Differential Geometry

Simon Rubinstein-Salzedo
Euler Circle, Mountain View, CA 94040
Email address: [email protected]
 Contents

Chapter 1. Parametrized Curves 5

1. Parametrized Curves 5
2. The Tangent Vector 6
3. Arclength 7
4. Reparametrization 8
5. Curvature 10
6. The Isoperimetric Inequality 12
7. Problems 15
Chapter 2. Curvature, Torsion, and the Frenet–Serret Frame 17
1. Curvature of Plane Curves 17
2. The Osculating Circle 21
3. Space Curves 21
4. The Osculating Plane and Circle 24
5. The Four Vertex Theorem 24
6. Problems 27
Chapter 3. Parametrized Surfaces 29
1. Parametrized Surfaces 29
2. Smooth Maps 30
3. The Tangent Plane 31
4. The Normal Vector 32
5. Level Surfaces 33
6. Quadric Surfaces 34
7. Ruled Surfaces 37
8. Surfaces of Revolution 38
9. The Inverse and Implicit Function Theorems 39
10. Problems 40
Chapter 4. The First Fundamental Form 43
1. Lengths of Curves on Surfaces 43
2. Isometries of Surfaces 44
3. Conformal Maps 45
4. Equiareal Maps 48
5. Problems 51
Chapter 5. Curvature of Surfaces 55
1. The Second Fundamental Form 55
2. The Gauß and Weingarten Maps 56
3. Normal and Geodesic Curvature 59
3
4 CONTENTS

4. The Covariant Derivative and Parallel Transport 60

5. Problems 63
Chapter 6. Principal Curvatures 65
1. Gaussian and Mean Curvatures 65
2. Principal Curvatures 68
3. Constant Gaussian Curvature 72
4. Flat Surfaces 73
5. Surfaces of Constant Mean Curvature 75
6. Problems 76
Chapter 7. Geodesics 79
1. Introduction to Geodesics 79
2. Geodesic Equations 80
3. Geodesics on Surfaces of Revolution 82
4. Geodesics as Shortest Paths 85
5. Geodesic Coordinates 88
6. Problems 88
Chapter 8. The Theorema Egregium 91
1. The Gauß and Codazzi–Mainardi Equations 91
2. The Theorema Egregium 93
3. Surfaces of Constant Gaussian Curvature 94
4. Geodesic Mappings 97
5. Problems 100
Chapter 9. The Gauß–Bonnet Theorem 103
1. Introduction to the Gauß–Bonnet Theorem 103
2. The Gauß–Bonnet Theorem for Simple Closed Curves 103
3. Gauß–Bonnet for Curvilinear Polygons 106
4. Gauss–Bonnet for Compact Surfaces 108
5. Holonomy 109
6. Problems 111
 CHAPTER 1

Parametrized Curves

1. Parametrized Curves
In this class, we will be primarily studying the differential geometry of curves and surfaces.
It is of course possible to study differential geometry in more dimensions, but this comes at
the cost of an extreme ramping up of technical difficulties, which I would like to avoid. Curves
and especially surfaces already illustrate most of the interesting phenomena in differential
geometry, and they can be handled in a relatively elementary manner.
We begin with curves, which are much simpler than surfaces. The first order of business
is to define what we mean by a curve in the context of differential geometry.
Definition 1.1. A parametrized curve, or simply a curve, in Rn is a continuous function
γ : (a, b) → Rn for some a, b with −∞ ≤ a < b ≤ ∞.
Example. The unit circle x2 + y 2 = 1 in R2 can be considered as a parametrized curve
by means of the parametrization γ : (−∞, ∞) → R2 defined by γ(t) = (cos t, sin t).
Example. The parabola y = x2 in R2 can be considered as a parametrized curve by
means of the parametrization γ : (−∞, ∞) → R2 defined by γ(t) = (t, t2 ).
Example. The astrid is the curve defined by the parametrization γ : (−∞, ∞) → R2
defined by γ(t) = (cos3 t, sin3 t). In rectangular coordinates, this is the curve x2/3 + y 2/3 = 1.
See Figure 1.

Figure 1. The astrid, defined by γ(t) = (cos3 t, sin3 t).

Example. The parametrized curve γ : (−∞, ∞) → R3 defined by γ(t) = (cos t, sin t, t)

is called a helix.
As we have seen, some curves close up, in the sense that over their domain of definition,
they repeat the same points over and over again. Such curves are known as closed curves.
Some curves, like the circle and astrid above, are closed curves, whereas the parabola and
helix are not.
Generally, we will consider parametrized curves γ such that each component of γ is a
smooth function, i.e. infinitely differentiable. We will assume this tacitly without making
5
6 1. PARAMETRIZED CURVES

 
cos t
Figure 2. The lemniscate, defined by γ(t) = , sin t cos t
1+sin2 t 1+sin2 t
.

any additional mention of it from here on out. Given a parametrized curve γ, defined by
γ(t) = (γ1 (t), . . . , γn (t)), we may differentiate it componentwise:
γ̇(t) = (γ10 (t), . . . , γn0 (t)).
Note that it is conventional to use the notation γ̇(t) for the derivative, rather than the more
familiar γ 0 (t).
2. The Tangent Vector
Definition 2.1. Let γ be a curve. Its tangent vector at the point γ(t) is defined to be
the vector γ̇(t).
Example. Consider the parabola γ(t) = (t, t2 ). Its tangent vector at the point γ(t) is
(1, 2t).
There is something a little bit unsettling about Definition 2.1. If we have a curve with
self-intersections, say with γ(t1 ) = γ(t2 ), then it might be the case that γ̇(t1 ) 6= γ̇(t2 ). Thus
the point γ(t1 ) = γ(t2 ) would have two different tangent vectors. For instance, consider the
lemniscate, defined by  
cos t sin t cos t
γ(t) = , ,
1 + sin2 t 1 + sin2 t
as shown in Figure 2. We have γ( π2 ) = γ( 3π 2
) = (0, 0), but
π       
1 1 3π 1 1
γ̇ = − ,− , γ̇ = ,− .
2 2 2 2 2 2
It would be technically more correct to talk about the tangent vector at time t rather than
the point γ(t), but we will be careless in this regard.
Let’s see why the terminology of “tangent vector” makes sense. For a small number δt,
consider the vector
γ(t + δt) − γ(t)
.
δt
This vector is parallel to the chord connecting the points γ(t) and γ(t + δt) of γ. As δt → 0,
the chord becomes parallel to the tangent to γ at γ(t), and this limiting vector is γ̇(t).
Proposition 2.2. If γ̇(t) is constant, then γ is contained in a straight line.
Proof. Suppose that γ̇(t) = a for all t, where a is a constant vector. Let us also suppose
that t0 is some point in the domain of γ. Then we have
Z t Z t
γ(t) = γ̇(u) du + γ(t0 ) = a du + γ(t0 ) = a(t − t0 ) + γ(t0 ).
t0 t0
 3. ARCLENGTH 7

If a 6= 0, this is the equation of a part of a line. If a = 0, then this is a single point. 

3. Arclength
Given a vector v = (v1 , . . . , vn ) ∈ Rn , its length kvk is given by the Pythagorean Theo-
rem: q
kvk = v12 + · · · + vn2 .
If γ is a curve and δt is a very small positive number, then the piece of γ between γ(t) and
γ(t + δt) is close to a segment of a line, so its length is approximately
kγ(t + δt) − γ(t)k.
Because δt is small, γ(t + δt) − γ(t) is very close to γ̇(t)δt, so the length of this segment is
approximately kγ̇(t)kδt.
If we wish to approximate the length of the curve γ between a and b, then we can divide
the interval from a to b into very small segments a = t0 < t1 < · · · < tn = b, calculate the
length of the line segment between each γ(ti−1 ) and γ(ti ), and add them all up. If each
ti − ti−1 is δt, then the length of the curve is approximately
n−1
X
kγ̇(ti )kδt.
i=0
Letting δt → 0, the sum turns into an integral, so the length of the curve becomes
Z b
kγ̇(t)k dt.
a

Definition 3.1. Let t0 be some point in the domain of γ. The arclength of γ starting
at the point γ(t0 ) is the function s(t) defined by
Z t
s(t) = kγ̇(u)k du.
t0

The arclength function s(t) is a differentiable function of t, for we have

d t
Z
ds
= kγ̇(u)k du = kγ̇(t)k
dt dt t0
by the Fundamental Theorem of Calculus.
If we imagine a particle moving along the curve, with its position being γ(t) at time t,
say starting with t0 = 0, then s(t) is the distance traveled by time t, so s0 (t) is the rate of
change of distance traveled, also known as the speed.
Definition 3.2. Let γ be a curve. Its speed at the point γ(t) is kγ̇(t)k. We say that γ
is a unit-speed curve if its speed is always 1, i.e. kγ̇(t)k = 1 for all t.
As we will see, many of our formulae simplify considerably for unit-speed curves. To see
why this is the case, we need to look at the derivative of a dot product of two vector-valued
functions.
Given two vectors a = (a1 , . . . , an ) and b = (b1 , . . . , bn ) in Rn , their dot product is
Xn
a·b= ai b i .
i=1
8 1. PARAMETRIZED CURVES

If a and b are differentiable functions of t, then we have

n n n
d X d X
0
X
(a · b) = (ai bi ) = ai b i + ai b0i = ȧ · b + a · ḃ.
dt i=1
dt i=1 i=1

Proposition 3.3. Suppose v(t) is a differentiable function of t such that v(t) is a unit
vector for all t. Then v̇(t) · v(t) = 0.
Thus v(t) is perpendicular to its derivative.
Proof. Since v(t) is a unit vector, we have v(t) · v(t) = 1 for all t. Differentiating both
sides with respect to t, we get v̇ · v + v · v̇ = 0, or 2v̇ · v = 0. Dividing by two yields the
desired result. 
The most important special case occurs when v(t) is the unit tangent vector to γ at γ(t).
If γ is a curve, then γ̇(t) is a tangent vector at γ(t). If γ̇(t) 6= 0, then we can divide by
kγ̇(t)k to get a unit tangent vector t(t) = kγ̇(t)
γ̇(t)k
. Its derivative n(t) = ṫ(t) is then orthogonal
to the curve. The reason that formulae simplify if γ is unit speed is that then γ̇(t) is already
a unit vector, so we do not need the denominator in t(t).
4. Reparametrization
Definition 4.1. Let γ : (a, b) → Rn be a curve. A reparametrization of γ is a curve
e : (ã, b̃) → Rn for which there exists a bijective smooth function φ : (ã, b̃) → (a, b) with a
γ
smooth inverse such that
γ
e (t̃) = γ(φ(t̃))
for all t̃ ∈ (ã, b̃)
e (φ−1 (t)) = γ(t) for all t ∈ (a, b).
Thus we also have γ
Example. We saw that γ(t) = (cos t, sin t) is a parametrization of the unit circle. A
reparametrization is γe (t) = (sin t, cos t). The reparametrization map φ can be taken to be
φ(t) = π2 − t, since we then have γ e (t̃) = γ(φ(t̃)).
Example. γ e (t) = (cos t3 , sin t3 ) is not a reparametrization
√ of the unit circle because if
we take φ(t) = t3 , then its inverse function φ−1 (t) = 3 t is not infinitely differentiable at
t = 0.
Definition 4.2. A point γ(t) on a curve γ is said to be a regular point if γ̇(t) 6= 0 and
a singular point if γ̇(t) = 0. A curve is said to be regular if all its points are regular.
Proposition 4.3. If γ is a regular curve, then any reparametrization of γ is also regular.
Proof. Suppose γ e (t) = γ(φ(t)). Let ψ = φ−1 . We
e is a reparametrization of γ, with γ
have φ(ψ(t)) = t, so, by differentiating, we get
dφ dψ
· = 1.
dt̃ dt
Thus dφ
dt̃
is never 0. Now, we have that
de
γ dγ dφ
= ·
dt̃ dt dt̃
is a product of two quantities that are each never 0, so the product is also never 0. 
 4. REPARAMETRIZATION 9

Proposition 4.4. If γ is a regular curve, then its arclength function s(t) is a smooth
function of t.
Proof. We have already seen that s is always differentiable, whether γ is regular or
not. So, we must show that it has higher derivatives as well when γ is regular. Suppose that
γ(t) = (γ1 (t), . . . , γn (t)). Then v
u n
ds u X
=t γi0 (t)2 .
dt i=1
√
P (0, ∞), the function x 7→ x is a smooth function of x. Since each γi is
On the interval
smooth and ni=1 γi0 (t)2 > 0 for all t, the function ds dt
is a composition of smooth functions
and hence itself smooth. 
Proposition 4.5. A curve γ has a unit-speed reparametrization if and only if it is
regular.
Proof. Suppose that γ has a unit-speed reparametrization γ e . Then γ
e is regular, so by
Proposition 4.3, γ is as well.
Now, suppose that γ is regular, i.e. the vector dγ dt
is never 0. Thus dsdt
> 0 for all t. Since
s is a smooth function with positive derivative everywhere, it is an injective function from
the interval of definition of γ to R, so it defines a bijection from the interval of definition
(a, b) to an open interval (ã, b̃) ⊆ R. Let φ = s−1 : (ã, b̃) → (a, b). Then
deγ ds dγ
= ,
ds dt dt
so
de
γ ds dγ ds
= = .
ds dt dt dt
de
γ
Thus k ds k = 1, i.e. γ
e is a unit-speed reparametrization of γ. 
While we now know that regular curves have unit-speed reparametrizations, the formulae
for these reparametrizations can be rather complicated, as they involve the inverse of the
arclength function, which is defined in terms of an integral. When the integral has no
elementary antiderivative, there is no way of simplifying the formula for the unit-speed
reparametrization. But even when the integral does have an elementary antiderivative, the
unit-speed reparametrization can be messy, as the following examples illustrates.
Example. Let γ(t) = (et cos t, et sin t). This curve is known as a logarithmic spiral, and
a variant of it is shown in Figure 3. We have γ̇(t) = (et (cos t − sin t), et (cos t + sin t)) and
kγ̇(t)k2 = 2e2t , which is never 0. The arclength is given by
Z t√ √
s(t) = 2et dt = 2(et − 1).
0
Thus, solving for t in terms of s, we get
 
s
t = log √ + 1 ,
2
which means that the unit-speed reparametrization is given by
       
s s s s
γ(s) = √ + 1 cos log √ + 1 , √ + 1 sin log √ + 1 .
2 2 2 2
10 1. PARAMETRIZED CURVES

 t t

Figure 3. The logarithmic spiral, defined by γ(t) = e 10 cos t, e 10 sin t .

While this is a closed formula, it might not be much fun to do calculations with it!

5. Curvature
The first major problem in the differential geometry of curves is to determine how much a
curve curves. Intuitively, it seems that a straight line doesn’t curve at all, and a large circle
only curves a little. However, a small circle curves a lot. The curvature should measure
something like the amount we have to zoom in on a curve to notice that it is curved. But of
course we need a precise definition, and the first challenge is come up with a good definition
that captures any intuition we might have appropriately. In the case of circles, it’s pretty
easy to define a notion of curvature that does what we want: we say that the curvature of a
circle of radius R is R1 . But what do we do for curves that aren’t circles?
One good way of capturing the notion of curvature is to look at the amount we move away
from the tangent line as we move a little on the curve. Suppose we have a unit-speed curve
γ in R2 . Then as we move a short distance on γ, say from γ(t) to γ(t + ∆t), we can measure
the vector between γ(t + ∆t) and its projection onto the tangent line at γ(t), as shown
in Figure 4. This displacement vector is perpendicular to the tangent line in the direction
of γ̇(t), hence in the direction of the unit normal vector n. The vector of displacement is
(γ(t + ∆t) − γ(t)) · n. By Taylor’s Theorem, we have
1
γ(t + ∆t) = γ(t) + γ̇(t)∆t + γ̈(t)(∆t)2 + higher order terms,
2
where the higher order terms grow more slowly than (∆t)2 as ∆t → 0. We can now take the
dot product with n, and since γ̇(t) · n = 0, we get
(∆t)2
(γ(t + ∆t) − γ(t)) · n = γ̈(t) · n + higher order terms.
2
2
If we now let ∆t → 0, we find that the displacement tends to (∆t)
2
γ̈(t)·n. By Proposition 3.3,
since γ̇(t) is a unit vector for all t, we have γ̈(t) · γ̇(t) = 0, so γ̈ is parallel to n, so
γ̈(t) · n = kγ̈(t)k. Thus we have
(∆t)2
k(γ(t + ∆t) − γ(t)) · nk ≈ kγ̈(t)k.
2
 5. CURVATURE 11

γ(t + ∆t)
(γ(t + ∆t) − γ(t)) · n

γ(t)

Figure 4. The displacement of the curve from the tangent line.

This suggests the following as a plausible definition:

Definition 5.1. Let γ be a unit-speed curve. Then its curvature κ(t) at time t is defined
to be κ(t) = kγ̈(t)k.
In order for this to be a good definition, it ought to agree with our proposed formula
for the curvature of a circle, i.e. that a circle of radius R has curvature R1 . So, let’s check.
Consider a circle centered at (x0 , y0 ) of radius R, which has unit-speed parametrization
 
t t
γ(t) = x0 + R cos , y0 + R sin .
R R
We have  
1 t 1 t
γ̈(t) = − cos , − sin ,
R R R R
so kγ̈(t)k = R1 , as claimed.
So far, we have only defined curvature for unit-speed curves. In theory, this takes care
of all regular curves, because there always exists a unit-speed reparametrization. But if the
unit-speed reparametrization involves a very complicated formula, then it may be challenging
to use it to compute kγ̈(t)k directly, so it would be nice to have a formula for curvature that
holds for any parametrization.
Proposition 5.2. Let γ be a regular curve in R3 . Then its curvature is given by
kγ̈ × γ̇k
κ= .
kγ̇k3
Proof. Let s be a unit-speed parameter for γ. Then we have
dγ dγ ds
γ̇ = = ,
dt ds dt
so
 
d dγ/dt
ds d2 γ 2
2 − ddt2s dγ
 
dγ d dγ/dt dt ds/dt dt dt2 dt
(5.1) κ= = = = .
ds2 ds ds/dt ds/dt (ds/dt)3

We have  2
ds
= kγ̇k2 = γ̇ · γ̇,
dt
12 1. PARAMETRIZED CURVES

so
ds d2 s
= γ̇ · γ̈.
dt dt2
Thus we have
ds 2 d2 γ 2
d s dγ ds 2 2
− ds γ̈ − ddt2s ds



dt dt2 dt dt2 dt dt dt
γ̇ k(γ̇ · γ̇)γ̈ − (γ̇ · γ̈)γ̇k
κ= = = .
(ds/dt)4 (ds/dt) 4 kγ̇k4
Using the identity
a × (b × c) = (a · c)b − (a · b)c
for a, b, c ∈ R3 , we get
γ̇ × (γ̈ × γ̇) = (γ̇ · γ̇)γ̈ − (γ̇ · γ̈)γ̇.
Since γ̇ and γ̈ are perpendicular, we have
kγ̇ × (γ̈ × γ̇)k = kγ̇kkγ̈ × γ̇k.
Thus we have
k(γ̇ · γ̇)γ̈ − (γ̇ · γ̈)γ̇k kγ̇ × (γ̈ × γ̇)k kγ̇kkγ̈ × γ̇k kγ̈ × γ̇k
κ= 4
= 4
= 4
= ,
kγ̇k kγ̇k kγ̇k kγ̇k3
as desired. 
Example. Consider the helix, defined by γ(θ) = (a cos θ, a sin θ, bθ), where a, b ∈ R are
constants. We have
γ̇(θ) = (−a sin θ, a cos θ, b), γ̈(θ) = (−a cos θ, −a sin θ, 0).
Thus p √
kγ̇(θ)k = a2 sin2 θ + a2 cos2 θ + b2 = a2 + b 2
and
p √
γ̈×γ̇ = (−ab sin θ, ab cos θ, −a2 ), kγ̈×γ̇k = a2 b2 sin2 θ + a2 b2 cos2 θ + a4 = |a| a2 + b2 .
Thus we have √
|a| a2 + b2 |a|
κ= 2 = .
(a + b2 )3/2 a2 + b 2

6. The Isoperimetric Inequality

We now turn to our first global fact about curves, i.e. something that depends on the
entire curve, rather than just an infinitesimal piece of it.
Definition 6.1. A simple closed curve is a curve with no self-intersections except at the
endpoints, which are equal.
In other words, a simple closed curve is a loop.
Theorem 6.2 (Isoperimetric Inequality). Let γ be a simple closed curve of length `,
enclosing an area of A. Then `2 ≥ 4πA, with equality if and only if γ is a circle.
There are many proofs of the isoperimetric inequality. We’ll give a proof based on
Wirtinger’s inequality. See problem 13 for another argument that isn’t quite a proof.
 6. THE ISOPERIMETRIC INEQUALITY 13

Theorem 6.3 (Wirtinger’s

R 2π Inequality). Let f (t) be a piecewise smooth, continuous 2π-
periodic function such that 0 f (t) dt = 0. Then
Z 2π Z 2π
0 2
f (t) dt ≥ f (t)2 dt,
0 0

with equality if and only if f (t) = a cos(t + b) for some a, b.

We’ll first prove the isoperimetric inequality assuming Wirtinger’s inequality, and then
we’ll prove Wirtinger’s inequality.
Proof of the Isoperimetric Inequality. Parametrize γ so as to have constant
`
speed 2π . Write γ(t) = (x(t), y(t)). By translating the curve if necessary, we may assume
R 2π
that 0 x(t) dt = 0. Now, we have
Z 2π Z 2π
`2
= 2
kγ̇(t)k dt = (x0 (t)2 + y 0 (t)2 ) dt.
2π 0 0

By Green’s Theorem, we have

Z Z 2π
A= dx dy = x(t)y 0 (t) dt,
int(γ) 0

where int(γ) is the interior of γ. Thus we have

Z 2π
2
` − 4πA = 2π (x0 (t)2 + y 0 (t)2 − 2x(t)y 0 (t)) dt
0
Z 2π Z 2π
0
= 2π 2 2
(x (t) − x(t) ) dt + 2π (y 0 (t) − x(t))2 dt
0 0
≥ 0,
where the first integral is nonnegative by Wirtinger’s inequality, and the second inntegral is
nonnegative because the integrand is nonnegative.
Equality holds if and only if x(t) = a cos(t
p + b) for some a, b and y 0 (t) = x(t), so that
` `
y(t) = a sin(t + b) + c for some c. Because x0 (t)2 + y 0 (t)2 = 2π , we have |a| = 2π . Thus γ
`
is a circle with center (0, c) and radius 2π . 
We now turn to the proof of Wirtinger’s Inequality. We begin by proving a weaker
version:
Theorem 6.4. Let f : [0, π] → R be a continuous function, smooth on (0, π), such that
f (0) = f (π) = 0 and such that limt→0+ f 0 (t) and limt→π− f 0 (t) exist. Then
Z π Z π
2
f (t) dt ≤ f 0 (t)2 dt,
0 0

with equality if and only if f (t) = a sin t for some a.

Rπ
Proof. We need to show that 0 [f 0 (t)2 − f (t)2 ] dt ≥ 0. To do this, we seek a function
ψ such that Z π Z π
[f 0 (t)2 − f (t)2 ] dt = [f 0 (t) − f (t)ψ(t)]2 dt,
0 0
14 1. PARAMETRIZED CURVES

i.e.
Z π
I= [2f (t)f 0 (t)ψ(t) − f (t)2 (1 + ψ(t)2 )] dt = 0.
0

Integration by parts gives

Z π
π
2
I = f (t)ψ(t) 0
− f 2 (t)[ψ 0 (t) + 1 + ψ(t)2 ] dt = 0.
0

(We should not necessarily conclude that the first term is 0 yet, because ψ could have poles
at 0 or π.) The differential equation ψ 0 + 1 + ψ 2 = 0 has solution ψ(t) = − tan(t + t0 ).
Letting t = π2 (so that ψ is defined everywhere on the interval (0, π), we get ψ(t) = cot t.
We find that
lim f 2 (t) cot t = lim− f 2 (t) cot = 0.
t→0+ t→π

Thus I = 0. Now, we have

Z π Z π
0 2 2
[f (t) − f (t) ] dt = [f 0 (t) − f (t) cot t]2 dt ≥ 0,
0 0

with equality if and only if f 0 (t) = f (t) cot t for all t, i.e. if f (t) = a sin t for some a. 

This doesn’t prove the full version of Wirtinger’s Inequality, because the assumption
there is that f is only piecewise smooth, and without the assumption that f is 0 at the
endpoints. So we need another small idea to prove the full theorem.

