = 0.
(𝑢,𝑣)→(0,0) 𝑢2 + 𝑣 2

67. A regular surface 𝑆 is a ruled surface if 𝑆 = {𝛼(𝑡) + 𝜆v(𝑡)∶ 𝑡 ∈ 𝐽, 𝜆 ∈ R }, where 𝐽

is an interval, 𝛼, v∶ 𝐽 → R 3 are 𝐶 ∞ , and 𝛼′ (𝑡), v(𝑡) are linearly independent vectors
for all 𝑡 ∈ 𝐽. Note that then 𝑆 = ⋃ 𝑟 𝑡 , where 𝑟 𝑡 is the line {𝛼(𝑡) + 𝜆v(𝑡)∶ 𝜆 ∈ R }.
𝑡∈𝐽
76 3 The Geometry of the Gauss Map

(a) Show that the hyperbolic paraboloid 𝑧 = 𝑥 2 −𝑦 2 and the hyperboloid 𝑥 2 +𝑦 2 −𝑧2 =
1 are ruled surfaces, and find out whether the equation 𝑒 𝑥 = 𝑧 + 𝑦 2 defines a ruled
surface;
(b) Show that the tangent planes to the ruled surface 𝑆 intersect 𝑆 along a line,
which is an asymptotic line; conclude that the curvature of 𝑆 is ≤ 0 at all points.
68. With the notation of Exercise 67, let 𝑆 be a ruled surface without planar points.
Show that the following conditions are equivalent:
i. 𝑆 has curvature 0 at all its points.
ii. The lines 𝜆 ↦ 𝛼(𝑡) + 𝜆v(𝑡) are lines of curvature of 𝑆.
iii. For every 𝑡 ∈ 𝐼, the vector v′ (𝑡) is a linear combination of v(𝑡) and 𝛼′ (𝑡).
69. Given a point 𝑝 on a compact surface 𝑆 ⊆ R 3 , we define the diameter of 𝑆 at 𝑝 by
𝑑(𝑝) = max {∣𝑞 − 𝑝∣∶ 𝑞 ∈ 𝑆}. Assume that 𝑆 has constant diameter 𝑑 — that is, that
𝑑(𝑝) = 𝑑 for every point 𝑝 of 𝑆. Show that:
(a) for every 𝑝 in 𝑆 there exists a single point 𝑓 (𝑝) in 𝑆 such that ∣ 𝑓 (𝑝) − 𝑝∣ = 𝑑;
(b) 𝑓 is an involutive diffeomorphism of 𝑆 (i.e., 𝑓 ○ 𝑓 = id);
(c) 𝑆 has strictly positive curvature, and therefore [Exercise 61-b)] admits an
orientation for which all principal curvatures are positive;
(d) if 𝑘 2 (𝑝) ≥ 𝑘 1 (𝑝) denote the principal curvatures of 𝑆 then

1 1
+ = 𝑑.
𝑘 1 (𝑝) 𝑘 2 ( 𝑓 (𝑝))

3.3 Vector Fields

A vector field assigns to each point in a surface a vector in the tangent space of the
surface to that point. If this assignment is made in a sufficiently regular way, the
vector field can be interpreted as a velocity field, and so it determines certain curves
(trajectories) on the surface. With this approach we can establish in this section the
existence of curves and parameterizations satisfying certain requirements (such as
asymptotic lines, lines of curvature, and orthogonal parameterizations).
A vector field of class 𝐶 𝑘 (𝑘 ≥ 1) in an open subset 𝑉 ⊆ R 𝑛 is a mapping
v∶ 𝑉 → R 𝑛 of class 𝐶 𝑘 , and a trajectory (or integral curve) of v is a curve 𝜑∶ 𝐼 → 𝑉
such that 𝜑′ (𝑡) = v(𝜑(𝑡)). In other words, a trajectory is a curve whose velocity
at each point is the vector v assigned to that point. The fundamental theorem of
differential equations, which we now state, asserts the existence and uniqueness of the
trajectory passing through each 𝑝 ∈ 𝑉 at a given instant; for its proof we suggest [23].

Theorem 3.3.1 Given a vector field v∶ 𝑉 ⊆ R 𝑛 → R 𝑛 of class 𝐶 𝑘 (𝑘 ≥ 1) and

𝑝 0 ∈ 𝑉, there exist 𝜀 > 0, an open neighborhood 𝑈 ⊆ 𝑉 of 𝑝 0 , and a 𝐶 𝑘 mapping
𝜑∶ ]− 𝜀, 𝜀[×𝑈 → 𝑉, such that, for every 𝑝 ∈ 𝑈, the curve 𝑡 ↦ 𝜑(𝑡, 𝑝) is the only
trajectory of v with initial condition 𝜑(0, 𝑝) = 𝑝 (in the sense that any other trajectory
with the same initial condition coincides with this one in the intersection of their
domains).
3.3 Vector Fields 77

At a point 𝑝 where the vector field v vanishes, one trajectory of v that passes
through 𝑝 is the constant curve 𝜑(𝑡) = 𝑝; the theorem guarantees that this is the only
trajectory that passes through 𝑝. This means that the singularities of the vector field

Figure 3.3

v correspond to the constant trajectories, and the trajectories of the other points are
regular curves: for if the derivative of 𝜑(𝑡) vanishes at 𝑡 0 , then 𝜑(𝑡 0 ) is a singularity
of v and the only trajectory through 𝜑(𝑡 0 ) is the constant one: so a non-constant
trajectory cannot have points of zero velocity.
The mapping 𝜑(𝑡, 𝑝) in 3.3.1 is called the local flow of the vector field v. A
frequent way to write the differential equation associated with a given field is in the
form
𝑋 = v(𝑋)
where 𝑋 = (𝑥 1 , . . . , 𝑥 𝑛 ) is a point of 𝑉. A differential equation of this type is called
autonomous. A non-autonomous equation is one where the right-hand side depends
explicitly on the independent variable 𝑡, i.e. of the form

𝑋 = v(𝑡, 𝑋), (3.12)

where v∶ 𝐼 × 𝑉 ⊆ R × R 𝑛 → R 𝑛 is of class 𝐶 𝑘 . A simple example is given by the

equation 𝑋 = 𝑡 𝑋: a curve 𝛼(𝑡) in R 𝑛 is a solution of this equation if and only if
𝛼′ (𝑡) = 𝑡𝛼(𝑡).
For non-autonomous equations there is also a theorem of existence, uniqueness and
differentiable dependence on initial conditions. In this instance, given (𝑡 0 , 𝑝 0 ) ∈ 𝐼 ×𝑉,
there exists a mapping 𝜑∶ ]𝑡0 − 𝜀, 𝑡 0 + 𝜀[×𝑈 → 𝑉 of class 𝐶 𝑘 (where 𝑈 is an open
78 3 The Geometry of the Gauss Map

neighborhood of 𝑝 0 ) such that, for every 𝑝 ∈ 𝑈, the function 𝑡 ↦ 𝜑(𝑡, 𝑝) is the only
solution of 3.1.2 with 𝜑(𝑡 0 , 𝑝) = 𝑝 defined in the interval ]𝑡 0 − 𝜀, 𝑡 0 + 𝜀[.
This result for non-autonomous equations, although seemingly more general
than 3.3.1, can in fact be easily deduced from it. To this end we consider the
autonomous equation
𝑌 = w(𝑌 ), (3.13)
where we put 𝑌 = (𝑡, 𝑋) ∈ 𝐼 ×𝑉, and where the vector field w∶ 𝐼 ×𝑉 ⊆ R 𝑛+1 → R 𝑛+1 is
given by w(𝑡, 𝑋) = (1, v(𝑡, 𝑋)). If 𝜓(𝑠) is the solution of (3.13) with 𝜓(0) = (𝑡0 , 𝑝),
then we have 𝜓(𝑠) = (𝑡 0 + 𝑠, 𝜑̃(𝑠)), and therefore 𝑡 = 𝑡0 + 𝑠. Putting 𝜑(𝑡) = 𝜑̃(𝑡 − 𝑡0 ),
we recognize that 𝜑(𝑡) is the solution of (3.12) with initial condition 𝜑(𝑡0 ) = 𝑝
if and only if 𝜓(𝑠) = (𝑡0 + 𝑠, 𝜑̃(𝑠)) is a solution of (3.13) with initial condition
𝜓(0) = (𝑡 0 , 𝑝).
Returning to surfaces, a vector field in an open subset 𝑉 ⊆ 𝑆 is a mapping
v such that v(𝑝) ∈ 𝑇 𝑝 𝑆 for every 𝑝 ∈ 𝑉. Using local coordinates, we can write
v ○ Φ(𝑢, 𝑣) = 𝛼(𝑢, 𝑣)Φ𝑢 + 𝛽(𝑢, 𝑣)Φ𝑣 for certain functions 𝛼 and 𝛽. We say that v is
of class 𝐶 𝑘 if every point of 𝑉 has a parametrized neighborhood Φ(𝑢, 𝑣) in which
the coordinate functions 𝛼 and 𝛽 of v are of class 𝐶 𝑘 (of course then the coordinate
functions of v in any parameterization are of class 𝐶 𝑘 ). Unless otherwise stated, our
vector fields are 𝐶 ∞ .
The equation 𝑋 = v(𝑋), when 𝑋 is contained in the image of Φ(𝑢, 𝑣), is equivalent
to the equation (𝑢, 𝑣) = (𝛼(𝑢, 𝑣), 𝛽(𝑢, 𝑣)) defined in an open subset of R 2 : in fact,
𝜑(𝑡) is a trajectory of the latter equation if and only if Φ ○ 𝜑(𝑡) is a trajectory of
the former. Theorem 3.3.1 is then carried over without changes to surfaces, since it
concerns local properties of trajectories.

Example 3.3.2 As an example of the application of these ideas, let us verify that
through each non-umbilical point 𝑝 pass exactly two lines of curvature, corresponding
to the two principal directions at 𝑝: to prove this, it suffices to show that there are two
differentiable vector fields w1 and w2 , defined in a neighborhood of 𝑝, such that, for
every 𝑞, the vectors w1 (𝑞) and w2 (𝑞) define the two principal directions in 𝑞: the
two lines of curvature through 𝑝 are then the integral curves of w1 and w2 that pass
through 𝑝. (The precise choice of the w𝑖 does not matter, since any other suitable
vector fields would have the same trajectories up to reparametrization.)
Let us consider a parameterization Φ(𝑢, 𝑣) in a neighborhood of the non-umbilical
point 𝑝 = Φ(0, 0). By the formulas (3.6)–(3.9), §2, the entries 𝑎 𝑖 𝑗 of the matrix of
−𝐷𝑁Φ(𝑢,𝑣) relative to the basis (Φ𝑢 , Φ𝑣 ) are differentiable functions of (𝑢, 𝑣), and
so are the principal curvatures 𝑘 1 < 𝑘 2 in a neighborhood of (0, 0). The eigenvectors
of −𝐷𝑁Φ(𝑢,𝑣) are 𝛼𝑖 Φ𝑢 + 𝛽𝑖 Φ𝑣 , where

𝑎 11 − 𝑘 𝑖 𝑎 12 𝛼 0
[ ][ 𝑖 ] = [ ] , 𝑖 = 1, 2. (3.14)
𝑎 21 𝑎 22 − 𝑘 𝑖 𝛽𝑖 0

If 𝑎 12 ∣(0,0) ≠ 0, then nonzero solutions of (3.14) in a neighborhood of (0, 0)

are given by (𝛼𝑖 , 𝛽𝑖 ) = (−𝑎 12 , 𝑎 11 − 𝑘 𝑖 ) for 𝑖 = 1, 2. If 𝑎 21 ∣(0,0) ≠ 0, then we
3.3 Vector Fields 79

take (𝛼𝑖 , 𝛽𝑖 ) = (𝑎 22 − 𝑘 𝑖 , −𝑎 21 ). If 𝑎 12 ∣(0,0) = 𝑎 21 ∣(0,0) = 0, then the 𝑘 𝑖 ∣ 𝑝 will

be precisely 𝑎 11 ∣(0,0) and 𝑎 22 ∣(0,0) , and therefore 𝑎 11 ∣(0,0) ≠ 𝑎 22 ∣(0,0) ; assuming
𝑘 1 ∣ 𝑝 = 𝑎 11 ∣(0,0) and 𝑘 2 ∣ 𝑝 = 𝑎 22 ∣(0,0) , we see that (𝛼1 , 𝛽1 ) = (𝑎 22 − 𝑘 1 , −𝑎 21 ) and
(𝛼2 , 𝛽2 ) = (−𝑎 12 , 𝑎 11 − 𝑘 2 ) are nonzero solutions of (3.14) in some neighborhood of
(0, 0).
In any case, we find non-trivial solutions of (3.14) that depend differentiably
on (𝑢, 𝑣), and the promised vector fields are given by w𝑖 = 𝛼𝑖 Φ𝑢 + 𝛽𝑖 Φ𝑣 , 𝑖 = 1, 2.
The local character of this construction is unavoidable, because there may not exist
differentiable vector fields, defined at all non-umbilical points in the surface, which
always point to a principal direction. ◻

What happens in the above example is that the principal directions do not define
vector fields, but direction fields: a direction field is a mapping D, defined in an open
subset 𝑈 ⊆ 𝑆, such that, for every 𝑝 ∈ 𝑈, D(𝑝) is a one-dimensional linear subspace
of 𝑇 𝑝 𝑆. The direction field D is called differentiable if every point of 𝑈 has an open
neighborhood 𝑉 in which a differentiable vector field v is defined such that v(𝑝)
generates D(𝑝) for all 𝑝 ∈ 𝑉. All direction fields we consider are differentiable. An
integral curve of the direction field is a curve whose velocity at each point has the
same direction as the field at that point.
Locally, the study of direction fields amounts to that of vector fields, and so it
readily follows that the integral curve of a direction field through a given point exists
and is unique (up to reparametrization). But, by example of the “hay fork” in Fig. 3.4,
we see that there are direction fields which globally do not arise from a vector field: it
is impossible to choose a coherent (continuous) orientation for all the trajectories
shown in the picture.

Figure 3.4

In local coordinates the direction fields are given by a linear relation 𝐴(𝑢, 𝑣)𝛼 +
𝐵(𝑢, 𝑣)𝛽 = 0 between the coordinates 𝛼 and 𝛽 of a vector in the basis (Φ𝑢 , Φ𝑣 ), so
that the integral curves have the form Φ(𝑢(𝑡), 𝑣(𝑡)), where 𝐴(𝑢, 𝑣)𝑢 ′ + 𝐵(𝑢, 𝑣)𝑣 ′ ≡ 0.
For example, the asymptotic directions at a point of negative Gaussian curvature
satisfy the equation 𝑒𝛼2 + 2 𝑓 𝛼𝛽 + 𝑔𝛽2 = 0. Since 𝑒𝑔 − 𝑓 2 < 0, the left-hand side of
80 3 The Geometry of the Gauss Map

the equation is the product of two linear factors: so if 𝑒 ≠ 0 then 𝑒𝛼2 + 2 𝑓 𝛼𝛽 + 𝑔𝛽2 =
𝑒(𝛼 − 𝜆1 𝛽)(𝛼 − 𝜆2 𝛽) with 𝜆1 ≠ 𝜆2 . Hence, the two asymptotic directions at each
point are given by the relations 𝛼 − 𝜆1 𝛽 = 0, 𝛼 − 𝜆2 𝛽 = 0, which define two direction
fields. The two families of asymptotic lines are the line integrals of these vector fields,
having equations 𝑢 ′ − 𝜆1 𝑣 ′ = 0, 𝑢 ′ − 𝜆2 𝑣 ′ = 0.

Example 3.3.3 The hyperbolic paraboloid parametrized by Φ(𝑢, 𝑣) = (𝑢, 𝑣, 𝑢 2 − 𝑣 2 )

has coefficients
2 −2
𝑒= √ , 𝑓 = 0, 𝑔= √ ,
1 + 4𝑢 2 + 4𝑣 2 1 + 4𝑢 2 + 4𝑣 2
so that the equation of the asymptotic lines is
2
√ (𝑢 ′2 − 𝑣 ′2 ) = 0
1 + 4𝑢 2 + 4𝑣 2
which is equivalent to
(𝑢 ′ − 𝑣 ′ )(𝑢 ′ + 𝑣 ′ ) = 0.
Thus, the two families of asymptotic lines are 𝑢 − 𝑣 = 𝑐 and 𝑢 + 𝑣 = 𝑐 for 𝑐 ∈ R , and
they can be parametrized by 𝑡 ↦ (𝑡 + 𝑐, 𝑡, 2𝑐𝑡 + 𝑐2 ) and 𝑡 ↦ (−𝑡 + 𝑐, 𝑡, −2𝑐𝑡 + 𝑡 2 ). ◻

We say that two direction fields D1 and D2 , defined in the same open subset
𝑈 ⊆ 𝑆, are independent if, for every 𝑝 ∈ 𝑈, the linear subspaces D1 (𝑝) and D2 (𝑝) of
𝑇 𝑝 𝑆 are distinct. Our next result, the most important one in this section, establishes
the existence of parameterizations in which tangent lines to coordinate curves have
directions fixed beforehand.

Theorem 3.3.4 Let D1 and D2 be two independent direction fields in 𝑈 ⊆ 𝑆. Then

each point of 𝑈 is contained in a parametrized neighborhood whose coordinate
curves are the integral curves of D1 and D2 .

Proof Given 𝑝 0 ∈ 𝑈, consider a parameterization Φ(𝑢, 𝑣) with Φ(0, 0) = 𝑝 0 . One of

the vectors Φ𝑢 ∣(0,0) and Φ𝑣 ∣(0,0) is not in D2 (𝑝 0 ), and we assume that it is the first
one. We can then find, in a neighborhood of 𝑝 0 , a differentiable vector field of the
form w = 𝛼Φ𝑢 + Φ𝑣 that generates D2 . We seek a parameterization of the form

Ψ(𝑥, 𝑡) = Φ(𝜑(𝑡, 𝑥), 𝑡)

and such that the coordinate curves 𝑥 = constant are line integrals of D2 ; the curves
𝑡 = constant of such a parameterization coincide with the curves 𝑣 = constant of
Φ(𝑢, 𝑣). We then have

𝜕𝜑
Ψ𝑥 = Φ𝑢 ,
𝜕𝑥
𝜕𝜑
Ψ𝑡 = Φ𝑢 + Φ 𝑣 ,
𝜕𝑡
3.3 Vector Fields 81

We thus see that, in order to be Ψ𝑡 = w, 𝜑 has to satisfy the equation

𝜕𝜑
= 𝛼(𝜑(𝑡, 𝑥), 𝑡). (3.15)
𝜕𝑡
Consider now the differential equation

𝑦 = 𝛼(𝑦, 𝑡). (3.16)

By the version for non-autonomous equations of Theorem 3.3.1, there exists a

differentiable mapping 𝜑∶ ]−𝜀, 𝜀[ × 𝐽 → R (where 𝐽 ⊆ R is an open interval containing
0) such that for 𝑥 ∈ 𝐽 the curve 𝑡 ↦ 𝜑(𝑡, 𝑥) is the solution of (3.16) with initial
condition 𝜑(0, 𝑥) = 𝑥. Of course, such a function 𝜑 satisfies (3.15); furthermore, we
𝜕𝜑 𝜕𝜑
have ∣(0,0) = ∣ = 1, so that Ψ𝑥 ∣(0,0) and Ψ𝑡 ∣(0,0) are independent. Hence,
𝜕𝑥 𝜕𝑥 (0, 𝑥)
the Jacobian of Ψ(𝑥, 𝑡) at (0, 0) has rank two and Ψ is a parameterization in a
neighborhood of 𝑝 0 .
In conclusion: keeping one of the families of coordinate curves, we modified the
other one so that it now corresponds to the integral curves of the D2 field; applying
this method again (fixing now the coordinate curves corresponding to D2 ), we obtain
a parameterization whose coordinate curves are the integral curves of D1 and of D2 .
◻

The next two results are an immediate consequence of 3.3.4 and the preceding
discussion.

Corollary 3.3.5 Every point in a surface has a neighborhood covered by an orthogo-

nal parameterization.

Corollary 3.3.6 In the neighborhood of any non-umbilical point (hyperbolic point)

there exists a parameterization whose coordinate curves are lines of curvature
(asymptotic lines).

Exercises
70. Show that in a neighborhood of a non-umbilical point (hyperbolic point) the
coordinate curves of Φ(𝑢, 𝑣) are lines of curvature (asymptotic lines) if and only if
the coefficients of this parameterization satisfy the condition 𝐹 ≡ 0 ≡ 𝑓 (𝑒 ≡ 0 ≡ 𝑔).
71. Consider T 2 parametrized by

Φ(𝑢, 𝑣) = ((2 + cos 𝑣) cos 𝑢, (2 + cos 𝑣) sin 𝑢, sin 𝑣).

cos 𝑣
(a) Show that the curvature of T 2 in Φ(𝑢, 𝑣) is 2+cos 𝑣 . What are the regions of T
2

with positive, zero and negative curvature?

(b) Show that the curves 𝑣 = 𝜋2 and 𝑣 = 32𝜋 are asymptotic lines, and that the other
asymptotic lines admit the parameterization 𝛼(𝑡) = Φ(𝑢(𝑡), 𝑣(𝑡)), where
82 3 The Geometry of the Gauss Map

⎛ 𝑡 𝑑𝑥 ⎞ 𝜋 𝜋
(𝑢(𝑡), 𝑣(𝑡)) = 𝑢 0 + ∫ √ , 𝜋 + 𝑡 , 𝑡 ∈ ]− , [
⎝ 0 cos 𝑥(2 − cos 𝑥) ⎠ 2 2

or
⎛ 𝑡 𝑑𝑥 ⎞ 𝜋 𝜋
(𝑢(𝑡), 𝑣(𝑡)) = 𝑢 0 − ∫ √ , 𝜋 + 𝑡 , 𝑡 ∈ ]− , [ ⋅
⎝ 0 cos 𝑥(2 − cos 𝑥) ⎠ 2 2
(c) Show that lim𝜋 𝛼(𝑡) and lim𝜋 𝛼(𝑡) exist and are points of the parallels 𝑣 = 𝜋
2
𝑡→− 2 𝑡→ 2
and 𝑣 = 3𝜋
2 . Deduce that all asymptotic lines of T 2 have finite length.
72. Let 𝑆 be a surface of constant zero Gaussian curvature and let 𝑝 be a parabolic
point of 𝑆. Show that there exists a line segment through 𝑝 which is entirely contained
in 𝑆.
Chapter 4
The Intrinsic Geometry of Surfaces

By intrinsic geometry of a surface we mean those properties that depend exclusively

on measurements made on the surface but not on how the surface is embedded in the
ambient space. Hence, the length of a curve on a sheet of paper is an intrinsic property,
because it doesn´t change when we fold the sheet; but the distance in space between
two points on that same sheet is not intrinsic. In this chapter we study the intrinsic
properties of surfaces and the mappings that preserve such properties (isometries).
All sections of this chapter are basic and should be studied consecutively, except
for 4.4, which can be omitted because it is only needed for sections 5.4 and 5.5; the
Rotation Index Theorem, proved in the Appendix, is used in Section 4.5. (Sections 4.5
and 4.6 are independent of each other and, from a logical point of view, can be studied
in any order; but the Gauss-Bonnet theorem should be included as early as possible.)

4.1 Conformal Mappings and Isometries

It is impossible to draw an entirely accurate map of the terrestrial globe on a plane:

all known maps distort the relative size of regions — making those further away from
the equator appear larger than they are in reality — and distort the shape of continents.
Still, a map is an approximate depiction of the real world, the more accurate as the
region so represented becomes smaller. In the terminology we now introduce, this
means that there exist conformal mappings of spherical regions into planes, but there
are no such mappings which are isometries.
A mapping 𝑓 ∶ 𝑈 ⊆ 𝑆1 → 𝑆2 (for an open subset 𝑈 of 𝑆1 ) is conformal if for every
𝑝 ∈ 𝑈 the derivative 𝐷 𝑓 𝑝 ∶ 𝑇 𝑝 𝑆1 → 𝑇 𝑓 ( 𝑝) 𝑆2 at 𝑝 is an isomorphism that preserves
angles — i.e., if, for all v, w in 𝑇 𝑝 𝑆1 , the angles ∠(𝐷 𝑓 𝑝 (v), 𝐷 𝑓 𝑝 (w)) and ∠(v, w)
are equal. This means that two regular curves that intersect at 𝑝 at a certain angle are

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 83

P. V. Araújo, Differential Geometry, https://doi.org/10.1007/978-3-031-62384-4_4
84 4 The Intrinsic Geometry of Surfaces

sent by the conformal mapping 𝑓 into curves that intersect at 𝑓 (𝑝) at the same angle.
(Note that we are referring to non-oriented angles).
We now give a brief interlude of linear algebra showing several characterizations
of linear maps which preserve angles.

Lemma 4.1.1 Let (𝐸 1 , ⟨⋅, ⋅⟩1 ) and (𝐸 2 , ⟨⋅, ⋅⟩2 ) be spaces of the same dimension n
equipped with an inner product and let 𝐿∶ 𝐸 1 → 𝐸 2 be a linear isomorphism. Then
the following conditions are equivalent:

i. 𝐿 preserves angles;
ii. 𝐿 is a similarity — that is, there exists 𝜆 > 0 such that ∣𝐿(v)∣2 = 𝜆∣v∣1 for all
v ∈ 𝐸1 ;
iii. there exists 𝜆 > 0 such that ⟨𝐿(v), 𝐿(w)⟩2 = 𝜆2 ⟨v, w⟩1 for all v, w ∈ 𝐸 1 ;
iv. there exist 𝜆 > 0 and a basis (v1 , . . . , v𝑛 ) of 𝐸 1 such that ⟨𝐿(v𝑖 ), 𝐿(v 𝑗 )⟩2 =
𝜆2 ⟨v𝑖 , v 𝑗 ⟩1 for all 1 ≤ 𝑖, 𝑗 ≤ 𝑛.

Proof The equivalence between iii. and iv. is simple to verify, and ii. is included for
informational purposes only; finally, to show that ii. ⇔ iii. may be left as an exercise.
i. ⇒ iii. Let (e1 , . . . , e𝑛 ) be an orthonormal basis of 𝐸 1 . The vectors
𝐿(e1 ), . . . , 𝐿(e𝑛 ) are then pairwise orthogonal. Given 𝑖 ≠ 𝑗, the angle 𝜃 ∈ [0, 𝜋]
between e𝑖 and e𝑖 + e 𝑗 is given by

⟨e𝑖 , e𝑖 + e 𝑗 ⟩1 1
cos 𝜃 = =√
∣e𝑖 ∣1 ∣e𝑖 + e 𝑗 ∣1 2
so that 𝜃 = 𝜋
4. Since 𝐿 preserves angles, we also have

⟨𝐿(e𝑖 ), 𝐿(e𝑖 + e 𝑗 )⟩2 1

=√ ,
∣𝐿(e𝑖 )∣2 ∣𝐿(e𝑖 + e 𝑗 )∣2 2
and from this
⟨𝐿(e𝑖 ), 𝐿(e𝑖 )⟩2 1
∣𝐿(e𝑖 )∣2 = = √ ∣𝐿(e𝑖 + e 𝑗 )∣2 .
∣𝐿(e𝑖 )∣2 2

Since the roles of e𝑖 and e 𝑗 are interchangeable, we also have

1
∣𝐿(e 𝑗 )∣2 = √ ∣𝐿(e𝑖 + e 𝑗 )∣2 ,
2
and so we deduce that ∣𝐿(e𝑖 )∣2 = ∣𝐿(e 𝑗 )∣2 for all 1 ≤ 𝑖, 𝑗 ≤ 𝑛. We now easily conclude
that iii. holds for 𝜆 = ∣𝐿(e𝑖 )∣2.
iii. ⇒. i. Just note that if iii. holds then we have, for all vectors v, w ∈ 𝐸 1

⟨𝐿(v), 𝐿(w)⟩2 𝜆2 ⟨v, w⟩1 ⟨v, w⟩1

= = ⋅◻
∣𝐿(v)∣2 ∣𝐿(w)∣2 𝜆∣v∣1 𝜆∣w∣1 ∣v∣1 ∣w∣1
4.1 Conformal Mappings and Isometries 85

From condition iii. of the lemma, it follows that 𝑓 ∶ 𝑈 ⊆ 𝑆1 → 𝑆2 is conformal if

and only if there exists a function 𝜆∶ 𝑈 → R + such that, for all 𝑝 ∈ 𝑈 and v, w ∈ 𝑇 𝑝 𝑆,
2
⟨𝐷 𝑓 𝑝 (v), 𝐷 𝑓 𝑝 (w)⟩ = 𝜆(𝑝) ⟨v, w⟩. (4.1)

By iv., condition (4.1) only needs to be checked for the vectors of a given basis of
𝑇 𝑝 𝑆. This has the consequence that in local coordinates Φ(𝑢, 𝑣) the condition that 𝑓
is conformal is equivalent to the combination of the three equalities
2
⟨( 𝑓 ○ Φ)𝑢 , ( 𝑓 ○ Φ)𝑢 ⟩ = 𝜆(𝑢, 𝑣) ⟨Φ𝑢 , Φ𝑢 ⟩
2
⟨( 𝑓 ○ Φ)𝑢 , ( 𝑓 ○ Φ)𝑣 ⟩ = 𝜆(𝑢, 𝑣) ⟨Φ𝑢 , Φ𝑣 ⟩
2
⟨( 𝑓 ○ Φ)𝑣 , ( 𝑓 ○ Φ)𝑣 ⟩ = 𝜆(𝑢, 𝑣) ⟨Φ𝑣 , Φ𝑣 ⟩,

that express that the coefficients of the first fundamental form of the parameterization
Φ(𝑢, 𝑣) of 𝑆1 are proportional to those of the parametrization Ψ(𝑢, 𝑣) = 𝑓 ○Φ(𝑢, 𝑣) of
𝑆2 . (There is a certain abuse of language here, since only locally can it be guaranteed
that Ψ is a parameterization.) It further follows from these formulas that 𝜆∶ 𝑈 → R +
is differentiable, because its expression 𝜆(𝑢, 𝑣) in local coordinates is differentiable.

Example 4.1.2 Let us verify that the stereographic projection (Exercise 30, Section 2.1)
𝜋∶ S1 ∖ {(0, 0, 1)} → R 2 is conformal. We recall that

2𝑥 2𝑦
𝜋(𝑥, 𝑦, 𝑧) = ( , );
1−𝑧 1−𝑧
so, using local coordinates
√ √
Φ(𝑢, 𝑣) = ( 1 − 𝑣 2 cos 𝑢, 1 − 𝑣 2 sin 𝑢, 𝑣) , 𝑣 ∈ ]−1, 1[ ,

in S2 , the coordinates Ψ = 𝜋 ○ Φ are given by

√ √
⎛ 1+𝑣 1+𝑣 ⎞
Ψ(𝑢, 𝑣) = 2 cos 𝑢, sin 𝑢 .
⎝ 1−𝑣 1−𝑣 ⎠

The coefficients of Φ(𝑢, 𝑣) are

1
𝐸 = 1 − 𝑣2, 𝐹 = 0, 𝐺= ,
1 − 𝑣2
and those of Ψ(𝑢, 𝑣) are

4(1 + 𝑣) 4
𝐸= , 𝐹 = 0, 𝐺= ⋅
1−𝑣 3
(1 − 𝑣) (1 + 𝑣)

Noting that
86 4 The Intrinsic Geometry of Surfaces
2 2 2
𝐸 = 𝜆(𝑢, 𝑣) 𝐸, 𝐹 = 𝜆(𝑢, 𝑣) 𝐹, 𝐺 = 𝜆(𝑢, 𝑣) 𝐺,

where 𝜆(𝑢, 𝑣) = 1−𝑣2

, we conclude that the restriction of 𝜋 to the image of Φ(𝑢, 𝑣)
is conformal. Such a parameterization excludes a meridian, but any point of S2 ,
except the poles, can be covered by Φ(𝑢, 𝑣) if we adjust the domain of 𝑢. We have
thus shown that 𝜋 is conformal at all points of S2 ∖ {(0, 0, 1), (0, 0, −1)}. But since
lim 𝜆(𝑢, 𝑣) = 1, the function 𝜆 extends continuously to S2 ∖{(0, 0, 1)}, and therefore
𝑣→−1
𝜋 is conformal on its entire domain. ◻

If 𝑓 ∶ 𝑈 ⊆ 𝑆1 → 𝑆2 is conformal, we have already observed that the function

𝜆∶ 𝑈 → R + given by (4.1) is differentiable, and hence continuous. This means that in
a neighborhood of 𝑝 0 ∈ 𝑈, the mapping 𝑓 is “approximately” a similarity of ratio
𝜆(𝑝 0 ), which is the reason why cartographic mappings based on conformal mappings
are more faithful to reality when representing small regions.
Given the importance of the concept, it is natural to ask whether between two
arbitrary surfaces there exist conformal mappings, at least locally. Since the inverse
and composite of conformal mappings are still conformal, the (affirmative) answer to
this question is a consequence of the fact that any point on a surface has a neighborhood
that can be parametrized by a conformal mapping. Such a parameterization is called
isothermal.
The condition for Φ(𝑢, 𝑣) to be isothermal is that its coefficients 𝐸, 𝐹, 𝐺 are
proportional to the coefficients of the parameterization (𝑢, 𝑣) ↦ (𝑢, 𝑣) of the plane,
which are 1, 0, 1 — that is, 𝐸 ≡ 𝐺, 𝐹 ≡ 0.
The existence of isothermal parameterizations for arbitrary surfaces is a deep
result, and we therefore omit its proof (which you can find, in all generality, in [25],
vol.IV, p. 455 ff.). Note that by 4.1.2 the inverse of the stereographic projection is an
isothermal parameterization of the sphere; in the exercises we give other examples.
Let us now talk about isometries. We say that 𝑓 ∶ 𝑈 ⊆ 𝑆1 → 𝑆2 is a local isometry
if for every 𝑝 ∈ 𝑈 the derivative 𝐷 𝑓 𝑝 ∶ 𝑇 𝑝 𝑆1 → 𝑇 𝑓 ( 𝑝) 𝑆2 is a linear isometry — that is,
if ⟨𝐷 𝑓 𝑝 (v), 𝐷 𝑓 𝑝 (w)⟩ = ⟨v, w⟩ for all v, w ∈ 𝑇 𝑝 𝑆1 . An isometry is a diffeomorphism
𝑓 ∶ 𝑆1 → 𝑆2 which is also a local isometry. Two surfaces are isometric if there exists
an isometry between them. They are locally isometric if every point of each surface
has an open neighborhood which is isometric to an open subset of the other surface.
A local isometry is thus a mapping that preserves the first fundamental form, and
therefore preserves all the quantities that depend on it: the length of curves, the
angle between two curves, and the area of small regions (i. e., such that the mapping,
restricted to them, is injective). In local coordinates, 𝑓 is a local isometry when
the coefficients of the first fundamental form of the parameterizations Φ(𝑢, 𝑣) and
Ψ(𝑢, 𝑣) = 𝑓 ○ Φ(𝑢, 𝑣) coincide.
Examples 4.1.3 A. The linear isometries of R 2 are isometries in the sense just
defined. The converse, which is more interesting, appears in the exercises of this
section: any isometry of R 2 is the composite of a linear isometry with a translation.
B. Let 𝑆 be a surface of revolution with axis 𝑟 and let 𝐿∶ R 3 → R 3 be a rotation
around 𝑟. Then 𝐿∣𝑆 ∶ 𝑆 → 𝑆 is an isometry of 𝑆.
4.1 Conformal Mappings and Isometries 87

C. The plane and cylinder C = {(𝑥, 𝑦, 𝑧) ∈ R 3 ∶ 𝑥 2 + 𝑦 2 = 1} are locally isometric: a

local isometry R 2 → C is 𝑓 (𝑢, 𝑣) = (cos 𝑢, sin 𝑢, 𝑣), and consists simply of wrapping
the plane around the cylinder. By identifying R 2 with R 2 × {0} ⊆ R 3 , this example
shows that not all isometries or local isometries between surfaces are restrictions of
isometries of R 3 .
√
D. The cone ℭ 𝑘 = {(𝑥, 𝑦, 𝑧)∶ 𝑧 = 𝑘 𝑥 2 + 𝑦 2 , 𝑧 ≠ 0}, for all 𝑘 > 0, is locally isometric
to the plane: if we cut ℭ 𝑘 along a generatrix and unroll it, then we obtain a circular
sector 𝑈 bounded by two half-lines with origin 𝑂, in which the arcs of the circle with
center 𝑂 correspond to the parallels of ℭ 𝑘 . If 𝛼 ∈ ]0, 𝜋2 [ is the angle that the cone’s
generatrix makes with the 𝑧-axis, and 𝛽 ∈ ]0, 2𝜋[ is the angle defined by 𝑈, then

1
tg 𝛼 = , 𝛽 = 2𝜋 sin 𝛼,
𝑘
and also 𝑈 = {(𝜌 cos 𝜑, 𝜌 sin 𝜑) ∈ R 2 ∶ 𝜌 > 0, 0 < 𝜑 < 𝛽}. (See Fig. 4.1.)

a b

Figure 4.1

We define a mapping 𝑓 ∶ 𝑈 → ℭ 𝑘 by sending the point with polar coordinates

(𝜌, 𝜑) to the point

𝜌𝛽 2𝜋𝜑 𝜌𝛽 2𝜋𝜑
( cos ( ), sin ( ) , 𝜌 cos 𝛼) ∶
2𝜋 𝛽 2𝜋 𝛽

𝑓 is a local isometry because the parameterizations

Φ(𝜌, 𝜑) = (𝜌 cos 𝜑, 𝜌 sin 𝜑)

and
𝜌𝛽 2𝜋𝜑 𝜌𝛽 2𝜋𝜑
𝑓 ○ Φ(𝜌, 𝜑) = ( cos ( ), sin ( ) , 𝜌 cos 𝛼)
2𝜋 𝛽 2𝜋 𝛽
have the same coefficients: 𝐸 = 1, 𝐹 = 0, 𝐺 = 𝜌 2 . ◻
88 4 The Intrinsic Geometry of Surfaces

Exercises

73. The inversion of the plane with respect to the circle with center 𝑂 and radius 𝑟
is the mapping that sends each point 𝑝 ≠ 𝑂 to the point 𝑝 ′ of the half-line 𝑂 𝑝 that
verifies the condition 𝑂 𝑝 ′ 𝑂 𝑝 = 𝑟 2 . Show that the inversion is a conformal mapping
and that it reverses orientation of the plane.

74. Consider S2 parametrized by two angles: the longitude 𝜑 ∈] − 𝜋, 𝜋[ and the

latitude 𝜃̃ ∈ ] − 𝜋2 , 𝜋2 [. The Mercator projection of the sphere into the plane (𝑢, 𝑣)
is defined by the following conditions: (i) the parallels 𝜃̃ = constant are sent to the
lines 𝑣 = constant, and the meridians 𝜑 = constant to the lines 𝑢 = constant; (ii) it is
a conformal mapping; (iii) the distances along the equator 𝜃̃ = 0 are converted into
proportional distances; (iv) the point 𝜃̃ = 0, 𝜑 = 0 is sent to the origin.

̃ is sent to the point 𝑢 = 𝜆𝜑, 𝑣 = 𝜆𝑔(𝜃),

Show that the point with coordinates (𝜑, 𝜃) ̃
where 𝜆 is a constant and

1 ̃
̃ =
𝑔 ′ (𝜃) ̃ = log tg ( 𝜋 + 𝜃 )]
[and therefore 𝑔(𝜃)
̃
cos 𝜃 4 2

75. Find a conformal mapping of T 2 into the plane.

76. The catenoid is the surface generated by the catenary

𝑣 ↦ (𝑎 cosh 𝑣, 0, 𝑎𝑣) (𝑎 > 0)

around the 𝑧-axis, and it can be parameterized by

Φ(𝑢, 𝑣) = (𝑎 cosh 𝑣 cos 𝑢, 𝑎 cosh 𝑣 sin 𝑢, 𝑎𝑣)

(see Figure 4.2). Consider also the helicoid parametrized by

𝑢, ̃
Ψ(̃ 𝑣 cos 𝑢̃, ̃
𝑣 ) = (̃ 𝑢 ).
𝑣 sin cos 𝑢, 𝑎̃

Using the change of coordinates 𝑢̃ = 𝑢, ̃ 𝑣 = 𝑎 sinh 𝑣, show that the helicoid is

locally isometric to the catenoid. What are the images under this local isometry of
the helices ̃
𝑣 = constant and the straight lines 𝑢̃ = constant?
4.1 Conformal Mappings and Isometries 89

Figure 4.2

77. Let 𝑓 ∶ R 2 → R 2 be an isometry such that 𝑓 (0, 0) = (0, 0).

(a) Show that ∣ 𝑓 (𝑝) − 𝑓 (𝑞)∣ ≤ ∣𝑝 − 𝑞∣ for all points 𝑝, 𝑞 of R 2 . Conclude that 𝑓
preserves the distance between points.
(b) Show that 𝑓 is linear. Hint: (i) use the equality ⟨ 𝑓 (𝑝), 𝑓 (𝑞)⟩ = 12 {∣∣ 𝑓 (𝑝)∣2 +
∣ 𝑓 (𝑞)∣2 − ∣ 𝑓 (𝑝) − 𝑓 (𝑞)∣2 } to show that 𝑓 preserves the inner product; (ii) computing
∣ 𝑓 (𝑝 + 𝑞) − ( 𝑓 (𝑝) + 𝑓 (𝑞))∣2 , show that 𝑓 (𝑝 + 𝑞) = 𝑓 (𝑝) + 𝑓 (𝑞).
(c) Deduce that any isometry of R 2 is the composite of a linear isometry and a
translation.

78. (a) Let 𝐸(̃ 𝑣 ) be 𝐶 ∞ functions defined on a compact interval 𝐼. Show that
𝑣 ), 𝐺(̃
there exists a parameterization

𝑢, ̃
Ψ(̃ 𝑣 ) = (𝜌(̃
𝑣 ) cos(𝑎̃
𝑢 ), 𝜌(̃
𝑣 ) sin(𝑎̃
𝑢 ), 𝑧(̃
𝑣 ))

of a surface of revolution whose coefficients of the first fundamental form are

𝐸 = 𝐸(̃𝑣 ), 𝐹 ≡ 0, 𝐺 = 𝐺(̃
𝑣 ).
(b) Consider the parametrized surface Φ(𝑢, 𝑣) = (3𝑢 + 3𝑢𝑣 2 − 𝑢 3 , −3𝑣 − 3𝑢 2 𝑣 +
𝑣 , 3𝑢 2 − 3𝑣 2 ), where (𝑢, 𝑣) ≠ (0, 0), and compute the coefficients 𝐸, 𝐹, 𝐺 of this
3

parameterization. Changing to the coordinates 𝑢̃, ̃ 𝑣 given by 𝑢 = ̃𝑣 cos 𝑢̃, 𝑣 = ̃

𝑣 sin 𝑢̃,
conclude that the given surface is locally isometric to a surface of revolution.

79. Construct an isometry between the surface defined by the equation 𝑧 = 𝑦 2 and the
plane 𝑧 = 0 which sends the parabolas 𝑧 = 𝑦 2 , 𝑥 = 𝑥 0 to the lines 𝑧 = 0, 𝑥 = 𝑥0 .

80. Show that if 𝑓 ∶ 𝑆1 → 𝑆2 is a diffeomorphism that preserves lengths of curves then

𝑓 is an isometry.
90 4 The Intrinsic Geometry of Surfaces

4.2 Gauss’s Theorema Egregium

In this section we give a necessary condition for two surfaces to be locally isometric,
from which it will follow, for example, the non-existence of a local isometry between
the plane and the sphere. In the proof we make use of Christoffel symbols, which are
functions of local parameters that are invariant under isometries.
Given a parameterization Φ(𝑢, 𝑣) of an oriented surface 𝑆, we consider at each point
the trihedron (Φ𝑢 , Φ𝑣 , 𝑁), where 𝑁(𝑢, 𝑣) is the normal field to 𝑆 compatible with
the orientation. [It suffices that 𝑁 be a local orientation, since all the considerations
we make are local in character — and therefore all the results in this section hold for
non-orientable surfaces.] We can express the second-order derivatives of Φ in terms
of this trihedron:
Φ𝑢𝑢 = Γ11
1
Φ𝑢 + Γ112
Φ𝑣 + 𝜆1 𝑁,

Φ𝑢𝑣 = Γ12
1
Φ𝑢 + Γ12
2
Φ𝑣 + 𝜆2 𝑁,
(4.2)
Φ𝑣𝑢 = 1
Γ21 Φ𝑢 + Γ21
2
Φ𝑣 + 𝜆3 𝑁,

Φ𝑣𝑣 = Γ22
1
Φ𝑢 + Γ22
2
Φ𝑣 + 𝜆4 𝑁.
The 𝜆𝑖 in (4.2) are just the coefficients of the second fundamental form: indeed, by the
formulas in Section 3.2, we have 𝜆1 = ⟨Φ𝑢𝑢 , 𝑁⟩ = 𝑒, 𝜆2 = 𝜆3 = ⟨Φ𝑢𝑣 , 𝑁⟩ = 𝑓 , 𝜆4 =
⟨Φ𝑣 , 𝑁⟩ = 𝑔. The coefficients Γ𝑖𝑘𝑗 are the Christoffel symbols of the parameterization
Φ(𝑢, 𝑣), and it follows from the definition that they are symmetric with respect to the
lower indices: thus, Γ𝑖𝑘𝑗 = Γ 𝑘𝑗𝑖 . To compute them, we form the inner product of the
equalities (4.2) with Φ𝑢 and Φ𝑣 , obtaining
⎧ 1
⎪
⎪Γ 𝐸 + Γ112
𝐹 = ⟨Φ𝑢𝑢 , Φ𝑢 ⟩ = 21 𝐸 𝑢
⎨ 11
⎪Γ11
⎪
1
𝐹 + Γ11
2
𝐺 = ⟨Φ𝑢𝑢 , Φ𝑣 ⟩ = 𝐹𝑢 − 12 𝐸 𝑣
⎩

⎧
⎪
⎪Γ12
1
𝐸 + Γ12
2
𝐹 = ⟨Φ𝑢𝑣 , Φ𝑢 ⟩ = 12 𝐸 𝑣
⎨ 1 (4.3)
⎪
⎪ Γ 𝐹 + Γ12
2
𝐺 = ⟨Φ𝑢𝑣 , Φ𝑣 ⟩ = 12 𝐺 𝑢
⎩ 12

⎧
⎪
⎪Γ22
1
𝐸 + Γ22
2
𝐹 = ⟨Φ𝑣𝑣 , Φ𝑢 ⟩ = 𝐹𝑣 − 12 𝐺 𝑢
⎨ 1
⎪
⎪ Γ 𝐹 + Γ22
2
𝐺 = ⟨Φ𝑣 , Φ𝑣 ⟩ = 21 𝐺 𝑣 .
⎩ 22

Each of the equation systems in (4.3) has determinant 𝐸𝐺 − 𝐹 2 > 0, which shows that
they suffice for computing the Christoffel symbols, and that these are functions of the
coefficients 𝐸, 𝐹, 𝐺 and their derivatives. In particular, and since local isometries
preserve 𝐸, 𝐹, 𝐺, they also preserve the coefficients Γ𝑖𝑘𝑗 — more precisely, if 𝑓 is a local
4.2 Gauss’s Theorema Egregium 91

isometry then the Christoffel symbols of the parameterization Ψ(𝑢, 𝑣) = 𝑓 ○ Φ(𝑢, 𝑣)

coincide with the Christoffel symbols of Φ(𝑢, 𝑣).
It is more sensible to solve the systems of equations (4.3) in each case instead
of determining general expressions for the Γ𝑖𝑘𝑗 . However, we observe that when we
deal with orthogonal parameterizations (𝐹 ≡ 0), the calculation of the Christoffel
symbols is greatly simplified. For example, the symbols of the parameterization
(𝜌(𝑣) cos 𝑢, 𝜌(𝑣) sin 𝑢, 𝑧(𝑣)) of a surface of revolution, where 𝐸 = 𝜌 2 , 𝐹 = 0,
𝐺 = 𝜌 2 + 𝑧 2 , are
−𝜌𝜌 𝜌
1
Γ11 = 0, Γ112
= 2 2 , Γ12 1
= ,
𝜌 +𝑧 𝜌
𝜌𝜌 + 𝑧𝑧 (4.4)
2
Γ12 = 0, Γ221
= 0, Γ22 2
= 2 2
𝜌 +𝑧
We now look for relations between the Christoffel symbols, starting from the
identity
(Φ𝑢𝑢 )𝑣 = (Φ𝑢𝑣 )𝑢 . (4.5)
Differentiating the first and second equalities of (4.2) with respect to 𝑣 and to 𝑢,
respectively, we obtain

(Φ𝑢𝑢 )𝑣 = (Γ11
1
)𝑣 Φ𝑢 + (Γ11
2
)𝑣 Φ𝑣 + Γ11
1
Φ𝑢𝑣 + Γ11
2
Φ𝑣𝑣 + 𝑒 𝑣 𝑁 + 𝑒𝑁 𝑣 ,
(Φ𝑢𝑣 )𝑢 = (Γ12
1
)𝑢 Φ𝑢 + (Γ12
2
)𝑢 Φ𝑣 + Γ12
1
Φ𝑢𝑢 + Γ12
2
Φ𝑢𝑣 + 𝑓𝑢 𝑁 + 𝑓 𝑁𝑢 .

