M3P5 Geometry of Curves and Surfaces

Lectured by Prof Tom Coates

Typed by David Kurniadi Angdinata
Autumn 2018

⇓
g−2

⇓
g−2

χ (Σ1,1 ) = −1 χ (Σ1,2 ) = −2 χ (Σ1,g−1 ) = −2 χ (Σ1,g ) = −1

X X
χ (Σg ) = χ (Σ1,i ) = (Vi − Ei + Fi ) = 2 − 2g
i i

Syllabus
Parametrisations of curves in three-dimensional space. Curvature. Torsion. Frenet–Serret formulae.
Winding number. Charts of surfaces. Tangent vectors. Tangent planes. Smooth maps between surfaces.
Normal vectors. The first fundamental form. The second fundamental form. Christoffel symbols. Normal
curvature. Gaussian curvature. Mean curvature. Gauss’s Theorema Egregium. Area of surfaces. Length-
minimising curves. Geodesic curvature. The Gauss–Bonnet theorem and applications.

1
M3P5 Geometry of Curves and Surfaces Contents

Contents
0 Introduction 3

1 Curves in three-dimensional space 4

1.1 What is a curve? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Space curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Plane curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2 Surfaces 12
2.1 Regular surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Tangent vectors and tangent planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Normal vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Curvature 18
3.1 The second fundamental form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Normal curvature, Gaussian curvature, and mean curvature . . . . . . . . . . . . . . . . . . . 19
3.3 What does Gaussian curvature mean? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.4 How to compute Gaussian and mean curvature? . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.5 The first fundamental form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.6 Christoffel symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.7 Theorema Egregium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

4 Area of surfaces 29
4.1 Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

5 Geodesics 31
5.1 Geodesic curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.2 Length-minimising curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

6 The Gauss-Bonnet theorem and applications 34

6.1 Local version of Gauss-Bonnet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.2 Gauss-Bonnet for curvilinear triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.3 Gauss-Bonnet theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2
M3P5 Geometry of Curves and Surfaces 0 Introduction

0 Introduction
Lecture 1
A question is what does it mean for a surface to be curved, more curved, less curved, or differently curved? Friday
In fact, what does it mean for a curve to be curved? The goal is to answer these questions. Touches on 05/10/18
ˆ manifolds, smooth shapes, and differential geometry,

ˆ topology, and

ˆ Riemannian geometry and general relativity.

The following are references.

ˆ C Bär, Elementary differential geometry, 2010

A tentative outline of the material that we will cover is as follows. This may change as the term progresses.
If so then an updated outline will be given during lectures.
ˆ Curves in three-dimensional space.

– Parametrisations.
– Curvature and torsion, Frenet–Serret formulae.
– Curves are determined by curvature and torsion.
– Winding number.
ˆ Surfaces.

– Charts.
– Tangent vectors and tangent planes.
– Smooth maps between surfaces.
– Normal vectors.
ˆ Curvature.

– The first and second fundamental forms.

– Christoffel symbols.
– Normal curvature, Gaussian curvature, and mean curvature.
– Gauss’s Theorema Egregium.

ˆ Area of surfaces.

ˆ Geodesics.

– Length-minimising curves.
– Existence, non-existence, and examples.
– Geodesic curvature.
ˆ The Gauss–Bonnet theorem and applications.

ˆ The topological classification of surfaces.

ˆ Vector fields and the Poincaré–Hopf theorem.

3
M3P5 Geometry of Curves and Surfaces 1 Curves in three-dimensional space

1 Curves in three-dimensional space

The ultimate goal is surfaces. First is curves. What does it mean for a curve to be curved? In fact, what is
a curve?

1.1 What is a curve?

Definition 1.1.1. The n-dimensional Euclidean space Rn consists of
 
x1
 .. 
(x1 , . . . , xn ) =  .  , xi ∈ R.
xn
Rn is a vector space and an inner product space. If x = (x1 , . . . , xn ) and y = (y1 , . . . , yn ) then
n
X
x·y = xi yi
i=1

is the inner product of x and y. The length of x ∈ Rn is

v
u n
√ uX
|x| = x · x = t x2 . i
i=1

Focus on R2 and R3 .
Definition 1.1.2. A parametrised curve in Rn is a smooth map
φ : [a, b] → Rn .
It is regular if
φ0 (t) 6= 0, t ∈ [a, b] .
The curve is φ ([a, b]). The parametrised curve says where a particle moving along this curve is positioned
at time t. This is φ (t).
Example.
φ : [0, 2π] −→ R2 φ : [0, 1] −→ R3
,

t 7−→ (cos t, sin t) t 7−→ cos 4πt, sin 4πt, t4
are curves.
Example.
φ : [−1, 1] −→ R2
t 7−→ (t,|t|)
is not a curve.
For us, all parametrised curves will be regular.
Example.
φ : [−1, 1] −→ R2

t 7−→ t2 , t3
is not a curve, since φ0 (t) = 2t, 3t2 and φ0 (0) = (0, 0).

We are not interested in the parametrisation, just the curve.

Example.
φ : [0, 1] −→ R2
t 7−→ (cos 2πt, sin 2πt)
is the curve above.

4
M3P5 Geometry of Curves and Surfaces 1 Curves in three-dimensional space

∼
Definition 1.1.3. Given a regular parametrised curve φ : [a, b] → Rn in Rn and a smooth map f : [c, d] −
→
[a, b], with f 0 (t) 6= 0 for all t, the curve

ϕ = φ ◦ f : [c, d] → Rn

is called a reparametrisation of φ.
∼
→ [a, b] is smooth with f 0 (t) 6= 0 for all t, then ϕ = φ ◦ f
Proposition 1.1.4. If φ is regular and f : [c, d] −
is also a regular parametrised curve.
Proof. Chain rule implies that
ϕ0 (t) = φ0 (f (t)) · f 0 (t) .
Then φ0 (f (t)) is never zero because φ is regular and f 0 (t) is never zero by assumption. Thus φ0 (t) 6= 0.
So will study parametrised curves up to reparametrisation and study reparametrisation-invariant properties
of parametrised curves.
∼
Remark. Reparametrisations f : [c, d] − → [a, b] with f 0 (t) 6= 0 form a groupoid under composition of
functions. The groupoid acts on parametrised curves.
Lecture 2
Say that φ ∼ ϕ if and only if φ is a reparametrisation of ϕ. Then ∼ is an equivalence relation. A curve Monday
is an equivalence class of parametrised curves. The trace of a curve φ : [a, b] → Rn is φ ([a, b]), that is the 08/10/18
image in Rn . If φ ∼ ϕ then the trace of φ is equal to the trace of ϕ, so this is well-defined.
Example. The tangent line to a curve at a point.
Definition 1.1.5. The tangent line L to φ at φ (t0 ) is

L = {φ (t0 ) + sφ0 (t0 ) | s ∈ R} .

∼
Check that this is a reparametrisation-invariant. Suppose f : [c, d] −
→ [a, b] is a reparametrisation and
ϕ = φ ◦ f . Let s0 ∈ [c, d] satisfy f (s0 ) = t0 . Then

ϕ (s0 ) = φ (f (s0 )) = φ (t0 ) , ϕ0 (s0 ) = φ0 (f (s0 )) · f 0 (s0 ) = φ0 (t0 ) · f 0 (s0 ) ,

so
L = {ϕ (s0 ) + s0 ϕ0 (s0 ) | s0 ∈ R} .
Definition 1.1.6. The length of a parametrised curve φ : [a, b] → Rn is
Z b
L (φ) = |φ0 (t)| dt.
a

Chopping up [a, b] into N intervals of size ∆t = (b − a) /N ,

N
X
L (φ) ≈ |φ0 (a + i∆t)| ∆t.
i=0

Example. Let
φ : [0, 2π] −→ R2
.
t 7−→ (cos t, sin t)
p
Then φ0 (t) = (− sin t, cos t), so |φ0 (t)| = sin2 t + cos2 t = 1. Thus
Z 2π
L (φ) = 1 dt = 2π,
0

so the length is 2π.

5
M3P5 Geometry of Curves and Surfaces 1 Curves in three-dimensional space

Proposition 1.1.7. L (φ) is invariant under reparametrisation.

Proof. For simplicity, suppose n = 3, so R3 . Then

φ : [a, b] −→ R3
.
t 7−→ (x (t) , y (t) , z (t))

Thus r
Z b  2
dx 2 dy dz 2



L (φ) = dt + dt + dt dt.
a
∼
→ [a, b], so f is smooth and f 0 (t) 6= 0 for all t, set ϕ = φ ◦ f and write
Given a reparametrisation f : [c, b] −

ϕ (s) = (X (t) , Y (t) , Z (t)) .

Then Z d q
dX 2

dY

2 dZ 2


L (ϕ) = ds + ds + ds ds.
c
Thus use the change of variable formula and
dX dx df
(s) = (f (s)) (s) ,
ds dt ds
by the chain rule.
If |φ0 (t)| = 1 for all t then
Z b
L (φ) = |φ0 (t)| dt = b − a,
a
 
and in fact L φ|[a,t] = t − a, so the curve is parametrised by arc-length. This is an arc-length paramet-
risation or a unit speed parametrisation.
Proposition 1.1.8. Let φ : [a, b] → Rn be a regular curve. Then there exists an arc-length parametrisation
of φ.
Proof. Let
l : [a, b] −→ [0, L]
Rt 0 .
t 7−→ 0
|φ (s)| ds
Set f = l−1 as the inverse function. Then if ϕ = φ ◦ f we have

|ϕ0 (t)| = |φ0 (f (t))| f 0 (t) .

Also l (f (t)) = t, so l0 (f (t)) f 0 (t) = 1. Thus

1 1
f 0 (t) = = 0 ,
l0 (f (t)) |φ (f (t))|

by the fundamental theorem of calculus, so we have |ϕ0 (t)| = 1 as required.

Lecture 3
Is an arc-length parametrisation unique? No. Suppose that φ : [a, b] → Rn is an arc-length parametrisation Tuesday
∼
of a curve, and that f : [c, d] −→ [a, b] is a reparametrisation such that ϕ = φ ◦ f is also an arc-length 09/10/18
parametrisation. Then |φ0 (t)| = 1 for all t ∈ [a, b] and |ϕ0 (t)| = 1 for all t ∈ [c, d], so

|φ0 (f (t))||f 0 (t)| = 1, t ∈ [c, d] ,

so |f 0 (t)| = 1. Thus f 0 (t) = ±1, so f (t) = ±t + C for some constant C.

6
M3P5 Geometry of Curves and Surfaces 1 Curves in three-dimensional space

1.2 Curvature
Definition 1.2.1. Let φ : [a, b] → Rn be a curve parametrised by arc-length. The curvature of φ at φ (t)
is
κ (t) = |φ00 (t)| .
The curvature vector is
~κ (t) = φ00 (t) .

Claim that this depends only on the curve not on the parametrisation φ. If ϕ = φ ◦ f is another arc-length
parametrisation then f (t) = ±t + C. Putting s = f (t) we have

dϕ dφ d2 ϕ d2 φ
=± , = ,
ds dt ds2 dt2
so ~κ (t) is independent of the choice of parametrisation.

Proposition 1.2.2. κ (t) = 0 if and only if the curve is a straight line.

Proof. Let φ : [a, b] → Rn be an arc-length parametrisation of the curve. Then κ (t) = 0, so φ00 (t) = 0. Thus

φ (t) = ~a + ~bt, ~a, ~b ∈ Rn .

Proposition 1.2.3. For curves in Rn , the vector curvature ~κ (t) is perpendicular to the tangent line at φ (t)
for all t, where φ is an arc-length parametrisation.
Proof. The tangent line to the curve at φ (t) is

L = {φ (t) + sφ0 (t) | s ∈ R} ,

where φ0 (t) is the direction vector of the tangent line. Need to show φ0 (t) · φ00 (t) = 0 for all t. Know
|φ0 (t)| = 1 for all t, that is φ0 (t) · φ0 (t) = 1 for all t. Differentiating,

φ00 (t) · φ0 (t) + φ0 (t) · φ00 (t) = 0,

so 2φ0 (t) · φ00 (t) = 0. Thus φ0 (t) · φ00 (t) = 0.

Example. The curvature of a circle in R2 centred at the origin of radius R. Need an arc-length paramet-
risation. Try
φ : [0, 2π] −→ R
.
t 7−→ (R cos t, R sin t)
Then
|φ0 (t)| = |(−R sin t, R cos t)| = R.
Oops. Try
φ : [0, 2πR] −→ R
.
t 7−→ R cos Rt , R sin Rt
Checking,
|φ0 (t)| = − sin Rt , cos Rt


= 1.
The vector curvature is
~κ (t) = − R1 cos Rt , − R1 sin Rt ,

so the curvature is
1
κ (t) = |~κ (t)| = .
R
The curvature κ (t) does not determine the curve.

7
M3P5 Geometry of Curves and Surfaces 1 Curves in three-dimensional space

1.3 Space curves

Let φ : [a, b] → R3 be an arc-length parametrisation of a curve. Set

T (t) = φ0 (t) ,

the unit tangent vector to the curve at φ (t) and assume T0 (t) 6= 0 for all t. Set

T0 (t)
N (t) = ,
|T0 (t)|

the principal normal vector to the curve at φ (t). Set

B (t) = T (t) × N (t) ,

the binormal vector to the curve at φ (t). Check that |B (t)| = 1. Thus (T, N, B) is a positively oriented
orthonormal basis for R3 , for each t ∈ [a, b]. This is the Frenet frame or Serret-Frenet frame. Lecture 4
Friday
T0 (t) = |φ00 (t)| N (t) = κ (t) N (t) . 12/10/18
B0 (t) = T0 (t) × N (t) + T (t) × N0 (t) = T (t) × N0 (t), so B0 (t) is perpendicular to T (t). Then B (t) · B (t) = 1
for all t, so B (t) is perpendicular to B0 (t). Thus

B0 (t) = −τ (t) N (t) ,

for some function τ (t). Then τ (t) is called the torsion of φ at the point φ (t). Then N (t) = B (t) × T (t) so

N0 (t) = B0 (t) × T (t) + B (t) × T0 (t) = −τ (t) N (t) × T (t) + B (t) × κ (t) N (t) = τ (t) B (t) − κ (t) T (t) .