Proof of Wirtinger’s Inequality. First, note that there exists some t0 with 0 ≤
t0 < π such that f (t0 ) = f (t0 + π), because the function h(t) = f (t) − f (t + π) has opposite
signs at t = 0 and t = π, so by the Intermediate Value Theorem, there must be some
t0 ∈ [0, π] such that h(t0 ) = 0. For this value of t0 , let f (t0 ) = f (t0 + π) = c. Then, applying
the argument in the proof of Theorem 6.4 to the function f (t) − c with ψ(t) = cot(t − t0 ),
we get
Z 2π Z 2π
0 2
2 2
[f (t) − (f (t) − c) ] dt − [f 0 (t) − (f (t) − c) cot(t − t0 )] dt
0 0
2π
= (f (t) − c)2 cot(t − t0 ) 0
= 0.

Thus we have
Z 2π
[f 0 (t)2 − (f (t) − c)2 ] dt ≥ 0,
0

R 2π only if f (t) − c = Ra2π

with equality if and sin(t − t0 ). To get rid of the constant c, we use the
assumption that 0 f (t) dt = 0, so 2c 0 f (t) dt = 0. Thus we have
Z 2π
[f 0 (t)2 − f (t)2 ] dt ≥ 2πc2 ≥ 0,
0

with equality if and only if c = 0, so that f (t) = a sin(t − t0 ). 

 7. PROBLEMS 15

7. Problems
x2 y2
(1) Consider the ellipse a2 + b2 = 1, where a > b > 0. Define the eccentricity of the
q
2
ellipse to be e = 1 − ab 2 . The foci 1 of the ellipse are the points f1 = (ea, 0) and
f2 = (−ea, 0).
(a) Show that there exists a constant c such that for all p on the ellipse, we have
kp − f1 k + kp − f2 k = c.
(b) Show that there exists a constant d such that for all points p on the ellipse,
the product of the distances from f1 and f2 to the tangent line is equal to d.
(c) Show that if p is any point on the ellipse, the lines joining f1 and f2 to p make
equal angles with the tangent line to the ellipse at p.
(2) A cycloid is the plane curve traced out by a point on the circumference of a circle
as it rolls without slipping along a straight line.
(a) Assume that the line is the x-axis, and the circle has radius 1. Find a parametriza-
tion of the cycloid.
(b) What is the arclength of one period of the cycloid?
(3) Find a parametrization for the curve of intersection of the cylinder x2 + y 2 = 1 and
the sphere (x + 1)2 + y 2 + z 2 = 4.
(4) A hypocycloid and an epicycloid are the curves obtained by tracing out the trajectory
of a fixed point on the circumference of a small circle as it rolls on the inside
(respectively, the outside) of a larger circle. Suppose the large circle has radius
c > 1, and the small circle has radius 1. If c is an integer, find the arclength of the
hypocycloid and epicycloid.
(5) Let γ be a curve in the plane not passing through the origin. Let t0 be such that
γ(t0 ) is the closest point of γ to the origin. Show that γ(t0 ) is orthogonal to γ̇(t0 ).
(6) Find all curves γ such that γ̈ = 0.
(7) Let γ : (0, π) → R2 be the curve given by
 
t
γ(t) = sin t, cos t + log tan .
2
This curve is called the tractrix. Show that for every point p on the tractrix, the
length of the segment of the tangent line from p to the y-axis is equal to 1. This
curve will be important later on, as a model of hyperbolic geometry.
(8) Find a unit-speed parametrization of the parabola y = x2 .
(9) As we have seen, the arclength of a curve γ(t) = (x(t), y(t)) in R2 in rectangular
coordinates is given by
Z bp
x0 (t)2 + y 0 (t)2 dt.
a

Find a similar formula for a curve in R2 in terms of polar coordinates. What about
cylindrical and spherical coordinates in for a curve in R3 ?
(10) Consider the ellipse parametrized by γ(t) = (a cos t, b sin t). At what points is the
speed maximized? Minimized?
(11) You know that a straight line is the shortest distance between two points, but do
you know how to prove it? This exercise will help you do so! Let p and q be two
1the singular of foci is focus
16 1. PARAMETRIZED CURVES

points, and let γ be a curve passing through both of them, say with γ(a) = p and
γ(b) = q, where a < b.
(a) Show that if u is any unit vector, then
γ̇ · u ≤ kγ̇k.
(b) Show that
Z b
(q − p) · u ≤ kγ̇k dt.
a
(c) Finish the proof that the shortest distance between two points is a straight line.
(12) Let γ be a regular curve in Rn . Suppose that the function t 7→ kγ(t)k has a local
maximum of r at time t0 . Show that κ(t0 ) ≥ 1r .
(13) In this exercise, we will give a heuristic argument for the isoperimetric inequality,
due to Steiner. Let γ be a unit-speed simple closed curve in the plane maximizing
the area subject to having a fixed arclength, say ` = `(γ).
(a) Show that the interior of γ must be convex. (A set S is said to be convex if
for any two points p, q ∈ S, the line segment connecting p and q is entirely
contained in S.)
(b) Show that, after a suitable translation, we may assume that γ(t) = γ(t + 2` )
for all t ∈ [0, 2` ].
(c) Let p and q be two distinct points in the plane. Show that if δ is a curve such
that for any point r ∈ δ, the angle ∠prq is a right angle, then δ is part of a
circle with diameter pq.
(d) Call γ 1 and γ 2 the restrictions of γ to [0, 2` ] and [ 2` , `], respectively. Show that
γ 1 and γ 2 must be semicircles.
(e) Why doesn’t this argument give a rigorous proof of the isoperimetric inequality?
(14) Prove that if a, b > 0, then
Z 2π p √
a2 sin2 t + b2 cos2 t dt ≥ 2π ab.
0
 CHAPTER 2

Curvature, Torsion, and the Frenet–Serret Frame

1. Curvature of Plane Curves

Let’s now restrict to the case where γ is a unit-speed plane curve, i.e. one contained in
R2 . Then t = γ̇ is a unit tangent vector to γ. There are two unit vectors perpendicular to
t, and we choose the signed unit normal vector ns to be the vector obtained by rotating t
counterclockwise by π2 . See Figure 1.

ns t

γ(t)

Figure 1. The tangent and signed normal vectors to a unit-speed plane

curve.

As we have already seen, ṫ = γ̈ is perpendicular to t, hence parallel (or antiparallel) to

ns . Thus there is a scalar κs such that
γ̈ = κs ns .
We call κs the signed curvature of γ. The relation to the notion of curvature we saw
previously is that
κ = kγ̈k = kκs ns k = |κs |,
so the curvature of γ is just the absolute value of the signed curvature. The signed curvature
is positive if the signed normal points inward, whereas it is negative if the signed normal
points outward. See Figure 2.
If γ is regular but not necessarily unit-speed, we can still define t, ns , and κs to be those
of the unit-speed reparametrization of γ. We have
dγ/dt dγ/dt
t= =
ds/dt kdγ/dtk
and
dt dt ds ds dγ
= = κs ns = κs ns .
dt ds dt dt dt
17
18 2. CURVATURE, TORSION, AND THE FRENET–SERRET FRAME

ns t
γ

ns
t

Figure 2. Left: κs > 0. Right: κs < 0.

The signed curvature has another interpretation as the angle at which the tangent vector
rotates. If γ is unit-speed, then for each s, there exists a number φ(s) such that
γ̇(s) = (cos φ(s), sin φ(s)).
The choice of φ(s) is not unique because we can always add 2π to it without changing cos φ(s)
or sin φ(s). However, it is possible to make a choice of φ(s) that varies smoothly with s.
Proposition 1.1. Let γ : (a, b) → R2 be a unit-speed curve, let s0 ∈ (a, b), and let φ0 be
such that γ̇(s0 ) = (cos φ0 , sin φ0 ). Then there exists a unique smooth function φ : (a, b) → R
such that φ(s0 ) = φ0 and such that
γ̇(s) = (cos φ(s), sin φ(s))
for all s ∈ (a, b).
Proof. Let γ̇(s) = (f (s), g(s)). Then we have f (s)2 + g(s)2 = 1 for all s, so, taking
derivatives, we have 2(f (s)f 0 (s) + g(s)g 0 (s)) = 0. Let
Z s
φ(s) = φ0 + (f (t)g 0 (t) − g(t)f 0 (t)) dt.
s0

Then φ(s0 ) = φ0 , and φ is smooth. Note that

φ0 (s) = f (s)g 0 (s) − g(s)f 0 (s).
Let
F (s) = f (s) cos φ(s) + g(s) sin φ(s), G(s) = f (s) sin φ(s) − g(s) cos φ(s).
Then
F 0 (s) = (f 0 (s) + g(s)φ0 (s)) cos φ(s) + (g 0 (s) − f (s)φ0 (s)) sin φ(s).
We have
f 0 + gφ0 = f 0 + g(f g 0 − gf 0 ) = f 0 (1 − g 2 ) + f gg 0 = f 0 f 2 + f gg 0 = f (f f 0 + gg 0 ) = 0.
Similarly, g 0 − f φ0 = 0. Thus F 0 = 0, so F is constant. Similarly, G is constant. Now, we
have
F (s0 ) = f (s0 ) cos φ0 + g(s0 ) sin φ0 = cos2 φ0 + sin2 φ0 = 1,
and similarly G(s0 ) = 0. Thus F is the constant 1 function, and G is the constant 0 function.
Thus we have
f cos φ + g sin φ = 1, f sin φ − g cos φ = 0
for all s. Thus f = cos φ and g = sin φ. This completes the proof of existence.
 1. CURVATURE OF PLANE CURVES 19

For uniqueness, suppose that ψ is another smooth function such that ψ(s0 ) = φ0 and
γ̇(s) = cos(ψ(s), sin ψ(s)) for all s ∈ (a, b). Then there exists an integer n(s) such that
ψ(s) − φ(s) = 2πn(s)
for all s ∈ (a, b). Since φ and ψ are smooth, so is n. The only smooth functions from (a, b)
to Z are constant functions, so n is a constant function. Since n(s0 ) = 0, n must be the
constant 0 function, i.e. ψ = φ. 
Definition 1.2. The function φ in Proposition 1.1 is called the turning angle of γ
determined by the condition φ(s0 ) = φ0 .
We can relate the turning angle to the signed curvature:
Proposition 1.3. Let γ(s) be a unit-speed plane curve, and let φ(s) be a turning angle
for γ. Then
dφ
κs = .
ds
In other words, the signed curvature is the rate at which the tangent vector rotates.
Proof. We have γ̇ = t = (cos φ, sin φ). Thus ṫ = (−φ0 sin φ, φ0 cos φ). Since ns =
(− sin φ, cos φ), the equation ṫ = κs ns implies the result. 
Example. Consider the catenary γ(t) = (t, cosh t). We have γ̇ = (1, sinh t), so
Z tp
s= 1 + sinh2 t dt = sinh t.
0
If φ is the angle between γ̇ and the x-axis, we have
tan φ = sinh t = s.
Thus
dφ
sec2 φ
= 1,
ds
so
dφ 1 1 1
κs = = 2
= 2
= .
ds sec φ 1 + tan φ 1 + s2
Now,
R ` suppose we have a closed curve γ of length `. We can ask about the total curvature,
i.e. 0 κs (s) ds.
Corollary 1.4. The total signed curvature of a closed plane curve is a multiple of 2π.
Proof. We have Z ` Z `
dφ
κs ds = ds = φ(`) − φ(0).
0 0 ds
Since γ is a closed curve, we have γ(s + `) = γ(s) for all s, so we have γ̇(s + `) = γ̇(s) as
well, so in particular γ̇(`) = γ̇(0). Thus we have (cos φ(`), sin φ(`)) = (cos φ(0), sin φ(0)), so
φ(`) − φ(0) is a multiple of 2π. 
Given a smooth function k : (a, b) → R, we might ask if there exists a plane curve whose
signed curvature is equal to k.
Proposition 1.5. Let k : (a, b) → R be a smooth function. Then there is a unit-
speed curve γ : (a, b) → R2 whose signed curvature is k. Furthermore, γ is unique up to
orientation-preserving isometry (i.e. a composition of a translation and a rotation).
20 2. CURVATURE, TORSION, AND THE FRENET–SERRET FRAME

Proof. Fix s0 ∈ (a, b), and define φ : (a, b) → R by

Z s
φ(s) = k(u) du
s0

and Z s Z s 
γ(s) = cos φ(t) dt, sin φ(t) dt .
s0 s0

Then the tangent vector at γ(s) is

γ̇(s) = (cos φ(s), sin φ(s)).
Thus γ is unit speed, and its signed curvature is
d s
Z
dφ
= k(u) du = k(s).
ds ds s0

For uniqueness, let φ(s)

e be a smooth turning angle for γ
e . Then
e˙ (s) = (cos φ(s),
γ e sin φ(s)),
e

so Z s Z s 
γ
e (s) = cos φ(t)
e dt, sin φ(t)
e dt + γ
e (s0 ).
s0 s0
dφ
Thus k(s) = , so
e
ds Z s
φ(s)
e = k(u) du + φ(s
e 0 ) = φ(s) + φ(s
e 0 ).
s0
Thus we have
Z s Z s 
γ
e (s) = cos(φ(t) + φ(s
e 0 )) dt, sin(φ(t) + φ(s
e 0 )) dt + γe (s0 )
s0 s0
 Z s Z s
= cos φ(s0 )
e cos φ(t) dt − sin φ(s0 )
e sin φ(t) dt,
s0 s0
Z s Z s 
sin φ(s0 )
e cos φ(t) dt + cos φ(s0 )
e sin φ(t) dt + γe (s0 ).
s0 s0

This is the result of first applying a rotation by φ(s

e 0 ) about the origin, then translating by
γ
e (s0 ). 
This implies, for instance, that any plane curve whose signed curvature is a positive
constant is a counterclockwise arc of a circle, whereas if the signed curvature is a negative
constant, then the curve is a clockwise arc of a circle.
While we have shown that any smooth function is the signed curvature of a plane curve,
the curve itself may be complicated to describe. For instance, a curve whose signed curvature
is κs (s) = s is given by
Z s Z s
t2 t2

γ(s) = cos dt, sin dt .
0 2 0 2
The integrals, known as Fresnel integrals, cannot be evaluated in closed form.
 3. SPACE CURVES 21

2. The Osculating Circle

So far, we have one interpretation of the curvature, as the change in the tangent angle.
Another important interpretation of the curvature is related to the best fit circle. Recall
from calculus that the tangent line to a curve is the best-fit line near a point, i.e. the line
carrying the same first-derivative information as the curve, so the derivative is the slope of
the tangent line. We can just as easily ask for the circle with the same second-derivative
information as the curve, and that circle is called the osculating circle: this circle is tangent
to the curve at the point in question, and it has the same curvature.
This gives us a new way of understanding the curvature. Just as we can approximate
the tangent line to a curve γ at a point p by choosing a point q on γ near p and drawing
the line between them, we can approximate the osculating circle to γ at p by picking two
nearby points q1 , q2 on γ and drawing the circle through p, q1 , q2 . The curvature is then
the reciprocal of the radius of the osculating circle.

3. Space Curves
Now let’s turn to the study of curves in R3 . Let γ be a unit-speed curve in R3 , and let
t = γ̇ be its tangent vector. Recall that if κ(s) 6= 0, then we can define the unit normal
vector of γ by
1
n(s) = ṫ(s),
κ(s)
which is a unit vector since kṫk = κ. Thus t and n are perpendicular unit vectors in R3 .
There is a third vector perpendicular to both of them, namely
b = t × n.
(Actually, there are two of them, with the other one being −b.) The vector b(s) is called
the binormal vector of γ at the point γ(s). Thus {t, n, b} is a right-handed orthonormal
basis of R3 , i.e. it satisfies the cross-product identities
b = t × n, n = b × t, t = n × b.
Since b is always a unit vector, ḃ is perpendicular to b, which means that it lies in
the plane spanned by t and n. We have the following formula for the derivative of a cross
product:
d du dv
(u × v) = ×v+u× .
ds ds ds
Thus we have
ḃ = ṫ × n + t × ṅ = t × ṅ,
because ṫ × n = κn × n = 0. Thus ḃ is perpendicular to t. Since it is perpendicular to both
t and b, ḃ must be parallel to n, so there is some scalar τ such that
ḃ = −τ n.
(The minus sign is just a convention that helps to improve some later formulae.) Note that
τ is only defined when κ 6= 0. We call τ the torsion of γ.
If we merely have a regular curve, rather than a unit-speed curve, we can still calculate
the torsion, thanks to the following formula:
22 2. CURVATURE, TORSION, AND THE FRENET–SERRET FRAME

Proposition 3.1. Let γ(t) be a regular curve in R3 such that κ(t) 6= 0 for all t. Then,
letting a dot denote a derivative with respect to t, we have
(γ̇ × γ̈) · γ
˙˙˙
τ= 2
.
kγ̇ × γ̈k
Proof. We first assume that γ is unit-speed. Then we have
d
τ = −n · ḃ = −n · (t × n) = −n · (ṫ × n + t × ṅ) = −n · (t × ṅ).
dt
1 1
Because n = κ ṫ = κ γ̈, we have
κ0
     
1 d 1 1 1 1
τ = − γ̈ · γ̇ × γ̈ = − γ̈ · γ̇ × ˙˙˙ − 2 γ̈
γ = 2γ˙˙˙ · (γ̇ × γ̈).
κ dt κ κ κ κ κ
This is what we wanted, since kγ̇ × γ̈k = κ.
Now, we consider the general case, where γ is not necessarily unit-speed. Let s be the
arclength. Then we have
dγ ds dγ
= ,
dt dt ds
 2 2
d2 γ ds d γ d2 s dγ
= + 2 ,
dt2 dt ds2 dt ds
 3 3
d3 γ ds d γ ds d2 s d2 γ d3 s dγ
= + 3 + 3 .
dt3 dt ds3 dt dt2 ds2 dt ds
Thus we have
 3 
dγ d2 γ

ds
γ̇ × γ̈ = × 2 ,
dt ds ds
 6  3
dγ d2 γ
 
ds dγ
˙˙˙ · (γ̇ × γ̈) =
γ · × 2 .
dt ds3 ds ds
Thus the torsion is  
d3 γ dγ d2 γ
ds3
· ds × ds2 ˙˙˙ · (γ̇ × γ̈)
γ
τ= 2 = ,
dγ 2
d γ kγ̇ × γ̈k2
ds
× ds2
as desired. 
Example. Consider the helix γ(θ) = (a cos θ, a sin θ, bθ). We have
γ̇(θ) = (−a sin θ, a cos θ, b),
γ̈(θ) = (−a cos θ, −a sin θ, 0),
˙˙˙ = (a sin θ, −a cos θ, 0),
γ(θ)
γ̇ × γ̈ = (ab sin θ, −ab cos θ, a2 )
kγ̇ × γ̈k2 − a2 (a2 + b2 ),
˙˙˙ = a2 b.
(γ̇ × γ̈) · γ
Thus we have
(γ̇ × γ̈) · γ
˙˙˙ a2 b b
τ= 2
= 2 2 2
= 2 .
kγ̇ × γ̈k a (a + b ) a + b2
 3. SPACE CURVES 23

Just as we can consider κ a measure of the extent to which γ is not contained in a line, we
can consider τ a measure of the extent to which γ is not contained in a plane. In particular,
if γ is contained in a plane, then τ = 0.
Proposition 3.2. Let γ be a regular curve in R3 with nowhere vanishing curvature.
Then γ is contained in a plane if and only if τ = 0 at every point of the curve.
Proof. Assume that γ is unit speed, by reparametrizing if necessary. We let s be the
d
parameter of γ, and we denote ds by a dot.
Suppose that γ is contained in the plane v · N = d, where N is a constant vector and d
is a constant scalar, i.e. γ(s) · N = d for all s. We may assume, by scaling if necessary, that
N is a unit vector. Differentiating the equation γ · N = d with respect to s, we get
t · N = 0,
and differentiating again, we get
ṫ · N = 0.
Thus
κn · N = 0,
so
n · N = 0.
Thus both t and n are perpendicular to N, so b = t × n is parallel to N. Since N and b
are both unit vectors and b(s) is a smoothly-varying function of s, we must have b = N or
b = −N for all s. In both cases, b is a constant vector, so ḃ = 0, so τ = 0.
Now, suppose that τ = 0 for all s. Then ḃ = 0, so b is a constant vector. We will show
that γ is contained in a plane of the form v · b = c. We have
d
(γ · b) = γ̇ · b = t · b = 0,
ds
so γ · b is some constant, say c. Thus γ is contained in the plane v · b = c. 
So far, we have computed ṫ = κn and ḃ = −τ n, but we have not yet calculated ṅ. Let
us do this now. We have n = b × t, so
ṅ = ḃ × t + b × ṫ = −τ n × t + κb × n = −κt + τ b.
The equations
ṫ = κn, ṅ = −κt + τ b, ḃ = −τ n
are called the Frenet–Serret equations, and the orthonormal basis {t, n, b} is called the
Frenet–Serret frame of γ.
We saw that the signed curvature of a plane curve determines the curve up to isometry.
There is an analogous result for space curves, which are determined by their curvature and
torsion.
Theorem 3.3. Let k and t be smooth functions with k > 0 everywhere. Then there is a
unit-speed curve γ in R3 with curvature k and torsion t. Furthermore, if γ and γ
e are two
such curves, then there is an orientation-preserving isometry mapping γ to γ
e.
You will prove Theorem 3.3 in problem 15.
24 2. CURVATURE, TORSION, AND THE FRENET–SERRET FRAME

4. The Osculating Plane and Circle

Just like in the case of curves in the plane, curves in space have osculating circles.
Definition 4.1. Let γ be a space curve, and let γ(t) be a point on γ. The plane through
γ(t) containing vectors parallel to t and n is called the osculating plane to γ at γ(t).
1
The osculating circle is then the circle with radius κ
in the osculating plane tangent to
γ at γ(t).

5. The Four Vertex Theorem

Last week, we encountered our first global result about differential geometry: the isoperi-
metric inequality. We now look at our second such result, known as the four vertex theorem.
dκs
Definition 5.1. Let γ be a plane curve. A vertex of γ is a point where dt
= 0.
Example. Consider the ellipse, parametrized by γ(t) = (a cos t, b sin t). Then we can
calculate that
ab
κs (t) = 2 2 .
(a sin t + b2 cos2 t)3/2
Thus we have
dκs 3ab(b2 − a2 ) sin t cos t
= 2 2 .
dt (a sin t + b2 cos2 t)5/2
This vanishes when t is a multiple of π2 . Thus we find that the ellipse has four vertices: the
ends of the two axes of the ellipse. Unless a = b, these are the only vertices.
It’s no accident that the ellipse has four vertices: this is the smallest possible number for
a simple closed curve.
Theorem 5.2 (Four Vertex Theorem). Every simple closed curve in R2 has at least four
vertices.
We present Osserman’s proof, which actually proves the following stronger result. Let γ
be a simple closed curve, and let C be the circumscribed circle of γ, i.e. the circle of smallest
radius that encloses γ.1 Then C intersects γ at at least 2 points. If γ intersects C at an arc,
then any point on the arc is a vertex. Otherwise, γ intersects C in two or more connected
components.
Theorem 5.3 (Osserman). If γ ∩ C has at least n components, then γ has at least 2n
vertices.
Proof. Let us suppose the radius of the circle C is K1 , so that its curvature is K. Then
at any point of C ∩ γ, the tangent angle to γ changes faster than that of C, which means
that κ ≥ K. Now, we claim that between any two points of intersection of γ and C, there is
a point with curvature κ ≤ K. To see this, let us parametrize γ and C by angle with respect
to some point in the interior of γ. Then if P1 and P2 are two points of intersection in γ ∩ C,
then the tangent lines to γ and C are the same at P1 and P2 , so the total turning angle
between P1 and P2 is the same for γ and C. Thus it cannot be the case that the turning
1Notethat this definition disagrees with the usual definition for triangles: the circumcircle of a triangle
need not be the circle of smallest radius containing the triangle.
 5. THE FOUR VERTEX THEOREM 25

angle for γ is everywhere greater than that of C, which means that there is some point on
γ between P1 and P2 such that κ < K.
Finally, we note that γ has n points at which κ ≥ K and n points at which κ ≤ K,
so there must be at least n local maxima and n local minima, for a total of at least 2n
vertices. 
From the four vertex theorem, we learn that the curvature of a simple closed curve cannot
be a completely arbitrary periodic function: it must have at least two local maxima and two
local minima. However, there is a converse to the four vertex theorem that tells us that this
is the only restriction.
Theorem 5.4 (Converse to the Four Vertex Theorem). Let κ : [0, 1] → R be a continuous
function such that κ(0) = κ(1), with at least two local maxima and two local minima. Then
there is a simple closed curve γ : [0, 1] → R2 whose curvature is κ.
We will prove this converse in the case where κ(t) > 0 for all t, and we have at least two
strict local maxima and two strict local minima. Let us parametrize the curve by the tangent
angle, so κ : [0, 2π] → R>0 with κ(0) = κ(2π). There is a unique curve γ : [0, 2π] → R2
beginning at the origin with unit tangent vector (cos θ, sin θ) and curvature κ(θ) at the point
γ(θ): if s is the arclength, then we have γ̇ = (cos θ, sin θ) and θ0 = κ(θ), where all derivatives
are with respect to s. Thus we have
Z θ
(cos u, sin u)
γ(θ) = du.
0 κ(u)
However, this usually won’t be a simple closed curve, because γ(0) need not be equal to
γ(2π). So, we define the error vector to be E = γ(2π) − γ(0), which measures the failure of
the curve to close up.
What we need to show is that if κ has at least two local maxima and two local minima,
then there is a modification of γ that closes up.
Definition 5.5. A diffeomorphism of the circle S1 is a smooth function φ : S1 → S1
that is bijective with a smooth inverse.
Think of a diffeomorphism of the circle as a reparametrization that traverses the same
points in the same order, but at a different speed.
We will find a continuous family (or a loop) of diffeomorphisms of the circle such that
the error vectors wind once around the origin. Then we will show that we can contract this
loop of diffeomorphisms to a point in the middle. This will correspondingly map the error
vector to the zero vector.
Given any diffeomorphism h of S1 , we get a new curve γ h : [0, 2π] → R2 by setting
γ h (θ) = γ(h(θ)). Thus the curvature of γ h at the point γ h (θ) is equal to the curvature of γ
at the point γ(h(θ)), i.e. κ(h(θ)).
So, how do we find this loop of diffeomorphisms? Since κ has two strict local minima
and two strict local maxima, we can find positive numbers a and b with 0 < a < b such
that κ takes the values a, b, a, b in that order as we travel around the loop. Then we find a
diffeomorphism h1 of the circle such that κ ◦ h1 is close to the step function κ0 taking on
the values a, b, a, b on the intervals (0, π2 ), ( π2 , π), (π, 3π
2
), ( 3π
2
, 2π), respectively. See Figure 3.
If the curvatures κ0 and κ ◦ h1 are close, then so are the corresponding curves. The curve
corresponding to κ0 consists of four quarter circles, of radii a1 , 1b , a1 , 1b .
26 2. CURVATURE, TORSION, AND THE FRENET–SERRET FRAME

ε
4
κ ◦ h1 ∼ b κ ◦ h1 ∼ a

ε ε
4 4

κ ◦ h1 ∼ a κ ◦ h1 ∼ b
ε
4

Figure 3. κ ◦ h1 is close to a step function.

