You are on page 1of 110

MATH329

Geometry of Curves and Surfaces


c Alexander C. R. Belton 2015
Copyright

Hyperlinked edition

All rights reserved

The right of Alexander Belton to be identified as the author


of this work has been asserted by him in accordance with
the Copyright, Designs and Patents Act 1988.
Contents

Contents i

Introduction iii
Preface to the expanded version . . . . . . . . . . . . . . . . . . . . . . iii
Preface to the original version . . . . . . . . . . . . . . . . . . . . . . . iii
Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

1 Curves 1
1.1 Review: vectors in R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
The scalar product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Orthonormal bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
The vector product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Lines and planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Spheres and circles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Parameterised curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Rectifiable curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Some calculus for vector-valued functions . . . . . . . . . . . . . . . . . 11
Best linear approximation . . . . . . . . . . . . . . . . . . . . . . . . . 11
Space-filling curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3 Unit-speed curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
The tangent vector field . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Smooth curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
The normal vector field . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Best circular approximation . . . . . . . . . . . . . . . . . . . . . . . . 18
The binormal vector field . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2 Surfaces – local theory 21


Partial derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.1 The first fundamental form . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Calculating area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Surface reparameterisations . . . . . . . . . . . . . . . . . . . . . . . . 32
Orthogonal reparameterisation . . . . . . . . . . . . . . . . . . . . . . . 35

i
Contents

Isometric surface patches . . . . . . . . . . . . . . . . . . . . . . . . . . 35


Isometries and reparameterisations . . . . . . . . . . . . . . . . . . . . 37
2.2 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Normal and geodesic curvature . . . . . . . . . . . . . . . . . . . . . . . 38
Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
The geodesic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Geodesics and surfaces of revolution . . . . . . . . . . . . . . . . . . . . 44
Meridians and parallels . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Geodesics and isometries . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Curves of extremal length . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3 The second fundamental form . . . . . . . . . . . . . . . . . . . . . . . . 49
Principal curvatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.4 Gaussian curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
The Gauss map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Theorema Egregium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3 Surfaces – global theory 63


3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.2 Plane curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.3 Simple closed curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Simple closed curves in surfaces . . . . . . . . . . . . . . . . . . . . . . 72
Curvilinear polygons . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.4 Global surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Charts and atlases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Subdivisions and the Euler characteristic . . . . . . . . . . . . . . . . . 82
The global Gauss–Bonnet theorem . . . . . . . . . . . . . . . . . . . . . 83

A Auxiliary results 85
A.1 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
A.2 Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

B Solutions to selected exercises 89

Bibliography 99

Index 101

ii
Introduction

Preface to the expanded version


This version of these notes contains a new chapter, on the global theory of surfaces as
typified by three variations on the Gauss–Bonnet theorem. Some minor amendments
have been made to the previous text; for example, the generating curve for a surface
of revolution is taken to lie in the x-z plane, so as to be consistent with the usual
longitude-latitude parameterisation of the sphere.

Alexander C. R. Belton
Lancaster, 6th January 2015

Preface to the original version


These notes are intended as a gentle introduction to the differential geometry of curves
and surfaces. Much effort has been expended to keep technicalities to a minimum, but
certain prerequisites are unavoidable; some of the necessary analytical and algebraic
results are collecting in two appendices. The focus is on local properties and we work
in R3 throughout. Unusually for a text on geometry, there are no figures: the reader is
invited to provide their own in the many gaps left for this purpose.
The book closest to the presentation given here is the excellent text of Andrew Pressley
[6], which contains far more material and goes into much greater detail. (Numbers
in square brackets refer to items in the bibliography.) It also differs in some respects
from the conventions and notation adopted here, so the reader should take care when
switching from one to the other. The Oxford University lecture notes of Graeme Segal
[8] were invaluable for the production of the second chapter of these notes, on surfaces.
John Roe’s book [7] is a pleasant exposition of geometry with a different emphasis (and
some overlap) with ours; a venerable but still excellent treatment of differential geometry
(both local and global) is [12].
The author happily acknowledges his debt to all those who tried to teach him differential
geometry, particularly Professors R. L. Hudson and N. J.Hitchen, and Dr P. J. Braam.
This document was typeset using LATEX 2ε with Peter Wilson’s memoir class and the
AMS-LATEX packages. The index was produced with the aid of the MakeIndex program.

iii
Introduction

Conventions
The notation “P := Q” means that the quantity P is defined to equal Q. For example,
if x ∈ R then (
x if x > 0,
|x| :=
−x if x < 0.
If f : A → B is a function and b ∈ B then f ≡ b means that f is the constant function
with value b, i.e., f (a) = b for all a ∈ A.

iv
One Curves

1.1 Review: vectors in R3


Recall that R3 is the three-dimensional vector space with scalar field R: elements of R3
are triples of real numbers, often written as column vectors, with vector-space operations
(addition and scalar multiplication) defined “coordinate-wise”:
         
v1 w1 v1 + w1 v1 λv1
 v2  +  w2  =  v2 + w2  and λ  v2  =  λv2 
v3 w3 v3 + w3 v3 λv3

for all v1 , v2 , v3 , w1 , w2 , w3 , λ ∈ R. For convenience, we will also denote vectors in R3


as row vectors,
v = (v1 , v2 , v3 ), w = (w1 , w2 , w3 )
and so on, and allow scalars to act on the right as well as the left:

vλ = (λv1 , λv2 , λv3 ) = λu.

The zero vector 0 = (0, 0, 0).


We regard R3 as a mathematical model for the space of everyday experience.
The standard basis of R3 consists of the unit vectors
     
1 0 0
 
i := 0 , j := 1   and k := 0 

0 0 1

and every vector v in R3 has a unique representation of the form

v = v1 i + v2 j + v3 k,

where v1 , v2 , v3 ∈ R.

1
1. Curves

The scalar product


Recall that R3 has a natural inner product, known as the scalar product or dot product:
algebraically, this is
v · w = v1 w1 + v2 w2 + v3 w3 ;
geometrically, we have
v · w = |v| |w| cos θ,
where q
|v| = v12 + v22 + v32
is the magnitude of the vector v (and similarly for |w|) and θ is the angle between
the vectors v and w. (The sense in which this angle is measured does not matter,
since cosine is an even function.) Two vectors are orthogonal (or perpendicular ) if their
scalar product is zero (so the angle between them equals π/2). The magnitude |v − w|
corresponds to the distance between v and w.

It is simple to check that the dot product is commutative (or symmetric),

v·w =w·v for all v, w ∈ R3 ,

and bilinear (linear in each argument):

(u + λv) · w = (u · w) + λ(v · w) and u · (v + λw) = (u · v) + λ(u · w)

for all u, v, w ∈ R3 and λ ∈ R.


Furthermore,
v · v = |v|2 for all v ∈ R3
and the Cauchy–Schwarz inequality holds:

|v · w| 6 |v| |w| for all v, w ∈ R3 ,

with equality if and only if the vectors v and w are linearly dependent. Recall that

|v + w| 6 |v| + |w| and |v| − |w| 6 |v − w| for all v, w ∈ R3 . (1.1)

Exercise 1.1. Prove the second of the two inequalities (1.1). [Hint: write v as
(v − w) + w and apply the first inequality.]

2
1.1. Review: vectors in R3

Orthonormal bases
An orthonormal basis for R3 is a set {u, v, w} consisting of three unit vectors, so that

|u| = |v| = |w| = 1,

which are mutually orthogonal :

u · v = v · w = w · u = 0.

Given such an orthonormal basis {u, v, w}, any vector a ∈ R3 can be written uniquely
as a linear combination of these vectors:

a = (u · a)u + (v · a)v + (w · a)w. (1.2)

Furthermore, if a = λu + µv + νw then Pythagoras’s theorem holds:

|a|2 = λ2 + µ2 + ν 2 . (1.3)

The standard basis {i, j, k} is an orthonormal basis, and there are many more.

Exercise 1.2. Prove (1.2) and (1.3). [Hint for the latter: recall that |a|2 = a · a.]
The vector product
Recall that R3 has a vector product, which is also known as the cross product. This is a
bilinear map
× : R3 × R3 → R3 ; (v, w) 7→ v × w,
where

i j k

v × w := v1 v2 v3 = (v2 w3 − v3 w2 )i + (v3 w1 − v1 w3 )j + (v1 w2 − v2 w1 )k
w1 w2 w3

= v2 w3 − v3 w2 , v3 w1 − v1 w3 , v1 w2 − v2 w1

and the determinant is calculated by expansion along the first row. (The notation v ∧ w
is used by some authors for the vector product of v with w.)
Note that the vector product is anti-commutative,

v × w = −w × v for all v, w ∈ R3 ,

but is not associative: in general,

(u × v) × w 6= u × (v × w). (1.4)

3
1. Curves

The vector product has the following geometrical interpretation: the magnitude of v ×w
is equal to
|v| |w| sin θ,
where θ ∈ [0, π] is the angle between v and w, and the direction of v×w is perpendicular
to both v and w and such that (v, w, v × w) is a right-handed triple. Physically, a
corkscrew turned from v to w will move in the direction of v × w.

Exercise 1.3. If v and w are linearly independent then they span a plane, so there
are only two directions which are orthogonal to both vectors. However, if v and w are
linearly dependent then they lie on a straight line (they are co-linear ) and there are
infinitely many directions which are perpendicular to this line: why is there no problem
with the geometrical interpretation in this case? [Hint: look at the magnitude of v × w.]

The product v × w is a vector which is orthogonal to both v and w. In particular,

i × j = k, j × k = i and k × i = j. (1.5)

If u and v are orthogonal unit vectors then (u, v, u × v) is a right-handed triple of


orthogonal unit vectors or orthonormal triple: this is a ordered collection of orthonormal
vectors, (u, v, w), such that

u × v = w, v × w = u and w × u = v.

By convention, our orthonormal triples will be right-handed unless otherwise specified.

Exercise 1.4. Use the vector-product identities (1.5) satisfied by i, j and k to find
vectors u, v and w which demonstrate the non-associativity of ×; that is, find u, v
and w such that (1.4) holds.

Exercise 1.5. Let v and w be two vectors in R3 . To what geometrical quantity does
the number |v| |w| sin θ correspond? [Hint: it is the product of two lengths, so is an
area: of what?]

4
1.1. Review: vectors in R3

Exercise 1.6. Use the answer to Exercise 1.5 to give a geometrical interpretation of
the scalar triple product
[u, v, w] := u · (v × w),
where u, v and w are vectors in R3 . [Hint: up to a choice of sign, this is the volume of
something.]

Show also that


u1 u2 u3

[u, v, w] = v1 v2 v3 .
w1 w2 w3
Deduce that the scalar triple product is unchanged by cyclically permuting its arguments,
but changes sign if two of its arguments are transposed.

Exercise 1.7. Let {u, v, w} be an orthonormal basis. Show that if [u, v, w] = 1 then
(u, v, w) is a right-handed triple. What other values can [u, v, w] take, and what
happens then?

Exercise 1.8. Prove the following identity:

|v × w|2 = |v|2 |w|2 − (v · w)2 for all v, w ∈ R3 .

Exercise 1.9. Let u, v, w, z ∈ R3 . By writing both sides in terms of coordinates, show


that
u × (v × w) = (u · w)v − (u · v)w.
Using this, deduce that

u·w u·z

(u × v) · (w × z) = (u · w)(v · z) − (u · z)(v · w) = .
v·w v·z

[Hint: for this deduction, consider the scalar triple product [u × v, w, z].]

5
1. Curves

Lines and planes


Given distinct points a and b in R3 , the straight line passing through them is the set of
points
{λa + (1 − λ)b : λ ∈ R},
with the subset
{λa + (1 − λ)b : 0 6 λ 6 1}
being the straight-line segment with end points a and b.
Alternatively, a straight line L may be specified by giving a direction, that is, a unit
vector u, and a point a through which the line L passes. Then

L = {a + λu : λ ∈ R}.

(More generally, the direction u can be any non-zero vector which is parallel to L.)

A plane P may be specified by giving two linearly independent vectors v and w which
are parallel to P and a point a which lies in P ; in this case

P = {a + λv + µw : λ, µ ∈ R}.

Alternatively, P may be specified by giving a unit vector c to which P is orthogonal and


the perpendicular distance d from the origin to P , measured in the direction c; in terms
of these quantities,
P = {r ∈ R3 : r · c = d}.

Exercise 1.10. Show that the different ways of specifying straight lines are equivalent,
by explaining how to go from each one to the other. Do the same thing for planes. [Hint
for the latter: given any unit vector u, there exist vectors v and w such that {u, v, w}
is an orthonormal basis.]

Exercise 1.11. Prove that a straight line L may be written as

{r ∈ R3 : r × c = d}

where c and d are two unit vectors. How does this representation relate to the other
two ways of describing L?

6
1.2. Parameterised curves

Spheres and circles


A sphere is the collection of all points in R3 equidistant from its centre, this distance being
called the radius. If d = (a, b, c) is the centre and r > 0 the radius then r = (x, y, z)
lies on the sphere if and only if

|r − d| = r ⇐⇒ |r − d|2 = r 2 ⇐⇒ (x − a)2 + (y − b)2 + (z − c)2 = r 2 .

A circle C is the collection of all points in a plane P equidistant from its centre, a point
in P . If d ∈ P is the centre and the circle has radius r > 0 then

C = {r ∈ R3 : |r − d| = r} ∩ {d + λv + µw : λ, µ ∈ R}
= {d + λv + µw : |λv + µw| = r}.

where the vectors v and w are linearly independent and parallel to P . If v and w are
orthogonal unit vectors then
 
C = d + λv + µw : λ2 + µ2 = r 2 = d + (r cos θ) v + (r sin θ) w : θ ∈ (−π, π] .

1.2 Parameterised curves


You are probably used to thinking of a curve as the graph of some function (for example,
the parabola y = x2 or the sinusoidal wave y = sin x) or, more generally, as the set of
points satisfying an equation, such as the circle

{(x, y) ∈ R2 : x2 + y 2 = 1}.

In these notes, we will prefer a dynamic perspective: we are not just interested in the
points which make up the curve, but how those points are traced out.
Another difference is that we will consider curves which live in R3 and not R2 ; it should be
clear that this is a straightforward generalisation – we can always restrict our attention
to planar curves (those which lie in a plane) if we wish. (We will study certain properties
peculiar to planar curves in Chapter 3.)

7
1. Curves

Definition 1.1. A parameterised curve is a continuous function



γ : (a, b) → R3 ; t 7→ γ(t) = γ1 (t), γ2 (t), γ3 (t) ,

where the open interval


(a, b) := {x ∈ R : a < x < b}
for a ∈ {−∞} ∪ R and b ∈ R ∪ {∞} such that a < b; more concisely, −∞ 6 a < b 6 ∞.
We include intervals such as (−∞, 0), (2, ∞) and (−∞, ∞) = R.

As t runs over the interval from a to b, we imagine the point γ(t) tracing out the curve
in R3 ; the parameter t can be thought of as time. The requirement of continuity ensures
that the curve is connected : informally, this means that the curve is made up of just one
piece, without any jumps or gaps as the parameter t moves through the interval (a, b).
It may seem more natural for the parameter interval to be closed, in other words, to
include the end points a and b (as long as these are finite). However, we will be interested
in curves which are differentiable functions, and excluding the end points simplifies
matters a little: there is no need to worry about one-sided derivatives.
Rectifiable curves
Suppose we have a parameterised curve γ : (a, b) → R3 . A natural question to ask
is, how long is it? If γ is part of a straight line, the answer is easy; if it is made
up of straight-lines segments (is piecewise linear or polygonal ) then the answer is also
straightforward. This suggests a way to approach the general case: by using polygonal
approximation.
We divide the interval (a, b) into subintervals, by taking t0 , t1 , . . . , tn such that

a < t0 < t1 < · · · < tn < b.

We then sum the lengths of each straight-line segment between consecutive points γ(ti−1 )
and γ(ti ) as i runs from 1 to n. As we include more and more points in the subdivision
of (a, b), we expect this approximation to become closer and closer to the “true” length
of the curve γ.

8
1.2. Parameterised curves

The approximate length of γ given by this procedure is


n
X
L = |γ(t1 ) − γ(t0 )| + |γ(t2 ) − γ(t1 )| + · · · + |γ(tn ) − γ(tn−1 )| = |γ(ti ) − γ(ti−1 )|.
i=1

If the set of all such approximations


nX
n o
S= |γ(ti ) − γ(ti−1 )| : a < t0 < · · · < tn < b, n > 1
i=1

is bounded above then the curve is said to be rectifiable and its length is defined to equal
the supremum (the least upper bound) of S.
It is not easy to see how, in practice, these approximations could be efficiently calculated
for all but the simplest of examples. However, for a large class of curves, there is a way.
The trick is to re-write L in the following manner:
Xn
γ(ti ) − γ(ti−1 )
L= (ti − ti−1 ).
t i − ti−1

i=1

If the points ti−1 and ti are close together then the ratio
γ(ti ) − γ(ti−1 )
≈ γ ′ (ti−1 ),
ti − ti−1
the derivative of γ at ti−1 . (Don’t worry too much about whether this makes sense for
the moment; if necessary, pretend that the function γ is scalar valued.) Hence
X n Z b

L≈ |γ (ti−1 )|(ti − ti−1 ) ≈ |γ ′ (t)| dt.
i=1 a

This gives a way to calculate the length of γ, provided that it is “sufficiently smooth”.

Definition 1.2. A curve γ : (a, b) → R3 is continuously differentiable if


γ(t + h) − γ(t)
γ ′ (t) := lim
h→0 h
exists for all t ∈ (a, b) and the derivative γ ′ : (a, b) → R3 is a continuous function;
thinking dynamically, the vector γ ′ (t) is the velocity of the curve at time t and |γ ′ (t)| is
its speed . If γ is continuously differentiable then it has length
Z b
ℓ(γ) := |γ ′ (t)| dt.
a

It may be shown that this definition of length agrees with the previous one. We will not
worry about this, as we will be dealing only with continuously differentiable curves (and
so can use this definition).

9
1. Curves

An immediate question presents itself: what do we mean by the limit of a function with
values in R3 ? If 
v : (a, b) → R3 ; t 7→ v1 (t), v2 (t), v3 (t)
is a vector-valued function and s ∈ (a, b) then the limit

lim v(t)
t→s

exists if and only if the limits of the coordinate functions do: that is,
 
lim v(t) = lim v1 (t), lim v2 (t), lim v3 (t) ,
t→s t→s t→s t→s

in the sense that the left-hand side exists if and only if each of the limits on the right-hand
side do, in which case the two sides are equal.
(If you know about metric spaces, you should convince yourself that this is equivalent
to regarding R3 as a metric space, where the metric d is such that d(v, w) = |v − w| for
all v, w ∈ R3 .)
In particular, the curve

γ : (a, b) → R3 ; t 7→ γ1 (t), γ2 (t), γ3 (t)

is continuously differentiable if and only if the real-valued functions γ1 , γ2 and γ3 are


differentiable on (a, b) and the derivatives γ1′ , γ2′ and γ3′ are continuous, in which case,

γ ′ (t) = γ1′ (t), γ2′ (t), γ3′ (t) for all t ∈ (a, b).

Example 1.3. If r > 0 is fixed and γ(t) := (r cos t, r sin t, 0) for all t ∈ (−π, π) then
this curve describes a circle in the x-y plane with radius r and centre (0, 0, 0), minus the
point (−r, 0, 0).

Since
γ ′ (t) = (−r sin t, r cos t, 0) for all t ∈ (−π, π),
the curve γ has length
Z π p Z π
ℓ(γ) = r 2 sin2 t+ r2 cos2 t+ 02 dt = r dt = 2πr,
−π −π

as we would expect.

10
1.2. Parameterised curves

Exercise 1.12. Show that the curve


√ 
γ : (−1, 1) → R3 ; t 7→ 1 + t2 , 0, 2

is continuously differentiable and find its length. Suggest a better parameterisation for
this curve.
Some calculus for vector-valued functions

Exercise 1.13. Let v : (a, b) → R3 and w : (a, b) → R3 be differentiable at t ∈ (a, b).


Show that v · w is differentiable at t, with

(v · w)′ (t) = v′ (t) · w(t) + v(t) · w′ (t).

Show also that v × w is differentiable at t and give a formula for (v × w)′ (t). Finally,
show that if v(t) 6= 0 then |v| is differentiable at t, with |v|′(t) = (v · v′ )(t)/|v|(t).

Exercise 1.14. Use the usual (scalar-valued) form of the chain rule to prove the following
version: if f : (c, d) → (a, b) is differentiable at t ∈ (c, d) and v : (a, b) → R3 is
differentiable at f (t) then

v ◦ f : (c, d) → R3 ; x 7→ v f (x)

is differentiable at t, with (v ◦ f )′ (t) = v′ f (t) f ′ (t).
Best linear approximation
Suppose the curve γ : (a, b) → R3 is differentiable at the point t ∈ (a, b) and let

v : R → R3 ; λ 7→ γ(t) + λd

be a straight-line segment through γ(t) in the direction of the vector d ∈ R3 \ {0}.

If h is small then
γ(t + h) = γ(t) + hγ ′ (t) + o(h),
simply by the definition of γ ′ (t) and o(h) (see Notation A.1). Hence

|γ(t + h) − v(h)| = |γ(t) + hγ ′ (t) + o(h) − γ(t) − hd| = h|γ ′ (t) − d| + o(h)

and the “error” between γ(t + h) and v(h) is smallest if and only if d = γ ′ (t). Thus,
as long as γ ′ (t) 6= 0, the straight line λ 7→ γ(t) + λγ ′ (t) is the best linear approximation
to γ at t.

11
1. Curves

Space-filling curves
A continuously differentiable curve cannot take up much room, as the following pair of
exercises demonstrates.

Exercise 1.15. Let γ : (a, b) → R3 be continuously differentiable and let c, d ∈ (a, b)


be such that c < d. Explain why
Z d
|γ ′ (t)| dt > |γ(d) − γ(c)|. (1.6)
c

[Hint: think geometrically.] Given n > 1, prove there exist points

c = t0 < t1 < · · · < tn = d

such that s(ti ) − s(ti−1 ) = s(d)/n for i = 1, . . . , n, where


Z t
s(t) := |γ ′ (r)| dr for all t ∈ [c, d].
c

[Hint: use the intermediate-value theorem, Theorem A.2.] Deduce that the set

γ [c, d] := {γ(t) : c 6 t 6 d}

can be covered by n spheres, each of radius s(d)/n. [Hint: centre the spheres at γ(ti )
for i = 1, . . . , n.]

Exercise 1.16. Prove that a continuously differentiable curve γ cannot “fill space”.
[Hint: suppose there exist c, d ∈ (a, b) such that c < d and γ [c, d] contains a cube of
positive volume. Use Exercise 1.15 to obtain a contradiction.]

However, if the smoothness condition is dropped then more pathological behaviour may
be obtained. In fact, there exists a continuous curve γ : [0, 1] → R3 which has image the
whole of the unit cube [0, 1]3 . (This was first proved by Peano in an article [5] published
in 1890.) Such space-filling curves were quite shocking when they first appeared but
they are now well understood. A good explanation of how to construct one can be found
in [4, Chapter 7, §44].

12
1.3. Unit-speed curves

1.3 Unit-speed curves


If two curves γ 1 : (a, b) → R3 and γ 2 : (c, d) → R3 have the same image, so that
{γ 1 (t) : t ∈ (a, b)} = {γ 2 (t) : t ∈ (c, d)},
then each is a different parameterisation of that image. For a given curve, length as
measured along it, called arc length, can be used to obtain a particularly well-behaved
parameterisation.

Definition 1.4. A continuously differentiable curve γ : (a, b) → R3 is regular if γ ′ (t) 6= 0


for all t ∈ (a, b). Geometrically, a regular curve may be approximated to first order at
every point by a straight line, as shown above.

Exercise 1.17. Show that the continuously differentiable curve



γ : (−1, 1) → R3 ; t 7→ t3 , t6 , t9
is not regular. Find a regular curve with the same image as γ.