Thus
T0 (t) = κ (t) · N (t) , B0 (t) = −τ (t) · N (t) , N0 (t) = τ (t) · B (t) − κ (t) · T (t) .
So the time evolution of the frame depends only on the curvature, the torsion, and the frame itself.

Proposition 1.3.1. Let φ : [a, b] → R3 be a curve parametrised by arc-length. Suppose φ00 (t) 6= 0 for all
t ∈ [a, b]. Need this for the Frenet frame to exist. Then τ (t) = 0 for all t if and only if φ is planar.
Proof.
=⇒ Suppose τ (t) = 0. Then B0 (t) = 0, so B (t) = v is constant. Claim that φ (t) · v is constant for all t.
0
(φ (t) · v) = φ0 (t) · v + φ (t) · 0 = φ0 (t) · v = T (t) · v = 0,

since T is perpendicular to B. Thus φ is moving in a plane perpendicular to v.

⇐= Suppose φ is planar. Then φ (t)·v = a for some constant v ∈ R3 and constant a ∈ R. Then φ0 (t)·v = 0,
so T (t) is perpendicular to v. Then φ00 (t) · v = 0, so κ (t) · N (t) is perpendicular to v. Then κ (t) 6= 0,
so N (t) is perpendicular to v. Thus B (t) = ±v/|v| is constant, so τ (t) = 0.

Exercise. The unit circle in the (x, y) plane is

φ (t) = (cos t, sin t, 0) , t ∈ [0, 2π] .

Compute T (t) , N (t) , B (t) , τ (t) , κ (t). Verify the Frenet formulae.

8
M3P5 Geometry of Curves and Surfaces 1 Curves in three-dimensional space

Example. The helix is  

φ (t) = cos √t2 , sin √t2 , √t2 , t ∈ R.
 
It is an arc-length parametrisation, since φ0 (t) = − √12 sin √t2 , √12 cos √t2 , √12 , so |φ0 (t)| = 1. Then
 
T (t) = φ0 (t) = − √12 sin √t2 , √12 cos √t2 , √12 ,
φ00 (t)    
N (t) = 00 = 2 − 21 cos √t2 , − 12 sin √t2 , 0 = − cos √t2 , − sin √t2 , 0 ,
|φ (t)|
 
B (t) = T (t) × N (t) = √12 sin √t2 , − √12 cos √t2 , √12 .
 
Then κ (t) = |φ00 (t)| = 12 and τ (t) = 21 , since B0 (t) = 21 cos √t2 , 12 sin √t2 , 0 = −τ (t) · N (t).
In general κ and τ are not constants. Lecture 5
Monday
Theorem 1.3.2 (Fundamental theorem of the local theory of curves). Given two differentiable functions
15/10/18
κ : [a, b] → R and τ : [a, b] → R such that κ (t) > 0 for all t, there exists a regular curve φ : [a, b] → R3
parametrised by arc-length with curvature κ (t) and torsion τ (t). Any other such curve ϕ : [a, b] → R3 differs
from φ by a rigid motion, that is
ϕ = g · φ + ~c, g ∈ SO (3) , ~c ∈ R3 .
Proof.
ˆ Proof of uniqueness. Rigid motions preserve arc-length, curvature, and torsion. If φe = g · φ + ~c then

– φe0 = |g · φ0 | = |φ0 | = 1,

– κφe = φe00 = |g · φ00 | = |φ00 | = κφ , and

– Tφe = φe0 = g · φ0 = g · Tφ and Nφe = T0φe/ T0φe = g · T0φ / T0φ = g · Nφ , so

Bφe = Tφe × Nφe = g · Tφ × g · Nφ = g · (Tφ × Nφ ) = g · Bφ ,

so
0
−τφe · Nφe = B0φe = (g · Bφ ) = g · B0φ = g · (−τφ · Nφ ) = −τφ · Nφe.
Thus τφe = τφ .
We can apply a rigid motion to ϕ which sends ϕ (a) to φ (a) and Frenet frame (Tϕ (a) , Nϕ (a) , Bϕ (a))
to (Tφ (a) , Nφ (a) , Bφ (a)). We can assume that these are equal without loss of generality. We compute
1 d
(hTφ − Tϕ , Tφ − Tϕ i + hNφ − Nϕ , Nφ − Nϕ i + hBφ − Bϕ , Bφ − Bϕ i)
2 dt
= Tφ − Tϕ , T0φ − T0ϕ + Nφ − Nϕ , N0φ − N0ϕ + Bφ − Bϕ , B0φ − B0ϕ
= κ hTφ − Tϕ , Nφ − Nϕ i + (τ hNφ − Nϕ , Bφ − Bϕ i − κ hNφ − Nϕ , Tφ − Tϕ i) − τ hBφ − Bϕ , Nφ − Nϕ i
= 0,
by the Frenet equations, and φ and ϕ are solutions to the problem in the statement. Then
hTφ − Tϕ , Tφ − Tϕ i + hNφ − Nϕ , Nφ − Nϕ i + hBφ − Bϕ , Bφ − Bϕ i
is constant in time and is zero at t = a, so it is zero for all t ∈ [a, b], so
Tφ = Tϕ , Nφ = Nϕ , Bφ = Bϕ .
In particular Z t Z t
φ (t) = φ (a) + Tφ (s) ds = ϕ (a) + Tϕ (s) ds = ϕ (t) .
a a
The conclusion is φ = ϕ.

9
M3P5 Geometry of Curves and Surfaces 1 Curves in three-dimensional space

ˆ Proof of existence. Given κ (t) > 0 and τ (t), we can pick any positive orthonormal frame (Ta , Na , Ba )
and use the existence theorem for solutions of linear differential equations to find T, N, B : [a, b] → R3
such that    

 T (a) Ta

N (a) = N
    
a

   


 B (a) Ba
     .
0


 T 0 κ 0 T
 0 
N  = −κ 0 τ  N

  



B0 0 −τ 0 B


We check that (T, N, B) is an orthonormal frame at all t ∈ [a, b]. This is true at t = a. We consider
   
0 −κ 0 0 −κ 0
M 0 = T0 N0 B0 = T N B κ 0 −τ  = M κ 0 −τ  .




M= T N B ,
0 τ 0 0 τ 0

As M is orthonormal if and only if M | · M = id, we want to prove that M | · M = id, so M | · M = id

at t = a, and
   
0 κ 0 0 −κ 0
d
(M | · M ) = M 0| · M + M | · M 0 = −κ 0 τ  · M | · M + M | · M · κ 0 −τ  .
dt
0 −τ 0 0 τ 0

The linear system     

 0 κ 0 0 −κ 0
 d (A) = −κ 0


τ  A + A κ 0 −τ 
  
dt



 0 −τ 0 0 τ 0
A (a) = id


has solution A (t) = id. Hence by uniqueness, A (t) = id = M | · M .

Exercise. Why is (T (t) , N (t) , B (t)) orthonormal for all t equivalent to the skew-symmetry of
 
0 κ 0
−κ 0 τ ?
0 −τ 0
Lecture 6
Tuesday
Example. Any curve φ : [a, b] → R3 with zero torsion and constant curvature c > 0 is an arc of a circle of
16/10/18
radius 1/c. An arc of a circle has this property, since

φ (t) = R cos Rt , R sin Rt , 0 ,

φ0 (t) = − sin Rt , cos Rt , 0 ,

φ00 (t) = − R1 cos Rt , − R1 sin Rt , 0 ,

so κ (t) = 1/R = c and τ (t) = 0, since the circle is a planar curve. Then apply the fundamental theorem of
the local theory of curves.

10
M3P5 Geometry of Curves and Surfaces 1 Curves in three-dimensional space

1.4 Plane curves

Let φ : [a, b] → R3 . Then φ0 (t) = 0 is a point, κ (t) = 0 is a line, and τ (t) = 0 is a plane. Let
φ : [a, b] −→ R2
t 7−→ (x (t) , y (t))
be parametrised by arc-length, so φ0 (t) = (x0 (t) , y 0 (t)) = 1 and N (t) = (−y 0 (t) , x0 (t)). Then φ00 (t) =
κ (t) N (t), so
κ (t) = κ (t) hN (t) , N (t)i = hκ (t) N (t) , N (t)i = hφ00 (t) , N (t)i
= h(x00 (t) , y 00 (t)) , (−y 0 (t) , x0 (t))i = x0 (t) y 00 (t) − y 0 (t) x00 (t) .
Proposition 1.4.1. For arbitrary φ,
φ00 (t) · N (t)
κ (t) = 2 .
|φ0 (t)|
Proof. Let ϕ (s) = φ (f (s)) = (x (f (s)) , y (f (s))) be a reparametrisation by arc-length. Then
0 00 0 00 3
κϕ (s) = (x (f (s))) (y (f (s))) − y (f (s)) x (f (s)) = (x0 (f (s)) y 00 (f (s)) − y 0 (f (s)) x00 (f (s))) · (f 0 (s))
0 0
2 2
= φ00 (f (s)) · − (y (f (s))) , (x (f (s))) · (f 0 (s)) = (φ00 (f (s)) · Nϕ (f (s))) · (f 0 (s)) .
Rt
We have f = l−1 and l (t) = a |φ0 (s)| ds, so in particular f 0 (s) = 1/|φ0 (f (s))|. Hence replace f (s) = t.
Example. Let
φ : [a, b] −→ R2
t 7−→ (t, f (t))
be a graph of a smooth function f . Then
2 (−f 0 (t) , 1)
φ0 (t) = (1, f 0 (t)) , φ00 (t) = (0, f 00 (t)) , |φ0 (t)| = 1 +|f 0 (t)| , N (t) = q ,
2
1 +|f 0 (t)|
so
f 00 (t)
κ (t) = q 3.
0 2
1 + (f (t))
Suppose that φ : [a, b] → R2 ∼ = C is a smooth closed curve, so φ (a) = φ (b) and φ(k) (a) = φ(k) (b) for all k.
The winding number of φ is an invariant measuring the number of clockwise rotations of the curve. Let
z = x + iy. We know from complex analysis
I Z b
1 1 1 1
w (φ) = dz = d(x (t) + iy (t))
2πi γ z 2πi a x (t) + iy (t)
Z b 0 Z b
1 x (t) + iy 0 (t) x (t) − iy (t) 1 (xx0 + yy 0 ) + i (y 0 x − x0 y)
= · dt = dt
2πi a x (t) + iy (t) x (t) − iy (t) 2πi a x2 + y 2
Z b 0 Z b 0
1  2 2 b

 1 y x − x0 y 1 y x − x0 y
= ln x + y a + dt = dt,
4πi 2π a x2 + y 2 2π a x2 + y 2
since φ is a closed curve. If φ is parametrised by arc-length, so |φ0 | = 1, then
Z b
0 1
w (φ ) = κ (t) dt.
2π a
Proposition 1.4.2. The winding number w (T (t)) of the unit tangent vector T (t) = φ0 (t) /|φ0 (t)| is equal
Rb
to the total curvature a κ (t) dt, divided by 2π.
Definition 1.4.3. The winding number w (T (t)) is called the index or turning number of φ, and it is
denoted Ind φ. If φ0 (t) = eiθt , then κ (t) = θ̇ (t). Thus
Z b
1
Ind φ = θ̇ (t) dt.
2π a

11
M3P5 Geometry of Curves and Surfaces 2 Surfaces

2 Surfaces
Lecture 7
2.1 Regular surfaces Friday
19/10/18
Definition 2.1.1. A regular surface S ⊂ R3 is a subset of R3 such that for all p ∈ S there exists
ˆ an open neighbourhood V ⊂ R3 of p,
ˆ an open set U ⊂ R2 , and
ˆ a smooth map φ : U → R3 such that
– φ (U ) = V ∩ S,
– φ is a homeomorphism onto its image, and
– for all q ∈ U , dφq : R2 → R3 is injective.
(U, φ) is called a chart near p or at p.
Write φ (u, v) = (x (u, v) , y (u, v) , z (u, v)), then dφq : R2 → R3 has a matrix
 ∂x ∂x
∂u ∂v
 ∂y ∂y 
,
∂u ∂v
∂z ∂z
∂u ∂v q

and is injective if and only if the matrix has rank two, if and only if the columns are linearly independent.
Then (1, 0) at q ∈ U ⊂ R2 maps to dφq (1, 0) at φ (q) ∈ S ⊂ R3 . Each column matrix of the matrix dφq is
a tangent vector to S at φ (q), so the regularity condition for surfaces ensures that a regular surface has a
tangent plane at every point, since the columns of dφq span a two-dimensional space, that is a plane in R3 .
Example.
ˆ Let R2 ⊂ R3 . Take the chart (U, φ) with U = R2 and φ (u, v) = (u, v, 0).
ˆ Let f : R2 → R be a smooth function. The graph of f is
Γf = (x, y, f (x, y)) x, y ∈ R2 .