Figure 4. Four quarter circles of radius 1, 4, 1, 4 closing up.

Figure 5. As we change the angles of the arcs slightly, the error vector
makes a loop around the origin.

Now, the curve corresponding to the actual step function closes up, since it’s four quarter
circles of radii a1 , 1b , a1 , 1b , as shown in Figure 4. However, that’s not quite what we have,
because we can only approximate the step function. But fortunately, we can change the
length of each arc of the step function so that the corresponding curves don’t close up.
Then, as we modify the angles slightly, the error vectors loop around the origin, as shown in
Figure 5.
 6. PROBLEMS 27

Now, we can shrink this loop of diffeomorphisms to a point by making the angle changes
smaller and smaller. Thus there must be some point in the middle where the error vector is
zero.
This argument can be found in the paper “The Four Vertex Theorem and Its Converse”
by DeTurck, Gluck, Pomerleano, and Vick.2

6. Problems
(1) Prove that if a curve has constant nonzero curvature, and its torsion is zero every-
where, then it lies in a circle.
(2) Prove that if γ is a unit-speed plane curve, then
ṅs = −κs t.
(3) Let γ be a curve (plane or space), and let γ a be the dilation, defined by γ a (t) =
aγ(t). Express the curvature of γ a in terms of that of γ.
(4) Let γ be a regular plane curve, and let λ be a constant. The parallel curve γ λ of γ
is defined by
γ λ (t) = γ(t) + λns (t).
Show that if λκs (t) 6= 1 for all t, then γ λ is a regular curve. What is its signed
curvature?
(5) Let γ be a unit-speed plane curve, and let γ(s0 ) be a point on γ. Write down a
formula for the center of the osculating circle.
(6) Let γ be a plane curve, and for each s, let ε be a curve such that ε(s) is the center
of the osculating circle of γ at γ(s). We call ε the evolute of γ. Suppose that
κ0s (s) > 0 for all s.
(a) Find a formula for the arclength function of ε.
(b) What is the signed curvature of ε?
(c) Let γ(t) = a(t − sin t, 1 − cos t) for 0 < t < 2π be a period of a cycloid.
Find the evolute of γ, and show that it is a translation of γ, after a suitable
reparametrization.
(d) Can you find other curves γ for which their evolutes are translates of γ?
(7) Suppose a string of length ` is attached to a point γ(0) of the unit-speed plane curve
γ. The curve obtained from winding the string onto the curve while keeping it taut
is called the involute of γ and is denoted ι. Find a formula for ι, and for its signed
curvature in the case that the signed curvature of γ is always positive.
(8) Show that the involute of the catenary γ(t) = (t, cosh t) with ` = 0 is the tractrix
x = cosh−1 y1 − 1 − y 2 .
p

(9) Two curves in the plane intersecting at a point p are said to have 0th order contact.
They are said to have 1st order contact if they are tangent. They are said to have 2nd
order contact if they are tangent and have equal signed curvatures. For n ≥ 3, they
are said to have nth order contact if they are tangent and the first n − 2 derivatives
of their curvatures at p are equal. Consider the parabola γ(t) = (t, t2 ) and its
osculating circle γ e at (0, 0). What is their contact order?
(10) Can you relate having 3rd order contact with a circle to other concepts we have
seen?
2https://www.ams.org/notices/200702/fea-gluck.pdf
28 2. CURVATURE, TORSION, AND THE FRENET–SERRET FRAME

(11) Find a curve γ such that there is no circle with 4th order contact.
(12) For each positive integer n, can you find a curve and a circle that have contact order
exactly n?
(13) Describe all space curves with constant curvature κ > 0 and constant torsion τ .
(14) Let γ be a unit-speed curve with κ(s) > 0 and τ (s) 6= 0 for all t. Show that if γ
lies on the surface of a sphere, then
τ d κ0
= .
κ ds τ κ2
Conversely, show that if the above equation holds, then
 2
1 1 d 1
+ = r2
κ2 τ ds κ
for some positive constant r.
(15) Prove Theorem 3.3.
(16) Show that the curve given by r = −1 − 2 sin θ has only two vertices. Why is this
not a counterexample to the four vertex theorem?
(17) Show that a plane curve γ has a vertex at γ(t0 ) if and only if the evolute ε of γ
has a singular point at t0 .
 CHAPTER 3

Parametrized Surfaces

1. Parametrized Surfaces
We now move on from curves to surfaces, which are more exciting. Informally, a surface
is something that locally looks like an open subset of R2 . To make this more formal, we need
to talk a bit about continuity. Informally, a function f : X → Y is continuous if nearby
points in X get mapped to nearby points in Y . The formal definition is a little different,
but it turns out to capture the same notion.
Definition 1.1. Let X ⊆ Rm and Y ⊆ Rn . A function f : X → Y is said to be
continuous if, for every open subset V ⊆ Y , its preimage f −1 (Y ) is an open subset of X.
Definition 1.2. Let X ⊆ Rm and Y ⊆ Rn . A function f : X → Y is said to be a
homeomorphism if f is a continuous bijection whose inverse is also continuous. When this
happens, we say that X and Y are homeomorphic.
Example. A sphere is homeomorphic to the surface of a pear. It might not be convenient
to write down a formula for the surface of a pear or a homeomorphism to a sphere, but
geometrically it’s clear that there must be one, since we can deform a rubber pear to a
sphere continuously.
Remark 1.3. A common misconception about a homeomorphism is that in order to
have a homeomorphism from X to Y , it must be possible to deform X continuously to get
to Y . If we have a deformation, then there is a homeomorphism, but such a deformation
need not exist. You are likely familiar with a Möbius strip, which is obtained by giving a
rectangle a half-twist before gluing opposite edges. The Möbius strip is not homeomorphic
to a cylinder (something that requires proof), but if we give the rectangle a full twist before
gluing opposite edges, then the resulting figure is homeomorphic to a cylinder. However,
there is no deformation in R3 from this figure to a cylinder.
Definition 1.4. A subset S ⊆ R3 is a surface if, for every point p ∈ S, there is an open
subset U ⊆ R2 and an open subset W ⊆ R3 containing p such that S ∩ W is homeomorphic
to U . A homeomorphism σ : U → S ∩W as above is called a surface patch or parametrization
of S ∩ W . A collection of surface patches whose images cover all of S is called an atlas of S.
Example. If U is an open subset of a plane in R3 , then U is a surface, because it can
be covered by a single surface patch.
Example. Let S = {(x, y, z) ∈ R3 : x2 + y 2 = 1}. The set S is a cylinder. It is a surface
thanks to the following two surface patches. The first is U1 = {(u, v) ∈ R2 : 0 < u < 2π}
and f : U1 → S given by f (u, v) = (cos u, sin u, v). The second is U2 = {(u, v) ∈ R2 : −π <
u < π} and g : U2 → S, also given by g(u, v) = (cos u, sin u, v). Between these two patches,
the entire cylinder is covered, because the first patch covers the cylinder except for one line,
and the second patch covers the cylinder except for a different (non-intersecting) line.
29
30 3. PARAMETRIZED SURFACES

Example. Let S = {(x, y, z) ∈ R3 : x2 + y 2 + z 2 = 1}. The set S is a sphere. We can

cover it with two patches, using spherical coordinates. The first is U1 = {(θ, φ) ∈ R2 : − π2 <
θ < π2 , 0 < φ < 2π} and f : U1 → S given by f (θ, φ) = (cos θ cos φ, cos θ sin φ, sin θ). This
covers the sphere except for one arc from the north pole to the south pole. We can cover the
rest of the sphere with a second surface patch on the same set U2 = U1 , but with a different
function g : U2 → S given by g(θ, φ) = (− cos θ cos φ, − sin θ, − cos θ sin φ).
In the future, we’ll assume that all the functions used to define surface patches are
smooth, i.e. have continuous partial derivatives of all orders. We also want to consider only
surface patches that are regular and allowable. We’ll assume that all surfaces are smooth
and that patches are regular, allowable, and so forth without mentioning it again.
Definition 1.5. A surface patch σ : U → R3 is said to be regular if it is smooth and
the vectors σ u (u, v) and σ v (u, v) are linearly independent for all (u, v) ∈ U . Equivalently,
σ is regular if it is smooth and the cross product σ u (u, v) × σ v (u, v) is nonzero at all points
(u, v) ∈ U .
Here, σ u and σ v denote the partial derivatives with respect to u and v, respectively.
We’ll use this notation throughout the rest of the class.
Definition 1.6. Suppose S is a surface. A surface patch σ : U → R3 for S is said to be
allowable if σ is a homeomorphism from U to an open subset of S. We say that S is smooth
if, for any p ∈ S, there is an allowable surface patch σ such that p ∈ σ(U ). A collection A
of allowable surface patches for S such that every point in S is in the image of at least one
patch in A is called an atlas for S.

2. Smooth Maps
It will be important for us to consider not just surfaces themselves, but also functions (or
maps) between surfaces. Since the notion of a smooth map will be local, i.e. depends only
on a neighborhood, we may freely assume that our surfaces are defined via single surface
patches, provided that everything we do is independent of the choice of parametrization.
So, let’s suppose we have two surface patches σ 1 : U1 → R3 and σ 2 : U2 → R3 , covering
surfaces S1 and S2 , respectively. If f : S1 → S2 is a function, then we get a function
σ −1
2 ◦ f ◦ σ 1 : U1 → U2 , as in the following diagram:

σ −1
2 ◦f ◦σ 1
U1 / U2 .
σ1 σ2
 
S1 / S2
f

Since σ −1 2
2 ◦ f ◦ σ 1 is a map between open subsets of R , we know what it means for this map
to be smooth.
Definition 2.1. With notation as above, we say that f is smooth if σ −1
2 ◦f ◦σ 1 : U1 → U2
is smooth.
One can check that the notion of smoothness does not depend on the choice of parametriza-
tion of the surfaces S1 and S2 , as you will show in problem 5.
The nicest smooth maps are the ones that are bijective with smooth inverses.
 3. THE TANGENT PLANE 31

Definition 2.2. A smooth map f : S1 → S2 is said to be a diffeomorphism if it is

bijective with a smooth inverse. We say that S1 and S2 are diffeomorphic if there exists a
diffeomorphism from S1 to S2 .
Sometimes we get maps that are not diffeomorphisms, but that are locally diffeomor-
phisms.
Definition 2.3. A smooth map f : S1 → S2 between surfaces is said to be a local
diffeomorphism if, for every point p ∈ S1 , there is an open subset O of S1 such that f (O) is
an open subset of S2 and f |O : O → f (O) is a diffeomorphism.
Example. We have a local diffeomorphism from the xz-plane P to the cylinder C =
{(x, y, z) ∈ R3 : x2 + y 2 = 1}, given by f (x, 0, z) = (cos x, sin x, z). This is not a global
diffeomorphism because it is not injective.
3. The Tangent Plane
Definition 3.1. Let S be a surface, and let p ∈ S be a point. A tangent vector to S at
p is the tangent vector at p of a curve in S passing through p. The tangent space Tp S of S
at p is the set of all tangent vectors to S at p.
At first, this definition is rather intimidating, because there are many curves on S passing
through p, so how are we to understand all the tangent vectors of curves on S passing
through p? But fortunately, there is a simpler way to understand the tangent space, in
terms of surface patches. Let σ : U → R3 be a surface patch of S such that p is in the image
of σ, say with σ(u0 , v0 ) = p. Suppose γ is a curve in S passing through p, say when t = t0 .
Thus there are functions u(t) and v(t) such that
γ(t) = σ(u(t), v(t))
for all t in a neighborhood of t0 , and u(t0 ) = u0 and v(t0 ) = v0 . The functions u and v are
necessarily smooth, as you will show in problem 14. Conversely, if u and v are smooth, then
γ(t) = σ(u(t), v(t)) is a curve in S.
Proposition 3.2. Let σ : U → R3 be a surface patch of S containing a point p ∈ S,
and let (u, v) be coordinates of U . The tangent space to S at p is the vector subspace of
R3 spanned by the vectors σ u (u0 , v0 ) and σ v (u0 , v0 ), where σ u and σ v denote the partial
derivatives with respect to u and v, respectively.
Proof. Let γ be a smooth curve in S, say with
γ(t) = σ(u(t), v(t)).
Letting a dot denote a derivative with respect to t, we have
γ̇(t0 ) = σ u (u0 , v0 )u0 (t0 ) + σ v (u0 , v0 )v 0 (t0 ).
Thus γ̇(t0 ) is a linear combination of σ u (u0 , v0 ) and σ v (u0 , v0 ). This shows that the tangent
space Tp S is contained in the subspace spanned by σ u (u0 , v0 ) and σ v (u0 , v0 ).
Now, suppose we have a vector in the subspace spanned by σ u and σ v , say λσ u + µσ v
for some λ, µ ∈ R. Let
γ(t) = σ(u0 + λt, v0 + µt).
Then γ is a smooth curve in S, and at t = 0, i.e. the point p ∈ S, we have
γ̇ = λσ u + µσ v .
32 3. PARAMETRIZED SURFACES

This shows that every vector in the span of σ u and σ v is the tangent vector at p of some
curve in S. 
Since σ is assumed to be regular, σ u and σ v are linearly independent, so the tangent
space Tp S is 2-dimensional. Thus we’ll call it the tangent plane from now on.
Using the tangent plane, we can define the derivative of a smooth map between surfaces.
Let f : S → Se be a smooth map between surfaces. The derivative of f is supposed to
measure how f (p) ∈ Se changes when p moves a little bit on S. If p and q are nearby points
on S, then the line through p and q should be nearly tangent to S at p, so the derivative
of f at p should send tangent vectors to S at p to tangent vectors to Se at f (p). In other
words, the derivative should be a map Dp f : Tp S → Tf (p) S.
e
To define this map, let w ∈ Tp S. Then there exists a curve γ in S through p with
tangent vector w; say w = γ̇(t0 ). Then γe = f ◦ γ is a curve in Se with γ
e (t0 ) = f (p). Thus
w
e =γ ˙
e (t0 ) ∈ Tf (p) S.
e

Definition 3.3. With notation as above, the derivative Dp f of f at p ∈ S is the map

Dp f : Tp S → Tf (p) Se such that Dp f (w) = w
e for all w ∈ Tp S.
Of course, there’s a lot we have to do, making sure that the derivative is independent of
the choice of γ and all that. We’ll omit these details, but we’ll at least state some of the key
facts about the derivative, leaving their proofs for the problems.
Proposition 3.4. If f : S → Se is a smooth map between surfaces and p ∈ S, then
Dp f : Tp S → Tf (p) Se is a linear map.
You will prove Proposition 3.4 in problem 6.
Proposition 3.5.
(1) If S is a surface and p ∈ S, the derivative at p of the identity map S → S is the
identity map Tp S → Tp S.
(2) If S1 , S2 , S3 are surfaces and f1 : S1 → S2 and f2 : S2 → S3 are smooth maps, then
for all p ∈ S1 , we have
Dp (f2 ◦ f1 ) = Df1 (p) f2 ◦ Dp f1 .
(3) If f : S1 → S2 is a diffeomorphism, then for all p ∈ S1 , the linear map Dp f :
Tp S1 → Tf (p) S2 is invertible.
You will prove Proposition 3.5 in problem 7.
Proposition 3.6. Let S and Se be surfaces, and let f : S → Se be a smooth map. Then
f is a local diffeomorphism if and only if, for all p ∈ S, the linear map Dp f : Tp S → Tf (p) Se
is invertible.
You will prove Proposition 3.6 in problem 8.

4. The Normal Vector

A plane through the origin in R3 can be described by a vector normal to it. By scaling
if necessary, we may assume that that normal vector is a unit vector. There are two choices
for the unit normal vector, which are negatives of each other.
 5. LEVEL SURFACES 33

>

<

Figure 1. Glue the top and bottom edges in the marked directions to obtain
a Möbius strip.

Let σ : U → R3 be a surface patch containing p. Then we can make an unambiguous

choice of the unit normal vector, namely
σu × σv
Nσ = .
kσ u × σ v k
We call the vector Nσ the standard unit normal of σ at p. Note that Nσ is not independent
e = {(v, u) ∈ R2 : (u, v) ∈ U } and
of the choice of surface patch σ. To see this, let us define U
σ
e :Ue → R3 by σ e (u, v) = σ(v, u). Then Nσe = −Nσ , because σ e u ×σ
e v = σ v ×σ u = −σ u ×σ v .
Definition 4.1. A surface S is said to be orientable if there exists an atlas admitting a
continuous family of unit normal vectors on S. Otherwise, S is said to be nonorientable.
The most famous example of a nonorientable surface is the Möbius strip, which is obtained
by starting with a rectangle, then gluing one pair of opposite sides with a twist, as shown
in Figure 1. If you had a typical childhood, then you know what happens when you cut a
Möbius strip down the middle, and perhaps what happens if you then make more cuts in it.
We can represent a Möbius strip by means of surface patches as follows. Let
   
2 1 1 2 1 1
U = (t, θ) ∈ R : − < t < , 0 < θ < 2π , U = (t, θ) ∈ R : − < t < , −π < θ < π .
e
2 2 2 2
Let σ : U → R3 be defined by
    
θ θ θ
σ(t, θ) = 1 − t sin cos θ, 1 − t sin sin θ, t cos ,
2 2 2
and define σ e → R3 by the same formula. Then σ and σ
e : U e form surface patches for a
Möbius strip. We can compute the unit normals: we have
 
θ θ θ
σ t = − sin , cos θ, − sin sin θ, cos , σ θ = (− sin θ, cos θ, 0),
2 2 2
so  
θ θ θ
σ t × σ θ = − cos θ cos , − sin θ cos , − sin .
2 2 2
This is already a unit vector, so it is the standard unit normal to the surface patch. However,
as we move from θ = 0 to θ = 2π, we replace this normal vector by its negative. Thus we
do not have a continuous family of unit normal vectors to the Möbius strip.

5. Level Surfaces
Often, surfaces are given to us not in terms of surface patches, but as level surfaces.
That is, we have a smooth function f of three variables x, y, z, and the surface is defined as
{(x, y, z) ∈ R3 : f (x, y, z) = 0}. Note that, even if f appears to be a perfectly well-behaved
function, this doesn’t necessarily give us a surface.
34 3. PARAMETRIZED SURFACES

Figure 2. The cone x2 + y 2 − z 2 = 0.

Example. Let f (x, y, z) = x2 + y 2 − z 2 . Then the level surface f (x, y, z) = 0 is a cone,

as shown in Figure 2. This cone is not a surface, because there is no neighborhood of the
origin that is homeomorphic to an open subset of R2 : locally, the cone does not look like a
disk, but rather like two disks glued at their center points.
It will be helpful to have a criterion under which a level surface is a smooth surface.
Theorem 5.1. Let S ⊆ R3 be such that for each p ∈ S, there is an open subset W ⊆ R3
containing p and a smooth function f : W → R such that
• S ∩ W = {(x, y, z) ∈ W : f (x, y, z) = 0}.
• The gradient ∇f = (fx , fy , fz ) does not vanish at p.
Then S is a smooth surface.
In the case of the cone, the second condition is not satisfied, because ∇f = (2x, 2y, −2z),
which is 0 at (x, y, z) = (0, 0, 0). However, we do learn that the cone with the origin removed
is a smooth surface. We can also easily see that other level surfaces are surfaces.
Example. Let S = S2 be the unit sphere, i.e. {(x, y, z) ∈ R3 : x2 + y 2 + z 2 = 1}. Then
we have f (x, y, z) = x2 + y 2 + z 2 − 1, and ∇f = (2x, 2y, 2z). But ∇f is never the 0 vector
on S.
We defer the proof of Theorem 5.1 to §9, once we have introduced the Inverse Function
Theorem.

6. Quadric Surfaces
Some of the simplest level surfaces are the quadric surfaces.
Definition 6.1. A quadric surface is the subset of R3 defined by an equation of the
form
v> Av + b> v + c = 0,
where A is a constant symmetric 3 × 3 matrix, b ∈ R3 is a constant vector, and c ∈ R is a
constant scalar.
 6. QUADRIC SURFACES 35

x y

Figure 3. An ellipsoid.

x y

Figure 4. A hyperboloid of one sheet.

More explicitly, let

   
a1 a4 a6 b1
A =  a4 a2 a5  , b = b2  .

a6 a5 a3 b3
Then the equation of the quadric becomes
a1 x2 + a2 y 2 + a3 z 2 + 2a4 xy + 2a5 yz + 2a6 xz + b1 x + b2 y + b3 z + c = 0.
As we have seen in the case of a cone, quadrics are not necessarily surfaces. There are
also other simple examples of quadrics that are not surfaces, such as x2 + y 2 + z 2 = 0 (a
point), x2 + y 2 = 0 (a line), and xy = 0 (a union of two intersecting planes). But most
quadrics are surfaces. Here are the basic examples:
Example. An ellipsoid has an equation of the form
x2 y 2 z 2
+ 2 + 2 = 1.
a2 b c
An example is shown in Figure 3.

Example. A hyperboloid of one sheet has an equation of the form

x2 y 2 z 2
+ 2 − 2 = 1.
a2 b c
See Figure 4.
36 3. PARAMETRIZED SURFACES

x y

Figure 5. A hyperboloid of two sheets.

x y

Figure 6. An elliptic paraboloid.

Example. A hyperboloid of two sheets has an equation of the form

x2 y 2 z 2
− 2 − 2 + 2 = 1.
a b c
See Figure 5.

Example. An elliptic paraboloid has an equation of the form

x2 y 2
+ 2 = z.
a2 b
See Figure 6.

Example. A hyperbolic paraboloid has an equation of the form

x2 y 2
− 2 = z.
a2 b
See Figure 7.

Example. An elliptic cylinder has an equation of the form

x2 y 2
+ 2 = 1.
a2 b
Example. A hyperbolic cylinder has an equation of the form
x2 y 2
− 2 = 1.
a2 b
 7. RULED SURFACES 37

y
x

Figure 7. A hyperbolic paraboloid.