Using again (4.2) and expressing 𝑁 𝑣 and 𝑁𝑢 in terms of Φ𝑢 and Φ𝑣 , we can write
each of the terms of (4.5) as a linear combination of the vectors Φ𝑢 , Φ𝑣 , 𝑁. Equating
the coefficients of Φ𝑣 in the two linear combinations, we obtain

(Γ11
2
)𝑣 + Γ11
1 2
Γ12 + Γ11
2 2
Γ22 − 𝑒𝑎 22 = (Γ12
1
)𝑢 + Γ12
1 2
Γ11 + Γ12
2 2
Γ12 − 𝑓 𝑎 21 . (4.6)

By the formulas (3.6) and (3.7) in Section 3.2, we have

𝑒(−𝐹 𝑓 + 𝐸𝑔) 𝑓 (−𝐹𝑒 + 𝐸 𝑓 )

𝑒𝑎 22 − 𝑓 𝑎 21 = −
𝐸𝐺 − 𝐹 2 𝐸𝐺 − 𝐹 2
(4.7)
𝑒𝑔 − 𝑓 2
=𝐸 = 𝐸𝐾.
𝐸𝐺 − 𝐹 2

Combining (4.6) and (4.7), we obtain the following expression for the Gaussian
curvature:
1
𝐾= ((Γ11
2
)𝑣 − (Γ12
2
)𝑢 + Γ12
1
(Γ11
1
− Γ12
2
) + Γ11
1
(Γ22
2
− Γ12
1
)). (4.8)
𝐸

Although its utility is questionable, formula (4.8) has this consequence: it is possible to
compute the Gaussian curvature knowing only the coefficients of the first fundamental
form and their derivatives. In the calculation of the principal curvatures 𝑘 1 and 𝑘 2 ,
the Gauss map and its derivative (i.e., the second fundamental form) are crucial
92 4 The Intrinsic Geometry of Surfaces

ingredients. It is remarkable that their product 𝐾 = 𝑘 1 𝑘 2 is after all an intrinsic

concept, for it depends only on the first fundamental form.
As a consequence of this discussion we have the following theorem, which the
author himself has called “egregious”.
Theorem 4.2.1 (Gauss).The Gaussian curvature is invariant under local isometries.
More precisely, if 𝑓 ∶ 𝑈 ⊆ 𝑆1 → 𝑆2 is a local isometry then, for all 𝑝 in 𝑈, the curvature
of 𝑆1 at 𝑝 is equal to the curvature of 𝑆2 at 𝑓 (𝑝).
This theorem already shows, in many cases, that two given surfaces — such as
the sphere (which has curvature > 0) and the plane (curvature 0) — are not locally
isometric.
The deduction of formula (4.8) seems like magic: how did we pull such a rabbit
out of such a hat? Later on we will deduce, by more transparent means, other formulas
that also establish the invariance of the curvature by local isometries.

Exercises
81. Let 𝑆1 = {(𝑥, 𝑦, 𝑧)∶ 𝑧 = 0} and 𝑆2 = {(𝑥, 𝑦, 𝑧)∶ 𝑥 2 + 𝑦 2 = 𝑧}.
(a) Check that 𝑓 ∶ 𝑆1 → 𝑆2 given by 𝑓 (𝑥, 𝑦, 0) = (𝑥, 𝑦, 𝑥 2 + 𝑦 2 ) is not an isometry.
(b) Show that 𝑆1 and 𝑆2 are not locally isometric.
82. Consider the surfaces 𝑆1 and 𝑆2 given by

Φ(𝑢, 𝑣) = (𝑢 cos 𝑣, 𝑢 sin 𝑣, log 𝑢), (𝑢 ∈ R + , 0 < 𝑣 < 2𝜋)

and
Ψ(𝑢, 𝑣) = (𝑢 cos 𝑣, 𝑢 sin 𝑣, 𝑣) (𝑢, 𝑣 ∈ R ).
−1
Show that Ψ ○ Φ ∶ 𝑆1 → 𝑆2 is not a local isometry, although the curvature of 𝑆2 at
Ψ(𝑢, 𝑣) is equal to that of 𝑆1 at Φ(𝑢, 𝑣).
83. Consider T 2 with the usual parameterization. Show that any isometry of T 2 sends
the equator 𝑣 = 0 to itself (use Exercise 71-a) in Section 3.3).
84. Show that a conformal mapping that preserves areas (see Exercise 71, in Section
2.5) is necessarily a local isometry (and therefore all maps of the Earth’s surface have
to choose one of two defects: either they distort areas, or they distort shapes).

4.3 Covariant Derivative, Parallel Transport, Geodesic Curvature

All the concepts we define in this section belong to intrinsic geometry, in the sense
that they are invariant under local isometries. The method, used here repeatedly, to
establish the intrinsic character of a certain concept consists of showing that it is a
function solely of the coefficients 𝐸, 𝐹, 𝐺 and their derivatives — which is the case
whenever we can express it by means of these quantities and the Christoffel symbols
only.
4.3 Covariant Derivative, Parallel Transport, Geodesic Curvature 93

Let v be a vector field defined on an open subset 𝑈 ⊆ 𝑆. Given a regular curve

𝛼∶ 𝐼 → 𝑈, the covariant derivative of v along 𝛼 is the orthogonal projection of
𝐷v 𝐷v
(v ○ 𝛼)′ (𝑡) on the tangent plane 𝑇𝛼(𝑡) 𝑆, denoted by (𝑡). So, to compute (𝑡),
′
𝑑𝑡 𝑑𝑡
we just have to subtract from (v ○ 𝛼) (𝑡) its component with respect to the normal
𝑁 ○ 𝛼(𝑡) to the surface.

Example 4.3.1 Consider a unit vector field v in S2 ∖ {(0, 0, 1), (0, 0, −1)} having
constant angle 𝜋4 with the parallels of the sphere. In spherical coordinates

Φ(𝜑, 𝜃) = (sin 𝜃 cos 𝜑, sin 𝜃 sin 𝜑, cos 𝜃),

we put
1 1
v(𝜑, 𝜃) = √ (−Φ 𝜃 + Φ𝜑 )
2 sin 𝜃
1
= √ (−cos 𝜃 cos 𝜑 − sin 𝜑, −sin 𝜃 sin 𝜑 + cos 𝜑, sin 𝜃).
2
Let us compute the covariant derivative of v along the parallel 𝜃 = 𝜃 0 , parametrized
by 𝛼 𝜃0 (𝑡) = Φ(𝑡, 𝜃 0 ). We now have

1
v′ (𝑡) = √ (cos 𝜃 0 sin 𝑡 − cos 𝑡, −sin 𝜃 0 cos 𝑡 − sin 𝑡, 0),
2
𝐷v
and we obtain (𝑡) by subtracting from v′ (𝑡) its normal component:
𝑑𝑡

(𝑡) = v′ (𝑡) − ⟨v′ (𝑡), 𝑁(𝑡, 𝜃 0 )⟩𝑁(𝑡, 𝜃 0 )

𝐷v
𝑑𝑡
1
= √ (cos 𝜃 0 sin 𝑡 − cos 𝑡, −sin 𝜃 0 cos 𝑡 − sin 𝑡, 0)+
2
1
+ √ sin 𝜃 0 (sin 𝜃 0 cos 𝑡, sin 𝜃 0 sin 𝑡, cos 𝜃 0 )
2
− cos 𝜃 0 1
= √ ( Φ 𝜑 ∣(𝑡 , 𝜃 ) + Φ 𝜃 ∣(𝑡 , 𝜃 ) ) . ◻
2 sin 𝜃 0 0 0

It is worthy to note that the covariant derivative of v along 𝛼 depends on the

parameterization of 𝛼, but not on the orientation of the surface. Let us now establish its
intrinsic character. In local coordinates, if we write v(𝑢, 𝑣) = 𝑎(𝑢, 𝑣)Φ𝑢 + 𝑏(𝑢, 𝑣)Φ𝑣 ,
𝛼(𝑡) = Φ(𝑢(𝑡), 𝑣(𝑡)), 𝑎(𝑡) = 𝑎(𝑢(𝑡), 𝑣(𝑡)) and 𝑏(𝑡) = 𝑏(𝑢(𝑡), 𝑣(𝑡)), we have

v(𝑡) = 𝑎(𝑡)Φ𝑢 + 𝑏(𝑡)Φ𝑣 ,

v′ (𝑡) = 𝑎 ′ Φ𝑢 + 𝑎(𝑢 ′ Φ𝑢𝑢 + 𝑣 ′ Φ𝑢𝑣 ) + 𝑏 ′ Φ𝑣 + 𝑏(𝑢 ′ Φ𝑢𝑣 + 𝑣 ′ Φ𝑣𝑣 )
= (𝑎 ′ + 𝑎𝑢 ′ Γ11
1
+ 𝑎𝑣 ′ Γ12
1
+ 𝑏𝑢 ′ Γ12
1
+ 𝑏𝑣 ′ Γ22
1
)Φ𝑢 (4.9)
′
+ (𝑏 + 𝑎𝑢 ′ Γ11
2
+ 𝑎𝑣 ′ Γ12
+ +
2
𝑏𝑢 ′ Γ12
2
𝑏𝑣 ′ Γ22
2
)Φ𝑣
′ ′ ′ ′
+ (𝑎𝑢 𝑒 + 𝑎𝑣 𝑓 + 𝑏𝑢 𝑓 + 𝑏𝑣 𝑔)𝑁.
94 4 The Intrinsic Geometry of Surfaces

Ignoring the normal component, the components of v′ (𝑡) with respect to Φ𝑢 and
𝐷v
Φ𝑣 give the expression of the covariant derivative (𝑡) in local coordinates. If
𝑑𝑡
we look at the coefficients of Φ𝑢 and Φ𝑣 , we see that they have “intrinsic character”
and are therefore preserved by isometries. To formalize this observation, let us
take a mapping 𝑓 ∶ 𝑈 → 𝑆2 that is a diffeomorphism onto the image, and consider
the vector field u = 𝐷 𝑓 (v) which is the transfer of v by 𝑓 , defined on 𝑓 (𝑈) by
𝐷u
u( 𝑓 (𝑝)) = 𝐷 𝑓 𝑝 (v(𝑝)). Let us denote by the covariant derivative of u along the
𝑑𝑡
curve 𝛽(𝑡) = 𝑓 ○ 𝛼(𝑡). We claim that if 𝑓 is an isometry, then
𝐷u 𝐷v
= 𝐷 𝑓 𝛼(𝑡) ( )⋅ (4.10)
𝑑𝑡 𝑑𝑡
In fact, using coordinates Φ(𝑢, 𝑣) on 𝑆 and Ψ(𝑢, 𝑣) = 𝑓 ○ Φ(𝑢, 𝑣) on 𝑆2 , we see by
𝐷u 𝐷v
(4.9) that the expressions of with respect to the basis (Ψ𝑢 , Ψ𝑣 ) and of with
𝑑𝑡 𝑑𝑡
respect to the basis (Φ𝑢 , Φ𝑣 ) have the same coefficients, which proves (4.10).
𝐷v
Another property of the covariant derivative of v along 𝛼 is that at each instant
𝑑𝑡
it only depends on the velocity vector of the curve at that instant.

𝐷v
Proposition 4.3.2 (𝑡) only depends on 𝛼′ (𝑡) = 𝑢 ′ (𝑡)Φ𝑢 + 𝑣 ′ (𝑡)Φ𝑣 and not on
𝑑𝑡
the curve 𝛼.

Proof In fact, noting that

𝑎 ′ (𝑡) =
𝜕𝑎 ′ 𝜕𝑎 ′
𝑢 (𝑡) + 𝑣 (𝑡),
𝜕𝑢 𝜕𝑣
𝑏 ′ (𝑡) =
𝜕𝑏 ′ 𝜕𝑏 ′
𝑢 (𝑡) + 𝑣 (𝑡),
𝜕𝑢 𝜕𝑣
we recognize that the functions 𝑢(𝑡), 𝑣(𝑡) only enter in (4.9) through the value they
and their first derivatives 𝑢 ′ and 𝑣 ′ take at time 𝑡. ◻

𝐷v
We further observe that the expression of is linear in 𝑢 ′ , 𝑣 ′ . Thanks to
𝑑𝑡
Proposition 4.3.2, for every 𝑝 ∈ 𝑈, a linear mapping 𝐷v 𝑝 ∶ 𝑇 𝑝 𝑆 → 𝑇 𝑝 𝑆 is defined as
follows: given w ∈ 𝑇 𝑝 𝑆, let 𝛼(𝑡) be a curve in 𝑈 such that 𝛼(𝑡 0 ) = 𝑝 and 𝛼′ (𝑡0 ) = w;
then 𝐷v 𝑝 (w) is the covariant derivative of v along 𝛼 computed at 𝑡0 . Intuitively,
𝐷v 𝑝 (w) is a kind of directional derivative of v in the direction of w.
Given a field v in 𝑈 ⊆ 𝑆, the divergence of v is the mapping Div v∶ 𝑈 ⊆ 𝑆 → R
such that, at every 𝑝 ∈ 𝑈, its value Div v(𝑝) is the trace of the linear mapping 𝐷v 𝑝 .

Example 4.3.3 If v(𝑢, 𝑣) = ( 𝑓1 (𝑢, 𝑣), 𝑓2 (𝑢, 𝑣)) is a vector field on an open subset
𝑈 ⊆ R 2 , its covariant derivative along a curve 𝛼∶ 𝐼 → 𝑈 is the usual derivative
(v ○ 𝛼)′ (𝑡), and therefore the linear mapping 𝑑v 𝑝 is just the derivative of v as the
4.3 Covariant Derivative, Parallel Transport, Geodesic Curvature 95

mapping 𝑈 → R 2 . The matrix of 𝐷v 𝑝 with respect to the canonical basis of R 2 is

the Jacobian
⎡ 𝜕 𝑓1 𝜕 𝑓1 ⎤
⎢ ⎥
⎢ 𝑑𝑢 𝑑𝑣 ⎥
⎢ ⎥
⎢ ⎥
⎢ 𝜕 𝑓2 𝜕 𝑓2 ⎥
⎣ 𝜕𝑢 𝜕𝑣 ⎦∣
𝑝

𝜕 𝑓1 𝜕 𝑓2
and its trace ∣ + ∣ is the divergence of v computed at 𝑝. ◻
𝜕𝑢 𝑝 𝜕𝑣 𝑝
Proposition 4.3.4 Divergence is invariant under isometries: if 𝑓 ∶ 𝑈 ⊆ 𝑆 → 𝑆2 is
an isometry into its image and v is a vector field on 𝑈, then Div(𝐷 𝑓 v)( 𝑓 (𝑝)) =
Div v(𝑝).

Proof We put u = 𝐷 𝑓 v, and start by rewriting formula (4.10) using the linear
mappings

𝐷u 𝑓 ( 𝑝) ∶ 𝑇 𝑓 ( 𝑝) 𝑆2 → 𝑇 𝑓 ( 𝑝) 𝑆2 and 𝐷v 𝑝 ∶ 𝑇 𝑝 𝑆 → 𝑇 𝑝 𝑆.

Let us fix a curve 𝛼(𝑡) such that 𝛼(0) = 𝑝, and let us put 𝛽(𝑡) = 𝑓 ○ 𝛼(𝑡), w1 = 𝛼′ (0)
𝐷u 𝐷v
and w2 = 𝛽′ (0). Then we have (0) = 𝐷u 𝑓 ( 𝑝) (w2 ) and (0) = 𝐷v 𝑝 (w1 ) =
𝑑𝑡 𝑑𝑡
−1
𝐷v 𝑝 (𝐷 𝑓 𝑝 (w2 )). By (4.10) we can write

𝐷u 𝑓 ( 𝑝) (w2 ) = 𝐷 𝑓 𝑝 ○ 𝐷v 𝑝 ○ 𝐷 𝑓 𝑝−1 (w2 )for everyw2 ∈ 𝑇 𝑓 ( 𝑝) 𝑆2 ,

i.e.
𝐷u 𝑓 ( 𝑝) = 𝐷 𝑓 𝑝 ○ 𝐷v 𝑝 ○ 𝐷 𝑓 𝑝−1 , (4.11)
which means that 𝐷u 𝑓 ( 𝑝) and 𝐷v 𝑝 are conjugate linear mappings, and therefore
have the same trace. ◻

We will revisit the notion of divergence in the next section, where we prove a
theorem that adapts the divergence theorem known from vector calculus to surfaces.
For now, we return to the covariant derivative.
The reader must have noticed that in the computation of the covariant derivative
of v along 𝛼, only the way v is defined on the trace of the curve plays a role, and so it
is not necessary that the field be defined on other points. We thus define a vector field
along a regular curve 𝛼(𝑡) as a function v(𝑡) such that, for every 𝑡, the vector v(𝑡)
belongs to 𝑇𝛼(𝑡) 𝑆; if v is differentiable (i.e., if, by writing v(𝑡) = 𝑎(𝑡)Φ𝑢 + 𝑏(𝑡)Φ𝑣 ,
the functions 𝑎(𝑡) and 𝑏(𝑡) are differentiable), we can as before compute the covariant
derivative of v along 𝛼.

Lemma 4.3.5 If v(𝑡), w(𝑡) are two vector fields along the curve 𝛼(𝑡) then

𝑑 𝐷v 𝐷w
⟨v, w⟩ = ⟨ , w⟩ + ⟨v, ⟩⋅
𝑑𝑡 𝑑𝑡 𝑑𝑡

Proof Denoting by the superscript 𝑛 the normal component of a vector, we have

96 4 The Intrinsic Geometry of Surfaces

⟨v, w⟩ = ⟨v′ w⟩ + ⟨v, w′ ⟩

𝑑
𝑑𝑡
+ (v′ ) , w⟩ + ⟨v, + (w′ ) ⟩
𝐷v 𝐷w
=⟨
𝑛 𝑛
𝑑𝑡 𝑑𝑡
𝐷v 𝐷w
=⟨ , w⟩ + ⟨v, ⟩⋅ ◻
𝑑𝑡 𝑑𝑡

We now introduce the concept of geodesic curvature. Given a regular curve

𝛼∶ 𝐼 → 𝑆 on an oriented surface, we can consider the unit tangent vector field along 𝛼,
given by 𝝉1 (𝑡) = ∣𝛼′1(𝑡)∣ 𝛼′ (𝑡), and also the vector field 𝝉2 = 𝑁 × 𝝉1 . This means that
at each instant 𝑡, the pair (𝝉1 (𝑡), 𝝉2 (𝑡)) is an orthonormal and positively oriented
basis of 𝑇𝛼(𝑡) 𝑆. By 4.3.5 we have

𝐷𝝉1 𝑑 1
⟨𝝉1 (𝑡), (𝑡)⟩ = ( ∣𝝉1 ∣2 ) = 0,
𝑑𝑡 𝑑𝑡 2

𝐷𝝉1
and so the vectors (𝑡) and 𝝉2 (𝑡) are collinear. The geodesic curvature 𝑘 𝑔 (𝑡) of
𝑑𝑡
𝛼 at the point 𝛼(𝑡) is defined by the equality

(𝑡) = ∣𝛼′ (𝑡)∣ 𝑘 𝑔 (𝑡)𝝉2 (𝑡).

𝐷𝝉1
(4.12)
𝑑𝑡
So the geodesic curvature gives us the tangential component of the curvature of 𝛼,
and is also the generalization, for curves on oriented surfaces, of the signed curvature
for planar curves (1.3). Like the signed curvature, the geodesic curvature depends
only (up to sign) on the point of the curve at which it is computed and not on the
parameterization. In fact, if 𝛼(𝑡) and 𝛼(𝑠) are two parameterizations of the same
curve and v a vector field along 𝛼, then
𝐷v 𝑑𝑠 𝐷v
= , (4.13)
𝑑𝑡 𝑑𝑡 𝑑𝑠
∣𝛼′ (𝑡)∣ = ∣ ∣ ∣𝛼′ (𝑠)∣.
𝑑𝑠
(4.14)
𝑑𝑡
𝑑𝑠
Assuming that 𝛼(𝑡) and 𝛼(𝑠) have the same orientation (so that > 0), we apply
𝑑𝑡
(4.13) to the vector field 𝝉1 and use (4.12) and (4.14), obtaining 𝑘 𝑔 (𝑡) = 𝑘 𝑔 (𝑠), which
proves our claim. For purposes of computation, we note that from (4.12) it follows
that
1 𝐷𝝉1 1
𝑘𝑔 = ′ ⟨ , 𝝉2 ⟩ = ′ ⟨𝝉1 ′ , 𝑁 × 𝝉1 ⟩
∣𝛼 ∣ 𝑑𝑡 ∣𝛼 ∣
(4.15)
1
= ′ 3 ⟨𝛼′′ , 𝑁 × 𝛼′ ⟩,
∣𝛼 ∣
and formula (4.15) further simplifies to 𝑘 𝑔 = ⟨𝛼′′ , 𝑁 × 𝛼′ ⟩ when 𝛼 is parametrized by
arc length.
4.3 Covariant Derivative, Parallel Transport, Geodesic Curvature 97

From formula (4.15) it follows that the sign of the geodesic curvature is changed
when we reverse orientation of the curve or change the orientation of the surface. In
our next proposition we gather some simple properties of the geodesic curvature.

Proposition 4.3.6 (i) The geodesic curvature is invariant under local isometries that
preserve orientation — that is, if 𝑓 ∶ 𝑈 ⊆ 𝑆1 → 𝑆2 is such a mapping and 𝛼∶ 𝐼 → 𝑈 is
a regular curve, then the geodesic curvatures of 𝛼 in 𝑆1 (computed at 𝛼(𝑡)) and of
𝑓 ○ 𝛼 in 𝑆2 (computed at 𝑓 ○ 𝛼(𝑡)) are equal.
(ii) If 𝛼(𝑡) is a regular curve in 𝑆, then
2 2 2
𝑘(𝑡) = 𝑘 𝑔 (𝑡) + 𝑘 𝑛 (𝑡) ,

where 𝑘, 𝑘 𝑛 , 𝑘 𝑔 are respectively the curvature, the normal curvature and the geodesic
curvature of 𝛼.

Proof The derivative 𝐷 𝑓 transforms the “moving” orthonormal frame (𝝉1 (𝑡), 𝝉2 (𝑡))
that appears in the calculation (4.12) of the geodesic curvature of 𝛼 into another
“moving” orthonormal frame 𝝉1 (𝑡), 𝝉2 (𝑡). Furthermore, 𝝉1 is the unit tangent vector
field along the curve 𝑓 ○ 𝛼 and, if 𝑓 preserves orientation, this second frame is also
positively oriented. Statement (i) is then an immediate consequence of (4.10).
To prove (ii), it suffices to note that from the decomposition

𝝉1 ′ = + (𝝉1 ′ )
𝐷𝝉1 𝑛
𝑑𝑡
into the tangential and normal components, it follows that

𝐷𝝉1 2 𝑛 2
∣𝝉1 ′ ∣2 = ∣ ∣ + ∣(𝝉1 ′ ) ∣ ,
i.e. 𝑑𝑡
∣𝛼′ (𝑡)∣2 𝑘(𝑡) = ∣𝛼′ (𝑡)∣2 (𝑘 𝑔 (𝑡) + 𝑘 𝑛 (𝑡) ).
2 2 2
◻

Now let v be any vector field along 𝛼. We say that v is parallel along 𝛼 if its
covariant derivative along 𝛼 is constantly zero. Assume that 𝛼 is defined on [𝑎, 𝑏],
that w1 ∈ 𝑇𝛼(𝑎) 𝑆, and let v be a parallel vector field along 𝛼 such that v(𝑎) = w1 .
Then the vector w2 = v(𝑏) in 𝑇𝛼(𝑏) 𝑆 is called the parallel transport of w1 along 𝛼
from 𝛼(𝑎) to 𝛼(𝑏). To justify the use of the definite article the, we have to show that
there is only one parallel vector field along 𝛼 with initial position w1 ; this is done in
the next proposition. However, we note first that by formula (4.13) the fact that v is
parallel along 𝛼 does not depend on the parameterization of the curve. Therefore the
notion of parallel transport is also independent of the parameterization.

Proposition 4.3.7 Let 𝛼∶ [𝑎, 𝑏] → 𝑆 be a regular curve in 𝑆. Then:

(i) if v and w are parallel vector fields along 𝛼, the norms ∣v(𝑡)∣, ∣w(𝑡)∣ and the
angle between v(𝑡) and w(𝑡) are constant;
(ii) given w1 ∈ 𝑇𝛼(𝑎) 𝑆, there exists one and only one parallel vector field v(𝑡)
along 𝛼 such that v(𝑎) = w1 .
98 4 The Intrinsic Geometry of Surfaces

Proof (i) If v is a parallel vector field then by 4.3.5 we have

𝑑 2 𝐷v
∣v∣ = 2⟨ , v⟩ = 0,
𝑑𝑡 𝑑𝑡
and so ∣v∣ is constant, and the same holds for ∣w∣ and for the inner product ⟨v, w⟩.
Therefore each of these vector fields is either always or never zero. Assuming they are
⟨v, w⟩
both nonzero, the angle 𝜃(𝑡) = ∠(v(𝑡), w(𝑡)) is constant, since cos 𝜃(𝑡) = ⋅
∣v∣ ∣w∣
(ii) It follows from (i) that two parallel vector fields v and u such that v(𝑎) = u(𝑎)
must be identical, since they both have constant norm equal to ∣v(𝑎)∣, and the angle
between them is always zero. Thus there exists at most one parallel vector field v(𝑡)
such that v(𝑎) = w1 .
Assuming that w1 ≠ 0, it remains to prove the existence of v. For this, consider the
vector fields 𝝉1 (𝑡) = ∣𝛼′1(𝑡)∣ 𝛼′ (𝑡) and 𝝉2 = 𝑁 × 𝝉1 . We write v(𝑡) in the form

v(𝑡) = 𝑎(𝑡)𝝉1 (𝑡) + 𝑏(𝑡)𝝉2 (𝑡).

Since v is parallel, v has constant norm 𝑟 = ∣w1 ∣. We can thus try to find 𝜑(𝑡) such
that 𝑎(𝑡) = 𝑟 cos 𝜑(𝑡) and 𝑏(𝑡) = 𝑟 sin 𝜑(𝑡). Note that

, 𝝉2 ⟩ = −𝑘 𝑔 ∣𝛼′ ∣
𝐷𝝉2 𝐷𝝉1
⟨ , 𝝉1 ⟩ = −⟨
𝑑𝑡 𝑑𝑡

𝐷𝝉2
and therefore = −𝑘 𝑔 ∣𝛼′ ∣𝝉1 . Noting that the covariant derivative obeys the usual
𝑑𝑡
rules of differentiation, we then have
1 𝐷v 𝐷𝝉1 𝐷𝝉2
= −𝜑 sin 𝜑 𝝉1 + cos 𝜑 + 𝜑 cos 𝜑 𝝉2 + sin 𝜑
𝑟 𝑑𝑡 𝑑𝑡 𝑑𝑡
= (𝜑 + 𝑘 𝑔 ∣𝛼′ ∣)(− sin 𝜑 𝝉1 + cos 𝜑 𝝉2 ),

so that v is a parallel vector field along 𝛼 if and only if 𝜑′ (𝑡) = −𝑘 𝑔 (𝑡)∣𝛼′ (𝑡)∣. To
finish, it is therefore sufficient to define 𝜑(𝑡) = 𝜑0 − ∫𝑎 𝑘 𝑔 (𝑠)∣𝛼′ (𝑠)∣ 𝑑𝑠, where 𝜑0
𝑡

satisfies w1 = 𝑟 cos 𝜑0 𝝉1 (𝑎) + 𝑟 sin 𝜑0 𝝉2 (𝑎). ◻

Observations and Examples 4.3.8 A. Every local isometry 𝑓 sends a parallel vector
field along 𝛼 to a parallel vector field along 𝑓 ○ 𝛼. This is a consequence of formula
(4.10).
B. In the plane a vector field is parallel along a certain curve if and only if it is
constant along that curve, which shows that in this case the parallel transport only
depends on the initial and final points of the curve, not on the path covered. This
property is shared (locally) by those surfaces which are locally isometric to the plane,
such as the cylinder and the cone: thus, each point of these surfaces has an open
neighborhood 𝑈 such that the parallel transport along any curve contained in 𝑈
depends only on the initial and final points of the curve. Later on we will conclude
that it is only on surfaces of zero curvature that this holds true.
4.3 Covariant Derivative, Parallel Transport, Geodesic Curvature 99

C. If the surfaces 𝑆1 and 𝑆2 are tangent along the regular curve 𝛾(𝑡), the covariant
derivative of a vector field v along 𝛾 is the same computed with respect to 𝑆1 or 𝑆2
(since the normal component we subtract from v′ (𝑡) is the same in both cases), and
therefore also the parallel transport along 𝛾 is the same on the two surfaces. Using
this observation, let us determine the parallel transport on the sphere S2 along the
parallel of colatitude 𝜃 0 , which we denote by 𝛾.
Consider the cone of revolution ℭ tangent to the sphere along 𝛾. The angle that
the generatrices of ℭ make with the axis is 𝜋2 − 𝜃 0 , and therefore (see example 4.1.3
D) ℭ is isometric to a planar region 𝑈 that defines an angle 𝛽 = 2𝜋 cos 𝜃 0 (see Figure
4.3). By such an isometry, the parallel transport along 𝛾 corresponds to the parallel
transport in 𝑈 along the arc with center 𝑂 and radius equal to tg 𝜃 0 . However, in
𝑈 parallel transport is simply translation; but a vector w1 which is tangent to the
generatrix of the cone is translated to a vector w2 which makes an angle of 𝛽 with
that same generatrix (the generatrices of ℭ correspond in 𝑈 to the half-lines of origin
𝑂, and the two half-lines bounding 𝑈 are identified with the same generatrix).
In conclusion: parallel transport in S2 along one complete turn of the parallel
𝜃 = 𝜃 0 makes each vector rotate through an angle of 2𝜋 cos 𝜃 0 at the end of a complete
turn. ◻

We end this section by defining one of the most important concepts of Differential
Geometry: a geodesic of the surface 𝑆 is a regular curve 𝛼(𝑡) on 𝑆 whose geodesic
curvature is constantly zero. By

p -q
2 0

w1
U
w2

b q0
w1 b w2 b

Figure 4.3

(4.12), this is equivalent to the unit vector field 𝝉1 = ∣𝛼1′ ∣ 𝛼′ tangent to 𝛼 being parallel.
This simply means that at points where the curvature of 𝛼 is nonzero, the principal
normal to the curve is also normal to the surface.
We leave the in-depth study of geodesics to a later section; to finish this one, we
give some simple examples.
100 4 The Intrinsic Geometry of Surfaces

Examples 4.3.9 Geodesics on the plane are simply straight lines, since in the plane
geodesic curvature is the same as signed curvature. On the sphere any maximal circle
is a geodesic, because the principal normal to a maximal circle passes through the
center of the sphere, and therefore coincides with the normal to the sphere itself (later
on we will see that there are no other geodesics on the sphere). More generally, the
meridians of any surface of revolution are geodesics. ◻

Exercises
85. Consider on the sphere S2 the meridian 𝜑 = 0, which we denote by 𝛾, and the
points N = (0, 0, 1) and 𝑆 = (0, 0, −1). We define a mapping 𝐿 𝛾 ∶ 𝑇N S2 → 𝑇S S2 as
follows: 𝐿 𝛾 (v) is the parallel transport of the vector v from 𝑁 to S along 𝛾.
(a) Compute 𝐿 𝛾 (1, 0, 0) and 𝐿 𝛾 (0, 1, 0).
(b) Show that 𝐿 𝛾 is linear.
(c) Does 𝐿 𝛾 depend on the choice of meridian?
86. (a) Describe all the geodesics of the cylinder 𝑥 2 + 𝑦 2 = 1.
(b) Compare the geodesic curvatures of the same helix in the cylinder and in the
helicoid.
87. (a) Let 𝑆 = {𝛼(𝑡) + 𝜆v(𝑡)∶ 𝑇 ∈ 𝐼, 𝜆 ∈ R } be a ruled surface. Show that the
lines 𝑟 𝑡 (𝜆) = 𝛼(𝑡) + 𝜆v(𝑡) are all geodesics, and find out whether the curves
𝑐 𝜆 (𝑡) = 𝛼(𝑡) + 𝜆v(𝑡) are geodesics.
(b) Show that on any surface an asymptotic line that is also a geodesic is necessarily
a (segment of) straight line.
88. Show that on the surface of revolution

Φ(𝑢, 𝑣) = (𝜌(𝑣) cos 𝑢, 𝜌(𝑣) sin 𝑢, 𝑧(𝑣)),

the parallel 𝑣 = 𝑣 0 is a geodesic if and only if the tangent line to the generating curve
at 𝑣 0 is parallel to the axis of revolution.
√
89. (a) Check that in the cone C 𝑘 = {(𝑥, 𝑦, 𝑧)∶ 𝑧 ≠ 0, 𝑧 = 𝑘 𝑥 2 + 𝑦 2 } there is always
some geodesic connecting each pair of points.
√
(b) The generatrices 𝑡 ↦ 𝑡(𝑥 0 , 𝑦 0 , 𝑘 𝑥02 + 𝑦 20 ) are geodesics of C 𝑘 that are not
defined for 𝑡 ≤ 0. Show that any other geodesic of C 𝑘 parametrized with constant
scalar velocity extends to all values of the parameter.
90. Let E be the intersection of the cylinder 𝑥 2 + 𝑦 2 = 1 with a plane which passes
through the 𝑥-axis and makes an angle 𝜃 with the 𝑥𝑦 plane. Show that E is an ellipse,
and compute the absolute value of the geodesic curvature of E (relative to the cylinder)
at the points where E intersects its axes.
91. Show that the geodesic curvature of the curve 𝛼 at 𝑝 ∈ 𝑆 is equal to the curvature
at 𝑝 of the curve obtained by projecting 𝛼 orthogonally on 𝑝 + 𝑇 𝑝 𝑆.
4.4 The Divergence Theorem. First Variation of Area 101

4.4 The Divergence Theorem. First Variation of Area

In this section we give a simple expression for the divergence of a vector field and
prove the divergence theorem. We also prove a formula for the variation of the area
of a surface when it undergoes a perturbation induced by a vector field (and mention
how this is related to minimal surfaces). These results are used in Section 5.4 in
the proof of the Minkowski integral formulas (which in turn we use to establish the
rigidity of the sphere and in the proof of Theorem 5.5.3) and otherwise are not used in
the remainder of this book. This being said, we must acknowledge that some subjects
treated here do not pertain to intrinsic geometry, but we could find no better place to
fit them in.
Let v be a vector field on an open subset 𝑈 of an oriented surface 𝑆. In the previous
section we defined the linear mapping 𝐷v 𝑝 as follows: given w ∈ 𝑇 𝑝 𝑆, 𝐷v 𝑝 (w)
is the covariant derivative of v, computed at 𝑝, along any curve in 𝑆 that passes
through 𝑝 with velocity w. Using coordinates Φ(𝑢, 𝑣), the vectors 𝐷vΦ(𝑢,𝑣) (Φ𝑢 )
𝜕v 𝜕v
and 𝐷vΦ(𝑢,𝑣) (Φ𝑣 ) are then the tangential components of and respectively —
𝜕𝑢 𝜕𝑣
that is, if we write
𝜕v
= 𝑏 11 Φ𝑢 + 𝑏 21 Φ𝑣 + 𝜆1 𝑁
𝜕𝑢 (4.16)
𝜕v
= 𝑏 12 Φ𝑢 + 𝑏 22 Φ𝑣 + 𝜆2 𝑁,
𝜕𝑣
then the matrix of 𝐷vΦ(𝑢,𝑣) relative to the basis (Φ𝑢 , Φ𝑣 ) of 𝑇Φ(𝑢,𝑣) 𝑆 is

𝑏 11 𝑏 12
𝐵=[ ]⋅
𝑏 21 𝑏 22

Divergence is the trace of this matrix. That is,

Div v(Φ(𝑢, 𝑣)) = tr 𝐵 = 𝑏 11 + 𝑏 22 .

Our goal is to obtain an explicit formula for divergence. Using (4.16), we have

𝜕v 𝜕v
× Φ𝑣 + Φ𝑢 × = (𝑏 11 + 𝑏 22 )(Φ𝑢 × Φ𝑣 ) + 𝜆1 (𝑁 × Φ𝑣 ) + 𝜆2 (Φ𝑢 × 𝑁),
𝜕𝑢 𝜕𝑣
and therefore
𝜕v 𝜕v
⟨ × Φ𝑣 + Φ𝑢 × , 𝑁⟩ = (𝑏 11 + 𝑏 22 )⟨Φ𝑢 × Φ𝑣 , 𝑁⟩
𝜕𝑢 𝜕𝑣 (4.17)
√
= (Div v) 𝐸𝐺 − 𝐹 2 ,

provided that Φ(𝑢, 𝑣) is compatible with the orientation of 𝑆.

The expression (4.17) for Div v can still be simplified. It is convenient, for the
calculations below, to work with a vector field v which is not necessarily tangent to
𝑆. By letting v = 𝛼Φ𝑢 + 𝛽Φ𝑣 + 𝛾𝑁, and using the matrix
102 4 The Intrinsic Geometry of Surfaces

𝑎 11 𝑎 12
[ ]
𝑎 21 𝑎 22

of −𝐷𝑁Φ(𝑢,𝑣) relative to the basis (Φ𝑢 , Φ𝑣 ) to express the vectors 𝑁𝑢 and 𝑁 𝑣 (see
(3.6) in Section 3.2), we have

Φ𝑣 × 𝑁𝑢 + 𝑁 𝑣 × Φ𝑢 = −Φ𝑣 × (𝑎 11 Φ𝑢 + 𝑎 21 Φ𝑣 ) − (𝑎 12 Φ𝑢 + 𝑎 22 Φ𝑣 ) × Φ𝑢
= (𝑎 11 + 𝑎 22 )(Φ𝑢 × Φ𝑣 )
√
= 2𝐻 𝐸𝐺 − 𝐹 2 𝑁 (4.18)

⟨v, Φ𝑣 × 𝑁⟩ = ⟨𝛼Φ𝑢 + 𝛽Φ𝑣 + 𝛾𝑁, Φ𝑣 × 𝑁⟩

= ⟨𝛼Φ𝑢 , Φ𝑣 × 𝑁⟩ = 𝛼⟨Φ𝑢 × Φ𝑣 , 𝑁⟩
√
= 𝛼 𝐸𝐺 − 𝐹 2 (4.19)

√
⟨v, 𝑁 × Φ𝑢 ⟩ = 𝛽 𝐸𝐺 − 𝐹 2 (4.20)

Pursuing our calculations, we further have

𝜕v 𝜕v 𝜕v 𝜕v
⟨ × Φ𝑣 + Φ𝑢 × , 𝑁⟩ = ⟨ , Φ𝑣 × 𝑁⟩ + ⟨ , 𝑁 × Φ𝑢 ⟩
𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣
𝜕 𝜕
= ⟨ −v, (Φ𝑣 × 𝑁) + (𝑁 × Φ𝑢 )⟩ +
𝜕𝑢 𝜕𝑣
𝜕 𝜕
+ ⟨v, Φ𝑣 × 𝑁⟩ + ⟨v, 𝑁 × Φ𝑢 ⟩
𝜕𝑢 𝜕𝑣
= −⟨v, Φ𝑣 × 𝑁𝑢 + 𝑁 𝑣 × Φ𝑢 ⟩ +
𝜕 𝜕
+ ⟨v, Φ𝑣 × 𝑁⟩ + ⟨v, 𝑁 × Φ𝑢 ⟩
𝜕𝑢 𝜕𝑣
√
= − 𝐸𝐺 − 𝐹 2 ⟨v, 2𝐻𝑁⟩ +
𝜕 √ 𝜕 √
+ (𝛼 𝐸𝐺 − 𝐹 2 ) + (𝛽 𝐸𝐺 − 𝐹 2 ),
𝜕𝑢 𝜕𝑣
where we apply (4.18), (4.19), and (4.20). Of course, when the vector field v is tangent
to 𝑆, the first summand of the last expression vanishes. Combining this formula with
(4.17), we obtain the desired formula for divergence:

Proposition 4.4.1 Let v be a (not necessarily tangent) vector field on an open subset
𝑈 covered by a parameterization Φ(𝑢, 𝑣) compatible with the orientation of 𝑆. Then:
(i) if we denote by v⊺ the tangential component of v, then we have

1
, 𝑁⟩ = −⟨v, 2𝐻𝑁⟩ + Div(v⊺ );
𝜕v 𝜕v
√ ⟨ × Φ𝑣 + Φ𝑢 ×
𝐸𝐺 − 𝐹 2 𝜕𝑢 𝜕𝑣
4.4 The Divergence Theorem. First Variation of Area 103

(ii) if v is a tangent field to 𝑆 and we write v = 𝛼Φ𝑢 + 𝛽Φ𝑣 , then we have

1 𝜕 √ 𝜕 √
Div v = √ { (𝛼 𝐸𝐺 − 𝐹 2 ) + (𝛽 𝐸𝐺 − 𝐹 2 )} .
𝐸𝐺 − 𝐹 2 𝜕𝑢 𝜕𝑣

The divergence theorem reduces the calculation of a certain surface integral on

a simple region to a line integral along its boundary. By simple region (or Jordan
region) Ω ⊆ 𝑆 we mean a connected region whose boundary is a simple closed curve
and whose closure (in 𝑆) is homeomorphic to a closed disk (in the plane); if the
boundary of Ω is piecewise regular, then Ω is also called a polygonal region. We now
state the most important result in this section.

Divergence Theorem 4.4.2 Let v be a field of tangent vectors on 𝑆, and let Ω ⊆ 𝑆 be

a polygonal region. Then

∫ Div v 𝑑𝜎 = ∫ ⟨v, −𝝉2 ⟩ 𝑑𝑠, (4.21)

Ω 𝛾

where 𝛾(𝑠) is the boundary of Ω and 𝝉2 (𝑠) ∈ 𝑇𝛾(𝑠) 𝑆 is the unit vector orthogonal to
𝛾 ′ (𝑠) that points to the interior of Ω.

Proof We assume that 𝛾(𝑠) is parametrized by arc length and runs in the positive
direction — i.e., in such a way that Ω is always to the left of 𝛾 (in other words, we
have 𝝉2 (𝑠) = 𝑁(𝑠) × 𝛾 ′ (𝑠)) (see Fig. 4.4). Of course, 𝝉2 is not defined at the vertices
of 𝛾, but this happens only for a finite number of values of 𝑠, which do not affect the
integration.

It is sufficient to prove the theorem under the hypothesis that the closure of Ω is
contained in some parametrized neighborhood. For if this is not the case, we can
𝑘
decompose Ω into a finite number of sufficiently small polygonal regions (Ω𝑖 )𝑖=1 .
Supposing that (4.21) holds for the Ω𝑖 , and
104 4 The Intrinsic Geometry of Surfaces

t2
t1

Figure 4.4

denoting by 𝛾𝑖 the boundary of Ω𝑖 moving in the positive direction and by 𝝉2 𝑖 the

associated unit vector, we obtain
𝑘
∫ Div v 𝑑𝜎 = ∑ ∫ Div v 𝑑𝜎 =
Ω 𝑖=1 Ω𝑖
𝑘
= ∑ ∫ ⟨v, −𝝉2 𝑖 ⟩ 𝑑𝑠 = ∫ ⟨v, −𝝉2 ⟩ 𝑑𝑠
𝑖=1 𝛾𝑖 𝛾

– since any “edge” (segment of some 𝛾𝑖 ) inside Ω is run through twice, with the
integrand function ⟨v, −𝝉2 𝑖 ⟩ taking up opposite signs in both instances, and therefore
we are left with only the integrals relative to the edges that are part of 𝛾.
We can thus assume that Ω is covered by the parameterization Φ(𝑢, 𝑣), and we
put 𝛾(𝑠) = Φ(𝑢(𝑠), 𝑣(𝑠)) for every 𝑠 ∈ [0, 𝐿]. Furthermore, we require that Φ(𝑢, 𝑣)
be compatible with the orientation of 𝑆. We thus have, with the notation of 4.4.1 and
using Green’s theorem

𝜕 √ 𝜕 √
∫ Div v 𝑑𝜎 = ∬ { (𝛼 𝐸𝐺 − 𝐹 2 ) + (𝛽 𝐸𝐺 − 𝐹 2 )} 𝑑𝑢𝑑𝑣
Ω Φ−1 (Ω) 𝜕𝑢 𝜕𝑣
𝐿 √
=∫ 𝐸𝐺 − 𝐹 2 (𝛼(𝑠)𝑣 ′ (𝑠) − 𝛽(𝑠)𝑢 ′ (𝑠)) 𝑑𝑠.
0

(Note that, since Φ is compatible with the orientation of 𝑆, the curve (𝑢(𝑠), 𝑣(𝑠))
moves along the boundary of Φ−1 (Ω) in the positive direction, and we have used this
fact when we applied Green’s theorem.) To simplify the latter integral, we form the
vector product of the two equalities

v(𝑠) = 𝛼(𝑠)Φ𝑢 + 𝛽(𝑠)Φ𝑣 , 𝛾 ′ (𝑠) = 𝑢 ′ (𝑠)Φ𝑢 + 𝑣 ′ (𝑠)Φ𝑣 ,

4.4 The Divergence Theorem. First Variation of Area 105

obtaining

v(𝑠) × 𝛾 ′ (𝑠) = (𝛼(𝑠)𝑣 ′ (𝑠) − 𝛽(𝑠)𝑢 ′ (𝑠))(Φ𝑢 × Φ𝑣 )

√
= 𝐸𝐺 − 𝐹 2 (𝛼(𝑠)𝑣 ′ (𝑠) − 𝛽(𝑠)𝑢 ′ (𝑠))𝑁.