Take a chart (U, φ) where U = R2 and φ (u, v) = (u, v, f (u, v)). This is smooth, a homeomorphism
onto its image, by projecting to (x, y) or (u, v) plane, and regular, since
 
1 0
dφ(u,v) =  0 1
∂f ∂f
∂u ∂v
has rank two.
ˆ S = z 2 = x2 + y 2 z ≥ 0 has no smooth chart near (0, 0, 0). Suppose there exists a smooth chart


(U, φ) near (0, 0, 0). Consider the curve z = |y| and x = 0. It has a discontinuous velocity at t = 0.
Any smooth curve through q ∈ U , when φ (q) = 0, maps to a smooth curve in R2 , by assumption, since
we assumed φ is smooth, a contradiction. Thus curves like this are not smooth.
ˆ Define
: (0, 2π) × (−1, 1) −→
φ R3
.
(u, v) 7−→ (sin u, sin 2u, v)
Then φ is not a homeomorphism onto its image.
Lecture 8
The following is a very common situation. Monday
Definition 2.1.2. Let F : R3 → R be a smooth function. We say that S ⊂ R3 is a regular level set of F 22/10/18
if S = F −1 (c) for some c ∈ R and ∇F (p) 6= 0 for all p ∈ S.
Example. Let
S = x 2 + y 2 + z 2 = 1 ⊂ R3 .


Here F = x2 + y 2 + z 2 for c = 1 and ∇F = (2x, 2y, 2z). Note ∇F (p) 6= 0 for all p ∈ S.

12
M3P5 Geometry of Curves and Surfaces 2 Surfaces

Proposition 2.1.3. Regular level sets are regular surfaces.

For this we need the inverse function theorem.
Theorem 2.1.4 (Inverse function theorem). Let f : Rn → Rn be smooth on an open neighbourhood V of
p ∈ Rn , and suppose that dfp : Rn → Rn is invertible. Then there exists an open neighbourhood U ⊂ V of p
such that f |U : U → f (U ) is a diffeomorphism, which is a smooth bijection with a smooth inverse.
Proof of Proposition 2.1.3. The goal is to construct a chart on S near each p ∈ S. Recall S = F −1 (c).
Now ∇F (p) 6= 0, so at least one of ∂∂xF (p) , ∂∂yF (p) , ∂∂zF (p) is non-zero. Without loss of generality assume
∂F
∂z (p) 6= 0. Consider
g : R3 −→ R3
.
(x, y, z) 7−→ (x, y, F (x, y, z))
Then g is smooth, and  
1 0 0
dgp =  0 1 0 
∂F ∂F ∂F
∂x (p) ∂y (p) ∂z (p)
is invertible. On a neighbourhood U of p, g is a diffeomorphism. Let
W = (u, v) ∈ R2 (u, v, c) ∈ g (U ) .


Identify this with {z = c} ∩ g (U ). Let V = U ∩ F −1 (c) and let

φ : W −→ V
.
(u, v) 7−→ g −1 (u, v, c)
Then φ is a chart on S near p, since φ is smooth, a homeomorphism onto its image, and dφq is injective for
all q ∈ W , guaranteed by the inverse function theorem.
Example. The torus is a regular surface. Consider the surface obtained by rotating the circle of radius one
in the xz plane, centred at (2, 0), about z-axis. Writing this in cylindrical polars,
n o p 2
2
S = (r − 2) + z 2 = 1 = F −1 (1) , F (x, y, z) = x2 + y 2 − 2 + z 2 .
Now,  √  √  
2x x2 +y 2 −2 2y x2 +y 2 −2
∇F = √ , √ , 2z .
x2 +y 2 x2 +y 2
p
If ∇F = 0 then 2z = 0 and either x = y = 0, so r = 0, or x2 + y 2 = 2, so r = 2. So ∇F (p) 6= 0 for all
p ∈ S. Thus S is a regular surface.
Proposition 2.1.5. If S is a regular surface then, for every p ∈ S, there exists a neighbourhood V of p in
S such that V is the graph of a smooth function z = f (x, y) or y = g (x, z) or x = h (y, z).
Proof. Take a chart on S at p. Write
φ : U ⊂ R2 −→ S ⊂ R3
.
(u, v) 7−→ (x (u, v) , y (u, v) , z (u, v))
Let q = φ−1 (p). Now dφq has rank two. So one of the 2 × 2 minors of dφq is non-zero. Without loss of
generality assume  ∂x ∂x

∂u (q) ∂v (q) 6= 0.
∂y ∂y
∂u (q) ∂v (q)
Consider
g φ
R2 ← − U −→ R3 ,
(x (u, v) , y (u, v)) ←[ (u, v) 7→ (x, y, F (x, y))
Then g is a diffeomorphism near q, by the inverse function theorem, so
φ ◦ g −1 (x, y) = (x, y, F (x, y)) ,


F (x, y) = z (u (x, y) , v (x, y)) .
Thus, near p, S is the graph z = F (x, y).

13
M3P5 Geometry of Curves and Surfaces 2 Surfaces

2.2 Tangent vectors and tangent planes

Lecture 9
Proposition 2.2.1. Let S be a regular surface, let p ∈ S, and let α : (−, ) → R3 be a regular curve in S Tuesday
with α (0) = p. Let φ : U → R3 be a chart on S near p. There exist smooth functions u, v : (−0 , 0 ) → U 23/10/18
such that α (t) = φ (u (t) , v (t)) for t ∈ (−0 , 0 ), so

(−, )
(u(t),v(t)) α .

q∈U φ
S3p

Proof. Without loss of generality S has the form z = f (x, y). Then
φ (u, v) = (x (u, v) , y (u, v) , f (x (u, v) , y (u, v))) .
−1
Let q = φ (p), and set q = (u0 , v0 ). Define
F : U × (−, ) −→ R3
.
(u, v, t) 7−→ (x (u, v) , y (u, v) , f (x (u, v) , y (u, v)) + t)
Then F (u0 , v0 , 0) = p, and
 ∂x  ∂x
∂u 0 ∂v
∂y ∂y
dF(u0 ,v0 ,0) =  ∂u 0 .
∂v
∂f ∂x ∂f ∂y ∂f ∂x ∂f ∂y
∂x ∂u + ∂y ∂u ∂x ∂v + ∂y ∂v 1
   
∂x ∂y
The first two columns of dφ(u0 ,v0 ) has rank two, so ∂u , ∂u and ∂∂vx , ∂∂vy are linearly independent, so
dF(u0 ,v0 ,0) is invertible. Applying the inverse function theorem, F is a diffeomorphism in a neighbourhood
of (u0 , v0 , 0). Let G = F −1 near p. Then
G (x (u, v) , y (u, v) , f (x (u, v) , y (u, v))) = (u, v, 0) .
Now α (t) = (x (t) , y (t) , f (x (t) , y (t))) for some smooth functions x and y. Setting (u (t) , v (t)) = (G ◦ α) (t)
we find u and v are smooth, because G and α are, and α (t) = φ (u (t) , v (t)).
Definition 2.2.2. Let S be a regular surface, and let p ∈ S. A tangent vector to S at p is a vector in R3
of the form α0 (0), where α : (−, ) → S is a regular curve with α (0) = p. The tangent plane Tp S to S at
p is the set of all tangent vectors.
This is independent of charts. How to compute this?
Proposition 2.2.3. If φ : U → R3 is a chart for S at p, with φ (q) = p, then Tp S is spanned by the columns
of dφq .
Proof. Write dφq = (w1 , w2 ). For all a, b ∈ R, need to show aw1 +bw2 ∈ Tp S. Consider α (t) = φ (q + t (a, b)).
Chain rule implies that α0 (0) = dφq · (a, b) = aw1 + bw2 . So aw1 + bw2 ∈ Tp S. Conversely, need to show any
element of Tp S is of the form aw1 + bw2 for some a, b ∈ R. This is of the form α0 (0) where α : (−, ) → S
is a regular curve with α (0) = p. Given the discussion earlier, we know that there exist smooth functions
u, v : (−0 , 0 ) → U such that α (t) = φ (u (t) , v (t)) for t ∈ (−0 , ). Then
 du   
dt (0) = dφ · a ,
du dv
α0 (0) = dφ(u(0),v(0)) · dv q a= (0) , b= (0) .
dt (0) b dt dt
Thus α0 (0) = aw1 + bw2 as claimed.
Exercise. Consider
S = (x, y, z) x2 + y 2 + z 2 = 1 ,

p = (0, 0, 1) .
Use the chart
φ : U −→ R3 √
,
(u, v) 7−→ u, v, 1 − u2 − v 2
where U is the open unit disc in R2 to compute Tp S as the xy plane.

14
M3P5 Geometry of Curves and Surfaces 2 Surfaces

Lecture 10
Proposition 2.2.4. Let S = F −1 (c) for F : R3 → R then for any p ∈ S,
Friday
⊥ 26/10/18
Tp S = (∇F (p)) .
Proof. Let α : (−, ) → S such that α (0) = p, so α0 (0) ∈ Tp S and F (α (t)) = c. Let α = (x (t) , y (t) , z (t)).
Then  
d ∂ F dx ∂ F dy ∂ F dz
0= F (α (t)) = + + = ∇F (α (0)) · α0 (0) ,
dt t=0 ∂x dt ∂y dt ∂z dt t=0
⊥ ⊥ ⊥
so α0 (0) ∈ (∇F (p)) , so Tp S ⊆ (∇F (p)) , which is two-dimensional. This implies that Tp S = (∇F (p)) .

Example. Let S be the paraboloid

z = x2 + y 2 = F −1 (0) , F = x2 + y 2 − z,


and let p = (1, 3, 10). Then ∇F = (2x, 2y, −1) 6= 0, and ∇F (p) = (2, 6, −1), so
⊥
Tp S = (∇F (p)) = {2x + 6y − z = 0} .
Definition 2.2.5. Let S1 and S2 be regular surfaces. A map F : S1 → R3 is smooth if for every chart
φ F
φ : U → S1 the composition U −
→ S1 −→ R3 is smooth. A map F : S1 → S2 is smooth if it is smooth when
3
viewed as a map F : S1 → R . The differential of a smooth map F : S1 → S2 at p ∈ S1 , denoted
dFp : Tp S1 → TF (p) S2 ,
0
is defined as follows. Let v = α (0) for α : (−, ) → S1 such that α (0) = p and β (t) = F (α (t)) : (−, ) →
S2 . This is a regular curve in S2 , so β 0 (0) ∈ TF (p) S2 , since β (0) = F (p). We define
dFp (v) = β 0 (0) .
Proposition 2.2.6. This definition is independent of the choice of curve α.
Proof. Let α1 , α2 : (−, ) → S1 such that α1 (0) = p and α10 (0) = α20 (0). Let α1 (t) = φ (u1 (t) , v1 (t)) and
α2 (t) = φ (u2 (t) , v2 (t)), so
∂ φ du1 ∂ φ dv1 ∂ φ du2 ∂ φ dv2
(0) + (0) = α10 (0) = α20 (0) = (0) + (0) .
∂u1 dt ∂v1 dt ∂u2 dt ∂v2 dt
dφ dφ
Recall du and dv are linearly independent. This implies that
du1 du2 dv1 dv2
(0) = (0) , (0) = (0) . (1)
dt dt dt dt
Want dFp (α10 (0)) = dFp (α20 (0)).
 
d ∂ (F ◦ φ) dui ∂ (F ◦ φ) dvi
dFp (αi (0)) = F (φ (ui , vi )) = + , i = 1, 2,
dt t=0 ∂u dt ∂v dt t=0

so dFp (α10 (0)) = dFp (α20 (0)) by (1). It follows that dFp (v) is the same if we define it via α1 or α2 .
Proposition 2.2.7. dFp : Tp S1 → TF (p) S2 is linear.
Proof. Let v, w ∈ Tp S1 and c, d ∈ R. Want dFp (cv + dw) = cdFp (v) + ddFp (w). Let α1 , α2 : (−, ) → S1
such that v = α10 (0) and w = α20 (0). Let φ : U → S1 be a chart. Without loss of generality we assume
(0, 0) ∈ U and φ (0, 0) = p. Let αi (t) = φ (ui (t) , vi (t)). Let
α3 (t) = φ (cu1 (t) + du2 (t) , cv1 (t) + dv2 (t)) ,
so α3 (0) = φ (0, 0) = p and
∂φ ∂φ
α30 (0) = (cu01 (0) + du02 (0)) + (cv10 (0) + dv20 (0))
∂u
 ∂v
  
0 ∂φ 0 ∂φ 0 ∂φ 0 ∂φ
= c u1 (0) + v1 (0) + d u2 (0) + v2 (0) = cα10 (0) + dα20 (0) = cv + dw.
∂u ∂v ∂u ∂v

15
M3P5 Geometry of Curves and Surfaces 2 Surfaces

Then
0
dFp (cv + dw) = dFp (α30 (0)) = ((F ◦ φ) (cu1 (t) + du2 (t) , cv1 (t) + dv2 (t))) (0)
∂ (F ◦ φ) ∂ (F ◦ φ)
= (cu01 (0) + du02 (0)) + (cv10 (0) + dv20 (0))
 ∂u ∂u   
∂ (F ◦ φ) 0 ∂ (F ◦ φ) 0 ∂ (F ◦ φ) 0 ∂ (F ◦ φ) 0
=c u1 (0) + v1 (0) + d u2 (0) + v2 (0)
∂u ∂v ∂u ∂v
0 0
= c (F ◦ α1 ) (0) + d (F ◦ α2 ) (0) = cdFp (v) + ddFp (w) ,
since    
0 ∂ (F ◦ φ) dui ∂ (F ◦ φ) dvi
dFp (αi0 (0)) = (F ◦ αi ) (0) = + .
∂u dt ∂v dt

Remark. We will use the following identity. Given a chart φ : U → S1 and F : S1 → S2 ,

   
∂φ ∂ (F ◦ φ) ∂φ ∂ (F ◦ φ)
dFp (u0 , v0 ) = (u0 , v0 ) , dFp (u0 , v0 ) = (u0 , v0 ) .
∂u ∂u ∂v ∂v
Similarly, we define the differential of a smooth function f : S → R if v ∈ Tp S satisfies v = α0 (0) for
α : (−, ) → S as
0
dfp (v) = f (α (t)) (0) .
Exercise. Show that dfp is well-defined, so independent of choice of α, and that it is linear.
Lecture 11
Example. Let Monday
29/10/18
F : R3 −→ R
S = x2 + y 2 + z 2 = 1 = F −1 (1) ,

,
(x, y, z) 7−→ x2 + y 2 + z 2
and let
f : S −→ R
p = (0, 1, 0) , .
(x, y, z) 7−→ z
Then ∇F (p) = (0, 2, 0), so
⊥
Tp S = (∇F (p)) = h(1, 0, 0) , (0, 0, 1)i .
But to compute dfp use the chart near p,
φp : U −→ S √
U = u2 + v 2 < 1 .