Example. A parabolic cylinder has an equation of the form

x2
= y.
a2
Example. A plane has an equation of the form ax + by + cz + d = 0.
Example. We can get two parallel planes, by an equation such as x2 = a2 .
These are all the smooth surfaces that can arise as quadric surfaces. Up to an isometry
of R3 , all smooth quadric surfaces are one of these types. They can all be represented
parametrically in terms of surfaces patches, with the help of the trigonometric and hyperbolic
trigonometric functions. For instance, the hyperboloid of two sheets z 2 − x2 − y 2 = 1 can be
parametrized as (x, y, z) = (cos u sinh v, sin u sinh v, cosh v), and the others are similar.

7. Ruled Surfaces
A ruled surface is a surface that is a union of straight lines. We call the lines the rulings
of the surface.
Suppose S is a ruled surface, and γ is a curve in R3 intersecting each of the rulings. Let
p be a point on S. Then p lies on some line, and this line must intersect γ, say at a point
q ∈ S. Suppose that γ(u) = q, and let δ be a nonzero vector in the direction of the line
passing through γ(u). Then there is some v ∈ R such that
p = γ(u) + vδ(u).
Let us write σ(u, v) = γ(u) + vδ(u). Then σ is a smooth map on some subset of R2 . Letting
d
a dot denote du , we have
σ u = γ̇ + v δ̇, σ v = δ.
Thus σ is regular if γ̇ + v δ̇ and δ are linearly independent. This holds in particular if γ̇ and
δ are linearly independent and v is sufficiently small. It follows that, to get a surface in this
way, γ must not be tangent to the rulings.
A special case occurs when all the rulings are parallel to each other. The resulting ruled
surface is called a generalized cylinder in this case. The parametrization takes the form
σ(u, v) = γ(u) + va
for some constant vector a.
38 3. PARAMETRIZED SURFACES

Another special case occurs when all the rulings pass through a fixed point, say p. Then
S is called a generalized cone with vertex p. When this happens, we have
σ(u, v) = (1 − v)γ(u) + vp.
This isn’t a surface, because it isn’t locally Euclidean at p = σ(0, 1). However, it becomes
a surface when we delete this cone point.
One other curious example of a ruled surface is a hyperboloid of one sheet, which is
actually doubly ruled, i.e. it has two families of lines on it, such that every point on the
surface lies on one line from each ruling. If we have a hyperboloid of one sheet defined by
x2 y 2 z 2
+ 2 − 2 = 1,
a2 b c
then the lines    
a cos θ −a sin θ
 b sin θ  + t ·  b cos θ 
0 ±c
for 0 ≤ θ < 2π are all contained in the surface, and this gives the double ruling. In a similar
way, hyperbolic paraboloids are also doubly ruled, as you will show in problem 12.

8. Surfaces of Revolution
Another important family of surfaces is the family of surfaces of revolution. Here, we take
a plane curve, called the profile curve, and rotate it around a line in the plane, thus forming
a surface in R3 . Thus in each plane perpendicular to the line of rotation, the intersection
of the surface with the plane is a circle, and these circles are called parallels. The image of
the profile curve under a fixed angle of rotation is called a meridian. This agrees with the
geographical terminology, if we think of a sphere as being a semicircle joining the two poles
rotated around the line containing the two poles.
Let’s take the z-axis to be the axis of rotation, and assume the profile curve lies in the
xz-plane. If γ(u) = (f (u), 0, g(u)) is a parametrization of the profile curve, then σ(u, v) =
(f (u) cos v, f (u) sin v, g(u)) is a parametrization of the corresponding surface of revolution.
Let us now determine the conditions under which a surface of revolution is regular. To
do this, we must compute the normal vector. Letting a prime denote the derivative with
respect to u, we have
σ u = (f 0 (u) cos v, f 0 (u) sin v, g 0 (u)), σ v = (−f (u) sin v, f (u) cos v, 0).
Thus we have
σ u × σ v = (f (u)g 0 (u) cos v, −f (u)g 0 (u) sin v, f (u)f 0 (u)),
so
kσ u × σ v k2 = f (u)2 (f 0 (u)2 + g 0 (u)2 ).
Thus σ is regular if f is never zero, i.e. γ does not intersect the z-axis, and f 0 and g 0 are
never simultaneously zero. If we do not allow f (u) to be zero, as is required for regularity,
we might as well assume that f (u) > 0 for all u. The parametrization σ is injective if γ
does not have any self-intersections.
 9. THE INVERSE AND IMPLICIT FUNCTION THEOREMS 39

9. The Inverse and Implicit Function Theorems

Now is as good a time as any to discuss two important theorems from multivariable
calculus that you likely missed if you took such a class in school. These two theorems are
intuitively fairly clear, but they are necessary to prove many results about surfaces and more
generally, higher-dimensional manifolds.
Let f : U → Rn be a smooth map, where U is an open subset of Rm . Let us write the
function in coordinates as
f (u1 , . . . , um ) = (v1 , . . . , vn ),
where v1 , . . . , vn are functions of u1 , . . . , um . Define the Jacobian matrix J(f ) of f to be
 ∂v1 ∂v1 ∂v1 
∂u1 ∂u2
· · · ∂u m
 ∂v2 ∂v2 · · · ∂v2 
 ∂u1 ∂u2 ∂um 
J(f ) =  . .. . . ..  .
 .. . . . 
∂vn ∂vn ∂vn
∂u1 ∂u2
· · · ∂um
Note that some people refer to the determinant of this matrix (or even the absolute value of
the determinant) as the Jacobian, but for us, the Jacobian will be the matrix itself.
Theorem 9.1 (Inverse Function Theorem). Let f : U → Rn be a smooth map defined
on an open subset U ⊆ Rn . (Note that the two n’s are equal.) Suppose that, at some point
x0 ∈ U , the Jacobian matrix J(f ) at x0 is invertible. Then there is an open subset V ⊆ Rn
and a smooth map g : V → Rn such that
• y0 = f (x0 ) ∈ V .
• g(y0 ) = x0 .
• g(V ) ⊆ U .
• g(V ) is an open subset of Rn .
• f (g(y)) = y for all y ∈ V .
Thus the functions g : V → g(V ) and f : g(V ) → V are inverse bijections.
In short, if the Jacobian is invertible, then f is locally invertible.
Theorem 9.2 (Implicit Function Theorem). Let W be an open subset of Rn+m , and let
f : W → Rm be a continuously differentiable function, where we treat Rn+m and Rn ×Rm and
write elements of Rn+m as (x, y), where x ∈ Rn and y ∈ Rm . Write J(f ) at (x0 , y0 ) ∈ W
as [X | Y ], where X is an m × n matrix and Y is an m × m matrix. If Y is invertible, then
there exists an open set U ⊆ Rn containing x0 , an open set V ⊆ Rm containing y0 , and a
function g : U → V such that
{(x, g(x)) | x ∈ U } = {(x, y) ∈ U × V : f (x, y) = 0}.
What this means it that, under the condition on the Jacobian, the level set defined by
f (x, y) = 0 is locally the graph of some function of some function g.
We will not give a proof of the Inverse and Implicit Function Theorems; see for instance
§5.4 of Pugh’s Real Mathematical Analysis for proofs.
We now return to Theorem 5.1, which we will prove as a consequence of the Inverse
Function Theorem. Recall the statement of the theorem:
Theorem 9.3. Let S ⊆ R3 be such that for each p ∈ S, there is an open subset W ⊆ R3
containing p and a smooth function f : W → R such that
40 3. PARAMETRIZED SURFACES

• S ∩ W = {(x, y, z) ∈ W : f (x, y, z) = 0}.

• The gradient ∇f = (fx , fy , fz ) does not vanish at p.
Then S is a smooth surface.
Proof. Let p = (x0 , y0 , z0 ). Since ∇f is nonzero at p, at least one of fx , fy , fz is nonzero
at p. We’ll assume that fz 6= 0; the other cases are similar. Consider the map F : W → R3
defined by
F (x, y, z) = (x, y, f (x, y, z)).
Its Jacobian is  
1 0 0
J(F ) =  0 1 0  ,
fx fy fz
which is clearly invertible at p because fz 6= 0. By the Inverse Function Theorem, there is
an open subset V ⊆ R3 containing F (x0 , y0 , z0 ) = (x0 , y0 , 0) and a smooth map G : V → W
such that Wf = G(V ) is open, and F : W f → V and G : V → W f are inverse bijections.
2
Since V is open, there are open subsets U1 ⊆ R containing (x0 , y0 ) and U2 ⊆ R containing
0 such that U1 ×U2 ⊆ V . Thus, by shrinking V if necessary, we may assume that V = U1 ×U2 .
Since F and G are inverse bijections, we have
G(x, y, w) = (x, y, g(x, y, w))
for some smooth map g : U1 × U2 → R, and
f (x, y, g(x, y, w)) = w
for all (x, y) ∈ U1 and w ∈ U2 .
Define σ : U1 → R3 by
σ(x, y) = (x, y, g(x, y, 0)).
Then σ is a homeomorphism from U1 to S ∩ W
f . Then σ is smooth, and it is regular because

σ x × σ y = (−gx , −gy , 1),

which is never 0. Thus σ is a regular surface patch on S containing p. Since p was arbitrary,
we have constructed an atlas for S making it into a smooth surface. 

10. Problems
(1) Show that the cylinder {(x, y, z) ∈ R3 : x2 + y 2 = 1} can be covered by a single
surface patch. Show that the sphere cannot be covered by a single surface patch.
(2) A torus is obtained by rotating a circle C in a plane Π around a line ` in Π that
does not intersect C. Let Π be the xz-plane, ` the z-axis, a > 0 the distance from
the center of C to `, and b < a the radius of C. Find surface patches covering the
torus.
(3) The surface
σ(u, v) = (v cos u, v sin u, λu),
where λ is a constant, is called a helicoid. Show that the cotangent of the angle the
standard unit normal of σ at a point p makes with the z-axis is proportional to the
distance from p to the z-axis.
 10. PROBLEMS 41

(4) Let γ be a unit-speed curve in R3 with nowhere vanishing curvature, and let a > 0.
Consider the surface described by
σ(s, θ) = γ(s) + a(n(s) cos θ + b(s) sin θ).
Explain what this surface is geometrically. Prove that σ is regular if the curvature
κ of γ is less than a1 everywhere. (Note that, even if σ is regular, the surface might
have self-intersections.)
(5) Suppose that f : S1 → S2 is a map between surfaces. Show that the smoothness of f
does not depend on the choice of parametrization, i.e. if it is smooth with respect to
some parametrization, then it is smooth with respect to any other parametrization.
(6) Prove Proposition 3.4.
(7) Prove Proposition 3.5.
(8) Prove Proposition 3.6.
(9) Suppose S is the level surface defined by f (x, y, z) = 0, where f is a smooth function
such that ∇f does not vanish at any point of S. Show that S is orientable.
(10) Show that if a quadric surface contains three points on a straight line, then it
contains the entire line. Conclude that if L1 , L2 , L3 are three nonintersecting lines
in R3 , there is a quadric containing all three lines.
(11) Show that any doubly ruled surface is part of a quadric surface.
(12) Show that hyperbolic paraboloids are doubly ruled.
(13) Show that
σ(u, v) = (sech u cos v, sech u sin v, tanh u)
is a regular surface patch for the unit sphere S2 . It is called the Mercator projection,
and it’s the most popular way of drawing maps in the plane. Show the meridians
and parallels on S2 correspond under σ to perpendicular lines in the plane.
(14) Show that if γ is a (smooth) curve whose image is contained in a surface patch
σ : U → R3 , then γ(t) = σ(u(t), v(t)) for some smooth map t 7→ (u(t), v(t)). (You
will need to use the Inverse Function Theorem.)
 CHAPTER 4

The First Fundamental Form

1. Lengths of Curves on Surfaces

Suppose we have a surface S parametrized by some surface patch σ, and we have a curve
γ : [a, b] → S on S. Then its arclength is given by
Z b
kγ̇(t)k dt.
a
Now, at a point γ(t) of γ, we can write γ̇(t) as a linear combination of σ u and σ v , say as
γ̇(t) = λσ u + µσ v ,
where λ and µ are scalars. Then
γ̇(t) · γ̇(t) = (λσ u + µσ v ) · (λσ u + µσ v ) = λ2 kσ u k2 + 2λµσ u · σ v + µ2 kσ v k2 .
Letting
E = kσ u k2 , F = σu · σv , G = kσ v k2 ,
we have
kγ̇(t)k2 = Eλ2 + 2F λµ + Gµ2 .
The three values E, F, G thus completely determine the length of curves on S.
More formally, we define linear maps du, dv : Tp S → R by setting
du(v) = λ, dv(v) = µ if v = λσ u + µσ v .
We call the expression
E du2 + 2F du dv + G dv 2
the first fundamental form of the surface patch σ. Note that E, F, G can vary from point to
point, so they are actually functions from σ to R.
In terms of E, F, G, the length of a curve γ : [a, b] → S on S is given by
Z b√
Eu02 + 2F u0 v 0 + Gv 02 dt.
a

Remark 1.1. An expression involving du2 , du dv, or dv 2 is called a quadratic differential.

For us, it won’t matter exactly where quadratic differentials live, i.e. what sorts of objects
they are, but if you like this sort of thing, they live in the symmetric square of the dual space
of the tangent plane.
Let’s calculate the first fundamental forms for certain surfaces.
Example. Consider the plane, given by the surface patch σ(u, v) = a + up + vq, where
a, p, q are constant vectors, and p and q are perpendicular unit vectors. Then we have
σ u = p, σ v = q,
43
44 4. THE FIRST FUNDAMENTAL FORM

so
E = kσ u k2 = 1, F = σ u · σ v = 0, G = kσ v k2 = 1.
Thus the first fundamental form is du2 + dv 2 .
Example. Consider a surface of revolution
σ(u, v) = (f (u) cos v, f (u) sin v, g(u)).
We will assume that f (u) > 0 for all u and that the profile curve u 7→ (f (u), 0, g(u)) is a
unit-speed curve, i.e. f 0 (u)2 + g 0 (u)2 = 1. Then
σ u = (f 0 (u) cos v, f 0 (u) sin v, g 0 (u)), σ v = (−f (u) sin v, f (u) cos v, 0).
Thus we have
E = kσ u k2 = f 0 (u)2 + g 0 (u)2 = 1,
F = σ u · σ v = 0,
G = kσ v k2 = f 2 .
Thus the first fundamental form is du2 + f (u)2 dv 2 . One special case is the sphere, given in
spherical coordinates by σ(θ, φ) = (cos θ cos φ, cos θ sin φ, sin θ), so that the first fundamental
form is dθ2 + cos2 θ dφ2 .
Example. Consider a generalized cylinder
σ(u, v) = γ(u) + va.
We may assume that γ is unit speed, a is a unit vector, and γ is contained in a plane
perpendicular to a. Then, letting a dot denote a derivative with respect to u, we have
σ u = γ̇, σ v = a,
so
E = kσ u k2 = kγ̇k2 = 1, F = σ u · σ v = γ̇ · a = 0, G = kσ v k2 = kak2 = 1,
so the first fundamental form is du2 + dv 2 .

2. Isometries of Surfaces
Note that the plane and the generalized cylinder have the same first fundamental forms.
This shouldn’t be too surprising, because if we wrap a plane around a generalized cylinder,
then we don’t change the lengths of curves drawn on it.
Definition 2.1. Let S1 and S2 be surfaces. A smooth map f : S1 → S2 is said to be a
local isometry if for any curve γ on S1 , the lengths of γ on S1 and f ◦ γ on S2 are equal. If
there exists a local isometry f : S1 → S2 , then we say that S1 and S2 are locally isometric.
A bijective local isometry is called an isometry.
In what follows, we’ll be using the notation h·, ·i for the inner product on R3 :
h(x1 , y1 , z1 ), (x2 , y2 , z2 )i = x1 x2 + y1 y2 + z1 z2 .
Generally, we’ll be restricting to the case where v = (x1 , y1 , z1 ) and w = (x2 , y2 , z2 ) are both
in Tp S for some surface S and some p ∈ S. In this case, we’ll write hv, wip .
Suppose f : S1 → S2 is a smooth map, and let p ∈ S1 . For v, w ∈ Tp S1 , we define
f ∗ hv, wip = hDp f (v), Dp f (w)if (p) .
 3. CONFORMAL MAPS 45

Note that this allows us to define a new inner product on Tp S1 out of the inner product on
Tf (p) S2 (in that order!).
Theorem 2.2. A smooth map f : S1 → S2 is a local isometry if and only if the symmetric
bilinear forms h·, ·ip and f ∗ h·, ·if (p) on Tp S1 are equal for all p ∈ S1 .
In other words, the inner product h·, ·ip (or equivalently the first fundamental form, as
we shall see) captures exactly the information of being a local isometry.
Proof. Suppose γ 1 is a curve on S1 . Then the length of the part of γ 1 with endpoints
γ 1 (t0 ) and γ 1 (t1 ) is
Z t1 p
hγ̇ 1 , γ̇ 1 i dt.
t0
The length of the corresponding part of the curve γ 2 = f ◦ γ 1 on S2 is
Z t1 p Z t1 p Z t1 p
hγ̇ 2 , γ̇ 2 i dt = hDf (γ̇ 1 ), Df (γ̇ 1 )i dt = f ∗ hγ̇ 1 , γ̇ 1 i dt.
t0 t0 t0

Thus if the two symmetric bilinear forms h·, ·ip and f ∗ h·, ·if (p) are equal, then the curves
have the same length. Since this is true for an arbitrary curve γ 1 , this implies that f ∗ is a
local isometry.
Now, suppose the two curves always have the same length, so the two integrals are equal
for all curves γ 1 on S. Then, by the fundamental theorem of calculus, the integrands must
be equal, so we have
hγ̇, γ̇i = f ∗ hγ̇, γ̇i
for all γ. Since any tangent vector v on S1 is the tangent vector of a curve on S1 , it follows
that hv, vi = f ∗ hv, vi for all v ∈ Tp S1 . It follows from the general formula
hv + w, v + wi = hv, vi + hw, wi + 2hv, wi
that knowledge of all norms determines the inner product, so hv, wi = f ∗ hv, wi for all
v, w ∈ Tp S1 . 
Corollary 2.3. A local diffeomorphism f : S1 → S2 is a local isometry if and only if,
for every surface patch σ 1 of S1 , the patches σ 1 and f ◦ σ 1 of S1 and S2 , respectively, have
the same first fundamental form.
Proof. It suffices to show that the first fundamental form determines the symmetric
bilinear form on the tangent space, and vice versa. But this is clear because E = hσ u , σ u i,
F = hσ u , σ v i, and G = hσ v , σ v i. 

3. Conformal Maps
The first fundamental form tells us about lengths of curves on surfaces. The next thing
we would like to know is about angles between curves. Give two curves γ and γ e on a surface
S intersecting at a point p ∈ S, we define the angle θ between them to be the angle between
their tangent vectors at p. Using the dot product formula, we have
e˙
γ̇ · γ
cos θ = .
kγ̇kkγ e˙ k
46 4. THE FIRST FUNDAMENTAL FORM

We can express this angle directly in terms of E, F, G from the first fundamental form:
suppose γ(t) = σ(u(t), v(t)) and γ e (t) = σ(ũ(t), ṽ(t)). If the first fundamental form of σ is
2 2
E du + 2F du dv + G dv , then we have
e˙ = Eu0 ũ0 + F (u0 ṽ 0 + ũ0 v 0 ) + Gv 0 ṽ 0 ,
γ̇ · γ
γ̇ · γ̇ = Eu02 + 2F u0 v 0 + Gv 02 ,
e˙ · γ
γ e˙ = E ũ02 + 2F ũ0 ṽ 0 + Gṽ 02 .
Thus we have
Eu0 ũ0 + F (u0 ṽ 0 + ũ0 v 0 ) + Gv 0 ṽ 0
cos θ = √ √ .
Eu02 + 2F u0 v 0 + Gv 02 E ũ02 + 2F ũ0 ṽ 0 + Gṽ 02
Example. Fix parameters u0 and v0 , and consider the parameter curves
γ(t) = σ(u0 , t), γ
e (t) = σ(t, v0 ).
Then we have
u(t) = u0 , v(t) = t, ũ(t) = t, ṽ(t) = v0 .
Thus we have
u0 = 0, v 0 = 1, ũ0 = 1, ṽ 0 = 0.
These parameter curves intersect at the point σ(u0 , v0 ). Their angle of intersection is thus
given by
F
cos θ = √ ,
EG
where E, F , and G are evaluated at (u0 , v0 ). In particular, the curves are orthogonal if and
only if F = 0.
Definition 3.1. Let S1 and S2 be surfaces. A local diffeomorphism f : S1 → S2 is said
to be a conformal map if, whenever γ 1 and γ e 1 are two curves on S1 that intersect at a point
p, and γ 2 and γ e 2 are their images in S2 under f , then the angle of intersection between γ 1
and γe 1 at p is equal to the angle of intersection between γ 2 and γe 2 at f (p).
That is, conformal maps are maps that preserve angles.
Theorem 3.2. A local diffeomorphism f : S1 → S2 is conformal if and only if there is a
function λ : S1 → R such that
f ∗ hv, wip = λ(p)hv, wip
for all p ∈ S1 and v, w ∈ Tp S1 .
Proof. Let γ and γ e be two curves on S1 intersecting at a point p ∈ S1 . The angle θ of
intersection of the two curves is given by
e˙ i
hγ̇, γ
cos θ = .
kγ̇kkγ e˙ k
We obtain the angle of intersection of f ◦ γ and f ◦ γ e is given by the same formula, but with
γ and γe replaced with f ◦ γ and f ◦ γ e . We have
 
d d
(f ◦ γ), (f ◦ γe) = hDp f (γ̇), Dp f (γe˙ )if (p) = f ∗ hγ̇, γ
e˙ ip .
ds ds f (p)

Thus if f ∗ h·, ·i = λh·, ·i, then the λ factors cancel, so the angles are equal, and f is conformal.
 3. CONFORMAL MAPS 47

Now, suppose that

e˙ i
hγ̇, γ e˙ i
f ∗ hγ̇, γ
q = q
e˙ , γ
e˙ i e˙ , γ
e˙ i
p p
hγ̇, γ̇i hγ f ∗ hγ̇, γ̇i f ∗ hγ
for all pairs of intersecting curves γ and γe on S1 . We must show that f ∗ h·, ·i is equal to
λh·, ·i for some scalar function λ. Equivalently, we have
hv, wi f ∗ hv, wi
p p =p p
hv, vi hw, wi f ∗ hv, vi f ∗ hw, wi
for all v, w ∈ Tp S1 .
Let {v1 , v2 } be an orthonormal basis of Tp S1 with respect to the bilinear form h·, ·i given
by its first fundamental form. Let
λ = f ∗ hv1 , v1 i, µ = f ∗ hv1 , v2 i, ν = f ∗ hv2 , v2 i.
Letting v = v1 and w = cos θ v1 + sin θ v2 , we get
λ cos θ + µ sin θ
cos θ = p .
λ(λ cos θ + 2µ sin θ cos θ + ν sin2 θ)
2

π
Letting θ = 2
gives µ = 0, which implies that
λ = λ cos2 θ + ν sin2 θ
for all θ ∈ R. Thus λ = ν. This implies that f ∗ hv, wi = λhv, wi whenever v and w are
basis vectors. Since both sides are bilinear forms, this implies that f ∗ h·, ·i = λh·, ·i for all
vectors. 
Corollary 3.3. A local diffeomorphism f : S1 → S2 is conformal if and only if, for
every surface patch σ of S1 , the first fundamental forms of σ on S1 and f ◦ σ on S2 are
proportional.
Example. Let S2 be the unit sphere in R3 , and let n = (0, 0, 1) be the north pole. We
can stereographically project the sphere other than the north pole to the plane, as follows:
given a point q 6= n on S2 , the line connecting n and q intersects the xy-plane at some point
p. Stereographic projection maps the point q ∈ S2 to the point p on the plane. Let us write
Π for the stereographic projection map from S2 to R2 .
We claim that Π is a conformal map. To see this, we need to write down a formula for
Π. We can check using basic geometry that
 
x y
Π(x, y, z) = , ,
1−z 1−z
and the surface patch of the sphere, which is the inverse of Π, is given by
u2 + v 2 − 1
 
2u 2v
σ(u, v) = , , .
u2 + v 2 + 1 u2 + v 2 + 1 u2 + v 2 + 1
The first fundamental form for the plane is of course du2 + dv 2 , whereas we can calculate
the first fundamental form of σ to be
4
2 2 2
(du2 + dv 2 ).
(u + v + 1)
48 4. THE FIRST FUNDAMENTAL FORM

Thus, letting λ(u, v) = (u2 +v42 +1)2 , we see that the first fundamental forms are proportional,
so the map Π is conformal.
It turns out, but it is not obvious, that given any surface, there is always an atlas
consisting of conformal surface patches. We will not prove this fact.