The above integral can then be rewritten as

𝐿 𝐿
∫ ⟨v(𝑠) × 𝛾 ′ (𝑠), 𝑁⟩ 𝑑𝑠 = ∫ ⟨v(𝑠), 𝛾 ′ (𝑠) × 𝑁⟩ 𝑑𝑠
0 0
𝐿
=∫ ⟨v(𝑠), −𝝉2 (𝑠)⟩ 𝑑𝑠 = ∫ ⟨v, −𝝉2 ⟩ 𝑑𝑠. ◻
0 𝛾

The divergence theorem can be generalized to regions bounded by more than one
closed curve, making the sum of all integrals relative to each curve constituting the
boundary of Ω appear in the right-hand side of (4.21). However, it is more interesting,
in the case of a compact surface 𝑆, to consider the integral of Div v over the whole
𝑘
surface. Choosing a “polygonal decomposition” (Ω𝑖 )𝑖=1 of 𝑆, and applying 4.4.2 to
each of the Ω𝑖 , what happens, when adding up the results, is that all the terms vanish,
because now none of the edges is run through only once. Thus we have just proved (i)
of the corollary below.

Corollary 4.4.3 Let v be a vector field on a compact surface 𝑆. Then:

(i) ∫𝑆 Div v 𝑑𝜎 = 0;

(ii) there exists 𝑝 ∈ 𝑆 such that Div v(𝑝) = 0.

To prove (ii) we can assume that 𝑆 is connected, because in any case its connected
components are compact. Hence, if Div v has no zeros, then it has constant sign and
therefore ∫𝑆 Div v 𝑑𝜎 ≠ 0, in contradiction to (i). Therefore Div v has some zero,
which proves (ii).

Given a differentiable vector field v, not necessarily tangent, on a compact surface

𝑆, let 𝑆𝑡 = {𝑝 + 𝑡v(𝑝)∶ 𝑝 ∈ 𝑆}. The family (𝑆𝑡 )𝑡 is called a variation of 𝑆 (when v
is a normal field to 𝑆, not necessarily unitary, (𝑆𝑡 )𝑡 is called a normal variation of
𝑆); see Fig. 4.5. Denoting by 𝐴(𝑡) the area of 𝑆𝑡 , our goal is to prove a formula for
𝐴′ (0) as an integral over 𝑆. First, however, we must guarantee that 𝑆𝑡 is a surface for
sufficiently small 𝑡, which we do in the proposition below.
106 4 The Intrinsic Geometry of Surfaces

Figure 4.5

Proposition 4.4.4 There exists 𝜂 > 0 such that, for all 𝑡 ∈ ]−𝜂, 𝜂[, 𝑆𝑡 is a compact
surface.

Proof We define the mapping 𝐿 𝑡 ∶ 𝑆 → R 3 by 𝐿 𝑡 (𝑝) = 𝑝 + 𝑡v(𝑝), and so 𝑆𝑡 is the

image of 𝑆 under 𝐿 𝑡 . Using this notation, let us start by proving two auxiliary claims:
Claim 1. For every point 𝑝 of 𝑆, there exist a parametrized neighborhood Φ(𝑢, 𝑣) of
𝑝 and 𝜀 > 0 such that, for all 𝑡 ∈ ]−𝜀, 𝜀[, the mapping

𝐿 𝑡 ○ Φ(𝑢, 𝑣) = Φ(𝑢, 𝑣) + 𝑡v(𝑢, 𝑣)

is injective and its Jacobian matrix has rank two.

Taking a parameterization Φ(𝑢, 𝑣) = (𝑥(𝑢, 𝑣), 𝑦(𝑢, 𝑣), 𝑧(𝑢, 𝑣)) with Φ(0, 0) = 𝑝,
we can assume that
𝜕(𝑥, 𝑦)
∣ ≠ 0.
𝜕(𝑢, 𝑣) (0,0)
Denoting by 𝑥 𝑡 , 𝑦 𝑡 , 𝑧 𝑡 the component functions of 𝐿 𝑡 ○ Φ, we define

𝐹(𝑢, 𝑣, 𝑡) = (𝑥 𝑡 (𝑢, 𝑣), 𝑦 𝑡 (𝑢, 𝑣), 𝑡) = (𝐺 𝑡 (𝑢, 𝑣), 𝑡).

It is easily seen that det(𝐽𝐹(𝑢,𝑣,𝑡) ) = det(𝐹(𝐺 𝑡 )(𝑢,𝑣) ). In particular, we have

𝜕(𝑥, 𝑦)
det(𝐽𝐹(0,0,0) ) = det(𝐽(𝐺 0 )(0,0) ) = ∣ ≠ 0.
𝜕(𝑢, 𝑣) (0,0)

Hence, by the inverse mapping theorem, there exists an open neighborhood 𝑈 × ]−𝜀, 𝜀[
of (0, 0, 0) such that the restriction of 𝐹 to this neighborhood is a diffeomorphism
−1
onto its image. The local inverse of 𝐹 has the form 𝐹 −1 (𝑥, 𝑦, 𝑡) = ((𝐺 𝑡 ) (𝑥, 𝑦), 𝑡),
and so, for every 𝑡 ∈ ]−𝜀, 𝜀[, the mapping 𝐺 𝑡 ∣𝑈 has a differentiable inverse, which
guarantees that it is injective and its Jacobian matrix has rank two. Since 𝐿 𝑡 ○Φ(𝑢, 𝑣) =
4.4 The Divergence Theorem. First Variation of Area 107

(𝐺 𝑡 (𝑢, 𝑣), 𝑧 𝑡 (𝑢, 𝑣)), the same holds for 𝐿 𝑡 ○ Φ∣𝑈 . The parameterization (𝑈, Φ) thus
satisfies the conditions of Claim 1.
Claim 2. For ∣𝑡∣ sufficiently small, 𝐿 𝑡 is injective.
By compactness of 𝑆, we can find 𝜀 > 0 and parameterizations (𝑈𝑖 , Φ𝑖 ), with
𝑖 = 1, . . . , 𝑘, which cover 𝑆 and such that each of them satisfies Claim 1 for the given 𝜀.
Next, we take 𝛿 > 0 such that, for every 𝑝 ∈ 𝑆, the ball 𝐵 𝛿 (𝑝, 𝑆) = {𝑞 ∈ 𝑆∶ ∣𝑞 − 𝑝∣ < 𝛿}
is contained in one of the open subsets Φ𝑖 (𝑈𝑖 ), and we put 𝑀 = max ∣v(𝑝)∣ and
𝑝∈𝑆
𝜂 = min { 4𝑀𝛿
, 𝜀}}. We claim that 𝐿 𝑡 is injective for ∣𝑡∣ < 𝜂. In fact, if ∣𝑞 − 𝑝∣ < 𝛿 it
is clear that 𝐿 𝑡 (𝑝) ≠ 𝐿 𝑡 (𝑞), since in this case 𝑝 and 𝑞 both belong to some Φ𝑖 (𝑈𝑖 )
and, by Claim 1, the restriction of 𝐿 𝑡 to Φ𝑖 (𝑈𝑖 ) is injective. If ∣𝑞 − 𝑝∣ ≥ 𝛿 then

∣𝐿 𝑡 (𝑝) − 𝐿 𝑡 (𝑞)∣ = ∣(𝑝 − 𝑞) − 𝑡(v(𝑝) − v(𝑞))∣

≥ ∣𝑝 − 𝑞∣ − ∣𝑡∣(∣v(𝑝)∣ + ∣v(𝑞)∣)
𝛿
≥ 𝛿 − 2𝜂𝑀 ≥ ,
2
which concludes the proof of Claim 2.
Finally we show that 𝑆𝑡 is a surface when ∣𝑡∣ < 𝜂. Since 𝐿 𝑡 is continuous and
injective and 𝑆 is compact, 𝐿 𝑡 is a homeomorphism 𝑆 → 𝑆𝑡 (in particular, 𝑆𝑡 is also
compact). Now, since the images of the mappings 𝐿 𝑡 ○ Φ𝑖 , 𝑖 = 1, . . . , 𝑘, cover 𝑆𝑡 , it
suffices to show that each of them is a parameterization. By Claim 1, the Jacobian
−1
of 𝐿 𝑡 ○ Φ𝑖 has rank two; furthermore, (𝐿 𝑡 ○ Φ𝑖 ) is continuous since it is given by
−1
the composite (Φ𝑖 ) ○ 𝐿 −1 𝑡 of continuous functions. This shows that 𝐿 𝑡 ○ Φ is a
𝑖

parameterization and concludes the proof of the proposition. ◻

It deserves mention that the mapping 𝐿 𝑡 is a diffeomorphism 𝑆 → 𝑆𝑡 . In fact, and
as we have seen, it is a homeomorphism. Its expression in local coordinates (𝑈𝑖 , Φ𝑖 )
on 𝑆 and (𝑈𝑖 , 𝐿 𝑡 ○ Φ𝑖 ) on 𝑆𝑡 is clearly a diffeomorphism, since it is just the identity
𝑈𝑖 → 𝑈𝑖 . This means that all surfaces of the variation (𝑆𝑡 )−𝜂<𝑡<𝜂 are diffeomorphic
to 𝑆.
Recalling that 𝐴(𝑡) denotes the area of the surface 𝑆𝑡 , let us now present the
formula for 𝐴′ (0).
Theorem 4.4.5 (First variation of area). 𝐴′ (0) = − ∫𝑆 ⟨v, 2𝐻𝑁⟩ 𝑑𝜎.

Proof We can consider a polygonal decomposition (Ω 𝑗 )1≤ 𝑗≤𝑟 of 𝑆 such that the
closure of each Ω 𝑗 is contained in the image of one of the parameterizations (𝑈𝑖 , Φ𝑖 )
defined in the proof of 4.4.4. Thus, for ∣𝑡∣ < 𝜂, the sets Ω𝑡𝑗 = 𝐿 𝑡 (Ω 𝑗 ) form a
decomposition of 𝑆𝑡 and each of them is covered by some parameterization

𝐿 𝑡 ○ Φ𝑖 (𝑢, 𝑣) = Φ𝑖 (𝑢, 𝑣) + 𝑡v(𝑢, 𝑣).

−1
We fix 𝑗 and the corresponding 𝑖, and we let 𝑊 𝑗 = (Φ𝑖 ) (Ω 𝑗 ). The area of Ω𝑡𝑗 is
given by
𝐴 𝑗 (𝑡) = ∬ ∣(𝐿 𝑡 ○ Φ𝑖 )𝑢 × (𝐿 𝑡 ○ 𝜙𝑖 )𝑣 ∣ 𝑑𝑢𝑑𝑣
𝑊𝑗
108 4 The Intrinsic Geometry of Surfaces

and therefore

𝐴′𝑗 (0) = ∬
𝜕
∣(𝐿 𝑡 ○ Φ𝑖 )𝑢 × (𝐿 𝑡 ○ 𝜙𝑖 )𝑣 ∣ 𝑑𝑢𝑑𝑣.
𝑊𝑗 𝜕𝑡 ∣𝑡=0

Now we have
𝜕v 𝜕v 𝜕v 𝜕v
∣(𝐿 𝑡 ○ Φ𝑖 )𝑢 × (𝐿 𝑡 ○ Φ𝑖 )𝑣 ∣ = ∣(Φ𝑖𝑢 × Φ𝑖𝑣 ) + 𝑡 ( × Φ𝑖𝑣 + Φ𝑖𝑢 × ) + 𝑡2 ( × )∣ .
𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣
Therefore, using 4.4.1 (i), we have

𝜕 ⟨ 𝜕 ((𝐿 𝑡 ○ 𝜙𝑖 )𝑢 × (𝐿 𝑡 ○ Φ𝑖 )𝑣 )∣𝑡=0 , Φ𝑖𝑢 × Φ𝑖𝑣 ⟩

∣(𝐿 𝑡 ○ Φ𝑖 )𝑢 × (𝐿 𝑡 ○ Φ𝑖 )𝑣 ∣∣𝑡=0 = 𝜕𝑡
𝜕𝑡 ∣Φ𝑖𝑢 × Φ𝑖𝑣 ∣
𝜕v 𝜕v
=⟨ × Φ𝑖𝑣 + Φ𝑖𝑢 × , 𝑁⟩
𝜕𝑢 𝜕𝑣
= {−⟨v, 2𝐻𝑁⟩ + Div(v⊺ )}∣Φ𝑖𝑢 × Φ𝑖𝑣 ∣,

where v⊺ denotes the tangential component of v. We thus conclude that

𝐴′𝑗 (0) = ∫ {−⟨v, 2𝐻𝑁⟩ + Div(v⊺ )} 𝑑𝜎

Ω𝑖

– and, using 4.4.3 (i), we finally obtain

𝑟
𝐴′ (0) = ∑ 𝐴′𝑗 (0) = ∫ {−⟨v, 2𝐻𝑁⟩ + Div(v⊺ )} 𝑑𝜎
𝑗=1 𝑆

= − ∫ ⟨v, 2𝐻𝑁⟩ 𝑑𝜎. ◻.

𝑆

Observation 4.4.6 If 𝑆 is not compact but the vector field v has compact support (the
support of v is the closure in 𝑆 of the set {𝑝 ∈ 𝑆∶ v(𝑝) ≠ 0}), then the sets 𝑆𝑡 are
still surfaces for ∣𝑡∣ small and, suitably interpreted, formula 4.4.5 remains valid. We
choose a compact region Ω ⊆ 𝑆 whose boundary is made up of a finite number of
piecewise regular closed curves and whose closure contains the support of v, and
we take a polygonal decomposition (Ω 𝑗 )1≤ 𝑗≤𝑟 of Ω. We denote by 𝐴(𝑡) the area of
𝐿 𝑡 (Ω) ⊆ 𝑆𝑡 . Proceeding as in 4.4.5, and using the divergence theorem, we obtain

𝐴′ (0) = − ∫ ⟨v, 2𝐻𝑁⟩ 𝑑𝜎 + ∫ ⟨v⊺ , −𝝉2 ⟩ 𝑑𝑠

Ω 𝜕Ω

= − ∫ ⟨v, 2𝐻𝑁⟩ 𝑑𝜎,

𝑆

where 𝜕Ω indicates the boundary of Ω (restricted to which v⊺ vanishes).

In particular, if 𝐻 ≡ 0 then 𝐴′ (0) = 0. A minimal surface is a surface whose mean
curvature 𝐻 is constantly zero (by Exercise 57 in Section 3.1, no such surface can be
compact). This name arose as follows: assume that 𝑆 has the property that, for every
4.4 The Divergence Theorem. First Variation of Area 109

region Ω ⊆ 𝑆 as above, the area of Ω is less than or equal to that of any other portion
̃ of the surface such that 𝜕 Ω
Ω ̃ = 𝜕Ω. Then 𝑆 is a minimal surface: indeed, denoting
by 𝐴(𝑡) the area of 𝐿 𝑡 (Ω) (for a variation with support contained in Ω), 𝐴(𝑡) has
a minimum at 0 and therefore 𝐴′ (0) = 0. This implies that for every vector field v
with compact support we have − ∫ ⟨v, 2𝐻𝑁⟩ 𝑑𝜎 = 0 — which is only possible with
𝑆
𝐻 ≡ 0.
Not all minimal surfaces minimize area in the sense just stated, but the name has
stuck. Besides the plane, the reader can verify (using formula (3.8) in the Section 3.2)
that the helicoid and the catenoid are minimal surfaces (Exercise 76). In the exercises
of this section we prove that besides the plane the catenoid is the only minimal surface
of revolution.
For more examples and an in-depth study of minimal surfaces we recommend
Osserman’s book [19].

Exercises
92. (a) Compute the divergence of the vector field v(𝜑, 𝜃) in S2 of example 4.3.1. For
that example, directly compute ∫S2 Div v 𝑑𝜎. Could you use corollary 4.4.3?
(b) Let 𝑆 be a compact surface, let 𝐹 ⊆ 𝑆 be a finite set, and let v be a field of
tangent vectors defined on 𝑆 ∖ 𝐹 such that the set sup {∣v(𝑝)∣∶ 𝑝 ∈ 𝑆 ∖ 𝐹} is bounded.
Show that ∫𝑆 Div v 𝑑𝜎 = 0.
93. If v is a tangent field to 𝑆 and 𝑓 ∶ 𝑆 → R a differentiable function, then

Div( 𝑓 v)(𝑝) = 𝐷 𝑓 𝑝 (v) + 𝑓 (𝑝) Div v(𝑝).

94. Given a differentiable function 𝑓 ∶ 𝑆 → R , the gradient of 𝑓 is the tangent

vector field ∇ 𝑓 defined on 𝑆 as follows: for each 𝑝 ∈ 𝑆 and all v ∈ 𝑇 𝑝 𝑆, we have
𝐷 𝑓 𝑝 (v) = ⟨∇ 𝑓 (𝑝), v⟩. Equivalently, ∇ 𝑓 is the gradient of 𝑓 if we have 𝛼∶ 𝐼 → 𝑆 for
every differentiable curve ( 𝑓 ○ 𝛼)′ (𝑡) = ⟨∇ 𝑓 (𝛼(𝑡)), 𝛼′ (𝑡)⟩.
(a) Show that in local coordinates Φ(𝑢, 𝑣) we have

1 𝜕𝑓 𝜕𝑓
∇𝑓 = √ 𝑁 ×(
Φ𝑢 − Φ𝑣 )
𝐸𝐺 − 𝐹2 𝜕𝑣 𝜕𝑢
1 𝜕𝑓 𝜕𝑓 𝜕𝑓 𝜕𝑓
= {(− 𝐹 + 𝐺) Φ𝑢 + (− 𝐸 − 𝐹) Φ𝑣 } .
𝐸𝐺 − 𝐹 2 𝜕𝑣 𝜕𝑢 𝜕𝑣 𝜕𝑢
(b) Conclude that the vector field w = 𝑁 × ∇ 𝑓 has zero divergence. Conversely,
show that if Div w ≡ 0 then v = w × 𝑁 is locally a gradient field (i.e., each point of 𝑆
has a neighborhood 𝑈 such that v∣𝑈 is the gradient of some function 𝑈 → R ).
(c) Show that a vector field v on 𝑆 is a gradient field if and only if, for every
curve piecewise differentiable 𝛼∶ [𝑎, 𝑏] → 𝑆, the line integral ∫𝑎 ⟨v ○ 𝛼(𝑡), 𝛼′ (𝑡)⟩ 𝑑𝑡
𝑏

depends only on the initial and final points of 𝛼.

110 4 The Intrinsic Geometry of Surfaces

95. Let Φ(𝑢, 𝑠) = (𝜌(𝑠) cos 𝑢, 𝜌(𝑠) sin 𝑢, 𝑧(𝑠)) be a parameterization of a surface of
revolution 𝑆, where 𝑠 is the arc length of the generating curve (i.e., 𝜌 2 + 𝑧 2 ≡ 1) and
𝜌(𝑠) > 0. Assume that 𝑆 is not a plane (so that 𝑧 is not constantly zero). Show that:
(a) 𝑆 is a minimal surface if and only if
𝑧
= 𝜌𝑧 − 𝜌𝑧 (apply 3.2.1)
𝜌

(b) assuming that 𝑧(𝑠0 ) ≠ 0, there exists 𝜀 > 0 such that, for 𝑠 ∈ ]𝑠0 − 𝜀, 𝑠0 + 𝜀[,
the above equation is equivalent to

𝑑2 1 2
𝜌𝜌 + 𝜌 2 = 1 ⇔ ( 𝜌 )=1
𝑑𝑠2 2
√
⇔ ∃ 𝐴, 𝐵 ∈ R ∶ 𝜌(𝑠) = 𝑠2 + 𝐴𝑠 + 𝐵;

guarantee that there exist 𝑎 > 0 and a

(c) by shifting the domain of 𝑠, we can √
non-empty open interval 𝐼 such that 𝜌(𝑠) = 𝑠2 + 𝑎 2 for all 𝑠 ∈ 𝐼;
√
(d) 𝜌(𝑠) = 𝑠2 + 𝑎 2 for all 𝑠 ∈ R ;
(e) assuming that 𝑧(0) = 0 (if not, apply a vertical translation to 𝑆) and that 𝑧 > 0,
𝑠
𝑑𝑡 = 𝑎 sinh−1 ( ) ;
𝑎 𝑠
𝑧(𝑠) = ∫ √
0 𝑡 +𝑎
2 2 𝑎

= sinh−1 ( ), we have 𝜌(𝑣) = 𝑎 cosh 𝑣, and therefore 𝑆 is the

𝑧 𝑠
(f) by letting 𝑣 =
𝑎 𝑎
catenoid.

4.5 The Gauss-Bonnet Theorem

The Gauss-Bonnet Theorem is one of the deepest results in the Differential Geometry
of surfaces establishing an unexpected connection between the Euler characteristic
of a compact surface (a purely topological concept) and its Gaussian curvature.
Moreover, it provides a general context for a seemingly rather particular result as
Girard’s formula for spherical triangles (example 2.5.2).
We work with an oriented surface 𝑆. Let Ω be a polygonal region (as defined
in Section 4.4) and 𝛼(𝑠), 𝑠 ∈ [0, 𝐿], a parameterization of 𝜕Ω by arc length with
positive orientation. Let 0 = 𝑠0 < 𝑠1 < ⋯ < 𝑠 𝑘 = 𝐿 be the 𝑘 vertices of 𝛼; 𝛾𝑖 ∈ [−𝜋, 𝜋],
for 𝑖 = 1, . . . , 𝑘 − 1, the oriented angle of 𝛼′ (𝑠𝑖− ) and 𝛼′ (𝑠𝑖+ ); and 𝛾 𝑘 ∈ [−𝜋, 𝜋] the
angle between 𝛼′ (𝑠−𝑘 ) and 𝛼′ (𝑠0+ ). These angles 𝛾𝑖 are called the exterior angles of
Ω.
4.5 The Gauss-Bonnet Theorem 111

p
gj
- W
-p
W
gi
a´ (si+ ) +
a´ (s-i )
W

Figure 4.6

(Note: if 𝛼′ (𝑠𝑖+ ) = −𝛼′ (𝑠𝑖− ), we have to decide which of the values 𝜋 or −𝜋 we choose
for 𝛾𝑖 . We take 𝛾𝑖 = 𝜋 when, for 𝑠 → 𝑠𝑖− , the trace of 𝛼∣[𝑠 ,𝑠 +𝜀] (for small 𝜀 > 0) stays,
𝑖 𝑖
like Ω, to the left of the curve; otherwise, we take 𝛾𝑖 = −𝜋. We can make this criterion
rigorous with the help of local coordinates, but Fig. 4.6 is more illuminating).
Let us state the first version of our theorem:

Gauss-Bonnet Theorem 4.5.1 (Local version). Let Ω ⊆ 𝑆 be a polygonal region

whose closure is covered by some parameterization. Let 𝛾𝑖 (𝑖 = 1, . . . , 𝑘) be the
exterior angles of Ω. Then
𝑘
∫ 𝐾 𝑑𝜎 + ∫ 𝑘 𝑔 (𝑠) 𝑑𝑠 + ∑ 𝛾𝑖 = 2𝜋.
Ω 𝜕Ω 𝑖=1

Let Φ(𝑢, 𝑣) be a parameterization compatible with the orientation of 𝑆 containing

the closure of Ω. Consider the unit tangent field v1 = √1 Φ𝑢 and define, for each
𝐸
interval [𝑠𝑖−1 , 𝑠𝑖 ], a continuous choice 𝜃 𝑖 (𝑠) of the oriented angle between v1 and
𝛼′ (𝑠). (See the Appendix to this chapter for the existence of such continuous choices.)
The Rotation Index theorem (Theorem A.4 of the Appendix) states that
𝑘 𝑘
∑{𝜃 𝑖 (𝑠𝑖 ) − 𝜃 𝑖 (𝑠𝑖−1 )} + ∑ 𝛾𝑖 = 2𝜋. (4.22)
𝑖=1 𝑖=1

Let w(𝑠), 𝑠 ∈ [0, 𝐿], be a unit vector field along 𝛼: the vector field w is therefore
continuous, and is parallel along each regular arc 𝛼([𝑠𝑖−1 , 𝑠𝑖 ]). Let 𝜑𝑖 (𝑠) (𝑠 ∈
[𝑠𝑖−1 , 𝑠𝑖 ]) be a continuous choice of the angle between 𝛼′ (𝑠) and w(𝑠). The proof of
Proposition 4.3.7 (ii) shows that

𝜑′𝑖 (𝑠) = −𝑘 𝑔 (𝑠). (4.23)

112 4 The Intrinsic Geometry of Surfaces

Further, let 𝜓(𝑠), 𝑠 ∈ [0, 𝐿], be a choice of the angle between v1 and w(𝑠). By the
equality between oriented angles

∠(v1 , 𝛼′ (𝑠)) = ∠(v1 , w(𝑠)) + ∠(w(𝑠), 𝛼′ (𝑠))

= ∠(v1 , w(𝑠)) − ∠(𝛼′ (𝑠), w(𝑠)),

we see that we can assume, for 𝑖 = 1, . . . , 𝑘, that

𝜃 𝑖 (𝑠) = 𝜓(𝑠) − 𝜑𝑖 (𝑠) for 𝑠 ∈ [𝑠𝑖−1 , 𝑠𝑖 ]. (4.24)

In view of (4.24), equality (4.22) can then be rewritten as

𝑘 𝑘
𝜓(𝐿) − 𝜓(0) − ∑{𝜑𝑖 (𝑠𝑖 ) − 𝜑𝑖 (𝑠𝑖−1 )} + ∑ 𝛾𝑖 = 2𝜋,
𝑖=1 𝑖=1

or, using (4.23),

𝐿 𝑘
𝜓(𝐿) − 𝜓(0) + ∫ 𝑘 𝑔 (𝑠) 𝑑𝑠 + ∑ 𝛾𝑖 = 2𝜋.
0 𝑖=1

In view of this formula, the next lemma concludes the proof of the Gauss-Bonnet
theorem.

Lemma 4.5.2
𝜓(𝐿) − 𝜓(0) = ∫ 𝐾 𝑑𝜎. (4.25)
Ω

Proof Let v2 = 𝑁 ×v1 , so that the pair (v1 , v2 ) forms at each point a direct orthonormal
𝐷v1
frame. Since v1 is a unit vector field, is orthogonal to v1 , and is therefore collinear
𝑑𝑠
𝐷v1 𝐷v1
with v2 . We thus have = 𝑎(𝑠)v2 , where 𝑎(𝑠) = ⟨ , v2 ⟩ = ⟨v1 ′ , v2 ⟩ − ⟨v1 , v2 ′ ⟩,
𝑑𝑠 𝑑𝑠
from which we conclude that

= ⟨v1 ′ , v2 ⟩v2 , = −⟨v1 ′ , v2 ⟩v1 .

𝐷v1 𝐷v2
(4.26)
𝑑𝑠 𝑑𝑠
From the equality
w(𝑠) = cos 𝜓(𝑠)v1 + sin 𝜓(𝑠)v2
we obtain, by taking the covariant derivative of both sides and using (4.26),

= (𝜓 ′ (𝑠) + ⟨v1 ′ v2 ⟩)(− sin 𝜓(𝑠)v1 + cos 𝜓(𝑠)v2 ).

𝐷w
𝑑𝑠
𝐷w
But since w is parallel, we have ≡ 0 and therefore
𝑑𝑠
𝜓 ′ (𝑠) = −⟨v1 ′ (𝑠), v2 (𝑠)⟩.
4.5 The Gauss-Bonnet Theorem 113

Hence
𝐿
𝜓(𝐿) − 𝜓(0) = − ∫ ⟨v1 ′ (𝑠), v2 (𝑠)⟩ 𝑑𝑠
0
𝐿
, v2 ⟩𝑢 ′ (𝑠) + ⟨ , v2 ⟩𝑣 ′ (𝑠)) 𝑑𝑠
𝜕v1 𝜕v1
= −∫ (⟨
0 𝜕𝑢 𝜕𝑣
𝜕 𝜕v1 𝜕 𝜕v1
= −∬ { ⟨ , v2 ⟩ − ⟨ , v2 ⟩} 𝑑𝑢𝑑𝑣
Φ−1 (Ω) 𝜕𝑢 𝜕𝑣 𝜕𝑣 𝜕𝑢
(by Green’s theorem)
𝜕v1 𝜕v2 𝜕v1 𝜕v2
=∬ {⟨ , ⟩−⟨ , ⟩} 𝑑𝑢𝑑𝑣.
Φ−1 (Ω) 𝜕𝑢 𝜕𝑣 𝜕𝑣 𝜕𝑢

To complete the proof of the lemma, it is now sufficient to show that

𝜕v1 𝜕v2 𝜕v1 𝜕v2 √

⟨ , ⟩−⟨ , ⟩ = 𝐾 𝐸𝐺 − 𝐹 2 . (4.27)
𝜕𝑢 𝜕𝑣 𝜕𝑣 𝜕𝑢

We begin by observing that from the equalities

𝜕v1 𝜕v2
𝑁𝑢 = ⟨𝑁𝑢 , v1 ⟩v1 + ⟨𝑁𝑢 , v2 ⟩v2 = −⟨𝑁, ⟩v1 − ⟨𝑁, ⟩v2 ,
𝜕𝑢 𝜕𝑢
𝜕v1 𝜕v2
𝑁 𝑣 = −⟨𝑁, ⟩v1 − ⟨𝑁, ⟩v2 ,
𝜕𝑣 𝜕𝑣

it ensures, taking into account the identity v1 × v2 = 𝑁, that

𝜕v1 𝜕v2 𝜕v1 𝜕v2

𝑁𝑢 × 𝑁 𝑣 = {⟨𝑁, ⟩⟨𝑁, ⟩ − ⟨𝑁, ⟩⟨𝑁, ⟩} 𝑁. (4.28)
𝜕𝑢 𝜕𝑣 𝜕𝑣 𝜕𝑢

If we write
𝜕v1 𝜕v1 𝜕v2 𝜕v2
= 𝑎v2 + ⟨𝑁, ⟩𝑁, = 𝑏v1 + ⟨𝑁, ⟩𝑁,
𝜕𝑢 𝜕𝑢 𝜕𝑣 𝜕𝑣

and form the scalar product, then we obtain

𝜕v1 𝜕v2 𝜕v1 𝜕v2

⟨ , ⟩ = ⟨𝑁, ⟩⟨𝑁, ⟩,
𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣

and analogously
𝜕v1 𝜕v2 𝜕v1 𝜕v2
⟨ , ⟩ = ⟨𝑁, ⟩⟨𝑁, ⟩⋅
𝜕𝑣 𝜕𝑢 𝜕𝑣 𝜕𝑢
Together with (4.28), the last two equalities give
114 4 The Intrinsic Geometry of Surfaces

𝜕v1 𝜕v2 𝜕v1 𝜕v2

𝑁𝑢 × 𝑁 𝑣 = {⟨ , ⟩−⟨ , ⟩} 𝑁,
𝜕𝑢 𝜕𝑣 𝜕𝑣 𝜕𝑢

and therefore
𝜕v1 𝜕v2 𝜕v1 𝜕v2
⟨ , ⟩−⟨ , ⟩ = ⟨𝑁𝑢 × 𝑁 𝑣 , 𝑁⟩
𝜕𝑢 𝜕𝑣 𝜕𝑣 𝜕𝑢
√
= 𝐾⟨Φ𝑢 × Φ𝑣 , 𝑁⟩ = 𝐾 𝐸𝐺 − 𝐹 2 .

This concludes the proof of (4.27) and hence that of the lemma. ◻

Observations 4.5.3 A. The difference 𝜓(𝐿) − 𝜓(0) in Lemma 4.5.2 is the angle
between the initial and final positions of a vector that is carried in parallel along a
closed curve. If 𝐾 has constant (nonzero) sign on Ω, Lemma 4.5.2 shows that this
angle is nonzero and becomes smaller as the region bounded by the curve becomes
smaller. It follows that a surface has constant zero curvature if parallel transport along
any curve on the surface depends only on the starting and ending points of the curve
(and not on the path t covered). The reader is invited to elaborate the argument in the
exercises in this section.

B. The formula (4.27) can be rewritten in the form

1 𝜕 𝜕v2 𝜕 𝜕v2
𝐾=√ { ⟨v1 , ⟩− ⟨v1 , ⟩} (4.29)
𝐸𝐺 − 𝐹2 𝜕𝑢 𝜕𝑣 𝜕𝑣 𝜕𝑢
1 𝜕 𝐷v2 𝜕 𝐷v2
=√ { ⟨v1 , ⟩− ⟨v1 , ⟩} , (4.30)
𝐸𝐺 − 𝐹2 𝜕𝑢 𝜕𝑣 𝜕𝑣 𝜕𝑢

and formula (4.30) gives another proof of Gauss’s Theorema Egregium, for it
expresses 𝐾 via intrinsic quantities: in fact, any isometry transforms (v1 , v2 ) into
another orthonormal dihedron and, by (4.10) in Section 4.3, preserves the covariant
derivative. We can further apply (4.29) to express 𝐾 as a function of the coefficients
𝐸, 𝐹 and 𝐺. For simplicity, we assume that Φ(𝑢, 𝑣) is an orthogonal parameterization
(𝐹 ≡ 0), so v1 = √1 Φ𝑢 and v2 = √1 Φ𝑣 . We then have
𝐸 𝐺

𝜕v2 𝜕 1 1 𝜕v2 𝜕 1 1
= ( √ ) Φ𝑣 + √ Φ𝑣 , = ( √ ) Φ𝑣 + √ Φ𝑣𝑢 ,
𝜕𝑣 𝜕𝑣 𝐺 𝐺 𝜕𝑢 𝜕𝑢 𝐺 𝐺
so that
𝜕v2 1 −1 −𝐺 𝑢
⟨v1 , ⟩= √ ⟨Φ𝑢 , Φ𝑣𝑣 ⟩ = √ ⟨Φ𝑢𝑣 , Φ𝑣 ⟩ = √ ,
𝜕𝑣 𝐸𝐺 𝐸𝐺 2 𝐸𝐺
𝜕v2 1 𝐸𝑣
⟨v1 , ⟩= √ ⟨Φ𝑢 , Φ𝑢𝑣 ⟩ = √ ⋅
𝜕𝑢 𝐸𝐺 2 𝐸𝐺
Finally, by (4.29), we have
4.5 The Gauss-Bonnet Theorem 115

−1 𝜕 𝐺𝑢 𝜕 𝐸𝑣
𝐾= √ { (√ )+ (√ )} , (4.31)
2 𝐸𝐺 𝜕𝑢 𝐸𝐺 𝜕𝑣 𝐸𝐺
which is the promised formula. ◻

The condition, in Theorem 4.5.1, that Ω is contained in a parametrized neigh-

borhood, is dispensable, since it holds for every polygonal region. To prove this,
we consider a triangulation (Δ 𝑗 )1≤𝑖≤𝑟 of Ω such that each Δ 𝑗 is a closed polygonal
region that is contained in some parametrized neighborhood (so that we can apply
Theorem 4.5.1 to it). Moreover:
(i) the closure of Ω is equal to the union of the Δ 𝑗 ;
(ii) for each 𝑗 the simple closed curve 𝜕Δ 𝑗 has three vertices; the portions of 𝜕Δ 𝑗
between each pair of consecutive vertices are regular curves called edges. The sets
Δ 𝑗 are called the faces;
(iii) the intersection of two distinct faces is either empty, or reduces to a vertex,
or is an edge common to both.

a triangulation not a triangulation

Figure 4.7

We denote by V the set of vertices of the triangulation, and by 𝑉, 𝐴, 𝐹 the numbers

of vertices, edges and faces, respectively. Denoting by 𝛾 𝑙𝑗 (𝑙 = 1, 2, 3) the exterior
angles of Δ 𝑗 , Theorem 4.5.1 gives, for each 1 ≤ 𝑗 ≤ 𝐹,
3
∫ 𝐾 𝑑𝜎 + ∫ 𝑘 𝑔 (𝑠) 𝑑𝑠 + ∑ 𝛾 𝑙𝑗 = 2𝜋.
Δ𝑗 𝜕Δ 𝑗 𝑙=1

Each interior edge of Ω is run through twice in opposite directions (the geodesic
curvature showing up with opposed sign), and by summing the above formulas for
𝑗 = 1, . . . , 𝐹 only the edges that make up 𝜕Ω are left. We thus obtain
116 4 The Intrinsic Geometry of Surfaces

𝐹 3
∫ 𝐾 𝑑𝜎 + ∫ 𝐾𝑔 (𝑠) 𝑑𝑠 + ∑ ∑ 𝛾 𝑙𝑗 = 2𝜋𝐹. (4.32)
Ω 𝜕Ω 𝑗=1 𝑙=1

For each vertex 𝑣 ∈ V, let:

I𝑣 = {( 𝑗, 𝑙)∶ 𝛾 𝑙𝑗 is adjacent to 𝑣};

S(𝑣) = ∑ 𝛾 𝑙𝑗 ;
( 𝑗,𝑙)∈I𝑣

𝐴(𝑣) = number of edges starting from 𝑣.

We break V into three subsets V1 , V2 , V3 defined as follows: V1 is the set of vertices

inside Ω; V2 contains the vertices that belong to 𝜕Ω but are not vertices of Ω; V3
contains the vertices of Ω. Denoting by 𝜂𝑙𝑗 = 𝜋 − 𝛾 𝑙𝑗 the interior angles of Δ 𝑗 , we have:

● if 𝑣 ∈ V1 then

S(𝑣) = ∑ (𝜋 − 𝜂𝑙𝑗 ) = 𝜋#I𝑣 − ∑ 𝜂𝑙𝑗 = 𝜋 𝐴(𝑣) − 2𝜋,

( 𝑗,𝑙)∈I𝑣 ( 𝑗,𝑙)∈I𝑣

● if 𝑣 ∈ V2 then

S(𝑣) = 𝜋I𝑣 − ∑ 𝜂𝑙𝑗 = 𝜋(𝐴(𝑣) − 1) − 𝜋 = 𝜋 𝐴(𝑣) − 2𝜋,

( 𝑗,𝑙)∈I𝑣

● if 𝑣 ∈ V3 then 𝑣 is one of the 𝑘 vertices of Ω and, if 𝛾𝑖 is the corresponding

exterior angle, we have

S(𝑣) = 𝜋#I𝑣 − ∑ 𝜂𝑙𝑗 = 𝜋(𝐴(𝑣) − 1) − (𝜋 − 𝛾𝑖 ) = 𝜋 𝐴(𝑣) − 2𝜋 + 𝛾𝑖 .

( 𝑗,𝑙)∈I𝑣

Adding up all these formulas, we obtain

𝐹 3 𝑘
∑ ∑ 𝛾 𝑗 = ∑ S(𝑣) = 𝜋 ∑ 𝐴(𝑣) − 2𝜋𝑉 + ∑ 𝛾𝑖
𝑙
𝑗=1 𝑙=1 𝑣∈V 𝑣∈V 𝑖=1
𝑘
= 2𝜋(𝐴 − 𝑉) + ∑ 𝛾𝑖
𝑖=1

— because each edge is counted twice (once for each one of its endpoints). Replacing
in (4.32) and using Lemma 4.5.4 below, we obtain
𝑘
∫ 𝐾 𝑑𝜎 + ∫ 𝑘 𝑔 (𝑠) 𝑑𝑠 + ∑ 𝛾𝑖 = 2𝜋(𝑉 − 𝐴 + 𝐹) = 2𝜋,
Ω 𝜕Ω 𝑖=1

which establishes the Gauss-Bonnet formula for any polygonal region Ω ⊆ 𝑆.

Lemma 4.5.4 (Euler’s formula). 𝑉 − 𝐴 + 𝐹 = 1.

4.5 The Gauss-Bonnet Theorem 117

Proof We have to show that for every triangulation of a polygonal region we have
𝑉 − 𝐴 + 𝐹 = 1. We proceed by induction on the number of faces 𝐹. If 𝐹 = 1, then
there are three vertices and three edges and the formula is true. If 𝐹 > 1, then let Δ 𝑗
̃ = Ω ∖ Δ 𝑗 is a
be a face where at least one of the edges is part of 𝜕Ω, and such that Ω
polygonal region. Consider in Ω ̃ the triangulation induced by the triangulation of Ω.
Denoting by 𝑉,̃ 𝐴̃ and 𝐹̃ the numbers of vertices, edges and faces of the triangulation
̃ we have:
of Ω,
̃ = 𝐴−1, 𝐹̃ = 𝐹−1;
̃ = 𝑉, 𝐴
● if 𝜕Ω contains a single edge of Δ 𝑗 , then 𝑉
● if 𝜕Ω contains two edges of Δ 𝑗 , then 𝑉 ̃
̃ = 𝑉−1, 𝐴 = 𝐴−2, 𝐹̃ = 𝐹−1.
̃ ̃ ̃
In both cases, 𝑉 − 𝐴 + 𝐹 = 𝑉 − 𝐴 + 𝐹 and the proof by induction is complete. ◻

There is a special case of the Gauss-Bonnet formula that is worthy to note: if the
boundary of Ω consists of geodesic arcs (𝑘 𝑔 ≡ 0), we are left with
𝑘
∫ 𝐾 𝑑𝜎 + ∑ 𝛾𝑖 = 2𝜋,
Ω 𝑖=1

or, denoting by 𝜂𝑖 = 𝜋 − 𝛾𝑖 the interior angles of Ω,

𝑘
∑ 𝜂𝑖 − (𝑘 − 2)𝜋 = ∫ 𝐾 𝑑𝜎.
𝑖=1 Ω

When 𝑘 = 3, Ω is called a geodesic triangle, and we have just obtained the promised
generalization of Girard’s formula:
Corollary 4.5.5 The difference between the sum of the interior angles of a geodesic
triangle Δ and 𝜋 is given by the integral, extended to Δ, of the Gaussian curvature:

(𝜂1 + 𝜂2 + 𝜂3 ) − 𝜋 = ∫ 𝐾 𝑑𝜎.
Δ

If in particular the curvature of the surface is constant, then this difference is

proportional to the area of the triangle, equal to (𝜂1 + 𝜂2 + 𝜂3 ) − 𝜋 = 𝐾⋅ area (Δ).
Let us now assume that 𝑆 is a compact surface, and consider a triangulation
T = (Δ𝑖 )1≤𝑖≤𝑟 of 𝑆: conditions (i) – (iii) are satisfied, but it is not required that each
Δ 𝑗 is contained in a parametrized neighborhood. (The existence of triangulations for
arbitrary surfaces is a deep result, and a proof is given in [17]; for regular compact
surfaces — the case we are concerned with — we will give a proof in Exercise 115.)
The Euler characteristic of 𝑆 is 𝜒(𝑆) = 𝑉 − 𝐴 + 𝐹, where 𝑉, 𝐴 and 𝐹 are the numbers
of vertices, edges and faces of T .
Gauss-Bonnet Theorem 4.5.6 (Global Version). If 𝑆 is a compact surface then
∫ 𝐾 𝑑𝜎 = 2𝜋 𝜒(𝑆).
𝑆

Proof Writing down the Gauss-Bonnet formula for each of the faces of a triangulation
T and summing them up, the integrals of the geodesic curvatures all cancel, because
118 4 The Intrinsic Geometry of Surfaces

each edge is run through twice in opposite directions. Furthermore, the vertices of T
are all inside 𝑆 (of type V1 ) and the sum of the interior angles adjacent to each of
them is 2𝜋. By the above calculations, we are left with

∫ 𝐾 𝑑𝜎 = 2𝜋(𝑉 − 𝐴 + 𝐹),
𝑆

which is the desired formula. ◻

The integral ∫𝑆 𝐾 𝑑𝜎 is called the total curvature of 𝑆. The equation now obtained
shows that 𝜒(𝑆) is welldefined, independent of the triangulation of 𝑆 chosen to
compute it. Furthermore, 𝜒(𝑆) is invariant under diffeomorphisms (because a
diffeomorphism 𝑓 ∶ 𝑆 → 𝑆2 maps any triangulation T of 𝑆 to another triangulation
𝑓 (T ) of 𝑆2 with the same number of vertices, edges and faces). We thus obtain the
following result.

Corollary 4.5.7 Any two diffeomorphic compact surfaces have the same total curva-
ture.

For example, any surface 𝑆 diffeomorphic to the sphere has total curvature 4𝜋.
This is not surprising if the curvature of 𝑆 is positive at all points (as in the ellipsoid),
because in this situation we will see later on that 𝑁∶ 𝑆 → S2 is a diffeomorphism,
and ∫𝑆 𝐾 𝑑𝜎 is nothing but the area of the image of 𝑆 under 𝑁 (cf. Exercise 64 in
Section 3.2), which in this case is S2 . But if we call to mind that 𝑆 can also have
regions of negative curvature we more readily appreciate the strength of the result.