,
(u, v) 7−→ u, 1 − u2 − v 2 , v
Then (f ◦ φ) (u, v) = v, so
   
∂φ ∂φ
dfp (0, 0) = 0, dfp (0, 0) = 1.
∂u ∂v
This determines dfp : Tp S → R, a linear map.
Proposition 2.2.8. Let f : S1 → S2 be a smooth map. If dfp : Tp S1 → Tf (p) S2 is an isomorphism, then
there exists an open neighbourhood V ⊆ S1 of p such that f : V → f (V ) is a diffeomorphism.
Proof. Take charts φ1 : U1 → S1 at p and φ2 : U2 → S2 at f (p), so
f
φ1 (q1 ) = p ∈ φ1 (U1 ) ⊆ S1 φ2 (q2 ) = f (p) ∈ φ2 (U2 ) ⊆ S2
φ1 φ2 .

q1 ∈ U1 ⊆ R2 g q2 ∈ U2 ⊆ R2
Check that the map
g : U1 → S1 → S2 → U2
satisfies g (q1 ) = q2 , and dgq1 is invertible. Use the inverse function theorem for g. Then f = φ2 ◦ g ◦ φ−1
1
will be a diffeomorphism on some neighbourhood of φ1 (q1 ) = p.

16
M3P5 Geometry of Curves and Surfaces 2 Surfaces

2.3 Normal vectors

A regular surface S ⊆ R3 has two normal vectors at p in S. If S = F −1 (c) for some smooth function
⊥
F : R3 → R, then Tp S = (∇F (p)) . We can define the unit normal vector
∇F (p)
N (p) = .
|∇F (p)|
In general we can only do this locally, in a chart. Given a chart φ : U → S such that φ (q) = p, we know
that Tp S is spanned by dφ dφ
du (q) and dv (q). Define
∂φ ∂φ
∂u (q) × ∂v (q)
N (p) = .
∂φ ∂φ
∂u (q) × ∂v (q)

This can always be done locally, but not globally. On the Möbius strip, you cannot pick N (p) continuously.
Definition 2.3.1. S is orientable if it admits a continuous choice of unit normal vector N (p) ∈ S. If S is
orientable, we get a map N : S → S2 . Then N is called the Gauss map. If this exists, it is smooth.
Example. Let S = S2 be the unit sphere. As S = F −1 (1) for F (x, y, z) = x2 + y 2 + z 2 ,
∇F (x, y, z) (2x, 2y, 2z)
N (x, y, z) = =p = (x, y, z) .
|∇F (x, y, z)| 4x2 + 4y 2 + 4z 2
√

Or, near the north pole (0, 0, 1) we have a chart φ (u, v) = u, v, 1 − u2 − v 2 , so
∂φ  −u
 ∂φ  −v

= 1, 0, √1−u 2 −v 2
, = 0, 1, √1−u 2 −v 2
,
∂u ∂u
so
∂φ ∂φ  
× = √1−uu2 −v2 , √1−uv2 −v2 , 1 .
∂u ∂v
Thus
∂φ
× ∂∂vφ  p 
N (φ (u, v)) = ∂u = u, v, 1 − u2 − v 2 = φ (u, v) .
∂φ ∂φ
∂u × ∂v

Example. Let
S = {ax + by + cz = d} = F −1 (d) , F (x, y, z) = ax + by + cz
be a plane. Then
∇F (x, y, z) (a, b, c)
N (x, y, z) = =√
|∇F (x, y, z)| a + b2 + c2
2

is a constant map.
What is dNp : Tp S → TN(p) S2 ? TN(p) S2 is all vectors orthogonal to N (p), which is Tp S. We use this
identification to write dNp : Tp S → Tp S.
Example. Let
S = (x, y, z) x2 + y 2 + z 2 = r2 = F −1 r2 , F = x2 + y 2 + z 2


be a sphere of radius r, and let p = (x, y, z). Then

∇F (p) x y z  p
N (p) = = , , = .
|∇F (p)| r r r r
Let α : (−, ) → S be a curve with α (0) = p, so
d ((N ◦ α) (t)) 1 0
dNp (α0 (0)) = = α (0) .
dt t=0 r
Thus dNp : Tp S → Tp S is dNp = id /r.

17
M3P5 Geometry of Curves and Surfaces 3 Curvature

3 Curvature
Lecture 12
We will try to define the curvature using the Gauss map. Tuesday
30/10/18
3.1 The second fundamental form
Example. Let
S = {ax + by + cz = d}
be a plane, and let p ∈ S. Then
(a, b, c)
N (p) = √
a2
+ b2 + c2
is constant with respect to p, so dNp : Tp S → TN(p) S2 is the zero map.
Exercise. Let
S = x2 + 2y 2 + z 2 = r2 .


Compute dNp at (1, 0, 0) and at (0, 1, 0).

Definition 3.1.1. Let S be a regular orientable surface, let p ∈ S. The second fundamental form at p is
A : Tp S × Tp S −→ R
,
(X, Y ) 7−→ −X · dNp (Y )
which is how much Np (Y ) is changing in the direction of X.
Example. If S is a sphere of radius r then
1
A (X, Y ) = − X · Y.
r
Proposition 3.1.2. A is a symmetric bilinear form.
Proof.
ˆ Bilinearity.
A (X, cY1 + dY2 ) = −X · dNp (cY1 + dY2 ) = −X · (cdNp (Y1 ) + ddNp (Y2 ))
= c (−X · dNp (Y1 )) + d (−X · dNp (Y2 )) = cA (X, Y1 ) + dA (X, Y2 ) .
Similarly A (cX1 + dX2 , Y ) = cA (X1 , Y ) + dA (X2 , Y ).
ˆ Symmetry. Take a chart φ : U → S near p. Then dφ dφ
du (q) and dv (q), where φ (q) = p, form a basis for
the tangent space Tp S. It is sufficient to prove that
   
∂φ ∂φ ∂φ ∂φ
A (q) , (q) = A (q) , (q) .
∂u ∂v ∂v ∂u
Observe that
∂φ ∂φ
N (φ (p)) · = 0, N (φ (p)) · = 0,
∂u ∂v
dφ dφ ∂ ∂
since du and dv are in Tp S and N (φ (p)) is normal to Tp S. Applying ∂v and ∂u ,

∂ ∂φ ∂2φ ∂ ∂φ ∂2φ
(N ◦ φ) · + (N ◦ φ) · = 0, (N ◦ φ) · + (N ◦ φ) · = 0.
∂v ∂u ∂v∂u ∂u ∂v ∂u∂v
Then
∂2φ ∂2φ

  
∂φ ∂φ ∂φ ∂φ
, A = (N ◦ φ) · , A , = (N ◦ φ) · .
∂u ∂v ∂v∂u ∂v ∂u ∂u∂v
Thus both these are equal, so    
∂φ ∂φ ∂φ ∂φ
A , =A , .
∂u ∂v ∂v ∂u

Lecture 13
Lecture 13 is a problems class. Friday
02/11/18
18
M3P5 Geometry of Curves and Surfaces 3 Curvature

3.2 Normal curvature, Gaussian curvature, and mean curvature

Lecture 14
There exists an orthonormal basis x1 and x2 for Tp S such that Monday
05/11/18
dNp (xi ) = −λi xi , i = 1, 2.
Then
ˆ A (x1 , x1 ) = −x1 · dNp (x1 ) = x1 · (λ1 x1 ) = λ1 ,
ˆ A (x1 , x2 ) = −x1 · dNp (x2 ) = x1 · (λ2 x2 ) = 0, and
ˆ A (x2 , x2 ) = −x2 · dNp (x2 ) = x2 · (λ2 x2 ) = λ2 .
We call x1 and x2 the principal directions in Tp S and λ1 and λ2 the principal curvatures.
Lemma 3.2.1. Let S be a regular surface and λ1 (p) ≤ λ2 (p) be the principal curvatures at p ∈ S. Then
λ1 = min {A (x, x) | x ∈ Tp S, |x| = 1} , λ2 = max {A (x, x) | x ∈ Tp S, |x| = 1} .
Proof. Consider x ∈ Tp S with |x| = 1. Write x = c1 x1 + c2 x2 where c ∈ R, then c21 + c22 = 1. Then
A (x, x) = A (c1 x1 + c2 x2 , c1 x1 + c2 x2 )
= c21 A (x1 , x1 ) + c1 c2 A (x1 , x2 ) + c2 c1 A (x2 , x1 ) + c22 A (x2 , x2 )
= λ1 c21 + λ2 c22 ≤ λ2 c21 + λ2 c22 = λ2 ,
with equality if and only if c1 = 0 and c2 = 1. Similarly
A (x, x) = λ1 c21 + λ2 c22 ≥ λ1 c21 + λ1 c22 = λ1 ,
with equality if and only if c1 = 1 and c2 = 0.
Example. Let S be a sphere of radius r. Then N (p) = p/r and dNp = id /r, so principal curvatures are
λ1 = λ2 = 1/r, which are independent of p.
Example. Suppose S is a connected regular surface and that λ1 (p) = λ2 (p) = 0 for all p ∈ S. Then
dNp = 0, so N (p) is constant for all p ∈ S. Suppose N (p) = ~v for all p ∈ S. Then S is in the plane
perpendicular to ~v through p. 1
Definition 3.2.2. Suppose C ⊂ S is a curve through p ∈ S. Suppose N is the unit normal vector to S at p
and n is the unit normal vector to C at p. The normal curvature of C at p is
kn (p) = κ cos θ,
where θ is the angle between N and n and κ is the curvature of C at p, so kn (p) is the length of the projection
of ~κ, the vector curvature of C at p, along N.
Proposition 3.2.3. Let S be a regular surface and C ⊂ S be a curve with unit tangent vector v ∈ Tp S.
Then
A (v, v) = kn (p) .
Proof. Let α : (−, ) → S be an arc-length parametrisation of C near p. Then α (0) = p and α0 (0) = v. We
have hα0 (t) , N (α (t))i = 0 for all t. Differentiating,
hα00 (t) , N (α (t))i + α0 (t) , dNα(t) (α0 (t)) ,
for all t. Setting t = 0,
hκnp , N (p)i + hv, dNp (v)i = 0.
Thus
A (v, v) = − hv, dNp (v)i = hκnp , N (p)i = kn (p) .

1 Exercise

19
M3P5 Geometry of Curves and Surfaces 3 Curvature

Definition 3.2.4. A point p ∈ S is umbilical if and only if the principal curvatures λ1 (p) and λ2 (p) are
equal.
Proposition 3.2.5. If S is a connected regular surface such that every point in S is umbilical then S is
contained in a plane or a sphere.
Lecture 15
Tuesday
Proof. By assumption, there exists a smooth function λ : S → R such that λ1 (p) = λ2 (p) = λ (p) for all
06/11/18
p ∈ S.
Step 1. Prove λ (p) is constant. Take a chart φ : U → S near p. Then

∂ ∂φ ∂φ ∂ ∂φ ∂φ
(N ◦ φ) = dNφ(u,v) · = − (λ ◦ φ) , (N ◦ φ) = dNφ(u,v) · = − (λ ◦ φ) ,
∂u ∂u ∂u ∂v ∂v ∂v
since dNφ(u,v) is some multiple of id. Now

∂2 ∂2
(N ◦ φ) = (N ◦ φ) ,
∂u∂v ∂v∂u
so
∂ ∂φ ∂ ∂φ
− (λ ◦ φ) =− (λ ◦ φ) .
∂u ∂v ∂v ∂u
dφ dφ
The vectors dv and du are linearly independent, so

∂ ∂
− (λ ◦ φ) = − (λ ◦ φ) = 0,
∂u ∂v
so λ ◦ φ is locally constant, that is λ is constant on φ (U ). Then S is connected, and covered by such
charts, so λ is constant.
Step 2. If λ (p) ≡ 0 for all p ∈ S then we already know that S is contained in a plane. Otherwise λ (p) ≡ λ0
with λ0 6= 0. Without loss of generality λ0 > 0. Now
∂ ∂φ ∂ ∂φ
(N ◦ φ) = −λ0 , (N ◦ φ) = −λ0 ,
∂u ∂u ∂v ∂v
so    
∂ 1 ∂ 1
(N ◦ φ) + φ = 0, (N ◦ φ) + φ = 0,
∂u λ0 ∂v λ0
so there exists c0 ∈ R3 such that (N ◦ φ) /λ0 + φ = c0 . Then (N ◦ φ) /λ0 has length 1/λ0 and c0 is
constant, so φ (u, v) lies in the sphere centred at c0 radius 1/λ0 .

Lecture 16
What if λ1 (p) 6= λ2 (p)? Friday
09/11/18
Definition 3.2.6. Let p ∈ S be a point on a regular surface. Let λ1 (p) and λ2 (p) be the principal curvatures.
Then
K (p) = λ1 (p) λ2 (p) = det dNp
is the Gaussian curvature at p and
1 1
H (p) = (λ1 (p) + λ2 (p)) = − Tr dNp
2 2
is the mean curvature at p.
Reversing the orientation of S, that is replacing N by −N, sends H (p) to −H (p) and K (p) to K (p).
Example. If S is a sphere of radius r then K = 1/r2 and H = 1/r. A plane has K = H = 0.