4. Equiareal Maps
So far, we have discussed lengths and angles. The natural next thing to consider is areas.
Suppose we have a surface patch σ : U → R3 on a surface S. The image of σ is covered by
two families of parameter curves: the ones where u is constant and v varies, and the ones
where v is constant and u varies. Let (u0 , v0 ) ∈ U , and let p = σ(u0 , v0 ). If we change u
by ∆u for sufficiently small ∆u, then we change σ by approximately σ u ∆u. Similarly, if we
change v by ∆v, then we change σ by approximately σ v ∆v. Thus the image of the rectangle
[u0 , u0 + ∆u] × [v0 , v0 × ∆v] in U is roughly a parallelogram in Tp S with sides being given
by the vectors σ u ∆u and σ v ∆v. The area of this parallelogram is
kσ u ∆u × σ v ∆vk = kσ u × σ v k∆u∆v.
In other words, the area is scaled by a factor of kσ u × σ v k.
Definition 4.1. The area Aσ (R) of the part σ(R) of a surface patch σ : U → R3
corresponding to a subset R ⊆ U is
Z
Aσ (R) = kσ u × σ v k du dv.
R

Just like lengths and angles, the area component kσ u × σ v k can be calculated directly
from the first fundamental form:
Proposition 4.2. We have
√
kσ u × σ v k = EG − F 2 .
Proof. We use the following fact about dot and cross products:
(a × b) · (c × d) = (a · c)(b · d) − (a · d)(b · c).
Thus we have
kσ u × σ v k2 = (σ u × σ v ) · (σ u × σ v ) = (σ u · σ u )(σ v · σ v ) − (σ u · σ v )2 = EG − F 2 .


For a regular surface, EG − F 2 > 0, because σ u × σ v is never zero.

Thus we have Z √
Aσ (R) = EG − F 2 du dv.
R
To make sure that this is a good definition, it is essential to know that the area depends
only on the surface itself, rather than the choice of surface patch.
Proposition 4.3. The area of a surface patch is unchanged by reparametrization.
 4. EQUIAREAL MAPS 49

Proof. Let σ : U → R3 be a surface patch and σe:U e → R3 a reparametrization, with

e → U the reparametrization map, so that σ = σ
Φ:U e ◦ Φ. Thus if Φ(ũ, ṽ) = (u, v), then
σ
e (ũ, ṽ) = σ(u, v).
e⊆U
Let R e be a region, and let R = Φ(R) e ⊆ U . We must prove that
Z Z
kσ u × σ v k du dv = ke
σ ũ × σ
e ṽ k dũ dṽ.
R R
e

We have
e ũ × σ
σ e ṽ = det(J(Φ))σ u × σ v .
Thus
Z Z Z
ke
σ ũ × σ
e ṽ k dũ dṽ = | det(J(Φ))|kσ u × σ v k dũ dṽ = kσ u × σ v k du dv,
R
e R
e R
as desired. 
This is good: it means that the area of a region on a surface depends only on the region,
not the choice of parametrization.
Definition 4.4. Let S1 and S2 be two surfaces. A local diffeomorphism f : S1 → S2
is said to be equiareal if it takes any region in S1 to a region of the same area in S2 . (We
assume that the region in S1 is sufficiently small, so that it and its image in S2 are contained
in surface patches in S1 and S2 , respectively.)
Remark 4.5. It is not possible to assign an area to any region in S1 and S2 : so-called
non-measurable sets exist. So, we ought to restrict to sufficiently nice regions in S1 and S2 ;
compact sets will do just fine.
Theorem 4.6. A local diffeomorphism f : S1 → S2 is equiareal if and only if, for any
surface patch σ on S1 , the first fundamental forms
E1 du2 + 2F1 du dv + G1 dv 2 and E2 du2 + 2F2 du dv + G2 dv 2
of the surface patches σ on S1 and f ◦ σ on S2 satisfy
E1 G1 − F12 = E2 G2 − F22 .
The proof is similar to that of the analogous theorems for lengths and angles.
A famous example of an equiareal map was discovered by Archimedes and inscribed on
his tombstone. Consider the unit sphere S2 = {(x, y, z) ∈ R3 : x2 + y 2 + z 2 + 1} and the
cylinder {(x, y, z) ∈ R3 : x2 + y 2 = 1}. There sphere is contained in the cylinder, and they
intersect at the unit circle in the xy-plane. See Figure 1. For each point p ∈ S2 other
than the poles (0, 0, ±1), we can project radially outward from the sphere to the cylinder,
resulting in a point q on the cylinder. Let f be the map from S2 with the two poles removed
to the cylinder that takes p to q.
In terms of coordinates, if we let p = (x, y, z) with (x, y) 6= (0, 0), then we have
!
x y
f (p) = p ,p ,z .
x2 + y 2 x2 + y 2
Theorem 4.7 (Archimedes). The map f as above is equiareal.
50 4. THE FIRST FUNDAMENTAL FORM

Figure 1. A sphere inside a cylinder.

Proof. Let σ 1 be the surface patch on the sphere given by σ 1 (θ, φ) = (cos θ cos φ, cos θ sin φ, sin θ).
The image of σ 1 under f is the surface patch σ 2 on the cylinder given by σ 2 (θ, φ) =
(cos φ, sin φ, sin θ). The first fundamental form of σ 1 is

dθ2 + cos2 θ dφ2 ,

whereas the first fundamental form of σ 2 is

cos2 θ dθ2 + dφ2 .

Thus in both cases, we have EG − F 2 = cos2 θ, so the map f is equiareal. 

As a consequence of Archimedes’s Theorem, we deduce an important result about the

sum of the angles of a spherical triangle.

Theorem 4.8. The area of a spherical triangle on S2 with angles α, β, γ is α + β + γ − π.

In other words, the angle excess (angle sum of a spherical triangle minus angle sum of
a Euclidean triangle) is equal to the area. By definition, a spherical triangle consists of
three arcs of great circles intersecting at three points, since great circles are the spherical
analogues of straight lines. (We will return to this topic in week 7 when we study geodesics.)
It follows, for instance, that if two spherical triangles are similar, then they are already
congruent, because we cannot have different-sized spherical triangles with the same angles.

Proof. We begin by computing the area of a lune, i.e. the region enclosed between
two great circles. By rotating the sphere if necessary, we can assume that the great circles
intersect at the north and south poles. If the angle between them is θ, then the image of
the lune under the equiareal projection to the cylinder is a curved rectangle of height 2 and
width θ. If we apply the isometry that unwraps the cylinder onto the plane, the image of
the lune is a Euclidean rectangle with height 2 and width θ, so it has area 2θ.
Now, let A, B, C be the vertices of a spherical triangle, with angles α, β, γ at A, B, C,
respectively. Consider the great circles obtained by extending the sides of the triangle. These
great circles divide S2 into eight triangles, as shown in Figure 2. Let A0 , B 0 , C 0 be the points
antipodal to A, B, C, respectively. Then observe that the spherical triangles ABC and A0 BC
 5. PROBLEMS 51

B0
A C0
C A0
B

Figure 2. Three great circles divide the sphere into eight triangles.

form a lune with angle α, and similar for other pairs. Thus we have
A(ABC) + A(A0 BC) = 2α,
A(ABC) + A(AB 0 C) = 2β,
A(ABC) + A(ABC 0 ) = 2γ.
Adding these equations, we have
2A(ABC) + (A(ABC) + A(A0 BC) + A(AB 0 C) + A(ABC 0 )) = 2(α + β + γ).
Now, the triangles ABC ,AB 0 C, AB 0 C 0 , and ABC 0 together make a hemisphere, so the sum
of their areas is 2π. Furthermore, the map sending each point on S2 to its antipode is clearly
equiareal, so we have A(A0 BC) = A(AB 0 C 0 ). Thus we have
A(ABC) + A(A0 BC) + A(AB 0 C) + A(AB 0 C) + A(ABC 0 ) = 2π.
Thus we have
2A(ABC) + 2π = 2(α + β + γ),
from which the result follows. 
It’s worth pointing out that this theorem is a very special case of the celebrated Gauß–
Bonnet Theorem, which we will discuss in week 9.

5. Problems
(1) Let σ be a surface patch with first fundamental form E du2 + 2F du dv + G dv 2 .
(a) Show that the following are equivalent:
(i) Ev = Gu = 0.
(ii) σ uv is parallel to N.
(iii) the opposite sides of any quadrilateral on σ formed by parameter curves
of σ have the same length.
When these conditions hold, the parameter curves of σ are said to form a
Chebyshev net.
52 4. THE FIRST FUNDAMENTAL FORM

(b) Show that if one (hence all) of the above conditions hold, then σ has a reparametriza-
tion σ
e (ũ, ṽ) with first fundamental form
dũ2 + 2 cos θ dũ dṽ + dṽ 2 ,
where θ is a smooth function of (ũ, ṽ).
(c) Show that θ is the angle between the parameter curves of σ e.
(d) Let û = ũ + ṽ and v̂ = ũ − ṽ. Show that the resulting reparametrization σ
b (û, v̂)
of σ
e (ũ, ṽ) has first fundamental form
θ 2 θ
cos2
dû + sin2 dv̂ 2 .
2 2
(2) Write down an isometry from the circular half-cone with a line removed
σ(u, v) = (u cos v, u sin v, u), u > 0, 0 < v < 2π
to an open subset of R2 .
(3) Let γ be a curve in R3 , and let v be a point in R3 . The generalized cone on γ
through v is the surface defined by σ(u, v) = (1 + v)γ(u) − vv.
(a) Under what conditions is σ a surface patch?
(b) Show that, when σ is a surface patch of a smooth surface, then it is a ruled
surface, with all rulings passing through the point v.
(c) Show that a generalized cone is locally isometric to the plane.
(4) Let γ be a unit-speed curve in R3 . The tangent developable to γ is the union of all
the tangent lines to γ.
(a) Show that σ(u, v) = γ(u)+v γ̇(u) is a surface patch for the tangent developable
to γ.
(b) Show that in order for σ to be regular, it is necessary that γ̈ is never 0.
(c) Show that we must exclude the curve itself, i.e. the portion corresponding to
v = 0, from the surface, in order for the tangent developable to be a regular
surface. From now on, we will restrict to one of the pieces v > 0 or v < 0.
(d) Show that a tangent developable is locally isometric to a plane.
(5) Let
σ(u, v) = (cosh u cos v, cosh u sin v, u), σ
e (u, v) = (u cos v, u sin v, v)
be surface patches of the catenoid and helicoid, respectively.
(a) Show that the map from the catenoid to the helicoid taking σ(u, v) to σ
e (sinh u, v)
is a local isometry.
(b) Which curves on the helicoid correspond to the parallels and meridians on the
catenoid?
(6) Show that every local isometry is conformal. Give an example of a conformal map
that is not a local isometry.
(7) Let
u3 v3
 
2 2 2 2
σ(u, v) = u − + uv , v − + vu , u − v .
3 3
Show that σ is a conformal map from the plane.
(8) Recall the latitude-longitude parametrization σ(θ, φ) of the sphere, given by
σ(θ, φ) = (cos θ cos φ, cos θ sin φ, sin θ),
 5. PROBLEMS 53

has first fundamental form dθ2 + cos2 θ dφ2 . Find a smooth function ψ such that the
reparametrization σe (u, v) = σ(ψ(u), v) is conformal. Verify that σe is the Mercator
projection.
(9) A spherical polygon on S2 is the region of positive area formed by the intersection
of n ≥ 3 hemispheres of S2 . Show that if α1 , . . . , αn are the interior angles of a
spherical polygon, its area is
Xn
αi − (n − 2)π.
i=1
This is a very special case of one version of the Gauß–Bonnet Theorem.
(10) Suppose that S2 is covered by spherical polygons in such a way that the intersection
of any two polygons, is empty, a common vertex, or a common edge. Suppose that
there are F polygons, E edges, and V vertices, where a shared edge or vertex is
only counted once.
(a) Show that the sum of all the angles of all the polygons is 2πV .
(b) Show that V − E + F = 2, a famous result of Euler.
This exercise gives another very special case of the Gauß–Bonnet Theorem.
(11) Let p and q be distinct points on S2 .
(a) Show that if p 6= −q, then the shorter arc of the great circle through p and q
is the unique curve of shortest length joining p and q.
(b) Show that if p = −q, then any great semicircle through p and q is a curve of
shortest length joining p and q.
(12) Show that if p, q ∈ S2 , and d(p, q) is the spherical distance between p and q, then
we have
cos d(p, q) = p · q.
(13) Suppose that a spherical triangle (where the sides are arcs of great circles) has sides
of length A, B, C and internal angles α, β, γ, with α opposite A and so forth.
(a) Show that
cos C − cos A cos B
cos γ = .
sin A sin B
(b) Show that
sin α sin β sin γ
= = .
sin A sin B sin C
(c) Show that similar triangles on a sphere must be congruent.
(14) A spherical circle with center p ∈ S2 and radius r is the set of points in S2 with
spherical distance r from p. Suppose that 0 < r < π2 .
(a) Show that a spherical circle of radius r is a Euclidean circle of radius sin r.
(b) Show that the spherical area of a circle of radius r is 2π(1 − cos r).
 CHAPTER 5

Curvature of Surfaces

1. The Second Fundamental Form

We now turn to the problem of defining the curvature of a surface. There are several
apparently different ways of doing this, but they all ultimately lead to the same thing: the
second fundamental form.
Suppose that σ is a surface patch of a surface in R3 , with standard unit normal N. As
we move from parameters (u, v) to (u + ∆u, v + ∆v) on σ, the surface moves away from the
plane tangent to σ at σ(u, v) by the amount
(σ(u + ∆u, v + ∆v) − σ(u, v)) · N.
By Taylor’s Theorem, we have
1
σ(u + ∆u, v + ∆v) − σ(u, v) = σ u ∆u + σ v ∆v + (σ uu (∆u)2 + 2σ uv ∆u∆v + σ vv (∆v)2 ) + R,
2
R
where lim(∆u)2 +(∆v)2 →0 (∆u)2 +(∆v)2 = 0. Now, note that σ u and σ v are tangent to the surface,
hence orthogonal to N. Thus we have
1
(σ(u + ∆u, v + ∆v) − σ(u, v)) · N = (σ uu (∆u)2 + 2σ uv ∆u∆v + σ vv (∆v)2 ) · N + R · N.
2
Let
L = σ uu · N, M = σ uv · N, N = σ vv · N.
Then the deviation from the tangent plane is roughly
1
(L(∆u)2 + 2M ∆u∆v + N (∆v)2 ).
2
We call the expression
L du2 + 2M du dv + N dv 2
the second fundamental form of σ. As we will see, the second fundamental form controls
the curvature of σ.
Example. Consider the plane, parametrized by
σ(u, v) = a + up + vq.
Then σ u = p and σ v = q, which are both constant vectors, so all the second derivatives are
0. Thus the second fundamental form is 0.
Example. Consider a surface of revolution
σ(u, v) = (f (u) cos v, f (u) sin v, g(u)),
where we assume that f (u) > 0 for all u and that the profile curve u 7→ (f (u), 0, g(u)) is
unit-speed (i.e. f 0 (u)2 + g 0 (u)2 = 1 for all u). Then we have
σ u = (f 0 (u) cos v, f 0 (u) sin v, g 0 (u)), σ v = (−f (u) sin v, f (u) cos v, 0),
55
56 5. CURVATURE OF SURFACES

so
σ u × σ v = (−f (u)g 0 (u) cos v, −f (u)g 0 (u) sin v, f (u)f 0 (u)) ⇒ kσ u × σ v k = f (u).
Thus
σu × σv
N= = (−g 0 (u) cos v, −g 0 (u) sin v, f 0 (u)).
kσ u × σ v k
Moving on to second derivatives, we have
σ uu = (f 00 (u) cos v, f 00 (u) sin v, g 00 (u)),
σ uv = (−f 0 (u) sin v, f 0 (u) cos v, 0),
σ vv = (−f (u) cos v, −f (u) sin v, 0).
Thus we have
L = σ uu · N = f 0 (u)g 00 (u) − f 00 (u)g 0 (u),
M = σ uv · N = 0,
N = σ vv · N = f (u)g 0 (u).
Thus the second fundamental form is
(f 0 (u)g 00 (u) − f 00 (u)g 0 (u)) du2 + f (u)g 0 (u) dv 2 .
In the case of the sphere, given by
σ(θ, φ) = (cos θ cos φ, cos θ sin φ, sin θ),
the second fundamental form reduces to
dθ2 + cos2 θ dφ2 ,
which is curiously equal to the first fundamental form. We’ll see an explanation for this in
week 6.
In the case of the cylinder, given by
σ(u, v) = (cos v, sin v, u),
we have L = M = 0 and N = 1, so the second fundamental form is dv 2 .
2. The Gauß and Weingarten Maps
The next approach to curvature of surface is through the spread of normal vectors over a
small region of a surface S. For a surface with small curvature, the normals should not change
very much as we move a bit on our surfaces, whereas for a surface with large curvature, the
normals should change a lot. Since the normal vectors are unit vectors, we can also consider
them as points on the unit sphere S2 .
Definition 2.1. Let S be a surface. The Gauß map is the map G : S → S2 sending a
point p ∈ S to its unit normal vector N. If the choice of surface is ambiguous, we write GS .
The rate at which N varies as we move along S is given by the derivative of the Gauß
map, which is the map
Dp G : Tp S → TG(p) S2 .
Note, however, that the tangent space to S at p is the same as the tangent space to S2 at
G(p), so we can also write this derivative as the map
Dp G : Tp S → Tp S.
 2. THE GAUSS AND WEINGARTEN MAPS 57

Definition 2.2. Let S be a surface, and let p ∈ S. The Weingarten map Wp of S at p

is defined to be
Wp = −Dp G.
The negative sign is a convention, adopted to simplify certain formulae later.
We can define the second fundamental form using the Weingarten map. We define a new
inner product on Tp S by setting
⟪v, w⟫p,S = hWp (v), wip,S ,
where v, w ∈ Tp S. The new inner product ⟪·, ·⟫ encodes the second fundamental form:
Proposition 2.3. Let S be a surface, let p ∈ S, and let σ(u, v) be a surface patch in S
containing p, and let L du2 + 2M du dv + N dv 2 be its second fundamental form. Then for
all v, w ∈ Tp S, we have
⟪v, w⟫ = L du(v) du(w) + M (du(v) dv(w) + du(w) dv(v)) + N dv(v) dv(w).
To prove Proposition 2.3, we need the following lemma:
Lemma 2.4. Let σ(u, v) be a surface patch with standard unit normal N(u, v). Then
Nu · σ u = −L, Nu · σ v = Nv · σ u = −M, Nv · σ v = −N.
Proof. Since σ u and σ v are tangent vectors, we have
N · σ u = N · σ v = 0.
Differentiating these equations with respect to u and v gives
Nu · σ u = −N · σ uu = −L,
Nu · σ v = −N · σ uv = −M,
Nv · σ u = −N · σ uv = −M,
Nv · σ v = −N · σ vv = −N.


Proof of Proposition 2.3. Both sides of the equation are bilinear forms on Tp S, so
it suffices to verify that they agree when v and w are σ u and σ v . Recall that du(σ u ) =
dv(σ v ) = 1 and du(σ v ) = dv(σ u ) = 0. Thus we must show that
⟪σ u , σ u ⟫ = L, ⟪σ u , σ v ⟫ = ⟪σ v , σ u ⟫ = M, ⟪σ v , σ v ⟫ = N.
Suppose that σ(u0 , v0 ) = p. Then, with all derivatives evaluated at (u0 , v0 ), we have
d d
W(σ u ) = − G(σ(u, v0 )) = − N(u, v0 ) = −Nu .
du u=u0 du u=u0

Similarly, W(σ v ) = −Nv . Thus we have

⟪σ u , σ u ⟫ = hW(σ u ), σ u i = −Nu · σ u = L.
The proofs of the other equations are similar. 
58 5. CURVATURE OF SURFACES

In the original definition of ⟪v, w⟫, it was not so obvious that we have ⟪v, w⟫ = ⟪w, v⟫,
but this is now clear from Proposition 2.3. It follows that W is a self-adjoint operator with
respect to the inner product h·, ·i, i.e. that

hW(v), wi = hv, W(w)i.

There is a considerable theory of self-adjoint operators. The main result is the Spectral
Theorem, which says that there is an orthonormal basis of eigenvectors of any self-adjoint
operator, and that all eigenvalues are real. The eigenvalues and eigenvectors will be very
important next week: the eigenvalues of the Weingarten operator are called the principal
curvatures of the surface, and the corresponding eigenvectors are called the principal vectors.

Example. Let U ⊆ R2 be open, and let f : U → R be a smooth function. Let G be the

graph of f , i.e. G = {(x, y, f (x, y)) : (x, y) ∈ U }. Then G is covered by the single surface
patch σ given by σ(x, y) = (x, y, f (x, y)). We have σ x = (1, 0, fx ) and σ y = (0, 1, fy ), so

σx × σy (−fx , −fy , 1)
N= =p 2 .
kσ x × σ y k fx + fy2 + 1

Let (x0 , y0 ) ∈ U , and let p = σ(x0 , y0 ) = (x0 , y0 , f (x0 , y0 )). Suppose that f has a critical
point at (x0 , y0 ), so that fx (x0 , y0 ) = fy (x0 , y0 ) = 0. Thus we have N(p) = (0, 0, 1), and Tp G
is spanned by σ x = (1, 0, 0) and σ y = (0, 1, 0).
Let γ 1 be a curve on G defined by γ 1 (t) = (x0 + t, y0 , f (x0 + t, y0 )). Then γ 1 is such that
γ 1 (0) = p and γ 01 (0) = (1, 0, 0). The restriction of N to γ 1 is
(−fx (x0 + t, y0 ), −fy (x0 + t, y0 ), 1)
N(γ 1 (t)) = p .
fx (x0 + t, y0 )2 + fy (x0 + t, y0 )2 + 1

We can check by computation (being more clever than computing the full derivative and
then evaluating) that

d d
N(γ 1 (t)) = (−fx (x0 + t, y0 ), −fy (x0 + t, y0 ), 1) = (−fxx (x0 , y0 ), −fyx (x0 , y0 ), 0).
dt t=0 dt t=0

Thus we have
Wp (1, 0, 0) = (fxx (x0 , y0 ), fyx (x0 , y0 ), 0).
A similar calculation shows that

Wp (0, 1, 0) = (fxy (x0 , y0 ), fyy (x0 , y0 ), 0).

If we let {(1, 0, 0), (0, 1, 0)} be our chosen basis for Tp G, then the Weingarten map has matrix
 
fxx (x0 , y0 ) fxy (x0 , y0 )
.
fyx (x0 , y0 ) fyy (x0 , y0 )

The Gaussian and mean curvatures, which we shall look at in more detail later, are the
determinant and half the trace, respectively, of the Weingarten operator. In this case, they
2
are K(p) = fxx fyy − fxy and H(p) = fxx +f
2
yy
, respectively.
 3. NORMAL AND GEODESIC CURVATURE 59

3. Normal and Geodesic Curvature

Suppose we have a curve γ on a surface patch σ. How does the curvature of σ (or,
say, the second fundamental form) influence the curvature of γ? The curvature of σ cannot
completely determine the curvature of γ because there are curves of different curvature
drawn on the flat plane. Still, it seems like some contribution to the curvature of γ should
come from that of σ.
Suppose that γ is unit speed on S. Then γ̇ is a unit vector that is tangent to S. Thus
γ̇ is perpendicular to the unit normal N of S. Thus the three vectors γ̇, N, and N × γ̇
are mutually perpendicular. Furthermore, γ̈ is perpendicular to γ̇. Thus it can be written
uniquely as a linear combination of the other two vectors:
(3.1) γ̈ = κn N + κg N × γ̇.
Definition 3.1. The scalars κn and κg in (3.1) are called the normal curvature and
geodesic curvature of γ, respectively.
The following formulae are easily derived from the definition:
Proposition 3.2. We have
κn = γ̈ · N,
κg = γ̈ · (N × γ̇),
κ2 = κ2n + κ2g ,
κn = κ cos ψ,
κg = ±κ sin ψ,
where κ is the curvature of γ and ψ is the angle between N and the principal normal n of γ.
The normal curvature can be expressed in terms of the second fundamental form:
Proposition 3.3. Suppose γ is a unit-speed curve on an oriented surface S. Then its
normal curvature is given by
κn = ⟪γ̇, γ̇⟫.
Equivalently, if σ is a surface patch on S and γ(t) = σ(u(t), v(t)), then
κn = Lu02 + 2M u0 v 0 + N v 02 .
A consequence is that if two curves γ and γ
e on S intersect at p and have the same tangent
vector at p, then they have the same normal curvature, because the normal curvature only
depends on information about the surface and the first derivative of the curve.
Proof. Since γ̇ is a tangent vector to S, N · γ̇ = 0. Thus N · γ̈ = −Ṅ · γ̇, so
κn = N · γ̈ = −Ṅ · γ̇ = hW(γ̇), γ̇i = ⟪γ̇, γ̇⟫
because
d
Ṅ = G(γ(t)) = −W(γ̇).
dt

The geodesic curvature, on the other hand, depends only on the first fundamental form,
as you will show in problem 10.
60 5. CURVATURE OF SURFACES

Theorem 3.4 (Meusnier). Let p be a point on a surface S, and let v be a unit tangent
vector to S at p. Let Πθ be the plane containing the line through p parallel to v and making
an angle θ with Tp S, and assume that Πθ is not parallel to Tp S. Let γ θ be the curve of
intersection between Πθ and S, and let κθ be its curvature. Then κθ sin θ is independent of
θ.
Proof. Suppose γ θ is parametrized so as to be unit speed. Then at p, we have γ̇ θ = ±v,
so γ̈ θ is perpendicular to v and parallel to Πθ . Then, in the notation of Proposition 3.2, we
have ψ = π2 − θ, so
κn = κθ sin θ.
But κn only depends on p and v, not on θ, so κn is independent of θ. 
Definition 3.5. A normal section to a surface S is the intersection of S with Π π2 , where
notation is as in Meusnier’s Theorem.
Corollary 3.6. The curvature, normal curvature, and geodesic curvature of a normal
section of a surface are related by
κn = ±κ, κg = 0.