Figure 4.8
The Euler characteristic (and hence the total curvature) of the torus is zero; that of
the double torus is −2. That of the 𝑛-torus (𝑛 ≥ 1) is 2 − 2𝑛. Together with the sphere,
and up to diffeomorphisms, this list exhausts all orientable compact surfaces (see
Fig. 4.8); this is a classical result whose proof you can find, for example, in [17]. In
particular, among the compact surfaces, only those which are diffeomorphic to the
sphere have non-negative total curvature.
4.5 The Gauss-Bonnet Theorem 119

Example 4.5.8 A non-compact surface may have finite total curvature. Consider, on
a surface of revolution 𝑆 given by 𝜌 = 𝜌(𝑧), 𝑧 ∈ R , the region Ω(𝑧0 , 𝑧 1 ) bounded
by the two parallels 𝑧 = 𝑧 0 and 𝑧 = 𝑧1 (𝑧0 < 𝑧1 ). We can break Ω(𝑧 0 , 𝑧 1 ) into two
“four-sided polygons” by two meridians and add up the two resulting Gauss-Bonnet
formulas. Since there are four vertices and the sum of the two exterior angles adjacent
to each vertex is 𝜋, we obtain

∫ 𝐾 𝑑𝜎 + ∫ 𝑘 𝑔 𝑑𝑠 + 4𝜋 = 4𝜋,
Ω(𝑧0 ,𝑧1 ) 𝜕Ω(𝑧0 ,𝑧1 )

i.e.,
∫ 𝐾 𝑑𝜎 = ∫ −𝑘 𝑔 𝑑𝑠.
Ω(𝑧0 ,𝑧1 ) 𝜕Ω(𝑧0 ,𝑧1 )

z1

z0

Figure 4.9

Taking in 𝑆 the orientation given by

1
𝑁(𝜑, 𝑧) = √ (cos 𝜑, sin 𝜑, −𝜌),
1 + 𝜌2

the parallel 𝑧 = 𝑧0 is run through counterclockwise, and 𝑧 = 𝑧1 clockwise. A quick

calculation then shows that their geodesic curvatures are

−𝜌(𝑧0 ) 𝜌(𝑧 1 )
𝑘 𝑔 (𝑧 0 ) = √ , 𝑘 𝑔 (𝑧1 ) = √ ,
2 2
𝜌(𝑧0 ) 1 + 𝜌(𝑧0 ) 𝜌(𝑧 1 ) 1 + 𝜌(𝑧1 )

and therefore
120 4 The Intrinsic Geometry of Surfaces

⎛ 𝜌(𝑧1 ) 𝜌(𝑧0 ) ⎞
∫ 𝐾 𝑑𝜎 = −2𝜋 ⎜ √ −√ ⎟⋅
Ω(𝑧0 ,𝑧1 ) ⎝ 1 + 𝜌(𝑧1 )2 1 + 𝜌(𝑧0 ) ⎠
2

This formula guarantees that in many cases the limit

∫ 𝐾 𝑑𝜎 = 𝑧1lim
→+∞ ∫
𝐾 𝑑𝜎
𝑆 Ω(𝑧0 ,𝑧1 )
𝑧0 →−∞

exists and is finite; we will call it the total curvature of 𝑆.

√
√ the total curvature of the hyperboloid 𝑥 + 𝑦 = 1 + 𝑧 is −2 2𝜋,
2 2 2
For example,
since 𝜌(𝑧) = 1 + 𝑧 2 and lim 𝜌(𝑧) = ±1.
𝑧→±∞

It would not be difficult to compute directly ∫Ω(𝑧0 ,𝑧1 ) 𝐾 𝑑𝜎 (see example 3.2.1) —
but, besides illustrating a use of the Gauss-Bonnet Theorem, this method is applicable
to surfaces other than surfaces of revolution (e.g., to those non-compact surfaces 𝑆
for which 𝑆 ∖ ℭ is a surface of revolution for some compact ℭ ⊆ 𝑆). ◻

Exercises

96. Assume that 𝑆 has non-positive curvature at all its points. Show that if two
geodesics start from the same point in 𝑆, they cannot meet again in such a way that
their traces constitute the boundary of a simple region of 𝑆 (in particular, no closed
geodesic of 𝑆 can be the boundary of a simple region).

97. Show that if 𝛾 is a regular closed simple curve in S2 then ∣∫𝛾 𝑘 𝑔 𝑑𝑠∣ < 2𝜋.

98. Let 𝑝 ∈ 𝑆 be such that 𝐾(𝑝) > 0, and let (𝑈, Φ) be a parameterization such
that 𝑝 ∈ Φ(𝑈) and 𝐾 ○ Φ(𝑢, 𝑣) > 0 for all (𝑢, 𝑣) ∈ 𝑈. Consider a family of circles
(C𝑟 )0<𝑟< 𝛿 such that:
i. each C𝑟 has radius 𝑟;
ii. for every 𝑟, Φ−1 (𝑝) ∈ C𝑟 and the closed disk bounded by C𝑟 is contained in 𝑈;
iii. if 𝑟 < ̃
𝑟 then C𝑟 is inside C̃𝑟 .
Further denote by 𝜓𝑟 ∈ [−𝜋, 𝜋] the oriented angle between the initial and final
positions of a vector carried parallel from 𝑝 to 𝑝 along the closed curve Φ(C𝑟 ) in 𝑆.
(a) Show that there exists 𝜀 ∈ ]0, 𝛿[ such that

0<𝑟 <̃
𝑟 < 𝜀 ⇒ 0 < ∣𝜓𝑟 ∣ < ∣𝜓̃𝑟 ∣ < 𝜋.

(b) Conclude that any surface on which parallel transport depends only on the
initial and final points of the curve has constant zero curvature:
Give an example of a surface with constant zero curvature that does not have this
property.
4.6 Minimizing Properties of Geodesics 121

Figure 4.10

99. Is it true that two non-compact diffeomorphic surfaces necessarily have the same
total curvature?

4.6 Minimizing Properties of Geodesics

We begin this section by establishing the local existence of geodesics on arbitrary

surfaces: to this end we will observe that, in local coordinates, geodesics are
characterized by second-order differential equations. We then introduce polar geodesic
coordinates that allow us to show that the shortest path on the surface between any
two points is, when it exists, given by a geodesic.
Recall that a curve 𝛼 of 𝑆 is a geodesic if the unit vector field tangent to the curve
𝝉1 = ∣𝛼1′ ∣ 𝛼′ is a parallel vector field. Taking coordinates (𝑈, Φ) and assuming that
𝛼(𝑡) = Φ(𝑢(𝑡), 𝑣(𝑡)) is parametrized with constant scalar velocity 𝑚 > 0, we have

1 ′ 1 ′
𝝉1 = 𝛼 = (𝑢 Φ𝑢 + 𝑣 ′ Φ𝑣 ).
𝑚 𝑚
From formula (4.9), we obtain
122 4 The Intrinsic Geometry of Surfaces

𝐷𝝉1 1
= {(𝑢 ′′ + 𝑢 ′2 Γ111
+ 2𝑢 ′ 𝑣 ′ Γ12
1
+ 𝑣 ′2 Γ22
1
)Φ𝑢
𝑑𝑡 𝑚
+ (𝑣 ′′ + 𝑢 ′2 Γ11
2
+ 2𝑢 ′ 𝑣 ′ Γ12
2
+ 𝑣 ′2 Γ22
2
)Φ𝑣 }

𝐷𝝉1
— and, if 𝛼 is a geodesic, then ≡ 0, so that
𝑑𝑡
⎧
⎪
⎪𝑢 ′′ + 𝑢 ′2 Γ11
1
+ 2𝑢 ′ 𝑣 ′ Γ12
1
+ 𝑣 ′2 Γ22
1
= 0,
⎨ ′′ (4.33)
⎪
⎪ + ′2
Γ 2
+ 2𝑢 ′ ′
Γ 2
+ ′2
Γ 2
=
22 0.
⎩
𝑣 𝑢 11 𝑣 12 𝑣

Conversely, if (𝑢(𝑡), 𝑣(𝑡)) is a non-constant solution of (4.33) then 𝛼(𝑡) =

Φ(𝑢(𝑡), 𝑣(𝑡)) is a regular curve and its scalar velocity is constant, since in this case
𝑑 𝐷𝛼′
the covariant derivative of the vector field 𝛼′ is zero and ∣𝛼′ ∣2 = 2⟨𝛼′ , ⟩ = 0.
𝑑𝑡 𝑑𝑡
1
It follows that 𝝉1 = 𝛼′ is a parallel vector field and therefore 𝛼 is a geodesic.
𝑚

We define parametrized geodesics to be either a parameterization 𝛼(𝑡) of a

geodesic with constant scalar velocity ∣𝛼′ (𝑡)∣, or a constant curve. This means that
the solutions of the system give us, locally, all parametrized geodesics of the surface:
we therefore call the equations in (4.33) the differential equations of the geodesics.

The system of equations (4.33) can be restated in the form

⎧
⎪
⎪𝑢 ′′ = 𝐻1 (𝑢, 𝑣, 𝑢 ′ , 𝑣 ′ ),
⎨ ′′ (4.34)
⎪
⎪ 𝑣 = 𝐻2 (𝑢, 𝑣, 𝑢 ′ , 𝑣 ′ ),
⎩
or otherwise
⎧
⎪𝑢 ′ = 𝑎,
⎪
⎪
⎪
⎪
⎪
⎪𝑣 ′ = 𝑏,
⎨ ′ (4.35)
⎪
⎪
⎪ 𝑎 = 𝐻1 (𝑢, 𝑣, 𝑎, 𝑏),
⎪
⎪
⎪𝑏 ′ = 𝐻2 (𝑢, 𝑣, 𝑎, 𝑏)
⎪
⎩
— where 𝐻1 , 𝐻2 are differentiable functions defined on 𝑈 × R 2 ⊆ R 4 where 𝑈
is the domain of the parameterization Φ. Theorem 3.3.1 guarantees that every
(𝑢 0 , 𝑣 0 , 𝑎 0 , 𝑏 0 ) ∈ 𝑈 × R 2 has an open neighborhood 𝑊 such that there exist 𝜀 > 0 and
a differentiable mapping

]−𝜀, 𝜀[ × 𝑊 → 𝑈 × R 2 , (𝑡; 𝑢, 𝑣, 𝑎, 𝑏) ↦ 𝜑̃(𝑡; 𝑢, 𝑣, 𝑎, 𝑏)

enjoying the following conditions: for every (𝑢, 𝑣, 𝑎, 𝑏) ∈ 𝑊, the curve

𝑡 ↦ 𝜑̃(𝑡; 𝑢, 𝑣, 𝑎, 𝑏)

is the only solution of the system (4.35) with initial condition 𝜑̃(0; 𝑢, 𝑣, 𝑎, 𝑏) =
(𝑢, 𝑣, 𝑎, 𝑏).
4.6 Minimizing Properties of Geodesics 123

Writing 𝜑̃ = (𝜑1 , 𝜑2 , 𝜑3 , 𝜑4 ), it is clear that 𝜑3 and 𝜑4 are redundant, since

𝜕𝜑1 𝜕𝜑2
𝜑3 = and 𝜑4 = . Letting 𝜑 = (𝜑1 , 𝜑2 ), the conclusion we obtain for the
𝜕𝑡 𝜕𝑡
system (4.34), and hence for (4.33), is as follows. The mapping 𝜑∶ ] − 𝜀, 𝜀[ × 𝑊 → 𝑈,
where 𝑊 is a neighborhood of (𝑢 0 , 𝑣 0 , 𝑎 0 , 𝑏 0 ), is such that, for (𝑢, 𝑣, 𝑎, 𝑏) ∈ 𝑊, the
curve 𝑡 ↦ 𝜑(𝑡; 𝑢, 𝑣, 𝑎, 𝑏) is the only solution of (4.34) with initial conditions

𝑑
𝜑(0; 𝑢, 𝑣, 𝑎, 𝑏) = (𝑢, 𝑣), 𝜑(𝑡; 𝑢, 𝑣, 𝑎, 𝑏)∣𝑡=0 = (𝑎, 𝑏).
𝑑𝑡
In summary, determining (𝑢(0), 𝑣(0)) and (𝑢 ′ (0), 𝑣 ′ (0)) completely determines
a solution (𝑢(𝑡), 𝑣(𝑡)) of (4.33). The solutions depend differentiably on these initial
conditions. Let us now define

𝐷Φ∶ 𝑈 × R 2 → ⋃ ({𝑝} × 𝑇 𝑝 𝑆)
𝑝∈Φ(𝑈)
(4.36)
(𝑢, 𝑣, 𝑎, 𝑏) ↦ (Φ(𝑢, 𝑣), (𝑎Φ𝑢 + 𝑏Φ𝑣 )∣(𝑢,𝑣) )

and note that 𝐷Φ is a continuous bijection. Given 𝑝 0 = Φ(𝑢 0 , 𝑣 0 ) ∈ Φ(𝑈), consider

the corresponding neighborhood 𝑊 associated with (𝑢 0 , 𝑣 0 , 0, 0) (which we can
assume to be of the form 𝑊 = 𝑊1 × 𝑊2 ), 𝜀 > 0 and the mapping 𝜑. Let us take 𝑉 ⊆ 𝑆,
an open neighborhood of 𝑝 0 whose closure is compact and contained in Φ(𝑊1 ), and
choose 𝛿 > 0 such that the set

𝐵 𝛿 (𝑉) = {(𝑝, v)∶ 𝑝 ∈ 𝑉, v ∈ 𝑇 𝑝 𝑆, ∣v∣ < 𝛿}

is contained in 𝐷Φ(𝑊) (that such a choice of 𝛿 is possible is an easy exercise). We

then define the mapping
−1
𝛾∶ ]−𝜀, 𝜀[ × 𝐵 𝛿 (𝑉) → 𝑆, 𝛾(𝑡; 𝑝, v) = Φ ○ 𝜑(𝑡; (𝐷Φ) (𝑝, v)),

and it follows from our analysis that for every (𝑝, v) ∈ 𝐵 𝛿 (𝑉), the curve 𝑡 ↦ 𝛾(𝑡; 𝑝, v)
is the only parametrized geodesic that at time 0 passes through 𝑝 with velocity v; the
constant parametrized geodesics are those of the form 𝛾(𝑡; 𝑝, 0). Furthermore, for
𝜆 ∈ R we have
𝛾(𝜆𝑡; 𝑝, v) = 𝛾(𝑡; 𝑝, 𝜆v), (4.37)
because the two terms of (4.37) are parametrized geodesics satisfying the same initial
conditions: at time 0, they both pass through 𝑝 with velocity 𝜆v. It follows that, up to
reparametrization, there is exactly one geodesic whose tangent line at a given point
has a given direction.
Observation 4.6.1 At this point we can already state that there are no geodesics
on the sphere other than the maximal circles (see example 4.3.9), since through
each point passes a maximal circle tangent to each given direction. The sphere thus
has the particularity that all its geodesics are closed. (A non-constant parametrized
geodesic 𝛾(𝑡) is called closed if it is periodic — that is, if there exists 𝑇 > 0 such
that 𝛾(𝑡 + 𝑇) = 𝛾(𝑡) for all 𝑡 ∈ R . A necessary and sufficient condition for 𝛾 to be
124 4 The Intrinsic Geometry of Surfaces

closed is that there exist 𝑡1 < 𝑡 2 such that 𝛾(𝑡 1 ) = 𝛾(𝑡2 ) and 𝛾 ′ (𝑡 1 ) = 𝛾 ′ (𝑡 2 ) — i.e., 𝛾
is closed if and only if it passes again through the same point with the same velocity
vector.) Surprisingly, there are other surfaces with the same property, as shown in [4].
◻

Let us now apply (4.37) to show that, choosing ∣v∣ sufficiently small, 𝛾(𝑡; 𝑝, v)
is defined for ∣𝑡∣ < 2. In fact, since 𝛾(𝑡; 𝑝, v) = 𝛾( 2𝜀 𝑡; 𝑝, 2𝜀 v), we can state that, for
∣ 2𝜀 v∣ < 𝛿 (i.e., for ∣v∣ < 𝜀2𝛿 ), 𝛾(𝑡; 𝑝, v) is defined whenever ∣ 2𝜀 𝑡∣ < 𝜀 — i.e., whenever
∣𝑡∣ < 2. In summary:
For every 𝑝 0 ∈ 𝑆 there exist 𝜂 > 0 and a neighborhood 𝑉 of 𝑝 0 such that whenever
𝑝 ∈ 𝑉 and v ∈ 𝐵 𝜂 (𝑝) = {v ∈ 𝑇 𝑝 𝑆∶ ∣v∣ < 𝜂} the geodesic 𝑡 ↦ 𝛾(𝑡; 𝑝, v) is defined for
𝑡 ∈ ] −2, 2[.
Given 𝑝 ∈ 𝑆, the exponential mapping exp 𝑝 is defined by exp 𝑝 (v) = 𝛾(1; 𝑝, v).
By the above reasoning, there exists some 𝜂 > 0 such that exp 𝑝 is defined on 𝐵 𝜂 (𝑝);
one can choose such an 𝜂 suitable for all points in a neighborhood of 𝑝.
The geodesic 𝑡 ↦ 𝛾(𝑡; 𝑝, v) has constant scalar velocity ∣v∣, and therefore its arc
length in the interval [0, 1] is also ∣v∣. The geometric meaning of the exponential
mapping is therefore as follows: exp 𝑝 (v) is the point that travels a distance of ∣v∣ on
the geodesic that begins at 𝑝 and whose direction and orientation is given by v. Note
that, by (4.37), we have exp 𝑝 (𝑡v) = 𝛾(1; 𝑝, 𝑡v) = 𝛾(𝑡; 𝑝, v) — which means that the
geodesics starting from 𝑝 are the image of the lines (or line segments) in 𝑇 𝑝 𝑆 that
pass through the origin under exp 𝑝 .
To make full use of the exponential mapping, we need the next proposition:
Proposition 4.6.2 Given 𝑝 0 ∈ 𝑆, there exist 𝛿 > 0 and an open neighborhood 𝑊 ⊆ 𝑆
of 𝑝 0 such that, for all 𝑝 ∈ 𝑊, exp 𝑝 ∣ 𝐵 ( 𝑝) is a diffeomorphism onto the image.
𝛿

Proof We know that there exist 𝜂 > 0 and an open neighborhood 𝑉 ⊆ 𝑆 of 𝑝 0 such
that, for all 𝑝 on 𝑉, the exponential mapping exp 𝑝 is defined on 𝐵 𝜂 (𝑝). We can thus
consider the differentiable mapping

𝐹∶ 𝐵 𝜂 (𝑉) Ð→ 𝑆 × 𝑆
(𝑝, v) z→ (𝑝, exp 𝑝 (v)).

(A caveat: both 𝐵 𝜂(𝑉) and 𝑆 × 𝑆 are spaces of dimension four, since 𝑆 × 𝑆 is the
product of two spaces of dimension two and 𝐵 𝜂 (𝑉) can be identified, via 𝐷Φ as
defined in (4.36), with an open subset of R 4 . Our proof can be made rigorous by
−1
applying the inverse mapping theorem to (Φ × Φ) ○ 𝐹 ○ 𝐷Φ, which is the expression
of 𝐹 in “local coordinates”.)
The tangent spaces to 𝐵 𝜂(𝑉) at (𝑝 0 , 0) and to 𝑆 ×𝑆 at (𝑝 0 , 𝑝 0 ) = 𝐹(𝑝 0 , 0) coincide:
both are 𝑇 𝑝0 𝑆 × 𝑇 𝑝0 𝑆. Given v ∈ 𝑇 𝑝0 𝑆, let 𝛼 be a curve in 𝑉 such that 𝛼(0) = 𝑝 0 and
𝛼′ (0) = v. Then 𝑡 ↦ (𝛼(𝑡), 0) is a curve in 𝐵 𝜂 (𝑉) that passes through (𝑝 0 , 0) with
velocity (v, 0). Thus

𝑑 𝑑
𝐷𝐹( 𝑝0 ,0) (v, 0) = 𝐹(𝛼(𝑡), 0)∣𝑡=0 = (𝛼(𝑡), 𝛼(𝑡))∣𝑡=0 = (v, v).
𝑑𝑡 𝑑𝑡
4.6 Minimizing Properties of Geodesics 125
𝜂
On the other hand, if w ∈ 𝑇 𝑝0 𝑆 ∖ {0} then the curve 𝑡 ↦ (𝑝 0 , 𝑡w), ∣𝑡∣ < , is
∣w∣
contained in 𝐵 𝜂 (𝑉) and passes through (𝑝 0 , 0) with velocity (0, w), and therefore

𝑑 𝑑
𝐷𝐹( 𝑝0 ,0) (0, w) = 𝐹(𝑝 0 , 𝑡w)∣𝑡=0 = (0, exp 𝑝0 (𝑡w)∣𝑡=0 ) = (0, w).
𝑑𝑡 𝑑𝑡
We thus conclude that

𝐷𝐹( 𝑝0 ,0) (v, w) = (v, v + w) for every (v, w) ∈ 𝑇 𝑝0 𝑆 × 𝑇 𝑝0 𝑆

– which shows that 𝐷𝐹( 𝑝0 ,0) is a linear isomorphism. The inverse mapping theorem
then guarantees that the restriction of 𝐹 to some neighborhood of (𝑝 0 , 0) in 𝐵 𝜂 (𝑉)
is a diffeomorphism onto the image, and we can choose such a neighborhood of the
form 𝐵 𝛿 (𝑊), where 𝑊 is an open subset of 𝑆 and 𝛿 > 0; it is easily verified that these
choices of 𝑊 and 𝛿 satisfy the desired condition. ◻

Given 𝑝 ∈ 𝑆, a neighborhood 𝑉 ⊆ 𝑆 of 𝑝 is called a normal neighborhood of 𝑝

if there exists 𝛿 > 0 such that exp 𝑝 is a diffeomorphism onto the image and
∣𝐵 ( 𝑝)
𝛿
𝑉 ⊆ exp 𝑝 (𝐵 𝛿 (𝑝)). For such a 𝛿, we write 𝐷 𝛿 (𝑝) = exp 𝑝 (𝐵 𝛿 (𝑝)): hence, 𝐷 𝛿 (𝑝) is
the neighborhood of 𝑝 covered by the geodesic rays of length 𝛿 starting from 𝑝, but
note that for now we only define 𝐷 𝛿 (𝑝) for sufficiently small 𝛿.
Let (v1 , v2 ) be an orthonormal basis of 𝑇 𝑝 𝑆. exp 𝑝 gives rise to several coordinate
systems in 𝐷 𝛿 (𝑝):

Φ(𝑢, 𝑣) = exp 𝑝 (𝑢v1 + 𝑣v2 ) (set on disk 𝑢 2 + 𝑣 2 < 𝛿2 );

Ψ(𝜌, 𝜑) = exp 𝑝 (𝜌 cos 𝜑 v1 + 𝜌 sin 𝜑 v2 ) (0< 𝜌 < 𝛿, 𝜑 ∈ ]𝜑0 − 𝜋, 𝜑0 + 𝜋[).

The coordinates Ψ(𝜌, 𝜑) are known as geodesic polar coordinates, and we call
Φ(𝑢, 𝑣) geodesic Cartesian coordinates. Of course, these coordinates depend on the
choice of the basis (v1 , v2 ). If we want them to respect the orientation of 𝑆, it suffices
that (v1 , v2 ) are positively oriented. In the case of geodesic polar coordinates, we
obtain different parameterizations by restricting 𝜑 to intervals of length 2𝜋, and each
of these parameterizations excludes a radial geodesic (𝜑 = constant). However, since
the excluded radial geodesic is arbitrary, the conclusions we draw with one of these
parameterizations are valid throughout the “punctured disc” 𝐷 𝛿 (𝑝) ∖ {𝑝}.
The next lemma says, among other things, that Ψ(𝜌, 𝜑) is an orthogonal parame-
terization. Geometrically, this means that the radial geodesics (𝜑 = constant) starting
from 𝑝 and the circumference geodesics (𝜌 = constant) with center 𝑝 intersect
orthogonally (see Fig. 4.11.).

Lemma 4.6.3 The coefficients 𝐸, 𝐹, 𝐺 of the geodesic polar coordinates satisfy

√ √
𝐸 ≡ 1, 𝐹 ≡ 0, lim 𝐺 = 0, lim ( 𝐺)𝜌 = 1.
𝜌→0 𝜌→0
126 4 The Intrinsic Geometry of Surfaces

TpS expp

S p

Figure 4.11

Proof To shorten the notation, we write v𝝋 = cos 𝜑 v1 + sin 𝜑 v2 and w𝝋 = − sin 𝜑 v1 +

cos 𝜑 v2 . We have
2
𝜕
𝐸 = ∣Ψ𝜌 ∣2 = ∣ exp 𝑝 (𝜌v𝝋 )∣ = ∣v𝝋 ∣2 = 1,
𝜕𝜌

since 𝜌 ↦ exp 𝑝 (𝜌v𝝋 ) is a parametrized geodesic, and so has constant scalar velocity.
𝐷Ψ𝜌
Furthermore, denoting by the derivative covariant of Ψ𝜌 along 𝜌 ↦ Ψ(𝜌, 𝜑) —
𝜕𝜌
which is zero because Ψ𝜌 is the velocity field of a parametrized geodesic — we have

𝜕 𝐷Ψ𝜌
𝐹𝜌 = ⟨Ψ𝜌 , Ψ𝜑⟩ = ⟨ , Ψ𝜑 ⟩ + ⟨Ψ𝜌 , Ψ𝜑𝜌 ⟩
𝜕𝜌 𝜕𝜌
𝜕 1
= ( ∣Ψ𝜌 ∣2 ) = 0 (because 𝐸 ≡ 1).
𝜕𝜑 2
We thus conclude that 𝐹 does not depend on 𝜌. Since

𝐹 = ⟨𝐷(exp 𝑝 )𝜌v𝝋 (v𝝋 ), 𝐷(exp 𝑝 )𝜌v𝝋 (𝜌w𝝋 )⟩

= 𝜌⟨𝐷(exp 𝑝 )𝜌v𝝋 (v𝝋 ), 𝐷(exp 𝑝 )𝜌v𝝋 (w𝝋 )⟩,

we have
∣𝐹∣ ≤ 𝜌∣𝐷(exp 𝑝 )𝜌v𝝋 (v𝝋 )∣ ∣𝐷(exp 𝑝 )𝜌v𝝋 (w𝝋 )∣,

which, together with the facts that 𝐷(exp 𝑝 )𝑂 is the identity (as we saw in the proof
of Proposition 4.6.2) and v𝝋 and w𝝋 are unit vectors, implies lim 𝐹 = 0 and therefore
𝜌→0
𝐹 ≡ 0.
It remains to prove the last two equalities. For this let us consider the geodesic
Cartesian coordinates Φ(𝑢, 𝑣), whose coefficients we denote by 𝐸, 𝐹, 𝐺. Noting that
for (𝑢, 𝑣) ≠ (0, 0), we have 𝑢 = 𝜌 cos 𝜑, 𝑣 = 𝜌 sin 𝜑 and that at (0, 0) the coefficients
𝐸, 𝐹, 𝐺 are equal to 1, 0, 1 (because Φ𝑢 ∣(0, 0) = v1 and Φ𝑣 ∣(0,0) = v2 ), we have
4.6 Minimizing Properties of Geodesics 127

√ √ √
√ 𝜕(𝑢, 𝑣) 2 2
𝐺 = 𝐸𝐺 − 𝐹 2 = ∣ ∣ 𝐸𝐺−𝐹 = 𝜌 𝐸𝐺−𝐹
𝜕(𝜌, 𝜑)
√
and therefore lim 𝐺 = 0. Finally, differentiating the last equality, we obtain
𝜌→0

√ √
√ 2 𝜕 2
( 𝐺)𝜌 = 𝐸𝐺−𝐹 +𝜌 ( 𝐸𝐺−𝐹 )
𝜕𝜌
√ √ √
2 𝜕 2 𝜕 2
= 𝐸𝐺 −𝐹 + 𝜌{ ( 𝐸 𝐺 − 𝐹 ) cos 𝜑 + ( 𝐸 𝐺 − 𝐹 ) sin 𝜑} ,
𝜕𝑢 𝜕𝑣

and thus √
√ 2
lim ( 𝐺)𝜌 = lim 𝐸 𝐺 − 𝐹 = 1. ◻
𝜌→0 (𝑢,𝑣)→(0,0)

Example 4.6.4 Let us make explicit a geodesic coordinate system in S2 with ori-
gin at the North Pole N = (0, 0, 1). Fixing the orthonormal basis v1 = (1, 0, 0),
v2 = (0, 1, 0), the geodesic 𝜑 = constant is the meridian that forms angle 𝜑
with the plane 𝑦 = 0; its parameterization by arc length is 𝑠 (0 ≤ 𝑠 < 𝜋) ↦
(sin 𝑠 cos 𝜑, sin 𝑠 sin 𝜑, cos 𝑠), and we find the point Ψ(𝜌, 𝜑) at distance 𝜌 from
N . Thus Ψ(𝜌, 𝜑) = (sin 𝜌 cos 𝜑, sin 𝜌 sin 𝜑, cos 𝜌) — which shows that these
geodesic coordinates are just the spherical coordinates. Let us also point out that
S2 ∖ {(0, 0, −1)} = 𝐷 𝜋 (N ) is a normal neighborhood of N , because it is the
diffeomorphic image of the disk 𝐵 𝜋 (N ) under expN . ◻

We can now almost specify under what conditions a geodesic describes a shortest
path on the surface. But first, let us define the intrinsic distance on the surface, which
is the distance “experienced” by those who move on it. For a connected surface 𝑆
and 𝑝, 𝑞 ∈ 𝑆, we define 𝑑(𝑝, 𝑞) to be the least of the lengths of the curves in 𝑆 that
connect 𝑝 to 𝑞:

𝑑(𝑝, 𝑞) = inf{𝑙(𝛼) ∣ 𝛼∶ [𝑎, 𝑏] → 𝑆 is piecewise regular,

𝛼(𝑎) = 𝑝 and 𝛼(𝑏) = 𝑞}.

The intrinsic distance 𝑑 is a true distance, since it enjoys the following three properties:
(i) 𝑑(𝑝, 𝑞) ≥ 0 and 𝑑(𝑝, 𝑞) = 0 if and only if 𝑝 = 𝑞 (positivity);
(ii) 𝑑(𝑝, 𝑞) = 𝑑(𝑞, 𝑝) (symmetry);
(iii) 𝑑(𝑝, 𝑞) ≤ 𝑑(𝑝, 𝑟) + 𝑑(𝑟, 𝑞) (triangular inequality).
These properties are easily verified; we prove property (iii) via an example. If
𝛼1 ∶ [𝑎, 𝑏] → 𝑆 is a curve from 𝑝 to 𝑟, and 𝛼2 ∶ [𝑐, 𝑑] → 𝑆 another from 𝑟 to 𝑞, then
their juxtaposition 𝛼1 ∗ 𝛼2 , defined on [𝑎, 𝑏 + 𝑑 − 𝑐] by
⎧
⎪
⎪𝛼1 (𝑡) if 𝑎 ≤ 𝑡 ≤ 𝑏
(𝛼1 ∗ 𝛼2 )(𝑡) = ⎨
⎪
⎪ 𝛼 (𝑡 + 𝑐 − 𝑏) if 𝑏 ≤ 𝑡 ≤ 𝑏 + 𝑑 − 𝑐,
⎩ 2
128 4 The Intrinsic Geometry of Surfaces

is a piecewise regular curve from 𝑝 to 𝑞, and 𝑙(𝛼1 ∗ 𝛼2 ) = 𝑙(𝛼1 ) + 𝑙(𝛼2 ). We then

have

𝑑(𝑝, 𝑞) ≤ inf 𝑙(𝛼1 ∗ 𝛼2 ) = inf {𝑙(𝛼1 ) + 𝑙(𝛼2 )}

𝛼1 , 𝛼2 𝛼1 , 𝛼2

= inf 𝑙(𝛼1 ) + inf 𝑙(𝛼2 ) = 𝑑(𝑝, 𝑟) + 𝑑(𝑟, 𝑞)

𝛼1 𝛼2

— which proves (iii).

We say that the curve 𝛼 in 𝑆 from 𝑝 to 𝑞, which is piecewise regular, minimizes
distance (or is minimizing) if 𝑙(𝛼) = 𝑑(𝑝, 𝑞) (indeed, by Proposition 4.6.8 below, any
minimizing curve is a geodesic and is therefore regular).

Example 4.6.5 There do not always exist minimizing curves in 𝑆. It is sufficient to

take for 𝑆 a punctured plane (i.e., a plane from which a point 𝑂 has been removed)
and consider for 𝑆 two points 𝑝, 𝑞 such that 𝑂 ∈ [𝑝, 𝑞]: the intrinsic distance in 𝑆
between 𝑝 and 𝑞 is ∣𝑝 − 𝑞∣, but there is in 𝑆 no curve from 𝑝 to 𝑞 with length ∣𝑝 − 𝑞∣. ◻

Proposition 4.6.6 Geodesics locally minimize distance. More exactly, if 𝐷 𝛿 (𝑝) is a

normal neighborhood of 𝑝, and 𝑞 ∈ 𝐷 𝛿 (𝑝), then the radial geodesic from 𝑝 to 𝑞 is
the only minimizing curve between 𝑝 and 𝑞.

Proof Take geodesic polar coordinates Ψ(𝜌, 𝜑) centered at 𝑝, and let (𝜌0 , 𝜑0 ) be
such that Ψ(𝜌0 , 𝜑0 ) = 𝑞. Given a curve 𝛼∶ [𝑎, 𝑏] → 𝑆, piecewise regular, such that
𝛼(𝑎) = 𝑝, 𝛼(𝑏) = 𝑞, our goal is to show that 𝑙(𝛼) > 𝜌0 , unless 𝛼 is a reparametrization
of the radial geodesic 𝜌 (0 ≤ 𝜌 ≤ 𝜌0 ) ↦ Ψ(𝜌, 𝜑0 ).
We first deal with the case where 𝛼([𝑎, 𝑏]) ⊆ 𝐷 𝛿 (𝑝). We may assume, truncating
the curve if necessary, that 𝛼(𝑡) ≠ 𝑝 for all 𝑡 ∈ ]𝑎, 𝑏]. Under this hypothesis, there
exist functions 𝜌(𝑡) and 𝜑(𝑡), piecewise differentiable, such that 𝜑(𝑏) = 𝜑0 and
𝛼(𝑡) = Ψ(𝜌(𝑡), 𝜑(𝑡)) for 𝑡 ∈ ]𝑎, 𝑏]. Using Lemma 4.6.3, we then have
𝑏√
𝑙(𝛼) = ∫ 𝐸 𝜌 2 + 2𝐹 𝜌𝜑 + 𝐺𝜑2 𝑑𝑡
𝑎
𝑏√ 𝑏√ 𝑏
=∫ 𝜌 2 + 𝐺𝜑2 𝑑𝑡 ≥ ∫ 𝜌 2 𝑑𝑡 ≥ ∫ 𝜌 𝑑𝑡 = 𝜌0 ,
𝑎 𝑎 𝑎

and equality holds only if 𝜑 ≡ 0 (i.e., if 𝜑 is constant) and 𝜌 ≥ 0, which implies that 𝛼
is the radial geodesic from 𝑝 to 𝑞.
If 𝛼([𝑎, 𝑏]) is not contained in 𝐷 𝛿 (𝑝) then, given 0 < 𝑟 < 𝛿, let 𝑡𝑟 be the first
point such that 𝛼(𝑡𝑟 ) belongs to the geodesic circumference 𝜌 = 𝑟. By the above
conclusion, we have 𝑙(𝛼) ≥ 𝑙(𝛼∣[𝑎,𝑡 ] ) ≥ 𝑟. This inequality, valid for all 𝑟 < 𝛿, implies
𝑟
that 𝑙(𝛼) ≥ 𝛿 > 𝜌0 . ◻

Observations 4.6.7 A. From the proof of Proposition 4.6.6 it also follows that, for
every point 𝑞 ∈ 𝑆 ∖ 𝐷 𝛿 (𝑝), we have 𝑑(𝑝, 𝑞) ≥ 𝛿. In short, if we put, for arbitrary
𝛿 > 0, 𝐷 𝛿 (𝑝) = {𝑞 ∈ 𝑆∶ 𝑑(𝑝, 𝑞) < 𝛿}, this agrees with our previous definition when
𝐷 𝛿 (𝑝) is a normal neighborhood of 𝑝.
4.6 Minimizing Properties of Geodesics 129

B. We should stress that we can only guarantee locally that geodesics are minimizing.
On the cylinder 𝑥 2 + 𝑦 2 = 1, for example, the geodesic 𝛼(𝑡) = (cos 𝑡, sin 𝑡, 𝑡) does
not minimize the distance between 𝛼(0) = (1, 0, 0) and 𝛼(2𝜋) = (1, 0, 2𝜋): the line
segment {(1, 0)} × [0, 2𝜋] is shorter. ◻

Proposition 4.6.8 Every minimizing curve is a geodesic.

Proof Let 𝛼∶ [0, 𝐿] → 𝑆 be a minimizing curve, piecewise regular, parametrized by

arc length. It suffices to see that, given any point 𝑝 0 = 𝛼(𝑠0 ) on the curve, there
exists 𝜀 > 0 such that the restriction from 𝛼 to [𝑠0 − 𝜀, 𝑠0 + 𝜀] ([0, 𝜀] or [𝐿 − 𝜀, 𝐿], if
𝑠0 = 0 or 𝑠0 = 𝐿) is a geodesic. We take 𝛿 > 0 and the neighborhood 𝑊 of 𝑝 0 given in
Proposition 4.6.2, and we choose 𝜀 > 0 so that 𝜀 < 2𝛿 and the points 𝑝 = 𝛼(𝑠0 − 𝜀) and
𝑞 = 𝛼(𝑠0 + 𝜀) are both in 𝑊. Since 𝛼∣[𝑠 −𝜀,𝑠 +𝜀] is minimizing, we have 𝑑(𝑝, 𝑞) = 2𝜀
0 0
and, by note 4.6.7, A, 𝑞 belongs to the normal neighborhood 𝐷 𝛿 (𝑝) of 𝑝. It follows
then from Proposition 4.6.6 that 𝛼∣[𝑠 −𝜀,𝑠 +𝜀] is a geodesic. ◻
0 0

Note that in the above proof, 𝑉 = 𝑊 ∩ 𝐷 𝛿/2 (𝑝 0 ) is a normal neighborhood of all

its points, therefore enjoying the following property: any two points of 𝑉 are joined
by a single minimizing curve, which is a geodesic of length 𝛿.
In note 4.5.3 B we gave a formula (4.31) for the curvature 𝐾 as a function of the
coefficients of an orthogonal parameterization. This formula simplifies substantially
for geodesic polar coordinates Ψ(𝜌, 𝜑), where 𝐸 ≡ 1:
√
−1 𝜕 𝐺 𝜌 −( 𝐺)𝜌𝜌
𝐾= √ (√ ) = √ ⋅ (4.38)
2 𝐺 𝜕𝜌 𝐺 𝐺
The formula (4.38) becomes even more interesting when 𝐾 is constant. We now need
an elementary result of Calculus:

Lemma 4.6.9 Let 𝑓 ∶ [𝑎, 𝑏[→ R be a continuous function, differentiable on ]𝑎, 𝑏[,
and such that lim+ 𝑓 ′ (𝑥) exists and is finite. Then 𝑓 is differentiable at 𝑎 and
𝑥→𝑎
𝑓 ′ (𝑎) = lim+ 𝑓 ′ (𝑥).
𝑥→𝑎

Fixing 𝜑, √ say that the continuous extension of 𝜌 (0 <

√ Lemmas 4.6.3 and 4.6.9
𝜌 < 𝛿) ↦ 𝐺(𝜌, 𝜑) defined by 𝐺(0, 𝜑) = 0 is differentiable and its derivative
at√0 has value
√1. Another application of Lemma 4.6.9 (together with the equality
( 𝐺)𝜌𝜌 + 𝐾 √𝐺 = 0) further ensures that it is twice differentiable at 0. We thus have
𝜌 (𝜌 ≥ 0) ↦ 𝐺(𝜌, 𝜑) as solution of the differential equation

𝑥(𝜌) + 𝐾 𝑥(𝜌) = 0 (4.39)

with initial conditions 𝑥(0) = 0 and 𝑥(0) = 1. There are three cases to consider:
√ 1 √ 1 √
i. if 𝐾 < 0 then 𝐺 = √ sinh( −𝐾 𝜌), and therefore 𝐺 = − sinh2 ( −𝐾 𝜌);
√ −𝐾 𝐾
ii. if 𝐾 = 0 then 𝐺 = 𝜌, 𝐺 = 𝜌 2 ;
130 4 The Intrinsic Geometry of Surfaces

√ 1 √ 1 √
iii. if 𝐾 > 0 then𝐺 = √ sin( 𝐾 𝜌), 𝐺 = sin2 ( 𝐾 𝜌).
𝐾 𝐾
We thus conclude that if 𝐾 is constant then the coefficients 𝐸, 𝐹, 𝐺 of the
coordinates Ψ(𝜌, 𝜑) only depend on 𝐾. As a consequence we have the following
result.

Theorem 4.6.10 Any two surfaces of equal constant Gaussian curvature are locally
isometric.

Proof Assume that 𝑆1 and 𝑆2 have the same constant curvature. Given 𝑝 ∈ 𝑆1 and
𝑞 ∈ 𝑆2 , let 𝛿 > 0 be such that 𝐷 𝛿 (𝑝) and 𝐷 𝛿 (𝑞) are normal neighborhoods of
𝑝 and 𝑞. Fixing orthonormal bases (v1 , v2 ) on 𝑇 𝑝 𝑆1 and (w1 , w2 ) on 𝑇𝑞 𝑆2 , let
𝐿∶ 𝑇 𝑝 𝑆1 → 𝑇𝑞 𝑆2 be the linear isometry such that 𝐿(v𝑖 ) = w𝑖 (𝑖 = 1, 2). We shall see
−1
that 𝑓 = exp𝑞 ○𝐿 ○ exp 𝑝 ∣ 𝐵 ( 𝑝) is an isometry of 𝐷 𝛿 (𝑝) onto 𝐷 𝛿 (𝑞): in fact, 𝑓 is
𝛿
clearly a diffeomorphism. Furthermore, the geodesic coordinates Ψ(𝜌, 𝜑) in 𝐷 𝛿 (𝑝)
̃
associated with (v1 , v2 ) are sent by 𝑓 to the geodesic coordinates Ψ(𝜌, 𝜑) in 𝐷 𝛿 (𝑞)
associated with (w1 , w2 ). From what we have seen above, Ψ and Ψ ̃ have the same
coefficients, and therefore 𝑓 ∣𝐷 ( 𝑝)∖{ 𝑝} = Ψ ̃ ○ Ψ−1 is an isometry. Thus, and since
𝛿
𝐷 𝑓𝑞 = 𝐿, we see that 𝐷 𝑓𝑟 is a linear isometry for all 𝑟 ∈ 𝐷 𝛿 (𝑝), which completes
the proof. ◻

Every surface of constant Gaussian curvature is thus locally isometric to the

pseudosphere (𝐾 < 0), to the plane (𝐾 = 0), or to the sphere (𝐾 > 0), but these
three examples do not exhaust all surfaces of constant curvature (see, in Section 3.2,
example 3.2.1 and Exercise 63). Later on, however, we will see that any compact
surface in R 3 of constant Gaussian curvature is a sphere.

Exercises
100. (a) Check that the differential equations of the geodesics of a surface of revolution
parametrized by
Φ(𝑢, 𝑣) = (𝜌(𝑣) cos 𝑢, 𝜌(𝑣) sin 𝑢, 𝑧(𝑣))
are
⎧
⎪ ′′ 2𝜌 ′ ′
⎪
⎪𝑢 + 𝜌 𝑢 𝑣 = 0,
⎪
⎨ 𝜌𝜌 + 𝑧𝑧
⎪
⎪
⎪
𝜌𝜌
𝑣 ′′ − 2 2 𝑢 ′2 + 2 2 𝑣 ′2 = 0
⎪
⎩ 𝜌 +𝑧 𝜌 +𝑧
— where 𝜌, 𝑧, etc. denote the derivatives with respect to 𝑣, and 𝑢 ′ , 𝑣 ′ , etc. the
derivatives with respect to the curve parameter.
(b) Use these equations to conclude again that any meridian 𝑢 = constant is a
geodesic, and that the parallel 𝑣 = 𝑣 0 is a geodesic if and only if the tangent line to
the generating curve at 𝑣 0 is parallel to the 𝑧-axis.
(c) Show that if 𝛾(𝑠) = Φ(𝑢(𝑠), 𝑣(𝑠)) is a geodesic parametrized by arc length
then (𝜌(𝑠)) 𝑢 ′ (𝑠) is constant. Check that (𝜌(𝑠)) 𝑢 ′ (𝑠) = 𝜌(𝑠) cos 𝜃(𝑠), where
2 2
4.6 Minimizing Properties of Geodesics 131

𝜌(𝑠) = 𝜌 ○ 𝑣(𝑠) is the radius of the parallel where 𝛾 lies at time 𝑠 and 𝜃(𝑠) is the
angle of intersection of the curve with that parallel. (The equality 𝜌 cos 𝜃 = constant
is the Clairaut Equation, and plays a fundamental role in the study of geodesics on
surfaces of revolution.)
101. (In this exercise use the Clairaut Equation from Exercise 100.) Consider the
geodesic 𝛾 that starts from a point 𝑝 on the upper half (𝑧 > 0) of the hyperboloid of
revolution 𝑥 2 + 𝑦 2 − 𝑧2 = 1 and makes angle 𝜃 given by cos 𝜃 = 𝜌1 with the parallel
(with radius 𝜌) passing through 𝑝. Put 𝛾(𝑠) = (𝜌(𝑠) cos 𝜑(𝑠), 𝜌(𝑠) sin 𝜑(𝑠), 𝑧(𝑠)),
𝛾(0) = 𝑝. Show that:
(a) while 𝛾(𝑠) stays in the upper half 𝑧 > 0, however 𝑧′ (𝑠) ≠ 0;
(b) if 𝑧′ (0) < 0 then

𝑧(𝑠) > 0 ∀ 𝑠 ∈ R , lim 𝑧(𝑠) = 0, lim 𝑧′ (𝑠) = 0, lim 𝜑′ (𝑠) = 1.