Definition 1.5. Suppose the continuously differentiable curve γ is regular and fix a
point m ∈ (a, b). If Z t
s(t) := |γ ′ (r)| dr for all t ∈ (a, b)
m
then s is the arc-length function for γ with starting point m. This function has a strictly
positive, continuous derivative, so is strictly increasing and thus a bijection from (a, b)
onto (c, d), where
Z m Z b

c := − |γ (r)| dr and d := |γ ′ (r)| dr.
a m

(Different choices of starting point give rise to arc-length functions which differ only by
a constant.)

By Theorem A.3, the inverse s−1 : (c, d) → (a, b) is continuously differentiable; let

γe := γ ◦ s−1 : (c, d) → R3 ; x 7→ γ s−1 (x) .
Then γ and γ e have the same image, but γ e is a unit-speed parameterisation of this image:
if x ∈ (c, d) then Exercise 1.14 and Theorem A.3 imply that
   −1
γ ′ (x)| = γ ′ s−1 (x) |(s−1 )′ (x)| = γ ′ s−1 (x) |s′ s−1 (x) = 1,
|e
since s′ (t) = |γ ′ (t)| for all t ∈ (a, b). This will be our preferred choice of parameterisation:
it makes many formulae more simple.
Note that if n ∈ (c, d) then the arc-length function se for γ
e with starting point n is such
that Z x Z x

se(x) = |e
γ (r)| dr = dr = x − n for all x ∈ (c, d) :
n n
e is parameterised by arc length.
the curve γ

13
1. Curves

Exercise 1.18. Prove the converse statement, that a curve parameterised by arc length
has unit speed.

Example 1.6. The curve



γ : R → R3 ; t 7→ 3 cosh t, 4 sinh t, 3t

has tangent vector 


γ ′ (t) = 3 sinh t, 4 cosh t, 3
with magnitude
p p
|γ (t)| = 9 sinh t + 16 cosh t + 9 = 25 cosh2 t = 5 cosh t,
′ 2 2

for all t ∈ R. Hence the arc-length function for γ with starting point 0 is
Z t
s : R → R; t 7→ 5 cosh r dr = 5 sinh t,
0

which has inverse x


−1 −1
s : R → R; x 7→ sinh .
5
The reparameterised curve
 x 
3 3√ 4
e := γ ◦ s
γ −1
: R → R ; x 7→ 25 + x2 , x, 3 sinh−1
5 5 5
d 1
has unit speed, which may be verified directly: if x ∈ R then, as sinh−1 z = √ ,
dz 1 + z2
9 x2 16 9 25
γ ′ (x)|2 =
|e + +
25 25 + x2 25 25 25 + x2
9x2 + 16(25 + x2 ) + 225
=
25(25 + x2 )
25x2 + 625
=
25x2 + 625
= 1.

This example shows that reparameterisation by arc length may make calculations more
difficult in specific cases, although it makes the general theory more straightforward.

Exercise 1.19. Let γ : (a, b) → R3 be a regular curve. Show that its length is unchanged
by reparameterisation: if f : (c, d) → (a, b) has strictly positive, continuous derivative
then 
e := γ ◦ f : (c, d) → R3 ; x 7→ γ f (x)
γ
is a regular curve with the same length as γ. Prove the same holds if f : (c, d) → (a, b)
has strictly negative, continuous derivative.

14
1.3. Unit-speed curves

The tangent vector field


If the regular curve γ : (a, b) → R3 is parameterised by arc length then the unit tangent
vector at s ∈ (a, b) is
t(s) := γ ′ (s);
note that t(s) is a unit vector because γ has unit speed. The function

t : (a, b) → R3 ; s 7→ t(s)

is the unit tangent vector field to the curve γ.

Given a curve γ : (a, b) → R3 , any related vector-valued function v : (a, b) → R3 defined


on the same parameter interval is called a vector field , and v(t) is the value of the vector
field at the point t ∈ (a, b). (We say “the point t”, rather than “the point γ(t)”, since
the curve γ need not be simple: distinct points s and t in (a, b) may correspond to the
same point γ(t) = γ(s). To put it another way, points on the curve are specified by
the parameter t rather than their location in R3 .) We will also be interested in scalar
fields: these are simply maps of the form f : (a, b) → R, where f depends on γ in some
manner.

Lemma 1.7. Suppose v : (a, b) → R3 has constant magnitude: there exists c ∈ R such
that |v(t)| = c for all t ∈ (a, b). If v is differentiable at t ∈ (a, b) then v(t) · v′ (t) = 0.

Proof. As v · v = |v|2 ≡ c2 , it follows that

0 = (v · v)′ (t) = v′ (t) · v(t) + v(t) · v′ (t) = 2v(t) · v′ (t).

15
1. Curves

Smooth curves
It will be convenient to strengthen the differentiability condition: a curve γ : (a, b) → R3
is smooth if it is infinitely differentiable; that is, the nth derivative γ (n) (t) exists for
all t ∈ (a, b) and n > 1, where γ (0) := γ and

γ (n−1) (t + h) − γ (n−1) (t)


γ (n) (t) := lim for all t ∈ (a, b).
h→0 h
In terms of coordinates, the curve γ is smooth if and only if each of its coordinate
functions is infinitely differentiable.

Exercise 1.20. Prove that if γ is smooth then so is any arc-length function for γ.
Deduce that a regular smooth curve has a smooth unit-speed reparameterisation.

Henceforth, curves will be taken to be smooth unless it is stated otherwise.


Curvature
Let γ : (a, b) → R3 be a unit-speed curve. It follows from Lemma 1.7 that t′ is everywhere
orthogonal to t, the unit tangent vector field. The curvature κ, a non-negative scalar
field, is defined by setting

κ(s) := |t′ (s)| = |γ ′′ (s)| for all s ∈ (a, b) :

the curvature of a unit-speed curve is the magnitude of the derivative of the tangent
vector field t. (More generally, curvature can be defined for any regular smooth curve:
see Exercise 1.26).

Exercise 1.21. Fix r > 0 and let

γ : (−π, π) → R3 ; t 7→ (r cos t, r sin t, 0)

be the circle in the x-y plane with radius r and centre (0, 0, 0) (minus one point).
Show that γ is regular. Find an arc-length function for γ and so obtain a unit-speed
e . Prove that γ
reparameterisation γ e has constant curvature 1/r.

Exercise 1.22. Prove that a unit-speed curve γ with zero curvature lies on a straight
line.

16
1.3. Unit-speed curves

The normal vector field


If γ : (a, b) → R3 is a unit-speed curve with curvature κ(s) 6= 0 at some point s ∈ (a, b)
then n(s) := t′ (s)/κ(s) is a unit vector, the unit normal vector . This is such that

t(s) · n(s) = 0 and t′ (s) = κ(s)n(s).

If the curvature scalar field κ is strictly positive then the unit normal vector field

n : (a, b) → R3 ; s 7→ n(s) := t′ (s)/κ(s)

is well defined, orthogonal to t and such that t′ = κn.

Exercise 1.23. Suppose the unit-speed curve γ : (a, b) → R3 has strictly positive
curvature and is planar: there exists a constant unit vector u and a constant d ∈ R such
that γ(s) · u = d for all s ∈ (a, b). Show that {t(s), n(s), u} is an orthonormal basis
of R3 , for all s ∈ (a, b). [Hint: differentiate γ · u twice.] Use this to prove that, in this
case, n′ = −κt. [Hint: let n′ = λt + µn + νu. Find λ, µ and ν.]

Exercise 1.24. e found in


Find the normal vector field n for the unit-speed circle γ
Exercise 1.21.

Example 1.8. If γ is a planar unit-speed curve with constant curvature κ > 0 then γ
lies on a circle.
To see this, suppose γ lies in the plane

P := {r ∈ R3 : r · u = d},

where the unit vector u and the scalar d are constant. By Exercise 1.23, if c := γ + κ−1 n
then
c · u = γ · u + κ−1 n · u ≡ d + 0 = d
and c lies in P . Furthermore, again by Exercise 1.23,

c′ = (γ + κ−1 n)′ = t + κ−1 n′ = t − t ≡ 0,

and therefore c is constant. Finally,

|γ − c| = |κ−1 n| ≡ κ−1 ,

so γ lies on the circle in the plane P with centre c and radius κ−1 .

17
1. Curves

Best circular approximation


Given a unit-speed curve γ : (a, b) → R3 , fix s ∈ (a, b). For sufficiently small h, Taylor’s
theorem (Theorem A.4) implies that

γ(s + h) = γ(s) + hγ ′ (s) + 12 h2 γ ′′ (s) + o(h2 ). (1.7)

If κ(s) = 0 then γ ′′ (s) = t′ (s) = 0 and

|γ(s + h) − γ(s) − ht(s)| = o(h2 ),

so the best linear approximation to γ has second-order contact at s (not just the usual
first-order contact). Otherwise, κ(s) > 0 and n(s) is well defined, so

γ(s + h) = γ(s) + ht(s) + 12 h2 κ(s)n(s) + o(h2 )

for small h. In the first-order situation we saw that the line through γ(s) in the direction
of t(s) was the best straight-line approximation to γ at s. Now we will show that the
circle C passing through γ(s) with centre γ(s) + rn(s), lying in the plane

P := {γ(s) + λt(s) + µn(s) : λ, µ ∈ R},

gives the best circular approximation to γ when the radius r = κ(s)−1 .

The point γ(s) + ht(s) + kn(s) in P will be close to γ(s + h) if k is chosen suitably;
by (1.7), the error

η := |γ(s + h) − γ(s) − ht(s) − kn(s)| = | 12 κ(s)h2 − k| + o(h2 ).

Furthermore, γ(s) + ht(s) + kn(s) lies on the circle

C = {v ∈ P : |v − γ(s) − rn(s)|2 = r 2 } = {γ(s) + λt(s) + µn(s) : λ2 + (µ − r)2 = r 2 }

if and only if √
h2 + (k − r)2 = r 2 ⇐⇒ k = r ± r 2 − h2 .

18
1.3. Unit-speed curves

The negative square root gives a point on C close to γ(s + h); the positive choice gives
the point at the other end of the chord parallel to n(s). With the negative choice, a
binomial expansion gives that

k = r 1 − (1 − r −2 h2 )1/2 = 12 r −1 h2 + o(h2 )

and therefore the error

η = | 21 κ(s)h2 − k| + o(h2 ) = 12 |κ(s) − r −1 |h2 + o(h2 ).

Hence C has second-order contact at γ(s) if and only if r = κ(s)−1 ; the error η is o(h2 )
(smaller than second order) in this case
The circle in the osculating plane P which has centre γ(s)+κ(s)−1 n(s) and radius κ(s)−1 ,
called the osculating circle at s, is the best circular approximation to γ at the point s.
The binormal vector field
Suppose the unit-speed curve γ has strictly positive curvature, so that n is well defined.
The cross product b := t × n is the binormal vector field associated with γ. At every
point s ∈ (a, b), the triple 
t(s), n(s), b(s)
is a (right-handed) orthonormal triple which evolves with the curve.
In particular, any vector field v can be decomposed into components in the t, n and b
directions:
v = (t · v)t + (n · v)n + (b · v)b.

We will use this observation to analyse how n and b evolve. Note that

b′ = (t × n)′ = t′ × n + t × n′ = κn × n + t × n′ = t × n′ ;

in particular, t · b′ = [t, t, n′ ] ≡ 0. By Lemma 1.7, b · b′ ≡ 0, and therefore

b′ = (n · b′ )n = [n, t, n′ ]n = [n′ , n, t]n = −[n′ , t, n]n = −τ n,

where τ := n′ · b is the torsion of γ; this scalar field τ measures how much the curve
deviates from the t-n plane.
Furthermore, n = b × t, so

n′ = (b × t)′ = b′ × t + b × t′ = −τ n × t + b × κn = τ b − κt.

In summary, we have the Serret–Frenet equations [1, 9, 10].

19
1. Curves

Theorem 1.9. (Serret–Frenet) If the smooth curve γ has unit speed and strictly
positive curvature then the following equations hold.

t′ = κn
n′ = −κt +τ b (S–F)
b′ = −τ n

In the above, t, n and b are the unit tangent, normal and binormal vector fields for γ,
whereas κ and τ are the curvature and torsion scalar fields for γ.

Exercise 1.25. Let γ : (a, b) → R3 be a unit-speed curve which has strictly positive
curvature and zero torsion: τ ≡ 0. Show that γ lies in a plane: there exist a constant
unit vector u and a constant scalar d such that γ(s) · u = d for all s ∈ (a, b). [Hint: use
the third Serret–Frenet equation.]

Example 1.10. If the unit-speed curve γ has strictly positive curvature and lies in a
plane then γ has zero torsion.
To see this, suppose γ · u ≡ d, where the unit vector u and the scalar d are constant.
It follows from Exercise 1.23 that u is orthogonal to t and n. Hence u is proportional
to b; as both are unit vectors, b = ±u. Since b is continuous, b · u = ±u · u = ±1 is
constant (by the intermediate-value theorem, Theorem A.2) and therefore b is constant
(because b = (b · u)u). By the third Serret–Frenet equation, τ ≡ 0.

Exercise 1.26. Let γ : (a, b) → R3 be a regular smooth curve with an arc-length


function s : (a, b) → (c, d). Suppose the unit-speed curve γ e = γ ◦ s−1 : (c, d) → R3
has unit tangent vector field e t, curvature scalar field κ
e (which is strictly positive), unit
normal vector field n e
e , unit binormal vector field b and torsion scalar field τe. Show that
 
γ ′ = |γ ′ |(et ◦ s), (et ◦ s)′ = |γ ′ | (e
κne ) ◦ s and γ ′ × γ ′′ = |γ ′ |3 (e e ◦s .
κb)

Deduce that
|γ ′ × γ ′′ |
e◦s=
κ .
|γ ′ |3
Show further that
 [γ ′ , γ ′′ , γ ′′′ ]
[γ ′ , γ ′′ , γ ′′′ ] = |γ ′ |6 (e
κ2 τe) ◦ s and τe ◦ s = .
|γ ′ × γ ′′ |2

Exercise 1.27. Let the unit-speed curve γ : (a, b) → R3 have constant curvature κ > 0
and constant torsion τ . Prove that γ describes a helix. [Hint: find n′′ .]

20
Two Surfaces – local theory

In this chapter, we will begin an investigation into the local theory of surfaces. The
objects of interest are surface patches: these are sufficiently differentiable, injective maps
from open subsets of R2 to R3 , and we think of them as equipping a region of space with
a two-dimensional coordinate system. In Chapter 3 we will consider the global theory
of how these patches fit together to form more complicated surfaces.
A parameterised curve is a map γ : I → R3 , where I is an open subinterval of R. For
a surface, this parameter interval needs to be replaced by a suitable subset of R2 , one
which is open. This is for the same reason we choose to define curves on open intervals
– those without endpoints; the key idea is that we must have room to define derivatives.
Recall that the open disc with centre (x, y) ∈ R2 and radius r > 0 is the set

D(x, y; r) := (u, v) ∈ R2 : (u − x)2 + (v − y)2 < r 2 .

A set U ⊆ R2 is open if every point in U is contained in some open disc which lies
completely within U. Formally, U is open if for every (x, y) ∈ U there exists r > 0 such
that D(x, y; r) ⊆ U. (Note that r can depend on x and y.) If a function is defined on an
open set then we can easily talk about the limit of that function at any point in the set.

Example 2.1. The empty set ∅ is open, since it satisfies the definition vacuously. The
whole of R2 is also an open set: given (x, y) ∈ R2 , the disc D(x, y; 1) ⊆ R2 (and 1 can
be replaced by any other choice of r > 0). Any open disc is open – draw a picture to
convince yourself of this – as is any open rectangle, a set of the form

(a, b) × (c, d) = {(x, y) ∈ R2 : a < x < b, c < y < d},

where −∞ 6 a < b 6 ∞ and −∞ 6 c < d 6 ∞.

21
2. Surfaces – local theory

We shall also require that the parameter set U ⊆ R2 is connected, so that every pair of
points in U can be joined by a path which lies within U. This requirement ensures that
a surface is also connected, and thus composed of only one piece.
Formally, the open set U ⊆ R2 is connected if, for all x, y ∈ U, there exists a continuous
function f : [0, 1] → U such that f (0) = x and f (1) = y. All the sets in Example 2.1
are connected. (For comparison, the connected subsets of R are precisely the intervals.)

Partial derivatives
Given a vector-valued map v : U → R3 , where U is an open set in R2 , the partial
derivatives ∂1 v and ∂2 v are defined by setting

∂v v(x + h, y) − v(x, y)
∂1 v(x, y) := (x, y) = lim
∂x h→0 h
and
∂v v(x, y + h) − v(x, y)
∂2 v(x, y) := (x, y) = lim
∂y h→0 h
for all (x, y) ∈ U. Similarly to the one-dimensional case, if v = (v1 , v2 , v3 ) then

∂v  ∂v ∂v2 ∂v3 
1
∂1 v(x, y) = (x, y) = (x, y), (x, y), (x, y) ,
∂x ∂x ∂x ∂x
in the sense that the left-hand side exists if and only if each derivative on the right-hand
side does, and then they are equal. Of course, the same holds for the partial derivative
with respect to the second variable.
A map v : U → R3 is smooth if it has partial derivatives of all orders, so that

∂11 v := ∂1 (∂1 v), ∂12 v := ∂1 (∂2 v), ∂21 v := ∂2 (∂1 v) and ∂22 v := ∂2 (∂2 v)

exist, and similarly for the higher derivatives. It is a standard result that if the mixed
second-order partial derivatives are continuous, as they must be for a smooth map, then
they are equal: ∂12 v = ∂21 v. A similar result holds for all higher derivatives: mixed
partial derivatives of a smooth map may be calculated in any order.

22
Definition 2.2. A surface patch is a smooth injection r : U → R3 , where U is a non-
empty connected open set in R2 . We think of the map (x, y) 7→ r(x, y) as introducing a
system of coordinates on the image r(U).

A surface patch r : U → R3 is regular if the tangent vector fields ∂1 r and ∂2 r are linearly
independent at every point of U.
(Given a surface patch r : U → R3 , any function v : U → R3 which depends on r is a
vector field , as in the one-dimensional case, and similarly for a scalar field.)

Let r : U → R3 be a surface patch. The tangent vector fields ∂1 r and ∂2 r are linear
approximations to the surface in the u and v directions; if they are linearly independent
at the point (x, y) ∈ U then they span a plane, the tangent plane

T(x,y) r := {λ ∂1 r(x, y) + µ ∂2 r(x, y) : λ, µ ∈ R}.

The plane r(x, y) + T(x,y) r is the best linear approximation to the regular surface patch r
at the point (x, y), in the same way that the tangent line {γ(t) + λγ ′ (t) : λ ∈ R} is
the best linear approximation to the regular curve γ at the point t. A proof of this fact
requires the two-variable version of Taylor’s theorem, so we omit it.

Exercise 2.1. Explain why the surface patch r is regular if and only if |∂1 r × ∂2 r| > 0
everywhere.

23
2. Surfaces – local theory

Example 2.3. Let γ : I → R3 be a regular curve which is simple (if s, t ∈ I are such
that γ(s) = γ(t) then s = t) and planar (there exist a unit vector c and a scalar d which
are constant and such that γ · c ≡ d). If

r : I × R → R3 ; (x, y) 7→ γ(x) + y c

then r is a surface patch, called a generalised cylinder or prism.

To see that r is injective, note that

r(u, v) = r(x, y) =⇒ r(u, v) · c = r(x, y) · c =⇒ v=y

and then γ(u) = γ(x) so, because γ is simple, u = x. Furthermore,

∂1 r(x, y) = γ ′ (x) and ∂2 r(x, y) = c.

Since γ(x) · c = d, for all x ∈ I, differentiating both sides of this equality gives that

0 = γ ′ (x) · c + γ(x) · c′ = γ ′ (x) · c,

so γ ′ (x) and c are orthogonal. Hence if (x, y) ∈ I × R then

|∂1 r × ∂2 r|(x, y) = |γ ′ (x) × c| = |γ ′ (x)| |c| = |γ ′ (x)| > 0,

as γ is regular, and the surface patch r is regular as well.

24
Exercise 2.2. Let γ : I → R3 be a regular curve which is simple and which lies in
a plane that does not contain the origin: there exists a unit vector c and a non-zero
scalar d such that γ · c ≡ d. Prove that

r : I × (0, ∞) → R3 ; (x, y) 7→ y γ(x)

is a regular surface patch. This type of surface is called a generalised cone.

Definition 2.4. If r : U → R3 is a regular surface patch then


∂1 r × ∂2 r
N :=
|∂1 r × ∂2 r|

is the unit normal vector field , made up of unit vectors which are orthogonal to the
surface. (The capital letter N is chosen to distinguish it from n, the unit normal to a
curve.)

Exercise 2.3. Show that the surface patch

r : R × (−π, π) → R3 ; (z, φ) 7→ (z cos φ, z sin φ, z)



is not regular. Describe the image r R × (−π, π) .

25
2. Surfaces – local theory

Example 2.5. The surface patch

r : (−π/2, π/2) × (−π, π) → R3 ; (θ, φ) 7→ (cos θ cos φ, cos θ sin φ, sin θ)

is a parameterisation of the unit sphere S 2 = {r ∈ R3 : |r| = 1}, minus the semicircular


arc from the north pole (0, 0, 1) to the south pole (0, 0, −1) which passes through
(−1, 0, 0); the coordinate θ corresponds to latitude (or elevation from the x-y plane)
and φ to longitude (or azimuth).

Since

∂1 r(θ, φ) = (− sin θ cos φ, − sin θ sin φ, cos θ)


and ∂2 r(θ, φ) = (− cos θ sin φ, cos θ cos φ, 0),

it follows that

i j k


(∂1 r × ∂2 r)(θ, φ) = − sin θ cos φ − sin θ sin φ cos θ

− cos θ sin φ cos θ cos φ 0

= (− cos2 θ cos φ, − cos2 θ sin φ, − sin θ cos θ).

As |∂1 r × ∂2 r|(θ, φ) = cos θ, we obtain the unit normal

N(θ, φ) = (− cos θ cos φ, − cos θ sin φ, − sin θ) = −r(θ, φ).

26
2.1. The first fundamental form

2.1 The first fundamental form


Suppose r : U → R3 is a regular surface patch which contains the curve γ : I → R3 ,
where I is an open interval:

γ(t) ∈ r(U) = {r(u, v) : (u, v) ∈ U} for all t ∈ I.


 
For each t ∈ I, there exists u(t), v(t) ∈ U such that γ(t) = r u(t), v(t) . In other
words, we can (and will) regard such a curve as a map of the form

r ◦ (u, v) : t 7→ r u(t), v(t) ,

where the pair 


(u, v) : I → U; t 7→ u(t), v(t) .

Example 2.6. The curve

γ : R → R3 ; t 7→ (cosh t, sinh t, 1)

lies in the hyperbolic paraboloid

r : R × R → R3 ; (x, y) 7→ (x, y, x2 − y 2)

and has the form r ◦ (u, v), where

u : R → R; t 7→ cosh t and v : R → R; t 7→ sinh t.

(The first claim follows from the identity cosh2 t − sinh2 t = 1, valid for all t ∈ R.)

27
2. Surfaces – local theory

Recall that a curve is said to be smooth if it is infinitely differentiable. Since r has


partial derivatives of all orders, in order to show that γ = r ◦ (u, v) is smooth, it suffices,
by the chain rule, to verify that the functions u : I → R and v : I → R are infinitely
differentiable; we call (u, v) a smooth pair if this is the case.

Exercise 2.4. Let r : U → R3 be a regular surface patch and let (u, v) : I → U be a


smooth pair. Use the identity

γ ′ (t) = ∂1 r(u(t), v(t)) u′(t) + ∂2 r(u(t), v(t)) v ′(t) for all t ∈ I (2.1)

to prove that γ = r ◦ (u, v) is regular if and only if u′(t)2 + v ′ (t)2 > 0 for all t ∈ I.