4. The Covariant Derivative and Parallel Transport

Definition 4.1. Let γ : (a, b) → S be a curve on a surface S. A tangent vector field
along γ is a map v : (a, b) → R3 such that v(t) ∈ Tγ(t) S for all t ∈ (a, b).
Suppose we have a tangent vector field v along γ. Then an observer moving along γ
will perceive the rate of change of v to be only the tangential component of v̇, meaning the
orthogonal projection of v̇ = dv
dt
onto Tγ(t) S. That is, let N be the unit normal to σ. Then
3
any vector in R can be written uniquely as t + aN for some t ∈ Tγ(t) S; the orthogonal
projection of this vector onto the tangent space Tγ(t) S is simply t. In the case of the rate of
change of a smooth tangent vector field, the tangential component is
∇γ v = v̇ − (v̇ · N)N.
Definition 4.2. Let γ be a curve on a surface S, and let v be a smooth tangent vector
field along γ. The covariant derivative of v along γ is the orthogonal projection ∇γ v of dv
dt
onto the tangent plane Tγ(t) S at a point γ(t).
To an observer living on S, v would appear to be constant along γ if ∇γ v = 0.
Definition 4.3. A tangent vector field v is said to be parallel along γ if ∇γ v = 0 at
every point of γ.
Another characterization of parallel vector fields is given in the following proposition,
which is obvious from the definition:
Proposition 4.4. A tangent vector field v is parallel along a curve γ on a surface S iff
v̇ is perpendicular to the tangent plane of S at all points of v.
Given a surface patch σ(u, v), the vectors σ u , σ v , and N form a basis for R3 , so any vector
in R3 can be expressed uniquely as a linear combination of them. It will be helpful to express
the second derivatives σ uu , σ uv , σ vv as linear combinations. The following proposition tells
us what the coefficients are, in terms of the first and second fundamental forms:
 4. THE COVARIANT DERIVATIVE AND PARALLEL TRANSPORT 61

Proposition 4.5 (Gauß Equations). Let σ(u, v) be a surface patch with first and second
fundamental forms
E du2 + 2F du dv + G dv 2 and L du2 + 2M du dv + N dv 2 ,
respectively. Then we have
σ uu = Γ111 σ u + Γ211 σ v + LN,
σ uv = Γ112 σ u + Γ212 σ v + M N,
σ vv = Γ122 σ u + Γ222 σ v + N N,
where
GEu − 2F Fu + F Ev
Γ111 = ,
2(EG − F 2 )
2EFu − EEv − F Eu
Γ211 = ,
2(EG − F 2 )
GEv − F Gu
Γ112 = ,
2(EG − F 2 )
EGu − F Ev
Γ212 = ,
2(EG − F 2 )
2GFv − GGu − F Gv
Γ122 = ,
2(EG − F 2 )
EGv − 2F Fv + F Gu
Γ222 = .
2(EG − F 2 )
The six coefficients Γkij are called the Christoffel symbols. Note that they only depend on
the first fundamental form of σ.

Proof. From linear algebra, as discussed above, there exist scalar functions α1 , . . . , γ3
such that
σ uu = α1 σ u + α2 σ v + α3 N,
σ uv = β1 σ u + β2 σ v + β3 N,
σ vv = γ1 σ u + γ2 σ v + γ3 N.
Taking dot products with N gives
α3 = L, β3 = M, γ3 = N.
Next, take dot products with σ u and σ v , which gives us six equations in six variables. We
omit the computation, but when we solve these simultaneous linear equations, we get the
formulae claimed. 

Given a tangent vector field v along a curve γ on a surface patch σ, it would be good to
know how to tell if it is parallel. Since v is a tangent vector field, it can be expressed as
v(t) = α(t)σ u + β(t)σ v .
62 5. CURVATURE OF SURFACES

Proposition 4.6. Let γ(t) = σ(u(t), v(t)) be a curve on a surface patch σ, and let
v(t) = α(t)σ u +β(t)σ v be a tangent vector field along γ, where α and β are smooth functions
of t. Then v is parallel along γ iff the following equations are satisfied:
α0 + (Γ111 u0 + Γ112 v 0 )α + (Γ112 u0 + Γ122 v 0 )β = 0,
β 0 + (Γ211 u0 + Γ212 v 0 )α + (Γ212 u0 + Γ222 v 0 )β = 0.
Note that these equations only involve the first fundamental form.
Proof. By the Gauß equations, we have
v̇ = α0 σ u + β 0 σ v + αu0 (Γ111 σ u + Γ211 σ v + LN)
+ (αv 0 + βu0 )(Γ112 σ u + Γ212 σ v + M N) + βv 0 (Γ122 σ u + Γ222 σ v + N N).
Now, v is parallel along γ iff v̇ is parallel to N, which means that the coefficients of σ u and
σ v must be 0. This implies the equations stated in the proposition. 
The equations are of the form
α0 = f (α, β, t), β 0 = g(α, β, t),
where f and g are smooth functions of three variables. By the theory of ordinary differential
equations, such equations have a unique solution given any set of initial conditions, i.e. if t0
is some particular value of t and α0 , β0 ∈ R, then there are unique smooth functions α(t)
and β(t) defined on an open interval containing t0 satisfying the given equations and such
that
α(t0 ) = α0 , β(t0 ) = β0 .
Corollary 4.7. Let γ be a curve on S, and let v0 be a tangent vector of S at the
point γ(t0 ). Then there is a unique tangent vector field v that is parallel along γ such that
v(t0 ) = v0 .
Example. For θ0 ∈ (− π2 , π2 ), and let γ be the circle of latitude θ0 on the unit sphere.
(Recall that the sphere has a surface patch σ(θ, φ) = (cos θ cos φ, cos θ sin φ, sin θ). Thus
γ is given by γ(t) = (cos θ0 cos t, cos θ0 sin t, sin θ0 ). The first fundamental form of σ is
dθ2 + cos2 θ dφ2 , so we can compute the Christoffel symbols to be
Γ111 = Γ211 = Γ222 = Γ112 = 0, Γ212 = − tan θ, Γ122 = sin θ cos θ.
Thus the differential equations above become
α0 = −β sin θ0 cos θ0 , β 0 = α tan θ0 .
If θ0 = 0, then α and β are constant. If θ0 6= 0, then we can eliminate β from the differential
equations to get a single differential equation in just α, namely
α00 + α sin2 θ0 = 0,
from which we get the general solution
α(φ) = A cos(φ sin θ0 ) + B sin(φ sin θ0 ),
where A and B are constants. We can then solve for β to get
sin(φ sin θ0 ) cos(φ sin θ0 )
β=A −B .
cos θ0 sin θ0
 5. PROBLEMS 63

Suppose that v = σ φ , so v is tangent to γ when φ = 0. Then α = 0 and β = 1 when

φ = 0, so we get A = 0 and B = − sin θ0 , so
v(t) = − sin θ0 sin(t sin θ0 )σ θ + cos(t sin θ0 )σ φ .
Note that v(t) is not tangent to γ at γ(t) for all t, unless θ0 = 0. Thus the tangent vector
of γ is parallel along γ if and only if γ is a great circle.
Suppose that p and q are two points on a curve γ on S. Then, given any v ∈ Tp S, we
can “transport” it to a vector in Tq S in a parallel manner along S. To do this, suppose that
p = γ(t0 ) and q = γ(t1 ), and let v(t) be the unique parallel vector field along γ such that
v(t0 ) = v0 . Then let v1 = v(t1 ) ∈ Tq S.
Definition 4.8. The map Πpq γ : Tp S → Tq S taking v0 ∈ Tp S to v1 ∈ Tq S is called
parallel transport from p to q along γ.

5. Problems
(1) Prove that if the second fundamental form of a surface patch σ is zero everywhere,
then σ is an open subset of a plane.
(2) Characterize the connected surfaces for which the image of the Gauß map is con-
tained in the equator of S2 .
(3) Prove that K ≤ H 2 at every point of a surface.
(4) Show that every compact surface has a point of positive Gaussian curvature.
(5) Let S be a compact surface, and let S + = {p ∈ S : K(p) ≥ 0}. Prove that the
restriction of the Gauß map to S + is surjective.
(6) Let γ be a regular, but not necessarily unit-speed, curve on a surface. Show that
the normal and geodesic curvatures are given by
⟪γ̇, γ̇⟫ γ̈ · (N × γ̇)
κn = , κg = .
hγ̇, γ̇i hγ̇, γ̇i3/2
(7) Show that the normal curvature of any curve on a sphere of radius r is ± 1r .
(8) Compute the geodesic curvature of any circle on a sphere (not necessarily a great
circle).
(9) More generally, compute the geodesic curvature of an arbitrary unit-speed curve γ
on S2 .
(10) Show that if γ(t) = σ(u(t), v(t)) is a unit-speed curve on a surface patch σ with
first fundamental form E du2 + 2F du dv + G dv 2 , the geodesic curvature of γ is
√
κg = (v 00 u0 − v 0 u00 ) EG − F 2 + Au03 + Bu02 v 0 + Cu0 v 02 + Dv 03 ,
where A, B, C, D can be expressed in terms of E, F, G and their derivatives. When
F = 0, find A, B, C, D explicitly.
(11) Suppose that γ is a unit-speed curve with κ > 0 and principal normal n, and that
γ is the intersection of two oriented surfaces S1 and S2 with unit normals N1 and
N2 , respectively.
(a) Show that if κ1 and κ2 are the normal curvatures of γ when considered as
curves in S1 and S2 , respectively, then
κ1 N2 − κ2 N1 = κ(N1 × N2 ) × n.
64 5. CURVATURE OF SURFACES

(b) Show that if α is the angle between the two surfaces, then
κ2 sin2 α = κ21 + κ22 − 2κ1 κ2 cos α.
(12) A curve γ on a surface S is said to be an asymptotic curve if its normal curvature
is zero everywhere.
(a) Show that if γ is a straight line, then it is an asymptotic curve.
(b) Show that a curve γ with κ > 0 is an asymptotic curve if and only if its
binormal b is parallel to the unit normal N of S at all points of γ.
(13) What are the asymptotic curves on the surface
σ(u, v) = (u cos v, u sin v, log u)?
(14) Let pqr be a triangle on a unit sphere, where the sides are arcs of great circles. Let
v0 be a nonzero tangent vector to the arc pq through p and q at p. Suppose we
parallel transport v0 along pq, then along qr, then along rp back to p. Show that
the resulting vector is a rotation of v0 by an angle of 2π − A, where A is the area
of triangle pqr. This is a first example of the phenomenon of holonomy.
 CHAPTER 6

Principal Curvatures

1. Gaussian and Mean Curvatures

Let W be the Weingarten map of a surface S at a point p ∈ S. Recall that W is a linear
map from Tp S to itself.

Definition 1.1. We define the Gaussian curvature and the mean curvature of S at p
to be K = det(W) and H = 21 Tr(W), respectively.

Recall from linear algebra that we can compute the determinant and trace of a linear
operator on a vector space V by choosing a basis for V and writing down the matrix  of
a b
the operator in terms of that matrix. Then the determinant of the 2 × 2 matrix is
c d
ad − bc, whereas the trace is a + d. However, the determinant and trace are independent of
the choice of basis.
It’s worth noting that if S is nonorientable, then K is still defined: replacing the unit
normal by its negative negates the linear operator, leaving the determinant unchanged.
However, H is only defined up to sign if S is nonorientable, because negating the Weingarten
map multiplies the trace by −1.
So far, our discussion of the Gaussian and mean curvatures has been abstract, since we
don’t have a natural choice of matrix to use to represent the Weingarten map. However, if
we choose a surface patch σ on S, then we can represent the Weingarten map in terms of
σ, which will give us a closed formula for the curvatures.
Let σ be a surface patch on S, and let

E du2 + 2F du dv + G dv 2 and L du2 + 2M du dv + N dv 2

be the first and second fundamentals forms of σ, respectively. Define two 2 × 2 matrices
   
E F L M
FI = , FII =
F G M N

associated to the two fundamental forms.

Proposition 1.2. With the above notation, the Weingarten map W with respect to the
basis σ u , σ v of Tp S is FI−1 FII .

Proof. As we saw last week, we haveW(σ u ) = −Nu and W(σ v ) = −Nv , so the matrix
a c
of W with respect to the basis σ u , σ v is , where
b d

−Nu = aσ u + bσ v , −Nv = cσ u + dσ v .
65
66 6. PRINCIPAL CURVATURES

To find a, b, c, d, take the dot product of each of the above equations with σ u and σ v to get
L = aE + bF,
M = aF + bG,
M = cE + dF,
N = cF + dG.
We can rewrite these four equations in matrix form as
    
L M E F a c
= ,
M N F G b d
i.e.  
a c
= FI−1 FII ,
b d
as claimed. 
Corollary 1.3. We have
LG − 2M F + N E LN − M 2
H= , K= .
2(EG − F 2 ) EG − F 2
Example. Let us compute the curvatures of a surface of revolution. Let
σ(u, v) = (f (u) cos v, f (u) sin v, g(u)),
where as usual we assume that f > 0 and f 02 + g 02 = 1. As we have already seen, we have
 0 00
f g − f 00 g 0 0
  
1 0
FI = , FII = .
0 f2 0 f g0
Thus the Gaussian curvature is
LN − M 2 (f 0 g 00 − f 00 g 0 )f g 0
K= = .
EG − F 2 f2
Since f 02 + g 02 = 1, we can differentiate to get f 0 f 00 + g 0 g 00 = 0, so we have
(f 0 g 00 − f 00 g 0 )g 0 = −f 02 f 00 − f 00 g 02 = −f 00 (f 02 + g 02 ) = −f 00 ,
so we can simplify our formula for K to get
−f 00 f f 00
K= = − .
f2 f
We can take a few special cases. If γ(u) = (u, 0, 0), then we have f (u) = u, so f 00 = 0,
and our formula for K simplifies to 0, which makes sense: the plane ought to have curvature
zero. If γ(u) = (1, 0, u), then σ is a cylinder, and we have f 00 = 0 again, so again K = 0.
If γ(u) = (cos u, 0, sin u), then σ is a sphere, and we have f (u) = cos u, so f 00 (u) = − cos u.
00
Thus we have K = − ff = 1.
Example. Let S be a ruled surface. This means that we can choose a patch σ(u, v) =
d
γ(u) + vδ(u). Letting a dot denote du , we have σ u = γ̇ + v δ̇ and σ v = δ, so
σ uv = δ̇, σ vv = 0.
Thus if
σu × σv
N=
kσ u × σ v k
 1. GAUSSIAN AND MEAN CURVATURES 67

is the standard unit normal, then

M = σ uv · N = δ̇ · N

and N = 0. Thus we have

LN − M 2 −(δ̇ · N)2
K= = ≤ 0.
EG − F 2 EG − F 2

In other words, the Gaussian curvature of a ruled surface is always nonpositive.

The Gaussian curvature is defined in terms of the Weingarten map, which is the derivative
of the Gauss map. However, the Gaussian curvature also has an interpretation in terms of
the Gauss map itself rather than its derivative. Recall that in the case of curves, if γ is a
unit-speed plane curve, then its signed curvature κs is the rate of change of the direction
of the tangent vector of γ per unit length. The Gaussian curvature measures something
similar, but up a dimension. Instead of the tangent vector, we have the tangent plane, which
carries the same information as the unit normal vector N, so we can hope that the Gaussian
curvature of σ measures the rate of change of N per unit area. Since the values of N are
recorded by the Gauss map G, we can consider the ratio

Area(G(R))
Area(R)

as R shrinks to just the single point p.

Theorem 1.4. Let σ : U → R3 be a surface patch of a surface S and let (u0 , v0 ) ∈ U .

For δ > 0, let Rδ = {(u, v) ∈ U : (u − u0 )2 + (v − v0 )2 ≤ δ 2 }. Then

AN (Rδ )
lim = |K|,
δ→0 Aσ (Rδ )

where K is the Gaussian curvature of σ at σ(u0 , v0 ).

By AN and Aσ , we mean
Z
AN (R) = kNu × Nv k du dv
R

and
Z
Aσ (R) = kσ u × σ v k du dv.
R

Proof. We have
R
AN (Rδ ) R
kNu × Nv k du dv
= Rδ .
Aσ (Rδ ) Rδ
kσ u × σ v k du dv
68 6. PRINCIPAL CURVATURES

In the notation of Proposition 1.2, we have

Nu × Nv = (aσ u + bσ v ) × (cσ u + dσ v )
= (ad − bc)σ u × σ v
= det(FI−1 FII )σ u × σ v
det FII
= σu × σv
det FI
LN − M 2
= σu × σv
EG − F 2
= Kσ u × σ v .
Let ε > 0. Since K(u, v) is a continuous function of u and v, there exists a δ > 0 such that
|K(u, v) − K(u0 , v0 )| < ε whenever (u, v) ∈ Rδ . Thus for all (u, v) ∈ Rδ , we have
|K(u0 , v0 )| − ε < |K(u, v)| < |K(u0 , v0 )| + ε.
Multiplying by kσ u × σ v k and integrating over Rδ , we have
Z Z
(|K(u0 , v0 )| − ε) kσ u × σ v k du dv < |K(u, v)|kσ u × σ v k du dv
Rδ Rδ
Z
< (K(u0 , v0 )| + ε) kσ u × σ v k du dv.
Rδ

Thus
AN (Rδ )
|K(u0 , v0 )| − ε < < |K(u0 , v0 )| + ε,
Aσ (Rδ )
or
AN (Rδ )
− |K(u0 , v0 )| < ε.
Aσ (Rδ )
This completes the proof. 
This only tells us about the unsigned Gaussian curvature, but we can also use a signed
version of area to recover an analogous statement about the signed Gaussian curvature.

2. Principal Curvatures
As we saw last week, the Weingarten map is a self-adjoint operator. It follows from the
Spectral Theorem that the tangent plane Tp S at p has an orthonormal basis of eigenvectors
of the Weingarten map: that is, there exist t1 , t2 ∈ Tp S such that
W(t1 ) = κ1 t1 , W(t2 ) = κ2 t2 , ht1 , t2 i = 0
for some scalars κ1 and κ2 .
The numbers κ1 , κ2 are the eigenvalues of W, and t1 , t2 are the corresponding eigenvec-
tors. In the context of differential geometry, we call κ1 and κ2 the principal curvatures of S,
and we call t1 and t2 the principal vectors. When κ1 = κ2 , then there is no canonical choice
of t1 , t2 , even up to sign, but otherwise these vectors are uniquely defined up to sign.
If κ1 = κ2 at a point p, then we call p an umbilic point of S. At these points, W is a scalar
multiple of the identity matrix, and every (unit) tangent vector is principal. Otherwise, the
principal vectors are necessarily orthogonal.
 2. PRINCIPAL CURVATURES 69

We can relate the principal curvatures to the Gaussian and mean curvatures:
κ1 + κ2
H= , K = κ1 κ2 .
2
This is because in a basis of principal tangent vectors, the Weingarten map has matrix
 
κ1 0
,
0 κ2
so the determinant K is κ1 κ2 , whereas the trace 2H is κ1 + κ2 .
We can read off the normal curvature of a curve on a surface in terms of the principal
curvatures of the surface and a bit more information about the angle between the curve and
the principal vectors:
Theorem 2.1. Let γ be a curve on S, and let κ1 and κ2 be the principal curvatures with
principal tangent vectors t1 and t2 . Let θ be the oriented angle between t1 and γ̇. Then
κn = κ1 cos2 θ + κ2 sin2 θ.
Proof. Assume, by replacing t2 with −t2 if necessary, that the oriented angle between
t1 and t2 is π2 (and not − π2 ). Then we have
γ̇ = cos θ t1 + sin θ t2 .
Then we have
κn = ⟪γ̇, γ̇⟫ = cos2 θ ⟪t1 , t1 ⟫ + 2 sin θ cos θ ⟪t1 , t2 ⟫ + sin2 θ ⟪t2 , t2 ⟫.
We have (
κi i = j,
⟪ti , tj ⟫ = hW(ti ), tj i = hκi ti , tj i =
0 i 6= j.
The result follows. 
Corollary 2.2. The principal curvatures of a surface S at a point p ∈ S are the
maximum and minimum values of the normal curvature of all curves on S passing through
p.
We can compute the principal curvatures directly from the fundamental forms. Since the
Weingarten matrix is FI−1 FII , the principal curvatures are the roots of the polynomial
det(FI−1 FII − κI) = 0.
Writing everything out in terms of E, F, G, L, M, N , we find that the principal curvatures
are the roots of the polynomial
 
L − κE M − κF
det = 0.
M − κF N − κG
The principal vectors corresponding to the principal curvature κ are (up to scaling) the
tangent vectors t = ξσ u + ησ v such that
    
L − κE M − κF ξ 0
= .
M − κF N − κG η 0
70 6. PRINCIPAL CURVATURES

Example. Let us compute the principal curvatures of the unit sphere. As we have seen,
we have  
1 0
FI = FII = .
0 cos2 θ
Thus the principal curvatures are the roots of
 
1−κ 0
det = 0,
0 cos2 θ − κ cos2 θ
which is just a double root at κ = 1. This makes sense: the sphere curves exactly the same
amount in every direction. This also “explains” why the first and second fundamental forms
of the sphere are equal.
Example. Consider the cylinder σ(u, v) = (cos v, sin v, u). Then the first and second
fundamental forms are
du2 + dv 2 and dv 2 ,
respectively. Thus the principal curvatures are the roots of
 
−κ 0
det = 0,
0 1−κ
which are κ = 0 and κ = 1. The principal vectors for κ = 0 and κ = 1 at σ(u, v) =
(cos v, sin v, u) are (0, 0, 1) and (− sin v, cos v, 0), respectively. This should be intuitively
clear: the most curved direction is around the circle, and the least curved direction is vertical.
On a sphere or a plane, every point is an umbilic. But these turn out to be essentially
the only surfaces with this property.
Proposition 2.3. Let S be a connected surface on which every point is an umbilic. Then
S is an open subset of a plane or sphere.
Proof. Suppose the principal curvatures are κ. Then for every tangent vector t, we
have W(t) = κt. Let σ : U → R3 be a surface patch of S, where U is an open subset of R2 .
Let t = σ u and t = σ v and recall that W(σ u ) = −Nu and W(σ v ) = −Nv . Thus we have
Nu = −κσ u , Nv = −κσ v .
Thus we have
(κσ u )v = (−Nu )v = −(Nv )u = (κσ v )u ,
so
κv σ u = κu σ v .
Since σ is regular, σ u and σ v are linearly independent, so κu = κv = 0. Thus κ is constant
on S.
Suppose now that κ = 0. Then N is constant. Thus
(N · σ)u = N · σ u = 0, (N · σ)v = N · σ v = 0,
so N · σ is constant, say c. Thus σ(U ) is an open subset of the plane v · N = c.
Now suppose that κ 6= 0. Then we have
N = −κσ + a
 2. PRINCIPAL CURVATURES 71

for some constant vector a. Thus

2 2
1 1 1
σ− a = − N = ,
κ κ κ2
so σ(U ) is an open subset of the sphere with center κ1 a and radius κ1 .
This argument shows that any single surface patch is an open subset of a plane or a
sphere. If the images of two such patches intersect, they have to be contained in the same
plane or sphere. Thus all of S is contained in some plane or sphere. 
So far, we have seen spheres, where both principal curvatures are positive (or both
negative if we parametrize them differently), cylinders where one is positive (or negative if
we parametrize them differently) and one is zero, and planes where both are zero. More
generally, the principal curvatures of S at p tell us the local shape of S near p. Let us
assume that p is the origin and that Tp S is the xy-plane; this can be achieved by performing
a suitable isometry of R3 , which does not affect its shape or principal curvatures. Let us
also assume that the principal vectors are t1 = (1, 0, 0) and t2 = (0, 1, 0), corresponding to
principal curvatures κ1 and κ2 . Finally, we’ll assume that N = (0, 0, 1).
Let σ be a surface patch on S with σ(0, 0) = 0. For any x, y ∈ R, there exist unique s, t
such that (x, y, 0) = sσ u + tσ v , where the derivatives are evaluated at (0, 0). By Taylor’s
Theorem, we have
1
σ(s, t) = σ(0, 0) + sσ u + tσ v + (s2 σ uu + 2stσ uv + t2 σ vv ),
2
plus higher order terms. Thus if x and y are small (hence s and t are small as well), then
σ(s, t) = (x, y, z), where
1 1
z ≈ (s2 σ uu + 2stσ uv + t2 σ vv ) · N = (Ls2 + 2M st + N t2 ).
2 2
If t = sσ u + tσ v , then
Ls2 + 2M st + N t2 = ⟪t, t⟫ = hW(t), ti.
Since t = xt1 + yt2 , we have
W(t) = xW(t1 ) + yW(t2 ) = κ1 xt1 + κ2 yt2 = (κ1 x, κ2 y, 0).
Thus
Ls2 + 2M st + N t2 = (κ1 x, κ2 y, 0) · (x, y, 0) = κ1 x2 + κ2 y 2 .
Thus near p, S is approximated by the quadric surface
1
z = (κ1 x2 + κ2 y 2 ).
2
Thus we have the following classification:
• If κ1 , κ2 are both nonzero and have the same sign, then S locally looks like an elliptic
paraboloid. We say that p is an elliptic point.
• If κ1 , κ2 are both nonzero but have opposite signs, then S locally looks like a hy-
perbolic paraboloid. We say that p is a hyperbolic point.
• One of κ1 , κ2 is zero, and the other is nonzero. Then S locally looks like a parabolic
cylinder, and we say that p is a parabolic point.
72 6. PRINCIPAL CURVATURES

x y

Figure 1. A surface with a monkey saddle.