𝑠→+∞ 𝑠→+∞ 𝑠→+∞

102. Consider, on the paraboloid of revolution 𝑧 = 𝑥 2 +𝑦 2 , the geodesic 𝛼(𝑡) = (0, 𝑡, 𝑡 2 ),

𝑡 ∈R.
(a) Show that there exists 𝜀 > 0 such that for all 0 < 𝑡0 ≤ 𝜀, 𝛼∣[−𝑡 ,𝑡 ] minimizes the
0 0
intrinsic distance between 𝛼(−𝑡 0 ) and 𝛼(𝑡0 ).
(b) Show that, for 𝑡0 sufficiently large, 𝛼∣[−𝑡 ,𝑡 ] no longer minimizes the distance
0 0
between 𝛼(−𝑡0 ) and 𝛼(𝑡0 ).
103. Let 𝑆 be a connected surface on which the sum of the interior angles of any
geodesic triangle is equal to 𝜋. Show that 𝑆 is locally isometric to a plane.
104. Show that if all geodesics of a connected surface are planar curves then that
surface is contained in a sphere or in a plane.
105. Let 𝑆 be a surface of constant curvature, and let Δ1 , Δ2 be geodesic triangles of
𝑆. Assuming that Δ1 , Δ2 are “sufficiently small”, show that:
(a) if two of the sides of Δ1 are equal then the angles opposite to those sides are
also equal;
(b) if Δ1 and Δ2 have pairwise equal sides then there exist open subsets 𝑊1 ⊇ Δ1 ,
𝑊2 ⊇ Δ2 and an isometry 𝑓 ∶ 𝑊1 → 𝑊2 for which 𝑓 (Δ1 ) = Δ2 . Hint: consider first the
case where the Δ𝑖 have the same angle and sides adjacent to that angle.
106. Show that on the surface of revolution (𝜌(𝑧) cos 𝜑, 𝜌(𝑧) sin 𝜑, 𝑧) (where 𝜌(𝑧) > 0
for all 𝑧 ∈ R ), the only minimizing geodesic between two points on the same meridian
𝜑 = constant is precisely that meridian.
107. Let 𝑆 be a connected surface, 𝑝 a point on 𝑆, and let 𝛼∶ ]𝑎, 𝑏[ → 𝑆 be a regular
curve that does not pass through 𝑝. Consider the intrinsic distance on 𝑆.
Show that if 𝛼(𝑡0 ) is a point of 𝛼 at the minimizing distance from 𝑝 and 𝛾 a
minimizing geodesic starting from 𝑝 to 𝛼(𝑡0 ) then 𝛾 intersects 𝛼 orthogonally.
108. Let 𝑆𝐶 (𝐶 ≠ 1) be the surface of revolution of constant curvature 1 given in
example 3.2.1 and 𝛾 ⊆ 𝑆𝐶 the meridian 𝑢 = 𝜋.
132 4 The Intrinsic Geometry of Surfaces

(a) Explicitly define a local isometry 𝑓 ∶ 𝑆𝐶 ∖ 𝛾 → S2 . Hint: send the equator of

𝑆𝐶 to the equator of S2 and use spherical coordinates.
(b) The equator 𝑣 = 0 is a closed (i.e., periodic) geodesic of 𝑆𝐶 . Show that 𝑆𝐶 has
other closed geodesics if and only if 𝐶 is rational.
109. Consider a geodesic polar coordinate system (𝜌, 𝜑) centered at a point 𝑝 0 ∈ 𝑆
with curvature 𝐾(𝑝 0 ). Prove that:
√ 𝐾(𝑝 0 ) 3 𝑜(𝜌 3 )
(a) 𝐺 = 𝜌− 𝜌 + 𝑜(𝜌 3 ), where lim = 0 uniformly on 𝜑.
6 𝜌→0 𝜌 3
Hint:√Lemma 4.6.9 and formula (4.38) should show that, given 𝜑, the function
𝜌 ↦ 𝐺(𝜌, 𝜑) admits a Taylor polynomial expansion around 0 as above — the
problem lies in proving that the said limit is uniform on 𝜑.
(b) by denoting by 𝑙(𝜌) the perimeter of the geodesic circle with radius 𝜌 centered
6𝜋𝜌 − 3𝑙(𝜌)
at 𝑝 0 at 𝐾(𝑝 0 ) = lim ⋅
𝜌→0 𝜋𝜌 3
110. (a) Show that in geodesic polar coordinates (𝜌, 𝜑) the geodesic curvature of the
𝐺𝜌
geodesic circumferences 𝜌 = constant is given by ⋅
2𝐺
(b) Conclude that these circles all have constant geodesic curvature if and only if
there exist differentiable and strictly positive functions 𝛽(𝜌) and 𝜆(𝜑) such that

𝐺(𝜌, 𝜑) = 𝛽(𝜌)𝜆(𝜑). (4.40)

(c) Show that if 𝐺 is of the form (4.40) then the Gaussian curvature along each
geodesic circumference 𝜌 = constant is constant.
(d) Conclude that the only oriented connected surfaces on which any geodesic
circumference has constant geodesic curvature are surfaces of constant Gaussian
curvature.
111. Show that a conformal mapping 𝑓 ∶ 𝑆1 → 𝑆2 that sends geodesics of 𝑆1 to
geodesics of 𝑆2 is necessarily a similarity — i.e., there exists 𝜆 > 0 such that
∣𝐷 𝑓 𝑝 (v)∣ = 𝜆∣v∣ for all 𝑝 ∈ 𝑆1 and v ∈ 𝑇 𝑝 𝑆1 .
112. Let 𝑝 be a point of 𝑆 and 𝛼(𝑢), ∣𝑢∣ < 𝜀, a regular curve in 𝑆 such that 𝛼(0) = 𝑝.
Choose along 𝛼 a unit vector field w(𝑢) of tangent vectors orthogonal to 𝛼′ (𝑢) and
write Φ(𝑢, 𝑣) = 𝛾(𝑣 − 𝑣 0 ; 𝛼(𝑢), w(𝑢)) (i.e., Φ(𝑢, ⋅) is the geodesic that at time 𝑣 0
passes through 𝛼(𝑢) with velocity w(𝑢)). Show that:
(a) Φ(𝑢, 𝑣) is a parameterization in a neighborhood of 𝑝 (the coordinates Φ(𝑢, 𝑣)
obtained this way are called semi-geodesic);
(b) Φ(𝑢, 𝑣) is an orthogonal parameterization;
(c) geodesic polar coordinates are an example of semi-geodesic coordinates.
113. Let 𝑆 be a connected oriented surface and let 𝑓 ∶ 𝑆 → 𝑆 be a local isometry. Show
that:
(a) if there exists 𝑝 ∈ 𝑆 such that 𝑓 (𝑝) = 𝑝 and 𝐷 𝑓 𝑝 = id ∣𝑇 𝑆 then 𝑓 is the identity
𝑝
(Hint: what are the geodesics that start from 𝑝 mapped to?);
4.6 Minimizing Properties of Geodesics 133

(b) if 𝑓 is not the identity and if there exists a regular curve 𝛼∶ 𝐼 → 𝑆 such that
𝑓 ○ 𝛼 = 𝛼 then 𝑓 reverses orientation of 𝑆;
(c) the curve 𝛼 in b) is a geodesic.

114. Let 𝑆1 , 𝑆2 be connected surfaces, 𝑓 ∶ 𝑆1 → 𝑆2 be a diffeomorphism, and 𝑑1 ,

𝑑2 be the intrinsic distances on 𝑆1 and 𝑆2 . Show that the following conditions are
equivalent:
(1) 𝑓 is an isometry;
(2) 𝑑1 (𝑝, 𝑞) = 𝑑2 ( 𝑓 (𝑝), 𝑓 (𝑞)) for all 𝑝, 𝑞 ∈ 𝑆1 .
(Hint for (2) ⇒ (1); a) show that 𝑓 transforms geodesics of 𝑆1 into geodesics of 𝑆2 ,
preserving scalar velocity; b) use the equivalence between (ii) and (iii) for 𝜆 = 1 in
Lemma 4.1.1, §1.)

115. In this exercise we show that any compact surface has a triangulation (a fact
used when establishing the global version of the Gauss-Bonnet theorem). Fix 𝛿 > 0
so that, for all 𝑝 ∈ 𝑆, 𝐷 𝛿 (𝑝) is a normal neighborhood of 𝑝, and consider a family
(R 𝑘 )1≤ 𝑗≤𝑘 of geodesic triangles such that:

● the interiors of R 𝑗 ( 𝑗 = 1, . . . , 𝑘) cover 𝑆;

● the diameter of each R 𝑗 is 𝛿 (i.e., if 𝑝, 𝑞 ∈ R 𝑗 then 𝑑(𝑝, 𝑞) < 𝛿).

(a) Show that any two edges of two distinct R 𝑗 are either disjoint, or intersect at a
single point, or intersect along an arc common to both. Hence, the intersections of
the R 𝑗 form a finite number of regions Ω𝑙 .
(b) By properly triangulating each region Ω𝑙 , obtain a triangulation of 𝑆.

Appendix: Rotation Index

In this appendix we prove the Rotation Index theorem that we used in the proof of the
Gauss-Bonnet theorem. First however let us explain what we mean by continuous
choice of the angle between two vector fields, an expression used repeatedly in Section
4.5. Let us take two unit vector fields v1 (𝑠) and w(𝑠) along a certain curve 𝛼(𝑠),
for 𝑠 ∈ [𝑎, 𝑏], and consider an auxiliary vector field v2 (𝑠) so that (v1 , v2 ) is, for
every 𝑠, an orthonormal and positively oriented basis of 𝑇𝛼(𝑠) 𝑆. We can then write
w(𝑠) = 𝜆1 (𝑠)v1 (𝑠) + 𝜆2 (𝑠)v2 (𝑠), where

[𝜆1 (𝑠)]2 + [𝜆2 (𝑠)]2 = ∣w(𝑠)∣2 = 1.

This shows that 𝑠 ↦ (𝜆 1 (𝑠), 𝜆 2 (𝑠)) is a mapping of [𝑎, 𝑏] into the unit circle S1 . If
the vector fields considered are differentiable, then also this mapping is differentiable,
and therefore (see note 1.3.1 and Exercise 7 in Section 1.3) there exists a differentiable
mapping 𝜑(𝑠) such that (𝜆1 (𝑠), 𝜆 2 (𝑠)) = (cos 𝜑(𝑠), sin 𝜑(𝑠)) — that is, such that

w(𝑠) = cos 𝜑(𝑠)v1 (𝑠) + sin 𝜑(𝑠)v2 (𝑠).

134 4 The Intrinsic Geometry of Surfaces

It is this function 𝜑(𝑠) that we call the continuous choice of the oriented angle
between v1 (𝑠) and w(𝑠). As we noted in note 1.3.1, any other choice of the same
angle is the sum of 𝜑(𝑠) with an integer multiple of 2𝜋.
For later use, it is convenient to obtain a description like the one we saw above for
functions [𝑎, 𝑏] → S1 that are not necessarily differentiable:
A.1 Lemma. Let 𝐹∶ [𝑎, 𝑏] → S1 be a continuous function. Then there is a lift of 𝐹,
that is, a continuous function 𝜑∶ [𝑎, 𝑏] → R such that 𝐹(𝑠) = (cos 𝜑(𝑠), sin 𝜑(𝑠)) for
all 𝑠 on [𝑎, 𝑏]. Every other lift of 𝐹 is the sum of 𝜑 with a constant integer multiple
of 2𝜋.

Proof Consider the mapping Π(𝑡) = (cos 𝑡, sin 𝑡) which wraps the line R into
S1 . Note that the restriction of Π to any interval [𝑡 1 , 𝑡 2 ] with 𝑡2 − 𝑡1 < 2𝜋 is a
homeomorphism onto its image, since [𝑡 1 , 𝑡 2 ] is compact and Π∣[𝑡 ,𝑡 ] is continuous
1 2
and injective. What we are looking for is a continuous function 𝜑 such that 𝐹 = Π ○ 𝜑.
The idea is to restrict 𝐹 to small intervals where we can apply a local inverse of Π to
both sides of this equality.
By uniform continuity of 𝐹, there exists 𝛿 > 0 such that for ∣𝑠 − 𝑡∣ < 𝛿, the
points 𝐹(𝑠) and 𝐹(𝑡) are never diametrically opposite in S1 . If we take a partition
𝑠0 < 𝑠1 < ⋯ < 𝑠 𝑘 of [𝑎, 𝑏] with 𝑠𝑖 − 𝑠𝑖−1 < 𝛿, then, for every 1 ≤ 𝑖 ≤ 𝑘, the image
𝐹([𝑠𝑖−1 , 𝑠𝑖 ]) is contained in a semi-circle. Let us now define 𝜑 recursively, starting
at the interval [𝑠0 , 𝑠1 ]. By construction, there exists an interval 𝐽1 , of amplitude
𝜋, such that the arc Π(𝐽1 ) contains 𝐹([𝑠0 , 𝑠1 ]): thus, for 𝑠 ∈ [𝑠0 , 𝑠1 ], we define
−1
𝜑(𝑠) = Π∣ 𝐽 ○ 𝐹(𝑠). Assuming we have defined 𝜑(𝑠) for all 𝑠 ∈ [𝑠0 , 𝑠𝑖−1 ], we take 𝐽𝑖 ,
1
of amplitude 𝜋, such that 𝜑(𝑠𝑖−1 ) ∈ 𝐽𝑖 and 𝐹([𝑠𝑖−1 , 𝑠𝑖 ]) ⊆ Π(𝐽𝑖 ), for 𝑠 ∈ [𝑠𝑖−1 , 𝑠𝑖 ], we
define 𝜑(𝑠) = Π∣−1
𝐽𝑖 ○ 𝐹(𝑠). This ends the construction of 𝜑. It is clear that 𝐹 = Π ○ 𝜑.
Since 𝜑 is continuous (because it is continuous on every interval [𝑠𝑖−1 , 𝑠𝑖 ]), 𝜑 is a lift
of 𝐹.
Regarding uniqueness of 𝜑 (minus a constant), the proof is given in note 1.3.1. ◻

In general, a lift(ing) of a continuous function 𝐹∶ Ω → S1 , where Ω is a domain

of R 𝑛 , is a continuous function 𝜑∶ Ω → R such that 𝐹 = Π ○ 𝜑. If we do not impose
conditions on Ω, it is not true that all such functions 𝐹 have a lifting. But we can
guarantee the existence of a lifting if, for example, Ω is a rectangle (i.e., the Cartesian
product of compact intervals): the next lemma proves this in the two-dimensional
case, the only one we need besides the case 𝑛 = 1 treated in A.1.

A.2. Lemma. Let 𝐹(𝑠, 𝑡) be a continuous function [𝑎, 𝑏] × [𝑐, 𝑑] → S1 . Then

𝐹 has a lifting — that is, there exists 𝜑(𝑠, 𝑡) continuous such that 𝐹(𝑠, 𝑡) =
(cos 𝜑(𝑠, 𝑡), sin 𝜑(𝑠, 𝑡)).

Proof Let 𝜃(𝑡) be a lifting of the mapping 𝑡 ↦ 𝐹(0, 𝑡), whose existence is guaranteed
by A.1. Again, using A.1, let us next take, for every 𝑡 ∈ [𝑐, 𝑑], a lifting 𝑠 ↦ 𝜑(𝑠, 𝑡) of
𝑠 ↦ 𝐹(𝑠, 𝑡) that satisfies 𝜑(0, 𝑡) = 𝜃(𝑡). The function 𝜑 obtained this way satisfies
the equality 𝐹 = Π ○ 𝜑, and it remains to show that 𝜑 is continuous. By construction,
4.6 Minimizing Properties of Geodesics 135

its restriction to each of the horizontal segments [𝑎, 𝑏] × {𝑡}, and to the vertical line
segment {0} × [𝑐, 𝑑], is continuous.
Given 𝜀 > 0 with 𝜀 < 𝜋2 , the uniform continuity of 𝐹 gives us 𝛿 > 0 such that
the angle between 𝐹(𝑠, 𝑡) and 𝐹(̃ 𝑠, ̃
𝑡 ) has amplitude 𝜀 whenever ∣(𝑠, 𝑡) − (̃ 𝑠, ̃
𝑡 )∣ < 𝛿.
Thus, given (𝑠0 , 𝑡 0 ), and under the assumption that ∣(𝑠, 𝑡) − (𝑠0 , 𝑡 0 )∣ < 𝛿, we can write

𝜑(𝑠, 𝑡) − 𝜑(𝑠0 , 𝑡 0 ) = 𝜀(𝑠, 𝑡) + 2𝜋𝑘(𝑠, 𝑡),

where 𝑘(𝑠, 𝑡) is an integer and ∣𝜀(𝑠, 𝑡)∣ < 𝜀. If we show that 𝑘(𝑠, 𝑡) = 0, the
continuity of 𝜑 on (𝑠0 , 𝑡 0 ) follows. Now, since 𝜑∣{0}×[𝑐,𝑑] is continuous, we have
∣𝜑(0, 𝑡) − 𝜑(0, 𝑡 0 )∣ < 𝜀 < 𝜋2 . Furthermore, since the difference 𝜑(⋅, 𝑡) − 𝜑(⋅, 𝑡 0 ) is
continuous and takes neither of the values ± 𝜋2 , it takes values in the interval ]− 𝜋2 , 𝜋2 [;
we conclude that ∣𝜑(𝑠, 𝑡) − 𝜑(𝑠, 𝑡 0 )∣ < 𝜋2 . But as also ∣𝜑(𝑠, 𝑡 0 ) − 𝜑(𝑠0 , 𝑡 0 )∣ < 𝜋2 , it
follows that

𝜋 > ∣𝜑(𝑠, 𝑡) − 𝜑(𝑠0 , 𝑡 0 )∣ = ∣𝜀(𝑠, 𝑡) + 2𝜋𝑘(𝑠, 𝑡)∣ ≥ 2𝜋∣𝑘(𝑠, 𝑡)∣ − ∣𝜀(𝑠, 𝑡)∣,

and therefore 𝑘(𝑠, 𝑡) = 0. This concludes the proof of the continuity of 𝜑. ◻

Here, in a simplified version, is the Rotation Index theorem:

A.3. Rotation Index theorem (first version). If 𝛼∶ [𝑎, 𝑏] → R 2 is a closed, regular,
simple curve, then ℜ(𝛼) = ±1, where the sign depends on the orientation of the
curve.

Proof We suppose, as usual, that 𝛼 is parametrized by arc length, and define a

continuous function 𝐹∶ [𝑎, 𝑏] × [𝑎, 𝑏] → S1 as follows: if 0 < ∣𝑠 − 𝑡∣ < 𝑏 − 𝑎, we put

𝛼(𝑠) − 𝛼(𝑡)
𝐹(𝑠, 𝑡) = ;
∣𝛼(𝑠) − 𝛼(𝑡)∣

in the other cases, we put 𝐹(𝑠, 𝑠) = 𝛼′ (𝑠) and 𝐹(𝑎, 𝑏) = −𝐹(𝑏, 𝑎) = 𝛼′ (𝑎). Denoting
by 𝜑(𝑠, 𝑡) a lifting of 𝐹(𝑠, 𝑡), the restriction of 𝜑 to the diagonal {(𝑠, 𝑠)∶ 𝑡𝑜 ≤ 𝑠 ≤ 𝑏}
is a lifting of 𝑠 ↦ 𝛼′ (𝑠), and therefore the rotation index is

1
ℜ(𝛼) = (𝜑(𝑏, 𝑏) − 𝜑(𝑎, 𝑎)).
2𝜋
Suppose that the initial point 𝛼(𝑎) is chosen so that the curve 𝛼 is all on the same
side of its tangent line at that point (such a choice is always possible: see for example
Section 1.6). Then the image of 𝐹(𝑠, 𝑎), for 𝑠 ∈ [𝑎, 𝑏], is contained in a semi-circle,
and therefore its lift 𝜑(𝑠, 𝑎) covers at most an interval of amplitude 𝜋. But since
𝐹(𝑎, 𝑎) = −𝐹(𝑏, 𝑎), we have

𝜑(𝑏, 𝑎) − 𝜑(𝑎, 𝑎) = 𝜀𝜋,

where 𝜀 = 1 if 𝛼 is positively oriented, and 𝜀 = −1 otherwise. Since 𝐹(𝑏, 𝑡), 𝑡 ∈ [𝑎, 𝑏],
runs exactly along the curve at S1 diametrically opposite to 𝐹(𝑠, 𝑎), we also have
136 4 The Intrinsic Geometry of Surfaces

𝜑(𝑏, 𝑏) − 𝜑(𝑏, 𝑎) = 𝜀𝜋.

Adding these two equalities, we obtain ℜ(𝛼) = 𝜀, which concludes the proof. ◻
Before we move to a generalization of Theorem A.3, we note that the above proof
works under the assumption that 𝛼 is only of class 𝐶 1 .
We will now show that A.3 remains valid for curves on surfaces, provided
such curves are contained in parametrized neighborhoods. Let 𝛼(𝑠), for 𝑠 ∈ [𝑎, 𝑏],
be a closed, simple, regular curve of class 𝐶 1 , contained in the image of the
parameterization Φ(𝑢, 𝑣) that we assume to be compatible with the orientation of the
1
surface. Let v1 = √ Φ𝑢 , and let 𝜃(𝑠) be a continuous choice of the oriented angle
𝐸
between v1 and 𝛼′ (𝑠). We claim that if 𝛼 is positively oriented, then 𝜃(𝑏) − 𝜃(𝑎) = 2𝜋.
The idea, of course, is to apply Theorem A.3 to the planar curve 𝛽 = Φ−1 ○ 𝛼.
If Φ were an isothermal parameterization, our statement would be an immediate
consequence of A.3, but it is unnecessary to invoke such a strong result as the
existence of isothermal parameterizations. Let 𝜑(𝑠) be a continuous choice of the
𝛽′ (𝑠)
angle between (1, 0) and 𝝉(𝑠) = ′ . For every instant 𝑠, and since Φ preserves
∣𝛽 (𝑠)∣
orientation and 𝐷Φ𝛽(𝑠) sends (1, 0) and 𝝉(𝑠) to vectors that are positive multiples of
v1 and 𝛼′ (𝑠), the dihedra ((1, 0), 𝝉(𝑠)) and (v1 , 𝛼′ (𝑠)) are both positively oriented
or both negatively oriented. Hence, 𝜑(𝑠) − 𝜃(𝑠) ≠ ±𝜋 for all 𝑠; choosing 𝜃(𝑎) and
𝜑(𝑎) with ∣𝜑(𝑎) − 𝜃(𝑎)∣ < 𝜋, we also have ∣𝜑(𝑏) − 𝜃(𝑏)∣ < 𝜋. It follows that

∣{𝜑(𝑏) − 𝜑(𝑎)} − {𝜃(𝑏) − 𝜃(𝑎)}∣ < 2𝜋

or, using A.3,

∣2𝜋 − (𝜃(𝑏) − 𝜃(𝑎)}∣ < 2𝜋,
and therefore, since 𝜃(𝑏) − 𝜃(𝑎) is an integer multiple of 2𝜋,

𝜃(𝑏) − 𝜃(𝑎) = 2𝜋,

which proves our assertion.

Let us now state the more general version of Theorem A.3. used in the proof of
the Gauss-Bonnet theorem. Consider a curve 𝛼(𝑠), for 𝑠 ∈ [𝑎, 𝑏], closed, simple,
piecewise regular, which is the boundary of a polygonal region Ω contained in a
neighborhood parametrized by Φ(𝑢, 𝑣). Let 𝑎 = 𝑠0 < 𝑠1 < ⋯ < 𝑠 𝑘 = 𝑏 be the 𝑘 + 1
vertices of 𝛼.

A.4. Rotation Index theorem (second version). Let 𝛾𝑖 , for 𝑖 = 1, . . . , 𝑘, be the

exterior angles of Ω and 𝜃 𝑖 (𝑠), for 𝑠 ∈ [𝑠𝑖−1 , 𝑠𝑖 ], a continuous choice of the angle
1
between v1 = √ Φ𝑢 and 𝛼′ (𝑠). Then, if 𝛼 runs through 𝜕Ω with positive orientation,
𝐸
we have
𝑘 𝑘
∑ {𝜃 𝑖 (𝑠𝑖 ) − 𝜃 𝑖 (𝑠𝑖−1 )} + ∑ 𝛾𝑖 = 2𝜋.
𝑖=1 𝑖=1
4.6 Minimizing Properties of Geodesics 137

For the proof, we approximate 𝛼 by curves 𝛼 𝜀 of class 𝐶 1 , obtained by rounding

each of the vertices of 𝛼. For sufficiently small 𝜀 > 0, 𝛼 𝜀 ∶ [0, 𝐿 𝜀 ] → 𝑆 is thus a
closed, simple, regular curve, parametrized by arc length, for which there exist
disjoint and consecutive intervals [𝑎 𝑖𝜀 , 𝑏 𝑖𝜀 ] ⊆ [0, 𝐿 𝜀 ] (𝑖 = 1, . . . , 𝑘) such that each arc
𝛼 𝜀 ([𝑎 𝑖𝜀 , 𝑏 𝑖𝜀 ]) is a segment of the regular arc 𝛼([𝑠𝑖−1 , 𝑠𝑖 ]), traversed with the same
orientation. Furthermore, the total length of the arcs of 𝛼 and 𝛼 𝜀 that are not common
to both curves, i.e., in the complement of
𝑘
⋃ 𝛼 𝜀 ([𝑎 𝑖 , 𝑏 𝑖 ]),
𝜀 𝜀
𝑖=1

is 𝜀.
Let us denote by 𝜃 𝜀 (𝑠) a continuous choice of the angle between v1 and 𝛼′𝜀 (𝑠). It
is clear then that each of the differences

𝜃 𝜀 (𝑏 𝑖𝜀 ) − 𝜃 𝜀 (𝑎 𝑖𝜀 )

comes arbitrarily close to 𝜃 𝑖 (𝑠𝑖 ) − 𝜃 𝑖 (𝑠𝑖−1 ), taking 𝜀 sufficiently small, and in this
way also
𝜃 𝜀 (𝑎 𝑖+1
𝜀
) − 𝜃 𝜀 (𝑏 𝑖𝜀 ), for 𝑖 = 1, . . . , 𝑘 − 1,
and 𝜃 𝜀 (𝑎 1𝜀 ) − 𝜃 𝜀 (0) + 𝜃 𝜀 (𝐿 𝜀 ) − 𝜃 𝜀 (𝑏 𝑘𝜀 ), for 𝑖 = 𝑘,
lie close to 𝛾𝑖 for 𝑖 = 1, . . . , 𝑘. We thus conclude that

2𝜋 = 𝜃 𝜀 (𝐿 𝜀 ) − 𝜃 𝜀 (0)
𝑘
= {𝜃 𝜀 (𝐿 𝜀 ) − 𝜃 𝜀 (𝑏 𝑘𝜀 )} + ∑ {𝜃 𝜀 (𝑏 𝑖𝜀 ) − 𝜃 𝜀 (𝑎 𝑖𝜀 )} +
𝑖=1
𝑘−1
+ ∑ {𝜃 𝜀 (𝑎 𝑖+1
𝜀
) − 𝜃 𝜀 (𝑏 𝑖𝜀 )} + {𝜃 𝜀 (𝑎 1𝜀 ) − 𝜃 𝜀 (0)}
𝑖=1

is arbitrarily close to
𝑘 𝑘
∑ {𝜃 𝑖 (𝑠𝑖 ) − 𝜃 𝑖 (𝑠𝑖−1 )} + ∑ 𝛾𝑖 ,
𝑖=1 𝑖=1

which proves the theorem.

Exercises

116. Lemma A.2 applies not only to rectangles, but also, obviously, to any regions that
are homeomorphic to rectangles. For example, any continuous function 𝐹∶ D2 → S1 ,
where D2 is the closed disk {𝑝 ∈ R 2 ∶ ∣𝑝∣ ≤ 1}, has a lifting. Using this fact, we next
give a proof of Brouwer’s fixed point theorem: any continuous function 𝑓 ∶ D2 → D2
has some fixed point (i.e., a point 𝑝 such that 𝑓 (𝑝) = 𝑝).
138 4 The Intrinsic Geometry of Surfaces

Assuming that 𝑓 has no fixed points, we define 𝐹∶ D2 → S1 by

𝑓 (𝑝) − 𝑝
𝐹(𝑝) = ,
∣ 𝑓 (𝑝) − 𝑝∣

and consider a lifting 𝜑∶ D2 → R of 𝐹.

(a) Write 𝛾(𝑡) = 𝐹(cos 𝑡, sin 𝑡), for 𝑡 ∈ [0, 2𝜋], and let 𝜃(𝑡) be a lifting of 𝛾 such
that 𝜃(0) ∈ ] 𝜋2 , 32𝜋 [. Show that 𝜃(𝑡) ∈ ]𝑡 + 𝜋2 , 𝑡 + 32𝜋 [ for all 𝑡, and conclude that
𝜃(2𝜋) − 𝜃(0) = 2𝜋.
̃ = 𝜑(cos 𝑡, sin 𝑡) is also a lift of 𝛾(𝑡), but that 𝜃(2𝜋)
(b) Note that 𝜃(𝑡) ̃ ̃
= 𝜃(0).
What follows?
117. Let Ω ⊆ R 2 be an open disk and let w∶ Ω → R 2 be a differentiable field of unit
vectors. Given a regular closed curve 𝛼∶ [𝑎, 𝑏] → Ω, show that the rotation index of 𝛼
is also equal to
1
(𝜃(𝑏) − 𝜃(𝑎)),
2𝜋
where 𝜃 is a continuous choice of the (oriented) angle of w(𝛼(𝑡)) and 𝛼′ (𝑡).
118. Consider, in T 2 with the parameterization of Exercise 31, the vector field
w(𝑢, 𝑣) = (− sin 𝑣 cos 𝑢, − sin 𝑣 sin 𝑢, cos 𝑣). Let the rotation index of a regular closed
curve 𝛼∶ [𝑎, 𝑏] → T 2 be the integer

1
Δ(𝛼) = (𝜃(𝑏) − 𝜃(𝑎)),
2𝜋
where 𝜃 is a continuous choice of the oriented angle of w and 𝛼′ (𝑡).
(a) Compute Δ(𝛼) for the curve 𝑡 ↦ Φ(𝑡, 𝑛𝑡), where 𝑛 ∈ Z is a constant and
𝑡 ∈ [0, 2𝜋].
(b) Does the result of (a) depend on the vector field w?
Chapter 5
The Global Geometry of Surfaces

Global geometry deals with those results that concern the surface as a whole. In
Chapter 4 we have already seen examples of global theorems, such as the divergence
theorem (Corollary 4.4.3.i) and the Gauss-Bonnet theorem (Theorem 4.5.6). Another
example is the sphere theorem that we will present in this chapter: any compact surface
of constant curvature in R 3 is a sphere. As we already observed, the assumption
that the curvature is constant is insufficient, and a global condition (in this case, the
compactness of the surface) is needed to draw such a conclusion.
Compact surfaces are inextensible, in the following sense: if 𝑆1 and 𝑆2 are
connected surfaces such that 𝑆1 is compact and 𝑆1 ⊆ 𝑆2 then 𝑆1 = 𝑆2 . The global
results must naturally deal with inextensible surfaces. In Section 5.1 we will define
the notion of a complete surface, which is a sufficient but not necessary condition for
a surface to be inextensible.
This chapter includes a mixed bag of topics unusual in introductory texts of
Differential Geometry: for instance, a Blaschke formula for surfaces of constant width
(Theorem 5.5.3) and the description of all complete surfaces of constant non-negative
curvature (Theorems 5.7.7 and 5.7.10). To help the readers in their choice of topics,
we mention that Sections 5.1 and 5.2 form the main body of the chapter, from which
two independent branches emerge, one consisting of 5.4 and 5.5 and the other one of
5.3, 5.6 and 5.7.
All surfaces in this chapter are connected.

5.1 Complete Surfaces

In Chapter 4 we gave an example (4.6.5) of a surface on which no minimizing geodesic

exists between given two points. Even though this example seems disingenuous (the
surface in question is a plane from which a point has been removed “unduly”), it
points out a possible problem: although there is a geodesic 𝛾 that starts from 𝑝 in the
direction of 𝑞, 𝛾 is not defined for some value of the parameter and therefore does
not reach 𝑞.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 139
P. V. Araújo, Differential Geometry, https://doi.org/10.1007/978-3-031-62384-4_5
140 5 The Global Geometry of Surfaces

We say that a surface 𝑆 is complete if any parametrized geodesic 𝛾(𝑡; 𝑝, v) of 𝑆

is defined for all 𝑡 ∈ R . Equivalently, 𝑆 is complete if exp 𝑝 is defined on the entire
tangent space 𝑇 𝑝 𝑆 for all 𝑝 ∈ 𝑆. The next result gives us many examples of complete
surfaces.
Proposition 5.1.1 If a surface 𝑆 ⊆ R 3 is closed in R 3 then 𝑆 is complete.
Proof Given 𝑝 ∈ 𝑆 and v ∈ 𝑇 𝑝 𝑆 with ∣v∣ = 1, we verify that the geodesic 𝛾(𝑠) =
𝛾(𝑠; 𝑝, v) is defined for every 𝑠 ∈ [0, +∞[ (replacing v by −v, it follows that 𝛾(𝑠) is
also defined on the entire interval ] −∞, 0]). For this, it suffices to check that if 𝛾 is
defined on [0, 𝑠0 [ then it is defined on some interval [0, 𝑠0 + 𝛿[ for 𝛿 > 0 as well.
Take a monotone sequence (𝑠 𝑛 )𝑛≥1 on [0, 𝑠0 [ that converges to 𝑠0 . Since

∣𝛾(𝑠 𝑛 ) − 𝛾(𝑠 𝑚 )∣ ≤ 𝑑(𝛾(𝑠 𝑛 ), 𝛾(𝑠 𝑚 )) ≤ ∣𝑠 𝑛 − 𝑠 𝑚 ∣ ,

it follows that (𝛾(𝑠 𝑛 ))𝑛≥1 is a Cauchy sequence in R 3 . Since 𝑆 is closed, the limit 𝑝
of this sequence is a point of 𝑆. We can then choose 𝜂 > 0 and a neighborhood 𝑈
of 𝑝 in 𝑆 such that, for all 𝑞 ∈ 𝑈, the radial geodesics starting from 𝑞 have length at
least 𝜂. We fix 𝑛 so that 𝑠 𝑛 > 𝑠0 − 𝜂2 and 𝛾(𝑠 𝑛 ) ∈ 𝑈, and let 𝑞 = 𝛾(𝑠 𝑛 ), w = 𝛾 ′ (𝑠 𝑛 ).
The geodesic 𝛾̃(𝑡) = 𝛾(𝑡; 𝑞, w) is defined for 𝑡 ∈ ]−𝜂, 𝜂[, and 𝛾(𝑠) = 𝛾̃(𝑠 − 𝑠 𝑛 ). Thus
𝛾(𝑠) is extensible to the interval [0, 𝑠 𝑛 + 𝜂[ ⊇ [0, 𝑠0 + 𝜂2 [. ◻

In particular, it follows from the above proposition that all compact surfaces are
complete. However, we observe that there are complete surfaces other than those
closed in R 3 :
Example 5.1.2 Let 𝑆 be the surface given by

Φ(𝑢, 𝑣) = ((1 + 𝑒 −𝑢 ) cos 𝑢, (1 + 𝑒 −𝑢 ) sin 𝑢, 𝑣),

𝑢, 𝑣 ∈ R . 𝑆 is complete because it is isometric to the plane, which is a complete

surface. However, 𝑆 is not a closed subset of R 3 : a point of the cylinder 𝑥 2 + 𝑦 2 = 1
lies outside of 𝑆, and is moreover the limit of a sequence of points in 𝑆 (see Fig. 5.1).

Figure 5.1
5.1 Complete Surfaces 141

The intrinsic distance 𝑑 in 𝑆 and the distance ∣ ∣ in R 3 are equivalent, in the sense
that a sequence (𝑝 𝑛 ) in 𝑆 converges to 𝑝 ∈ 𝑆 for one of these distances if and only if
it converges for the other one. However, this example shows that they do not have to
be uniformly equivalent: there can be a sequence of points on 𝑆 that is Cauchy for ∣ ∣
but is not so for 𝑑. ◻

The major advantage of complete surfaces is that there exist minimizing geodesics
between any two points on them. Before we prove this, we note that it follows from the
triangle inequality (see Section 4.6) that, for every 𝑝 ∈ 𝑆, the mapping 𝑞 ↦ 𝑑(𝑝, 𝑞)
is continuous (since ∣𝑑(𝑝, 𝑞) − 𝑑(𝑝, 𝑟)∣ ≤ 𝑑(𝑞, 𝑟) ∀ 𝑞, 𝑟 ∈ 𝑆) — and therefore its
restriction to any compact 𝑆 attains a maximum and a minimum.

Theorem 5.1.3 Let 𝑝, 𝑞 be two points of a complete surface 𝑆. Then there is on 𝑆

some minimizing geodesic from 𝑝 to 𝑞.

Proof Let 𝐷 𝛿 (𝑝) be a normal neighborhood of 𝑝. If 𝑞 ∈ 𝐷 𝛿 (𝑝), there is nothing

to show. Otherwise, letting 𝑙 = 𝑑(𝑝, 𝑞), we have 𝑙 ≥ 𝛿. Consider now the geodesic
circumference S1 (𝑝, 2𝛿 ) with center 𝑝 and radius 2𝛿 , and let 𝑟 be a point of S1 (𝑝, 2𝛿 )
at the shortest possible distance from 𝑞. By the triangle inequality, we have
𝛿
𝑑(𝑟, 𝑞) ≥ 𝑑(𝑝, 𝑞) − 𝑑(𝑝, 𝑟) = 𝑙 − ⋅ (5.1)
2
Let 𝛼(𝑡), for 𝑡 ∈ [𝑎, 𝑏], be any piecewise regular curve from 𝑝 to 𝑞, and let 𝑡0 be the
first instant at which 𝛼(𝑡 0 ) ∈ S1 (𝑝, 2𝛿 ). Then

𝛿
𝑙(𝛼) = 𝑙(𝛼∣[𝑎,𝑡 ] ) + 𝑙(𝛼∣[𝑡 ) ≥ 𝑑(𝑝, 𝛼(𝑡 0 )) + 𝑑(𝛼(𝑡0 ), 𝑞) ≥ + 𝑑(𝑟, 𝑞),
0 0 ,𝑏] 2
which implies 𝑙 ≥ 2𝛿 + 𝑑(𝑟, 𝑞). Together with (5.1), this gives 𝑑(𝑟, 𝑞) = 𝑙 − 2𝛿 . Let
𝛾(𝑠) be the radial geodesic, parametrized by arc length, such that 𝛾(0) = 𝑝 and
𝛾( 2𝛿 ) = 𝑟. Note that for 𝑠 ≥ 0 we have 𝑑(𝛾(𝑠), 𝑞) ≥ 𝑑(𝑝, 𝑞) − 𝑑(𝑝, 𝛾(𝑠)) ≥ 𝑙 − 𝑠.
Defining
𝐼 = {𝑠 ∈ [0, 𝑙]∶ 𝑑(𝛾(𝑠), 𝑞) = 𝑙 − 𝑠},
the preceding inequality shows that 𝑠 ∈ 𝐼 if and only if 𝑠 ≥ 0 and 𝑑(𝛾(𝑠), 𝑞) ≤ 𝑙 − 𝑠.
It follows that 𝐼 is a (necessarily closed) interval: in fact, if 𝑠 ∈ 𝐼, and 0 ≤ 𝑡 < 𝑠, then
𝑑(𝛾(𝑡), 𝑞) ≤ 𝑑(𝛾(𝑡), 𝛾(𝑠)) + 𝑑(𝛾(𝑠), 𝑞) ≤ (𝑠 − 𝑡) + (𝑙 − 𝑠) = 𝑙 − 𝑡, and therefore 𝑡 ∈ 𝐼.
We have already seen that [0, 2𝛿 ] ⊆ 𝐼; let us now show that 𝐼 = [0, 𝑙]. To this end, it
suffices to show that if 𝑠0 ∈ ]0, 𝑙[ is in 𝐼 then also 𝑠0 + 𝜂 ∈ 𝐼 for some 𝜂 > 0 — because
then necessarily sup 𝐼 = 𝑙. Let us fix a normal neighborhood 𝐷 2𝜂 (𝛾(𝑠0 )) of 𝛾(𝑠0 )
so that 𝑑(𝛾(𝑠0 ), 𝑞) > 𝜂, and let ̃ 𝑟 be a point of S1 (𝛾(𝑠0 ), 𝜂) at the shortest possible
distance from 𝑞: the above argument shows that

𝑟 , 𝑞) = 𝑑(𝛾(𝑠0 ), 𝑞) − 𝜂 = 𝑙 − (𝑠0 + 𝜂) (because 𝑠0 ∈ 𝐼),

𝑑(̃ (5.2)

and therefore
142 5 The Global Geometry of Surfaces

𝑑(𝛾(𝑠0 − 𝜂), ̃ 𝑟 , 𝑞) = 2𝜂.

𝑟 ) ≥ 𝑑(𝛾(𝑠0 − 𝜂), 𝑞) − 𝑑(̃ (5.3)

The restriction 𝛾∣[𝑠 −𝜂,𝑠 ] , followed by the radial geodesic from 𝛾(𝑠0 ) to ̃
𝑟 , is a
0 0
piecewise regular curve from 𝛾(𝑠0 − 𝜂) to ̃ 𝑟 whose length is exactly 2𝜂. By (5.3) the
curve is minimizing and therefore a geodesic
q

p
g r~

g (s0 )

Figure 5.2
(Proposition 4.6.8), so that it coincides with 𝛾∣[𝑠 −𝜂,𝑠 +𝜂] . It follows that 𝛾(𝑠0 + 𝜂) = ̃
𝑟
0 0
and, by (5.2), 𝑠0 + 𝜂 ∈ 𝐼. We then have 𝐼 = [0, 𝑙]; in particular, 𝑑(𝛾(𝑙), 𝑞) = 0 and
therefore 𝛾(𝑙) = 𝑞. Thus, and since 𝑑(𝑝, 𝑞) = 𝑙 as well, 𝛾∣[0,𝑙] is a minimizing
geodesic from 𝑝 to 𝑞. ◻
If we read the above proof carefully, we obtain: if 𝑝 ∈ 𝑆 is such that all geodesics
starting from 𝑝 extend to all values of the parameter then, for every point 𝑞 of 𝑆,
there is some minimizing geodesic from 𝑝 to 𝑞. This observation lets us easily assert
that any complete surface 𝑆 is inextensible: if 𝑆 is contained in another surface 𝑆1
then 𝑆1 = 𝑆. In fact, 𝑆 is necessarily open in 𝑆1 , and therefore the geodesics of 𝑆 are
also geodesics of 𝑆1 . Let us fix 𝑝 ∈ 𝑆: the geodesics of 𝑆1 that start from 𝑝, because
they are geodesics of 𝑆, are defined for all values of the parameter. This means that,
given 𝑞 ∈ 𝑆1 , there exists some minimizing geodesic 𝛾 from 𝑝 to 𝑞. But 𝛾 ⊆ 𝑆 and
therefore 𝑞 ∈ 𝑆 — which shows that 𝑆1 ⊆ 𝑆.

√ 5.1.4 Besides the punctured plane, also the cone C given by the equation
Example
𝑧 = 𝑥 2 + 𝑦 2 (𝑧 > 0) is a non-complete surface, since the generating lines are
geodesics that are not defined for all parameter values. Nevertheless, between every
pair of points of C there exists a minimizing geodesic (Exercise 89 of Section 4.3) —
and C is inextensible. To prove the latter statement let us suppose, to the contrary,
that there exists a connected surface 𝑆 such that C ⊆ 𝑆 but C ≠ 𝑆. Let 𝑝 be a point on
the boundary of C in 𝑆: 𝑝 does not belong to C because C is open in 𝑆. Let (𝑝 𝑛 )𝑛≥1
be a sequence of points of C that converges to 𝑝.

Lemma Given 𝜀 > 0, there exists 𝑛0 such that 𝑝 𝑛 is below the plane 𝑧 = 𝜀 for all
𝑛 ≥ 𝑛0 .
5.1 Complete Surfaces 143

If the lemma is false, then some infinite subsequence of (𝑝 𝑛 )𝑛≥1 is contained

in C ∩ {𝑧 ≥ 𝜀} — which is a closed subset of R 3 and therefore contains all the
accumulation points of the mentioned subsequence; then 𝑝 ∈ C ∩ {𝑧 ≥ 𝜀} ⊆ C, which
is absurd. We have thus proved the lemma.
It immediately follows that lim 𝑝 𝑛 = (0, 0, 0). Thus the boundary of C on 𝑆
𝑛→+∞
consists only of the point (0, 0, 0), which implies 𝑆 = C ∪ {(0, 0, 0)}, otherwise
𝑆 ∖ {(0, 0, 0)} would not be connected. But C ∪ {(0, 0, 0)} is not a surface, which
proves the non-existence of 𝑆. ◻

Exercises
119. (a) Show that 𝑆 is a complete surface if and only if (𝑆, 𝑑) (where 𝑑 is the
intrinsic distance on 𝑆) is a complete metric space. (Complete means that any Cauchy
sequence converges. Try to prove the stronger statement that any bounded sequence
has a convergent subsequence.)
(b) Show that if there exists 𝑝 ∈ 𝑆 such that any geodesic passing through 𝑝 is
defined in R , then 𝑆 is complete.
120. If 𝑆 is a non-compact, complete surface, and 𝑝 a point of 𝑆, then there exists a
geodesic 𝛾(𝑠) of 𝑆 such that 𝛾(0) = 𝑝 and that minimizes the distance between 𝑝
and 𝛾(𝑠) for all 𝑠 ∈ R .
121. Let 𝑆 be a complete surface, 𝑝 a point of 𝑆, and v ∈ 𝑇 𝑝 𝑆 a unit vector. Write
𝛾(𝑠) = exp 𝑝 (𝑠v), and suppose that there exists 𝑠 > 0 such that 𝛾∣[0,𝑠] does not
minimize the distance between 𝑝 and 𝛾(𝑠). Consider the set 𝐼 = {𝑠 ≥ 0∶ 𝑑(𝑝, 𝛾(𝑠)) =
𝑠}. Show that:
(a) 𝐼 is a closed interval [0, 𝑠0 ];
(b) for 𝑠 > 𝑠0 there exists a geodesic 𝛾̃ of length 𝑠 connecting 𝑝 to 𝛾(𝑠);
(c) for 0 < 𝑠 < 𝑠0 , 𝛾∣[0,𝑠] is the only minimizing geodesic connecting 𝑝 with 𝛾(𝑠);
(d) for 𝑠 = 𝑠0 two cases could apply:
● there is another minimizing geodesic connecting 𝑝 with 𝛾(𝑠0 );
● 𝑠0 v is not a regular point of exp 𝑝 .
122. Consider in T 2 , parametrized by

Φ(𝑢, 𝑣) = ((2 + cos 𝑣) cos 𝑢, (2 + cos 𝑣) sin 𝑢, sin 𝑣),

the points 𝑝 = Φ(0, 0), 𝑞 = Φ(𝑢 0 , 𝑣 0 ) and 𝑟 = Φ(𝑢 0 , 𝜋), where 0 < 𝑢 0 < 𝜋, for
0 < 𝑣 0 < 𝜋. Let 𝛾(𝑠) = Φ(𝑢(𝑠), 𝑣(𝑠)), for 𝑠 ∈ [0, 𝑎], be a minimizing geodesic from
𝑝 to 𝑞 with 𝑢(0) = 𝑣(0) = 0. Show that:
(a) 0 ≤ 𝑢(𝑠) ≤ 𝑢 0 , 0 ≤ 𝑣(𝑠) < 𝜋 for all 𝑠 ∈ [0, 𝑎];
(b) there are two and only two minimizing geodesics from 𝑝 to 𝑟 (use the Clairaut
Equation, Exercise 100, to conclude that there is only one such geodesic in the region
0 ≤ 𝑢 ≤ 𝑢 0 , 0 ≤ 𝑣 ≤ 𝜋);
(c) if 𝛾̃(𝑠) is a geodesic such that 𝛾̃(0) = 𝑝 and 𝛾̃(𝑏) = 𝑟, then 𝛾̃(2𝑏) = Φ(2𝑢 0 , 0);
(d) if 𝑢𝜋0 is rational then the geodesic 𝛾̃ is periodic, otherwise it is dense in T 2 ;
144 5 The Global Geometry of Surfaces

(e) all geodesics (except the parallel 𝑣 = 𝜋) intersect the parallel 𝑣 = 0;

(f) there exist in T 2 geodesics that are neither periodic nor dense.