We shall insist henceforth that every smooth pair (u, v) satisfies this regularity condition,
so that the resulting curve r ◦ (u, v) is always regular.
It follows from (2.1) that

γ ′ · γ ′ = ∂1 r(u, v) · ∂1 r(u, v)(u′)2 + ∂1 r(u, v) · ∂2 r(u, v)u′v ′


+ ∂2 r(u, v) · ∂1 r(u, v)u′v ′ + ∂2 r(u, v) · ∂2 r(u, v)(v ′)2
= E(u, v)(u′)2 + 2F (u, v)u′v ′ + G(u, v)(v ′)2 ,

where the parameter t is omitted for clarity and the scalar fields

E := ∂1 r · ∂1 r,
F := ∂1 r · ∂2 r = ∂2 r · ∂1 r
and G := ∂2 r · ∂2 r.

(Here and in what follows, if v : U → R3 is a vector field and (u, v) : I → U is a smooth


pair then, for convenience and clarity, we will write v ◦ (u, v) as v(u, v) et cetera, and
similarly for scalar fields. Hence we have the function

v(u, v) : I → R3 ; t 7→ v u(t), v(t) .

It is essential to remember that v(u, v) is a function on I which depends on the curve


parameter t, and is not the value of v at a point (u, v) ∈ R3 .)
Thus if I = (a, b) then the curve γ has length
Z b
ℓ(γ) := |γ ′ (t)| dt
a
Z bp
= E(u, v)(u′)2 + 2F (u, v)u′v ′ + G(u, v)(v ′)2 (t) dt
a
Z bq
  
= E u(t), v(t) u′ (t)2 + 2F u(t), v(t) u′ (t)v ′ (t) + G u(t), v(t) v ′ (t)2 dt.
a
(2.2)

28
2.1. The first fundamental form

The scalar fields E, F and G are the coefficients of the first fundamental form,
E du2 + 2F du dv + G dv 2 . (2.3)
For convenience, we will sometimes call the triple (E, F, G) the first fundamental form.
We shall regard (2.3) as a purely formal expression. However, the following calculation
is suggestive: if ds is an infinitesimal piece of arc length (whatever that may mean) then
ds2 = |dγ|2 = dγ · dγ = (∂1 r du + ∂2 r dv) · (∂1 r du + ∂2 r dv)
= ∂1 r · ∂1 r du2 + (∂1 r · ∂2 r + ∂2 r · ∂1 r) du dv + ∂2 r · ∂2 r dv 2
= E du2 + 2F du dv + G dv 2 ,
from which it follows that
Z Z √ Z p
s= ds = 2 2
E du + 2F du dv + G dv = E(u′)2 + 2F u′v ′ + G(v ′ )2 dt.

Example 2.7. The surface patch


r : R × R → R3 ; (x, y) 7→ (x, y, x2 + y 2)
is an elliptic paraboloid . This has tangent vectors
∂1 r(x, y) = (1, 0, 2x) and ∂2 r(x, y) = (0, 1, 2y),
so is regular, and the coefficients of the first fundamental form are
E(x, y) = 1 + 4x2 , F (x, y) = 4xy and G(x, y) = 1 + 4y 2 .

Example 2.8. If r > 0 is fixed and


γ : (−π, π) → R3 ; t 7→ (r cos t, r sin t, r 2 )
is a curve in the elliptic paraboloid r of Example 2.7 then γ = r ◦ (u, v), where
u : (−π, π) → R; t 7→ r cos t and v : (−π, π) → R; t 7→ r sin t.
Hence u′ (t) = −r sin t, v ′ (t) = r cos t and
  
E u(t), v(t) = 1+4r 2 cos2 t, F u(t), v(t) = 4r 2 cos t sin t, G u(t), v(t) = 1+4r 2 sin2 t.
This curve has length
Z πq
ℓ(γ) = (1 + 4r 2 cos2 t)r 2 sin2 t − 8r 4 cos2 t sin2 t + (1 + 4r 2 sin2 t)r 2 cos2 t dt
Z−π
π p
= r 2 sin2 t + r 2 cos2 t dt
−π
= 2πr;
since γ is a parameterisation of the circle x2 + y 2 = r 2 , except for one point, this is as
expected.

29
2. Surfaces – local theory

Exercise 2.5. Let r : U → R3 be a regular surface patch. Express |∂1 r × ∂2 r| in terms


of the first fundamental form. [Hint: Exercise 1.8 of Section 1.1 may be useful.]

Exercise 2.6. Let r : U → R3 be a regular surface patch and let (x, y) ∈ U. Show
that the angle θ between the tangent vectors ∂1 r(x, y) and ∂2 r(x, y) is such that

F
cos θ = √ (x, y). (2.4)
EG
An orthogonal parameterisation is one such that F ≡ 0; use (2.4) to explain why this
terminology is appropriate.
Calculating area
As well as determining the length of curves, the first fundamental form can be used to
calculate the area of part of a surface patch. The key to this is the following interpretation
of the vector product.

Consider the parallelogram P QRS, with vertices labelled clockwise starting with the
bottom left corner. The line QS divides this figure into two congruent triangles, P QS
and QRS, so the parallelogram has area
−→  −→ 

AP QRS = 21 × P S × vertical height + 12 × QR × vertical height .
−→
−→
The vertical height equals P Q sin θ, where θ is the angle between the vectors P S
−→ −→ −→
and P Q, and P S = QR, so
−→ −→ −→ −→

AP QRS = P S P Q sin θ = P S × P Q ,

−→ −→
the magnitude of the vector product of P S and P Q. Furthermore, by Exercise 1.8 of
Section 1.1,
−→ −→ r −→ 2 −→ 2  −→ −→ 2

AP QRS = P S × P Q = P S P Q − P S · P Q .

30
2.1. The first fundamental form

Hence an infinitesimally small parallelogram with sides dx = ∂1 r dx and dy = ∂2 r dy


will have area
p p
dAr = |dx|2 |dy|2 − (dx · dy)2 = ∂1 r · ∂1 r dx2 ∂2 r · ∂2 r dy 2 − (∂1 r · ∂2 r dx dy)2.

Thus if V ⊆ U then the area of r(V ) equals


ZZ ZZ √
Ar (V ) = dAr := EG − F 2 (x, y) dx dy. (2.5)
V V

Example 2.9. If

r : (−π/2, π/2) × (−π, π) → R3 ; (θ, φ) 7→ (cos θ cos φ, cos θ sin φ, sin θ)

is the parameterisation of the unit sphere S 2 from Example 2.5 then

∂1 r(θ, φ) = (− sin θ cos φ, − sin θ sin φ, cos θ)


and ∂2 r(θ, φ) = (− cos θ sin φ, cos θ cos φ, 0),

so the first fundamental form has coefficients

E(θ, φ) = 1, F (θ, φ) = 0 and G(θ, φ) = cos2 θ.

Hence the sphere has area


Z π Z π/2  π/2

Ar (−π/2, π/2) × (−π, π) = cos θ dθ dφ = 2π sin θ = 4π,
−π −π/2 −π/2

as expected.

Example 2.10. Consider a (circular) cylinder which exactly encloses the sphere of the
Example 2.5: this has a natural coordinate system obtained from the sphere by projection
parallel to the x-y plane. The corresponding surface patch is

r : (−π/2, π/2) × (−π, π) → R3 ; (θ, φ) 7→ (cos φ, sin φ, sin θ),

for which

∂1 r(θ, φ) = (0, 0, cos θ) and ∂2 r(θ, φ) = (− sin φ, cos φ, 0).

It follows that

E(θ, φ) = cos2 θ, F (θ, φ) = 0 and G(θ, φ) = 1;

in particular, (EG − F 2 )(θ, φ) = cos2 θ, the same as for the sphere.


(This is not the “usual” parameterisation of the cylinder; vertical height is measured
through the angle of elevation θ rather than directly.)

31
2. Surfaces – local theory

Surface reparameterisations
Let r : U → R3 be a surface patch. A reparameterisation of this surface patch is a
e → U, where U
bijection Φ : U e is a non-empty connected open set in R2 , such that Φ
and Φ−1 are smooth (that is, have partial derivatives of all orders). Informally, this is
just a change of coordinate system; the composite map r ◦ Φ is a regular surface patch
which describes the same region of R3 as r does.
The Jacobian of

e → U; (z, w) 7→ φ(z, w), ψ(z, w)
Φ := (φ, ψ) : U
is the 2 × 2-matrix-valued function
" #
∂1 φ(z, w) ∂2 φ(z, w)
JΦ : (z, w) 7→ ,
∂1 ψ(z, w) ∂2 ψ(z, w)
∂φ
where ∂1 φ(z, w) = (z, w) et cetera. (See Definition A.5 for more about Jacobians.)
∂z
Exercise 2.7. Explain why the Jacobian determinant, det JΦ, is either everywhere
strictly positive or everywhere strictly negative. [Hint: explain why JΦ is invertible,
then apply the intermediate-value theorem, Theorem A.2.]

A reparameterisation Φ : U e → U such that det JΦ(e e∈U


u) > 0 for all u e is orientation
preserving; one with det JΦ(eu) < 0 for all u e is orientation reversing. (Exercise 2.8
e∈U
explains this terminology.)
r := r ◦ Φ, where the reparameterisation Φ = (φ, ψ), then
If e
∂1e
r = ∂1 r(φ, ψ) ∂1 φ + ∂2 r(φ, ψ) ∂1ψ and ∂2e
r = ∂1 r(φ, ψ) ∂2φ + ∂2 r(φ, ψ) ∂2 ψ,
so
      
∂1e
r ∂1 φ ∂1 ψ ∂1 r(φ, ψ) t ∂1 r ◦ Φ
= = (JΦ) , (2.6)
∂2e
r ∂2 φ ∂2 ψ ∂2 r(φ, ψ) ∂2 r ◦ Φ
where At denotes the transpose of the matrix A.

Exercise 2.8. Prove that


(
r × ∂2e N ◦ Φ if Φ is orientation preserving,
e := ∂1e
N
r
=
|∂1e
r × ∂2e
r| −N ◦ Φ if Φ is orientation reversing.
(In other words, the unit normal vector field N is unchanged by a orientation-preserving
reparameterisation, and changes sign, so is reflected in the tangent plane, after an
orientation-reversing reparameterisation.)

It follows from (2.6) that


" # " # " #
e
E F e ∂1er · ∂1er ∂1er · ∂2er E ◦ Φ F ◦ Φ
e e = = (JΦ)t JΦ. (2.7)
F G ∂2er · ∂1er ∂2er · ∂2er F ◦Φ G◦Φ

32
2.1. The first fundamental form

In particular,

eG
E e − Fe2 = (det JΦ)2 (EG − F 2 ) ◦ Φ . (2.8)

Exercise 2.9. Use the identity (2.8) to prove that if the surface patch r is regular then
r := r ◦ Φ.
so is the reparameterised patch e

Another consequence of (2.8) is that the area


ZZ q ZZ √
Aer (V ) = e e e 2
E G − F dz dw = ( EG − F 2 ◦ Φ) | det JΦ| dz dw
V V
ZZ √
= EG − F 2 dx dy
Φ(V )

= Ar ◦ Φ (V ),

where the penultimate identity follows from the change-of-variables formula for double
integrals (Theorem A.6). Thus area, like length, is unchanged by reparameterisation.

Example 2.11. The circular cylinder of Example 2.10 admits the reparameterisation

Φ : (−1, 1) × (−π, π) → (−π/2, π/2) × (−π, π); (z, φ) 7→ (sin−1 z, φ).

This coordinate system corresponds to our normal notion of distance on the cylinder:
note that

r := r ◦ Φ : (−1, 1) × (−π, π) → R3 ; (z, φ) 7→ (cos φ, sin φ, z)


e

and a short calculation (Exercise) shows that

e ≡ 1,
E Fe ≡ 0 and G
e ≡ 1,

the same first fundamental form as the plane. We saw above that area is invariant under
reparameterisation, so combining this observation with Example 2.10 gives a celebrated
result of Archimedes: the area of a sphere is equal to area of its circumscribing cylinder.
Archimedes was so proud of this theorem that he had its statement inscribed upon his
tomb.

In fact, we have shown a stronger result: projection (parallel to the x-y plane) from the
sphere onto the cylinder preserves area. This gives a way of producing a map of the
Earth which represents area correctly; many atlases contain maps obtained by such a
cylindrical projection.

33
2. Surfaces – local theory

Exercise 2.10. A great circle C is the curve described by the intersection of the unit
sphere S 2 with a plane through the origin; in vector terms,

C = {r ∈ R3 : |r|2 = 1} ∩ {r ∈ R3 : r · u = 0},

where u is a unit vector. Show that two distinct great circles meet at exactly two points
which are antipodal : they lie on a straight line through the origin. [Hint: let

C1 = {r ∈ R3 : |r|2 = 1, r · u1 = 0} and C2 = {r ∈ R3 : |r|2 = 1, r · u2 = 0},

where u1 and u2 are linearly independent. Let v be a unit vector which is orthogonal to
both u1 and u2 , and show that if r ∈ C1 ∩ C2 then r = νv for some ν ∈ R.] A lune is
a region of the unit sphere bounded by two semicircles from distinct great circles; any
two great circles divide the sphere into four lunes. Explain why a lune has area 2α,
where α is the interior angle of the lune made by the semicircles at their intersection
(the antipodes). Verify that this is consistent with Example 2.9.

Exercise 2.11. A spherical triangle is a region on the unit sphere S 2 bounded by


(parts of) three great circles. By using the result of Exercise 2.10 on the area of lunes,
prove that a spherical triangle has area

∆ := α + β + γ − π,

where α, β and γ are the interior angles of the triangle. [Hint: let A, B and Γ be the
vertices of the triangle, with interior angles α, β and γ respectively, and let A′ , B′ and Γ′
be the corresponding antipodal points. Show that

A(ABΓ) + A(AB′ Γ) = 2β

and
3∆ + A(A′ BΓ) + A(AB′ Γ) + A(ABΓ′ ) = 2(α + β + γ),
where A(ABΓ) denotes the area of the triangle ABΓ et cetera. Now consider the region
covered by the triangles ABΓ, A′ BΓ, A′ B′ Γ and AB′ Γ.]

34
2.1. The first fundamental form

Orthogonal reparameterisation
As seen in Section 1.3, any regular curve may be reparameterised by arc length, and this
simplifies many formulae. It is natural to wonder if something similar can be done for
surfaces.
Given a regular surface patch r : U → R3 and a point (x, y) ∈ U, it is possible to
obtain a “good” reparameterisation about this point, although it may be necessary to
shrink U. More formally, there exist non-empty connected open sets V and Ve , with
(x, y) ∈ V ⊆ U, and a reparameterisation Φ : Ve → V which makes e r := r ◦ Φ an
orthogonal parameterisation: the new first fundamental form (E, e Fe, G)
e has Fe ≡ 0. In
other words, we can find a coordinate system for a neighbourhood of r(x, y) such that,
in this coordinate system, the tangent vector fields are orthogonal there.
To establish this result, however, would take us too far afield and we will have no use
for it in what follows. Some authors make use of it (for example, [7, §§12.3–4]) but the
simplifications it brings are rather nullified by the complications of its proof.
Isometric surface patches

Informally, two surface patches r1 : U1 → R3 and r2 : U2 → R3 are isometric if the notion


of distance measured along curves in them is the same, so that curves corresponding to
the same smooth pair have the same length. More formally, for this to make sense the
domains U1 and U2 of the patches must be identical, U1 = U2 = U, and then we say the
patches are isometric if, for any smooth pair

(u, v) : I → U; t 7→ u(t), v(t)
 
the curves γ 1 : I → R3 ; t 7→ r1 u(t), v(t) and γ 2 : I → R3 ; t 7→ r2 u(t), v(t) have the
same length, i.e., Z Z
|(γ 1 ) (t)| dt = |(γ 2 )′ (t)| dt.

I I

Since this must hold for any smooth pair (u, v), it follows that the first fundamental
forms are equal:

E1 (x, y) = E2 (x, y), F1 (x, y) = F2 (x, y) and G1 (x, y) = G2 (x, y)

at all points (x, y) ∈ U. (The proof of this is Exercise 2.12.) The converse is immediate
– if the first fundamental forms are the same then ℓ(γ 1 ) = ℓ(γ 2 ), by (2.2) – and so we
have the following theorem.

Theorem 2.12. Two regular surface patches r1 : U → R3 and r2 : U → R3 are isometric


if and only if their first fundamental forms are equal.

35
2. Surfaces – local theory

Exercise 2.12. Let r1 : U → R3 and r2 : U → R3 be regular surface patches with first


fundamental forms (E1 , F1 , G1 ) and (E2 , F2 , G2 ) respectively. Prove that if
Z p
E1 (u, v)(u′)2 + 2F1 (u, v)u′v ′ + G1 (u, v)(v ′)2 dt
I Z p
= E2 (u, v)(u′)2 + 2F2 (u, v)u′v ′ + G2 (u, v)(v ′)2 dt
I

for any open interval I ⊆ R and for any smooth pair (u, v) : I → U then (E1 , F1 , G1 )
and (E2 , F2 , G2 ) are equal everywhere on U. [Hint: let (a, b) ∈ U and choose ε > 0 such
that the square [a − ε, a + ε] × [b − ε, b + ε] ⊆ U; explain why this is possible. Then
consider (u, v) : (−ε, ε) → U; t 7→ (a + t, b) and two similar smooth pairs.]

Example 2.13. Since a cylinder can be obtained by rolling up a piece of paper, the
plane and the cylinder should be isometric surfaces: the act of rolling up does not distort
distances. Let
rS : R × (−π, π) → R3 ; (x, y) 7→ (x, y, 0)
be a strip in the x-y plane and let

rC : R × (−π, π) → R3 ; (z, φ) 7→ (cos φ, sin φ, z)

be the cylinder x2 + y 2 = 1, minus the line {(−1, 0, z) : z ∈ R}. Then

∂1 rS ≡ (1, 0, 0), ∂2 rS ≡ (0, 1, 0) and (ES , FS , GS ) ≡ (1, 0, 1),

whereas

∂1 rC ≡ (0, 0, 1), ∂2 rC (z, φ) = (− sin φ, cos φ, 0) and (EC , FC , GC ) ≡ (1, 0, 1).

Thus rS and rC are isometric, as expected.

36
2.1. The first fundamental form

Example 2.14. Now consider the surface of revolution obtained by rotating about the
z axis the circle (x − 2)2 + z 2 = 1 with centre (2, 0, 0) and unit radius in the x-z plane:
this is a torus, with parameterisation

r : (−π, π) × (−π, π) → R3 ; (θ, φ) 7→ (2 + cos θ) cos φ, (2 + cos θ) sin φ, sin θ .

(The circles x2 + y 2 = 1 and (x + 2)2 + z 2 = 1 are omitted.) The coordinate θ gives


the angle on the generating circle, measured from the positive x axis, and φ corresponds
to the angle through which this generating circle has been rotated about the z axis,
measured in the positive sense (anticlockwise) from the x-z plane.

It is straightforward to verify that the tangent vectors

∂1 r(θ, φ) = (− sin θ cos φ, − sin θ sin φ, cos θ),


∂2 r(θ, φ) = (−(2 + cos θ) sin φ, (2 + cos θ) cos φ, 0)

and the first fundamental form



(E, F, G)(θ, φ) = 1, 0, (2 + cos θ)2 .

Hence the torus is not isometric to the open square (−π, π) × (−π, π): if you try to
construct one by taking a piece of paper and rolling it into a cylinder then bending the
cylinder into a torus, you will find the second step impossible without squashing and
trying to stretch the paper.
Isometries and reparameterisations
Do not confuse the idea of being isometric with that of reparameterisation. The latter
just corresponds to a change of coordinate system, and all the geometrical features are
unchanged; the former preserves some geometrical quantities, such as length and area,
but other features may be completely different. For example, the cylinder and the plane
are isometric, but a cylinder contain geodesics (see below) which intersect themselves,
and this does not occur in the plane.

37
2. Surfaces – local theory

2.2 Geodesics
Normal and geodesic curvature
Let the regular surface patch r : U → R3 contain the smooth curve

γ = r ◦ (u, v) : I → R3 ; t 7→ r u(t), v(t) .

Suppose γ has unit speed; recall that this is always possible, by reparameterising the
curve using its arc length.
We can use the tools developed in Chapter 1 to analyse this curve, ignoring the surface
in which it lies, but we can also use properties of the surface to give a more detailed
picture. For example, the curve γ has tangent vector
 
t(s) = γ ′ (s) = ∂1 r u(s), v(s) u′ (s) + ∂2 r u(s), v(s) v ′ (s) ∈ T(u(s),v(s)) r,

the tangent plane to the surface. Since N, the unit normal to the surface, is orthogonal
to the tangent plane, the vector fields t and N(u, v) are orthogonal and
 
t(s), N u(s), v(s) , B(s) := t(s) × N u(s), v(s)

is a right-handed orthonormal basis of R3 at every point s ∈ I. Using this basis to


express the derivative of the tangent vector field, and recalling that t′ · t ≡ 0, we see that

t′ = (t′ · t)t + t′ · N(u, v) N(u, v) + (t′ · B)B = κn N(u, v) + κg B, (2.9)

where the scalar field


κn := t′ · N(u, v)
is the normal curvature of the curve γ and the scalar field

κg := t′ · B = [t′ , t, N(u, v)]

is the geodesic curvature of γ. Note that, by Pythagoras’s theorem,

κ2 = κ2n + κ2g ,

where κ := |t′| is the curvature of γ.

Warning. Pressley defines geodesic curvature [6, p.127] with the opposite sign to us.
For some reason, the most common convention in the literature is to work with the left-
handed orthonormal triple t, B, N(u, v) rather than the right-handed choice made
above, which we prefer for its consistency with the Serret–Frenet apparatus.

38
2.2. Geodesics

Exercise 2.13. Suppose γ = r ◦ (u, v) : I → R3 is a unit-speed curve with strictly


positive curvature, which lies in theregular surface patch r : U → R3 . Let ψ(s) denote
the angle from n(s) to N u(s), v(s) for all s ∈ I, so that

N(u, v) = n cos ψ + b sin ψ.

(The existence of such a smooth function ψ follows from Lemma 3.6.) Show that the
normal curvature κn = κ cos ψ, the surface binormal

B = −n sin ψ + b cos ψ

and the geodesic curvature κg = −κ sin ψ. Deduce that

t′ = κn N(u, v) + κg B,
N(u, v)′ = −κn t + τg B
and B′ = −κg t − τg N(u, v)

where τg := τ + ψ ′ is the geodesic torsion. (This is an analogue of the Serret–Frenet


equations for the orthonormal triple of vector fields t, N(u, v), B .)
Geodesics
A particle which is free to move in R3 and which has zero acceleration will either remain
at rest or travel with constant velocity in a straight line: this is Newton’s first law of
motion. However, if the particle is required to move upon some surface patch, then it
may be necessary for the particle to accelerate in order that it stays in contact with the
surface. As long as there is no component of acceleration in the tangent plane (so that,
infinitesimally, the particle does not appear to change velocity at all) then the particle
is accelerating as little as possible, subject to the constraint of remaining on the surface.
The curve traced out by such a particle is called a geodesic.
Formally, the curve γ = r ◦ (u, v) : I → R3 in the surface patch r is a geodesic if and
only if
 
γ ′′ (t) · ∂1 r u(t), v(t) = γ ′′ (t) · ∂2 r u(t), v(t) = 0 for all t ∈ I.

Equivalently, γ is a geodesic if and only if γ ′′ is a scalar multiple of N everywhere.

Exercise 2.14. Suppose the unit-speed curve γ : I → R3 lies on a straight line. Prove
that γ is a geodesic.

39
2. Surfaces – local theory

Example 2.15. The parameterisation of the unit sphere S 2 given in Example 2.5,

r : (−π/2, π/2) × (−π, π) → R3 ; (θ, φ) 7→ (cos θ cos φ, cos θ sin φ, sin θ),

has tangent vectors

∂1 r(θ, φ) = (− sin θ cos φ, − sin θ sin φ, cos θ)


and ∂2 r(θ, φ) = (− cos θ sin φ, cos θ cos φ, 0).