20
M3P5 Geometry of Curves and Surfaces 3 Curvature

3.3 What does Gaussian curvature mean?

Let S be a regular surface, and let p ∈ S. Choose a chart φ : U → S on S near p. Choose a curve through
p, γ (t) = φ (ct, dt). Now γ 0 (t) · N (γ (t)) = 0 for all t. Differentiating,

γ 00 (t) · N (γ (t)) + γ 0 (t) · dNγ(t) (γ 0 (t)) = 0,

or in other words
A (γ 0 (t) , γ 0 (t)) = γ 00 (t) · N (γ (t)) .
At t = 0,
∂φ ∂φ ∂2φ ∂2φ 2
2 ∂ φ
γ 0 (t) = c +d , γ 00 (t) = c2 + 2cd + d .
∂u ∂v ∂u2 ∂u∂v ∂v 2
Thus
2
∂2φ 2
   
∂φ ∂φ ∂φ ∂φ 2 ∂ φ 2 ∂ φ
A c +d ,c +d = c + 2cd +d · N (p) .
∂u ∂v ∂u ∂v ∂u2 ∂u∂v ∂v 2
Proposition 3.3.1. If K (p) > 0 then all points of S near p lie on the same side of Tp S. If K (p) < 0 then
there are points of S near p on each side of Tp S.
Proof. Consider the Taylor series

1 ∂2φ ∂2φ ∂2φ

   
∂φ ∂φ 2 2
φ (u, v) = p + (0, 0) u + (0, 0) v + (0, 0) u + 2 (0, 0) uv + 2 (0, 0) v + R (u, v) ,
∂u ∂v 2 ∂u2 ∂u∂v ∂v
where
R (u, v)
lim = 0.
(u,v)→(0,0) u2 + v 2
Consider
1 ∂2φ ∂2φ ∂2φ
 
2 2
hφ (u, v) − p, N (p)i = (0, 0) u + 2 (0, 0) uv + (0, 0) v , N (p) + hR (u, v) , N (p)i
2 ∂u2 ∂u∂v ∂v 2
1 ∂φ ∂φ
= A (w, w) + hR (u, v) , N (p)i , w= (0) u + (0) v.
2 ∂u ∂v
A (w, w) is quadratic in (u, v) and R (u, v) is worse than quadratic in (u, v). So for u and v small, when we
are close to p, the function hφ (u, v) − p, N (p)i has the same sign as A (w, w). Write w = ax1 + bx2 where
x1 and x2 are the principal directions then A (w, w) = λ1 a2 + λ2 b2 .
ˆ λ1 , λ2 > 0 gives A (w, w) > 0.
ˆ λ1 > 0 and λ2 < 0 gives that A (w, w) can be positive or negative.
ˆ λ1 , λ2 < 0 gives A (w, w) < 0.

In fact we can do slightly better than this. Consider the chart map φ : U → S. After a translation of S,
without loss of generality φ (0, 0) = (0, 0, 0), after a rigid motion to S ⊆ R3 , without loss of generality
∂φ ∂φ
(0, 0) = (1, 0, 0) , (0, 0) = (0, 1, 0) .
∂u ∂v
Taylor series implies that

1 ∂2φ ∂2φ ∂2φ

   
∂φ ∂φ 2 2
φ (u, v) = (0, 0, 0) + (0, 0) u + (0, 0) v + (0, 0) u + 2 (0, 0) uv + 2 (0, 0) v + . . .
∂u ∂v 2 ∂u2 ∂u∂v ∂v
2 2 2
 
1 ∂ φ ∂ φ ∂ φ
= (u, v, 0) + u2 2
(0, 0) + 2uv (0, 0) + v 2 2 (0, 0) + . . . .
2 ∂u ∂u∂v ∂v

21
M3P5 Geometry of Curves and Surfaces 3 Curvature

Near (u, v) = (0, 0) the quadratic term satisfies

2
∂2φ 2
 
2 ∂ φ 2 ∂ φ
u (0, 0) + 2uv (0, 0) + v (0, 0) · N (0, 0, 0) = A ((u, v, 0) , (u, v, 0)) = λ1 u2 + λ2 v 2 ,
∂u2 ∂u∂v ∂v 2
because we chose (1, 0, 0) and (0, 1, 0) as our principal directions. So near (u, v) = (0, 0), the surface is
approximated by the graph  
1
λ1 u2 + λ2 v 2 .


φ (u, v) ≈ u, v,
2
Lecture 17
Say p ∈ S is Monday
ˆ elliptic if K (p) > 0, so either λ1 , λ2 > 0 or λ1 , λ2 < 0, 12/11/18

ˆ hyperbolic if K (p) < 0, so either λ1 < 0 and λ2 > 0 or vice versa,

ˆ parabolic if K (p) = 0 and H (p) 6= 0, so λ1 = 0 and λ2 6= 0 or vice versa, and
ˆ planar if K (p) = H (p) = 0, so λ1 = λ2 = 0.
Example. Consider the Monkey saddle with equation z = x3 − 3x2 y, so
3
z = Re (x + iy) = Re r3 ei3θ = r3 cos 3θ,
using polar coordinates x + iy = reiθ . At (0, 0, 0), the principal curvatures are λ1 = λ2 = 0. So the previous
theorem says, to second order near p = (0, 0, 0), the surface S looks like the plane z = 0. True, but not so
informative. This has an umbilical point with λ1 = λ2 = 0 at (0, 0, 0), and hyperbolic points elsewhere.

3.4 How to compute Gaussian and mean curvature?

We start with a simple observation.
Lemma 3.4.1. Let V be a two-dimensional vector space with inner product h·, ·i. Let e1 and e2 be an
orthonormal basis for V . Let x and y be elements of V and write x = x1 e1 + x2 e2 and y = y1 e1 + y2 e2 . Then
 

y1
hx, yi = x1 x2 .
y2
That is, we can compute inner products in V using the ordinary matrix multiplication between row vectors
and column vectors, provided that we remember to use an orthonormal basis when writing elements of V as
vectors.
Proof. Just write everything out, so
hx, yi = hx1 e1 + x2 e2 , y1 e1 + y2 e2 i = x1 y1 he1 , e1 i + x1 y2 he1 , e2 i + x2 y1 he2 , e1 i + x2 y2 he2 , e2 i = x1 y1 + x2 y2 ,
as required. The final equality here uses the fact that e1 and e2 are orthonormal.
Remark. There is nothing special about the fact that V is two-dimensional here. The obvious generalisation
of Lemma 3.4.1 holds for any finite-dimensional inner product space.
Proposition 3.4.2. Let φ : U → S be a chart on a regular surface S. Define 2 × 2 matrices
   
φu · φu φu · φv A (φu , φu ) A (φu , φv ) ∂φ ∂φ
g= , A= , φu = , φv = .
φv · φu φv · φv A (φv , φu ) A (φv , φv ) ∂u ∂v
Then, writing σ = g−1 A, we have
det A 1
K = det σ = , H= Tr σ.
det g 2
Remark. If φu and φv are the principal directions at p ∈ S then this is immediate, because in that case g
is the identity matrix and Proposition 3.4.2 just restates the definitions of K and H.

22
M3P5 Geometry of Curves and Surfaces 3 Curvature

Proof. Let p be a point on S. Let us work in a basis for Tp S given by the principal directions x1 (p) and
x2 (p). Then, because this is an orthonormal basis, we can apply Lemma 3.4.1 to find,
 
φu

A= −dNp (φu ) −dNp (φv ) .
φv
Here (φu ) is a row vector of size two and (−dNp (φu )) is a column vector of size two. Now write the matrix
of dNp : Tp S → Tp S with respect to the basis x1 (p) and x2 (p) too,
 
φu

A= M φu φv ,
φv
where M , the matrix of dNp with respect to the basis x1 (p) and x2 (p), is a 2 × 2 matrix. Then
   
φ φ
det A = det u det M det φu φv = det M det u φu φv = det M det g,



φv φv
where for the last equality we used Lemma 3.4.1 again. Since det M = K, by definition, this proves that
K = det A/ det g. Furthermore,
 
φu

−1

A= φu φv φu φv M φu φv = gB,
φv

−1
M φu φv is conjugate to M . Thus σ = g−1 A is conjugate to M , and so H =


where B = φu φv
Tr M = Tr σ because conjugate matrices have the same trace.
Lecture 18
Example. Let S be the surface in R3 defined by the equation z = x2 − y 2 . Choose the chart Tuesday
13/11/18
φ : U = R2 −→ S
.
(u, v) 7−→ u, v, u2 − v 2
Now φu = (1, 0, 2u) and φv = (0, 1, −2v), so
1 + 4u2
   
φu · φu φu · φv −4uv
g= = .
φv · φu φv · φv −4uv 1 + 4v 2
Also,  
i j k
φu × φv =  1 0 2u  = (−2u, 2v, 1) ,
0 1 −2v
and so
φ u × φv 
−2u 2v

N= = √1+4u2 +4v 2
, √1+4u2 +4v 2
, √1+4u12 +4v2 .
|φu × φv |
Now φuu = (1, 0, 2) , φuv = (0, 0, 0) , φvv = (1, 0, −2), so
!
2
0
  √
 
N · φuu N · φuv 1+4u2 +4v 2 1 2 0
A= = −2 =√ .
N · φvu N · φvv 0 √
1+4u2 +4v 2
2
1 + 4u + 4v 2 0 −2

Then
1 + 4v 2 1 + 4v 2
   
1 4uv 2 −4uv
σ = g−1 A = ·A= √ .
1 + 4u2 + 4v 2 4uv 1 + 4u2 1 + 4u + 4v 2
3 4uv −1 − 4u2

So

!
− 1 + 4u2 1 + 4v 2 + 16u2 v 2 −4
K = det σ = 4 3 = 2,
(1 + 4u2 + 4v 2 ) (1 + 4u2 + 4v 2 )
and
1 4v 2 − 4u2
H= Tr σ = √ 3.
2 1 + 4u2 + 4v 2

23
M3P5 Geometry of Curves and Surfaces 3 Curvature

3.5 The first fundamental form

Definition 3.5.1. Let S ⊂ R3 be a regular surface, and let p ∈ S. The first fundamental form at p, also
called the metric at p, is the bilinear map
g : Tp S × Tp S −→ R
.
(v, w) 7−→ hv, wi
Then g is symmetric, bilinear, and non-degenerate. If φ : U → S is a chart near p ∈ S, then φu and φv are
a basis for Tp S and in this basis, g has the form
g (aφu + bφv , cφu + dφv ) = acg (φu , φv ) + adg (φu , φv ) + bcg (φu , φv ) + bdg (φu , φv )
    

g (φu , φu ) g (φu , φv ) c
c
= a b = a b g ,
g (φv , φu ) g (φv , φv ) d d
that is the metric is represented by the symmetric matrix that we called g before. Metric determines arc-
length. If α : [a, b] → S is a curve contained in φ (U ), so that α (t) = φ (u (t) , v (t)) for some smooth functions
u, v : [a, b] → R, then
Z b Z b Z bs  du 
0 ∂ φ du ∂ φ dv du dv


L (α) = |α (t)| dt = ∂u dt + ∂v dt dt = dt dt
dt
· g · dv dt.
a a a dt

Definition 3.5.2. A smooth map F : S1 → S2 is called a local isometry if it preserves the first fundamental
form, so
hdFp (x) , dFp (y)i = hx, yi , x, y ∈ Tp S1 , p ∈ S1 .
F is an isometry if it is also bijective.
Proposition 3.5.3. Local isometries are local diffeomorphisms.
Lecture 19
Friday
Proof. If F is a local isometry then, for each p ∈ S, dFp is injective. Suppose not. Then there exists
16/11/18
0 6= v ∈ Tp S such that dFp (v) = 0, so hdFp (v) , dFp (w)i = 0 for all w ∈ Tp S, so hv, wi = 0 for all w ∈ Tp S,
so v = 0, a contradiction. Now dFp : Tp S1 → TF (p) S2 is linear and injective, and dim Tp S1 = dim Tp S2 = 2,
so dFp is an isomorphism. From before, since dFp is an isomorphism, F is a local diffeomorphism, that is,
there is an open neighbourhood U of p in S1 such that F |U : U → F (U ) is a diffeomorphism.
Proposition 3.5.4. A smooth map F : S1 → S2 between regular surfaces is a local isometry if and only if
it preserves lengths of curves, that is if and only if for all α : [a, b] → S1 smooth we have
L (α) = L (F ◦ α) .
Proof.
=⇒ Suppose that F is a local isometry. Then
Z b Z b q
0 0 0
L (F ◦ α) = (F ◦ α) (t) dt = (F ◦ α) (t) , (F ◦ α) (t) dt
a a
Z b q Z b p
= dFα(t) (α0 (t)) , dFα(t) (α0 (t)) dt = hα0 (t) , α0 (t)i dt
a a
Z b
= |α0 (t)| dt = L (α) ,
a
by definition of dFα(t) and as F is a local isometry.
⇐= Given v ∈ Tp S1 and a curve α : (−, ) → S1 with α (0) = p and α0 (0) = v. Consider the curve
α|[−δ,t] : [−δ, t] → S1 , so
F
p ∈ S1 S2
α α| .
[−δ,t]

(−, ) ⊂
[−δ, t]

24
M3P5 Geometry of Curves and Surfaces 3 Curvature

   
Know L α|[−δ,t] = L (F ◦ α)|[−δ,t] , that is
Z t Z t
|α0 (s)| ds = dFα(s) (α0 (s)) ds.
−δ −δ

Taking d
dt and setting t = 0, |α0 (0)| = |dFp (α0 (0))|, that is

hv, vi = hdFp (v) , dFp (v)i , v ∈ Tp S1 , p ∈ S1 .