123. Consider in T 2 the parameterization Φ(𝑢, 𝑣) from Exercise 122. Let Δ be

the geodesic triangle of vertices 𝑝 = Φ(0, 0), 𝑞 = Φ( 𝜋2 , 0), 𝑟 = Φ( 𝜋2 , 𝑣 0 ), where
⌢ ⌢
0 < 𝑣 0 < 𝜋, and where 𝑝𝑞 = {Φ(𝑢, 0)∶ 0 ≤ 𝑢 ≤ 𝜋2 }, 𝑞𝑟 = {Φ( 𝜋2 , 𝑣)∶ 0 ≤ 𝑣 ≤ 𝑣 0 }, and
⌢
𝑝𝑟 is a minimizing geodesic from 𝑝 to 𝑟.

Use the Clairaut Equation to show that the sum of the interior angles of Δ is greater
than 𝜋. Can you draw the same conclusion from the Gauss-Bonnet theorem?

124. Let 𝑆 be a connected surface such that for every 𝑝 on 𝑆, there exists an isometry
𝜉 𝑝 ∶ 𝑆 → 𝑆 with 𝜉 𝑝 (𝑝) = 𝑝 and 𝐷(𝜉 𝑝 ) 𝑝 = − id. Show that:

(a) the sphere satisfies that condition;

(b) such a surface is complete and has constant curvature.

√
125. Let C be the cone 𝑧 = 𝑥 2 + 𝑦 2 , 𝑧 > 0. Show that there is no mapping 𝑓 ∶ C → 𝑆
such that 𝑆 is a complete surface and 𝑓 an isometry onto the image (i.e., there is no
isometric embedding of C into a complete surface).

5.2 Coverings

The notion of a covering is one of the most fruitful in topology — but, to save time,
we will restrict ourselves to coverings of surfaces, though they prove useful in more
general topological spaces. The theory developed here will allow us to show that a
good number of surfaces are (globally) images of local diffeomorphisms of standard
surfaces like the plane or the sphere.

A mapping 𝑓 ∶ 𝑆1 → 𝑆2 between two surfaces is called a covering (map) of 𝑆2 if

each 𝑞 ∈ 𝑆2 has an open neighborhood 𝑈 with the following property: 𝑓 −1 (𝑈) is a
collection of disjoint open subsets (𝑈𝑖 )𝑖 such that, for every 𝑖, the restriction 𝑓 ∣𝑈
𝑖
is a diffeomorphism of 𝑈𝑖 on 𝑈. Such an open subset 𝑈 is called an evenly covered
neighborhood of 𝑞 — or simply an evenly covered open subset; see Fig. 5.3.
5.2 Coverings 145

f
U

Figure 5.3

Example 5.2.1 The mapping R 2 → {𝑥 2 + 𝑦 2 = 1} given by (𝑢, 𝑣) → (cos 𝑢, sin 𝑢, 𝑣)

is a covering; as is the usual mapping R 2 → T 2 given by

Φ(𝑢, 𝑣) = ((2 + cos 𝑣) cos 𝑢, (2 + cos 𝑣) sin 𝑢, sin 𝑣).

It follows from the definition that any covering is a locally surjective diffeomor-
phism, but it is worthy to note that not all locally surjective diffeomorphisms are
coverings. An example is the restriction of Φ to the square ]−2𝜋, 2𝜋[ × ]−2𝜋, 2𝜋[:
in Figure 5.4, the preimage of the marked open subset 𝑈 joins nine disjoint open
subsets, but only one of them is surjectively sent onto 𝑈 under Φ. ◻

Figure 5.4
146 5 The Global Geometry of Surfaces

There is however an additional condition under which a local diffeomorphism is

guaranteed to be a covering:

Proposition 5.2.2 If 𝑆1 is compact and 𝑓 ∶ 𝑆1 → 𝑆2 a local diffeomorphism then 𝑓 is

a covering.

Proof Let us first show that 𝑓 is surjective. Since 𝑓 is a local diffeomorphism, its
image 𝑓 (𝑆1 ) is open in 𝑆2 . Given a sequence ( 𝑓 (𝑝 𝑛 ))𝑛≥1 on 𝑓 (𝑆1 ) that converges
to 𝑞 ∈ 𝑆2 , if 𝑝 ∈ 𝑆1 is an accumulation point of (𝑝 𝑛 )𝑛≥1 (which exists because 𝑆1 is
compact), then 𝑓 (𝑝) = 𝑞. Thus 𝑓 (𝑆1 ) is also closed in 𝑆2 and, since 𝑆2 is connected,
necessarily 𝑓 (𝑆1 ) = 𝑆2 .
We now observe that each point 𝑞 of 𝑆2 has a finite number of preimages, otherwise
they would accumulate in the neighborhood of some point of 𝑆1 , in contradiction
to 𝑓 being a local diffeomorphism. Let 𝑝 1 , . . . , 𝑝 𝑘 be the preimages of 𝑞, and let us
choose open neighborhoods 𝑉𝑖 of 𝑝 𝑖 such that each 𝑓 ∣𝑉 is a diffeomorphism onto the
𝑖
image and 𝑉𝑖 ∩ 𝑉 𝑗 is empty if 𝑖 ≠ 𝑗.
𝑘
Claim: There is 𝜀 > 0 such that 𝑓 −1 ({𝑟 ∈ 𝑆2 ∶ ∣𝑟 − 𝑞∣ < 𝜀}) ⊆ ⋃ 𝑉𝑖 .
𝑖=1

In fact, the negation of this statement implies the existence of a sequence (̃

𝑟 𝑛 )𝑛≥1
𝑘
𝑟 𝑛 ) = 𝑞, and any accumulation point of (̃
in 𝑆1 ∖ ( ⋃ 𝑉𝑖 ) such that lim 𝑓 (̃ 𝑟 𝑛 )𝑛≥1 is
𝑖=1 𝑛→+∞
a preimage of 𝑞 distinct from 𝑝 1 , . . . , 𝑝 𝑘 . Finally, for the 𝜀 given by the claim, the
𝑘
open subset {𝑟 ∈ 𝑆2 ∶ ∣𝑝 − 𝑞∣ < 𝜀} ∩ ( ⋂ 𝑓 (𝑉𝑖 )) is an evenly covered neighborhood of
𝑖=1
𝑞. ◻

We say that a differentiable mapping 𝑓 ∶ 𝑆1 → 𝑆2 lifts curves if, for every curve
𝛼∶ [𝑎, 𝑏] → 𝑆2 and 𝑝 ∈ 𝑆1 such that 𝑓 (𝑝) = 𝛼(𝑎), there exists a single curve
̃∶ [𝑎, 𝑏] → 𝑆1 such that 𝛼
𝛼 ̃(𝑡) = 𝛼(𝑡) for all 𝑡 ∈ [𝑎, 𝑏]. That is, for
̃(𝑎) = 𝑝 and 𝑓 ○ 𝛼
every preimage 𝑝 of the final point 𝛼(𝑎) of 𝛼, there exists a single curve in 𝑆1 that
starts from 𝑝 and whose image under 𝑓 is 𝛼. Note that any mapping that lifts curves
is necessarily surjective.

Proposition 5.2.3 Every covering lifts curves.

Proof Let 𝑓 ∶ 𝑆1 → 𝑆2 be a covering, 𝛼∶ [𝑎, 𝑏] → 𝑆2 a curve, and 𝑝 ∈ 𝑆1 a preimage

of 𝛼(𝑎). By compactness of [𝑎, 𝑏] there exists a partition

𝑎 = 𝑡0 < 𝑡1 < ⋯ < 𝑡 𝑘 = 𝑏

such that each 𝛼([𝑡𝑖−1 , 𝑡 𝑖 ]) is contained in an evenly covered open subset 𝑊𝑖 . We

̃ step by step, starting with the interval [𝑡0 , 𝑡 1 ]: denoting by 𝑊
construct 𝛼 ̃1 the
−1 −1
component of 𝑓 (𝑊1 ) that contains 𝑝, we define 𝛼 ̃∣[𝑡 ,𝑡 ] = 𝑓 ∣𝑊 ̃ is
̃1 ○ 𝛼∣[𝑡0 ,𝑡1 ] : 𝛼
0 1
continuous and obviously satisfies 𝛼 ̃ = 𝛼.
̃(𝑡 0 ) = 𝑝 and 𝑓 ○ 𝛼
5.2 Coverings 147

̃∣[𝑡 ,𝑡 ] and defined 𝑊

Assuming that we constructed 𝛼 ̃𝑖 for 𝑖 = 0, . . . , 𝑗 − 1,
0 𝑗−1
̃ 𝑗 the component of 𝑓 −1 (𝑊 𝑗 ) that contains 𝛼
we denote by 𝑊 ̃(𝑡 𝑗−1 ) and we put
−1
̃∣[𝑡 ,𝑡 ] = 𝑓 ∣𝑊
𝛼 ̃ 𝑗 ○ 𝛼∣[𝑡 𝑗−1 ,𝑡 𝑗 ] . Repeating this until 𝑗 = 𝑘, we complete the construction
𝑗−1 𝑗
̃.
of 𝛼
Regarding uniqueness of 𝛼 ̃, it follows from the following general observation:
if 𝑓 ∶ 𝑆1 → 𝑆2 is a local diffeomorphism and 𝛼 ̃∶ [𝑎, 𝑏] → 𝑆1 are curves such that
̂, 𝛼
̂(𝑎) = 𝛼
𝛼 ̃(𝑎) and 𝑓 ○ 𝛼 ̂= 𝑓 ○𝛼 ̃, then 𝛼
̂=𝛼 ̃.
To see this, we observe that 𝐼 = {𝑡 ∈ [𝑎, 𝑏]∶ 𝛼 ̂(𝑡) = 𝛼 ̃(𝑡)} is open and closed on
[𝑎, 𝑏], by which 𝐼 = [𝑎, 𝑏] (𝐼 is closed because 𝛼 ̂ and 𝛼 ̃ are continuous, and is open
because 𝑓 is a local diffeomorphism). ◻

A remarkable property of coverings 𝑓 ∶ 𝑆1 → 𝑆2 is that all points of 𝑆2 have the

same number of preimages, which is a simple consequence of Proposition 5.2.3.
Indeed, given 𝑝 and 𝑞 in 𝑆2 , consider a curve 𝛼 in 𝑆2 from 𝑝 to 𝑞. For each preimage
𝑝 𝑖 of 𝑝, let 𝛼𝑖 be the lifting of 𝛼 that starts from 𝑝 𝑖 . The endpoint 𝑞 𝑖 of 𝛼𝑖 is a
preimage of 𝑞. By Proposition 5.2.3, if 𝑖 ≠ 𝑗 then 𝑞 𝑖 ≠ 𝑞 𝑗 — since, obviously, there is
a unique lifting for a given endpoint. Thus # 𝑓 −1 ({𝑝}) ≤ # 𝑓 −1 ({𝑞}), and swapping
the roles of 𝑝 and 𝑞 we obtain the opposite inequality.
The number of preimages of each point of 𝑆2 is usually called the number of sheets
of the covering.
By Proposition 5.2.3, we see that any covering 𝑓 ∶ 𝑆1 → 𝑆2 enjoys the following
properties:
(i) 𝑓 is a local diffeomorphism,
(ii) 𝑓 lifts curves,
— and the converse is also true:

Proposition 5.2.4 If 𝑓 ∶ 𝑆1 → 𝑆2 has properties (i) and (ii) then 𝑓 is a covering.

Note: In topological spaces other than surfaces — where, instead of diffeomorphisms,

we would speak of local homeomorphisms — (i) and (ii) no longer guarantee that a
mapping is a covering.
We say that two curves 𝛼0 , 𝛼1 ∶ [𝑎, 𝑏] → 𝑆2 with 𝛼0 (𝑎) = 𝛼1 (𝑎) = 𝑞 0 and 𝛼0 (𝑏) =
𝛼1 (𝑏) = 𝑞 1 are homotopic with fixed endpoints if there exists a continuous function

𝐻∶ [0, 1] × [𝑎, 𝑏] → 𝑆2

such that 𝐻(0, 𝑡) = 𝛼0 (𝑡), 𝐻(1, 𝑡) = 𝛼1 (𝑡) for all 𝑡 ∈ [𝑎, 𝑏], and 𝐻(𝑠, 𝑎) = 𝑞 0 ,
𝐻(𝑠, 𝑏) = 𝑞 1 for all 𝑠 ∈ [0, 1]. Letting 𝛼𝑠 = 𝐻(𝑠, ⋅), the family of curves (𝛼𝑠 )0≤𝑠≤1 is
called a continuous deformation of 𝛼0 with fixed endpoints. Note that if 𝛼0 is closed
then all 𝛼𝑠 curves are also closed.

Lemma 5.2.5 With the above notation, let 𝛼 ̃𝑠 , for 0 ≤ 𝑠 ≤ 1, be a lifting of 𝛼𝑠 in 𝑆1 .

̃𝑠 have the same initial point 𝑝 0 , then they also all have the same
If all the curves 𝛼
endpoint.
148 5 The Global Geometry of Surfaces

Proof It suffices to show that, given 𝑠0 ∈ [0, 1], there exists 𝛿 > 0 such that if
∣𝑠 − 𝑠0 ∣ < 𝛿, for all 𝑠 ∈ [0, 1] then 𝛼̃𝑠 (𝑏) = 𝛼 ̃𝑠0 (𝑏). Let 𝑎 = 𝑡 0 < 𝑡1 < ⋯ < 𝑡𝑙 = 𝑏 be a
partition of [𝑎, 𝑏] such that 𝛼 ̃𝑖 , where for each 𝑖:
̃𝑠0 ([𝑡 𝑖−1 , 𝑡 𝑖 ]) ⊆ 𝐷

• 𝐷̃𝑖 is an open subset diffeomorphic to a disk of the plane;

̃𝑖 ∪ 𝐷
• if 𝑖 < 𝑘 then there exists an open subset including the closure of 𝐷 ̃𝑖+1 such
that the restriction of 𝑓 to it is a diffeomorphism onto the image.

Every 𝐷 𝑖 = 𝑓 ( 𝐷 ̃𝑖 ) is an open subset of 𝑆2 diffeomorphic to a disk. By (uniform)

continuity of 𝐻(𝑠, 𝑡) = 𝛼𝑠 (𝑡), there exists 𝛿 > 0 such that if ∣𝑠 − 𝑠0 ∣ < 𝛿, for all
𝑠 ∈ [0, 1], then 𝛼𝑠 (𝑡) ∈ 𝐷 𝑖 for all 𝑡 ∈ [𝑡 𝑖−1 , 𝑡 𝑖 ] and 𝑖 = 1, . . . , 𝑘.
Fixing 𝑠 ∈ [0, 1] ∩ ]𝑠0 − 𝛿, 𝑠0 + 𝛿[, let us show that 𝛼 ̃𝑠 ([𝑡𝑖−1 , 𝑡 𝑖 ]) ⊆ 𝐷 ̃𝑖 for 𝑖 =
1, . . . , 𝑘. For 𝑖 = 1, we have 𝛼 ̃𝑠 (𝑡0 ) = 𝑝 0 ∈ 𝐷̃𝑖 ; if not 𝛼
̃𝑠 ([𝑡0 , 𝑡 𝑖 ]) ⊆ 𝐷 ̃1 , there is
̃𝑠 (𝑡) is on the boundary of 𝐷
𝑡 ∈ ]𝑡 0 , 𝑡 1 ] such that 𝛼 ̃1 . For this 𝑡, the point 𝛼𝑠 (𝑡) is on
the boundary of 𝐷 1 , which contradicts the choice of 𝑠. We then have 𝛼 ̃1 .
̃𝑠 ([𝑡 0 , 𝑡 1 ]) ⊆ 𝐷
Assuming now that 𝛼 ̃𝑠 ([𝑡𝑖−1 , 𝑡 𝑖 ]) ⊆ 𝐷̃𝑖 (𝑖 < 𝑘), we observe that 𝛼𝑠 (𝑡 𝑖 ) = 𝑓 (̃ 𝛼𝑠 (𝑡 𝑖 )) is
in 𝐷 𝑖 ∩ 𝐷 𝑖+1 . Since, by the second condition above, the restriction of 𝑓 to 𝐷 ̃𝑖 ∪ 𝐷 ̃𝑖+1
−1
is a bijection onto 𝐷 𝑖 ∪ 𝐷 𝑖+1 , we have 𝑓 ∣ ̃ ̃ (𝐷 𝑖 ∩ 𝐷 𝑖+1 ) = 𝐷 ̃𝑖 ∩ 𝐷 ̃𝑖+1 and
𝐷𝑖 ∪𝐷𝑖+1
therefore 𝛼 ̃𝑖+1 . From this, as above, we conclude that 𝛼
̃𝑠 (𝑡 𝑖 ) ∈ 𝐷 ̃𝑖+1
̃𝑠 ([𝑡𝑖 , 𝑡 𝑖+1 ]) ⊆ 𝐷
— and the proof by induction is finished. Finally, from the fact that 𝛼 ̃
̃𝑠 (𝑏) ∈ 𝐷 𝑘 and
𝑓 (̃
𝛼𝑠 (𝑏)) = 𝑞 1 = 𝑓 (̃ ̃𝑠 (𝑏) = 𝛼
𝛼𝑠0 (𝑏)), it follows, just as desired, that 𝛼 ̃𝑠0 (𝑏). ◻

We now prove Proposition 5.2.4. Let us fix 𝑞 0 ∈ 𝑆2 and an open neighborhood 𝑉

of it that is diffeomorphic to a disk. We say that two points of 𝑓 −1 (𝑉) are in the same
connected component if there exists a curve in 𝑓 −1 (𝑉) that connects them, and we
write
𝑓 −1 (𝑉) = ⋃ 𝑉 𝑗 ,
𝑗

where the 𝑉 𝑗 are the connected components. Each 𝑉 𝑗 is open: in fact, if 𝑝 ∈ 𝑉 𝑗 and
𝑊 ⊆ 𝑓 −1 (𝑉) is a neighborhood of 𝑝 diffeomorphic to a disk, then 𝑊 ⊆ 𝑉 𝑗 .
Let us now verify that 𝑓 ∣𝑉 ∶ 𝑉 𝑗 → 𝑉 is bijective. Take 𝑝 ∈ 𝑉 𝑗 and let 𝑞 = 𝑓 (𝑝).
𝑗
Given 𝑟 ∈ 𝑉, let 𝛼∶ [𝑎, 𝑏] → 𝑉 be a curve with 𝛼(𝑎) = 𝑞 and 𝛼(𝑏) = 𝑟, and let 𝛼 ̃ be
̃(𝑏) is in the same component 𝑉 𝑗
the lifting of 𝛼 that starts from 𝑝. By definition, 𝛼
as 𝑝 and 𝑓 (̃𝛼(𝑏)) = 𝑟. Therefore 𝑓 ∣𝑉 is surjective. Regarding injectivity, assume
𝑗
that 𝑝̃, 𝑝̂ ∈ 𝑉 𝑗 have the same image 𝑞, and 𝛼̃0 ∶ [𝑎, 𝑏] → 𝑉 𝑗 is a curve from 𝑝̃ to 𝑝̂, and
𝛼0 = 𝑓 ○ 𝛼 ̃0 . This curve 𝛼0 is then closed. Since 𝑉 is diffeomorphic to a disk, there
exists a continuous deformation (𝛼𝑠 )0≤𝑠≤1 of 𝛼0 with fixed endpoints such that the
trace of each 𝛼𝑠 is in 𝑉 and 𝛼1 is the constant curve equal to 𝑞. Denoting by 𝛼 ̃𝑠 the
lifting of 𝛼𝑠 that starts from 𝑝̃, Lemma 5.2.5 guarantees that the curves 𝛼 ̃𝑠 all have
the same endpoints 𝑝̃ and 𝑝̂. But 𝛼 ̃1 is constant, and therefore 𝑝̃ = 𝛼̃1 (𝑎) = 𝛼
̃1 (𝑏) = 𝑝̂.
Thus 𝑓 ∣𝑉 is injective.
𝑗
In conclusion: each component 𝑉 𝑗 is diffeomorphically sent into 𝑉, and therefore
𝑉 is an evenly covered neighborhood of 𝑞 0 . This shows that 𝑓 is a covering and
concludes the proof of Proposition 5.2.4.
5.3 Complete Surfaces of Non-Positive Curvature 149

In this proof we used that, for every open subset 𝑈 diffeomorphic to a disk, every
closed curve in 𝑈 is homotopic (with fixed endpoints) to a constant curve by a
homotopy that only takes values in 𝑈. A set 𝑈 with such a property is called simply
connected. Examples of simply connected surfaces are the plane and the sphere (recall
that S2 minus one point is diffeomorphic to the plane via stereographic projection).
If 𝑆2 is simply connected and 𝑓 ∶ 𝑆1 → 𝑆2 a covering then 𝑓 −1 (𝑆2 ) has only one
connected component, which is all of 𝑆1 — and, by the proof of Proposition 5.2.4, 𝑓
is injective, hence a diffeomorphism. This proves our next result.

Proposition 5.2.6 Every covering of a simply connected surface is a diffeomorphism.

We now have gathered all necessary tools about coverings and it is good to start at
once to make interesting use of it.

Theorem 5.2.7 (Hadamard)

Let 𝑆 ⊆ R 3 be a compact surface with positive curvature at all points. Then 𝑆 is
diffeomorphic to the sphere.

In fact, such a surface is orientable (e.g. Example 61-b in Section 3.2), and the
hypothesis implies that the normal mapping 𝑁∶ 𝑆 → S2 is a local diffeomorphism,
hence (by Proposition 5.2.2 and the compactness of 𝑆) a covering, hence (by
Proposition 5.2.6 and S2 being simply connected) a diffeomorphism.
In Section 5.4 we will discuss in more depth the compact surfaces of positive
curvature in R 3 (called ovals) and show that ovals are convex (in the sense that they
bound convex regions of R 3 ).

Exercises
̃ 𝑡) = 𝛼
126. Show that in Lemma 5.2.5 the function defined by 𝐻(𝑠, ̃𝑠 (𝑡) is continuous.
127. Let 𝑆1 and 𝑆2 be two connected surfaces and 𝑓 ∶ 𝑆1 → 𝑆2 a covering. Further, let
ℑ( 𝑓 ) be the set of diffeomorphisms 𝑔∶ 𝑆1 → 𝑆1 such that 𝑓 ○ 𝑔 = 𝑔. Show that:
(a) ℑ( 𝑓 ) is a group with respect to composition of functions;
(b) if 𝑆1 is simply connected, then for every pair of points 𝑝 0 , 𝑝 1 in 𝑆1 such that
𝑓 (𝑝 0 ) = 𝑓 (𝑝 1 ), there exists a unique 𝑔 ∈ ℑ( 𝑓 ) such that 𝑔(𝑝 0 ) = 𝑝 1 ;
(c) if 𝑆1 is not simply connected, then the property in (b) may fail.

5.3 Complete Surfaces of Non-Positive Curvature

In this section, 𝑆 is a complete surface of non-positive curvature. For every 𝑝 ∈ 𝑆, the

exponential mapping exp 𝑝 is then defined on the entire tangent space 𝑇 𝑝 𝑆. We shall
show the assumption that the curvature of 𝑆 is non-positive implies that exp 𝑝 is a
covering.
150 5 The Global Geometry of Surfaces

Fixing 𝑝 ∈ 𝑆 and an orthonormal basis (v1 , v2 ) of 𝑇 𝑝 𝑆, we note that the coefficients

of the geodesic polar coordinates Ψ(𝜌, 𝜑) centered at 𝑝 have a well-defined meaning
for all (𝜌, 𝜑) with 𝜌 > 0, even if globally Ψ(𝜌, 𝜑) is not a parameterization. For
example,
𝐺(𝜌, 𝜑) = ∣𝐷(exp 𝑝 )𝜌v 𝜑 (𝜌w 𝜑 )∣2 , (5.4)
where v𝝋 = cos 𝜑v1 + sin 𝜑v2 , w𝝋 = − sin 𝜑v1 + cos 𝜑v2 . Furthermore, the equalities
𝐸 ≡ 1, 𝐹 ≡ 0 are valid for all (𝜌, 𝜑) without having to change anything in the proof
of Lemma 4.6.3.
Ψ is a true parameterization in the neighborhood of any point (𝜌, 𝜑) such that
𝐺(𝜌, 𝜑) ≠ 0, and hence formula (4.38) is valid. That is,
√ √
( 𝐺)𝜌𝜌 + 𝐾 𝐺 = 0 (5.5)
√
whenever 𝐺 ≠ 0. We know that with 𝜑 fixed the function 𝜌 ↦ 𝐺(𝜌, 𝜑) extends
continuously to 𝜌 = 0,√and such an extension
√ is twice differentiable at 0 (see
Lemma 4.6.9) — with 𝐺(0, 𝜑) = 0 and ( 𝐺)𝜌 (0, 𝜑) = 1. Assuming that 𝐾 ≤ 0
√
on 𝑆, we obtain from (5.5) that ( 𝐺)𝜌𝜌 ≥ 0; thus
√ √
( 𝐺)𝜌 (𝜌, 𝜑) ≥ ( 𝐺)𝜌 (0, 𝜑) = 1
√ √ (5.6)
⇒ 𝐺(𝜌, 𝜑) ≥ 𝐺(0, 𝜑) + 𝜌 = 𝜌.

Both these inequalities hold, as does (5.5), up to the first 𝜌0 > 0 where 𝐺(𝜌0 , 𝜑) = 0.
But (5.6) ensures that such a 𝜌0 does not exist, and therefore the inequalities hold
for all 𝜌 ≥ 0. Since 𝐺(𝜌, 𝜑) is nonzero for all 𝜌 > 0, it follows that exp 𝑝 is a local
diffeomorphism. This proves one part of the following theorem:

Theorem 5.3.1 Let 𝑆 be a complete surface of non-positive curvature at all points.

Then, for every 𝑝 ∈ 𝑆, the local diffeomorphism exp 𝑝 ∶ 𝑇 𝑝 𝑆 → 𝑆 is a covering.

In view of Proposition 5.2.4, to complete the proof of Theorem 5.3.1, it is only

left to show that exp 𝑝 lifts curves. The next result says that exp 𝑝 increases the length
of curves:

Lemma 5.3.2 If 𝛽∶ [𝑎, 𝑏] → 𝑇 𝑝 𝑆 is a differentiable curve then 𝑙(𝛽) ≤ 𝑙(exp 𝑝 ○𝛽).

Proof It suffices to show that for every 𝑟 ∈ 𝑇 𝑝 𝑆, and vector v, we have ∣𝐷(exp 𝑝 )𝑟 (v)∣ ≥
∣v∣, since then
𝑏 𝑏
𝑙(𝛽) = ∫ ∣𝛽′ (𝑡)∣ 𝑑𝑡 ≤ ∫ ∣𝐷(exp 𝑝 )𝛽(𝑡) (𝛽′ (𝑡))∣ 𝑑𝑡 = 𝑙(exp 𝑝 ○𝛽).
𝑎 𝑎

Since 𝐷(exp 𝑝 )𝑂 is the identity, we can assume that 𝑟 = 𝜌v𝝋 with 𝜌 > 0. The vectors
v𝝋 and w𝝋 form an orthonormal basis of 𝑇 𝑝 𝑆, and the vectors 𝐷(exp 𝑝 )𝑟 (v𝝋 ) and
𝐷(exp 𝑝 )𝑟 (w𝝋 ) are also orthogonal (𝐹 ≡ 0). Moreover,
5.3 Complete Surfaces of Non-Positive Curvature 151

∣𝐷(exp 𝑝 )𝑟 (v𝝋 )∣ = 1 = ∣v𝝋 ∣,

√
𝐺
∣𝐷(exp 𝑝 )𝑟 (w𝝋 )∣ = ≥ 1 = ∣w𝝋 ∣ [by (5.4) and (5.6)]
𝜌
– and therefore

∣𝐷(exp 𝑝 )𝑟 (𝜆v𝝋 + 𝜇w𝝋 )∣2 = 𝜆2 ∣𝐷(exp 𝑝 )𝑟 (v𝝋 )∣2 + 𝜇2 ∣𝐷(exp 𝑝 )𝑟 (w𝝋 )∣2
≥ 𝜆2 + 𝜇2 = ∣𝜆v𝝋 + 𝜇w𝝋 ∣2 .◻

Returning to the proof of Theorem 5.3.1, consider a differentiable curve 𝛼∶ [𝑎, 𝑏] →

𝑆, and let 𝑟 be a point of 𝑇 𝑝 𝑆 such that exp 𝑝 (𝑟) = 𝛼(𝑎). Taking a neighborhood of 𝑟
that is diffeomorphically sent onto the image, we see that there exists 𝑐 ∈ ]𝑎, 𝑏] such
̃ of 𝛼 starting at 𝑟 is defined on [𝑎, 𝑐] (in the proof of Proposition 5.2.3
that the lifting 𝛼
we saw that for a local diffeomorphism 𝑆1 → 𝑆2 the lifting in 𝑆1 of a curve in 𝑆2 for a
given fixed initial point is unique, whenever it exists). Let ̃ 𝑡 be the supremum of the
set
̃ is defined on [𝑎, 𝑡]} ∶
𝐼 = {𝑡 ∈ [𝑎, 𝑏] ∶ 𝛼
̃ is therefore defined on [𝑎, ̃
𝛼 𝑡 [. If 𝑎 ≤ 𝑠 < 𝑡 < ̃
𝑡 then, using Lemma 5.3.2,

∣̃
𝛼(𝑡) − 𝛼 𝛼∣[𝑠,𝑡] ) ≤ 𝑙(𝛼∣[𝑠,𝑡] )
̃(𝑠)∣ ≤ 𝑙(̃

— and therefore
lim ∣̃ ̃(𝑠)∣ = 0,
𝛼(𝑡) − 𝛼
𝑠,𝑡→̃
𝑡−

̃(𝑡). We thus conclude that ̃

which implies the existence of lim 𝛼 𝑡 ∈ 𝐼, and it follows
𝑡→̃
𝑡−
that ̃
𝑡 = 𝑏, otherwise the fact that exp 𝑝 is a local diffeomorphism would allow 𝛼 ̃
to be extended beyond ̃ 𝑡 . Hence exp 𝑝 lifts curves, which concludes the proof of
Theorem 5.3.1.
Combining Theorem 5.3.1 with Proposition 5.2.6 we obtain the following corollary.
Theorem 5.3.3 (Hadamard)
If 𝑆 is a complete, simply connected surface of curvature 𝐾 ≤ 0, then, for every
𝑝 ∈ 𝑆, the local diffeomorphism exp 𝑝 ∶ 𝑇 𝑝 𝑆 → 𝑆 is a diffeomorphism.

We saw in Chapter 4 how to construct local isometries between two surfaces of

equal constant curvature 𝐾 (Theorem 4.6.10): given 𝑝 ∈ 𝑆1 , 𝑞 ∈ 𝑆2 , and a linear
isometry 𝐿∶ 𝑇 𝑝 𝑆1 → 𝑇𝑞 𝑆2 , the mapping exp𝑞 ○ 𝐿 ○ exp−1
𝑝 is defined in a neighborhood
of 𝑝 and is an isometry onto the image. Let us now assume that 𝐾 ≤ 0, that both 𝑆1
and 𝑆2 are complete, and that 𝑆1 is simply connected. Then 𝑓 = exp𝑞 ○ 𝐿 ○ exp−1 𝑝 is
defined on the entire surface 𝑆1 (by Theorem 5.3.3), is a local isometry, and is still
a covering (by Theorem 5.3.1). In case 𝑆2 is also simply connected, 𝑓 is a (global)
isometry between 𝑆1 and 𝑆2 . To summarize, we proved the following theorem:
Theorem 5.3.4 Let 𝑆1 be a complete, simply connected surface of constant curvature
𝐾 ≤ 0, and let 𝑆2 be another complete surface with the same curvature 𝐾. Then:
152 5 The Global Geometry of Surfaces

i. there exists a covering 𝑓 ∶ 𝑆1 → 𝑆2 which is a local isometry;

ii. if 𝑆2 is simply connected, the covering 𝑓 is an isometry.
For every 𝐾 ≤ 0, there is therefore, up to isometry, only one complete surface,
simply connected, of constant curvature 𝐾, and all other complete surfaces of
equal curvature are images of this one under some isometric covering. The cylinder
𝑥 2 + 𝑦 2 = 1, for example, is (as we already knew) the image of the plane under an
isometric covering: the function that wraps the plane around the cylinder.
We now face a problem: we do not yet know any complete surface on R 3 , simply
connected or not, with negative constant curvature. Indeed, a famous theorem by
Hilbert (of which a proof is found in [6]) states that such a surface does not exist.
Do we have to conclude that Theorem 5.3.4 is void when 𝐾 < 0? The solution is to
consider, as we will do in Section 5.6, abstract surfaces whose metric structure is not
induced by any ambient space.
Theorem 5.3.4 also holds for 𝐾 > 0, with a slightly different proof; but in this case,
for surfaces in R 3 , Theorem 5.3.4 is a triviality, since 𝑆1 and 𝑆2 have to be compact
(see Exercise 129), and therefore are spheres of equal radius (Section 5.4). Again we
have to consider abstract surfaces for the result to be interesting.

Exercises
128. Let 𝑆 be a complete surface, of non-positive curvature, and 𝑝 a point of 𝑆.
(a) Show that the equations of geodesics in geodesic polar coordinates are

1 𝐺𝜌 𝐺𝜑 2
𝜌− 𝐺 𝜌 𝜑2 = 0, 𝜑+ 𝜌𝜑 + 𝜑 = 0.
2 𝐺 2𝐺
(b) Let 𝛾(𝑠) be a geodesic of 𝑆 that does not pass through 𝑝, and denote by
𝛾̃∶ R → 𝑇 𝑝 𝑆 a lifting of 𝛾. Show that the function 𝜌(𝑠) = ∣̃
𝛾 (𝑠)∣ is convex (i.e., 𝜌 ≥ 0)
and has at most one local minimum.
(c) Assume now that 𝑆 is simply connected. Show that the trace of any geodesic
of 𝑆 is a closed set on 𝑆 and that, given a geodesic 𝛾(𝑠) that does not pass through 𝑝,
there exists a single point of 𝛾 at the minimal distance from 𝑝.
(d) Compare the result of (c) with the case of the sphere. Give an example of a
complete surface of non-positive curvature where not all geodesics are closed sets.
129. Let 𝑆 be a complete surface of constant curvature 𝐾 > 0. Using polar geodesic
𝜋
coordinates, show that any minimizing geodesic of 𝑆 has length ≤ √ . Conclude
𝐾
that 𝑆 is compact.

5.4 Ovals (First Part): The Rigidity of the Sphere

The oval surfaces (or simply ovals) are the compact surfaces in R 3 with positive
curvature at all points. Such surfaces are, as we saw in Section 5.2 (Theorem 5.2.7),
5.4 Ovals (First Part): The Rigidity of the Sphere 153

diffeomorphic to the sphere, the field of normal vectors 𝑁∶ 𝑆 → S2 being a diffeomor-

phism.
A surface 𝑆 is strictly convex if, for every 𝑝 ∈ 𝑆, the intersection of 𝑆 with the
tangent plane {𝑝} + 𝑇 𝑝 𝑆 with 𝑆 reduces to the point 𝑝; equivalently, if 𝑆 ∖ {𝑝} is
entirely in one of the two half-spaces into which {𝑝} + 𝑇 𝑝 𝑆 divides R 3 .
Proposition 5.4.1 Every oval 𝑆 is strictly convex.
Proof Fixing 𝑝 ∈ 𝑆, we must check that the mapping 𝑆 → R given by ℎ 𝑝 (𝑞) =
⟨𝑁(𝑝), 𝑞− 𝑝⟩ has constant sign, vanishing only at 𝑝. Otherwise ℎ 𝑝 reaches a minimum
and a maximum at points 𝑞 0 , 𝑞 1 ∈ 𝑆 ∖ {𝑝}. Each of the tangent spaces 𝑇𝑞𝑖 𝑆 (𝑖 = 0, 1)
is then orthogonal to 𝑁(𝑝), which implies that two of the vectors 𝑁(𝑝), 𝑁(𝑞 0 ) and
𝑁(𝑞 1 ) are equal — in contradiction to 𝑁 being a diffeomorphism.
Now suppose that 𝑝 0 ∈ 𝑆 satisfies ⟨𝑁(𝑝 0 ), 𝑞 − 𝑝 0 ⟩ > 0 for all 𝑞 ∈ 𝑆 ∖ {𝑝 0 }. Let
us verify that one also has ⟨𝑁(𝑝), 𝑞 − 𝑝⟩ > 0 for all distinct points 𝑝, 𝑞 ∈ 𝑆. Since
ℎ 𝑝 has constant sign, we can assume 𝑞 ≠ 𝑝 0 and consider a curve 𝛼∶ [𝑎, 𝑏] → 𝑆
such that 𝛼(𝑎) = 𝑝 0 , 𝛼(𝑏) = 𝑝, and 𝛼(𝑡) ≠ 𝑞 for all 𝑡 ∈ [𝑎, 𝑏]. The function
𝑡 ↦ ⟨𝑁(𝛼(𝑡)), 𝑞 − 𝛼(𝑡)⟩ never vanishes, and therefore takes the same sign for 𝑡 = 𝑎
and 𝑡 = 𝑏, which proves that ⟨𝑁(𝑝), 𝑞 − 𝑝⟩ > 0. ◻
From now on we fix an orientation 𝑁 of 𝑆 such that ⟨𝑁(𝑝), 𝑞 − 𝑝⟩ > 0 for all
𝑝 ≠ 𝑞 ∈ 𝑆. With this orientation the principal curvatures (and the mean curvature) at
each point are positive.
Our goal now is to show that the region Ω of R 3 bounded by 𝑆 is convex in the
usual sense: the line segment joining each pair of points in Ω is also contained in Ω.
Let us consider the sets

C = {𝑞 ∈ R 3 ∶ ⟨𝑁(𝑝), 𝑞 − 𝑝⟩ > 0 ∀ 𝑝 ∈ 𝑆},

D = {𝑞 ∈ R 3 ∶ ∃ 𝑝 ∈ 𝑆 such that ⟨𝑁(𝑝), 𝑞 − 𝑝⟩ < 0}.

Proposition 5.4.2 The sets C and D are connected open, C is convex and bounded, 𝑆
is the boundary of both sets C and D, and R 3 ∖ 𝑆 = C ∪ D.
Proof Each of the sets D 𝑝 = {𝑞 ∈ R 3 ∶ ⟨𝑁(𝑝), 𝑞 − 𝑝⟩ < 0}, for 𝑝 ∈ 𝑆, is open, and D
is the union of them all. Therefore D is open, and it is easily seen to be connected.
The set C is the intersection of the convex sets C 𝑝 = {𝑞 ∈ R 3 ∶ ⟨𝑁(𝑝), 𝑞 − 𝑝⟩ > 0}
(𝑝 ∈ 𝑆), and is therefore convex; and any convex set is connected.
Given 𝑞 ∈ C, let us consider 𝛿 = min ⟨𝑁(𝑝), 𝑞 − 𝑝⟩ > 0. If ∣𝑟 − 𝑞∣ < 𝛿 then we have,
𝑝∈𝑆
for every 𝑝 ∈ 𝑆,

⟨𝑁(𝑝), 𝑟 − 𝑝⟩ = ⟨𝑁(𝑝), 𝑞 − 𝑝⟩ − ⟨𝑁(𝑝), 𝑞 − 𝑟⟩ ≥ 𝛿 − ∣𝑞 − 𝑟∣ > 0,

so that 𝑟 ∈ C — which proves that C is open.

We shall show that 𝑆 ⊆ 𝜕C ∩ 𝜕D. Let us fix 𝑝 0 ∈ 𝑆 and a unit vector v such
that ⟨𝑁(𝑝 0 ), v⟩ > 0, and let 𝑝 𝑡 = 𝑝 0 + 𝑡v. When 𝑡 < 0, we have ⟨𝑁(𝑝 0 ), 𝑝 𝑡 − 𝑝 0 ⟩ =
𝑡⟨𝑁(𝑝 0 ), v⟩ < 0 and therefore 𝑝 𝑡 ∈ D. This shows that 𝑝 0 ∈ 𝜕D, and we thus conclude
that 𝑆 ⊆ 𝜕D.
154 5 The Global Geometry of Surfaces

Now let 𝑈 ⊆ 𝑆 be an open neighborhood of 𝑝 0 such that ⟨𝑁(𝑝), v⟩ > 0 whenever

𝑝 ∈ 𝑈, and let us put 𝜀 = min ⟨𝑁(𝑝), 𝑝 0 − 𝑝⟩. It follows immediately that if 𝑝 ∈ 𝑈
𝑝∈𝑆∖𝑈
and 𝑡 > 0 then ⟨𝑁(𝑝), 𝑝 𝑡 − 𝑝⟩ > 0. On the other hand, if 𝑝 ∈ 𝑆 ∖ 𝑈 and 0 < 𝑡 < 𝜀 then

⟨𝑁(𝑝), 𝑝 𝑡 − 𝑝⟩ = ⟨𝑁(𝑝), 𝑝 0 − 𝑝⟩ + 𝑡⟨𝑁(𝑝), v⟩ ≥ 𝜀 − 𝑡∣𝑁(𝑝)∣ ∣v∣ > 0

— which shows that 𝑝 𝑡 ∈ C for 𝑡 ∈ ]0, 𝜀[. Thus 𝑝 0 ∈ 𝜕C, and therefore 𝑆 ⊆ 𝜕C.
We now show that R 3 ∖ (C ∪ D) ⊆ 𝑆. Indeed, if 𝑞 ∈ R 3 ∖ (C ∪ D) then there exists
some 𝑝 0 ∈ 𝑆 such that ⟨𝑁(𝑝 0 ), 𝑞 − 𝑝 0 ⟩ = 0. If 𝑞 ≠ 𝑝 0 then we can consider a curve
𝛼∶ ]− 𝜂, 𝜂[ → 𝑆 such that 𝛼(0) = 𝑝 0 and 𝛼′ (0) = 𝐷𝑁 −1 𝑝0 (𝑞 − 𝑝 0 ), where

⟨𝑁 ○ 𝛼(𝑡), 𝑞 − 𝛼(𝑡)⟩∣𝑡=0 = ⟨𝐷𝑁 𝑝0 (𝛼′ (0)), 𝑞 − 𝑝 0 ⟩ = ∣𝑞 − 𝑝 0 ∣2 > 0.

𝑑
𝑑𝑡
This implies that ⟨𝑁 ○ 𝛼(𝑡), 𝑞 − 𝛼(𝑡)⟩ < 0 for 𝑡 < 0 near 0, and therefore 𝑞 ∈ D, in
contradiction to our hypothesis. We thus conclude that 𝑝 = 𝑝 0 ∈ 𝑆 and R 3 ∖(C∪D) ⊆ 𝑆
— which, together with what we have shown above, also proves 𝜕C = 𝜕D = 𝑆.
Finally, let us take 𝜆 > 0 such that 𝑆 ⊆ 𝐷 𝜆 = {𝑞 ∈ R 3 ∶ ∣𝑞∣ ≤ 𝜆}. The complement
of 𝐷 𝜆 , being connected and disjoint to 𝑆, is contained in either of the sets C or D; but
as D is not bounded, D ∩ (R 3 ∖ 𝐷 𝜆 ) is non-empty and therefore R 3 ∖ 𝐷 𝜆 ⊆ D. Thus
C ⊆ 𝐷 𝜆 and C is bounded. ◻

The next result is used immediately afterwards in the proof of the rigidity of the
sphere, and in the proof of a Blaschke formula in the next section (Theorem 5.5.3).

Theorem 5.4.3 (Minkowski integral formulas)

Let 𝑆 be an oval and 𝑝 0 ∈ R 3 . Then, denoting by 𝐴 the area of 𝑆, we have

𝐴 = − ∫ 𝐻(𝑝)⟨𝑝 − 𝑝 0 , 𝑁(𝑝)⟩ 𝑑𝜎, (5.7)

𝑆

∫ 𝐻(𝑝) 𝑑𝜎 = − ∫ 𝐾(𝑝)⟨𝑝 − 𝑝 0 , 𝑁(𝑝)⟩ 𝑑𝜎. (5.8)

𝑆 𝑆

Both formulas (5.7) and (5.8) are valid for every compact surface 𝑆 ⊆ R 3 , and
indeed we prove (5.7) [but not (5.8)] in this generality. We will make use of the results
of Section 4.4, in particular of Theorem 4.4.5 (first variation of area).
Let us consider the vector field on 𝑆 given by v(𝑝) = 𝑝 − 𝑝 0 , and let 𝑆𝑡 be the
variation of 𝑆 induced by v. Since 𝑆𝑡 is the image of 𝑆 under the homothety with
center 𝑝 0 and ratio 1 + 𝑡, its area is 𝐴(𝑡) = (1 + 𝑡) 𝐴, which implies 𝐴′ (0) = 2𝐴.
2

Comparing with Theorem 4.4.5, we obtain

2𝐴 = − ∫ ⟨v(𝑝), 2𝐻(𝑝)𝑁(𝑝)⟩ 𝑑𝜎 = −2 ∫ 𝐻(𝑝)⟨𝑝 − 𝑝 0 , 𝑁(𝑝)⟩ 𝑑𝜎,

𝑆 𝑆

which proves (5.7).

The proof of (5.8) follows the steps of the previous one. We denote by
𝑀 = ∫𝑆 𝐻(𝑝) 𝑑𝜎 the total mean curvature of 𝑆, and we denote by 𝑀(𝑡) the
5.4 Ovals (First Part): The Rigidity of the Sphere 155

analogous quantity for 𝑆𝑡 . Note that the parameterization Φ(𝑢, 𝑣) of 𝑆 is transformed,

when composed with the homothety just mentioned, into the parameterization
(1 + 𝑡)Φ(𝑢, 𝑣) − 𝑡 𝑝 0 of 𝑆𝑡 ; the coefficients of the first fundamental form are multiplied
2
by (1 + 𝑡) and the coefficients of the second by 1 + 𝑡. Hence the area element is
2
multiplied by (1 + 𝑡) and, by formula (3.8) in Section 3.2, 𝐻 is divided by 1 + 𝑡.
We hence have 𝑀(𝑡) = (1 + 𝑡)𝑀 and therefore 𝑀 ′ (0) = 𝑀; (5.8) is an immediate
consequence of the lemma below:
Lemma 5.4.4 Let 𝑆 be an oval and let 𝑆𝑡 be the variation of 𝑆 induced by a vector
field v. Then, if 𝑀(𝑡) denotes the total mean curvature of 𝑆𝑡 , we have

𝑀 ′ (0) = − ∫ 𝐾(𝑝)⟨v(𝑝), 𝑁(𝑝)⟩ 𝑑𝜎.