If γ : (−π, π) → R3 ; s 7→ r(0, s) is the equator, so that u ≡ 0 and v(s) = s, then

γ(s) = (cos s, sin s, 0),


γ ′ (s) = (− sin s, cos s, 0)
and γ ′′ (s) = (− cos s, − sin s, 0) = −γ(s)

for all s ∈ (−π, π). Hence

γ ′′ (s) · ∂1 r(0, s) = (− cos s, − sin s, 0) · (0, 0, 1) = 0

and
γ ′′ (s) · ∂2 r(0, s) = (− cos s, − sin s, 0) · (− sin s, cos s, 0) = 0,
which shows that γ is a geodesic.
There is nothing particularly special about the equator; any great circle, the intersection
of the sphere with a plane through its centre, is a geodesic when parameterised by arc
length. The converse is also true; see the following exercise.

Exercise 2.15. Let r : U → R3 be a regular surface patch contained in the unit


sphere S 2 . Prove that the unit normal vector field N is such that N = ±r. Deduce that
a unit-speed curve γ = r ◦ (u, v) : I → S 2 is a geodesic if and only if γ ′′ + γ ≡ 0, and so

γ(s) = a cos s + b sin s

for all s ∈ I, where a and b are orthogonal unit vectors. [Thus γ is part of the great
circle in the plane spanned by a and b.]

Proposition 2.16. Any geodesic has constant speed and so a very simple unit-speed
reparameterisation.

Proof. If γ = r ◦ (u, v) : I → R3 is a geodesic then

d ′ 2 d ′ 
|γ (t)| = γ (t) · γ ′ (t) = 2γ ′ (t) · γ ′′ (t) = 0 for all t ∈ I,
dt dt
since γ ′ lies in the tangent plane. Hence |γ ′ | is constant.

40
2.2. Geodesics

Suppose γ = r ◦ (u, v) is a unit-speed geodesic. Then γ ′′ = t′ is proportional to the unit


normal to the surface N(u, v), hence its geodesic curvature

κg = t′ · B = [t′ , t, N(u, v)] ≡ 0.

The converse is clear from (2.9) and so we have the following proposition, which explains
why κg bears the name it does.

Proposition 2.17. A unit-speed curve in a surface is a geodesic if and only if it has


zero geodesic curvature everywhere: κg ≡ 0.

Exercise 2.16. Show that the torsion τ of a unit-speed geodesic γ with strictly positive
curvature is equal to its geodesic torsion τg , defined in Exercise 2.13.
The geodesic equations
The smooth curve γ = r ◦ (u, v) is a geodesic if and only if

γ ′′ · ∂1 r(u, v) = γ ′′ · ∂2 r(u, v) = 0

everywhere on γ. Now, (v · w)′ = v′ · w + v · w′ for any smooth vector fields v and w, so


′ ′
γ ′ · ∂1 r(u, v) = γ ′′ · ∂1 r(u, v) + γ ′ · ∂1 r(u, v)

and rearranging this shows that


′ ′
γ ′′ · ∂1 r(u, v) = γ ′ · ∂1 r(u, v) − γ ′ · ∂1 r(u, v) .

Since γ ′ = ∂1 r(u, v) u′ + ∂2 r(u, v) v ′, it follows that

γ ′′ · ∂1 r(u, v)
′
= (∂1 r · ∂1 r)(u, v) u′ + (∂2 r · ∂1 r)(u, v) v ′
− (∂1 r(u, v) u′ + ∂2 r(u, v) v ′) · (∂11 r(u, v) u′ + ∂21 r(u, v) v ′)
= (E(u, v) u′ + F (u, v) v ′)′

− (∂1 r · ∂11 r)(u, v)(u′)2 + (∂2 r · ∂11 r + ∂1 r · ∂21 r)(u, v)u′v ′ + (∂2 r · ∂21 r)(u, v)(v ′)2 ,
(2.10)

where ∂11 r := ∂1 (∂1 r) et cetera:

∂2r ∂r ∂2r
∂11 r(x, y) = (x, y), ∂21 r(x, y) = (x, y) and ∂22 r(x, y) = 2 (x, y).
∂x2 ∂y∂x ∂y
Moreover,

∂1 E = ∂1 (∂1 r · ∂1 r) = ∂11 r · ∂1 r + ∂1 r · ∂11 r = 2∂1 r · ∂11 r,


∂1 F = ∂1 (∂1 r · ∂2 r) = ∂11 r · ∂2 r + ∂1 r · ∂12 r = ∂2 r · ∂11 r + ∂1 r · ∂21 r
and ∂1 G = ∂1 (∂2 r · ∂2 r) = ∂12 r · ∂2 r + ∂2 r · ∂12 r = 2∂2 r · ∂21 r.

41
2. Surfaces – local theory

Applying these to (2.10) yields the identity

γ ′′ · ∂1 r(u, v)
= (E(u, v) u′ + F (u, v) v ′)′ − 21 (∂1 E(u, v)(u′)2 + 2∂1 F (u, v)u′v ′ + ∂1 G(u, v)(v ′)2 );

similar working gives the equation

γ ′′ · ∂2 r(u, v)
= (F (u, v) u′ + G(u, v) v ′)′ − 21 (∂2 E(u, v)(u′)2 + 2∂2 F (u, v)u′v ′ + ∂2 G(u, v)(v ′)2 ).

Hence the smooth curve γ = r ◦ (u, v) is a geodesic if and only if the geodesic equations
are satisfied:

(E(u, v)u′ + F (u, v)v ′)′ = 21 (∂1 E(u, v)(u′)2 + 2∂1 F (u, v)u′v ′ + ∂1 G(u, v)(v ′)2 ) (2.11)

and

(F (u, v)u′ + G(u, v)v ′)′ = 21 (∂2 E(u, v)(u′)2 + 2∂2 F (u, v)u′v ′ + ∂2 G(u, v)(v ′)2 ). (2.12)

These equations are usually very hard to solve explicitly.

Example 2.18. For the plane

r : R2 → R3 ; (x, y) 7→ (x, y, 0),

the first fundamental form has coefficients E = G ≡ 1 and F ≡ 0. The geodesic


equations become
u′′ = 0 and v ′′ = 0,
which have the general solutions u = at + b and v = ct + d for constants a, b, c, d ∈ R.
Hence γ = r ◦ (u, v) has the form

γ : t 7→ vt + w,

where v = (a, c, 0) and w = (b, d, 0). In other words, for the plane, geodesics and
straight lines are the same thing.

42
2.2. Geodesics

Example 2.19. The cylinder

r : R × (−π, π) → R3 ; (z, φ) 7→ (cos φ, sin φ, z)

has first fundamental form dz 2 + dφ2 , the same as for the plane. As in Example 2.18
above, the geodesic equations become u′′ = 0 and v ′′ = 0, so u(t) = at+b and v(t) = ct+d
for constants a, b, c and d. In this case, the geodesic γ is such that

γ(t) = r u(t), v(t) = r(at + b, ct + d) = (cos(ct + d), sin(ct + d), at + b)

and the geodesics on the cylinder fall into three classes:

(i) if a = 0 and c 6= 0 then γ lies on the parallel r(b, ·) : φ 7→ r(b, φ), a circle parallel
to the x-y plane;

(ii) if c = 0 and a 6= 0 then γ lies on the meridian r(·, d) : z 7→ r(z, d), a straight line
parallel to the z axis;

(iii) if a 6= 0 and c 6= 0 then γ describes a helix .

(If a = c = 0 then γ is not a curve.)

43
2. Surfaces – local theory

Geodesics and surfaces of revolution


Let 
γ : I → R3 ; s 7→ f (s), 0, g(s)
be a simple unit-speed curve in the x-z plane, such that f (s) > 0 for all s ∈ I. As γ has
unit speed, if s ∈ I then

1 = |γ ′ (s)|2 = | f ′ (s), 0, g ′ (s) |2 = f ′ (s)2 + g ′ (s)2 .

A surface of revolution is formed by rotating this curve about the z axis; it may be
parameterised in the following fashion:

r : I × (−π, π) → R3 ; (s, φ) 7→ f (s) cos φ, f (s) sin φ, g(s) . (2.13)

The coordinate s corresponds to the point γ(s) on the curve γ, and the coordinate φ
describes the angle through which this point has been rotated.

Exercise 2.17. Verify that the map r is injective.

Exercise 2.18. Verify that the surface of revolution r given by (2.13) is regular.

Since 
∂1 r(s, φ) = f ′ (s) cos φ, f ′ (s) sin φ, g ′ (s)
and 
∂2 r(s, φ) = −f (s) sin φ, f (s) cos φ, 0 ,
the coefficients of the first fundamental form are

E(s, φ) = f ′ (s)2 + g ′(s)2 = 1, F (s, φ) = 0 and G(s, φ) = f (s)2 .

44
2.2. Geodesics

Exercise 2.19. Show that the geodesic equations for the surface of revolution r given
by (2.13) have the form

u′′ = f (u)f ′ (u)(v ′)2 and f (u)2 v ′ )′ = 0. (2.14)

Meridians and parallels


A meridian on the surface of revolution r is a curve of the form

r(·, φ0) : I → R3 ; s 7→ r(s, φ0 ) = f (s) cos φ0 , f (s) sin φ0 , g(s) ,

where φ0 ∈ (−π, π) is constant; it is a copy of the curve γ which generates r, rotated


through an angle of φ0 radians about the z axis.
Similarly, a parallel of the surface of revolution r has the form

r(s0 , ·) : (−π, π) 7→ R3 ; φ 7→ r(s0 , φ) = f (s0 ) cos φ, f (s0 ) sin φ, g(s0) ,

where s0 ∈ I is constant. This is a circle (minus one point) parallel to the x-y plane,
formed by rotating the point γ(s0 ) through 2π radians.

Exercise 2.20. Use Exercise 2.19 to prove that the meridian r(·, φ0) is a geodesic for
any φ0 ∈ (−π, π), but a parallel r(s0 , ·) is a geodesic if and only if f ′ (s0 ) = 0.

45
2. Surfaces – local theory

Geodesics and isometries


A unit-speed curve γ = r ◦ (u, v) in the surface patch r is a geodesic if and only if its
geodesic curvature κg := t′ · B = 0. However,

B := t × N(u, v)
∂1 r × ∂2 r
= (∂1 r(u, v) u′ + ∂2 r(u, v) v ′) × (u, v)
|∂1 r × ∂2 r|
 
(∂1 r · ∂2 r)∂1 r − (∂1 r · ∂1 r)∂2 r (u, v) u′ + (∂2 r · ∂2 r)∂1 r − (∂2 r · ∂1 r)∂2 r (u, v) v ′
= p
|∂1 r|2 |∂2 r|2 − (∂1 r · ∂2 r)2 (u, v)
(F (u, v)u′ + G(u, v)v ′)∂1 r(u, v) − (E(u, v)u′ + F (u, v)v ′)∂2 r(u, v)
= √ ,
EG − F 2 (u, v)
since, by Exercises 1.9 and 1.8 in Section 1.1,

u × (v × w) = (u · w)v − (u · v)w for all u, v, w ∈ R3

and
|u × v|2 = |u|2 |v|2 − (u · v)2 for all u, v ∈ R3 .
Moreover, the identity t = ∂1 r(u, v) u′ + ∂2 r(u, v) v ′ implies that

t′ = ∂11 r(u, v)(u′)2 + 2∂12 r(u, v) u′v ′ + ∂22 r(u, v)(v ′)2 + ∂1 r(u, v) u′′ + ∂2 r(u, v) v ′′

and therefore EG − F 2 (u, v) κg equals

(F (u, v)u′ + G(u, v)v ′)


× (∂1 r · ∂11 r)(u, v)(u′)2 + 2(∂1 r · ∂12 r)(u, v) u′v ′ + (∂1 r · ∂22 r)(u, v)(v ′)2

+ E(u, v) u′′ + F (u, v) v ′′
− (E(u, v) u′ + F (u, v) v ′)
× (∂2 r · ∂11 r)(u, v)(u′)2 + 2(∂2 r · ∂12 r)(u, v) u′v ′ + (∂2 r · ∂22 r)(u, v)(v ′)2

+ F (u, v) u′′ + G(u, v) v ′′
= (u′ )3 (F ∂1 r · ∂11 r − E ∂2 r · ∂11 r)(u, v)
+ (u′ )2 v ′ (G ∂1 r · ∂11 r + 2F ∂1 r · ∂12 r − 2E ∂2 r · ∂12 r − F ∂2 r · ∂11 r)(u, v)
+ u′ (v ′ )2 (F ∂1 r · ∂22 r + 2G ∂1 r · ∂12 r − 2F ∂2 r · ∂12 r − E ∂2 r · ∂22 r)(u, v)
+ (v ′ )3 (G ∂1 r · ∂22 − F ∂2 r · ∂22 r)(u, v)
+ (EG − F 2 )(u, v)(u′′v ′ − u′ v ′′ ).

Since

∂1 E = 2∂1 r · ∂11 r, ∂1 F = ∂1 r · ∂12 r + ∂2 r · ∂11 r, ∂1 G = 2∂2 r · ∂12 r (2.15)

and

∂2 E = 2∂1 r · ∂12 r, ∂2 F = ∂1 r · ∂22 r + ∂2 r · ∂12 r, ∂2 G = 2∂2 r · ∂22 r, (2.16)

46
2.2. Geodesics

it follows that

κg = A(u, v)(u′)3 + B(u, v)(u′)2 v ′ + C(u, v)u′(v ′ )2 + D(u, v)(v ′)3



+ EG − F 2 (u, v)(u′′v ′ − u′ v ′′ ),

where
1
2
F ∂1 E − E∂1 F + 12 E∂2 E 1
2
G∂1 E + 32 F ∂2 E − E∂1 G − F ∂1 F
A= √ , B= √ ,
EG − F 2 EG − F 2
F ∂2 F − 32 F ∂1 G + G∂2 E − 12 E∂2 G G∂2 F − 21 G∂1 G − 12 F ∂2 G
C= √ and D = √ .
EG − F 2 EG − F 2
The precise form of A, B, C and D is unimportant; the key thing to observe is that
geodesic curvature depends only on the coefficients of the first fundamental form and
their partial derivatives. Consequently, geodesics are preserved by isometry.

Theorem 2.20. If the surface patches r1 : U → R3 and r2 : U → R3 are isometric then,


for any smooth pair (u, v) : I → U, the curve γ 1 := r1 ◦ (u, v) is a geodesic in r1 if and
only if γ 2 := r2 ◦ (u, v) is a geodesic in r2 .

Exercise 2.21. Use the geodesic equations to give another proof of Theorem 2.20, that
that geodesics are preserved by isometry.

This leads us to consider another characterisation of geodesics, in terms of length.


Curves of extremal length
Let a and b be distinct points in the regular surface patch r and suppose γ : I → R3 is
a regular curve in r which passes through a and b. There there exist a, b ∈ I such that
γ(a) = a and γ(b) = b; without loss of generality, suppose a < b. The curve γ|(a,b) is a
geodesic if and only if its length is an extremum of the lengths of all curves between the
points a and b, in the following sense.
Let 
γ · : (−ε, ε) × I → R3 ; (c, t) 7→ γ c (t) := r uc (t), vc (t)
be a family of smooth curves in r : U → R3 which depends smoothly on the parameter c,
i.e.,

u· : (−ε, ε) × I → U; (c, t) 7→ uc (t) and v· : (−ε, ε) × I → U; (c, t) 7→ vc (t)

have partial derivatives of all orders. Suppose γ c (a) = a and γ c (b) = b for all c ∈ (−ε, ε),
and suppose also that γ 0 = γ.

47
2. Surfaces – local theory

The length of γ c is
Z bp
l(c) = E(uc , vc )(u′c )2 + 2F (uc , vc )u′c vc′ + G(uc , vc )(vc′ )2 dt
a

for all c ∈ (−ε, ε), by (2.2).

Theorem 2.21. The curve γ|(a,b) is a geodesic if and only if l′ (0) = 0 for every family
of smooth curves γ · as above.

Proof. This is a little involved so we omit it; details can be found in [6, Chapter 8].

If γ is the shortest path between a and b then it is a geodesic, since l′ (0) = 0 in this
case. The converse does not hold: a geodesic must be a stationary point for length, but
it need not be a minimum. For example, if a and b are points on a sphere then the great
circle on which they lie (the intersection of the sphere with the plane through a, b and
the origin) gives two paths from a to b; unless these points are antipodal (that is, they
lie on a straight line in R3 through the origin) one of these paths will be longer than the
other, but both are geodesics.
Furthermore, there may not exist a shortest path between two points. To see this,
consider the punctured plane P∗ = {(x, y, 0) : x, y ∈ R} \ {(0, 0, 0)}. If a = (−1, 0, 0)
and b = (1, 0, 0) then any path from a to b has length at least 2, and there is no path
of length 2 between a and b in P∗ . (Think of the origin as a deep hole which must be
avoided by anyone walking from a to b.)

Exercise 2.22. Show that any regular curve from a = (−1, 0, 0) to b = (1, 0, 0) which
lies in P∗ = (R2 \ {(0, 0)}) × {0} has length strictly greater than two, but if ε > 0 then
there exists a smooth curve from a to b of length less than 2 + ε.

48
2.3. The second fundamental form

2.3 The second fundamental form


If γ = r ◦ (u, v) is a unit-speed curve in the regular surface patch r then the tangent
vector field

t := γ ′ = ∂1 r(u, v) u′ + ∂2 r(u, v) v ′ (2.17)

and

t′ = ∂11 r(u, v)(u′)2 + ∂21 r(u, v) u′v ′ + ∂1 r(u, v) u′′


+ ∂12 r(u, v) u′v ′ + ∂22 r(u, v)(v ′)2 + ∂2 r(u, v) v ′′
= ∂11 r(u, v)(u′)2 + 2∂12 r(u, v) u′v ′ + ∂22 r(u, v)(v ′)2 + ∂1 r(u, v) u′′ + ∂2 r(u, v) v ′′.

It follows that

κn := t′ · N(u, v) = (∂11 r · N)(u, v)(u′)2 + 2(∂12 r · N)(u, v)u′v ′ + (∂22 r · N)(u, v)(v ′)2
= L(u, v)(u′)2 + 2M(u, v)u′ v ′ + N(u, v)(v ′)2 , (2.18)

where L, M and N are the coefficients of the second fundamental form,

L(u, v) du2 + 2M(u, v) du dv + N(u, v) dv 2, (2.19)

with

L := ∂11 r · N, M := ∂12 r · N and N := ∂22 r · N. (2.20)

The normal curvature κn measures how much the tangent vector t is moving out of
the tangent plane as we move along γ. More generally, the second fundamental form
gives a measure of how much the surface r deviates from its tangent planes. (The first
fundamental form governs the local intrinsic geometry of the surface: what can be seen
to first order, by looking at the tangent plane. The second fundamental form, on the
other hand, is extrinsic: it depends on how the surface sits inside R3 . The plane and
the cylinder have the same first fundamental form, but their second fundamental forms
differ.)

49
2. Surfaces – local theory

Example 2.22. Let r : U → R3 be a regular surface patch with vanishing second


fundamental form: L = M = N ≡ 0. Show that ∂1 N = ∂2 N ≡ 0. Deduce that if U is
an open rectangle then N is constant and r lies in a plane.
Since ∂1 r · N ≡ 0 and L ≡ 0, it follows that

0 ≡ ∂1 (∂1 r · N) = ∂11 r · N + ∂1 r · ∂1 N = L + ∂1 r · ∂1 N = ∂1 r · ∂1 N.

Similarly, the facts that ∂2 r · N ≡ 0 and M ≡ 0 imply that ∂2 r · ∂1 N ≡ 0. Thus ∂1 N is


everywhere orthogonal to the tangent plane, so is a scalar multiple of the unit normal N.
However, N is a unit vector, so

0 ≡ ∂1 (N · N) = ∂1 N · N + N · ∂1 N = 2N · ∂1 N

and therefore ∂1 N ≡ 0. Similar working shows that ∂2 N ≡ 0.


As ∂1 N = ∂2 N ≡ 0 and any two points in U can be connected by a path made up of a
horizontal line segment and a vertical line segment, the function N is constant. (This is
the two-variable version of a very familiar result: a function which has zero derivative
on some interval is constant on that interval.) Furthermore,

∂1 (r · N) = ∂1 r · N + r · ∂1 N ≡ 0 and ∂2 (r · N) = ∂2 r · N + r · ∂2 N ≡ 0,

so r · N is also constant (for the same reason); let r · N ≡ d. This working shows that r
lies in the plane {v ∈ R3 : v · N = d}.

In Example 2.22, the restriction that U is an open rectangle may be dropped; the result
holds for any connected open set U.

r = r ◦ Φ, where Φ is a reparameterisation of the regular surface


Exercise 2.23. Let e
patch r. Show that
" # " #
e M
L f L ◦ Φ M ◦ Φ
= ±(JΦ)t JΦ,
f
M N e M ◦Φ N ◦Φ

where the positive sign is taken if Φ preserves orientation and the negative sign if Φ is
orientation reversing, and JΦ is the Jacobian of Φ: see p.32.

50
2.3. The second fundamental form

Principal curvatures
Let r : U → R3 be a regular surface patch; throughout this section, for clarity, we suppose
the point (x, y) ∈ U to be fixed and omit mentioning it explicitly: ∂1 r = ∂1 r(x, y) et
cetera.

If

t1 = λ1 ∂1 r + µ1 ∂2 r and t2 = λ2 ∂1 r + µ2 ∂2 r (2.21)

are any two tangent vectors then their scalar product t1 · t2 equals
h i 
  E F λ2
λ1 λ2 ∂1 r · ∂1 r + λ1 µ2 ∂1 r · ∂2 r + µ1 λ2 ∂2 r · ∂1 r + µ1 µ2 ∂2 r · ∂2 r = λ1 µ 1 ;
F G µ2

with respect to the ordered basis (∂1 r, ∂2 r), the matrix comprising the coefficients of the
first fundamental form corresponds to the inner product on the tangent plane T(x,y) r.
Hence
  h i 
λ   E F λ
QI : T(x,y) r → R; λ ∂1 r + µ ∂2 r = 7→ λ µ
µ F G µ

is a quadratic form on the tangent plane which gives the squared magnitude of a tangent
vector. Similarly, the quadratic form
  h i 
λ   L M λ
QII : T(x,y) r → R; λ ∂1 r + µ ∂2 r = 7→ λ µ
µ M N µ

gives the normal curvature κn of a unit-speed curve through r(x, y) which has the tangent
vector λ ∂1 r(x, y) + µ ∂2r(x, y) at that point, by (2.17) and (2.18). Using diagonalisation,
we will find the maximum and minimum values of κn over all such curves, that is, the
extreme values of QII under the constraint that QI ≡ 1.

51
2. Surfaces – local theory

Suppose the tangent vectors t1 and t2 from (2.21) are chosen so that {t1 , t2 } is an
orthonormal basis of the tangent plane T(x,y) r. If
h i
λ1 λ2
A :=
µ1 µ2
then
h i h i h i
E F t1 ·t1 t1 ·t2 1 0
At A= = = I; (2.22)
F G t2 ·t1 t2 ·t2 0 1
in particular, the matrix A is invertible. Since
 h i t h it h i
L M L M L M
At A = At (At )t = At A,
M N M N M N
the principal-axis theorem (Theorem A.8) yields a real 2×2 matrix B which is orthogonal
(that is, B t B = BB t = I) and such that
h i h i
L M κ1 0
B t At AB = , (2.23)
M N 0 κ2
where κ1 and κ2 are real numbers with κ1 6 κ2 .
Furthermore, by (2.22), it follows that
h i
t t E F
BA AB = B t IB = B t B = I; (2.24)
F G
the matrix C := AB simultaneously diagonalises the matrices corresponding to the first
and the second fundamental forms. Let the linear transformation
   
λ λ
TC : T(x,y) r → T(x,y) r; λ ∂1 r + µ ∂2 r = 7→ C = λC ∂1 r + µC ∂2 r
µ µ
and note that TC is invertible because the matrix C is. If s = λ ∂1 r + µ ∂2r is any tangent
vector then
h i    
2
  t E F λ   λ
|TC s| = (TC s) · (TC s) = λ µ C C = λ µ = λ2 + µ 2 ,
F G µ µ
so
{unit vectors in T(x,y) r} = {TC s : λ2 + µ2 = 1}.
Furthermore, the value of the second fundamental form at TC s is
h i   h i 
  t L M λ   κ1 0 λ
QII (TC s) = λ µ C C = λ µ = κ1 λ2 + κ2 µ2 .
M N µ 0 κ2 µ
Hence
{QII (t) : t is a unit tangent vector} = {κ1 λ2 + κ2 µ2 : λ2 + µ2 = 1}
= {κ1 x + κ2 (1 − x) : 0 6 x 6 1} = [κ1 , κ2 ]
and κ1 and κ2 are the principal curvatures, scalar fields on r giving the minimum and
maximum values of the normal curvature κn .