Now,

hX + Y, X + Y i = hX, Xi + hX, Y i + hY, Xi + hY, Y i = hX, Xi + 2 hX, Y i + hY, Y i ,

so
1
(hX + Y, X + Y i − hX, Xi − hY, Y i) .
hX, Y i =
2
Set X = dFp (v) and Y = dFp (w), so
1
hdFp (v) , dFp (w)i = (hdFp (v + w) , dFp (v + w)i − hdFp (v) , dFp (v)i − hdFp (w) , dFp (w)i)
2
1
= (hv + w, v + wi − hv, vi − hw, wi) = hv, wi .
2
Therefore F is a local isometry.

Example. Let S ⊂ R3 be any regular surface. Let B ∈ SO (3), 3 × 3 matrices B such that B | B = id and
det B = 1. Set
S 0 = {B~x | ~x ∈ S} .
Then
B : S −→ S0
~x 7−→ B~x
is bijective. Let p ∈ S and v, w ∈ Tp S be arbitrary. Then dFp (v) = Bv and
|
hdFp (v) , dFp (w)i = hBv, Bwi = (Bv) Bw = v | B | Bw = v | w = hv, wi ,

so B is a local isometry and an isometry.

Example. Let

S1 = (x, y, 0) ∈ R3 S2 = (x, y, z) ∈ R3 x2 + y 2 = 1 .
 
x ∈ R, y ∈ R ,

Define
F : S1 −→ S2
.
(u, v, 0) 7−→ (cos u, sin u, v)
At p = (u, v, 0) ∈ S1 a tangent vector ~v = (a, b, 0) can be written as α0 (0) where α (t) = (u + at, v + bt, 0).
Now
(F ◦ α) (t) = (cos (u + at) , sin (u + at) , v + bt) ,
0
so (F ◦ α) (0) = (−a sin u, a cos u, b). Then

hdFp (~v ) , dFp (~v )i = h(−a sin u, a cos u, b) , (−a sin u, a cos u, b)i = a2 + b2 = h~v , ~v i , ~v ∈ Tp S1 ,

and so, as before,

h~v , wi
~ = hdFp (~v ) , dFp (w)i
~ , ~ ∈ Tp S1 .
~v , w
Thus F is a local isometry. It is not an isometry because it is not bijective.

25
M3P5 Geometry of Curves and Surfaces 3 Curvature

3.6 Christoffel symbols

Lecture 20
dφ dφ dφ dφ
Now, dx1 and dx2 span Tp S and N (p) is orthogonal to Tp S so dx1 , dx2 , and N form a basis for R3 . Write Monday
19/11/18
∂2φ ∂φ ∂φ
= Γ1ij + Γ2ij + Aij N, (2)
∂xi ∂xj ∂x1 ∂x2

for some coefficients Γkij and Aij . Then

∂2φ ∂2φ
= ,
∂xi ∂xj ∂xj ∂xi
so Γkij = Γkji for all i, j, k.
Definition 3.6.1. The Γkij are called Christoffel symbols.
Taking (2) · N, we get
∂2φ
· N = Aij .
∂xi ∂xj
Note. ∇i ej = Γkij ek , where ∇i is the metric connection in the local frame ej .
Proposition 3.6.2. The Christoffel symbols are determined only by the first fundamental form, or metric.
dφ
Proof. From (2) · dxk ,

∂2φ
 
∂φ g11 g12
· = Γ1ij g1k + Γ2ij g2k , g= .
∂xi ∂xj ∂xk g21 g22
Also, !
∂ ∂φ ∂φ ∂2φ ∂φ ∂φ ∂2φ
· = · + · .
∂xi ∂xj ∂xk ∂xi ∂xj ∂xk ∂xj ∂xi ∂xk
Taking j = k,
∂
(gjj ) = 2 Γ1ij g1j + Γ2ij g2j .


∂xi
6 k, then either i = j or i = k. Taking i = j 6= k,
If j =
∂ ∂2φ ∂φ ∂φ ∂2φ ∂2φ ∂φ 1 ∂
(gik ) = · + · = · + (gii ) .
∂xi ∂xi ∂xi ∂xk ∂xi ∂xi ∂xk ∂xi ∂xi ∂xk 2 ∂xk
Rewriting,
∂2φ ∂φ ∂ 1 ∂
Γ1ij g1k + Γ2ij g2k = · = (gik ) − (gii ) .
∂xi ∂xi ∂xk ∂xi 2 ∂xk
We have lots of equations involving Γkij , gij , and derivatives of gij . In fact we can solve these for Γkij . Then
i = j = 1 and k = 1, 2 gives
1 ∂ g11 ∂ g12 1 ∂ g11
Γ111 g11 + Γ211 g21 = , Γ111 g12 + Γ211 g22 = − .
2 ∂x1 ∂x1 2 ∂x2
That is, !
1 ∂ g11
Γ111
 
2 ∂x1
g = .
Γ211 ∂ g12 1 ∂ g11
∂x1 − 2 ∂x2
dφ dφ
Then g is invertible, because dx1 and dx2 are linearly independent. Similarly,
∂ g11
! ∂ g21
!
1 1 ∂ g22
Γ121 −
   1 
2 ∂x2 Γ
g = , g 22
= ∂x2 1 ∂ g222 ∂x1 .
Γ221 1
2
∂ g22
∂x
Γ222 2 ∂x
1 2

∂ gjk
Thus can solve these equations to write all Γkij in terms of gij ’s and ∂xi ’s.

26
M3P5 Geometry of Curves and Surfaces 3 Curvature

3.7 Theorema Egregium

Lecture 21
Recall that K = det Aij / det gij . The goal is that K depends only on g. Tuesday
Notation. Set x1 = u and x2 = v. Write φu = dφ
= dφ 20/11/18
dx1 du , without confusion with indices.
Mixed partial derivatives are equal, so
∂2φ ∂2φ
   
∂ ∂
= ,
∂x2 ∂x1 ∂x1 ∂x1 ∂x1 ∂x2
so    
∂ ∂φ ∂φ ∂ ∂φ 2 ∂φ
Γ111 + Γ211 + A11 N = Γ112 + Γ12 + A12 N .
∂x2 ∂x1 ∂x2 ∂x1 ∂x1 ∂x2
Thus
Γ111 φuv + Γ111v φu + Γ211 φvv + Γ211v φv + A11v N + A11 Nv = Γ112 φuu + Γ112u φu + Γ212 φvu + Γ212u φv + A12u N + A12 Nu .
(3)
Will take the φu -component of (3), and use
φuv = Γ112 φu + Γ212 φv + A12 N, φvv = Γ122 φu + Γ222 φv + A22 N, Γuu = Γ111 φu + Γ211 φv + A11 N,
∂φ ∂φ ∂φ ∂φ
A11 = − · Nu , A21 = − · Nu , A12 = − · Nv , A22 = − · Nv .
∂u ∂v ∂u ∂v
Now
Nu = aφu + bφv , Nv = cφu + dφv ,
since N · N = 1, so Nu · N + N · Nu = 0 and Nv · N + N · Nv = 0, so Nu , Nv ∈ Tp S. Then
−A11 = ag11 + bg12 , −A21 = ag21 + bg22 , −A12 = cg11 + dg12 , −A22 = cg21 + dg22 ,
so  
a c
−A = g ,
b d
so     
a c −1 g22 −g12 A11 A12
= −g−1 A = ,
b d 2
g11 g12 − g12 −g21 g11 A21 A22
so
g12 A21 − g22 A11 g12 A22 − g22 A12
a= 2 , c= 2 .
g11 g22 − g12 g11 g22 − g12
Now, take (3), dot with φu , and use these equations above. We find,
   
g12 A22 − g22 A12 g12 A21 − g22 A11
Γ111 Γ112 + Γ111v + Γ211 Γ122 + A11 = Γ112 Γ111 + Γ112u + Γ212 Γ112 + A12 .
det g det g
Rearranging,
(A12 A21 − A11 A22 ) g12
Γ111v − Γ112u = Γ212 Γ112 − Γ211 Γ122 + .
det g
In other words,
Kg12 = Γ212 Γ112 − Γ211 Γ122 + Γ112u − Γ111v .
Similarly, taking the φv component of (3)
Kg11 = Γ111 Γ221 + Γ211 Γ222 − Γ112 Γ211 − Γ212 Γ212 + Γ211v − Γ212u .
These are the Gauss equations.
Note. Kg12 and Kg11 depend only on g. But g is invertible, so g11 and g12 cannot simultaneously vanish.
We have proved the following.
Theorem 3.7.1 (Theorema Egregium, Gauss). The Gaussian curvature K depends only on the first funda-
mental form.
In other words, it is an intrinsic invariant of the surface (S, g), as a Riemannian manifold.

27
M3P5 Geometry of Curves and Surfaces 3 Curvature

Corollary 3.7.2. If f : S1 → S2 is a local isometry then KS1 (p) = KS2 (f (p)) for all p ∈ S1 .
Corollary 3.7.3. There is no local isometry from a plane to a sphere.
Proof. A plane has K = 0 everywhere and a sphere has K = 1/r2 everywhere, where r is the radius.
In particular, there is no way to draw a map of the Earth on a flat sheet of paper without distorting lengths
or angles. This is also why it is hard to gift-wrap a ball. Lecture 22
Friday
Theorem 3.7.4. Let S be a compact surface with K > 0, such that K is constant, then S is umbilical.
23/11/18
Proof. Let x ∈ S, and λ1 (x) ≤ λ2 (x) be principal curvatures at x. The aim is that λ1 (x) = λ2 (x). The idea
of the proof is to assume λ1 < λ2 for a contradiction. Let p ∈ S be a point maximising λ2 (x), so minimises
λ1 (x) and
λ1 (p) ≤ λ1 (x) ≤ λ2 (x) ≤ λ2 (p) .
The aim is if λ1 (p) = λ2 (p), then λ1 (x) = λ (x), for all x ∈ S. Assume λ1 (p) < λ2 (p), and want to get a
contradiction. After a rigid motion, around p = (0, 0, 0),
1
λ 1 u2 + λ 2 v 2 + . . . ,


φ (u, v) = (u, v, F (u, v)) + · · · =
2
by Taylor expansion. At (0, 0), F (0, 0) = Fu (0, 0) = Fv (0, 0) = 0, so
∂φ ∂φ
(0, 0) = (1, 0, Fu (0, 0)) = (1, 0, 0) , (0, 0) = (0, 1, Fv (0, 0)) = (0, 1, 0) ,
∂u ∂u
so
φu × φv (−Fu , −Fv , 1)
N= = q = (0, 0, 1) ,
|φu × φv | 2
1 +|∇F |
at (0, 0). Then
       
Fuu Fuv λ1 0 φu · φu φu · φv 1 0
A= = , g= = .
Fvu Fvv 0 λ2 φv · φu φv · φv 0 1
Thus K = det A/ det g = λ1 λ2 . Let
(1, 0, Fu ) (0, 1, Fv )
E1 = p , E2 = p ,
1 + Fu2 1 + Fv2
and let
h1 (t) = A (E1 (0, t) , E1 (0, t)) , h2 (t) = A (E2 (t, 0) , E2 (t, 0)) .
2
Check
1 Fuu 1 Fvv
h1 (t) = ·q , h2 (t) = ·q .
1 + Fu2 2 1 + Fv2 2
1 +|∇F | 1 +|∇F |
(0,t) (t,0)
Recall that

λ1 (p) ≤ λ1 (φ (0, t)) = min A (v, v) v ∈ Tφ(0,t) S, |v| = 1 ≤ A (E1 (0, t) , E1 (0, t)) = h1 (t) ,
and

λ2 (p) ≥ λ2 (φ (t, 0)) = max A (v, v) v ∈ Tφ(t,0) S, |v| = 1 ≥ A (E2 (t, 0) , E2 (t, 0)) = h2 (t) ,
so h1 (t) is minimum at t = 0 and h2 (t) is maximum at t = 0, so h001 (0) ≥ 0 and h002 (0) ≤ 0, so
h001 (0) − h002 (0) ≥ 0.
3
Assume λ1 < λ2 . Thus λ1 λ2 (λ1 − λ2 ) < 0, a contradiction.
Corollary 3.7.5. Let K > 0 be constant. Then S is compact and connected implies that S = S2 .
2 Exercise
3 Exercise

28
M3P5 Geometry of Curves and Surfaces 4 Area of surfaces

4 Area of surfaces
4.1 Area
Lecture 23
Let S ⊂ R3 be a regular surface and φ : U → S be a chart. How to measure the area on S? Consider a tiny Monday
rectangle in U with sides (δu, 0) and (0, δv). The area of this, in U , is δuδv. If δu and δv are small then the 26/11/18
3
areaof the image
  of this rectangle in φ (U ) is approximately the area of the parallelogram in R with sides
dφ dφ
du δu and dv δv, that is
   
∂φ ∂φ ∂φ ∂φ
δu × δv = × δuδv.
∂u ∂v ∂u ∂v

Definition 4.1.1. If D ⊂ U is compact, then

ZZ
∂φ ∂φ
area φ (D) = ∂u × ∂v du dv.
D

Need to show this integral

ˆ converges, and

ˆ is independent of choice of chart.

Proposition 4.1.2. This definition does not depend on the choice of coordinate chart φ : U → S.