𝑆

Proof Given a parameterization (𝑈, Φ) of 𝑆, compatible with the orientation, let

Φ𝑡 (𝑢, 𝑣) = Φ(𝑢, 𝑣) + 𝑡v(𝑢, 𝑣);

(𝑈, Φ𝑡 ) is a parameterization of 𝑆𝑡 for sufficiently small ∣𝑡∣. Let us denote by 𝑁 𝑡 (𝑢, 𝑣)

the field of normal vectors in Φ𝑡 (𝑈). By formula (4.18) of Section 4.4 we have

1
𝐻 𝑡 ∣Φ𝑢𝑡 × Φ𝑡𝑣 ∣ = − ⟨Φ𝑢𝑡 × 𝑁 𝑣𝑡 + 𝑁𝑢𝑡 × Φ𝑡𝑣 , 𝑁 𝑡 ⟩. (5.9)
2
We now look at the derivative of the second term in (5.9). We have
𝑑
⟨Φ𝑡 × 𝑁 𝑣𝑡 + 𝑁𝑢𝑡 × Φ𝑡𝑣 , 𝑁 𝑡 ⟩∣𝑡=0
𝑑𝑡 𝑢
𝑑 𝑑
=⟨Φ𝑢 × 𝑁 𝑣 + 𝑁𝑢 × Φ𝑣 , 𝑁 𝑡 ∣𝑡=0 ⟩ + ⟨ {Φ𝑢𝑡 × 𝑁 𝑣𝑡 + 𝑁𝑢𝑡 × Φ𝑡𝑣 }∣𝑡=0 , 𝑁⟩
𝑑𝑡 𝑑𝑡
𝜕v 𝜕v 𝑑 𝑑
=⟨ × 𝑁 𝑣 + 𝑁𝑢 × , 𝑁⟩ + ⟨Φ𝑢 × ( 𝑁 𝑣𝑡 ∣𝑡=0 ) + ( 𝑁𝑢𝑡 ∣𝑡=0 ) × Φ𝑣 , 𝑁⟩ (5.10)
𝜕𝑢 𝜕𝑣 𝑑𝑡 𝑑𝑡

(because 𝑑𝑡 𝑁 ∣𝑡=0 is orthogonal to 𝑁, and 𝜙𝑢 × 𝑁 𝑣 + 𝑁𝑢 × Φ𝑣 is collinear with 𝑁).

𝑑 𝑡

If we consider the field of tangent vectors w = 𝑑𝑡

𝑑 𝑡
𝑁 ∣𝑡=0 and use Proposition 4.4.1,
the right-hand side of (5.10) becomes

𝜕w 𝜕w
⟨Φ𝑢 × + × Φ𝑣 , 𝑁⟩ = (Div w) ∣Φ𝑢 × Φ𝑣 ∣. (5.11)
𝜕𝑣 𝜕𝑢

Regarding the left-hand side in (5.9), we note that 𝑁∶ 𝑆 → S2 induces in S2 the orien-
tation for which the mean curvature is negative, equal to −1. Using Proposition 4.4.1
we obtain

, 𝑁⟩ = ∣𝑁𝑢 × 𝑁 𝑣 ∣ {Div∗ (v⊺ ) + 2⟨v, 𝑁⟩}

𝜕v 𝜕v
⟨ × 𝑁 𝑣 + 𝑁𝑢 ×
𝜕𝑢 𝜕𝑣 (5.12)
∗ ⊺
= Div (v )∣𝑁𝑢 × 𝑁 𝑣 ∣ + 2𝐾⟨v, 𝑁⟩∣Φ𝑢 × Φ𝑣 ∣,
156 5 The Global Geometry of Surfaces

where Div∗ (v⊺ ) indicates the divergence of v⊺ as the tangent vector field to S2
— which is fine since 𝑁 is a diffeomorphism and the tangent spaces 𝑇 𝑝 𝑆 and
𝑇𝑁 ( 𝑝) S2 are parallel. Now the total mean curvature of Φ𝑡 (𝑈) ⊆ 𝑆𝑡 is 𝑀Φ (𝑡) =
∫𝑈 𝐻 ∣Φ𝑢 × Φ𝑣 ∣ 𝑑𝑢 𝑑𝑣, so that
𝑡 𝑡 𝑡

𝑀Φ′ (0) = ∫
𝑑
{𝐻 𝑡 ∣Φ𝑢𝑡 × Φ𝑡𝑣 ∣}∣𝑡=0 𝑑𝑢 𝑑𝑣. (5.13)
𝑈 𝑑𝑡
Combining formulas (5.9)-(5.13) and using Corollary 4.4.3, we finally obtain
1 1
𝑀 ′ (0) = − ∫ Div w 𝑑𝜎 − ∫ Div∗ (v⊺ ) 𝑑𝜎 − ∫ 𝐾⟨v, 𝑁⟩ 𝑑𝜎
2 𝑆 2 S2 𝑆

= − ∫ 𝐾⟨v, 𝑁⟩ 𝑑𝜎. ◻
𝑆

Theorem 5.4.5 (Rigidity of the Sphere)

Let 𝑆 ⊆ R 3 be a compact surface with constant curvature 𝐾. Then 𝑆 is a sphere.

Proof Being compact, 𝑆 has some point of positive curvature (see Exercise 57 in
Section 3.1), and therefore 𝐾 > 0. 𝑆 is then an oval, and we can fix the orientation
𝑁∶ 𝑆 → S2 such that ⟨𝑁(𝑝), 𝑞 − 𝑝⟩ > 0 for every pair of points 𝑝 ≠ 𝑞 on 𝑆. With this
orientation we have√ 𝐻(𝑝) > 0 at every point 𝑝 ∈ 𝑆.
Let us put 𝑘 = 𝐾. By the inequality between the arithmetic and geometric means,
we have 𝑘 𝐻(𝑝) − 𝐾 ≥ 0, and the equality holds at 𝑝 if and only if 𝑝 is an umbilical
point. Let 𝑝 0 ∈ R 3 (whose existence is guaranteed by Proposition 5.4.2) be such that
⟨𝑁(𝑝), 𝑝 0 − 𝑝⟩ > 0 for all 𝑝 ∈ 𝑆, and let us put

Δ0 = ∫ (𝑘 𝐻 − 𝐾) 𝑑𝜎 = 𝑘 𝑀 − 𝑘 2 𝐴,
𝑆

Δ1 = ∫ (𝑘 𝐻 − 𝐾)⟨𝑁, 𝑝 0 − 𝑝⟩ 𝑑𝜎
𝑆

= 𝑘 ∫ 𝐻⟨𝑁, 𝑝 0 − 𝑝⟩ 𝑑𝜎 − ∫ 𝐾⟨𝑁, 𝑝 0 − 𝑝⟩ 𝑑𝜎
𝑆 𝑆
= 𝑘𝐴 − 𝑀 (by Minkowski’s formulas).

We thus have Δ0 = −𝑘Δ1 , but their definitions ensure that Δ0 and Δ1 are both non-
negative. Hence, we have Δ0 = Δ1 = 0, and therefore the integrand function 𝑘 𝐻 − 𝐾
is constantly zero. All points of 𝑆 are therefore umbilical and, by Proposition 3.1.5,
𝑆 is a sphere. ◻

In particular, all surfaces in R 3 that are isometric to the sphere are themselves
spheres, and hence the theorem speaks of the rigidity of the sphere. The theorem has
the following generalization (see [13]): if 𝑓 ∶ 𝑆1 → 𝑆2 is an isometry between two oval
surfaces then 𝑓 is the restriction of an isometry of R 3 (which, as is known, is the
composite of a linear isometry with a translation).
We now give another result of the same kind as Theorem 5.4.5.
5.5 Ovals: Areas and Volumes; Surfaces of Constant Width 157

Theorem 5.4.6 Let 𝑆 ⊆ R 3 be an oval surface of constant mean curvature 𝐻. Then

𝑆 is a sphere.

We choose the orientation 𝑁 and the point 𝑝 0 as above, and notice that 𝐻 2 − 𝐾 ≥ 0,
with equality only at the umbilical points. Then

∫ (𝐻 − 𝐾)⟨𝑁, 𝑝 0 − 𝑝⟩ 𝑑𝜎 = 𝐻 ∫ 𝐻⟨𝑁, 𝑝 0 − 𝑝⟩ 𝑑𝜎 − ∫ 𝐾⟨𝑁, 𝑝 𝑜 − 𝑝⟩ 𝑑𝜎

2
𝑆 𝑆 𝑆

= 𝐻 𝐴 − ∫ 𝐻 𝑑𝜎 = 0
𝑆

— and from this, since the integrand function is non-negative, it follows that 𝐻 2 −𝐾 ≡ 0,
and therefore 𝑆 is a sphere.
Let us point out that Theorem 5.4.6 remains valid without the assumption that 𝑆
has positive curvature: any compact surface 𝑆 ⊆ R 3 with constant mean curvature is
a sphere. An accessible proof of this result appears in Osserman’s paper [20].

Exercise
130. Show that the open subset C bounded by the oval surface 𝑆 (Proposition 5.4.2) is
given by {(1 − 𝑡)𝑝 + 𝑡𝑞∶ 𝑝, 𝑞 ∈ 𝑆, 0 < 𝑡 < 1}.

5.5 Ovals: Areas and Volumes; Surfaces of Constant Width

In this section we prove a number of formulas involving the area, the total mean
curvature, and the volume bounded by an oval surface. Some of these formulas
concern convex bodies in R 3 — which include not only solids bounded by oval
surfaces but also by convex polyhedra. As we noted in Section 1.1, the length of a
regular curve 𝛾 can be computed by considering polygonal lines inscribed in 𝛾 with
increasing numbers of segments. The generalization to surfaces is not so obvious:
there are approximations of the cylindrical surface C = {𝑥 2 + 𝑦 2 = 1, 0 ≤ 𝑧 ≤ 1} by
polyhedra whose vertices are all in C but whose areas do not converge to that of C
(see Exercise 131).
The solution, for a compact convex surface 𝑆, is to approximate 𝑆 by convex
polyhedra: if P1 and P2 are convex polyhedra such that P1 ≺ 𝑆 ≺ P2 , where the sign
≺ means “is in”, then we have the expected inequality of areas

𝐴(P1 ) < 𝐴(𝑆) < 𝐴(P2 ). (5.14)

Knowing that we can find P1 and P2 such that P1 ≺ 𝑆 ≺ P2 and with 𝐴(P2 ) − 𝐴(P1 )
as small as we want, it follows from (5.14) that

𝐴(𝑆) = sup 𝐴(P1 ) = inf 𝐴(P2 ). (5.15)

P1 ≺𝑆 𝑆≺P2
158 5 The Global Geometry of Surfaces

Formula (5.15) suggests that for certain formulas involving areas of convex surfaces,
it is sufficient to prove them for polyhedra; moreover, it provides a definition of area
that does not depend on whether the surface in question is regular or not.
The interested reader can find the details of this construction in several books on
convexity (e.g., [9]). For now, we will make use of (5.15) to prove a Cauchy formula.
Let 𝑆 be a compact convex surface. Given 𝑝 ∈ S2 , let us denote by 𝐴𝑆 (𝑝) the area of
the orthogonal projection of 𝑆 onto 𝑇 𝑝 S2 .
Theorem 5.5.1 (Cauchy)
The area of 𝑆 is given by
1
𝐴= ∫ 𝐴𝑆 (𝑝) 𝑑𝜎. (5.16)
𝜋 S2

Proof Let 𝐹 be a polygon included in a plane Π, and let v be a unit vector orthogonal
to Π. Let 𝜃 ∈ [0, 𝜋] be the angle between v and Op (𝑝 ∈ S2 ) ; the area of the
orthogonal projection of 𝐹 onto 𝑇 𝑝 S2 is 𝐴𝐹 (𝑝) = 𝐴(𝐹)∣ cos 𝜃∣. Using spherical
coordinates (𝜑, 𝜃) relative to an orthonormal frame in which the third vector is v, we
then have
2𝜋 𝜋
∫ 𝐴𝐹 (𝑝) 𝑑𝜎 = ∫ (∫ 𝐴(𝐹)∣ cos 𝜃∣ sin 𝜃 𝑑𝜃) 𝑑𝜑
S2 0 0
𝜋
= 2𝜋 𝐴(𝐹) ∫ ∣ cos 𝜃∣ sin 𝜃 𝑑𝜃 = 2𝜋 𝐴(𝐹).
0

Thus
1
𝐴(𝐹) = ∫ 𝐴𝐹 (𝑝) 𝑑𝜎. (5.17)
2𝜋 S2
Now let P be a convex polyhedron and 𝐹1 , . . . , 𝐹𝑘 its faces. Except for the points on
the boundary (which are negligible), each point of the orthogonal projection of P
in the plane 𝑇 𝑝 S2 is the image of exactly two points of P, belonging to two distinct
faces. We then have
1 𝑘
𝐴P (𝑝) = ∑ 𝐴𝐹𝑖 (𝑝);
2 𝑖=1
and, using (5.17), we obtain
𝑘
1 1 𝑘
𝐴(P) = ∑ 𝐴(𝐹𝑖 ) = ∫ 2 ( ∑ 𝐴𝐹𝑖 (𝑝)) 𝑑𝜎
𝜋 S 2 𝑖=1
𝑖=1 (5.18)
1
= ∫ 𝐴P (𝑝) 𝑑𝜎.
𝜋 S2
Noting that if P1 ≺ 𝑆 ≺ P2 then 𝐴P1 (𝑝) ≤ 𝐴𝑆 (𝑝) ≤ 𝐴P2 (𝑝) for every 𝑝 ∈ S2 , it
follows from (5.18) that
1
𝐴(P1 ) ≤ ∫ 𝐴𝑆 (𝑝) 𝑑𝜎 ≤ 𝐴(P2 ),
𝜋 S2
5.5 Ovals: Areas and Volumes; Surfaces of Constant Width 159

which, together with (5.15), proves the theorem. ◻

As in the case of curves (see Section 1.6), we say that an oval surface 𝑆 has
constant curvature 𝔏 if, for every pair of parallel planes tangent to 𝑆 at two distinct
points, the distance between them is equal to 𝔏.
Corollary 5.5.2 If 𝑆 has constant curvature 𝔏 and area 𝐴 then

𝐴 ≤ 𝜋 𝔏2 ,

with equality if and only if 𝑆 is a sphere.

Sketch of the proof. The orthogonal projection of 𝑆 onto the plane 𝑇 𝑝 S2 , which we
denote by 𝑆(𝑝), is bounded by a curve of constant width 𝔏, having (by Theorem 1.6.3)
perimeter 𝜋𝔏. By the isoperimetric inequality (Theorem 1.8.2), the area of 𝑆(𝑝)
satisfies the inequality
𝜋𝔏2
𝐴𝑆 (𝑝) ≤ (5.19)
4
— with equality only in the case where 𝑆(𝑝) is a disk with radius 𝔏/2. Combining
(5.19) and (5.16), we obtain

1 𝜋𝔏2
𝐴≤ ∫ 𝑑𝜎 = 𝜋𝔏2 ,
𝜋 S2 4
with equality if and only if every orthogonal projection 𝑆(𝑝) of 𝑆 is a disk with radius
𝔏/2 — and this last condition implies that 𝑆 is a sphere (see Exercise 132). ◻
Let us further consider an oval surface 𝑆, and on it the field of normal vectors 𝑁
pointing into 𝑆. The variation of 𝑆 induced by −𝑁 is, as we have already defined,
the family 𝑆𝑡 = {𝑝 − 𝑡𝑁(𝑝)∶ 𝑝 ∈ 𝑆}. But in this case, the surfaces 𝑆𝑡 are parallel to
𝑆, either surrounding it (for 𝑡 > 0) or being surrounded by it (for 𝑡 < 0) at a fixed
distance equal to ∣𝑡∣. Furthermore 𝑆𝑡 is a surface for all 𝑡 ≥ 0 (Exercise 135), though
not for all 𝑡 < 0.
Given a parameterization Φ(𝑢, 𝑣) of 𝑆, let us put

Φ𝑡 (𝑢, 𝑣) = Φ(𝑢, 𝑣) − 𝑡𝑁(𝑢, 𝑣).

Now we have

Φ𝑢𝑡 × Φ𝑡𝑣 = Φ𝑢 × Φ𝑣 − 𝑡{Φ𝑢 × 𝑁 𝑣 + 𝑁𝑢 × Φ𝑣 } + 𝑡 2 {𝑁𝑢 × 𝑁 𝑣 }

(5.20)
= (1 + 2𝑡𝐻 + 𝑡 2 𝐾){Φ𝑢 × Φ𝑣 } [per (4.18), Section 4.4].

For 𝑡 > 0 (and for negative 𝑡 near 0), the expression 1 + 2𝑡𝐻 + 𝑡 2 𝐾 is positive. As a
consequence, and taking absolute values on both sides of (5.20), we conclude that
the area of 𝑆𝑡 is given by

𝐴(𝑡) = ∫ (1 + 2𝑡𝐻 + 𝑡 2 𝐾) 𝑑𝜎 = 𝐴 + 2𝑀𝑡 + 4𝜋𝑡 2 , (5.21)

𝑆
160 5 The Global Geometry of Surfaces

where 𝐴 and 𝑀 are the area and the total mean curvature of 𝑆 (where we recall
that as 𝑆 is diffeomorphic to the sphere its total curvature ∫𝑆 𝐾 𝑑𝜎 is 4𝜋). Formula
(5.21) thus expresses the remarkable fact that the area of a surface parallel to 𝑆 is a
polynomial function of its distance to 𝑆.

We can apply (5.21) to deduce an analogous formula involving volumes. Let us

denote by 𝑉(𝑡) the volume of the region bounded by 𝑆𝑡 , where 𝑉 = 𝑉(0). Assuming
that 𝑡 > 0, the difference 𝑉(𝑡) − 𝑉 is the volume of the region between 𝑆 and 𝑆𝑡 .
This region is the union of the images of the functions Ψ(𝑢, 𝑣, 𝑠) = Φ𝑠 (𝑢, 𝑣), with
(𝑢, 𝑣, 𝑠) ∈ 𝑈 × [0, 𝑡], where (𝑈, Φ) is a parameterization of 𝑆. Now

𝜕Ψ 𝜕Ψ 𝜕Ψ
∣ det 𝐽Ψ∣ = ∣⟨ × , ⟩∣ = ∣⟨Φ𝑢𝑠 × Φ𝑠𝑣 , 𝑁⟩∣ = ∣Φ𝑢𝑠 × Φ𝑠𝑣 ∣,
𝜕𝑢 𝜕𝑣 𝜕𝑠

and therefore the volume of the image of Ψ is

𝑡
𝑉(Ψ) = ∫ (∬ ∣ det 𝐽Ψ∣ 𝑑𝑢 𝑑𝑣) 𝑑𝑠
0 𝑈
𝑡 𝑡 (5.22)
=∫ (∬ ∣Φ𝑢𝑠 × Φ𝑠𝑣 ∣ 𝑑𝑢 𝑑𝑣) 𝑑𝑠 = ∫ 𝐴(Φ𝑠 ) 𝑑𝑠,
0 𝑈 0

where 𝐴(Φ𝑠 ) denotes the area of Φ𝑠 (𝑈). Using (5.22) and (5.21), we finally obtain
𝑡 4𝜋 3
𝑉(𝑡) = 𝑉 + ∫ 𝐴(𝑠) 𝑑𝑠 = 𝑉 + 𝐴𝑡 + 𝑀𝑡 2 + 𝑡 , (5.23)
0 3
and here it is, the promised formula, which is also valid for negative 𝑡 near 0. There
is a formula analogous to (5.23) for the area of planar regions bounded by parallel
curves as well, which we will give in Exercise 133.

To conclude the section, we again assume that 𝑆 has constant width 𝔏. As in the
case of curves, the antipode A(𝑝) of 𝑝 ∈ 𝑆 is given by

A(𝑝) = 𝑝 + 𝔏𝑁(𝑝),

and A∶ 𝑆 → 𝑆 is an involutive diffeomorphism (i.e., A ○ A = id). Since 𝑆 is inside the

sphere with center A(𝑝) and radius 𝔏, and is tangent to the sphere at the point 𝑝, the
principal curvatures at 𝑝 are both bounded by 1/𝔏 (see Exercise 57 in Section 3.1).
̃ = A ○ Φ is
If Φ(𝑢, 𝑣) is a parameterization of 𝑆 then, since A is a diffeomorphism, Φ
another parameterization. But from the expression of A, we see that with the above
̃ = Φ−𝔏 , and from this, using (5.20), we take
notation Φ
̃𝑢 × Φ
Φ ̃ 𝑣 = (1 − 2𝐻𝔏 + 𝐾𝔏2 ){Φ𝑢 × Φ𝑣 } (5.24)

— a formula that ensures that the expression 1 − 2𝐻𝔏 + 𝐾𝔏2 never vanishes. In terms
of the principal curvatures 𝑘 1 , 𝑘 2 , we have

1 − 2𝐻𝔏 + 𝐾𝔏2 = (𝑘 1 𝔏 − 1)(𝑘 2 𝔏 − 1) > 0,

5.5 Ovals: Areas and Volumes; Surfaces of Constant Width 161

since both factors are non-negative. It then follows from (5.24) that for every
continuous function 𝑓 ∶ 𝑆 → R we have

∫ 𝑓 𝑑𝜎 = ∫ ( 𝑓 ○ A)(1 − 2𝐻𝔏 + 𝐾𝔏 ) 𝑑𝜎.

2
(5.25)
𝑆 𝑆

The formula (5.25) has interesting consequences. For example, taking 𝑓 ≡ 1, and
using the fact that the total curvature of 𝑆 is 4𝜋, we obtain

𝐴 = 𝐴 − 2𝑀𝔏 + 4𝜋𝔏2 ⇒ 𝑀 = 2𝜋𝔏

— that is, the total mean curvature of a surface of constant width 𝔏 is 𝑀 = 2𝜋𝔏.
For another application of (5.25), we will prove a Blaschke formula that relates
the area 𝐴 of a surface of constant width to the volume 𝑉 of the region bounded by it.
Theorem 5.5.3 (Blaschke)
If 𝑆 has constant width 𝔏 then 𝑉 = 1
2 𝐴𝔏 − 13 𝜋𝔏3 .
Proof Let 𝑝 0 be a point of the region Ω bounded by 𝑆. The volume of Ω is given by
1
𝑉= ∫ ⟨𝑁(𝑝), 𝑝 0 − 𝑝⟩ 𝑑𝜎(𝑝)
3 𝑆
(this formula corresponds to regarding Ω as a union of “infinitesimal cones” of vertex
𝑝 0 and base in 𝑆, and is a particular case of the divergence theorem in R 3 — see e.g.
[16], p. 493). By (5.25), denoting by 𝑝̃ the antipode of 𝑝, we also have
1
𝑉= ∫ ⟨𝑁( 𝑝̃), 𝑝 0 − 𝑝̃⟩(1 − 2𝐻𝔏 + 𝐾𝔏 ) 𝑑𝜎(𝑝).
2
3 𝑆
Since 𝑁( 𝑝̃) = −𝑁(𝑝) and 𝑝̃ = 𝑝 + 𝔏𝑁(𝑝), we obtain from this, using Minkowski’s
formulas (Theorem 5.4.3) and the fact that 𝑀 = 2𝜋𝔏,
1
𝑉= ∫ (−⟨𝑁(𝑝), 𝑝 0 − 𝑝⟩ + 𝔏)(1 − 2𝐻𝔏 + 𝐾𝔏 ) 𝑑𝜎(𝑝)
2
3 𝑆
1 2 1
= −𝑉 + 𝐴𝔏 + (∫ 𝐻⟨𝑝 0 − 𝑝, 𝑁⟩ 𝑑𝜎) 𝔏− (∫ 𝐾⟨𝑝 0 − 𝑝, 𝑁⟩ 𝑑𝜎) 𝔏2
3 3 𝑠 3 𝑆
1 2 1 2
= −𝑉 + 𝐴𝔏 + 𝐴𝔏 − 𝑀𝔏2 = −𝑉 + 𝐴𝔏 − 𝜋𝔏3 ,
3 3 3 3
which concludes the proof of the theorem. ◻

Theorem 5.5.3 shows that among surfaces with a certain constant width 𝔏 those
with the largest (or smallest) area are also those that enclose the largest (or smallest)
volume. From Corollary 5.5.2 it then follows that the maximal volume is that of the
sphere. The problem of finding the surface of a given constant width with minimal
area (or minimal volume) is still open (see [7]).

Exercises
162 5 The Global Geometry of Surfaces

131. The segment C of the cylinder 𝑥 2 + 𝑦 2 = 1 bounded by the planes 𝑧 = 0 and 𝑧 = 1

has area 2𝜋. Given 𝑘 ≥ 1 and 𝑛 ≥ 3, consider the 𝑘 + 1 circles in C given by
1 2 𝑘 −1
𝑧 = 0, , ,..., , 1.
𝑘 𝑘 𝑘
Divide each of these circles into 𝑛 equal arcs, so that the ends of the arcs in each circle
are vertically at the midpoints of the arcs of the preceding circle. The ends of these
arcs define regular polygons of 𝑛 sides inscribed in each of the circles. Joining each
vertex to the two vertices closest to the neighboring polygons, we obtain a polyhedron
P(𝑘, 𝑛) whose faces are triangles, all of which are congruent.
(a) Show that the area of P(𝑘, 𝑛) is
√
𝜋 𝜋
𝐴(𝑘, 𝑛) = 2𝑛 sin ( ) 1 + 4𝑘 2 sin4 ( )⋅
𝑛 2𝑛

(b) Compute lim 𝐴(𝑛𝑟 , 𝑛) for 𝑟 = 1, 2, 3. What conclusion can you draw?
𝑛→+∞

Figure 5.5

132. Let 𝑆 be an oval surface and 𝑟 > 0 such that the orthogonal projection of 𝑆 onto
each plane P ⊆ R 3 is a disk with radius 𝑟. Show that:
(a) for every plane P, the surface 𝑆 is inscribed in a straight circular cylinder C of
height 2𝑟 whose base has radius 𝑟 and is parallel to P;
(b) if 𝛾 is the equator of C then 𝛾 ⊆ 𝑆 and the normals to C and to 𝑆 along 𝛾
coincide;
(c) each normal section of 𝑆 is a circle with radius 𝑟, and therefore 𝑆 is a sphere.
133. Let 𝛼 be a convex, regular, closed planar curve. Denote by 𝛼𝑡 the curve parallel
to 𝛼 at distance 𝑡 from 𝛼, and by 𝔘(𝑡) the area of the region bounded by 𝛼𝑡 . Show
that
𝔘(𝑡) = 𝔘 + 𝑙(𝛼)𝑡 + 𝜋𝑡 2 [where 𝔘 = 𝔘(0)].
Hint: verify that 𝑙(𝛼𝑡 ) = 𝑙(𝛼) + 2𝜋𝑡 and 𝔘(𝑡) = 𝔘 + ∫0 𝑙(𝛼𝑠 ) 𝑑𝑠.
𝑡
5.6 Abstract Surfaces. The Hyperbolic Plane 163

Note: since any convex curve can be approximated by a regular convex curve, the
obtained formula is valid for any convex curve.
134. Let 𝑆 be an oval surface, and denote by 𝑙 𝑆 (𝑝) the perimeter of the orthogonal
projection of 𝑆 onto 𝑇 𝑝 S2 . Show that the total mean curvature of 𝑆 is given by

1
𝑀= ∫ 𝑙 𝑆 (𝑝) 𝑑𝜎.
2𝜋 S2
Hint: use Exercise 133 and Theorem 5.5.1 to obtain an expression for the area 𝐴(𝑡)
of the surface 𝑆𝑡 parallel to 𝑆, and compare (5.21) with the formula obtained.
135. Let 𝑆 be an oval surface and 𝑁 be the normal field pointing into 𝑆. Show that,
for all 𝑡 ≥ 0, the surface 𝑆𝑡 = {𝑝 − 𝑡𝑁(𝑝)∶ 𝑝 ∈ 𝑆} is diffeomorphic to 𝑆.

5.6 Abstract Surfaces. The Hyperbolic Plane

All measures intrinsic to a surface depend, as we know, on the inner product defined
on the tangent space (i.e., the first fundamental form) — which, so far, has simply been
the usual scalar product restriction on R 3 . But nothing prevents us from considering
other inner products, obtaining in this way surfaces whose metric structure is not
inherited from R 3 . The next step is to dispense with the ambient space, defining
abstract surfaces that are not even diffeomorphic to surfaces of R 3 .
Let 𝑈 ⊆ R 2 be an open and connected region, and let 𝐸, 𝐹, 𝐺∶ 𝑈 → R be
∞
𝐶 -functions such that
● 𝐸 and 𝐺 are strictly positive;
● 𝐸𝐺 − 𝐹 2 > 0 at all points of 𝑈.
For every point (𝑢, 𝑣) ∈ 𝑈 we define an inner product ⟨⋅, ⋅⟩(𝑢,𝑣) on 𝑇(𝑢,𝑣)𝑈 = R 2 as
follows:

⟨(𝑥 1 , 𝑦 1 ), (𝑥 2 , 𝑦 2 )⟩(𝑢,𝑣) = 𝑥1 𝑥 2 𝐸(𝑢, 𝑣) + (𝑥 1 𝑦 2 + 𝑦 1 𝑥 2 )𝐹(𝑢, 𝑣) + 𝑦 1 𝑦 2 𝐺(𝑢, 𝑣).

If 𝛼∶ [𝑎, 𝑏] → 𝑈 is a regular curve then, in this inner product, the norm of 𝛼′ (𝑡) is
given by
√ √
2 2
𝐼 𝛼(𝑡) (𝛼′ (𝑡)) = 𝑢 ′ (𝑡) 𝐸(𝛼(𝑡)) + 2𝑢 ′ (𝑡)𝑣 ′ (𝑡)𝐹(𝛼(𝑡)) + 𝑣 ′ (𝑡) 𝐺(𝛼(𝑡)),

and the length of 𝛼 is

𝑏√
𝑙(𝛼) = ∫ 𝐼 𝛼(𝑡) (𝛼′ (𝑡)) 𝑑𝑡.
𝑎
The area of a region Δ ⊂ 𝑈 is computed, as usual, by
√
∬ 𝐸𝐺 − 𝐹 2 𝑑𝑢 𝑑𝑣.
𝑈
164 5 The Global Geometry of Surfaces

The inner product thus imposed on 𝑈 determines therefore all intrinsic measures:
lengths of curves, angles between vectors, areas of regions. We say that 𝑈 has been
equipped with a Riemannian metric. We can define, for such a metric, the notions of
Gaussian curvature, covariant derivative, geodesic — and, under the assumption that
𝑈 is oriented, that of geodesic curvature. To this end, we make use of the formulas in
Chapter 4 that express such notions using 𝐸, 𝐹, 𝐺, Christoffel’s symbols, and their
derivatives [such as (4.8), (4.9), (4.31)].

Examples 5.6.1 A. Let us take R 2 with the Riemannian metric given by

𝑑𝑠2 = 𝑑𝑢 2 + {2 + cos(𝑢 + 𝑣)}2 𝑑𝑣 2

(which is a shorthand for 𝐸 ≡ 1, 𝐹 ≡ 0, 𝐺(𝑢, 𝑣) = {2 + cos(𝑢 + 𝑣)}2 ). Using formula

(4.38) in Section 4.6, we find that the Gaussian curvature of this metric is

cos(𝑢 + 𝑣)
𝐾(𝑢, 𝑣) = ⋅
{2 + cos(𝑢 + 𝑣)}

B. Let us consider, also in R 2 , the metric

1
𝑑𝑠2 = 2
(𝑑𝑢 2 + 𝑑𝑣 2 ).
1 + (𝑢 2 + 𝑣2)

This is a conformal metric, in the sense that angles are measured as usual in R 2 ,
since 𝐹 ≡ 0 and 𝐸 = 𝐺. Another particularity is that with this metric R 2 has finite
area and is not a complete surface (the proof of this statement is Exercise 136). ◻

An abstract surface (of class 𝐶 ∞ ) is a connected topological space 𝑆 equipped

with an atlas A = ((𝑈 𝛼 , Φ 𝛼 )) 𝛼∈I such that:

● every 𝑈 𝛼 is an open subset of R 2 , Φ 𝛼 (𝑈 𝛼 ) is an open subset of 𝑆, and

𝑆 = ⋃ Φ 𝛼 (𝑈 𝛼 );
𝛼∈I
● Φ 𝛼 ∶ 𝑈 𝛼 → 𝑆 is a homeomorphism onto its image;
● if Φ 𝛼 (𝑈 𝛼 ) ∩ Φ𝛽 (𝑈𝛽 ) = 𝑉 ≠ ∅ then Φ−1 −1 −1 ∞
𝛽 ○ Φ 𝛼 ∶ Φ 𝛼 (𝑉) → Φ 𝛽 (𝑉) is 𝐶 .

To avoid pathologies (see the appendix to volume I of [25] for a catalogue of them)
we further impose certain conditions on the topology of the surface 𝑆, namely: any
two distinct points of 𝑆 have disjoint open neighborhoods (𝑆 is Hausdorff ) and there
exists a subset of 𝑆 which is dense and countable (𝑆 is separable).
By Proposition 2.2.1, and since the mentioned topological requirements are
satisfied by any subset of R 𝑛 , any regular connected surface of R 3 is an abstract
surface. What we have done was to require the coordinate changes on 𝑆 to be
𝐶 ∞ -diffeomorphisms, which allows us to define all sorts of notions using local
coordinates. Note that a surface does not determine an atlas uniquely: we can add to
or take away from a given atlas a few mappings, and as long as the remaining ones
still cover 𝑆 and coordinate changes remain 𝐶 ∞ , we obtain several different atlases;
5.6 Abstract Surfaces. The Hyperbolic Plane 165

what matters is that they all define the same differentiable structure on 𝑆, and thus the
same abstract surface.
Given an abstract surface 𝑆, it is not easy, now that we have no ambient space, to
define a tangent space 𝑇 𝑝 𝑆. Our approach, instead of defining it, is to explain how to
work with it in local coordinates. So if we have a curve that in local coordinates is
written 𝛼(𝑡) = Φ(𝑢(𝑡), 𝑣(𝑡)), we would still like it to be

𝛼′ (𝑡) = 𝑢 ′ (𝑡)Φ𝑢 + 𝑣 ′ (𝑡)Φ𝑣 (5.26)

— whatever the meaning of the vectors Φ𝑢 and Φ𝑣 . If we use other coordinates

𝑢, ̃
Ψ(̃ 𝑣 ), we should have

𝜕̃
𝑢 𝜕̃
𝑣
Φ𝑢 = Ψ𝑢̃ + Ψ̃𝑣 ,
𝜕𝑢 𝜕𝑢
𝜕̃
𝑢 𝜕̃
𝑣
Φ𝑣 = Ψ𝑢̃ + Ψ̃𝑣 ;
𝜕𝑣 𝜕𝑣
and, replacing in (5.26), we obtain

𝛼′ (𝑡) = (
𝜕̃𝑢 ′ 𝜕̃𝑢 ′ 𝑣 ′
𝜕̃ 𝑣 ′
𝜕̃
𝑢 (𝑡) + 𝑣 (𝑡)) Ψ𝑢̃ + ( 𝑢 (𝑡) + 𝑣 (𝑡)) Ψ̃𝑣
𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣
= 𝑢̃′ (𝑡)Ψ𝑢̃ + ̃
𝑣 ′ (𝑡)Ψ̃𝑣

— which shows that (5.26) is compatible with the change of coordinates. Hence for us
the tangent space to 𝑆 at the point Φ(𝑢, 𝑣) is simply the space generated by the two
independent vectors Φ𝑢 and Φ𝑣 . The transition matrix from the basis (Ψ𝑢̃, Ψ̃𝑣 ) to
(Φ𝑢 , Φ𝑣 ) is the Jacobian of Ψ−1 ○ Φ.
A Riemannian metric on an abstract surface 𝑆 is given by an inner product ⟨⋅, ⋅⟩ 𝑝
in the tangent space 𝑇 𝑝 𝑆 for every 𝑝 ∈ 𝑆. In local coordinates Φ(𝑢, 𝑣), the matrix
𝐸 𝐹
of this inner product with respect to the basis (Φ𝑢 , Φ𝑣 ) is [ ], and we usually
𝐹𝐺
̃ 𝐹,
require that 𝐸, 𝐹, 𝐺 are 𝐶 ∞ functions; the coefficients 𝐸, ̃ 𝐺 ̃ of other coordinates
Ψ(̃𝑢, ̃
𝑣 ) are obtained from these by the relation

𝐸̃ 𝐹̃
[ ̃ ̃] = {𝐽(Φ−1 ○ Ψ)}⊺ [ ] 𝐽(Φ−1 ○ Ψ),
𝐸 𝐹
𝐹𝐺 𝐹𝐺

where {𝐽(Φ−1 ○ Ψ)}⊺ denotes the transpose of 𝐽(Φ−1 ○ Ψ).

We can define, using the Riemannian metric (⟨⋅, ⋅⟩ 𝑝 ) 𝑝∈𝑆 , intrinsic notions such as
those of Gaussian curvature, covariant derivative, etc., using the formulas that express
such concepts via the coefficients 𝐸, 𝐹, 𝐺 and the Christoffel symbols Γ𝑖𝑘𝑗 . Of course,
we should now check that the definitions do not depend on the coordinates used
(which was previously unnecessary since such concepts had been defined without any
use of local coordinates), but we just assure the reader that such a check is possible;
for details, see [13].
Examples 5.6.2 A. Let us consider the equivalence relation on R 2 given by
166 5 The Global Geometry of Surfaces

𝑢, ̃
(𝑢, 𝑣) ∼ (̃ 𝑣)

if and only if both differences 𝑢 − 𝑢̃ and 𝑣 − ̃ 𝑣 are integers, and let [𝑢, 𝑣] denote
the equivalence class of the pair (𝑢, 𝑣). Let R 2 / ∼ be the set of equivalence classes
and Π∶ R 2 → R 2 / ∼ the quotient mapping, given by Π(𝑢, 𝑣) = [𝑢, 𝑣]. We define a
topology on R 2 / ∼ by 𝑈 ⊆ R 2 / ∼ being open if and only if Π −1 (𝑈) is an open subset
of R 2 . Thus R 2 / ∼ is a surface, since a restriction of Π to all squares of the form
]𝑢 0 , 𝑢 0 + 1[×]𝑣 0 , 𝑣 0 + 1[ constitutes an atlas of R 2 / ∼.

Figure 5.6

It is not difficult to verify that R 2 / ∼ is diffeomorphic to the torus T 2 . The

mapping Π induces a Riemannian metric on R 2 / ∼, for which Π𝑢 , Π𝑣 constitute an
orthonormal basis of 𝑇[𝑢,𝑣] (R 2 / ∼) (𝐸 ≡ 𝐺 ≡ 1, 𝐹 ≡ 0). This is the only metric such
that Π∶ R 2 → R 2 / ∼ is a local isometry, and therefore R 2 / ∼ is a surface with zero
constant curvature, commonly called a flat torus. In fact Π is an isometric covering.
We suggest as an exercise to show that the equivalence relation

𝑢, ̃
(𝑢, 𝑣) ≃ (̃ 𝑣 ) ⇔ ∃𝑛 ∈ Z ∶ 𝑢̃ = 𝑢 + 𝑛 and ̃
𝑣 = (−1) 𝑣
𝑛

defines a surface of zero curvature diffeomorphic to the Möbius strip.

B. Let us define a complete, simply connected surface of constant curvature equal

to −1. This surface, which we denote by D (hyperbolic plane), is by Theorem 5.3.3
diffeomorphic to the plane, and it suffices therefore to find a Riemannian metric
on R 2 of curvature −1. But this is easy if we use polar coordinates and recall the
formulas given at the end of Section 4.6. Then the coefficients of the parameterization
Φ(𝜌, 𝜑) = (𝜌 cos 𝜑, 𝜌 sin 𝜑) become

𝐸 ≡ 1, 𝐹 ≡ 0, 𝐺 ≡ sinh2 𝜌 (5.27)
5.6 Abstract Surfaces. The Hyperbolic Plane 167

— and we can take the coefficients of this metric in Cartesian coordinates. However,
this is not the best model of D, because the metric defined this way is not conformal.
Let 𝑈 be the open disk with radius 𝜆 (0 < 𝜆 ≤ +∞) centered at the origin, and
𝑓 ∶ [0, 𝜆[ → [0, +∞[ a differentiable, strictly monotone function such that 𝑓 (0) = 0,
and consider the mapping Ψ∶ 𝑈 → R 2 which transforms the polar coordinate point
(𝜌, 𝜑) into the coordinate point ( 𝑓 (𝜌), 𝜑). We want to find 𝑓 so that the Riemannian
metric induced by Ψ on 𝑈, from the metric defined by (5.27) on R 2 , is conformal. If
𝑝 has coordinates (𝜌, 𝜑) then, letting v𝝋 = (cos 𝜑, sin 𝜑) and w𝝋 = (−sin 𝜑, cos 𝜑),
we have
𝑓 (𝜌)
𝐷Ψ 𝑝 (v𝝋 ) = 𝑓 ′ (𝜌)v𝝋 , 𝐷Ψ 𝑝 (w𝝋 ) = w𝝋 (5.28)
𝜌
— by which we obtain

⟨v𝝋 , v𝝋 ) 𝑝 = [ 𝑓 ′ (𝜌)]2 ,
⟨v𝝋 , w𝝋 ) 𝑝 = 0,
(5.29)
𝐺( 𝑓 (𝜌), 𝜑) sinh2 ( 𝑓 (𝜌))
⟨w𝝋 , w𝝋 ) 𝑝 = = ⋅
𝜌2 𝜌2
For the metric (5.29) to be conformal, necessarily

sinh( 𝑓 (𝜌))
𝑓 ′ (𝜌) = ⋅
𝜌

The general solution of this equation is 𝑓 (𝜌) = 2tgh−1 (𝑐𝜌), where 𝑐 is a real constant,
and is defined for 𝜌 ∈ [0, ∣𝑐∣
1
[. We take 𝑐 = 1, so that 𝜆 = 1 and 𝑈 is the unit disk.
In this case 𝑓 ′ (𝜌) = 2
1−𝜌2
, and therefore (5.29) defines in Cartesian coordinates the
metric
4
𝑑𝑠2 = (𝑑𝑢 2 + 𝑑𝑣 2 ). (5.30)
{1 − (𝑢 2 + 𝑣 2 )}2

From now on, let D denote the disk 𝑢 2 + 𝑣 2 < 1 (or the disk ∣𝑧∣ < 1 in the complex
plane C) equipped with the Riemannian metric (5.30). D is usually called Poincaré’s
disk. We next give some of its properties:
(1) D has constant curvature equal to −1. We verify this using formula (4.31) in
Section 4.5. To obtain a surface of constant curvature 𝐾 < 0, we would take, on the
same disk, the metric
4
𝑑𝑠2 = 2
(𝑑𝑢 2 + 𝑑𝑣 2 ).
∣𝐾∣(1 − (𝑢 2 + 𝑣 2 ))

(2) The diameters of D are geodesic. This can be seen by checking that its geodesic
curvature is zero, but we prefer to show that the diameters are minimizing curves.
Indeed, if 𝛼(𝑡) = (𝑢(𝑡), 𝑣(𝑡)) is a curve from 𝛼(0) = (0, 0) to 𝛼(1) = (𝑢 1 , 0) (𝑢 1 > 0)
then
168 5 The Global Geometry of Surfaces
√
1√
2 2
1 2 𝑢 ′ (𝑡) + 𝑣 ′ (𝑡)
𝑙(𝛼) = ∫ 𝐼 𝛼(𝑡) (𝛼′ (𝑡)) 𝑑𝑡 = ∫ 2 2
𝑑𝑡
0 0 1 − [𝑢(𝑡) + 𝑣(𝑡) ] (5.31)
1 2∣𝑢 ′ (𝑡)∣ 1 2𝑢 ′ (𝑡) 1 + 𝑢1
≥∫ 𝑑𝑡 ≥ ∫ 𝑑𝑡 = log ( ),
0 1 − 𝑢(𝑡)
2 0 1 − 𝑢(𝑡)
2 1 − 𝑢1

and equality holds if and only if 𝑣(𝑡) = 0 and 𝑢 ′ (𝑡) ≥ 0 for all 𝑡 ∈ [0, 1] — i.e., if and
only if the trace of 𝛼 is the line segment [0, 𝑢 1 ] × {0}, which is therefore a minimizing
curve. The diameter ] − 1, 1[ × {0} is thus a geodesic of D. Since rotations around
the origin are isometries of D, all other diameters are also geodesics.
It follows from (5.31) that geodesics starting from the origin have infinite length,
and therefore (see Exercise 119) D is complete. Furthermore, the intrinsic distance
between 0 and 𝑧 ∈ D is
1 + ∣𝑧∣
𝑑D (0, 𝑧) = log ( )⋅
1 − ∣𝑧∣

(3) Let 𝑎, 𝑏 ∈ C be such that ∣𝑎∣ > ∣𝑏∣, and consider the mapping

𝑎𝑧 + 𝑏
ℎ 𝑎,𝑏 (𝑧) = ⋅
𝑏𝑧 + 𝑎
The reader may check that ℎ 𝑎,𝑏 sends D bijectively onto itself. More remarkable is
that ℎ 𝑎,𝑏 is an isometry of D. Indeed, for 𝑧 ∈ D and 𝑤 ∈ C, we have (abbreviating
ℎ 𝑎,𝑏 to ℎ)

𝐼 ℎ(𝑧) (𝐷ℎ 𝑧 (𝑤)) = 𝐼 𝑧 (𝑤) ⇔ 𝐼 ℎ(𝑧) (ℎ′ (𝑧)𝑤) = 𝐼 𝑧 (𝑤)

4∣ℎ′ (𝑧)∣2 4
⇔ 2
∣𝑤∣2 = 2
∣𝑤∣2
(1 − ∣ℎ(𝑧)∣2 ) (1 − ∣𝑧∣2 ) (5.32)
1 − ∣ℎ(𝑧)∣2
⇔ ∣ℎ′ (𝑧)∣ = .
1 − ∣𝑧∣2

Checking this last equality is a simple calculation.