52
2.3. The second fundamental form

The corresponding principal directions are unit tangent vectors fields, t1 and t2 , on r
corresponding to the directions in which the minimum and maximum values of the
normal curvature, κ1 and κ2 , are attained, so that QII (t1 ) = κ1 and QII (t2 ) = κ2 . A
point where κ1 = κ2 is called an umbilic, and every unit tangent vector is principal at
such a point.
Suppose the principal curvatures are distinct: κ1 6= κ2 . If t = TC s is a unit tangent
vector, where s = λ ∂1 r + µ ∂2 r, then λ2 + µ2 = QI (t) = 1 and

QII (t) = κ1 ⇐⇒ κ1 λ2 + κ2 µ2 = κ1
⇐⇒ κ1 λ2 + κ2 (1 − λ2 ) = κ1
⇐⇒ (κ1 − κ2 )λ2 = κ1 − κ2
⇐⇒ λ2 = 1.

Hence t is a principal direction corresponding to κ1 if and only if t = ±TC ∂1 r, and


similar working shows that t is a principal direction corresponding to κ2 if and only
if t = ±TC ∂2 r. In particular, up to a choice of sign, the principal directions are unique
wherever the principal curvatures are distinct.

Exercise 2.24. Suppose the principal curvatures κ1 and κ2 are distinct at some point.
Show that the principal directions t1 and t2 are orthogonal there.

Since the matrix C was obtained non-constructively, with the assistance of the principal-
axis theorem, there remains the question of how to calculate in practice the principal
curvatures and principal directions. We proceed as follows. h i
L M
By (2.23), the principal curvatures κ1 and κ2 are the eigenvalues of C t C.
M N
Moreover, from (2.24) and the fact that C is invertible,
 h i    h i h i 
L M E F L M 
det κI − C t C = 0 ⇐⇒ det C t κ − C =0
M N F G M N
 h i h i
E F L M
⇐⇒ det κ − = 0.
F G M N

Thus the principal curvatures κ1 and κ2 may be calculated in a straightforward manner,


as the roots of the quadratic equation
" #
κE − L κF − M
det = 0. (2.25)
κF − M κG − N

53
2. Surfaces – local theory

To find the principal directions, note that if

s = λ ∂1 r + µ ∂2 r and t = TC s = λC ∂1 r + µC ∂2 r

then
" #   
κE − L κF − M λC 0
=
κF − M κG − N µC 0
 h i h
i    
t L M E F λ 0
⇐⇒ C − κ C =
M N F G µ 0
h i   
κ−κ1 0 λ 0
⇐⇒ =
0 κ−κ2 µ 0
   
(κ1 − κ)λ 0
⇐⇒ = .
(κ2 − κ)µ 0

Recall that t = TC s is a principal direction corresponding to κ1 if and only if λ = ±1


and µ = 0. From the previous working, this is equivalent to
    
κ1 E − L κ1 F − M λC 0
= ,
κ1 F − M κ1 G − N µC 0

and a similar result holds for κ2 .


To summarise, the tangent vector t = z ∂1 r + w ∂2 r is a principal direction corresponding
to κi (where i = 1 or i = 2) if and only if
" #   
κi E − L κi F − M z 0
= , (2.26)
κi F − M κi G − N w 0

and the principal directions may be found by solving this equation, with the constraint
that Ez 2 + 2F zw + Gw 2 = 1.

Exercise 2.25. Let e r = r ◦ Φ, where Φ is a reparameterisation of the regular surface


patch r. Prove that the principal curvatures κ
e1 and e r are equal to κ1 ◦ Φ and κ2 ◦ Φ,
κ2 of e
where κ1 and κ2 are the principal curvatures of r.

54
2.3. The second fundamental form

Exercise 2.26. Let E, F , G and L, M, N be constants. Use the method of Lagrange


multipliers to show that the extreme values of

κ(x, y) := Lx2 + 2Mxy + Ny 2 ,

subject to the constraint

f (x, y) := Ex2 + 2F xy + Gy 2 = 1,

satisfy  
κE − L κF − M
det = 0.
κF − M κG − N
[Hint: let 
F (x, y) := κ(x, y) − λ f (x, y) − 1
∂F ∂F
and show first that λ = κ if F is stationary, so that = = 0, and f (x, y) = 1.]
∂x ∂y

Example 2.23. The sphere of radius a > 0, minus the poles and one meridian, has the
parameterisation

r : (−π/2, π/2) × (−π, π) → R3 ; (θ, φ) 7→ (a cos θ cos φ, a cos θ sin φ, a sin θ).

(When a = 1, this is the parameterisation of Example 2.5.) This surface has first
fundamental form a2 dθ2 +a2 cos2 θ dφ2 and second fundamental form a dθ2 +a cos2 θ dφ2 ,
so the principal curvatures are the roots of the equation
 2 
κa − a 0
0 = det = a2 (κa − 1)2 cos2 θ.
0 κa2 cos2 θ − a cos2 θ

Hence the principal curvatures are equal and have the constant value 1/a. In particular,
every point is an umbilic and every tangent vector is a principal direction, as we should
expect: at each point, the sphere curves equally in every direction.

55
2. Surfaces – local theory

Example 2.24. Fix a > 0. The cylinder

r : R × (−π, π) → R3 ; (z, φ) 7→ (a cos φ, a sin φ, z)

has first fundamental form dz 2 + a2 dφ2 and second fundamental form a dφ2 . Hence the
principal curvatures are the roots of
" #
κ 0
det = 0 ⇐⇒ κ(a2 κ − a) = 0,
0 a2 κ − a

so κ1 ≡ 0 and κ2 ≡ 1/a. If we visualise the situation, the minimum deviation from the
tangent plane should occur along a meridian (as this is a straight line in the cylinder, so
lies in the tangent plane) and the maximum along a parallel.
To see if this conjecture holds, we solve the equations
" #    " #   
0 0 λ1 0 1/a 0 λ2 0
= and =
0 −a µ 1 0 0 0 µ 2 0

subject to the constraints λ2i + a2 µ2i = 1 for i = 1 and i = 2.


The first yields (λ1 , µ1 ) = (±1, 0) and the second yields (λ2 , µ2) = (0, ±1/a), so the
principal directions are

t1 (z, φ) = ±∂1 r(z, φ) = (0, 0, ±1)

and
1
t2 (z, φ) = ± ∂2 r(z, φ) = ±(− sin φ, cos φ, 0).
a
It is an exercise to verify that the meridian and the parallel through r(z, φ) have tangent
vectors which are proportional to these, and so show that our conjecture is correct.

Exercise 2.27. Do isometric surfaces have the same principal curvatures? [Hint:
consider a plane and a cylinder.]

56
2.4. Gaussian curvature

2.4 Gaussian curvature

Definition 2.25. The Gaussian curvature of a regular surface patch r : U → R3 is the


scalar field K := κ1 κ2 , the product of the principal curvatures.

If the surface r : U → R3 has Gaussian curvature K(x, y) > 0 at some point (x, y) ∈ U
then the principal curvatures κ1 (x, y) and κ2 (x, y) have the same sign, so the surface
near to r(x, y) is convex or concave, lying on one side of the tangent plane T(x,y) r.
If K(x, y) < 0 then the principal curvatures have different signs and r(x, y) is a saddle
point, where the surface crosses the tangent plane.

Theorem 2.26. The Gaussian curvature


LN − M 2
K= , (2.27)
EG − F 2
where E, F , G and L, M, N are the coefficients of the first and second fundamental
forms, respectively.

Proof. If b and c are the roots of a quadratic polynomial then

a(x − b)(x − c) = ax2 − a(b + c)x + abc = a2 x2 + a1 x + a0 ,

so their product bc equals a0 /a2 . Since the principal curvatures are roots of the quadratic
 h i h i  
E F L M κE − L κF − M
det κ − = det
F G M N κF − M κG − N
= (κE − L)(κG − N) − (κF − M)2

= (EG − F 2 )κ2 − (EN + LG − 2F M)κ + LN − M 2 ,

it follows that
K = κ1 κ2 = (LN − M 2 )/(EG − F 2 ).

Exercise 2.28. Why is EG − F 2 > 0 at all points in a regular surface patch?

57
2. Surfaces – local theory

Example 2.27. For the sphere with radius a > 0, parameterised as in Example 2.23,
the formula (2.27) becomes

a2 cos2 θ 1
K(θ, φ) = 4 2
= 2,
a cos θ a
which agrees with the values of the principal curvatures found in that example.

An important property of Gaussian curvature is its invariance under reparameterisation.

Exercise 2.29. Let er = r ◦ Φ, where Φ is a reparameterisation of the regular surface


patch r. Prove that the Gaussian curvature Ke of e
r is equal to K ◦ Φ, where K is the
Gaussian curvature of r.

The Gauss map


If r : U → R3 is a regular surface patch then the unit normal vector field N is a smooth
map from U to the unit sphere S 2 . When thought of in this way, N is called the Gauss
map. As the point (x, y) varies over U, the vector r(x, y) traces out the surface r(U)
and N(x, y) traces out a region on the sphere.

Exercise 2.30. Explain why the vectors ∂1 N(x, y) and ∂2 N(x, y) lie in the tangent
plane T(x,y) r for all (x, y) ∈ U, so that

∂1 N = A ∂1 r + B ∂2 r and ∂2 N = C ∂1 r + D ∂2 r,

where A, B, C and D are scalar fields.


By considering the partial derivatives of ∂1 r · N and ∂2 r · N, prove that

∂1 r · ∂1 N = −L, ∂1 r · ∂2 N = ∂2 r · ∂1 N = −M and ∂2 r · ∂2 N = −N.

Deduce that the Weingarten matrix


   −1  
−A −C E F L M
= .
−B −D F G M N

Hence show that ∂1 N × ∂2 N = K ∂1 r × ∂2 r, where K is the Gaussian curvature.

58
2.4. Gaussian curvature

Exercise 2.31. Explain why [N, ∂1 N, ∂2 N] = ±|∂1 N × ∂2 N|.

Exercise 2.32. Let Dε := D(z, w; ε) ⊆ U be the open disc of radius ε and centre (z, w).
Explain why the area traced out on the sphere by N(x, y) as (x, y) varies over Dε is
ZZ
AN (Dε ) = |(∂1 N × ∂2 N)(x, y)| dx dy.

We let the signed area


ZZ
sAN (Dε ) = [N, ∂1 N, ∂2 N](x, y) dx dy;

Exercise 2.31 shows why this is a sensible definition. By using the fact that
ZZ
1
lim f (x, y) dx dy = f (z, w)
ε→0 πε2 Dε

for any continuous function f : U → R, show that


sAN (Dε )
lim
ε→0 Ar (Dε )

exists and equals K(z, w), where Ar (Dε ) is the area of the region on r with coordinates
in Dε , as defined by (2.5). [Use the usual “rules for limits”.]
Theorema Egregium
Although it appears from Theorem 2.26 that Gaussian curvature depends on both the
first and the second fundamental forms, this is not in fact the case: Gaussian curvature
depends only on the first fundamental form and its derivatives. This result is Gauss’s
Theorema Egregium (the Latin for “remarkable theorem”).

Theorem 2.28. (Theorema Egregium) If the regular surface patches r1 : U → R3


and r2 : U → R3 are isometric then their Gaussian curvatures are equal.

The key step in the proof of Theorem 2.28 is the following lemma, which is also a vital
component in the proof of the Gauss–Bonnet Theorem: see Theorem 3.13.

Lemma 2.29. Let (e, f, N) be a orthonormal triple of vector fields on the regular
surface patch r : U → R3 , where N is, as usual, the unit normal to the surface. Then
LN − M 2
∂1 e · ∂2 f − ∂2 e · ∂1 f = √ ,
EG − F 2
where E, F , G and L, M, N are the coefficients of the first and second fundamental
forms, respectively.

Proof. By Exercise 2.30, we have that ∂1 N × ∂2 N = K ∂1 r × ∂2 r. Therefore


LN − M 2
(∂1 N × ∂2 N) · N = K (∂1 r × ∂2 r) · N = K |∂1 r × ∂2 r| N · N = √ ,
EG − F 2

by Theorem 2.26 and the fact that |∂1 r × ∂2 r| = EG − F 2 .

59
2. Surfaces – local theory

Now,

(∂1 N × ∂2 N) · N = (∂1 N × ∂2 N) · (e × f) = (∂1 N · e)(∂2 N · f) − (∂2 N · e)(∂1 N · f),

by Exercise 1.9. Furthermore,

N·e≡0 =⇒ 0 ≡ ∂i (N · e) = ∂i N · e + N · ∂i e =⇒ ∂i N · e = −N · ∂i e,

for i = 1 and i = 2, and the same holds with e replaced by f. Hence

LN − M 2
√ = (N · ∂1 e)(N · ∂2 f) − (N · ∂2 e)(N · ∂1 f).
EG − F 2
Finally, note that when we write the vector fields ∂i e and ∂i f in terms of the orthonormal
triple (e, f, N), certain terms do not appear: as e · e ≡ 1, so e · ∂i e ≡ 0, and similarly
for ∂i f. Hence

∂1 e = (f · ∂1 e)f + (N · ∂1 e)N,
∂2 e = (f · ∂2 e)f + (N · ∂2 e)N,
∂1 f = (e · ∂1 f)e + (N · ∂1 f)N
and ∂2 f = (e · ∂2 f)e + (N · ∂2 f)N,

so
LN − M 2
∂1 e · ∂2 f − ∂2 e · ∂1 f = (N · ∂1 e)(N · ∂2 f) − (N · ∂2 e)(N · ∂1 f) = √ ,
EG − F 2
as claimed.

Proof of Theorem 2.28. It follows from Lemma 2.29 and Theorem 2.26 that

∂1 e · ∂2 f − ∂2 e · ∂1 f
K= √ .
EG − F 2
Thus it suffices to show that ∂1 e · ∂2 f − ∂2 e · ∂1 f depends only on the coefficients of the
first fundamental form and their partial derivatives, when e and f are suitably chosen.
Let
∂1 r ∂2 r − (e · ∂2 r)e
e := and f := ;
|∂1 r| |∂2 r − (e · ∂2 r)e|
since {∂1 r, ∂2 r} is linearly independent everywhere, the vector fields e and f are well
defined. (They are the result of applying the Gram-Schmidt orthogonalisation procedure
to this set.)
By construction, {e, f, N} is an orthonormal basis everywhere; to see that (e, f, N) is
a right-handed triple, it suffices to verify that [e, f, N] = [N, e, f] = N · (e × f) ≡ 1, by
Exercise 1.7.

60
2.4. Gaussian curvature

For this, note that

|∂1 r × ∂2 r| = (EG − F 2 )1/2 ,

|∂1 r| = E 1/2
and |∂2 r − (e · ∂2 r)e| = |∂2 r − |∂1 r|−2 (∂1 r · ∂2 r)∂1 r|
= |∂2 r − E −1 F ∂1 r|
= (G − 2E −1 F 2 + E −2 F 2 E)1/2
= E −1/2 (EG − F 2 )1/2 .

As ∂1 r × e ≡ 0, it follows that

(∂1 r × ∂2 r) · (∂1 r × ∂2 r)
N · (e × f) = ≡ 1,
(EG − F 2 )1/2 E 1/2 E −1/2 (EG − F 2 )1/2

as required.
Now,

∂2 (∂1 e · f) − ∂1 (∂2 e · f) = ∂21 e · f + ∂1 e · ∂2 f − ∂12 e · f − ∂2 e · ∂1 f


= ∂1 e · ∂2 f − ∂2 e · ∂1 f,

so it suffices to use (2.15) and (2.16) to note that


 
∂1 E ∂1 r ∂11 r E 1/2 (∂2 r − E −1 F ∂1 r)
∂1 e · f = + 1/2 ·
2E 3/2 E (EG − F 2 )1/2
∂11 r · ∂2 r F ∂11 r · ∂1 r 2∂1 F − ∂2 E F ∂1 E
= − = −
(EG − F 2 )1/2 E(EG − F 2 )1/2 2(EG − F 2 )1/2 2E(EG − F 2 )1/2
 
∂2 E ∂1 r ∂21 r E 1/2 (∂2 r − E −1 F ∂1 r)
and ∂2 e · f = + 1/2 ·
2E 3/2 E (EG − F 2 )1/2
∂21 r · ∂2 r F ∂21 r · ∂1 r ∂1 G F ∂2 E
= 2 1/2
− 2 1/2
= 2 1/2
− ,
(EG − F ) E(EG − F ) 2(EG − F ) 2E(EG − F 2 )1/2

so they depend only on E, F , G and their partial derivatives.

Corollary 2.30. If the parameterisation of the regular surface patch r : U → R3 is


orthogonal, so that the coefficients (E, F, G) of the first fundamental form are such
that F ≡ 0, then the Gaussian curvature

−1  ∂E   ∂ G 
2 1
K= √ ∂2 √ + ∂1 √ .
2 EG EG EG

61
2. Surfaces – local theory

Proof. If F ≡ 0 then, following the working in the proof of Theorem 2.28,


1 
K=√ ∂1 e · ∂2 f − ∂2 e · ∂1 f
EG
1 
=√ ∂2 (∂1 e · f) − ∂1 (∂2 e · f)
EG
1  −∂ E   ∂ G 
2 1
=√ ∂2 √ − ∂1 √ ,
EG 2 EG EG
as claimed.

Example 2.31. The sphere and the plane have different Gaussian curvature, which is
invariant under reparameterisation. Hence there is no isometric map from the sphere to
the plane, no matter how the plane is parameterised. In particular, any flat map of the
Earth must distort distances.

62
Three Surfaces – global theory

3.1 Overview
In this chapter we present three closely related theorems, named after Carl Gauss and
Pierre Bonnet. The first is the Gauss–Bonnet theorem for smooth simple closed curves.

Theorem 3.1. Let γ be a unit-speed simple closed curve in the regular surface patch r,
and suppose γ is positively oriented. Then
ZZ Z
K dAr − κg ds = 2π,
int γ γ
where int γ is the interior of the curve γ, K is the Gaussian curvature of r, dAr is the
area element of r and κg is the geodesic curvature of γ.

Informally, a simple closed curve is one which starts at some point and travels back to
that point without crossing itself; such a curve γ lies in the surface patch r as long as
the whole of γ and its interior are contained in r, and the curve is positively oriented if
its interior lies on the left of the curve as we proceed around it.

It is not straightforward to make sense of these seemingly simple concepts in a rigorous


fashion. For example, given the equator on the unit sphere, is the northern or southern
hemisphere its interior?

63
3. Surfaces – global theory

The second theorem is a generalisation of the first, called the Gauss–Bonnet theorem for
curvilinear polygons.

Theorem 3.2. Let γ be a unit-speed curvilinear polygon in the regular surface patch r,
and suppose γ is positively oriented and has n vertices. Then
ZZ Z n
X
K dAr − κg ds = (2 − n)π + αi ,
int γ γ i=1

where int γ, K, dAr and κg are as in Theorem 3.1 and α1 , . . . , αn are the interior angles
at the vertices of γ.

A curvilinear polygon is a simple closed curve which is smooth apart from a finite number
of corners, or vertices, where the curve changes direction sharply (so is not differentiable
there). A polygon in the plane, made up of a finite number of straight-line segments,
provides a familiar example of such a curve. Note that taking n = 0 reduces Theorem 3.2
to Theorem 3.1, so this is a generalisation of the former result.

The third and final theorem is the Gauss–Bonnet theorem for closed surfaces.

Theorem 3.3. If S is a closed surface then


ZZ
K dA = 2πχ, (3.1)
S

where χ is the Euler characteristic of S.

A surface S is a subset of R3 made up of one or more regular surface patches; it is closed


if S consists of only one piece (so is connected ) and is compact: given any collection of
open sets which covers S, a finite number of them can be selected which still covers S.
The plane
{(x, y, 0) : x, y ∈ R}
and the cylinder
{(x, y, z) : x2 + y 2 = 1, z ∈ R}
are connected but non-compact surfaces, as is the open unit disc

{(x, y, 0) : x2 + y 2 < 1}.

64
3.1. Overview

The unit sphere


{(x, y, z) : x2 + y 2 + z 2 = 1}
and the torus  p 2
(x, y, z) : x2 + y 2 − 2 + z 2 = 1
are closed surfaces.
[It is difficult to appreciate the definition of compactness at first, so an alternative
characterisation may be helpful: a set S ⊆ R3 is compact if and only if it is bounded (S
lies in some ball centred at the origin) and contains its limit points (a point x ∈ R3 is
a limit point of S if, for all ε > 0, there exists p ∈ S such that |x − p| < ε; hence S
contains its limit points if every point not in S lies a positive distance away from S).]
The Euler characteristic of a closed surface is defined by the formula

χ = V − E + F,

where V is the number of vertices, E is the number of edges and F is the number of
faces in a subdivision of S. This is a way of splitting up the surface into a finite number
of curvilinear polygons, the interiors of which form the faces, which are either disjoint
or share common edges or vertices. The precise choice does not matter; any subdivision
of a given surface will produce the same value for χ.
If a surface is orientable (so there is a consistent choice of an outward direction from the
surface and hence the surface has two sides, like the sphere and torus, but unlike the
Möbius strip) then
χ = 2 − 2g,
where g is the genus of S, the number of holes in it: the unit sphere has genus 0 and
the torus has genus 1. Every closed surface is orientable and homeomorphic to a torus
with g holes, for some g > 0: two surfaces are homeomorphic if there is a bijection
between them which is continuous and has continuous inverse.

The most remarkable thing about Theorem 3.3 is the different character of the two
sides of (3.1). The left-hand side is analytical and geometrical, defined using area and
curvature; the right-hand side is topological and combinatorial, defined by counting
vertices, edges and faces or holes. The fact that these two coincide is the source of
inspiration for a significant amount of recent mathematical research.

65
3. Surfaces – global theory

3.2 Plane curves

Definition 3.4. A plane curve is a map



c : (a, b) → R2 ; t 7→ u(t), v(t) ,

where −∞ 6 a < b 6 ∞, such that the coordinate functions

u : (a, b) → R and v : (a, b) → R

are continuous. The plane curve c is smooth if the coordinate functions u and v have
derivatives of all orders (i.e., are infinitely differentiable). [Thus a smooth plane curve
is the same thing as a smooth pair, considered earlier.]

A smooth plane curve is regular if the tangent vector field



c′ : (a, b) → R2 ; t 7→ (u′ (t), v ′ (t)

is nowhere zero; equivalently,

|c′ (t)|2 = u′ (t)2 + v ′ (t)2 > 0 for all t ∈ (a, b),

where |x| denotes the magnitude of the vector x ∈ R2 .


If c is a regular plane curve and m ∈ (a, b) then the bijection
Z t
ℓ : (a, b) → (c, d); t 7→ |c′(x)| dx
m

is the arc-length function for c with starting point m, where


Z a Z m Z b
′ ′
c := |c (t)| dt = − |c (t)| dt and d := |c′ (t)| dt;
m a m

note that c could equal −∞ and d could equal ∞. The arc-length function ℓ is smooth
and has strictly positive derivative, so the inverse

ℓ−1 : (c, d) → (a, b)

is also a smooth bijection. The curve ec := c ◦ ℓ−1 is a reparameterisation of c which has


unit speed, i.e.,
c′(s)| = 1
|e for all s ∈ (c, d).