Proof. Let φ : U → S and ψ : U 0 → S be charts whose images contain φ (D) = ψ (D). After shrinking U
and U 0 if necessary, we have that f = φ−1 ◦ ψ : U 0 → U is well-defined and smooth. Write

f (u, v) = (x (u, v) , y (u, v)) ,

so
φ
D⊂U S
f .
ψ
D0 ⊂ U 0
Then ψ = φ ◦ f so
     
∂ψ ∂ψ ∂φ ∂x ∂φ ∂y ∂φ ∂x ∂φ ∂y ∂φ ∂φ ∂x ∂y ∂y ∂x
× = + × + =∆ × , ∆ (u, v) = − .
∂u ∂v ∂x ∂u ∂y ∂u ∂x ∂v ∂y ∂v ∂x ∂y ∂u ∂v ∂u ∂v

So
∂y
ZZ ZZ  ∂x 
0 ∂ψ ∂ψ ∂φ ∂φ
area ψ (D ) = ∂u × ∂v du dv = ∂u × ∂v det ∂u
∂x
∂u
∂y du dv
D0 D0 ∂v ∂v
ZZ
∂φ ∂φ
= ∂x × ∂y dx dy = area φ (D) ,
D

by the change of variable formula.

Note. S ⊂ R3 , and R3 has a distinguished 3-form dx ∧ dy ∧ dz. If S is oriented, so we have chosen a normal
vector field N on S, then can consider the 2-form ω = 2N (dx ∧ dy ∧ dz), where 2N is the contraction, and
measure areas on S using this 2-form. Up to sign, this gives the same notion of area.

29
M3P5 Geometry of Curves and Surfaces 4 Area of surfaces

Lemma 4.1.3. ZZ p
area φ (D) = det g du dv.
D
2 2 2 2
Proof. |a × b| = |a| |b| − (a · b) , so
2 2 2  2
∂φ ∂φ ∂φ ∂φ ∂φ ∂φ
× = − · .
∂u ∂v ∂u ∂v ∂u ∂v
But ∂φ ∂φ ∂φ ∂φ

∂u · ∂u ∂u · ∂v
g= ∂φ ∂φ ∂φ ∂φ ,
∂v · ∂u ∂v · ∂v
so
2
∂φ ∂φ 2
× = g11 g22 − g21 = det g.
∂u ∂v
Thus ZZ ZZ
∂φ ∂φ
p
area φ (D) = ∂u × ∂v du dv = det g du dv.
D D

Definition 4.1.4. Assume S is compact. If f : S → R is a smooth function on S, φ : U → S is a chart, and

D ⊂ U is a compact subset, then we define
Z ZZ ZZ p
f dA = (f ◦ φ) (u, v) ∂∂uφ × ∂∂vφ du dv = (f ◦ φ) det g du dv.
φ(D) D D

This does not depend on chart φ : U → S. To integrate over all of S we divide S into pieces
S = S1 ∪ · · · ∪ Sk , Si ∩ Sj ⊂ (∂Si ) ∩ (∂Sj ) , i 6= j,
such that for each i there exists a chart φ : Ui → S such that Si ⊂ φ (Ui ) and Si = φ (Di ) for some compact
set Di ⊂ Ui . By compactness of S, such a partition S = S1 ∪ · · · ∪ Sk always exists.
Definition 4.1.5. Define
Z k Z
X
f dA = f dA.
S i=1 Si

Claim that this is independent of the choice of decomposition. Lecture 24

3 Tuesday
Example. What is the surface area of the unit sphere in R ? Use spherical polar coordinates
27/11/18
ψ (θ, φ) = (cos θ sin φ, sin θ sin φ, cos φ) ,  ≤ θ ≤ 2π − ,  ≤ φ ≤ π − .
Are the columns of the matrix
 

− sin θ sin φ cos θ cos φ
ψθ ψφ =  cos θ sin φ sin θ cos φ  .
0 − sin φ
linearly independent? If sin φ 6= 0 then the first two rows of the first column is not both zero and the last
row of the second column is not zero. So this surface is regular, because sin φ 6= 0 in our chart, since we took
φ ∈ (0, π). Then
ZZ Z π− Z 2π−
area ψ (D) = |ψθ × ψφ | dθ dφ = sin φ dθ dφ
D  

= (2π − 2) [cos φ]π− = (2π − 2) (cos  − cos (π − )) ,
so
lim area ψ (D) = 4π.
→0

Compute the area of the rest by using another chart, which tends to zero as  → 0. Thus area S2 = 4π.

30
M3P5 Geometry of Curves and Surfaces 5 Geodesics

5 Geodesics
5.1 Geodesic curvature
Geodesics are the shortest paths. Let γ : [a, b] → S be a regular curve in a regular surface S. Suppose that
S is oriented, with normal vector field N, and that γ is parametrised by arc-length. Then
(γ 0 (t) , N (γ (t)) × γ 0 (t) , N (γ (t)))
is an orthonormal basis for R3 . Recall that the curvature of γ is ~κ (t) = γ 00 (t) and that γ 00 is orthogonal to
γ 0 , because γ 0 (t) · γ 0 (t) = 1, so dt
d
(γ 00 (t) · γ 0 (t)) = 0, so 2γ 00 (t) · γ 0 (t) = 0. So
~κ (t) = kn (t) N + kg (t) (N × γ 0 ) ,
where
kn (t) = ~κ (t) · N
is the normal curvature of γ at t, and
kg (t) = ~κ (t) · (N × γ 0 )
is the geodesic curvature of γ at t, the amount that γ is curving along S. Then γ is a geodesic if the
geodesic curvature is zero. Lecture 25
Friday
Example. Let S be the xy plane and N = (0, 0, 1). Consider a curve
30/11/18
γ (t) = (x (t) , y (t) , 0)

2 2 
dy
parametrised by arc-length, so dx
dt + dt = 1. Then the normal curvature is

kn = ~κ · N = (x00 , y 00 , 0) · (0, 0, 1) = 0,
and N × γ 0 = (0, 0, 1) × (x0 , y 0 , 0) = (−y 0 , x0 , 0), so the geodesic curvature is
kg = ~κ · (N × γ 0 ) = (x00 , y 00 , 0) · (−y 0 , x0 , 0) = x0 y 00 − x00 y 0 .
Then ~κ = kg (N × γ 0 ), so kg = 0 if and only if ~κ = 0, if and only if x00 = y 00 = 0. That is, geodesics in the
plane take the form
(x (t) , y (t)) = (a0 , b0 ) + t (a1 , b1 ) ,
which are straight lines.
Example. Let S be the unit sphere
(x, y, z) ∈ R3 x2 + y 2 + z 2 = 1 .


Consider a latitude of radius r,

 p 
γ (t) = r cos rt , r sin rt , 1 − r2 , 0 ≤ t ≤ 2πr.

Then  p 
γ 0 (t) = − sin rt , cos rt , 0 , N (γ (t)) = γ (t) = r cos rt , r sin rt , 1 − r2 ,

so  p p  1
N × γ 0 = − 1 − r2 cos rt , − 1 − r2 sin rt , r , ~κ (t) = γ 00 (t) = − cos rt , − sin rt , 0 .


r
Thus
1
 p p 
kg = − cos rt , − sin rt , 0 · − 1 − r2 cos rt , − 1 − r2 sin rt , r
r
√ √
1 − r2 2 t 2 t

1 − r2
= cos r + sin r = .
r r
So a latitude is a geodesic if and only if r = 1, if and only if it is the equator.

31
M3P5 Geometry of Curves and Surfaces 5 Geodesics

Proposition 5.1.1. Local isometries send geodesics to geodesics.

Proof. Let γ : [a, b] → S be a geodesic parametrised by arc-length, and F : S → S 0 a local isometry, so

F
γ (t) ⊂ S (F ◦ γ) (t) ⊂ S 0
φ .
ψ=F ◦φ
U ⊂ R2

From
~κ (t) = kn N (γ (t)) + kg (N (γ (t)) × γ 0 (t)) ,
we see that γ is a geodesic if and only if γ 00 (t) is a multiple of N (γ (t)), if and only if

γ 00 (t) · φu = γ 00 (t) · φv = 0,

for φ : U → R2 a chart at γ (t). Write

γ (t) = φ (u (t) , v (t)) ,
so
γ 0 (t) = φu u0 + φv v 0 , γ 00 (t) = (φuu u0 + φuv v 0 ) u0 + (φuv u0 + φvv v 0 ) v 0 ,
so
2 2
0 = γ 00 · φu = (φu · φuu ) (u0 ) + 2 (φu · φuv ) (u0 v 0 ) + (φu · φvv ) (v 0 )
 
1 ∂ 2 ∂ ∂ 1 ∂ 2
= (φu · φu ) (u0 ) + (φu · φu ) (u0 v 0 ) + (φu · φv ) − (φv · φv ) (v 0 )
2 ∂u ∂v ∂v 2 ∂u
 
1 0 2 1 2
0 0
= g11u (u ) + g11v (u v ) + g12v − g22u (v 0 ) ,
2 2

and similarly for γ 00 · φv . These are determined by u and v, and the first fundamental form g. Since F is an
isometry, it preserves g with respect to the chart ψ = F ◦ φ, and F (γ (t)) = ψ (u (t) , v (t)) implies that
00 00
(F ◦ γ) · ψu = γ 00 · φu = 0, (F ◦ γ) · ψv = γ 00 · φv = 0.

Let φ : U → S be a chart on a regular surface, and let γ (t) = φ (u (t) , v (t)) be a curve on S. Then γ is a
geodesic if and only if 4

0 1 2 2

(g11 u0 + g12 v 0 ) = g11u (u0 ) + 2g12u (u0 v 0 ) + g22u (v 0 ) ,
2
and
0 1 2 2

(g21 u0 + g22 v 0 ) = g11v (u0 ) + 2g12v (u0 v 0 ) + g22v (v 0 ) .
2
These are the geodesic equations.
4 Exercise

32
M3P5 Geometry of Curves and Surfaces 5 Geodesics

5.2 Length-minimising curves

Lecture 26
Claim that geodesics minimise arc-length, locally. That is, for any small perturbation β of γ, L (β) ≥ L (γ). Monday
Definition 5.2.1. A variation of γ : [0, L] → S is a smooth map 03/12/18

[0, L] × [−, ] −→ S
,
(t, s) 7−→ γs (t)
such that

γ0 (t) = γ (t) , γs (0) = γ (0) , γs (L) = γ (L) , t ∈ [0, L] , s ∈ [−, ] .

Proposition 5.2.2. Geodesics minimise arc-length locally. That is, if γ : [0, L] → S is a local minimum for
arc-length between γ (0) and γ (L) and is parametrised by arc-length, then γ is a geodesic.
Proof. Let γs (t) be a variation of γ. Then s 7→ L (γs ) is minimised at s = 0. This is a smooth function of s.
So
Z Lp
d d
0= (L (γs )) = γs0 (t) · γs0 (t) dt
ds s=0 ds 0
s=0
Z L p p
d
= ds γs0 (t) · γs0 (t) dt γs0 (t) · γs0 (t) is smooth
0 s=0
d 0 0
ds (γs (t)·γs (t))
Z L Z L
= √ s=0
dt = d
ds (γs0 (t)) · γs0 (t) s=0
dt |γ00 (t)| = |γ 0 (t)| = 1
0 2 γs0 (t)·γs0 (t) 0
s=0
Z L   Z L  
d dγs dγ0 d dγs dγ0
= ds dt · dt dt = dt ds · dt dt (s, t) 7→ γs (t) is smooth
0 s=0 0 s=0
h iL Z L
dγs dγ0 dγs d2 γ0
= ds · dt − ds · dt2 dt integration by parts
s=0 0 0 s=0
Z L
=− dγs
ds · (kn (t) N (γ (t)) + kg (t) (N (γ (t)) × γ 0 (t))) dt dγs
ds (0) = dγs
ds (L) = 0
0 s=0
Z L
= dγs
ds · (kg (t) (N (γ (t)) × γ 0 (t))) dt N (γ (t)) ⊥ Tγ(t) S.
0 s=0

Let g : [0, L] → R be any smooth function such that g (0) = g (L) = 0. Can find a variation γs (t) with
dγs
= g (t) (N (γ (t)) × γ 0 (t)) .
ds s=0

Thus we have Z L
kg (t) g (t) dt = 0,
0
and, since kg is continuous as a function of t, and g is arbitrary, it follows that kg (t) = 0 for all t, that is γ
was a geodesic.
Remark.
ˆ Geodesics need not be global minima for arc-length between two points. In S2 with usual metric, a
major arc is a geodesic but not a global minimum for arc-length.
ˆ Geodesics between two points need not be unique. All great circles are geodesics from N to S.
ˆ Geodesics connecting two points need not exist. In R2 \ {0}, 1 and −1 has no geodesic, and R2 \ {0} is
locally isometric to R2 . Local isometries preserve geodesics, so geodesics in R2 \ {0} are straight lines.
In general, it is hard to find geodesics explicitly. Need to solve the geodesic equation.