Using Proposition 5.7.3 of the next section, we can show that all isometries of D
that preserve orientation are of this form (one obtains the rotations by letting 𝑏 = 0),
𝑎𝑧+𝑏
and those that reverse it are the conjugates of these (𝑧 ↦ , with ∣𝑎∣ > ∣𝑏∣).
𝑏𝑧 + 𝑎

Now given 𝑧0 ∈ D,
𝑧 − 𝑧0
ℎ(𝑧) =
1 − 𝑧0 𝑧
is an isometry that maps 𝑧0 to 0, which allows us to deduce a formula for the distance
between 𝑧0 and another point 𝑧1 ∈ D:
5.6 Abstract Surfaces. The Hyperbolic Plane 169

𝑧1 − 𝑧0
𝑑D (𝑧 0 , 𝑧 1 ) = 𝑑D (ℎ(𝑧 0 ), ℎ(𝑧1 )) = 𝑑D (0, )
1 − 𝑧0 𝑧1
(5.33)
∣1 − 𝑧0 𝑧1 ∣ + ∣𝑧1 − 𝑧0 ∣
= log ( )⋅
∣1 − 𝑧0 𝑧1 ∣ − ∣𝑧1 − 𝑧0 ∣

(4) Mappings of the type

𝑎𝑧 + 𝑏
𝑧↦ ,
𝑐𝑧 + 𝑑
where 𝑎, 𝑏, 𝑐, 𝑑 are complex numbers such that 𝑎𝑑 − 𝑏𝑐 ≠ 0, are called Möbius
transformations, and are bijections of C ∪ {∞} (the Riemann sphere) onto itself. We
can assume that 𝑎𝑑 − 𝑏𝑐 = 1, since if we multiply each of the numbers 𝑎, 𝑏, 𝑐, 𝑑 by
the same nonzero factor the transformation does not change.
Möbius transformations form a group 𝔐 with respect to the composition of
𝑎𝑏
functions; if to each matrix 𝑀 = [ ] with 𝑎𝑑 − 𝑏𝑐 = 1 we attach
𝑐𝑑

𝑎𝑧 + 𝑏
𝑓 𝑀 (𝑧) = ,
𝑐𝑧 + 𝑑
we have 𝑓 𝑀1 𝑀2 = 𝑓 𝑀1 ○ 𝑓 𝑀2 . We hence obtain a group homomorphism Sl(2, C) → 𝔐
(where Sl(2, C) is the multiplicative group of the complex 2 × 2 matrices with
determinant equal to 1).
We are now interested in the geometric properties of these transformations. Writing

𝑎𝑧 + 𝑏 𝑎 𝑏 − 𝑎𝑑 𝑎𝑧 + 𝑏 𝑎 𝑏
= + 𝑐
(if 𝑐 ≠ 0), = 𝑧+ ,
𝑐𝑧 + 𝑑 𝑐 𝑐(𝑧 + 𝑑𝑐 ) 𝑑 𝑑 𝑑

we recognize that any Möbius transformation is written as a composition of:

● translations 𝑧↦𝑧+𝜁 (𝜁 ∈ C)

● rotations 𝑧 ↦ 𝑒𝑖 𝜃 𝑧 (𝜃 ∈ [0, 2𝜋])

● homotheties 𝑧 ↦ 𝜆𝑧 (𝜆 ∈ ]0, +∞[ )

1
● inversions 𝑧↦
𝑧
Transformations of the first three types have known properties: in particular, they
transform straight lines into straight lines and circles into circles. Regarding the
inversions, we have the following:
● if 𝑟 is a line passing through 0, its inverse is still a line. If 𝑟 does not pass through
0, its inverse is a circle passing through 0;
● the inverse of a circle C is a straight line if 0 ∈ C, otherwise it is also a circle.

In fact, the inverse of 𝑢 + 𝑖𝑣 is the point 𝑢̃ + 𝑖̃

𝑣 given by
170 5 The Global Geometry of Surfaces
𝑢 −𝑣
𝑢̃ = , ̃
𝑣= ⋅
𝑢2 + 𝑣2 + 𝑣2
𝑢2
If 𝑟 is the line given by the equation 𝛼𝑢 + 𝛽𝑣 + 𝛾 = 0 then, for 𝑢 + 𝑖𝑣 ∈ 𝑟, we have
𝛼𝑢 + 𝛽𝑣 + 𝛾
𝑢 − 𝛽̃
𝛼̃ 𝑢2 + ̃
𝑣 + 𝛾(̃ 𝑣2) = =0
𝑢2 + 𝑣 2
— which shows that 𝑓 (𝑟) is a straight line if 𝛾 = 0, and is a circle otherwise. The
second statement allows for an analogous verification.
We thus conclude that any Möbius transformation preserves the family of lines
and circles in the plane. Moreover, every 𝑓 ∈ 𝔐 is a conformal mapping and therefore
keeps the orthogonality relation between two curves. Returning to the hyperbolic
plane D, the geodesics passing through a point 𝑧 0 ≠ 0 are, in particular, images of the
geodesics passing through 0 (which are line segments and intersect S1 orthogonally)
by elements of 𝔐 which fix S1 . Hence, every geodesic of D is either a diameter of S1
or an arc of some circumference that intersects S1 orthogonally (see Fig. 5.7).

Figure 5.7

(5) If “lines” mean geodesics, Poincaré’s disc provides a model of the non-
Euclidean geometry of Lobachevski and Bolyai. In this geometry all the axioms of
Euclidean geometry are valid, except the axiom of parallels: in D, through a point
outside a “line” 𝑟, pass infinitely many “lines” that do not intersect 𝑟 — and not only
one, as in the Euclidean case.

Exercises

136. Show that R 2 , with the metric of example 5.6.1.B, has finite area and is not
complete.
In the remaining exercises in this section we work with the hyperbolic plane D.
5.6 Abstract Surfaces. The Hyperbolic Plane 171

137. Consider two geodesics 𝛾1 (𝑠) and 𝛾2 (𝑠) which, at time 𝑠 = 0, start from 𝑧0 ∈ D
in directions that make an angle 𝜃 with each other. Show that

𝛿(𝜃) = lim {𝑑D (𝛾1 (𝑠), 𝛾2 (𝑠)) − 2𝑠}

𝑠→+∞

exists.
138. Check that for all 𝑧0 , 𝑧 1 ∈ D, the set {𝑧 ∈ D∶ 𝑑D (𝑧, 𝑧 0 ) = 𝑑D (𝑧, 𝑧1 )} is a geodesic.
139. (a) Show that the hyperbolic circle S(𝑧 0 ; 𝑟) = {𝑧 ∈ D∶ 𝑑D (𝑧, 𝑧 0 ) = 𝑟} is also a
Euclidean circle.
(b) Check that any equilateral triangle in D (geodesic triangle with all sides equal)
can be inscribed into a hyperbolic circle, but that this is no longer true for every
geodesic triangle.
140. (a) Check that 𝑔(𝑧) = − 2𝑖 + 𝑧−𝑖
1
sends D into the half-plane {𝑧 ∈ C∶ ℑ𝑚(𝑧) > 0}.
(b) Let H be the half-plane {𝑧 ∈ C∶ ℑ𝑚(𝑧) > 0} with the metric given by
2
𝐸 = 𝐺 = 1/ℑ𝑚(𝑧) and 𝐹 ≡ 0. Show that 𝑔 is an isometry of D on H .
(c) Show that the geodesics of H are the vertical semicircles ℜ𝑒(𝑧) = constant
and the half circumferences with center on the axis ℑ𝑚(𝑧) = 0.
(d) Show that the isometries of H that preserve orientation are the functions
𝑎𝑧 + 𝑏
𝑧↦ ,
𝑐𝑧 + 𝑑
where 𝑎, 𝑏, 𝑐, 𝑑 are real numbers such that 𝑎𝑑 − 𝑏𝑐 = 1 (you can use Proposition
5.7.3 of the next section).
(e) Denote by ℑ(D) and ℑ(H ) the groups of the isometries of D and H that
preserve orientation. Note that ℑ(H ) = 𝑔 ○ ℑ(D) ○ 𝑔 −1 , and therefore ℑ(D) and
ℑ(H ) are conjugate subgroups of the group 𝔐 of Möbius transformations.
𝑎𝑏 𝑎𝑧 + 𝑏
141. Check that 𝑀 = [ ] ↦ 𝑓 𝑀 (𝑧) = is a homomorphism of Sl(2, C) on
𝑐𝑑 𝑐𝑧 + 𝑑
𝔐, and that the kernel of this homomorphism is {𝐼, −𝐼}. Conclude that 𝑓 𝑀 = 𝑓 𝑁 if
and only if 𝑀 = ±𝑁.
142. Two elements 𝑓 , 𝑔 ∈ 𝔐 are conjugate if there exists ℎ ∈ 𝔐 such that 𝑓 = ℎ○𝑔○ℎ−1 .
(a) Given 𝑓 ∈ 𝔐, define 𝜏( 𝑓 ) = ∣trace(𝑀)∣ if 𝑓 = 𝑓 𝑀 for 𝑀 ∈ Sl(2, C). Show that
if 𝑓 , 𝑔 ≠ id then 𝑓 and 𝑔 are conjugate if and only if 𝜏( 𝑓 ) = 𝜏(𝑔).
(b) Let Γ = ℑ(D) or Γ = ℑ(H ). Show that if 𝑓 , 𝑔 ∈ Γ are conjugate by an element
of 𝔐, then they are conjugate by some element of Γ.
143. Denote by Γ the group ℑ(D) or ℑ(H ), and let 𝑓 ∈ Γ. Show that:
(a) if√𝜏( 𝑓 ) > 2 then 𝑓 is conjugate to 𝑧 ↦ 𝜆𝑧 in H , where 𝜆 > 0 satisfies
𝜏( 𝑓 ) = 𝜆 + √1 ;
𝜆
(b) if 𝜏( 𝑓 ) = 2 and 𝑓 ≠ id then ℎ is conjugate to 𝑧 ↦ 𝑧 + 1 in H ;
(c) if 𝜏( 𝑓 ) < 2 then 𝑓 is conjugate to 𝑧 ↦ 𝑒 𝑖 𝜃 𝑧 on D, where 2 cos ( 2𝜃 ) = 𝜏( 𝑓 ).
Note: 𝑓 ≠ id is called hyperbolic, parabolic or elliptic according to whether a), b) or
c) of Ex. 143 holds.
172 5 The Global Geometry of Surfaces

144. Let 𝑓 ∈ ℑ(D) ∖ id. Show that:

(a) inf 𝑑D ( 𝑓 (𝑧), 𝑧) > 0 if and only if 𝑓 is hyperbolic.
𝑧∈D
(b) there exists 𝑧0 such that 𝑑D ( 𝑓 (𝑧 0 ), 𝑧 0 ) = inf 𝑑D ( 𝑓 (𝑧), 𝑧) if and only if 𝑓 is
𝑧∈D
elliptic or hyperbolic.

5.7 Complete Surfaces of Constant Curvature

In this section we will study the isometry groups of the complete and simply connected
surfaces of constant curvature, and show that the other complete surfaces of equal
curvature are obtained from these as quotients by a certain subgroup of the isometry
group. This approach allows us to describe all complete surfaces of non-negative
constant curvature.
We begin with a lemma that will be applied repeatedly:

Lemma 5.7.1 Let 𝑓 , 𝑔∶ 𝑆1 → 𝑆2 be local isometries such that for a certain 𝑝 ∈ 𝑆1 we

have 𝑓 (𝑝) = 𝑔(𝑝) and 𝐷 𝑓 𝑝 = 𝐷𝑔 𝑝 . Then 𝑓 = 𝑔.

Proof Given v ∈ 𝑇 𝑝 𝑆1 , let us consider the geodesic 𝛾(𝑡) = exp 𝑝 (𝑡v). Then 𝑓 ○ 𝛾 and
𝑔 ○ 𝛾 are parametrized geodesics with the same initial conditions, since 𝑓 ○ 𝛾(0) =
𝑔 ○ 𝛾(0) = 𝑓 (𝑝) and ( 𝑓 ○ 𝛾)′ (0) = 𝐷 𝑓 𝑝 (v) = 𝐷𝑔 𝑝 (v) = (𝑔 ○ 𝛾)′ (0) — and therefore
𝑓 ○𝛾(𝑡) = 𝑔○𝛾(𝑡) whenever 𝛾(𝑡) is defined. We thus conclude that 𝑓 ○ exp 𝑝 = 𝑔○ exp 𝑝 ,
which implies that 𝑓 and 𝑔 coincide in a neighborhood of 𝑝. This shows that the set
𝑈 = {𝑞 ∈ 𝑆1 ∶ 𝑓 (𝑞) = 𝑔(𝑞), 𝐷 𝑓𝑞 = 𝐷𝑔𝑞 } is non-empty and open. But its definition
ensures that 𝑈 is also closed, and therefore, 𝑆1 being connected, 𝑈 = 𝑆1 . ◻

We denote by 𝑆 𝐾 the complete, simply connected surface of constant curvature 𝐾.

𝑆 𝐾 is thus the sphere with radius √1 (if 𝐾 > 0), the Euclidean plane (if 𝐾 = 0), or
𝐾
the hyperbolic plane of curvature 𝐾 (if 𝐾 < 0). We can then rewrite Theorem 5.3.4 as
follows:

Theorem 5.7.2 Let 𝑆 be a complete surface of constant curvature 𝐾. Then there

exists an isometric covering 𝑓 ∶ 𝑆 𝐾 → 𝑆.

Proof The case 𝐾 ≤ 0 was treated in Theorem 5.3.4. Only the case 𝐾 > 0 is left.
Given 𝑝 ∈ 𝑆 𝐾 , the mapping exp 𝑝 sends the disk 𝐵 𝜋/√𝐾 (𝑝) diffeomorphically into
𝑆 𝐾 ∖{−𝑝}. Thus, if 𝑞 is a point of 𝑆 and 𝐿∶ 𝑇 𝑝 𝑆 𝑘 → 𝑇𝑞 𝑆 a linear isometry, the mapping
𝑔∶ 𝑆 𝐾 ∖{−𝑝} → 𝑆 defined by 𝑔 = exp𝑞 ○ 𝐿 ○ exp−1 𝑝 is a local isometry. Let us now
̃
take 𝑝̃ ∈ 𝑆 𝐾 ∖{𝑝, −𝑝}, and let 𝑞̃ = 𝑔( 𝑝̃), 𝐿 = 𝐷𝑔 ̃𝑝 and ℎ∶ 𝑆 𝐾 ∖{− 𝑝̃} → 𝑆 be the
mapping given by ℎ = exp𝑞̃ ○ ̃ 𝐿 ○ exp−1
𝑝 . The mapping ℎ is also a local isometry, and
̃
furthermore 𝑔( 𝑝̃) = ℎ( 𝑝̃) and 𝐷𝑔 ̃𝑝 = 𝐷ℎ ̃𝑝 . By Lemma 5.7.1, 𝑔 and ℎ coincide on
the intersection of their domains, and we can thus define a local isometry 𝑓 ∶ 𝑆 𝐾 → 𝑆
by 𝑓 (𝑟) = 𝑔(𝑟) if 𝑟 ≠ −𝑝, and 𝑓 (𝑟) = ℎ(𝑟) if 𝑟 ≠ − 𝑝̃. Since 𝑆 𝐾 is compact, by
Proposition 5.2.2 this local isometry is a covering. ◻
5.7 Complete Surfaces of Constant Curvature 173

The following result is a consequence of Lemma 5.7.1 and the proofs of Theo-
rems 5.3.4 and 5.7.2.

Proposition 5.7.3 Given 𝑝, 𝑞 ∈ 𝑆 𝑘 and a linear isometry 𝐿∶ 𝑇 𝑝 𝑆 𝐾 → 𝑇𝑞 𝑆 𝐾 , there

exists one and only one isometry 𝑓 ∶ 𝑆 𝐾 → 𝑆 𝐾 such that 𝑓 (𝑝) = 𝑞 and 𝐷 𝑓 𝑝 = 𝐿.

Since the composite and inverse of isometries are still isometries, the set of
isometries of a given surface forms a group. By the above proposition, the group of
isometries of 𝑆 𝐾 , which we denote by ℑ𝐾 , is exceptionally large.
Let us now assume that 𝑓 ∶ 𝑆 𝐾 → 𝑆 is an isometric covering. Given 𝑞 ∈ 𝑆, let
𝑝 0 and 𝑝 1 be two of the preimages of 𝑞, and 𝐿 𝑖 (𝑖 = 0, 1) the linear isometry
𝐷 𝑓 𝑝𝑖 ∶ 𝑇 𝑝𝑖 𝑆 𝐾 → 𝑇𝑞 𝑆. By Proposition 5.7.3, there exists an isometry 𝑔∶ 𝑆 𝐾 → 𝑆 𝐾 such
that 𝑔(𝑝 0 ) = 𝑝 1 and 𝐷𝑔 𝑝0 = 𝐿 1−1 ○ 𝐿 0 . Now 𝑓 ○ 𝑔∶ 𝑆 𝐾 → 𝑆 is a local isometry such
that

𝑓 ○ 𝑔(𝑝 0 ) = 𝑞 = 𝑓 (𝑝 0 ),
𝐷( 𝑓 ○ 𝑔) 𝑝0 = 𝐷 𝑓 𝑝1 ○ 𝐷𝑔 𝑝0 = 𝐿 1 ○ (𝐿 1−1 ○ 𝐿 0 ) = 𝐷 𝑓 𝑝0

— it follows, by Lemma 5.7.1, that 𝑓 ○ 𝑔 = 𝑔.

It is easily checked that the set ℑ𝐾 ( 𝑓 ) = {𝑔 ∈ ℑ𝐾 ∶ 𝑓 ○ 𝑔 = 𝑓 } is a subgroup of ℑ𝐾 ;
we call it the covering group of 𝑓 .

Proposition 5.7.4 For every 𝑞 ∈ 𝑆, the group ℑ𝐾 ( 𝑓 ) acts transitively on 𝑓 −1 ({𝑞})

— that is, for every pair of points 𝑝 0 , 𝑝 1 ∈ 𝑓 −1 ({𝑞}) there exists one and only one
𝑔 ∈ ℑ𝐾 ( 𝑓 ) such that 𝑔(𝑝 0 ) = 𝑝 1 .
The action of ℑ𝐾 ( 𝑓 ) on 𝑆 𝐾 is discontinuous. This means that every point 𝑝 of 𝑆 𝐾
has an open neighborhood 𝑉 such that 𝑔(𝑉) ∩ 𝑉 is empty for all 𝑔 ∈ ℑ𝐾 ( 𝑓 )∖{id}.

The first statement (apart from the uniqueness of 𝑔, which is easy) has already
been proved. Regarding the second, let us take an evenly covered neighborhood 𝑈
of 𝑓 (𝑝) and let 𝑉 be the component of 𝑓 −1 (𝑈) that contains 𝑝. If 𝑔 ∈ ℑ𝐾 ( 𝑓 )∖{id}
and 𝑞 ∈ 𝑉 then 𝑞 ≠ 𝑔(𝑞), since the identity is the only element of ℑ𝐾 ( 𝑓 ) with some
fixed point. Since 𝑓 (𝑞) = 𝑓 (𝑔(𝑞)), necessarily 𝑔(𝑞) ∉ 𝑉. Thus 𝑔(𝑉) ∩ 𝑉 = ∅, which
proves the statement.
It deserves mention that what we call discontinuous action is called proper
discontinuous action by most authors, who reserve the former name for a weaker
condition, of which we will make no use. We further say that a subgroup Γ of ℑ𝐾 is
discrete if its action on 𝑆 𝐾 is discontinuous (this terminology is also not the usual
one).
The covering group completely determines the surface in the following sense: if
we have two isometric coverings 𝑓𝑖 ∶ 𝑆 𝐾 → 𝑆𝑖 (𝑖 = 1, 2) such that ℑ𝐾 ( 𝑓1 ) = ℑ𝐾 ( 𝑓2 )
then 𝑆1 and 𝑆2 are isometric. We can further ask which subgroups of ℑ𝐾 are covering
groups. The answer is simple:

Proposition 5.7.5 A subgroup Γ of ℑ𝐾 is a covering group if and only if it is discrete.

174 5 The Global Geometry of Surfaces

Proof It remains to prove that such a subgroup is a covering group. We note

that the action of Γ on 𝑆 𝐾 induces an equivalence relation whose classes are
[𝑝] = {𝑔(𝑝)∶ 𝑔 ∈ Γ}. Let 𝑆 𝐾 /Γ be the set of equivalence classes and let Π∶ 𝑆 𝐾 → 𝑆 𝐾 /Γ
be the quotient mapping. Given 𝑝 ∈ 𝑆, let (𝑈, Φ) be a parameterization in a
neighborhood of 𝑝 such that 𝑔(Φ(𝑢)) ∩ Φ(𝑈) = ∅ for all 𝑔 ∈ Γ ∖ {id}, and let us
put Ψ = Π ○ Φ: the set of mappings Ψ so defined constitutes an atlas of 𝑆 𝐾 /Γ, which
is therefore a surface; moreover, Ψ(𝑈) is an evenly covered neighborhood of [𝑝],
which shows that Π is a covering. With the Riemannian metric induced by Π on
𝑆 𝐾 /Γ the mapping Π is an isometric covering whose covering group is obviously Γ.
◻

Consider, for example, the case of the sphere 𝑆 𝐾 given by the equation 𝑥 2 + 𝑦 2 +𝑧2 =
1
𝐾 (𝐾 > 0). The isometries of 𝑆 𝐾 are the restrictions of the linear isometries of
R , and correspond to the orthogonal 3 × 3 matrices over R — the group of which
3

is denoted by 𝑂(3, R ). An example of a discrete subgroup of 𝑂(3, R ) is {−𝐼, 𝐼},

where 𝐼 is the identity matrix. The surface 𝑆 𝐾 /{−𝐼, 𝐼} is called the projective plane,
and is the surface obtained by identifying in 𝑆 𝐾 the pairs of diametrically opposed
points. The projective plane is non-orientable, and in fact contains a Möbius strip
(see Fig. 5.8)

Figure 5.8
More generally, we have the following result (the proof of which is Exercise 145
of this section):

Proposition 5.7.6 Let Γ be a discrete subgroup of ℑ𝐾 . Then 𝑆 𝐾 /Γ is orientable if

and only if every isometry 𝑔 ∈ Γ preserves the orientation of 𝑆 𝐾 .

We can now describe all complete surfaces of constant positive curvature.

Theorem 5.7.7 The only complete surfaces of constant curvature 𝐾 > 0 are, up to
isometry, the sphere and the projective plane.
5.7 Complete Surfaces of Constant Curvature 175

Proof In view of Proposition 5.7.5, it suffices to show that the only discrete subgroups
of ℑ𝐾 ≃ 𝑂(3, R ) are {𝐼, −𝐼} and the trivial group {𝐼}. Let Γ be such a subgroup:
given 𝐴 ∈ Γ, the isometry 𝐴 has some real eigenvalue 𝜆, which is necessarily 1 or −1.
If 𝜆 = 1, then 𝐴 has some fixed point in 𝑆 𝐾 , which implies that 𝐴 = 𝐼. If 𝜆 = −1 then
𝐴2 has the eigenvalue 𝜆2 = 1 and therefore 𝐴2 = 𝐼. The eigenvalues of 𝐴 are thus all
equal to 1 or −1; but none of them can be 1, and therefore 𝐴 = −𝐼. We thus have
Γ = {𝐼} or Γ = {𝐼, −𝐼}. ◻

In example 5.6.2.A we described the torus as the quotient of R 2 by the group Γ of

the translations 𝑇(v) associated with vectors v of integer coordinates. Γ is generated
by the two independent translations 𝑇(1, 0) and 𝑇(0, 1); as we will see below, this
is a typical situation. The square ]0, 1[ × ]0, 1[ is a fundamental region for Γ. In
general, we say that an open subset 𝑈 of 𝑆 𝐾 is a fundamental region for the action of
a discrete subgroup Γ of ℑ𝐾 if
● 𝑔(𝑈) ∩ ℎ(𝑈) = ∅ for all 𝑔 ≠ ℎ in Γ;
● 𝑆 𝐾 is the union of the closures of 𝑔(𝑈) for 𝑔 ∈ Γ.
The most interesting fundamental regions are polygons, where the sides are geodesic
segments. It is possible to reconstruct the surface 𝑆 𝐾 /Γ if one knows how to identify
the sides of the polygon. In Fig. 5.9 we illustrate a fundamental region P for a certain
discrete subset Γ of ℑ𝐾 (𝐾 < 0): it is a regular polygon of eight sides whose sum
of interior angles equals 2𝜋; the pairs of sides to be identified are designated by the
same letter.

c b

d a

c b

d a

Figure 5.9

The surface 𝑆 𝐾 /Γ is, in this case, the double torus. We would obtain the 𝑛-torus,
𝑛 ≥ 2, from a regular polygon P𝑛 of 4𝑛 sides and sum of angles equal to 2𝜋 (𝑛 = 1
is impossible because, by the Gauss-Bonnet theorem, the sum of the angles of any
four-sided polygon in the hyperbolic plane is < 2𝜋). A theorem of Poincaré ensures
176 5 The Global Geometry of Surfaces

that the elements of Γ that identify the sides of P𝑛 generate Γ. The interested reader
can find details in [10], [2] and [3].
To conclude this section and the book, we will determine all discrete subgroups of
the group ℑ0 of the isometries of R 2 — which is equivalent, by Proposition 5.7.5, to
determining all complete surfaces of constant zero curvature.
Lemma 5.7.8 The isometries of R 2 without fixed points are translations and glide
reflections (reflection in an axis 𝑟 followed by a translation in the direction of 𝑟).
Translations preserve orientation, while glide reflections reverse it.
Proof Every isometry of R 2 is of the form 𝑓 (𝑝) = 𝐿(𝑝) + v, where 𝐿 is a linear
isometry and v a vector. If 𝐼 − 𝐿 were an isomorphism then 𝑓 would have some fixed
point, and so for 𝑓 to have no fixed points, 𝐿 must have eigenvalue 1. Let (e1 , e2 ) be
an orthonormal basis of R 2 such that 𝐿(e1 ) = e1 . Since 𝐿 is an isometry, necessarily
𝐿(e2 ) = ±e2 . If 𝐿(e2 ) = e2 then 𝐿 is the identity and 𝑓 a translation. If 𝐿(e2 ) = −e2 ,
then 𝐿 is the reflection in the straight line generated by e1 . Letting v = 𝛼e1 + 𝛽e2 , we
recognize that 𝑓 is the composite of the reflection in the line 𝜆 ↦ 𝜆e1 + 𝛽2 e2 with the
translation associated with the vector 𝛼e1 (in particular, 𝑓 has no fixed points if and
only if 𝛼 ≠ 0).
Regarding the orientation, just note that it is preserved by 𝑓 if and only if it is by
𝐿. ◻

Let Γ ≠ {id} be a discrete subgroup of ℑ0 . Assume first that R 2 /Γ is orientable:

by Proposition 5.7.6, this means that all elements of Γ preserve orientation — that
is, that they are all translations. To simplify notation, we identify the translation
associated with v with the vector v itself.
Since Γ is discrete, we can choose v1 ≠ 0 in Γ such that ∣v1 ∣ is as small as possible.
Then any other v ∈ Γ that is collinear with v1 is an integer multiple of v1 (since, if
v = 𝜆v1 , then, denoting by ⌊𝜆⌋ the integer part of 𝜆, {𝜆 − ⌊𝜆⌋ is an element of Γ with
norm less than that of v1 , and is therefore zero). If Γ contains only such vectors, it is
therefore of the form
Γ = {𝑛v1 ∶ 𝑛 ∈ Z } (5.34)
— in which case R 2 /Γ is a cylinder.
If Γ contains elements that are non-collinear with v1 , then let v2 be one such vector
with minimal norm. Then

Γ = {𝑛v1 + 𝑚v2 ∶ 𝑛, 𝑚 ∈ Z }. (5.35)

Indeed, given v ∈ Γ, it belongs to some parallelogram P = {𝛼v1 + 𝛽v2 ∶ 𝑛1 ≤ 𝛼 ≤

𝑛1 + 1, 𝑚 1 ≤ 𝛽 ≤ 𝑚 1 + 1}. Denoting by w the vertex of P closest to v and by 𝑑 the
length of the longest diagonal of P, we have
𝑑 1
∣v − w∣ ≤ < {∣v1 ∣ + ∣v2 ∣} ≤ ∣v2 ∣
2 2
— which implies, by the choice of v2 , that v − w is collinear with v1 , and therefore an
integer multiple of v1 . We have therefore proved equality (5.35). If Γ is of this form,
5.7 Complete Surfaces of Constant Curvature 177

R 2 /Γ is a torus — and a fundamental region for Γ is the parallelogram of sides v1

and v2 .
Then 𝑓 ○ 𝑔(𝑝) = 𝐿 1 ○ 𝐿 2 (𝑝) + w, where w = u + 𝐿 1 (v). Now 𝑓 ○ 𝑔 preserves
orientation and, by Lemma 5.7.8, it must be a translation. Therefore 𝐿 1 ○ 𝐿 2 is the
identity, hence 𝐿 2 = 𝐿 1−1 = 𝐿 1 . In conclusion:
there is a linear isometry 𝐿 and a set Γ𝑅 ⊆ R 2 such that {𝑝 ↦ 𝐿(𝑝) + v∶ v ∈ Γ𝑅 }
is the set of the elements of Γ that reverse orientation.
We further denote by Γ𝑇 the set of vectors associated with the translations of Γ.
Γ𝑇 is a discrete subgroup and, from the discussion above, the only two possibilities
are that Γ𝑇 is either one-dimensional (of the form {𝑛v1 ∶ 𝑛 ∈ Z }) or two-dimensional
(of the form {𝑛v1 + 𝑚v2 ∶ 𝑛, 𝑚 ∈ Z }).

Lemma 5.7.9 If u, v ∈ Γ𝑅 and w ∈ Γ𝑇 , then:

(i) u + 𝐿(v) ∈ Γ𝑇 ;
(ii) −𝐿(v) ∈ Γ𝑅 ;
(iii) u − v ∈ Γ𝑇 ;
(iv) v + w ∈ Γ𝑅 ;
(v) 𝐿(w) ∈ Γ𝑇 .

Proof The composite of 𝑓 (𝑝) = 𝐿(𝑝) + u with 𝑔(𝑝) = 𝐿(𝑝) + v is given by

𝑓 ○ 𝑔(𝑝) = 𝑝 + {u + 𝐿(v)}, which proves (i). Regarding (ii), we observe that the
inverse of 𝑝 ↦ 𝐿(𝑝) + v is 𝑝 ↦ 𝐿(𝑝) − 𝐿(v). For (iii), we write u − v = u + 𝐿(−𝐿(v))
and apply (i) and (ii). The composite of ℎ(𝑝) = 𝑝 + w with 𝑔(𝑝) = 𝐿(𝑝) + v is
ℎ○𝑔(𝑝) = 𝐿(𝑝)+(v+w), which proves (iv). Finally, we have 𝐿(w) = −𝐿(v)+𝐿(v+w)
— and since, by (ii) and (iv), we have −𝐿(v), v+w ∈ Γ𝑅 , it follows by (i) that 𝐿(w) ∈ Γ𝑇 ,
which proves (v). ◻

Statement (v) says that 𝐿(Γ𝑇 ) ⊆ Γ𝑇 . Applying 𝐿 to both sides, we obtain the
opposite inclusion Γ𝑇 ⊆ 𝐿(Γ𝑇 ); we thus conclude that Γ𝑇 is invariant under 𝐿.
Moreover, (iii) and (iv) say that Γ𝑅 is the result of a translation on Γ𝑇 : for all v ∈ Γ𝑅
we have Γ𝑅 = {v} + Γ𝑇 . This means that to obtain all the glide reflections of Γ, we
just need to compose one of these transformations with each of the translations of Γ.
Let (e1 , v2 ) be an orthonormal basis of R 2 such that 𝐿(e1 ) = e1 , 𝐿(e2 ) = −e2 . An
important observation, contained in the proof of Lemma 5.7.8, is that if 𝛼e1 + 𝛽e2 ∈ Γ𝑅
then 𝛼 ≠ 0. With these remarks in mind, let us now look at the forms that the group Γ
can take.
If Γ𝑇 is of the form {𝑛v1 ∶ 𝑛 ∈ Z }, then v1 is an eigenvector of 𝐿, which is collinear
with e1 or with e2 . Take v = 𝛼e1 + 𝛽e2 ∈ Γ𝑅 : then 𝛼 ≠ 0 and, by (i), 2𝛼e1 = v + 𝐿(v)
belongs to Γ𝑇 . We thus have v1 = 𝜆e1 , where we suppose 𝜆 > 0. Adding to v, if
necessary, an integer multiple of v1 , we can assume that 0 < 𝛼 < 𝜆. Since 2𝛼e1 is an
integer multiple of v1 , it follows that 𝜆 = 2𝛼 and v1 = v + 𝐿(v), which means that the
translation associated with v1 is the composite of 𝑓 (𝑝) = 𝐿(𝑝) + v with itself. Since
Γ𝑅 = {v} + Γ𝑇 , we conclude that Γ is the cyclic group generated by 𝑓 (𝑝). Therefore:
If Γ contains elements that reverse orientation and Γ𝑇 is one-dimensional, then Γ
is generated by a single glide reflection. In this case R 2 /Γ is a Möbius strip.
178 5 The Global Geometry of Surfaces

The case where Γ𝑇 is two-dimensional is left. Since 𝐿(Γ𝑇 ) = Γ𝑇 , there are two
possibilities:
(a) Γ𝑇 = {𝑛v1 + 𝑚𝐿(v1 )∶ 𝑛, 𝑚 ∈ Z };
(b) Γ𝑇 = {𝑛𝜆 1 e1 + 𝑚𝜆2 e2 ∶ 𝑛, 𝑚 ∈ Z }.
Suppose (a). Writing v1 = 𝜆e1 + 𝜂e2 (𝜆 > 0, 𝜂 ≠ 0), we have Γ𝑇 ∩⟨e1 ⟩ = {2𝑛𝜆e1 ∶ 𝑛 ∈
Z }. Take v = 𝛼e1 + 𝛽e2 ∈ Γ𝑅 : we may suppose, by adding to v a multiple of v1 , that
0 < 𝛼 < 𝜆. But 2𝛼e1 = v + 𝐿(v) is in Γ𝑇 ∩ ⟨e1 ⟩, implying 𝛼 = 𝑛𝜆, for some 𝑛 ∈ Z ,
which is absurd.
Thus (b) must be true. Repeating the argument above, we see that there exists
v = 𝛼e1 + 𝛽e2 ∈ Γ𝑅 such that 2𝛼 = 𝜆1 , and therefore v+𝐿(v) = 𝜆1 e1 . As Γ𝑅 = {v}+Γ𝑇 ,
the group Γ is generated by 𝑓 (𝑝) = 𝐿(𝑝) + v and the translation associated with the
vector 𝜆2 e2 . To summarize:

f(p)

p+ l2e2

p p+ ae1

Figure 5.10

If Γ contains elements that reverse orientation and Γ𝑇 is two-dimensional, then Γ

is generated by a glide reflection, with axis on a line 𝑟 say, and by a translation in the
direction perpendicular to 𝑟. In this case R 2 /Γ is a Klein bottle.
In Fig. 5.10 we show several copies of a fundamental region (a rectangle) for this
group, marking the identifications to be made.
Let us summarize our findings:

Theorem 5.7.10 Every complete surface of constant zero curvature is diffeomorphic

to one of the following surfaces:
● the plane, the cylinder, or the torus, if it is orientable;
● the Möbius strip or the Klein bottle, if non-orientable.
5.7 Complete Surfaces of Constant Curvature 179

Exercises
145. Prove Proposition 5.7.6. Hint: if 𝑆 𝐾 /Γ is orientable and an orientation is fixed
on it then the quotient mapping 𝑆 𝑘 → 𝑆 𝐾 /Γ induces an orientation on 𝑆 𝐾 .
146. (a) Let Γ1 and Γ2 be discrete subgroups of ℑ𝐾 . Show that 𝑆 𝐾 /Γ1 and 𝑆 𝐾 /Γ2 are
isometric if and only if Γ1 and Γ2 are conjugate subgroups.
(b) Give an example of two tori of equal area and constant zero curvature that are
not isometric.
147. Show that if Γ is a discrete subgroup of ℑ(D), then Γ ∖{id} contains only
hyperbolic isometries (see, in Section 5.6, exercises 140, 143 and 144).
148. (a) Consider on the torus of revolution T 2 ⊆ R 3 , given by Φ(𝑢, 𝑣) = ((2 +
cos 𝑣) cos 𝑢, (2+cos 𝑣) sin 𝑢, sin 𝑣), the equivalence relation ∼ that identifies (𝑥, 𝑦, 𝑧)
and (−𝑥, −𝑦, −𝑧). Show that T 2 / ∼ is a Klein bottle.
(b) If ≃ is the equivalence relation that identifies (𝑥, 𝑦, 𝑧) and (−𝑥, −𝑦, 𝑧), what
surface is T 2 / ≃?
References

1. E. Barbier: Note sur le problème de l’aiguille et le jeu du joint couvert. J. Math. Pures Appl.
(2) 5 (1860).
2. A.F. Beardon: The Geometry of Discrete Groups. Springer-Verlag 1983.
3. A.F. Beardon: An introduction to Hyperbolic Geometry [in Bedford, Keane, Series (eds.):
Ergodic Theory, Symbolic Dynamics, and Hyperbolic Spaces]. Oxford U.P. 1991.
4. A.L. Besse: Manifolds all of whose Geodesics are Closed. Springer-Verlag 1978.
5. J.H. Cadwell: Topics in Recreational Mathematics. Cambridge University Press 1966.
6. M.P. do Carmo: Differential Geometry of Curves and Surfaces. Prentice-Hall 1976.
7. C.D. Charkerian, H. Groemer: Convex bodies of constant width [in P.M. Gruber, J.M. Wills
(eds.): Convexity and its applications]. Birkhäuser 1983.
8. S.S. Chern: Curves and Surfaces in Euclidean Space [in S.S. Chern (ed.): Global Differential
Geometry]. Math. Association of America 1988.
9. H.G. Eggleston: Convexity. Cambridge University Press 1958.
10. R. Fenn: What is the Geometry of a Surface? Am. Math. Monthly 90 (1983)
11. D.G. Figueiredo: Análise de Fourier e equações diferenciais parciais. Projeto Euclides, IMPA
1977.
12. A. Goetz: Introduction to Differential Geometry. Addison-Wesley 1970.
13. W. Klingenberg: A course in Differential Geometry. Springer-Verlag 1978.
14. E. Kreyszig: Introduction to Differential Geometry and Riemannian Geometry. University of
Toronto Press 1968.
15. E.L. Lima: Duas novas demonstrações do teorema de Jordan-Brouwer no caso diferenciável.
Matemática Universitária 4 (1986).
16. E.L. Lima: Curso de Análise, vol. 2 (3a edição). Projeto Euclides, IMPA 1989.
17. E.E. Moise: Geometric Topology in dimensions 2 and 3. Springer-Verlag 1977.
18. I. Niven: Maxima and minima without Calculus. Math. Association of America 1981.
19. R. Osserman: A survey of Minimal Surfaces. Dover Publications 1986.
20. R. Osserman: Curvature in the Eighties. Am. Math. Monthly 97 (1990).
21. R. Osserman: The four-or-more vertex theorem. Am. Math. Monthly 92 (1985).
22. L.A. Santaló: Integral Geometry [in S.S. Chern (ed.): Global Differential Geometry]. Math.
Association of America 1988.
23. J. Sotomayor: Lições de equações diferenciais ordinárias. Projeto Euclides, IMPA 1979.
24. M. Spivak: Calculus. Benjamin 1967.
25. M. Spivak: A comprehensive introduction to Differential Geometry, vol. I-IV. Publish or Perish
1970-75.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 181
P. V. Araújo, Differential Geometry, https://doi.org/10.1007/978-3-031-62384-4
Index

Discontinuous group Action, 174 Curve

Angle, 1, 53 convex, 15 – 18
oriented, 10, 53 coordinate, 32, 57, 81
Area, 54, 157 strictly convex, 18
of an oval surface, 158 closed, 14
of surfaces parallel to an oval, 160 of constant width, 19 – 24
between parallel curves, 162 integral of a vector field, 76
Atlas, 164 – 165 parallel, 24, 162
oriented, 49, 49 regular, 2
regular in parts, 23
Klein bottle, 179 simple, 14
Curvature
Euler’s Characteristics, 116 of a curve, 5, 8
Catenary, 90 of Gauss, 62, 69, 93, 112, 127,131
Catenoid, 90, 108 geodesic, 98, 99, 110, 131
Center of curvature, 6 mean, 62, 70, 101, 156
Change of coordinates, 38, 166 total mean, 155, 160, 161, 163
Clairaut equation, 129, 143 normal, 60, 69
Clover, 2 principal, 60, 70
Cylinder, 89, 102, 143, 151, 177, 179 with sign, 10, 10
Geodesic circumference, 123, 131, 170 total, 116 – 118, 119
Connected component, 147
Cone, 39, 89, 100, 141, 143 Derivative
Contact of a differentiable mapping, 44, 44
of order 𝑛 between planar curves, 12 of the Gauss mapping, 59, 69
of order ≥ 2 between surfaces, 73, 75 covariant, 95 – 96, 97
Coordinates Diffeomorphism, 42, 45
spherical, 33 local, 46
isothermal, 88 orientation-preserving, 50, 52
local (or parametrization), 32, 41 area-preserving, 57, 94
orthogonal, 54, 82, 112 Continuous Deformation of a curve, 146
polar, 35 Direction
polar geodesic, 123, 127, 130 asymptotic, 63, 67
semi-geodesic, 131 principal, 60, 70
Covering, 144, 145, 148 Poincaré Disk, 167
isometric, 151, 172 geodesic circumferences, 170
number of sheets, 146 distance between two points, 169

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024 183
P. V. Araújo, Differential Geometry, https://doi.org/10.1007/978-3-031-62384-4
184 Index

geodesics, 170 Length, 3, 53

and non-Euclidean geometry, 170 of an arc, 4
isometries, 168, 171, 172 of a curve of constant width, 22, 24
Intrinsic Distance, 126, 131 of a parallel curve, 162
Divergence, 96, 100, 102, 103 Lifting, 132 – 133, 146
Lifting curves, 146, 146
Ellipsoid, 64 – 65 Lines
Symmetric Endomorphism (or self-adjoint), 60 asymptotic, 74, 75, 81, 82
Evolute, 11 of curvature, 74, 79, 82
integrals, 80
Direction Field, 80, 81
Gradient Field, 107 – 108 Mapping
Field of vectors conformal, 86, 94, 131
along a curve, 97 differentiable, 42
divergence, 96, 100, 102, 103 exponential, 122, 150, 151
of integral curves or paths, 76 linear symmetric (or self-adjoint), 60
of normal vectors, 47 Gauss normal, 59, 148
of parallel vectors, 99 – 100 quotient, 166
of tangent vectors on a surface, 79 Riemannian Metric, 164, 166
First fundamental form, 52 conformal, 164
Formula Möbius strip, 47, 52, 178, 179
Blaschke’s, 161
Cauchy’s, 158 Chebyshef Net, 57
Euler’s, 115 Evenly covered Neighborhood, 144
Girard’s, 55, 115 – 116 Normal Neighborhood, 123
Formulas Principal Norm, 5
Frenet’s, 7, 10
Minkowski integrals, 154 Open subset of a surface, 32
connected, 40
Geodesics, 100 evenly covered, 144
surfaces of revolution, 128 simply connected, 148
differential equations, 120, 128, 151 Orientation, 47, 49, 175
closed, 118, 122 Oval, 148, 152
minimizers, 126, 140
parametrized, 120 Osculating Paraboloid, 74
Gradient, 107 Parametrization: see Coordinates
Group Plane, 32, 179
acting discontinuously, 174 geodesics, 101
discrete, 174 isometries, 88, 90
isometries of a surface, 173 Hyperbolic Plane: see Poincaré disk
covering, 173 Osculating Plane, 6
of Möbius transformations, 169 Projective Plane, 175
Tangent Plane, 44
Helix, 2, 102 Point
Helicoid, 35, 90, 107 antipodal, 20, 55, 57, 76, 160
Homotopy with fixed ends, 146 elliptical, 62, 71
hyperbolic, 62, 71
Index of Rotation, 14, 18, 110, 134, 136 parabolic, 62, 67, 83
Isometry, 88, 131, 151, 173 planar, 62, 67
elliptical, 172 umbilical, 62, 65
hyperbolic, 172, 180 Projection
local, 88, 93, 128, 151, 172 of Archimedes, 57
parabolic, 172 of Mercator, 90
Isoperimetric inequality, 28 stereographic, 36, 87
Index 185

gnomonic, 57 Theorem
Pseudo-sphere, 36, 75 of Barbier, 22
of Brouwer’s fixed point, 136
Radius of curvature, 6 of divergence, 102, 103, 161 – 161
Glide reflection, 176 of Gauss, 93
Fundamental region, 175 of Gauss-Bonnet, 109, 116
Polygonal region, 102 of Hadamard, 148, 151
of the four vertices, 26
Normal Section, 61 of the rigidity of the sphere, 156
Second fundamental form, 67, 74
of the rotation index, 134, 136
Similarity, 86, 87, 131
Torsion, 7, 8, 14
Poincaré Semi-plane, 171
Torus, 36, 43, 82, 94, 143, 144, 166, 177, 179
Christoffel Symbols, 91
Double Torus, 117, 175
Conjugate Subgroups, 171, 180
Trace of a curve, 1
Sphere, 33, 63, 70
Tractrix, 36
geodesics, 122
isometries, 174 – 175 Trajectory of a vector field, 76
rigidity theorem, 156 Möbius Transformation, 169
parallel transport, 100 Parallel Transport, 99, 100, 112, 119
Tangent Space, 44, 165 – 166 Triangulation, 113, 131
Surfaces Geodesic triangle, 115, 130, 143, 171
abstract, 164 Reuleaux triangle, 23
complete, 139, 140, 142 Frenet’s triangle, 6
connected, 40
convex, 152 Regular Value, 40
of constant curvature, 70, 75, 123, 128, 130, Variation
131, 156, 172, 175, 179 of area, 106
of non-positive curvature, 150, 151 normal, 103
inextensible, 141 of a surface, 103, 159
of constant width, 75 – 76, 159 160 – 161 Velocity, 2
minimal, 107, 108 scalar, 3
level, 40, 42 Volume
orientable, 46 – 47, 49, 75, 175 between parallel surfaces, 160
oval, 148, 152
bounded by a surface of constant width, 161
parallel, 160
parameterized, 34
ruled, 75, 102 Width, 19
regular, 31, 40 Constant Width
of revolution, 34, 70, 75, 102, 128 of a curve, 19 – 24
simply connected, 148, 151, 151 of a surface, 71 – 76, 159, 160 – 161
Spindle, 55 Wirtinger’s lemma, 27