66
3.2. Plane curves

Exercise 3.1. Prove that an arc-length function ℓ for a regular plane curve c = (u, v) is
infinitely differentiable. Prove further that the inverse ℓ−1 is also infinitely differentiable.

For a unit-speed plane curve



c : (a, b) → R2 ; s 7→ u(s), v(s) ,

the unit normal vector field is



n : (a, b) → R2 ; s 7→ −v ′ (s), u′ (s) .

Since the map " # 


 
x cos θ − sin θ x
7 →
y sin θ cos θ y

rotates the point (x, y) ∈ R2 about the origin through θ radians in the positive (anti-
clockwise) sense, geometrically the normal vector

n(s) = −v ′ (s), u′ (s)

is obtained by rotating the tangent vector



t(s) = u′ (s), v ′ (s)

through π/2 radians anticlockwise.

Note that
t(s) · n(s) = −u′ (s)v ′ (s) + v ′ (s)u′ (s) = 0
and
1 = |t(s)|2 = u′ (s)2 + v ′ (s) = |n(s)|,
so, for all s ∈ (a, b) the set {t(s), n(s)} is an orthonormal basis of R2 , analogous to the
Serret–Frenet basis {t(s), n(s), b(s)} attached to a unit-speed curve in R3 .

67
3. Surfaces – global theory

This orthonormal basis may be used to decompose the vector t′(s):

t′(s) = (t(s) · t′ (s))t(s) + (n(s) · t′ (s))n(s).

However, as t has unit magnitude,


d d 
0= |t(s)|2 = t(s) · t(s) = t′ (s) · t(s) + t(s) · t′ (s) = 2t(s) · t′ (s),
ds ds
so
t′ (s) = κs (s)n(s) for all s ∈ (a, b),
where the scalar field

κs : (a, b) → R; s 7→ κs (s) := n(s) · t′ (s)

is the signed curvature of the curve c. Note that

|t′| = |κs n| = |κs |,

so the absolute value |κs | is the magnitude of the derivative of the tangent vector (the
same as for curves in R3 ).

If the scalar curvature is positive then the curve is moving in the direction of the normal
vector; if negative, the curve is bending away from the normal vector. The following
theorem makes this idea more precise.

Theorem 3.5. Let u ∈ R2 be a fixed unit vector. If c is a unit-speed plane curve and ψ
is the angle from u to t, measured in the positive sense, then

κs = .
ds

Proof. Let v be the unit vector obtained by rotating u through π/2 radians in the
positive sense and note that {u, v} is an orthonormal basis, so

t = (u · t) u + (v · t) v = cos ψ u + sin ψ v.

Then
t′ = (− sin ψ u + cos ψ v)ψ ′
and
n = cos ψ v − sin ψ u
(as rotating u and v through π/2 radians gives v and −u, respectively) so

κs = n · t′ = (sin2 ψ + cos2 ψ)ψ ′ = ψ ′ .

68
3.3. Simple closed curves

Those of a cautious frame of mind may wonder if the angle ψ in Theorem 3.5 is well
defined. The following lemma should reassure them.

Lemma 3.6. Let


x : (a, b) → R and y : (a, b) → R
be smooth functions such that x(s)2 + y(s)2 = 1 for all s ∈ (a, b). If s0 ∈ (a, b) is fixed
and ψ0 ∈ R is chosen so that

(cos ψ0 , sin ψ0 ) = x(s0 ), y(s0)

then Z s
ψ : (a, b) → R; s 7→ ψ0 + x(t)y ′ (t) − x′ (t)y(t) dt
s0

is a smooth function such that


 
cos ψ(s), sin ψ(s) = x(s), y(s) for all s ∈ (a, b).

Proof. Consider the function


2 2
F : (a, b) → R; s 7→ x(s) − cos ψ(s) + y(s) − sin ψ(s) ;

since F (s0 ) = 0, the result holds if F ′ ≡ 0. However, as x2 + y 2 ≡ 1, it follows by


differentiating that xx′ + yy ′ ≡ 0 and therefore
1 ′
2
F = (x − cos ψ)(x′ + ψ ′ sin ψ) + (y − sin ψ)(y ′ − ψ ′ cos ψ)
= xx′ + yy ′ − (x′ cos ψ + y ′ sin ψ) + ψ ′ (x sin ψ − y cos ψ)
= 0 − (x2 + y 2)(x′ cos ψ + y ′ sin ψ) + (xy ′ − x′ y)(x sin ψ − y cos ψ)
= −x(xx′ + yy ′) cos ψ − y(xx′ + yy ′) sin ψ
≡ 0,

as required. That ψ is smooth follows from the fundamental theorem of calculus and
the smoothness of x and y.

3.3 Simple closed curves

Definition 3.7. Let c : R → R2 be a plane curve. If there exists a > 0 such that

(i) c|[0,a) is injective, so that c(s) = c(t) if and only if s = t, for all s, t ∈ [0, a)
and
(ii) c(t) = c(t + a) for all t ∈ R

then c is a simple closed plane curve with period a.

69
3. Surfaces – global theory

Theorem 3.8. A simple closed plane curve c partitions the plane R2 into three disjoint
non-empty sets, the image of c, c(R), the interior of c, int c, and the exterior of c, ext c;
the latter two sets are open and have c(R) as their boundary. [The boundary of a set S is
the set of limit points of S which do not lie in S.] Each of these three sets is connected:
any two points in the set may be joined by a path which lies within that set. The interior
of c is bounded and the exterior of c is unbounded: there exists r > 0 such that

int c ⊆ {(x, y) ∈ R2 : x2 + y 2 < r}

and, for all R > 0,


ext c ∩ {(x, y) ∈ R2 : x2 + y 2 > R} =
6 ∅.

Proof. This is the Jordan curve theorem. A proof is surprisingly difficult, so we omit it.
The concerned reader may look at Theorem 63.4 in [4]; a simpler proof, valid only for
piecewise-smooth curves, is contained in [2, Section VIII.5].

A unit-speed simple closed plane curve is positively oriented if the normal vector field
points to the interior of the curve at every point, and negatively oriented if the normal
vector field points to the exterior of the curve at every point. The normal vector field
cannot flip from interior to exterior, or vice versa, for reasons of continuity; hence every
unit-speed simple closed plane curve has an orientation, either positive or negative.

We extend the notation of orientation to any regular simple closed plane curve, by saying
such a curve is positively oriented if a unit-speed reparameterisation is, and similarly for
negatively oriented.

Exercise 3.2. Prove that any unit-speed reparameterisation of a regular simple closed
plane curve c with period a is a regular simple closed plane curve with period
Z a
ℓ(c) := |c′ (t)| dt.
0

Exercise 3.3. Suppose c is a regular simple closed plane curve and let e c(t) := c(−t)
for all t ∈ R. Show that ec is also a regular simple closed plane curve, with the same
period as c. Show further that c and ec have opposite orientations.

70
3.3. Simple closed curves

The following result is known as Green’s theorem.

Theorem 3.9. Let c = (u, v) : R → R2 be a regular simple closed curve with period a,
which is positively oriented. Suppose f : U → R and g : U → R are smooth functions
defined on the open set U ⊆ R2 which contains c and its interior. Then
Z ZZ
f du + g dv = ∂1 g − ∂2 f dx dy,
c int c

where the notation for the first integral is shorthand for


Z a
 
f u(t), v(t) u′ (t) + g u(t), v(t) v ′ (t) dt.
0

Proof. The strategy is to prove the theorem for particularly nice choices of c, then to
show that any c can be reduced to this case by suitable subdivision of the region int c;
this is somewhat similar to the proof of Cauchy’s theorem in MATH215. Details can be
found in [3, Section 5.2].

Corollary 3.10. The area A(int c) enclosed by the positively oriented regular simple
closed curve c = (u, v) : R → R2 with period a equals
Z
1 a
u(t)v ′(t) − u′ (t)v(t) dt.
2 0

Proof. Let f (x, y) = −y and g(x, y) = x for all x, y ∈ R. Then, by Green’s theorem,
Z a ZZ
′ ′
−v(t)u (t) + u(t)v (t) dt = 1 + 1 dx dy = 2A(int c).
0 int c

Example 3.11. The simple closed curve

c : R → R2 ; t 7→ (r cos t, r sin t)

is positively oriented and has period 2π; it describes in an anticlockwise motion the circle
about the origin with radius r > 0.

By Corollary 3.10, the interior of this circle has area


Z
1 2π 2
r cos2 t + r 2 sin t dt = πr 2 ,
2 0
as expected.

71
3. Surfaces – global theory

Simple closed curves in surfaces

Definition 3.12. Let r : U → R3 be a regular surface patch and let



c : R → R2 ; t 7→ u(t), v(t)

be a regular simple closed plane curve with period a. If the open set U contains c and
its interior then 
γ := r ◦ c : R → R3 ; t 7→ r u(t), v(t)
is a simple closed curve in r with period a and interior int γ := r(int c). By definition,
the curve γ has the same orientation as c.

Exercise 3.4. Prove that a simple closed curve γ in r with period a is a regular smooth
curve such that γ|[0,a) is injective and γ(t) = γ(t + a) for all t ∈ R.

Exercise 3.5. Let γ be a simple closed curve in r with period a, let m ∈ R and let
Z t
ℓ : R → R; t 7→ |γ ′ (x)| dx.
m

e := γ ◦ ℓ−1 is a
Prove that ℓ is a smooth bijection with smooth inverse. Deduce that γ
unit-speed smooth curve in r with period
Z a
ℓ(γ) := |γ ′ (x)| dx.
0

72
3.3. Simple closed curves

We can now state the first Gauss–Bonnet theorem. The minus sign on the left-hand
side below appears as we have defined the geodesic curvature in the opposite sense to
Pressley: compare this with [6, Theorem 11.1].

Theorem 3.13. Let γ := r ◦ c be a unit-speed simple closed curve in the regular surface
patch r : U → R3 and suppose γ is positively oriented. Then
ZZ Z
K dAr − κg ds = 2π,
int γ γ
where K is the Gaussian curvature of r, dAr is the area element of r and κg is the
geodesic curvature of γ. The integrals are defined as follows:
ZZ ZZ √
K dAr := K(u, v) EG − F 2 (u, v) du dv
int γ int c
Z Z a
and κg ds := κg (s) ds,
γ 0

where E, F and G are the coefficients of the first fundamental form of r and a is the
period of γ.

Proof. Let (e, f, N) be the orthonormal triple obtained by applying the Gram–Schmidt
procedure to {∂1 r, ∂2 r, N}, as in the proof of Theorem 2.28. The proof here consists of
calculating Z Z a
I := e · f ′ ds = e(u, v) · f(u, v)′ ds
γ 0

in two different ways, where



c = (u, v) : R → R2 ; s 7→ u(s), v(s) .

First, note that


Z a Z a

e(u, v) · f(u, v) ds = e(u, v) · (∂1 f(u, v)u′ + ∂2 f(u, v)v ′) ds
0 0
Z
= e · ∂1 f du + e · ∂2 f dv.
c

By Green’s theorem, Theorem 3.9, this equals


ZZ ZZ
∂1 (e · ∂2 f) − ∂2 (e · ∂1 f) du dv = ∂1 e · ∂2 f − ∂2 e · ∂1 f du dv
int c int c

and, from Lemma 2.29, it follows that


ZZ ZZ √ ZZ
LN − M 2
I= √ du dv = 2
K EG − F du dv = K dAr .
int c EG − F 2 int c int γ

73
3. Surfaces – global theory


Now use Lemma 3.6 to let θ(s) be the angle between t(s) := γ ′ (s) and e u(s), v(s) , so
that
t = e(u, v) cos θ + f(u, v) sin θ.
As (e, f, N) is an orthonormal triple of vector fields,

t × N(u, v) = −f(u, v) cos θ + e(u, v) sin θ

and

t′ = e(u, v)′ cos θ − e(u, v) θ′ sin θ + f(u, v)′ sin θ + f(u, v) θ′ cos θ
= e(u, v)′ cos θ + f(u, v)′ sin θ − t × N(u, v) θ′.

It follows that

κg := t′ · t × N(u, v) = −θ′ + (−e(u, v)′ · f(u, v) cos2 θ + f(u, v)′ · e(u, v) sin2 θ)
= −θ′ + e(u, v) · f(u, v)′,

using the facts that t × N(u, v) is a unit vector and

e(u, v) · e(u, v) ≡ 1 =⇒ e(u, v)′ · e(u, v) ≡ 0,


f(u, v) · f(u, v) ≡ 1 =⇒ f(u, v)′ · f(u, v) ≡ 0
and e(u, v) · f(u, v) ≡ 0 =⇒ e(u, v)′ · f(u, v) = −e(u, v) · f(u, v)′ .

The result now follows, since


Z a Z a Z a

− κg ds = θ ds − e(u, v) · f(u, v)′ ds = 2π − I
0 0 0
R Ra
because γ θ′ ds := 0 θ′ ds = θ(a) − θ(0) = 2π.

For this final step, we argue as follows. It may be shown that the interior of any simple
closed curve is simply connected: it has no holes. Consequently,
R ′ the curve γ may be
1
continuously deformed until it is a small circle γ 0 . As 2π γ θ ds is integer valued and
R R
depends on γ in a continuous manner, it must be the case that γ θ′ ds = γ θ′ ds. The
0
curve γ 0 is very small, so e is almost constant on γ 0 and θ is almost the angle made by
the tangent vector γ ′0 with a fixed direction. Finally, as the tangent vector travels once
round a circle anticlockwise, it turns through an angle of 2π.

74
3.3. Simple closed curves

Curvilinear polygons
If the simple closed curve γ is only piecewise smooth then a modified version of the Gauss–
Bonnet theorem still holds. Most of the working above in the proof R of′ Theorem 3.13
remains valid; Green’s theorem still applies, but the final claim, that γ θ ds = 2π, must
be modified.

Definition 3.14. The simple closed curve c : R → R2 with period a is a curvilinear


polygon with vertices 0 = t1 < · · · < tn < a if

(i) for i = 1, . . . , n, the edge c|(ti ,ti+1 ) is a regular smooth curve , where tn+1 := a, and
the limit tangent vectors

c′ (ti +) := lim c′ (t) and c′ (ti+1 −) := lim c′ (t)


t→ti + t→ti+1 −

at the start and the end of the edge are well defined

and

(ii) the vectors c′ (ti +) and c′ (ti −) are linearly independent for i = 1, . . . , n, where
c′ (0−) := c′ (a−).

Since a curvilinear polygon is a simple closed curve, it has an interior. A curvilinear


polygon c has unit speed if each of its edges do, and is positively oriented if the unit
normal to each edge, reparameterised by arc length if necessary, points to the interior
of c.
Let r : U → R3 be a regular surface patch. If c : R → R2 is a curvilinear polygon such
that U contains c and its interior then γ := r ◦ c is a curvilinear polygon in r with the
same orientation, the same period a and the same vertices 0 = t1 < t2 < · · · < tn as c,
and with edges γ|(ti ,ti+1 ) for i = 1, . . . , n.

75
3. Surfaces – global theory

R
Consider γ θ′ ds, where γ is a unit-speed curvilinear polygon in r. At each vertex ti
the value of θ jumps by the exterior angle δi , where cos δi = γ ′ (ti −) · γ ′ (ti +).

Hence the integral


Z n
X n
X

θ ds = 2π − δi = 2π − (π − αi ),
γ i=1 i=1

where αi := π − δi is the interior angle at the vertex ti . (If γ is a polygon in the plane
then θ′P≡ 0 on each side and we have the familiar result that the sum of the interior
angles ni=1 αi equals (n − 2)π.) This extension of the previous result gives the following
variant of the Gauss–Bonnet theorem.

Theorem 3.15. Let γ be a positively oriented unit-speed curvilinear polygon in the


regular surface patch r. If γ has interior angles α1 , . . . , αn at its n vertices then
ZZ Z n
X
K dAr − κg ds = (2 − n)π + αi .
int γ γ i=1

Example 3.16. A curvilinear triangle is a curvilinear polygon with three vertices; a


geodesic triangle is a curvilinear triangle with edges which are geodesics. For a unit-
speed geodesic triangle γ with interior angles α1 , α2 and α3 , the Gauss–Bonnet theorem
gives that ZZ
K dAr = −π + (α1 + α2 + α3 ).
int γ

For the plane, where K ≡ 0, the edges of a geodesic triangle are straight lines and we
have the familiar result that
α1 + α2 + α3 = π;
the sum of the interior angles of a plane triangle is π.
For the unit sphere, where K ≡ 1, the edges of a geodesic triangle are great circles and
the Gauss–Bonnet theorem gives that

A(int γ) = α1 + α2 + α3 − π.

In particular, α1 + α2 + α3 > π: the sum of the interior angles of a spherical triangle


exceed π.

76
3.4. Global surfaces

3.4 Global surfaces


Charts and atlases
The reader may wonder why the regular surface patches we have considered often have
pieces missing; the sphere of Example 2.5 is lacking a line of longitude, for example.
This is not due to carelessness or lack of effort; topological methods may be used to
show that there is no continuous bijection r : U → S 2 , where U is an open set in R2 and

S 2 := {(x, y, z) ∈ R3 : x2 + y 2 + z 2 = 1}

is the unit sphere.


If we want to deal with surfaces such as S 2 as a whole, it is necessary to build on the
foundations from Chapter 2; the idea is to cover the surface with patches, which introduce
systems of local coordinates. The systems, or charts, are required to be compatible where
they overlap, so that the change of coordinates is a smooth function.
Formally, we proceed as follows. Let S be a subset of R3 , and suppose there exists a
family of regular surface patches {ri : Ui → S | i ∈ I} which covers S, i.e., the union
∪i∈i Si = S, where Si := ri (Ui ).

Definition 3.17. Since a surface patch ri : Ui → Si := ri (Ui ) ⊆ S is injective, the


coordinate chart 
1 2
φi : Si → Ui ; p 7→ r−1
i (p) = (φi (p), φi (p)

is well defined, and the real numbers φ1i (p) and φ2i (p) are the coordinates of the point p
with respect to this chart.

It is required that these charts be compatible: if Si ∩ Sj 6= ∅ then the transition function

φj ◦ φ−1
i : φi (Si ∩ Sj ) → φj (Si ∩ Sj )

must be smooth, i.e., have partial derivatives of all orders, for all i, j ∈ I. For this to
make sense, the domain of the transition function must be an open subset of R2 .

77
3. Surfaces – global theory

Definition 3.18. A set U ⊆ R3 is open if every point in U is contained in some open


ball lying within U: for all u ∈ U there exists r > 0 such that the ball
B(u; r) := {v ∈ R3 : |u − v| < r} ⊆ U.
[This is the same as the definition for R2 , but with “disc” replaced by “ball”.]
Let S be a subset of R3 . A set V ⊆ S is said to be relatively open in S if there exists an
open set W ⊆ R3 such that W ∩ S = V .

[If you know about such things, you should verify that the open sets in R3 are those
which come from the standard Euclidean metric, and that S has the relative topology
obtained by restricting this metric.]

Exercise 3.6. [For those who know some topology.] Prove that if V is relatively open
2
in S then r−1
i (V ) is open in R . Deduce that if Si is relatively open in S then φi (Si ∩ Sj )
is open in R2 for all j ∈ I.

This leads us to the following definition; in practice, it turns out to be slightly more
convenient to work with the coordinate charts than their inverses, the surface patches.

Definition 3.19. A surface (S, A) is a set S ⊆ R3 with an atlas


A := {φi : Si → Ui | i ∈ I}
of coordinate charts; for all i ∈ I, the map φi is a continuous bijection from the set Si ,
which is relatively open in S, to the set Ui , which is open in R2 , and its inverse
3
ri := φ−1
i : Ui → Si ⊆ R

is a regular surface patch. Moreover, these charts cover S, so that ∪i∈I Si = S.

The following theorem, which states that transition functions are automatically smooth,
is an application of the inverse-function theorem, Theorem A.7. [The proof may safely
be skipped by those without the necessary knowledge of topology.]

Theorem 3.20. Let φ1 : S1 → U1 and φ2 : S2 → U2 be coordinate charts. Then the


transition function
φ1 ◦ φ−1
2 : φ2 (S1 ∩ S2 ) → φ1 (S1 ∩ S2 )
is smooth.

Proof. It suffices to show that the transition function is infinitely differentiable at an


arbitrary point (x0 , y0 ) ∈ φ2 (S1 ∩ S2 ). Fix this point and note that

(z0 , w0 ) := r−1
1 r2 (x0 , y0 ) ∈ φ1 (S1 ∩ S2 ) ⊆ U1
and, by regularity, the 2 × 3 matrix
" #
∂1 r1 (z0 , w0 )
∂2 r1 (z0 , w0 )
has linearly independent rows.

78
3.4. Global surfaces

As row rank and column rank are equal, this matrix must have (at least) two linearly
independent columns; let these be the ith and jth columns, where 1 6 i < j 6 3. If

π : R3 → R2 ; (u1 , u2 , u3 ) 7→ (ui , uj )

then the 2 × 2 matrix "  #


π ∂1 r1 (z0 , w0 )
A= 
π ∂2 r1 (z0 , w0 )
is invertible and the map
Φ := π ◦ r1 : U1 → R2
has Jacobian JΦ(z0 , w0 ) = A. By the inverse-function theorem, Theorem
A.7, there
exist open sets V ⊆ U1 and V ⊆ R such that (z0 , w0 ) ∈ V and Φ V : V → Ve is a
e 2
smooth bijection with smooth inverse.
Now note that −1
r−1
1 r1 (V ) = Φ V ◦ π r1 (V ) ,

since if v ∈ V and p = r1 (v) then Φ(v) ∈ Ve and


−1  −1   −1 
r−1
1 (p) = v = Φ Φ(v) = Φ π r 1 (v) = Φ V π(p) .
V V


Letting W := r−1
2 r 1 (V ) , it follows that W is open (since V is open and both r−1
1 = φ1
and r2 are continuous) and
−1 −1
φ1 ◦ φ−1 −1
2 W = r1 ◦ r2 W = r1 r1 (V ) ◦ r2 W = Φ V ◦ π r1 (V ) ◦ r2 W ;

the right-hand side is the composition of three smooth maps, so is smooth. Finally,
 
(x0 , y0) = r−1
2 r1 (z0 , w0 ) ∈ r−1
2 r1 (V ) = W

so the result follows.

Exercise 3.7. Explain why, by increasing the number of charts in a given atlas if
necessary, one may assume that each codomain Ui is an open disc in R2 . Explain further
why, with a little more work, this may be taken to be the open unit disc

D := {(u, v) ∈ R2 : u2 + v 2 < 1}.

We will only consider embedded surfaces in these notes: those which exist as subsets
of R3 . There is a more general theory, where a surface is an abstract topological space,
and which considers curves, surfaces and higher-dimensional spaces on the same footing:
this is the theory of differentiable manifolds.

79
3. Surfaces – global theory

Example 3.21. The sphere S 2 has an atlas consisting of six coordinate charts,

φz+ : {(x, y, z) ∈ S 2 : z > 0} → D; (x, y, z) 7→ (x, y),


φz− : {(x, y, z) ∈ S 2 : z < 0} → D; (x, y, z) 7→ (x, y),
φy+ : {(x, y, z) ∈ S 2 : y > 0} → D; (x, y, z) 7→ (x, z),
φy− : {(x, y, z) ∈ S 2 : y < 0} → D; (x, y, z) 7→ (x, z),
φx+ : {(x, y, z) ∈ S 2 : x > 0} → D; (x, y, z) 7→ (y, z)
and φx− : {(x, y, z) ∈ S 2 : x < 0} → D; (x, y, z) 7→ (y, z),

where the common codomain

D := D(0, 0; 1) = {(u, v) ∈ R2 : u2 + v 2 < 1}

is the open unit disc; the domain of each chart is a hemisphere, and is relatively open
in S 2 because the sets

{(x, y, z) ∈ R3 : x > 0}, {(x, y, z) ∈ R3 : x < 0},


{(x, y, z) ∈ R3 : y > 0}, {(x, y, z) ∈ R3 : y < 0},
{(x, y, z) ∈ R3 : z > 0} and {(x, y, z) ∈ R3 : z < 0}

are all open in R3 .