33
M3P5 Geometry of Curves and Surfaces 6 The Gauss-Bonnet theorem and applications

6 The Gauss-Bonnet theorem and applications

Lecture 27
Equates geometry and topology. Will prove this from its local version. Tuesday
04/12/18
6.1 Local version of Gauss-Bonnet
Theorem 6.1.1 (Local
Gauss-Bonnet). Let φ : U → S be a chart which is

smooth in U , the closure of U in
R2 , such that S = φ U is an oriented surface with boundary ∂S = φ ∂U . Suppose that U is diffeomorphic
to a disc. Then, Z Z
K dA + kg ds = 2π,
S ∂S
where A is the area, s is the arc-length, and ∂S is positively oriented with respect to S. A surface with
boundary is the same as a regular surface except that some points p ∈ S have charts φ : U ⊂ R2 → S, where
U is the intersection of an open neighbourhood of (0, 0) in R2 and {y ≥ 0}. φ here is smooth, which means
there exists an open set V ⊂ R3 and Φ : V → R3 smooth such that φ = Φ|U and U ⊆ V . Then ∂S is the
collection of all points with charts like this, a collection of regular curves, and ∂S is positively oriented if
in some parametrisation γ : [a, b] → ∂S, N × γ 0 points into S.
Proof. The idea is to recall Z ZZ
∂φ ∂φ
K dA = (K ◦ φ) ∂u × ∂v du dv.
S U
We will find functions M, L : U → R such that
ZZ ZZ Z Z
(K ◦ φ) ∂∂uφ × ∂∂vφ du dv = ∂M
∂u ×
∂L
∂v du dv = L du + M dv = 2π − kg ds,
U U ∂U ∂U
by applying Green’s theorem. For a precise version, make an orthonormal basis for Tp S. Take
φu
E1 = , E2 = N × E1 ,
|φu |
so N = E1 × E2 . Have an orthonormal basis E1 and E2 for Tp S along ∂S. So there exists a continuous
function θ : [0, L] → R such that
γ 0 (t) = cos θ (t) E1 (γ (t)) + sin θ (t) E2 (γ (t)) .
Here L is the length of the boundary, and γ : [0, L] → ∂S is an arc-length parametrisation of ∂S.
ˆ Claim that
kg (γ (t)) = θ0 (t) − E1 (t) · E20 (t) .
kg = γ 00 · (N × γ 0 ). Then
γ 00 (t) = (− sin θE1 + cos θE2 ) θ0 (t) + cos θE10 + sin θE20 , N × γ 0 (t) = − sin θE1 + cos θE2 ,
so
kg (γ (t)) = γ 00 (t) · (N × γ 0 (t)) = θ0 (t) + (cos θE10 + sin θE20 ) · (− sin θE1 + cos θE2 ) .
Now,
E1 (t) · E1 (t) = 1, E1 (t) · E2 (t) = 0, E2 (t) · E2 (t) = 1.
Differentiating,
E1 · E10 = 0, E1 · E20 + E10 · E2 = 0, E2 · E20 = 0,
so kg = θ0 − E1 · E20 , which was the claim. Lecture 28
Friday
ˆ Claim that Z L ZZ 07/12/18
E1 (t) · E20 (t) dt = (K ◦ φ)|φu × φv | du dv.
0 U
This will help because it implies
Z Z Z L
K dA + kg ds = θ0 (t) dt.
S ∂S 0

34
M3P5 Geometry of Curves and Surfaces 6 The Gauss-Bonnet theorem and applications

Let us show that

∂ E1 ∂ E2 ∂ E1 ∂ E2
· − · = K|φu × φv | (4)
∂u ∂v ∂v ∂u
Assuming (4) then, setting γ (t) = φ (u (t) , v (t)), we have
Z L Z L Z
0 ∂ E2 0 ∂ E2 0
E1 (t) · ∂∂u
E2 ∂ E2


E1 (t) · E2 (t) dt = E1 (t) · ∂u u + ∂v v dt = du + E1 (t) · ∂v dv
0 0 ∂U
ZZ
∂ E2
∂ ∂
E1 (t) · ∂∂u
E2




= ∂u E1 (t) · ∂v − ∂v du dv
Z ZU ZZ
∂ E1 ∂ E2 ∂ E1 ∂ E2


= ∂u · ∂v − ∂v · ∂u du dv = K|φu × φv | du dv,
U U

by Green’s theorem and (4), which proves the claim. It remains to prove (4). Note
∂ E1 ∂ E2 ∂ E1
E1 · = 0, E1 · + · E2 = 0, ....
∂u ∂u ∂u
So      
∂ E1 ∂ E1 ∂N ∂φ
= aE2 + · N N = aE2 − E1 · N = aE2 + A E1 , N,
∂u ∂u ∂u ∂u
for some constant a. Similarly,
     
∂ E1 ∂φ ∂ E2 ∂φ ∂ E2 ∂φ
= bE2 + A E1 , N, = cE1 + A E2 , N, = dE1 + A E2 , N,
∂v ∂v ∂u ∂u ∂v ∂v
for some constants b, c, d. Thus
       
∂ E1 ∂ E2 ∂ E1 ∂ E2 ∂φ ∂φ ∂φ ∂φ
· − · = A E1 , A E2 , − A E2 , A E1 , .
∂u ∂v ∂v ∂u ∂u ∂v ∂u ∂v
Writing
∂φ ∂φ
= c11 E1 + c12 E2 , = c21 E1 + c22 E2 ,
∂u ∂v
this is
∂ E1 ∂ E2 ∂ E1 ∂ E2
· − · = (c11 A (E1 , E1 ) + c12 A (E1 , E2 )) · (c21 A (E2 , E1 ) + c22 A (E2 , E2 ))
∂u ∂v ∂v ∂u
− (c11 A (E2 , E1 ) + c12 A (E2 , E2 )) · (c21 A (E1 , E1 ) + c22 A (E1 , E2 ))
= (c11 c22 − c12 c21 ) · (A (E1 , E1 ) A (E2 , E2 ) − A (E1 , E2 ) A (E2 , E1 ))
∂φ ∂φ ∂φ ∂φ
= × det A = × K,
∂u ∂v ∂u ∂v
as claimed.
So at this point we know
Z Z Z L
dθ
K dA + kg ds = dt dt.
S ∂S 0
It remains to show that the right hand side is 2π. Since
Z L
dθ
dt dt = θ (L) − θ (0)
0

is the total rotation of γ 0 (t) along the path ∂S with respect to the frame (E1 (t) , E2 (t)), the fact that this
is 2π follows from the fact that U is only diffeomorphic to a disc, so replace it by a disc. There is a regular
homotopy
H : [0, 1] × [0, 1] −→ X
,
(s, t) 7−→ γs (t)
smooth in s and t, from the boundary curve to an approximate tiny circle in Tp S. Since Ind γ = θ (L) − θ (0)
is invariant under regular homotopy, Ind γ is the index of a small circle in the plane, which is 2π.

35
M3P5 Geometry of Curves and Surfaces 6 The Gauss-Bonnet theorem and applications

6.2 Gauss-Bonnet for curvilinear triangles

Lecture 29
Definition 6.2.1. A curvilinear triangle in R2 is a continuous map β : R → R2 such that Monday
ˆ β (t) = β (t + 3) for all t, and 10/12/18

ˆ there exist t0 , t1 , t2 ∈ [0, 3) such that 0 ≤ t0 < t1 < t2 < 3 and

– β is smooth on (t0 , t1 ), (t1 , t2 ), and (t2 , t3 ) where t3 = t0 + 3,
– β|[0,3) is injective, and
0 0
– β− (ti ) and β+ (ti ) exist and meet at an angle θi that is not a multiple of π.
Let T ⊂ U ⊂ R2 be a curvilinear triangle and let φ : U → S be a chart on a regular surface S. At each
vertex of φ (T ), the edges incident to that vertex are parametrised by γin : (−, 0] → S and γout : [0, ) → S.
The tangent vectors meet at some angle θ for −π < θ < π, so
0 0
γin (0) · γout (0)
cos θ = 0 0 .
|γin (0)||γout (0)|
0 0
Choose the sign of θ such that θ > 0 if (γin (0) , γout (0) , N) is a positive basis of R3 and θ < 0 if
0 0 3
(γin (0) , γout (0) , N) is a negative basis of R . Then θ is the exterior angle at the vertex. The interior
angle is π − θ.
Theorem 6.2.2 (Gauss-Bonnet for curvilinear triangles). Let T ⊂ U ⊂ R2 be a curvilinear triangle and let
φ : U → S be a chart on the regular surface S. Suppose that φ (∂T ) ⊂ S is oriented positively with respect to
the orientation of S. Let θ1 , θ2 , θ3 be the exterior angles and let γ1 , γ2 , γ3 be the edges of φ (T ) parametrised
by arc-length. Then,
X3 Z X3 Z
kg ds + θi = 2π − K dA.
i=1 γi i=1 φ(T )
R
Think of the left hand side as φ(∂T )
kg ds where φ (∂T ) is thought as having a δ-function of curvature of
size θi at the vertex i.
Proof. Essentially the same as in local Gauss-Bonnet. We still integrate kg = θ0 − E1 · E20 , apply Green’s
theorem to the triangle T , and get
Z X3 Z Z
K dA + kg ds = θ0 (t) dt.
φ(T ) i=1 γi ∂S

But now, by topology, we have

Z 3
X
θ0 (t) dt = 2π − θi ,
∂S i=1

because the tangent vectors to the γi Ralong φ (T ) = ∂S still rotate a total of 2π, but this includes jumps of
θ1 , θ2 , θ3 at the three corners, so that ∂S θ0 (t) dt does not include these jumps.
The following are consequences. Consider a geodesic triangle on S of geodesic edges, and interior angles
α1 , α2 , α3 . Gauss-Bonnet says,
Z 3 Z
X 3
X X3 X3
K dA = 2π − kg ds − θi = 2π − (π − αi ) = αi − π.
φ(T ) i=1 γi i=1 i=1 i=1

For plane triangles, K = 0, so

3
X
αi = π.
i=1
For geodesic triangles T on a sphere of radius one, K = 1, so
3
X
αi = π + area T ≥ π.
i=1

36
M3P5 Geometry of Curves and Surfaces 6 The Gauss-Bonnet theorem and applications

6.3 Gauss-Bonnet theorem

Proposition 6.3.1. Every compact regular surface admits a triangulation
N
[
S= Ti ,
i=1

that is a partition of S into curvilinear triangles Ti such that whenever Ti ∩ Tj is non-empty then Ti ∩ Tj is
an edge or vertex of both Ti and Tj , and each edge belongs to
ˆ exactly two of the Ti if that edge lies in the interior of S, and
ˆ in exactly one of the Ti if that edge lies in ∂S.
S is sewn together from curvilinear triangles.
Proof. By compactness, we can cover S by finitely many charts φi : Ui → S such that Ui is homeomorphic
to a disc. Without loss of generality we may assume
[
φi (Ui ) * φj (Uj ) ,
j6=i

because otherwise just remove (Ui , φi ) from our cover. Then

[
Vi = φi (Ui ) \ φj (Uj )
j6=i

is non-empty, and closed in S. Draw a curve Ci ⊂ φi (Ui ) which bounds a neighbourhood of Vi . The union
of the Ci divides S into regions, each of which is contained in a single chart, and that chart is homeomorphic
to a disc in R2 . Now work in Ui . Now have a closed curve in Ui ⊂ R2 . Triangulate the interior of this.
Lecture 30
Lemma 6.3.2. Three curves with the same vertex of interior angles α1 and α2 imply that the sum of the Tuesday
interior angles is α1 + α2 . 11/12/18
0 0
Proof. Without loss of generality we can take the velocity vectors γin (0) and γout (0) as unit vectors. Let
(cos α1 , sin α1 ) be a velocity vector of interior angle α1 , (cos (α1 + α2 ) , sin (α1 + α2 )) be a velocity vector of
interior angle α2 , and φ be the difference of the interior angles. Then
cos φ = (cos (α1 + α2 ) , sin (α1 + α2 )) · (cos α1 , sin α1 )
= cos (α1 + α2 ) cos α1 + sin (α1 + α2 ) sin α1 = cos ((α1 + α2 ) − α1 ) = cos (α2 ) ,
so φ = α2 .
What is the Euler characteristic? Take a triangulation of S. Then
χ (S) = V − E + F,
where V is the number of vertices in the triangulation, E is the number of edges in the triangulation, and
F is the number of triangles in the triangulation. This is independent of the choice of triangulation of S.
Follows from the Gauss-Bonnet theorem and can prove this topologically.
Example.
ˆ Triangulate S2 by an inflated tetrahedron. Then V = 4, E = 6, F = 4, so χ S2 = 4 − 6 + 4 = 2.

ˆ Triangulate T2 by two triangles in a square. Then V = 1, E = 3, F = 2, so χ T2 = 1 − 3 + 2 = 0.

Exercise.
ˆ Triangulate T2 by eight triangles in a square, and compute χ T2 using this triangulation.

ˆ Triangulate Σg , a surface of genus g, by two tori without a disc and g − 2 tori without two discs, and
show χ (Σg ) = 2 − 2g.

37
M3P5 Geometry of Curves and Surfaces 6 The Gauss-Bonnet theorem and applications

Theorem 6.3.3 (Gauss-Bonnet). Let S ⊂ R3 be a compact oriented surface, with, a possibly empty, posit-
ively oriented boundary. Then Z Z
K dA + kg ds = 2πχ (S) ,
S ∂S

where χ is the Euler characteristic. If ∂S = ∅, then

Z
K dA = 2πχ (S) .
S

Proof. Triangulate S, so
N
[
S= Ti ,
i=1

where Ti is a curvilinear triangle with exterior angles θij for 1 ≤ j ≤ 3. Set the interior angles αij = π − θij .
Then
Z Z X3 X3
K dA + kg ds = 2π − θij = αij − π,
Ti ∂Ti j=1 j=1

by Gauss-Bonnet for Ti . The integral of kg along the edge Ti ∩ Tj cancels because it occurs for Ti and Tj
with opposite signs. All of the interior edges cancel out. At each interior vertex, the interior angles add up
to 2π. Boundary edges do not cancel. At boundary vertices, the interior angles sum to π. Adding everything
up,
XN Z XN Z XN X 3 XN
K dA + kg ds = αij − π,
i=1 Ti i=1 ∂Ti i=1 j=1 i=1
so Z Z
K dA + kg ds = 2πV − π# {boundary vertices} − πF.
S ∂S
But counting edges,

3F = # {interior edges} + # {boundary edges} = 2E − # {boundary edges} = 2E − # {boundary vertices} .

So Z Z
K dA + kg ds = 2π (V + F ) − π (3F + # {boundary vertices}) = 2π (V − E + F ) .
S ∂S

Corollary 6.3.4. Let S ⊂ R3 be a compact oriented surface with ∂S = ∅, such that K ≥ 0 on S. Then
S∼
= S2 .
Proof. There exists some point with K > 0 because there exists an elliptic point on S, so
Z
1
χ (S) = K dA > 0.
2π S

But χ (S) = 2 − 2g with g the genus and so g = 0 because χ (S) > 0.

38