• κ1 = κ2 = 0. Then the second-order approximation of S has the equation of a plane,

and we say that p is a planar point. In this case, the higher order derivatives dictate
the shape of S near p, and more complicated behaviors can occur. For instance,
the surface defined by z = x3 − 3xy 2 shown in Figure 1 has a monkey saddle at the
origin.

3. Constant Gaussian Curvature

So far, we have seen that spheres have constant positive Gaussian curvature: a sphere
with radius R has constant Gaussian curvature equal to R12 . We have also seen that planes
and cylinders have constant Gaussian curvature equal to zero. Are there surfaces with
constant negative Gaussian curvature? And can we classify all the surfaces with constant
Gaussian curvature?
Let us consider surfaces of revolution, given by a surface patch
σ(u, v) = (f (u) cos v, f (u) sin v, g(u)),
with the usual conventions on f and g. We have seen that its Gaussian curvature is
f 00
K=− .
f
Let us first consider the case where K = 0. This implies that f 00 = 0, so √ we have
f (u) = au +√b for some constants a and b. Since f 02 + g 02 = 1, we have g 0 = ± 1 − a2 ,
so g(u) = ± 1 − a2 u + c for some c. By translating along the z-axis if necessary, we may
assume that c = 0, and by rotation by π around the x-axis if necessary, we may assume that
the sign is positive. Thus we obtain the ruled surface
√
σ(u, v) = (b cos v, b sin v, 0) + u(a cos v, a sin v, 1 − a2 ).
If a = 0, this is a circular cylinder; if |a| = 1, then it’s the xy-plane; if 0 < |a| < 1, then it’s
a circular cone.
Now let’s consider the case K > 0, say with K = R12 for some constant R. Then we have
f
f 00 + = 0.
R2
We can solve the differential equation to get
u 
f (u) = a cos +b ,
R
 4. FLAT SURFACES 73

where a and b are constants. By replacing u with u + Rb if necessary, we may assume that
b = 0. Thus, up to a sign and an additive constant, we have
Z ur
a2 t
g(u) = 1 − 2 sin2 dt.
0 R R
The integral only has an elementary antiderivative when a = 0 or a = ±R. If a = 0, then we
do not get a surface, and when a = ±R, we have f (u) = R cos Ru and g(u) = R sin Ru , which
is a sphere of radius R. If we choose other values of a, we still get a surface of revolution with
constant Gaussian curvature, at least locally: it may have self-intersections that prevent it
from being a surface in R3 globally.
Finally, consider the case K < 0. We’ll focus on the case K = −1; the general case can
be derived from it by a scaling. We want to end up with something like a sphere of radius
i, i.e. a “pseudosphere.” This time, we end up with the differential equation
f 00 − f = 0,
which has the general solution
f (u) = aeu + be−u ,
where a and b are constants. We can only express g in elementary terms if ab = 0. If b = 0,
then we can assume that a = 1 by replacing u with u + c for some suitable c, and the case
a = 0 can be reduced to the case b = 0 by replacing u with −u. So, we suppose that a = 1
and b = 0, so that f (u) = eu and
Z u√
g(u) = 1 − e2t dt.
0

The integral only makes sense when u ≤ 0, since otherwise the stuff inside the square root
is sometimes negative. If we set et = cos θ, then we have
Z u√ Z cos−1 eu
2t
sin2 θ
1 − e dt = − dθ
0 0 cos θ
= sin cos−1 eu − log(sec cos−1 eu + tan cos−1 eu )
√ √
= 1 − e2u − log(e−u + e−2u + 1).
√
Letting x = f (u) and z = g(u) and recalling that cosh−1 (v) = log(v + v 2 − 1), we find that
the profile curve has equation
√ 1
z = 1 − x2 − cosh−1 .
x
This curve is called a tractrix ; see Figure 2 Left. After rotating it around the z-axis, we end
up with a surface called a pseudosphere; see Figure 2 Right.

4. Flat Surfaces
We now turn to the classification of surfaces with K = 0.
Definition 4.1. A surface is said to be flat if K = 0 everywhere.
We begin with the following proposition that helps us simplify our analysis:
74 6. PRINCIPAL CURVATURES

Figure 2. Left: A tractrix. Right: A pseudosphere.

Proposition 4.2. Let p be a point on a surface S, and suppose that p is not an umbilic.
Then there is a surface patch σ(u, v) of S containing p whose first and second fundamental
forms satisfy F = 0 and M = 0, i.e. they take the form E du2 + G dv 2 and L du2 + N dv 2 ,
respectively.
We call such a patch a principal patch. You will prove Proposition 4.2 in problem 13.
Proposition 4.3. Let p be a point on a flat surface S, and assume that p is not an
umbilic. Then there is a patch of S containing p that is a ruled surface.
Proof. Take a principal patch σ : U → R3 containing p, say with p = σ(u0 , v0 ). Thus
the Gaussian curvature is K = LN EG
. Since S is flat, either L = 0 or N = 0 at each point of
U , and since p is not an umbilic, L and N are not both 0. Suppose that L(u0 , v0 ) 6= 0; the
case N (u0 , v0 ) 6= 0 is similar. Then L(u, v) 6= 0 for (u, v) in some open neighborhood U of
(u0 , v0 ). Thus, by shrinking U if necessary, we may assume that L 6= 0 on all of U . Thus
N = 0 everywhere, and the second fundamental form of σ is L du2 .
We will now prove that the parameter curves with constant values of u are straight lines,
which shows that S is a ruled surface, at least locally. If we have the curve u = u0 , then we
can parametrize it by v 7→ σ(u0 , v). A unit tangent vector to this curve is t = √σG v
, so we
have to prove that tv = 0.
We have
L
Nu = − σ u , Nv = 0.
E
Thus
E
tv · σ u = − tv · Nu .
L
Now, t · Nu = 0 and Nuv = 0, so tv · Nu = −t · Nuv = 0. Thus tv · σ u = 0. Now, tv · t = 0
because t is a unit vector, so tv · σ v = 0. Finally, tv · N = −t · Nv = 0. Since the vectors
σ u , σ v , N form a basis of R3 , it follows that tv = 0. 
 5. SURFACES OF CONSTANT MEAN CURVATURE 75

Thus, in order to classify flat surfaces, we just have to describe flat ruled surfaces. We
can parametrize ruled surfaces as
σ(u, v) = γ(u) + vδ(u).
So, we need to determine which ruled surfaces are flat. We have σ u = γ̇ + v δ̇ and σ v = δ,
where the dot denotes a derivative with respect to u. The Gaussian curvature of σ is zero if
and only if
δ̇ · (σ u × σ v ) = 0.
Since
σ u × σ v = γ̇ × δ + v δ̇ × δ
and δ̇ · (δ̇ × δ) = 0, we have that K = 0 if and only if δ̇ · (γ̇ × δ) = 0. In other words, K = 0
if and only if the vectors γ̇, δ, δ̇ are everywhere linearly dependent.
Suppose, as we are free to do, that δ is a unit vector for all u. Then δ · δ̇ = 0. If δ̇(u) = 0
for all u, then δ is a constant vector, so σ is a generalized cylinder.
Now suppose that δ̇ is never zero. Then δ and δ̇ are linearly independent since they
are nonzero and perpendicular, so if γ̇, δ, δ̇ are linearly dependent, then there exist smooth
functions f and g such that
γ̇(u) = f (u)δ(u) + g(u)δ̇(u).
If f = g 0 everywhere, then γ̇ = (gδ)˙, so γ = gδ + a for some constant vector a. Thus we
have
σ(u, v) = a + (v + g(u))δ(u).
Letting ũ = u and ṽ = v + g(u), we find that this is a generalized cone.
Now, suppose that δ̇ and f − g 0 are both nonzero everywhere. Let
v + g(u)
e = γ(u) − g(u)δ(u),
γ ṽ = .
f (u) − g 0 (u)
Then we find that
σ(u, v) = γ e˙ (u),
e (u) + ṽ γ
so σ is a reparametrization of an open subset of the tangent developable of γ e . (Recall that
the tangent developable of a space curve γ is the union of all the tangent lines to γ.)
Note that it might be the case that none of these cases holds everywhere, but only in
neighborhoods. Thus we have only shown that open subsets of a flat surface are parts of
generalized cylinders, generalized cones, or tangent developables. Indeed, we can patch up
a surface using pieces of these types along straight lines, so we have a surface that includes
pieces of two or three of these types.

5. Surfaces of Constant Mean Curvature

Soap bubbles always form surfaces with constant mean curvature, so these surfaces are
physically interesting. The main case is H = 0, which deserves more study, which alas we
do not have time for in this class. Such surfaces are called minimal surfaces. Due to lack of
time, we’ll just mention, without proof, a few of the main results about surfaces of constant
nonzero mean curvature.
76 6. PRINCIPAL CURVATURES

Definition 5.1. Let S be an oriented surface, and let λ ∈ R. The parallel surface S λ of
S is defined to be
S λ = {p + λNp : p ∈ S},
where Np is the unit normal of S at p.
Proposition 5.2. Let κ1 and κ2 be the principal curvatures of an oriented surface S,
let λ ∈ R, and let S λ be the parallel surface of S. Suppose that κ1 and κ2 are never equal to
1
λ
at any point of S. Then
(1) S λ is a smooth oriented surface. The unit normal S λ at p + λNp is equal to εNp ,
where ε is the sign of (1 − λκ1 )(1 − λκ2 ).
εκ1 εκ2
(2) The principal curvatures of S λ are 1−λκ 1
and 1−λκ 2
. The corresponding principal
vectors are the same as those for S.
(3) The Gaussian and mean curvatures of S λ are
H ε(H − λK)
2
and ,
1 − 2λH + λ K 1 − 2λH + λ2 K
where K and H are the Gaussian and mean curvatures of S, respectively.
Corollary 5.3. If S has constant Gaussian curvature R12 , then the parallel surfaces
S ±R have constant mean curvature 2R1 1
. Conversely, if S has constant mean curvature 2R ,
R 1
then the parallel surface S has constant Gaussian curvature R2 .
Proposition 5.4. Place an ellipse tangent to the x-axis in R2 , and roll it without slipping
along the x-axis. Let γ be the trajectory of one of its foci. If we rotate γ around the x-axis,
the resulting surface has constant nonzero mean curvature.
6. Problems
(1) Consider the surface z = f (x, y), where f is a smooth function. Show that the
Gaussian and mean curvature are given by
2
fxx fyy − fxy (1 + fy2 )fxx − 2fx fy fxy + (1 + fx2 )fyy
K= , H = .
(1 + fx2 + fy2 )2 2(1 + fx2 + fy2 )3/2
(2) Compute the Gaussian curvature of the helicoid σ(u, v) = (v cos u, v sin u, λu) and
the catenoid σ(u, v) = (cosh u cos v, cosh u sin v, u).
(3) Consider the ruled surface σ(u, v) = γ(u) + vδ(u). Show that if δ is the principal
normal n or binormal b, then K = 0 if and only if γ is planar.
(4) Let σ : U → R3 be a surface patch of a surface S. Let R ⊆ U be a region such that
the Gauß map is injective on σ(R). Show that its image under the Gauß map of
σ(R) has area Z
|K| dAσ .
R
(5) Let S be the torus defined by
σ(θ, φ) = ((a + b cos θ) cos φ, (a + b cos θ) sin φ, b sin θ),
where 0 < b < a. Let S + and S − be the parts of the torus where K > 0 and K < 0,
respectively. Show, as painlessly as possible, that
Z Z
K dA = − K dA = 4π.
S+ S−
 6. PROBLEMS 77
R
It follows that S K dA = 0, another special case of the Gauß–Bonnet Theorem.
(6) A curve γ on a surface S is called a line of curvature if the tangent vector of γ is
a principal vector of S at all points of γ. (Note that it need not be a straight line!)
Show that γ is a line of curvature if and only if
Ṅ = −λγ̇
for some scalar function λ. In this case, the corresponding principal curvature is λ.
This is called Rodrigues’s formula.
(7) Show that a curve γ(t) = σ(u(t), v(t)) on a surface patch σ is a line of curvature iff
(EM − F L)u02 + (EN − GL)u0 v 0 + (F N − GM )v 02 = 0.
Deduce that all parameter curves are lines of curvature if and only if one of the
following holds:
(a) The second fundamental form of σ is proportional to the first fundamental
form.
(b) F = M = 0.
Find all surfaces for which (i) holds.
(8) Show that all meridians and parallels of a surface of revolution are lines of curvature.
(9) Suppose a, b, c are distinct positive numbers. How many umbilic points are there on
the ellipsoid
x2 y 2 z 2
+ 2 + 2 = 1.
a2 b c
Can you describe them? What happens if a, b, c are√not all distinct?
(10) Let γ be the tractrix in the xz-plane, given by z = 1 − x2 − cosh−1 x1 , and let p be
a point on γ. Suppose the tangent line to γ at p intersects the z-axis at q. Show
that the distance from p to q is always equal to 1.
(11) Show that setting w = e−u gives a parametrization of the pseudosphere with first
fundamental form
dv 2 + dw2
.
w2
This is called the upper half-plane model of the hyperbolic plane.
(12) In the previous problem, let
v 2 + w2 − 1 −2v
V = 2 2
, W = 2 .
v + (w + 1) v + (w + 1)2
Show that this parametrization has first fundamental form
4(dV 2 + dW 2 )
.
(1 − V 2 − W 2 )2
This is called the Poincaré disk model of the hyperbolic plane.
(13) Prove Proposition 4.2.
 CHAPTER 7

Geodesics

1. Introduction to Geodesics
Geodesics on a surface are the curves that behave like straight lines. For example, the
shortest path between two points is always a geodesic. The definition of geodesic doesn’t
appear to be related to this notion, but it turns out to have the properties we want.
Definition 1.1. A curve γ on a surface S is said to be a geodesic if γ̈(t) is perpendicular
to the tangent plane of S at γ(t), for all t.
In other words, γ̈ is parallel or antiparallel to N for all parameters t. We allow the
possibility that γ̈ = 0. An equivalent statement is that γ is a geodesic if and only if γ̇ is a
parallel vector field along γ.
Proposition 1.2. A geodesic is constant speed.
Proof. Let γ be a geodesic on S. Then we have
d d
kγ̇k2 = (γ̇ · γ̇) = 2γ̈ · γ̇.
dt dt
Since γ is a geodesic, γ̈ is perpendicular to the tangent plane, so in particular it is perpen-
dicular to the tangent vector γ̇. Thus γ̈ · γ̇ = 0, which implies that kγ̇k is constant. 
Thus a unit-speed reparametrization of a geodesic is still a geodesic, because a unit-speed
reparametrization has the form γ(ct) for some fixed c. Thus we may freely restrict to the
case of unit-speed geodesics. Note that an arbitrary reparametrization of a geodesic is not a
geodesic, since a typical reparametrization will not be constant speed.
We can also express the notion of being a geodesic in terms of the geodesic curvature κg ,
which is the reason for the name.
Proposition 1.3. A unit-speed curve on a surface is a geodesic if and only if its geodesic
curvature is zero everywhere.
Proof. Let γ be a unit-speed curve on S, and let p ∈ S. Let σ be a surface patch of S
with p in its image, and let N be the standard unit normal of σ, so that
κg = γ̈ · (N × γ̇).
If γ̈ is parallel to N, then it is perpendicular to N × γ̇, so κg = 0.
Now, suppose that κg = 0. Then γ̈ is perpendicular to N × γ̇. Since γ̇, N, N × γ̇ are
perpendicular unit vectors and γ̈ is perpendicular to γ̇, it follows that γ̈ is parallel to N. 
Let’s now see some examples of geodesics.
Example. Any part of a straight line on any surface is a geodesic. To see this, note that
a line has constant speed parametrization γ(t) = a + bt for some constant vectors a and b,
79
80 7. GEODESICS

so γ̈ = 0. Thus all straight lines in the plane are geodesics, as are the rulings on any ruled
surface.
Example. Recall that a normal section of a surface S is the intersection C of S with
a plane Π, such that Π is perpendicular to S at each point of C. We saw in week 5, as
a consequence of Meusnier’s Theorem, that κg = 0 for a normal section. For example, it
follows that all great circles on a sphere are geodesics. The intersection of a generalized
cylinder with a plane perpendicular to the rulings is also a geodesic.

2. Geodesic Equations
The methods we saw of finding geodesics in the previous section do not allow us to find
geodesics on most surfaces. For that, we need some differential equations.
Theorem 2.1. A curve γ on a surface S is a geodesic if and only if, for any part of γ
contained in a surface patch σ as σ(u(t), v(t)), the equations
d 1
(Eu0 + F v 0 ) = (Eu u02 + 2Fu u0 v 0 + Gu v 02 ),
dt 2
d 1
(F u0 + Gv 0 ) = (Ev u02 + 2Fv u0 v 0 + Gv v 02 )
dt 2
are satisfied, where E du + 2F du dv + G dv 2 is the first fundamental form of σ.
2

The equations above are called the geodesic equations.

Proof. Since {σ u , σ v } is a basis of the tangent plane of σ, γ is a geodesic if and only
if γ̈ is perpendicular to both σ u and σ v . We have γ̇ = u0 σ u + v 0 σ v , so this is equivalent to
   
d 0 0 d 0 0
(u σ u + v σ v ) · σ u = 0 and (u σ u + v σ v ) · σ v = 0.
dt dt
We will show that these equations are equivalent to the geodesic equations.
The left side of the first equation is equal to
 
d 0 d dσ u
(u σ u + v σ v ) · σ u = ((u0 σ u + v 0 σ v ) · σ u ) − (u0 σ u + v 0 σ v ) ·
0
dt dt dt
d
= (Eu0 + F v 0 ) − (u0 σ u + v 0 σ v ) · (u0 σ uu + v 0 σ uv )
dt
d
= (Eu0 + F v 0 ) − (u02 (σ u · σ uu ) + u0 v 0 (σ u · σ uv + σ v · σ uu ) + v 02 (σ v · σ uv )).
dt
Now,
Eu = (σ u · σ u )u = σ uu · σ u + σ u · σ uu = 2σ u · σ uu ,
so σ u · σ uu = 2 Eu . Similarly, σ v · σ uv = 12 Gu . Also,
1

σ u · σ uv + σ v · σ uu = (σ u · σ v )u = Fu .
Substituting these values, we get
 
d 0 d 1
(u σ u + v σ v ) · σ u = (Eu0 + F v 0 ) − (Eu u02 + 2Fu u0 v 0 + Gu v 02 ).
0
dt dt 2
This shows that the first equation here is equivalent to the first geodesic equation. The
equivalence of the other equations is proven similarly. 
 2. GEODESIC EQUATIONS 81

Unfortunately, we have a system of nonlinear differential equations, so it’s usually im-

possible to solve them explicitly. But occasionally we are able to do so. Let’s see how to
solve them in the case of geodesics on a sphere (which, of course, we already knew how to
compute).
Example. The sphere in its usual parametrization is
σ(θ, φ) = (cos θ cos φ, cos θ sin φ, sin θ),
and the first fundamental form is
dθ2 + cos2 θ dφ2 .
Let us restrict to the case of unit-speed curves on the sphere, i.e. curves γ(t) = σ(θ(t), φ(t))
such that θ02 + φ02 cos2 θ = 1. If γ is a geodesic, then the second geodesic equation becomes
d 0
(φ cos2 θ) = 0,
dt
i.e. φ0 cos2 θ = C for some constant C. If C = 0, then φ0 = 0 (the case cos θ = 0 is not
contained in the surface patch), so φ is constant, and γ is part of a meridian.
Let’s now assume that φ0 6= 0. The unit-speed condition gives us

02 C2
θ =1− .
cos2 θ
Thus along the geodesic, we have
 2
θ02 C2 cos4 θ
   2 
dθ 2 cos θ
= 02 = 1 − = cos θ −1 .
dφ φ cos2 θ C2 C2
Thus we have
Z θ
dψ
±(φ − φ0 ) = p ,
θ0 cos ψ C −2 cos2 ψ − 1
for some constants φ0 and θ0 . If we let u = tan ψ, then the integral becomes
Z tan θ tan θ
du u tan θ tan θ0
±(φ−φ0 ) = √ = sin−1 √ = sin−1 √ −sin−1 √ .
tan θ0 C −2 − 1 − u2 C −2 − 1 tan θ0 C −2 − 1 C −2 − 1
We can absorb the second term on the right into the φ0 term on the left, so we have
tan θ
±(φ − φ0 ) = sin−1 √ ,
C −2 − 1
so
√
tan θ = ± C −2 − 1 sin(φ − φ0 ).
After some computations, we find that this implies each point γ(t) satisfies the equation
z = ax + by,
√ √
where a = ∓ C −2 − 1 sin φ0 and b = ± C −2 − 1 cos φ0 , since x = cos θ cos φ, y = cos θ sin φ,
and z = sin θ. This implies that γ is contained in the intersection of S2 with a plane passing
through the origin, i.e. γ is contained in a great circle.
82 7. GEODESICS

There is another set of equivalent differential equations for geodesics, in terms of the
Christoffel symbols. Recall that the Christoffel symbols are defined by
σ uu = Γ111 σ u + Γ211 σ v + LN,
σ uv = Γ112 σ u + Γ212 σ v + M N,
σ vv = Γ122 σ u + Γ222 σ v + N N.
Proposition 2.2. A curve γ on a surface S is a geodesic if and only if, for any part
γ(t) = σ(u(t), v(t)) of γ contained in a surface patch σ of S, we have
u00 + Γ111 u02 + 2Γ112 u0 v 0 + Γ122 v 02 = 0,
v 00 + Γ211 u02 + 2Γ212 u0 v 0 + Γ222 v 02 = 0.
This follows from the geodesic equations thanks to the conditions on a vector field being
parallel along a curve that we saw in week 5. This set of equations, while no easier to solve
explicitly, is in a form that allows us to say something about the existence and uniqueness
of solutions.
Proposition 2.3. Let p be a point on a surface S, and let t be a unit tangent vector to
S at p. Then there exists a unique unit-speed geodesic γ on S that passes through p and has
tangent vector t there.
This follows from the general theory of existence and uniqueness of solutions to (families
of) differential equations: we have a family of the form
u00 = f (u, v, u0 , v 0 ), v 00 = g(u, v, u0 , v 0 ).
Thus there exists a solution with every set of initial conditions (i.e. the values of u(t0 ) and
v(t0 ) and the tangent vectors, or equivalently u0 (t0 ) and v 0 (t0 )). In short, if we pick a point on
S and a direction, then there’s a unique geodesic stating at that point in the given direction.
This is a generalization to the obvious statement in R2 that there’s a unique line through a
given point in a given direction.