To compute the transition functions, consider φy− ◦ φ−1 z+ ; the other cases are similar.
Letting Sy− denote the domain of φy− , and so on, it follows that

Sy− ∩ Sz+ = {(x, y, z) ∈ S 2 : y < 0 and z > 0}

and the regular surface patch inverse to φz+ is


√ 
rz+ := φ−1
z+ : D → Sz+ ; (u, v) 7→ u, v, 1 − u2 − v 2 .

Hence the transition function


√ 
φy− ◦ φ−1
z+ : {(u, v) ∈ D : v < 0} → {(u, v) ∈ D : v > 0}; (u, v) 7→ u, 1 − u2 − v 2 ,

which is smooth.

80
3.4. Global surfaces

Exercise 3.8. Use stereographic projection to obtain an atlas for the sphere S 2 which
consists of only two charts.

Exercise 3.9. From its realisation as a surface of revolution, it may seem that the
circular cylinder

S 1 × R := {(x, y, z) ∈ R3 : x2 + y 2 = 1, z ∈ R}

requires an atlas with at least two coordinate charts. Find a single chart which suffices.
[Hint: S 1 × R can be seen as C \ {0}; how?]

A surface S is connected if, for every pair of points x, y ∈ S there exists a path from x
to y, i.e., a continuous function f : [0, 1] → S such that f (0) = x and f (1) = y.
A surface S is closed if it is connected and compact: if S is a family of relatively open
sets in S which covers S then there exist S1 , . . . , Sn ∈ S such that S = S1 ∪ · · · ∪ Sn . In
particular, any atlas for S may be reduced to an atlas containing only a finite number of
charts. [In Rn , compact sets are those which are bounded and contain their limit points,
as mentioned earlier.]

Exercise 3.10. Show that the plane

{(x, y, 0) ∈ R3 : x, y ∈ R}

is not compact. [Hint: consider the atlas consisting of the charts

φn : {(x, y, 0) ∈ R3 : x2 + y 2 < n} → {(x, y) : x2 + y 2 < n}; (x, y, 0) 7→ (x, y),

for all n > 1.]

A surface is orientable if there is a compatible atlas such that the transition maps between
charts are always orientation preserving; it is non-orientable otherwise. The sphere and
torus are orientable; the Möbius strip is not. In fact, every closed surface is orientable.
Surfaces S1 and S2 are homeomorphic if there exists a continuous bijection f : S1 → S2
with continuous inverse f −1 : S2 → S1 .

Theorem 3.22. Every closed surface is homoeomorphic to a torus with g holes, for
some g > 0. The number g is the genus of the surface.

Proof. This is another big theorem, with a complicated proof which we shall omit.

81
3. Surfaces – global theory

There is a similar classification theorem for non-orientable surfaces; see [11, Section 5].
Non-orientable closed surfaces do exist, but not as subsets of R3 . A Klein bottle is the
closed surface obtained by identifying pairs of opposite points on the edges of a Möbius
strip; if realised in three dimensions, it must intersect itself. The abstract theory of
manifolds avoids issues like this.

Subdivisions and the Euler characteristic

Definition 3.23. Let S be a closed surface. A subdivision of S is a finite collection of


curvilinear polygons such that

(i) each face, the image of a curvilinear polygon and its interior, lies in a regular
surface patch with inverse contained in the atlas for S,

(ii) every point in S belongs to a face,

(iii) faces are disjoint, or meet at a common vertex or a common edge

and

(iv) each edge belongs to exactly two faces.

Given such a subdivision, the Euler characteristic

χ := V − E + F,

where V is the number of distinct vertices, E is the number of distinct edges and F is
the number of faces in the subdivision.

82
3.4. Global surfaces

Theorem 3.24. Any subdivision of a closed surface S has Euler characteristic χ = 2−2g,
where g is the genus of S.

Proof. See [11, Theorem 5.1].

Hence the Euler characteristic depends only on the surface, and not a particular choice
of subdivision; hence we speak of the Euler characteristic of a surface. Furthermore, one
may determine the genus of a surface intrinsically, by constructing a subdivision and
computing the Euler characteristic.

Theorem 3.25. Every closed surface possesses a triangulation: a subdivision where


every polygon has three vertices.

Proof. See [11, Section 4].

It is even possible to show that a geodesic triangulation exists for any closed surface, i.e.,
a subdivision where every edge is a geodesic and every face is a geodesic triangle.

The global Gauss–Bonnet theorem

Theorem 3.26. If S is a closed surface then


ZZ
K dA = 2πχ,
S

where K is the Gaussian curvature and χ is the Euler characteristic of S.

This is quite an astonishing result; the quantity on the right has an analytical nature,
depending on area and curvature, whereas the left is a topological quantity, depending
only on the number of holes in the surface.

83
3. Surfaces – global theory

Proof of Theorem 3.26. Suppose S has a geodesic triangulation, and let γ be a geodesic
triangle bounding a face in this triangulation. By the Gauss–Bonnet theorem for curvi-
linear polygons, ZZ
K dAr = α1 + α2 + α3 − π,
int γ

where r is a regular surface patch containing γ and its interior; it follows from earlier
results that this quantity is independent of the parameterisation of int γ. Summing over
all faces gives ZZ
K dA = 2πV − πF,
S
where V is the number of vertices and F the number of faces in the subdivision, since
the sum of interior angles at a vertex is 2π. Furthermore, each edge in the subdivision
is the boundary of two faces, so
2E = 3F
and therefore
ZZ
K dA = 2π(V − 12 F ) = 2π(V − 32 F + F ) = 2π(V − E + F ) = 2πχ,
S

as required.

84
A Auxiliary results

A.1 Analysis

Notation A.1. (Landau) Let f andg be functions defined on an open interval which
contains 0. We write f (h) = o g(h) if limh→0 f (h)/g(h) = 0. For example, if the
function f is differentiable at x, with f ′ (x) = c, then
f (x + h) − f (x)
− c → 0 as h → 0,
h
so f (x + h) − f (x) − hc = o(h). We can re-arrange this to give the following proposition,
a simple re-writing of the definition of differentiability: a function f is differentiable at x
with derivative c if and only if f (x + h) = f (x) + hc + o(h).

Theorem A.2. (Intermediate-value theorem) If f : [a, b] → R is continuous and y


lies between f (a) and f (b), so either f (a) 6 y 6 f (b) or f (b) 6 y 6 f (a), then there
exists x ∈ [a, b] such that f (x) = y.

Proof. MATH210.

Theorem A.3. If f : I → J is a differentiable bijection between open subintervals of R


with f ′ (x) > 0 for all x ∈ I then f −1 : J → I is also differentiable, with

(f −1 )′ (y) = 1/f ′ f −1 (y) for all y ∈ J.

If f ′ is continuous then so is (f −1 )′ .

Proof. MATH210. For the final comment, note that if f ′ is continuous then so is f ′ ◦f −1 ,
and f ′ (x) > 0 for all x ∈ I.

Theorem A.4. (Taylor’s theorem) Let f : (a − h, a + h) → R be n + 1-times


differentiable, where h > 0. If x ∈ (a − h, a + h) then there exists c between x and a
such that
f ′′ (a) f (n) (a) f (n+1) (c)
f (x) = f (a) + f ′(a)(x − a) + (x − a)2 + · · · + (x − a)n + (x − a)n+1 .
2 n! (n + 1)!

85
A. Auxiliary results

Definition A.5. (The Jacobian) Suppose Φ = (φ, ψ) : U → R2 has continuous partial


derivatives, where U ⊆ R2 is open. The Jacobian of Φ is the matrix-valued function JΦ
defined on U by setting
" #
∂1 φ(u, v) ∂2 φ(u, v)
JΦ(u, v) = for all (u, v) ∈ U,
∂1 ψ(u, v) ∂2 ψ(u, v)
∂φ
where ∂1 φ(u, v) := (u, v) and so on. If u ∈ U then it may be shown that
∂u
Φ(u + h) = Φ(u) + JΦ(u)h + o(|h|),
with the magnitude of vectors in R2 defined in the usual way, in the sense that
|Φ(u + h) − Φ(u) − JΦ(u)h|
lim = 0.
h→0 |h|
(Note that this makes sense: JΦ(u)h is the product of the matrix JΦ(u) with the
column vector h.) Hence the Jacobian is the appropriate generalisation, for functions of
two variables, of the derivative considered as the best linear approximation.
If Ψ : V → U and Φ : U → R2 have continuous partial derivatives, where U and V are
open subsets of R2 , then it may be shown that

J(Ψ ◦ Φ)(v) = JΨ Φ(v) JΦ(v) for all v ∈ V.
Compare this with the one-dimensional version: if f : I → R and g : J → I are
differentiable on the open intervals I and J, respectively, then

(f ◦ g)′ (x) = f ′ g(x) g ′ (x) for all x ∈ J.
Hence if Ψ : V → U is invertible and Φ = Ψ−1 then
h i
1 0 
= JΨ Ψ−1 (v) JΨ−1 (v),
0 1

so JΨ Ψ−1 (v) is invertible and
 −1
JΨ−1 (v) = JΨ Ψ−1 (v) for all v ∈ V.
(This is an exact analogue of the formula

(f −1 )′ (y) = 1/f ′ f −1 (y) ,
valid for any invertible function f which has non-zero derivative at f −1 (y).)

Theorem A.6. If V ⊆ R2 is open and Φ : V → R2 has continuous partial derivatives


and Jacobian JΦ such that det JΦ(v) 6= 0 for all v ∈ V then
ZZ ZZ

f (x, y) dx dy = f Φ(u, v) | det JΦ(u, v)| du dv
Φ(V ) V

for any sufficiently nice function f : Φ(V ) → R.

Proof. MATH113. “Sufficiently nice” includes all the functions in these notes.

86
A.2. Algebra

Theorem A.7. (Inverse-function theorem) Let Φ : U → R2 be smooth, where the


set U ⊆ R2 is open, and suppose v ∈ U is such that JΦ(v) is invertible. There exist
open sets V and Ve such that v ∈ V ⊆ U, Φ(v) ∈ Ve and Φ|V : V → Ve is invertible with
smooth inverse.
A.2 Algebra

Theorem A.8. (Principal-axis theorem) Let C be a real 2 × 2 matrix. If C is


symmetric (that is, the transpose C t = C) then there exists a real 2 × 2 matrix D which
is orthogonal (i.e., D t D = DD t = I) and such that
h i
t
λ1 0
D CD = ,
0 λ2

where λ1 6 λ2 .

87
B Solutions to selected exercises

Solution to the first part of Exercise 1.9


Let u = (u1, u2 , u3 ) et cetera. Then u × (v × w) equals

i j k i j k


u × v1 v2 v3 = u1 u2 u3

w1 w2 w3 v2 w3 − v3 w2 v3 w1 − v1 w3 v1 w2 − v2 w1

= (u2 v1 w2 − u2 v2 w1 − u3 v3 w1 + u3 v1 w3 ,
u1 v2 w1 − u1 v1 w2 − u3 v3 w2 + u3 v2 w3 ,
u1 v3 w1 − u1 v1 w3 − u2 v2 w3 + u2 v3 w2 )
= (u1 w1 + u2 w2 + u3 w3 )v1 − (u1 v1 + u2 v2 + u3 v3 )w1 ,
(u1 w1 + u2 w2 + u3 w3 )v2 − (u1 v1 + u2 v2 + u3 v3 )w2 ,

(u1 w1 + u2 w2 + u3 w3 )v3 − (u1 v1 + u2 v2 + u3 v3 )w3 )
= (u · w)v − (u · v)w,

as claimed.

Solution to the first part of Exercise 1.15 Z d


Since |γ(d) − γ(c)| is the straight-line distance from γ(c) to γ(d), whereas |γ ′ (t)| dt
is the distance measured along the curve γ, the inequality c

Z d
|γ ′ (t)| dt > |γ(d) − γ(c)|
c

holds because a straight line gives the shortest distance in R3 between two points. For
a rigorous proof of this fact, we proceed as follows.
Let u ∈ R3 be a unit vector such that

γ(d) − γ(c) · u = |γ(d) − γ(c)|

and note that


d 
γ(t) · u = γ ′ (t) · u + γ(t) · u′ = γ ′ (t) · u.
dt

89
B. Solutions to selected exercises

By the fundamental theorem of calculus,


Z Z d
d
d 
|γ(d) − γ(c)| = γ(d) · u − γ(c) · u = γ(t) · u dt = γ ′ (t) · u dt
c dt c

and the Cauchy–Schwarz inequality implies that


γ ′ (t) · u 6 |γ ′ (t) · u| 6 |γ ′ (t)| |u| = |γ ′ (t)|.
Thus Z Z
d d

|γ(d) − γ(c)| = γ (t) · u dt 6 |γ ′ (t)| dt.
c c

Solution to Exercise 1.18


If γ : (a, b) → R3 is parameterised by arc length then there exists m ∈ (a, b) such that
Z t
s(t) := |γ ′ (r)| dr = t
m

for all t ∈ (a, b). Hence |γ (t)| = s (t) = 1 for all t ∈ (a, b) and γ has unit speed.
′ ′

Solution to Exercise 1.22


If κ ≡ 0 then t′ ≡ 0 and t ≡ c for some constant vector c. It follows that γ ′ = t ≡ c
and γ(s) = s c + d for all s, where d is a constant vector: γ lies on a straight line.

Solution to Exercise 1.23


Since t and n are orthogonal, it suffices to prove that each of these vector fields is
orthogonal to u. Differentiating,
d≡γ·u =⇒ 0 ≡ γ ′ · u + γ · u′ =⇒ 0≡t·u
and t is orthogonal to u. Differentiating once more,
0≡t·u =⇒ 0 ≡ t′ · u + t · u′ =⇒ 0 ≡ κn · u;
as κ(s) 6= 0 for all s, it follows that n and u are orthogonal, as required.
For the final claim, note that
n·n≡1 =⇒ n′ · n + n · n′ ≡ 0 =⇒ n′ · n ≡ 0
and
n·u≡0 =⇒ n′ · u + n · u′ ≡ 0 =⇒ n′ · u ≡ 0,
so
n′ = (t · n′ )t + (n · n′ )n + (u · n′ )u = (t · n′ )t.
Furthermore, t · n ≡ 0 and therefore
0 ≡ t′ · n + t · n′ = κn · n + t · n′ =⇒ −κ = t · n′ ,
from which it follows that n′ = −κt.

90
Solution to Exercise 1.25
Note that b is constant, by the third Serret–Frenet equation, and

(γ · b)′ = γ ′ · b + γ · b′ = t · b + γ · 0 ≡ 0.

Hence γ · b ≡ c for some constant c and the result follows.

Solution to Exercise 1.26


The chain rule gives that

(v ◦ s)′ = (v′ ◦ s)s′ = |γ ′ |(v′ ◦ s)

for any vector field v : (a, b) → R3 . Taking v = γ


e = γ ◦ s−1 , we have

γ ′ = (e γ ′ ◦ s) = |γ ′ |(et ◦ s);
γ ◦ s)′ = |γ ′ |(e (B.1)

taking v = et shows that



(t ◦ s)′ = |γ ′ |(t′ ◦ s) = |γ ′ | (e
κne) ◦ s , (B.2)

by the definition of n. It follows from differentiating (B.1) that

γ ′′ = |γ ′ |′ (et ◦ s) + |γ ′ |(et ◦ s)′



= |γ ′ |′ (et ◦ s) + |γ ′ |2 (e
κne) ◦ s , (B.3)

where the last equality holds by (B.2). Hence



γ ′ × γ ′′ = |γ ′ |(et ◦ s) × γ ′′ = |γ ′ |3 (e e ◦s ,
κb) (B.4)

e has unit magnitude.


as required. The deduction is immediate, since b
For the next part, it follows by differentiating (B.4) that

γ ′ × γ ′′′ = γ ′′ × γ ′′ + γ ′ × γ ′′′
= (γ ′ × γ ′′ )′

= |γ ′ |3 (e e ◦ s) + |γ ′ |3 (e
κ ◦ s )′ (b e ◦ s)′
κ ◦ s)(b
′ 
= |γ ′ |3 (e
κ ◦ s) (b e ◦ s) − |γ ′ |4 (eκτen) ◦ s ,

by the initial observation and the third Serret–Frenet equation. Thus



[γ ′ , γ ′′ , γ ′′′ ] = −γ ′′ · (γ ′ × γ ′′′ ) = |γ ′ |6 (e
κ2 τe) ◦ s ,

by (B.3), as claimed. The final result follows.

91
B. Solutions to selected exercises

Solution to Exercise 2.2


For injectivity, note that
v γ(u) = x γ(y) =⇒ vd = xd =⇒ v = x,
where the first implication follows from taking the scalar product of each side with c,
and then γ(u) = γ(y), whence u = y.
For regularity, note that
∂1 r(u, v) = v γ ′ (u) and ∂2 r(u, v) = γ(u),
so (∂1 r×∂2 r)(u, v) = v γ ′ (u)×γ(u), which has positive magnitude unless γ ′ (u) and γ(u)
are linearly dependent. However, γ(u) · c = d and γ ′ (u) · c = 0; if a γ(u) + b γ ′ (u) = 0
then
0 = 0 · c = a γ(u) · c + b γ ′ (u) · c = ad,
so a = 0, but then |b γ ′ (u)| = |0| = 0 and (since γ is regular) b = 0.

Solution to Exercise 2.3


As
∂1 r(z, φ) = (cos φ, sin φ, 1) and ∂2 r(z, φ) = (−z sin φ, z cos φ, 0),
it follows that

i j k


(∂1 r × ∂2 r)(z, φ) = cos φ sin φ 1 = (−z cos φ, −z sin φ, z)
−z sin φ z cos φ 0

and q √
|∂1 r × ∂2 r|(z, φ) = z 2 cos2 φ + z 2 sin2 φ + z 2 = 2 |z|.
Hence ∂1 r(0, φ) and ∂2 r(0, φ) are linearly dependent (which may also be seen directly,
since ∂2 r(0, φ) is the zero vector) and r is not regular.
This surface is the double cone x2 + y 2 = z 2 , minus the half lines {(−z, 0, z) : z 6= 0};
it fails to be regular at (0, 0, 0), where the apices of the two cones meet.

Solution to Exercise 2.13


The first Serret–Frenet equation and the fact that (t, n, b) is an orthonormal triple of
vector fields imply that
κn := t′ · N(u, v) = κn · (n cos ψ + b sin ψ) = κ cos ψ.
Furthermore,
B := t × N(u, v) = t × (n cos ψ + b sin ψ) = b cos ψ − n sin ψ,
as required; it follows from this that
κg := t′ · B = κn · (b cos ψ − n sin ψ) = −κ sin ψ.

92
For the last part, note that the first equation holds by definition. For the others, the
second and third Serret–Frenet equations and the previous working give that

N(u, v)′ = n′ cos ψ − ψ ′ n sin ψ + b′ sin ψ + ψ ′ b cos ψ


= (−κt + τ b) cos ψ + ψ ′ B − τ n sin ψ
= −κ cos ψ t + (τ + ψ ′ )B
= −κn t + τg B

and

B′ = −n′ sin ψ − ψ ′ n cos ψ + b′ cos ψ − ψ ′ b sin ψ


= (κt − τ b) sin ψ − ψ ′ N(u, v) − τ n cos ψ
= κ sin ψ t − (τ + ψ ′ )N(u, v)
= −κg t − τg N(u, v).

Solution to Exercise 2.15


For the first part, note that

r·r ≡1 =⇒ ∂1 r · r = ∂2 r · r ≡ 0,

so the unit vector field r is orthogonal to the tangent plane everywhere, and thus N = ±r.
The unit-speed curve γ = r ◦ (u, v) is a geodesic if and only if the component of γ ′′ in
the tangent plane is zero:

0 ≡ γ ′′ − (N(u, v) · γ ′′ )N(u, v) = γ ′′ − (γ · γ ′′ )γ,

by the first part. However,

γ·γ ≡1 =⇒ γ′ · γ ≡ 0 =⇒ γ ′′ · γ + γ ′ · γ ′ ≡ 0

and γ ′ · γ ′ ≡ 1, so γ is a geodesic if and only if γ ′′ + γ ≡ 0, as claimed.


Solving for each component separately,

γ(s) = a1 cos s + b1 sin s, a2 cos s + b2 sin s, a3 cos s + b3 sin s) = a cos s + b sin s

for all s ∈ I. If 0 ∈ I then

|a| = |γ(0)| = 1, |b| = |γ ′ (0)| = 1 and a · b = (γ · γ ′ )(0) = 0,

so a and b are orthogonal unit vectors. For the general case, fix s0 ∈ I and note that

e = r ◦ (e
γ v ) : I − s0 → S 2 ; s 7→ γ(s + s0 ) = r u(s + s0 ), v(s + s0 )
u, e

is a geodesic if and only if γ is, because


 
e ′′ (s) = γ ′′ (s + s0 )
γ and N u(s + s0 ), v(s + s0 ) = N u
e(s), e
v(s) .

93
B. Solutions to selected exercises

Hence, by the previous working, there exist orthogonal unit vectors e e such that
a and b
e (s − s0 ) = e
γ(s) = γ e sin(s − s0 ) = a cos s + b sin s
a cos(s − s0 ) + b
for all s ∈ I, where

a := e e sin s0
a cos s0 − b and b := e e cos s0 ;
a sin s0 + b

since e e are orthogonal unit vectors, so are a and b.


a and b

Solution to Exercise 2.16


If γ = r ◦ (u, v) is a geodesic then κg ≡ 0, therefore

κn = t′ = κn N(u, v) + κg B = κn N(u, v).

Hence n = ±N(u, v) (with the sign fixed everywhere on γ) and

b = t × n = ±t × N(u, v) = ±B
(with the same sign as for n). By the third Serret–Frenet equation and its analogue from
Exercise 2.13,
∓τ n = ±b′ = B′ = −κg t − τg N(u, v) = ∓τg N(u, v)

and thus τg = τ , as required.

Solution to Exercise 2.18


The surface of revolution r given by (2.13) is such that


′ i j k

(∂1 r × ∂2 r)(s, φ) = f (s) cos φ f (s) sin φ g (s)
′ ′

−f (s) sin φ f (s) cos φ 0

= −f (s)g ′ (s) cos φ, −f (s)g ′(s) sin φ, f (s)f ′ (s) .

Hence the squared magnitude

|∂1 r × ∂2 r|2 (s, φ) = f (s)2 (g ′(s)2 + f ′ (s)2 ) = f (s)2 > 0


for all (s, φ) ∈ I × (−π, π) and therefore r is regular.

Solution to Exercise 2.20


For a meridian, u′ ≡ 1 and v ′ ≡ 0, so both of the geodesic equations (2.14) are satisfied.
For a parallel, u ≡ s0 (so u′ ≡ 0) and v ′ ≡ 1; the equations become

0 = f (s0 )f ′ (s0 ) and 2f (s0 )f ′ (s0 )u′ = 0.


The latter is always satisfied and, because f (s0 ) > 0, the former is satisfied if and only
if f ′ (s0 ) = 0.

94
Solution to Exercise 2.22
Let γ : I → P∗ be a regular curve such that

γ(a) = a = (−1, 0, 0) and γ(b) = b = (1, 0, 0)

for some a, b ∈ I with a < b. It follows from (1.6) that


Z b
ℓ(γ|(a,b) ) = |γ ′ (t)| dt > |γ(b) − γ(a)| = 2;
a

suppose for contradiction that we have equality.



Letting γ(t) = x(t), y(t), z(t) , we see that
Z b Z bp Z b
2 = x(b) − x(a) = ′
x (t) dt 6 x′ (t)2 + y ′(t)2 + z ′ (t)2 dt = |γ ′ (t)| dt = 2,
a a a

so equality holds throughout. Hence


Z bp
x′ (t)2 + y ′ (t)2 + z ′ (t)2 − x′ (t) dt = 0,
a

and as the integrand is non-negative, it must be zero everywhere on [a, b]. In particular,
y ′ (t) = z ′ (t) = 0 for all t ∈ (a, b), so y(t) = y(a) = 0 and z(t) = z(a) = 0 for all
t ∈ [a, b]. Since x(a) < 0 and x(b) > 0, continuity and the intermediate-value theorem
(Theorem A.2) give the existence of c ∈ (a, b) such that x(c) = 0, but then it follows
that γ(c) = x(c), y(c), z(c) = 0 6∈ P∗ . This is the desired contradiction.
For the second part, let δ > 0, define the smooth curve

γ : R → P∗ ; t 7→ t, 21 δ(1 − t2 ), 0

and note that


γ ′ (t) = (1, −δt, 0) for all t ∈ R.

Then γ(−1) = a = (−1, 0, 0), γ(1) = b = (1, 0, 0) and


Z 1√ Z 1 √
ℓ := ℓ(γ|(−1,1) ) = 2 2
1 + δ t dt = 2 1 + δ 2 t2 dt. (B.5)
−1 0

If α := sinh−1 δ > 0 and x := sinh−1 (δt) then


Z Z
2 α 2 1 α 1h iα α
ℓ= cosh x dx = 1 + cosh 2x dx = x + 12 sinh 2x = + cosh α,
δ 0 δ 0 δ 0 sinh α

since 2 cosh2 x = 1 + cosh 2x and sinh 2x = 2 sinh x cosh x for all x ∈ R. It is clear from
(B.5) that ℓ is an increasing function of δ, and as δ → 0+, so α → 0+ and ℓ → 2. Hence
for all ε > 0 there exists δ > 0 such that ℓ < 2 + ε, as required.

95
B. Solutions to selected exercises

Solution to Exercise 2.24


As the principal directions t1 = ±TC ∂1 r and t2 = ±TC ∂2 r, it follow that
h i   h i 
  t E F 0   1 0 0
t1 · t2 = ±1 0 C C = ±1 0 = 0,
F G ±1 0 1 ±1
as claimed.

Solution to Exercise 2.26


If F is stationary at (x, y) then
∂F
0= = 2Lx + 2My − λ(2Ex + 2F y) (B.6)
∂x
∂F
and 0 = = 2Mx + 2Ny − λ(2F x + 2Gy). (B.7)
∂y
Adding x times the first to y times the second gives that

0 = 2(Lx2 + 2Mxy + Ny 2 ) − 2λ(Ex2 + 2F xy + Gy 2) ⇐⇒ λ = κ(x, y)

if f (x, y) = 1. Hence the equations (B.6) and (B.7) may be written as the matrix
equation " #   
L − κE M − κF x 0
2 = .
M − κF N − κG y 0
Note that x and y cannot both be zero, for then f (x, y) = 0. Hence the matrix on the
left-hand side above must be singular and so
" #
L − κE M − κF
det = 0,
M − κF N − κG

as required.

Solution to Exercise 2.29


Note that 
eN
L e −Mf2 (det JΦ) 2
(LN − M 2
) ◦ Φ
e=
K =  = K ◦ Φ,
eG
E e − Fe2 (det JΦ)2 (EG − F 2 ) ◦ Φ
by Theorem 2.26, Exercise 2.23 and (2.8).

Solution to Exercise 2.31


Since ∂1 N and ∂2 N lie in the tangent plane, their cross product is orthogonal to the
tangent plane, so is a multiple of the unit normal N. Hence

∂1 N × ∂2 N = ±|∂1 N × ∂2 N| N.

The claim follows.

96
Solution to Exercise 2.32
Note that ZZ
sAN (Dε ) = [N, ∂1 N, ∂2 N](x, y) dx dy

and ZZ
Ar (Dε ) = |∂1 r × ∂2 r|(x, y) dx dy.

Hence, as ε → 0,
ZZ
1
[N, ∂1 N, ∂2 N](x, y) dx dy
sAN (Dε ) πε2 Dε
= ZZ
Ar (Dε ) 1
|∂1 r × ∂2 r|(x, y) dx dy
πε2 Dε

[N, ∂1 N, ∂2 N]
→ (z, w)
|∂1 r × ∂2 r|
N · (K ∂1 r × ∂2 r)
= (z, w) (by Exercise 2.30)
|∂1 r × ∂2 r|
= (K N · N)(z, w)
= K(z, w).

Solution to Exercise 3.6


If V = W ∩ S, where W ⊆ R3 is open, then r−1 −1 −1
i (V ) = ri (W ∩ S) = ri (W ), which is
open because ri is continuous.
2
Hence φi (Si ∩ Sj ) = r−1 −1
i (Si ∩ Sj ) = ri (Sj ) is open in R if Sj is relatively open in S.

97
Bibliography

[1] F. Frenet, Sur les courbes à double courbure, J. de Math. Pures et Appliquées 17
(1852), 437–447. [Cited on page 19.]

[2] T. W. Gamelin, Complex analysis, Springer, New York, 2001. [Cited on page 70.]

[3] P. C. Matthews, Vector calculus, Springer, London, 1998. [Cited on page 71.]

[4] J. R. Munkres, Topology, second edition, Prentice-Hall, New Jersey, 2000. [Cited
on pages 12 and 70.]

[5] G. Peano, Sur une courbe, qui remplit toute une aire plane, Math. Ann. 36 (1890),
157–160. [Cited on page 12.]

[6] A. Pressley, Elementary differential geometry, Springer, London, 2001. [Cited on


pages iii, 38, 48 and 73.]

[7] J. Roe, Elementary geometry, Oxford University Press, Oxford, 1993. [Cited on
pages iii and 35.]

[8] G. B. Segal, Geometry of surfaces, Mathematical Institute lecture notes, Univer-


sity of Oxford, 1986. [Cited on page iii.]

[9] J.-A. Serret, Sur quelques formules relatives à la théorie des courbes à double
courbure, J. de Math. Pures et Appliquées 16 (1851), 193–207. [Cited on page 19.]

[10] J.-A. Serret, Sur un théorème relatif aux courbes à double courbure, J. de Math.
Pures et Appliquées 16 (1851), 499–500. [Cited on page 19.]

[11] C. Thomassen The Jordan–Schönflies theorem and the classification of surfaces,


Amer. Math. Monthly 99 (1992), 116–130. [Cited on pages 82 and 83.]

[12] T. J. Willmore, An introduction to differential geometry, Oxford University


Press, Oxford, 1959. [Cited on page iii.]

99
Index

Numbers set in italic type refer to the page where the term is first defined.

A chart
anti-commutative . ................. 3 coordinate . . . . . . . . . . . . . . . . . . . . 77
antipodal . . . . . . . . . . . . . . . . . . . . . . . 34 circle . . . . . . . . . . . . . . . . . . . . . 7, 10, 16
antipodes . . . . . . . . . . . . . . . . . . . . . . . 34 great . . . . . . . . . . . . . . . . . . . . . 34, 40
approximation osculating . . . . . . . . . . . . . . . . . . . . 19
best circular . . . . . . . . . . . . . . . . . . . 19 closed surface . . . . . . . . . . . . . . . . . 64, 81
best linear . . . . . . . . . . . . . . . . . 11, 23 co-linear . . . . . . . . . . . . . . . . . . . ...... 4
polygonal . . . . . . . . . . . . . . . . . . . . . . 8 compact . . . . . . . . . . . . . . . . . . . . . 64, 81
arc length . . . . . . . . . . . . . . . . . . . . 13, 16 cone
arc-length function . . . . . . . . . . . . . 13, 16 generalised . . . . . . . . . . . . . . . . . . . . 25
Archimedes . . . . . . . . . . . . . . . . . . . . . . 33 connected . . . . . . . . . . . . . . . . . . . . . . . 22
area . . . . . . . . . . .. . . . . . . . . . . . . 31, 33 connected surface . . . . . . . . . . . . . . . . . 81
signed . . . . . . . . . . . . . . . . . . . . . . . 59 continuously differentiable . . . . . . . . . 9, 10
. . . . . . . . . . . . . . . . 78 coordinate chart . . . . . . .
atlas . . . . . . . . . . . . . . . . . . . . . . 77
azimuth . . . . . . . . . . . . . . . . . . . . . . . . 26 cross product . . . . . . . . . . . . . . . ...... 3
curvature . . . . . . . . . . . . . . . . . . . . . . . 16
B Gaussian . . . . . . . . . . . . . . . . . . . . . 57
basis geodesic . . . . . . . . . . . . . . . . 38, 41, 46
orthonormal . . . . . . . . . . . . . . . . . . . . 3 normal . . . . . . . . . . . . . . . . . . . . 38, 51
standard . . . . . . . . . . . . . . . . . . . . . . 1 principal . . . . . . . . . . . . . . . . . . . . . 52
best circular approximation . . . . . . . . . . 19 signed . . . . . . . . . . . . . . . . . . . . . . . 68
bilinear . . . . . . . . . . . . . . . . . . . . . . . . . 2 zero . . . . . . . . . . . . . . . . . . . . . . . . . 16
binormal vector field . . . . . . . . . . . . . . . 19 curve
bottle in a surface . . . . . . . . . . . . . . . . 27, 38
Klein . . . . . . . . . . . . . . . . . . . . . . . . 82 parameterised . . . . . . . . . . . . ...... 8
boundary . . . . . . . . . . . . . . . . . . . . . . . 70 planar . . . . . . . . . . . . . . . . . . 7, 17, 24
bounded . . . . . . . . . . . . . . . . . . . . . . . . 65 plane . . . . . . . . . . . . . . . . . . . . . . . . 66
rectifiable . . . . . . . . . . . . . . . . ...... 9
C regular . . . . . . . . . . . . . . . . . . . . . . . 13
Cauchy–Schwarz inequality . . . . . . . . . . . 2 simple . . . . . . . . . . . . . . . . . . . . . . . 24
chain rule simple closed . . . . . . . . . . . . . . . . . . 63
vector-valued . . . . . . . . . . . . . . . . . . 11 smooth . . . . . . . . . . . . . . . . . . . . . . 16
characteristic space-filling . . . . . . . . . . . . . . . . . . . 12
Euler . . . . . . . . . . . . . . . . . . . . . 65, 82 curvilinear polygon . . . . . . . . . . . . . 64, 75

101
Index

curvilinear triangle . .... ........... 76 generalised cylinder . . . . . . . . . . ...... 24


cylinder . . . . . . . . . .... 31, 33, 36, 43, 56 genus . . . . . . . . . . . . . . . . . . . . . . . 65, 81
generalised . . . . . .... ........... 24 geodesic . . . . . . . . . . . . . . . . . . . . . 39, 47
cylindrical projection ... ........... 33 geodesic curvature . . . . . . . . . . . 38, 41, 46
geodesic equations . . . . . . . . . . . . . . 42, 45
D geodesic torsion . . . . . . . . . . . . . . . 39, 41
derivative geodesic triangle . . . . . . . . . . . . ...... 76
of a vector-valued function ......... 9 geodesic triangulation . . . . . . . . ...... 83
determinant geometry
Jacobian . . . . . . . . . . . . . . . . . . . . . 32 extrinsic . . . . . . . . . . . . . . . ...... 49
differentiable intrinsic . . . . . . . . . . . . . . . . ...... 49
continuously . . . . . . . . . . . . . . . . . 9, 10 great circle . . . . . . . . . . . . . . . . 34, 40, 48
differentiable manifold . . . . . . . . . . . . . . 79 Green’s theorem . . . . . . . . . . . . ...... 71
direction
principal . . . . . . . . . . . . . . . . . . . . . 53 H
disc helix . . . . . . . . . . . . . . . . . . . . . . . . . . 43
open . . . . . . . . . . . . . . . . . . . . . . . . 21 homeomorphic . . . . . . . . . . . . . . . . 65, 81
dot product . . . . . . . . . . . . . . . . . . . . . . 2 hyperbolic paraboloid . . . . . . . . . . . . . . 27

E I
elevation . . . . . . . .. . . . . . . . . . . . . . . . 26 inequality
elliptic paraboloid . . . . . . . . . . . . . . . . . 29 Cauchy–Schwarz . . . . . . . .......... 2
equations Intermediate-value theorem . . . . . . . . . . 85
geodesic . . . . . . . . . . . . . . . . . . . 42, 45 interval
Serret–Frenet . . . . . . . . . . . . . . . . . . 19 open . . . . . . . . . . . . . . . . . . ....... 8
Euler characteristic . . . . . . . . . . . . . 65, 82 intrinsic geometry . . . . . . . . . . . . . . . . . 49
extrinsic geometry . . . . . . . . . . . . . . . . . 49 Inverse-function theorem . . . . . . . . . . . . 87
isometric . . . . . . . . . . . . . . . . . 35, 37, 47
F
field J
scalar . . . . . . . . . . . . . . . . . . . . . . . 23 Jacobian . . . . . . . . . . . . . . . . . . . . . 32, 86
vector . . . . . . . . . . . . . . . . . . . . . . . 23 Jacobian determinant . . . . . . . . . . . . . . 32
first fundamental form . . . . . . . . . . 29, 51 Jordan curve theorem . . . . . . . . . . . . . . 70
form
first fundamental . . . . . . . . . . . . 29, 51 K
quadratic . . . . . . . . . . . . . . . . . . . . . 51 K (Gaussian curvature) . . . . . . . . . . . . 57
second fundamental . . . . . . . . . . . . . 49 Klein bottle . . . . . . . . . . . . . . . . . . . . . 82
function
arc-length . . . . . . . . . . . . . . . . . 13, 16 L
transition . . . . . . . . . . . . . . . . . . 77, 78 Landau notation . . . ...... . . . . . . . . . 85
latitude . . . . . . . . . . . . . . . . . . . . . . . . 26
G length . . . . . . . . . . . . . . . . . . . . . . 14, 47
Gauss map . . . . . . . . . . . . . . . . . . . . . . 58 approximate . . . . . . . . . . . . . . . . .... 9
Gauss–Bonnet theorem of a rectifiable curve . . . . . . . . . . .... 9
for closed surfaces . . . . . . . . . . . . . . 64 of a smooth curve . . . . . . . . . . . . . 9, 28
for curvilinear polygons . . . . . . . . . . 64 limit
for smooth simple closed curves . . . . . 63 of a vector-valued function . . . . . . . . 10
Gaussian curvature . . . . . . . . . . . . . . . . 57 limit point . . . . . . . . . . . . . . . . . . . . . . 65
generalised cone . . . . . . . . . . . . . . . . . . 25 line . . . . . . . . . . . . . . . . . . . . . . . . . 6, 39

102
longitude . . . . . . . . . . . . . . . . . . . . . . . 26 piecewise linear . . . . . . . . . . . . . . . . . . . . 8
lune . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 planar curve . . . . . . . . . . . . . . . . 7, 17, 24
plane . . . . . . . . . . . . . . . . . . 6, 36, 42, 62
M osculating . . . . . . . . . . . . . . . . . . . . 19
magnitude . . . . . . . . . . . . . . . . . . . . . . . 2 tangent . . . . . . . . . . . . . . . . . . . . . . 23
constant . . . . . . . . . . . . . . . . . . . . . 15 plane curve . . . . . . . . . . . . . . . . . . . . . . 66
manifold simple closed . . . . . . . . . . . . . . . . . . 69
differentiable . . . . . . . . . . . . . . . . . . 79 point
map limit . . . . . . . . . . . . . . . . . . . . . . . . 65
Gauss . . . . . . . . . . . . . . . . . . . . . . . 58 saddle . . . . . . . . . . . . . . . . . . . . . . . 57
smooth . . . . . . . . . . . . . . . . . . . . . . 22 polygon
matrix curvilinear . . . . . . . . . . . . . . . . . 64, 75
Weingarten . . . . . . . . . . . . . . . . . . . 58 polygonal . . . . . . . . . . . . . . . . . . . . . . . . 8
meridian . . . . . . . . . . . . . . . . . . . . . 43, 45 polygonal approximation . . . . . . . . . . . . . 8
mutually orthogonal . . . . . . . . . . . . . . . . 3 positively oriented . . . . . . . . . . . . . . 63, 70

N principal curvature . . . . . . . . . . . . . . . . 52
negatively oriented . . . . . . . . . . . . . . . . 70 principal direction . . . . . . . . . . . . . . . . . 53
Newton . . . . . . . . . . . . . . . . . . . . . . . . 39 principal-axis theorem . . . . . . . . . . . . . . 52
normal curvature . . . . . . . . . . . . . . 38, 51 prism . . . . . . . . . . . . . . . . . . . . . . . . . . 24
product
O cross . . . . . . . . . . . . . . . . . . . . . . . . . 3
open disc . . . . . . . . . . . . . . . . . . . . . . . 21 dot . . . . . . . . . . . . . . . . . . . . . . . . . . 2
open interval . . . . . . . . . . . . . . . . . . . . . 8 scalar . . . . . . . . . . . . . . . . . . . . . . . . 2
open rectangle . . . . . . . . . . . . . . . . . . . 21 scalar triple . . . . . . . . . . . . . . . . . . . . 5
open set . . . . . . . . . . . . . . . . . . . . . 21, 78 vector . . . . . . . . . . . . . . . . . . . . . . . . 3
orientable . . . . . . . . . . . . . . . . . . . . . . . 65 product rule . . . . . . . . . . . . . . . . . . . . . 11
orientable surface . . . . . . . . . . . . . . . . . 81 projection
orthogonal . . . . . . . . . . . . . . . . . . . . . . . 2 cylindrical . . . . . . . . . . . . . . . . . . . . 33
orthogonal parameterisation . . . . . . 30, 35 Pythagoras’s theorem . . . . . . . . . . . . . . . 3
orthonormal basis . . . . . . . . . . . . . . . . . . 3
orthonormal triple . . . . . . . . . . . . . . . 4, 19 Q
osculating circle . . . . . . . . . . . . . . . . . . 19 quadratic form . . . . . . . . . . . . . . . . . . . 51
osculating plane . . . . . . . . . . . . . . . . . . 19
R
3
R ............................. 1
P
pair rectangle
smooth . . . . . . . . . . . . . . . . . . . . . . 28 open . . . . . . . . . . . . . . . . . . . . . . . . 21
paraboloid rectifiable curve . . . . . . . . . . . . . . . . . . . 9
elliptic . . . . . . . . . . . . . . . . . . . . . . . 29 regular curve . . . . . . . . . . . . . . . . . . . . 13
hyperbolic . . . . . . . . . . . . . . . . . . . . 27 regular surface patch . . . . . . . . . . . . 23, 33
parallel . . . . . . . . . . . . . . . . . . . . . . 43, 45 relatively open set . . . . . . . . . . . . . . . . . 78
parameterisation reparameterisation . . 14, 32, 33, 37, 40, 50
orthogonal . . . . . . . . . . . . . . . . . 30, 35 orientation-preserving . . . . . . . . . . . . 32
parameterised curve . . . . . . . . . . . . . . . . 8 orientation-reversing . . . . . . . . . . . . . 32
patch smooth . . . . . . . . . . . . . . . . . . . . . . 16
surface . . . . . . . . . . . . . . . . . . . . . . . 23 right-handed triple . . . . . . . . . . . . . . . . . 4
period . . . . . . . . . . . . . . . . . . . . . . . . . 69 rule
perpendicular . . . . . . . . . . . . . . . . . . . . . 2 product . . . . . . . . . . . . . . . . . . . . . . 11

103
Index

S tomb . . . . . . . . . .. . . . . . . . . . . . . ... 33
saddle point . . . . . . . . . . . . . . . . . . . . . 57 torsion . . . . . . . . .. . . . . . . . . . . . . ... 19
scalar field . . . . . . . . . . . . . . . . . . . 15, 23 geodesic . . . . . .. . . . . . . . . . . . . 39, 41
scalar product . . . . . . . . . . . . . . . . . . . . . 2 zero . . . . . . . . .. . . . . . . . . . . . . ... 20
scalar triple product . . . . . . . . . . . . . . . . 5 torus . . . . . . . . . .. . . . . . . . . . . . . ... 37
second fundamental form . . . . . . . . . . . . 49 transition function . . . . . . . . . . . . . 77, 78
Serret–Frenet equations . . . . . . . . . . 19, 39 transpose . . . . . . .. . . . . . . . . . . . . ... 32
set triangle
open . . . . . . . . . . . . . . . . . . . . . 21, 78 curvilinear . . . . . . . . . . . . . . . . . . . . 76
relatively open . . . . . . . . . . . . . . . . . 78 geodesic . . . . . . . . . . . . . . . . . . . . . . 76
signed area . . . . . . . . . . . . . . . . . . . . . . 59 spherical . . . . . . . . . . . . . . . . . . . . . 34
signed curvature . . . . . . . . . . . . . . . . . . 68 triangulation . . . . . . . . . . . . . . . . . . . . 83
simple closed curve . . . . . . . . . . . . . . . . 63 geodesic . . . . . . . . . . . . . . . . . . . . . . 83
simple curve . . . . . . . . . . . . . . . . . . . . . 24 triple
smooth curve . . . . . . . . . . . . . . . . . . . . 16 orthonormal . . . . . . . . . . . . . . . . . 4, 19
smooth map . . . . . . . . . . . . . . . . . . . . . 22 right-handed . .................. 4
smooth pair . . . . . . . . . . . . . . . . . . . . . 28
space-filling curve . . . . . . . . . . . . . . . . . 12 U
speed . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 umbilic . . . . . . . . . . . . . . . . . . . . . . 53, 55
constant . . . . . . . . . . . . . . . . . . . . . 40 unit normal vector . . . . . . . . . . . . . ... 17
sphere . . . . . . . . . . . . 7, 31, 40, 55, 62, 80 unit normal vector field . . . . . . . . . . 17, 32
unit . . . . . . . . . . . . . . . . . . . . . . . . . 26 unit sphere . . . . . . . . . . . . . . . . . . . ... 26
spherical triangle . . . . . . . . . . . . . . . . . 34 unit tangent vector . . . . . . . . . . . . . ... 15
standard basis . . . . . . . . . . . . . . . . . . . . 1 unit tangent vector field . . . . . . . . . ... 15
subdivision . . . . . . . . . . . . . . . . . . . 65, 82
surface . . . . . . . . . . . . . . . . . . . . . . . . . 78 V
closed . . . . . . . . . . . . . . . . . . . . 64, 81 vector
connected . . . . . . . . . . . . . . . . . . . . 81 unit normal . . . . . . . . . . . . . . . . . . . 17
of revolution . . . . . . . . . . . . . . . . . . 44 unit tangent . . . . . . . . . . . . . . . . . . . 15
orientable . . . . . . . . . . . . . . . . . . . . 81 zero . . . . . . . . . . . . . . . . . . . . ...... 1
surface patch . . . . . . . . . . . . . . . . . . . . 23 vector field . . . . . . . . . . . . . . . . . . . 15, 23
regular . . . . . . . . . . . . . . . . . . . . 23, 33 binormal . . . . . . . . . . . . . . . . . . . . . 19
unit normal . . . . . . . . . . . . . 17, 25, 32
T unit tangent . . . . . . . . . . . . . . . . . . . 15
tangent plane . . . . . .. . . . . . . . . . . . . . 23 vector product . . . . . . . . . . . . . ....... 3
tangent vector fields .. . . . . . . . . . . . . . 23 velocity . . . . . . . . . . . . . . . . . . ....... 9
Taylor’s theorem . . . .. . . . . . . . . . . . . . 85
theorem W
Green’s . . . . . . . . . . . . . . . . . . . . . . 71 Weingarten matrix . . . . . . . . . . . . . . . . 58
Jordan curve . . . . . . . . . . . . . . . . . . 70
principal-axis . . . . . . . . . . . . . . . . . . 52 Z
Pythagoras’s . . . . . . . . . . . . . . . . . . . 3 zero torsion . . . . . . . . . . . . . . . . . . . . . 20
Theorema Egregium . . . . . . . . . . . . . . . 59 zero vector . . . . . . . . . . . . . . . . . . . . . . . 1

104

You might also like