3. Geodesics on Surfaces of Revolution

We cannot typically solve the geodesic equations on a surface of revolution explicitly, but
we can still learn something about their behavior.
As usual, we parametrize a surface of revolution via the surface patch
σ(u, v) = (f (u) cos v, f (u) sin v, g(u)),
where f > 0 and f 02 + g 02 = 1. The first fundamental form is du2 + f (u)2 dv 2 . Thus the
geodesic equations are
df d
(3.1) u00 (t) = f (u(t)) v 0 (t)2 , (f (u(t))2 v 0 (t)) = 0.
du dt
We may restrict to the case of unit-speed geodesics, where we have u0 (t)2 +f (u(t))2 v 0 (t)2 = 1.
The following are easy consequences.
Proposition 3.1. Suppose we have a surface of revolution
σ(u, v) = (f (u) cos v, f (u) sin v, g(u)),
with the usual assumptions. Then
 3. GEODESICS ON SURFACES OF REVOLUTION 83

(1) Every meridian (i.e. a parameter curve where v is constant) is a geodesic.

df
(2) A parallel with u = u0 is a geodesic if and only if du = 0 at u = u0 .
This proposition only describes some of the geodesics on surfaces of revolution. The
following result helps to understand the rest of them.
Theorem 3.2 (Clairaut). Let γ be a unit-speed curve on a surface of revolution S, let
ρ : S → R be the distance of a point of S to the axis of rotation, and let ψ be the angle
between γ̇ and the meridians of S. If γ is a geodesic, then ρ sin ψ is constant along γ.
Conversely, if ρ sin ψ is constant along γ, and if no part of γ is part of some parallel of S,
then γ is a geodesic.
Proof. Parametrize S by the surface patch
σ(u, v) = (f (u) cos v, f (u) sin v, g(u)),
with the usual assumptions. Then we have ρ = f (u). Note that σ u and ρ−1 σ v are unit
vectors tangent to the meridians and parallels, respectively, and they are perpendicular
because F = 0. Since γ is unit-speed, we have
γ̇ = cos ψ σ u + ρ−1 sin ψ σ v .
Furthermore, γ̇ = u0 σ u + v 0 σ v . Since σ u and σ v are linearly independent, we have
u0 = cos ψ, v 0 = ρ−1 sin ψ.
Thus ρv 0 = sin ψ, so
ρ sin ψ = ρ2 v 0 .
By (3.1), this is a constant, say C, along the geodesic.
For the converse, suppose that ρ sin ψ is a constant C along γ. Then the second equation
in (3.1) holds, and we must show that the first equation holds as well. Since
sin ψ C
v0 = = 2,
ρ ρ
the unit-speed assumption on the reference curve gives
C2
u02 = 1 −
.
ρ2
Differentiating both sides with respect to t, we get
2C 2 0 2C 2 dρ 0
2u0 u00 = ρ = 3 u.
ρ3 ρ du
Thus  
0 dρ 02
00
u u −ρ v = 0.
du
If the bracketed term does not vanish at some point, say γ(t0 ) = σ(u0 , v0 ), then there is
some neighborhood where it does not vanish, say when |t − t0 | < ε. But then u0 = 0 for
|t − t0 | < ε, so γ coincides with the parallel u = u0 when |t − t0 | < ε, contrary to our
assumption. Thus the bracketed term must vanish everywhere on γ, i.e.
dρ
u00 = ρ v 02 ,
du
which shows that the first geodesic equation holds, so γ is a geodesic. 
84 7. GEODESICS

We can use Clairaut’s Theorem to determine the geodesics on the pseudosphere, originally
parametrized by
√
σ(u, v) = (eu cos v, eu sin v, 1 − e2u − cosh−1 e−u ).
Let us reparametrize by setting w = e−u , so that the new parametrization is
r !
1 1 1
σ
e (v, w) = cos v, sin v, 1 − 2 − cosh−1 w ,
w w w
and its first fundamental form is
dv 2 + dw2
.
w2
We must have w > 1 for σ e to be well defined and smooth.
If γ(t) = σ
e (v(t), w(t)) is a unit-speed geodesic, then the unit speed condition gives
v 02 (t) + w02 (t) = w(t)2 ,
and Clairaut’s Theorem gives
1 1 0
sin ψ = v (t) = C
w(t) w(t)2
for some constant C, because ρ = w1 . Thus we have v 0 = Cw2 . If C = 0, we get a meridian,
of the form v = D for some constant D. Now, assume that C 6= 0. Then we have
√
w0 = ±w 1 − C 2 w2 .
Thus along the geodesic, we have
dv v0 Cw
= 0 = ±√ ,
dw w 1 − C 2 w2
or
1
(v − v0 )2 + w2 = 2 ,
C
where v0 is a constant. In other words, the geodesics are the images under σ e of parts of
circles lying in the region w > 1. All these circles are centered on the v-axis, hence intersect
the v-axis perpendicularly. The meridians are the straight lines perpendicular to the v-axis.
Note that we cannot extend these geodesics forever, because they run into the circular
edge of the pseudosphere. This is because the pseudosphere ends at the image of the line
w = 1.
Clairaut’s Theorem can tell us about other behavior of the geodesics on a surface S of
revolution. For any p ∈ S and a constant C with |C| ≤ ρ, there are generally two geodesics
through p with ρ sin ψ = C, because v 0 = sinρ ψ is uniquely determined, whereas there are
q
2
two choices for u0 = ± 1 − Cρ2 . These two geodesics are reflections of each other, followed
by changing the parameter t of the geodesic to −t. Thus from now on, we’ll assume that
C > 0, which does not entail any loss of generality. Observe that this geodesic is confined
to the part of the surface of distance ≥ C from the axis. In the case of the pseudosphere,
this implies that geodesics corresponding to semicircles (as opposed to straight lines) must
go up the pseudosphere for a while before turning back, and they turn back once they have
reached a point where ρ = C. In other words, the geodesic is confined to the portion of S
where ρ ≥ C.
 4. GEODESICS AS SHORTEST PATHS 85

If all of S is of distance > C from the axis, then the geodesic crosses every parallel of S.
To see this, observe that otherwise u would be bounded above or below on S, say above. Let
u0 be the least upper bound of u on the geodesic, and let C + 2ε be the radius of the parallel
corresponding to u = u0 . For u sufficiently close to u0 , the radius of the corresponding
parallel is ≥ C + ε, and on the part of the geodesic lying in this region, we have
s  2
0 C
|u | ≥ 1 − > 0.
C +ε
But this implies that the geodesic crosses u0 , contradicting the assumption.
Thus the relevant part of S is exactly the part with a distance of ≥ C from the axis.
Next, let’s turn to geodesics on the hyperboloid of one sheet, obtained by rotating the
hyperbola x2 − z 2 = 1 for x > 0 in the xz-plane about the z-axis. The entire surface has
distance ≥ 1 from the axis. Thus if 0 ≤ C < 1, a geodesic with ρ sin ψ = C crosses every
parallel and so extends from z = −∞ to z = +∞. √
Suppose now that √ C > 1. Then the geodesic is confined to either the region z ≥ C2 − 1
2 + −
or the region z ≤ − C − 1, which are bounded by circles Γ and Γ respectively, of radius
C. Let p be a point on Γ+ , and consider the geodesic γ that passes through p and is tangent
to Γ+ there. Then ψ = π2 and ρ = C at p, so γ satisfies ρ sin ψ = C. Now, γ cannot be
contained in Γ+ because Γ+ isn’t a geodesic, so C must head into the region above Γ+ as it
leaves p. Furthermore, γ must be symmetric about p. Since u0 6= 0 in the region above Γ+ ,
the geodesic crosses every parallel above Γ+ and z → +∞ as t → ±∞. This describes all
the geodesics with C > 1.
Finally, suppose that C = 1. Let γ be a geodesic with C = 1 passing through a point p.
If p is a point with z = 0, on the waist of the hyperboloid, then ρ = 1 there, so sin ψ must
be 1, so ψ = π2 and γ is tangent to the z = 0 circle at p. Since this circle is a geodesic, in
this case γ is just the unit circle in the xy-plane.
Now suppose that C = 1 and p is not in the xy-plane, say with z > 0. Then 0 < ψ < π2 ,
so it leaves p, it gets arbitrarily close to the xy-plane, for if it always stayed in the region
with radius > 1 + ε with ε > 0, we would have
s  2
0 1
|u | ≥ 1 −
1+ε
everywhere along γ, which implies that γ crosses every parallel, which contradicts our as-
sumptions. So in this case, γ spirals around, getting closer and closer to the xy-plane.

4. Geodesics as Shortest Paths

We now explain another interpretation geodesics: that they are (at least locally) distance-
minimizing paths between two points. Suppose we have a surface S and two points p, q ∈ S.
Let γ be a unit-speed curve on S passing through p and q. If γ is the shortest path on S
from p to q, then the part of γ must be the shortest path between any two of its points
between p and q: if p0 , p0 are two points on γ on the path from p to q, and there were a
shorter path, then we could find a shorter path than γ between p and q by replacing the
part of γ between p0 and q0 with the shorter path.
Suppose γ is contained in a surface patch σ. To see if γ has a shorter length than any
other path in σ from p to q, we consider γ as part of a smooth family of curves on σ from
p to q: that is, we have a curve γ τ on σ for each τ ∈ (−δ, δ) for some δ > 0, such that
86 7. GEODESICS

• there exists some ε > 0 such that γ τ (t) is defined for all t ∈ (−ε, ε) and all τ ∈
(−δ, δ),
• for some a, b with −ε < a < b < ε, we have
γ τ (a) = p and γ τ (b) = q for all τ ∈ (−δ, δ),
• The map (−δ, δ) × (−ε, ε) given by (τ, t) 7→ γ τ (t) is smooth,
• γ 0 = γ.
The length of the part of γ τ between p and q is
Z b
L(τ ) = kγ̇ τ k dt.
a

Theorem 4.1. With the above notation, the unit-speed curve γ is a geodesic if and only
if
d
L(τ ) =0
dτ τ =0
for all families of curves γ τ with γ 0 = γ.
Note that, while we assume that γ = γ 0 is unit-speed, we cannot make the same as-
sumption about γ τ , or else that would imply that the length of each γ τ is the same as that
of γ.
Proof. We differentiate under the integral sign:
Z b
d d
L(τ ) = kγ̇ τ k dt
dτ dτ a
Z b√
d
= Eu02 + 2F u0 v 0 + Gv 02 dt
dτ a
Z b
∂ p
= g(τ, t)
a ∂τ
1 b
Z
1 ∂g
= p dt,
2 a g(τ, t) ∂τ
where
g(τ, t) = Eu02 + 2F u0 v 0 + Gv 02
and all unspecified derivatives are with respect to t. We have
0
 0 0
∂v 0

∂g ∂E 02 ∂F 0 0 ∂G 02 0 ∂u ∂u 0 0 ∂v
= u +2 uv + v + 2Eu + 2F v +u + 2Gv 0
∂τ ∂τ ∂τ ∂τ ∂τ ∂τ ∂τ ∂τ
     
∂u ∂v ∂u ∂v ∂u ∂v
= Eu + Ev u02 + 2 Fu + Fv u0 v 0 + Gu + Gv v 02
∂τ ∂τ ∂τ ∂τ ∂τ ∂τ
2
 2 2
 2
∂ u ∂ u ∂ v ∂ v
+ 2Eu0 + 2F + u0 + 2Gv 0
∂τ ∂t ∂τ ∂t ∂τ ∂t ∂τ ∂t
∂u ∂v
= (Eu u02 + 2Fu u0 v 0 + Gu v 02 ) + (Ev u02 + 2Fv u0 v 0 + Gv v 02 )
∂τ ∂τ
2 2
∂ u ∂ v
+ 2(Eu0 + F v 0 ) + 2(F u0 + Gv 0 ) .
∂τ ∂t ∂τ ∂t
 4. GEODESICS AS SHORTEST PATHS 87

The contribution to the integral dτd L(τ ) coming from the terms with second partial derivatives
is
Z b  2 2

1 0 0 ∂ u 0 0 ∂ v
√ (Eu + F v ) + (F u + Gv ) dt
a g ∂τ ∂t ∂τ ∂t
 t=b
1 0 0 ∂u 0 0 ∂v
= √ (Eu + F v ) + (F u + Gv )
g ∂τ ∂τ t=a
Z b     
∂ 1 0 0 ∂u ∂ 1 0 0 ∂v
− √ (Eu + F v ) + √ (F u + Gv ) dt.
a ∂t g ∂τ ∂t g ∂τ
Since γ τ (a) and γ τ (b) do not depend on τ (being always equal to p and q, respectively), we
have
∂γ τ
=0
∂τ
when t = a or t = b. Since
∂γ τ ∂u ∂v
= σu + σv ,
∂τ ∂τ ∂τ
we find that
∂u ∂v
= =0
∂τ ∂τ
when t = a or t = b. Thus the contribution to dτd L(τ ) coming from the part without second
derivatives is zero. Thus we have
Z b 
d ∂u ∂v
L(τ ) = U +V dt,
dτ a ∂τ ∂τ
where
 
1 02 0 0 02 d 1 0 0
U = √ (Eu u + 2Fu u v + Gu v ) − √ (Eu + F v ) ,
2 g dt g
 
1 02 0 0 02 d 1 0 0
V = √ (Ev u + 2Fv u v + Gv v ) − √ (F u + Gv ) .
2 g dt g
Now, γ 0 = γ is a unit-speed curve, so since kγ̇ τ k2 = g(τ, t), we have g(0, t) = 1 for all t.
Comparing with the geodesic equations, we find that if γ is a geodesic, then U = V = 0
when τ = 0, so
∂
L(τ ) = 0.
∂τ τ =0
For the converse, we must show that if
Z b 
∂u ∂v
U +V dt = 0
a ∂τ ∂τ
when τ = 0 for all families of curves γ τ , then U = V = 0 when τ = 0, since this shows
that γ satisfies the geodesic equations. Assume that this condition holds, and suppose
that, say, U 6= 0 when τ = 0. Thus there is some t0 ∈ (a, b) such that U (0, t0 ) 6= 0, say
U (0, t0 ) > 0. Thus there exists an η > 0 such that if |t − t0 | < η, then U (0, t) > 0. Let φ
be a smooth function that is positive when |t − t0 | < η and 0 if |t − t0 | ≥ η. Suppose that
γ(t) = σ(u(t), v(t)), and consider the family of curves γ τ (t) = σ(u(τ, t), v(τ, t)), where
u(τ, t) = u(t) + τ φ(t), v(τ, t) = v(t).
88 7. GEODESICS

∂u ∂v
Then ∂τ
= φ and = 0 for all τ and t, so
∂τ
Z b  Z t0 +η
∂u ∂v
0= U +V dt = U (0, t)φ(t) dt.
a ∂τ ∂τ τ =0 t0 −η

But the integrand on the right side is positive, as it is the product of two positive functions.
This contradicts the assumption that there exists a t0 such that U (0, t0 ) > 0. Other cases
are very similar. 
Note that Theorem 4.1 does not imply that a geodesic from p to q is the shortest path
between p and q. For instance, a great circle on a sphere is a geodesic, but if we take the
long arc of the great circle between two points, that’s clearly not distance-minimizing.
It is also possible for there not to be a shortest path between two points on a surface. For
example, there’s no shortest path between (−1, 0) and (1, 0) on the surface R2 \ {(0, 0)}. We
can find paths whose lengths are as close as desired to 2, but there is no path of length 2 since
we are not allowed to pass through the origin. However, if S is a closed and path-connected
subset of R3 , then there is always a distance-minimizing path between any two points. This
follows, with some work, from a compactness argument, but we shall omit the details here.
Another interesting property of closed subsets of R3 is that geodesics can be extended
indefinitely. This is a celebrated theorem called the Hopf–Rinow Theorem.

5. Geodesic Coordinates
The existence of geodesics on a surface S can be used to construct a useful atlas on S. Let
p ∈ S, and let γ be a unit-speed geodesic on S with parameter v and γ(0) = p. For any v in
the domain of γ, let γe v be a unit-speed geodesic with parameter u such that γ e v (0) = γ(v)
which is perpendicular to γ at γ(v). This curve γ e v is unique up the reparametrization
v
u 7→ −u. Define a surface patch σ(u, v) = γ e (u).
Proposition 5.1. With the above notation, there is an open subset U ⊆ R2 containing
(0, 0) such that σ : U → R3 is an allowable surface patch of S. The first fundamental form
of σ is
du2 + G(u, v) dv 2 ,
where G is a smooth function on U such that G(0, v) = 1 and Gu (0, v) = 0 whenever
(0, v) ∈ U .
You will prove Proposition 5.1 in problem 13.
We call the patch σ a geodesic patch, and u and v are called geodesic coordinates.
Example. Let p be a point on the equator of S2 , let γ be the equator with parameter
e φ be the meridian parametrized by the latitude θ and passing
being the longitude φ, and let γ
through the point on the equator with longitude φ. The corresponding geodesic patch is the
usual latitude-longitude patch, and the first fundamental form is dθ2 + cos2 θ dφ2 .

6. Problems
(1) A regular curve γ with nowhere vanishing curvature on a surface S is called a pre-
geodesic on S if it is a reparametrization of a geodesic. Show that a curve γ is a
pre-geodesic if and only if γ̈ · (N × γ̇) = 0 everywhere on γ.
 6. PROBLEMS 89

(2) Let γ be a curve in R3 , and consider the tube of radius a > 0 around γ, given by
σ(s, θ) = γ(s) + a(cos θ n(s) + sin θ b(s)).
Show that the parameter curves where s is fixed are circular geodesics on σ.
(3) Show that a geodesic with nowhere vanishing curvature is a plane curve if and only
if it is a line of curvature. (Recall that a line of curvature is a curve whose tangent
vector is tangent to a principal vector.)
(4) Let S1 and S2 be two surfaces intersecting at a curve C. Let γ be a unit-speed
parametrization of C.
(a) Show that if γ is a geodesic on both S1 and S2 , and the curvature on γ is
nowhere zero, then S1 and S2 have the same tangent plane at each point of C.
(b) Give an example of where the situation in part (i) happens.
(c) Show that if S1 and S2 intersect orthogonally at each point of C (i.e. their
normal vectors are orthogonal), then γ is a geodesic on S1 if and only if Ṅ2 is
parallel to N1 at each point of C (where N1 and N2 are the unit normals of S1
and S2 , respectively).
(d) Show that, in the situation of part (iii), γ is a geodesic on both S1 and S2 if
and only if C is part of a straight line.
(5) Show that if p and q are distinct points on the unit cylinder x2 + y 2 = 1 in R3 , then
there are either two or infinitely many geodesics on the cylinder with endpoints p
and q. When are there infinitely many?
(6) Find all the geodesics on the circular cone x2 + y 2 = z 2 and z > 0.
(7) Let γ(t) be a unit-speed curve on the helicoid
σ(u, v) = (u cos v, u sin v, v).
(a) Show that
u02 + (1 + u2 )v 02 = 1.
(b) Show that if γ is a geodesic on σ, then
a
v0 = ,
1 + u2
where a is a constant. Describe the geodesics corresponding to a = 0 and a = 1.
(c) Suppose that a geodesic γ on σ intersects a ruling at a point p at a distance D >
0 from the z-axis, and that the angle between γ and the ruling at p is α, where
0 < α < π2 . Show that the geodesic intersects the z-axis if D > cot α, but that
p
if D < cot α, then its smallest distance from the z-axis is D2 sin2 α − cos2 α.
Find the equation of the geodesic if D = cot α.
(8) An ellipsoid of rotation is an ellipsoid with two equal axes, say
x2 y 2 z 2
+ + = 1.
a2 a2 c 2
Describe as well as you can the geodesics on an ellipsoid of rotation.
(9) Describe as well as you can the geodesics on a torus.
1
(10) Show that a geodesic on a pseudosphere intersects itself if and only if 0 < C < √1+π 2.

(11) Find a parametrization of the pseudosphere by the unit disk such that geodesics
on the pseudosphere correspond to segments of straight lines and circles in the pa-
rameter plane that intersect the boundary of the disk orthogonally. Find another
90 7. GEODESICS

parametrization of the pseudosphere such that all geodesics on the pseudosphere cor-
respond to straight lines in the parameter plane. The existence of such parametriza-
tions is a very special property of the pseudosphere that few other surfaces have.
(12) Let p and q be two points on S2 with p 6= ±q. Is the long arc of the great circle
connecting p and q a local minimum of the length function?
(13) Prove Proposition 5.1.
(14) Redo §5 for polar geodesic patches.
 CHAPTER 8

The Theorema Egregium

1. The Gauß and Codazzi–Mainardi Equations

Are there any relations between the first and second fundamental forms of a surface
patch? Recall that the plane and the cylinder, with a suitable parametrization, have the
same first fundamental forms but different second fundamental forms, so it cannot be possible
to determine the second fundamental form from the first fundamental form. But there could
still be some relations between the two, and indeed there are. Recall the Christoffel symbols,
in terms of the first fundamental form:
GEu − 2F Fu + F Ev
Γ111 = ,
2(EG − F 2 )
2EFu − EEv − F Eu
Γ211 = ,
2(EG − F 2 )
GEv − F Gu
Γ112 = ,
2(EG − F 2 )
EGu − F Ev
Γ212 = ,
2(EG − F 2 )
2GFv − GGu − F Gv
Γ122 = ,
2(EG − F 2 )
EGv − 2F Fv + F Gu
Γ222 = .
2(EG − F 2 )
These are the coefficients such that
σ uu = Γ111 σ u + Γ211 σ v + LN,
σ uv = Γ112 σ u + Γ212 σ v + M N,
σ vv = Γ122 σ u + Γ222 σ v + N N.
The first relations between the fundamental forms are called the Codazzi–Mainardi equa-
tions.
Proposition 1.1 (Codazzi–Mainardi Equations). Let E du2 + 2F du dv + G dv 2 and
L du2 +2M du dv +N dv 2 be the first and second fundamental forms, respectively, of a surface
patch σ of a surface S. Then we have
Lv − Mu = LΓ112 + M (Γ212 − Γ111 ) − N Γ211 ,
Mv − Nu = LΓ122 + M (Γ222 − Γ112 ) − N Γ212 .
The other relations are the Gauß equations.

91
92 8. THE THEOREMA EGREGIUM

Proposition 1.2 (Gauß Equations). If K is the Gaussian curvature of the surface patch
σ(u, v), then we have
EK = (Γ211 )v − (Γ212 )u + Γ111 Γ212 + Γ211 Γ222 − Γ112 Γ211 − (Γ212 )2 ,
F K = (Γ112 )u − (Γ111 )v + Γ212 Γ112 − Γ211 Γ122 ,
F K = (Γ212 )v − (Γ222 )u + Γ112 Γ212 − Γ122 Γ211 ,
GK = (Γ122 )u − (Γ112 )v + Γ122 Γ111 + Γ222 Γ112 − (Γ112 )2 − Γ212 Γ122 .
We’ll prove both of these propositions together.
Proof. We have (σ uu )v = (σ uv )u , which implies that
(Γ111 σ u + Γ211 σ v + LN)v = (Γ112 σ u + Γ212 σ v + M N)u .
Expanding out this last set of derivatives, we have
 1
∂Γ11 ∂Γ112
 2
∂Γ11 ∂Γ212
 
− σu + − σ v + (Lv − Mu )N
∂v ∂u ∂v ∂u
= Γ112 σ uu + (Γ212 − Γ111 )σ uv − Γ211 σ vv − LNv + M Nu
= Γ112 (Γ111 σ u + Γ211 σ v + LN) + (Γ212 − Γ111 )(Γ112 σ u + Γ212 σ v + M N)
− Γ211 (Γ122 σ u + Γ222 σ v + N N) − LNv + M Nu .
Now, Nu and Nv are perpendicular to N, and so are linear combinations of σ u and σ v . Thus
we may equate N components on both sides to get
Lv − Mu = LΓ112 + M (Γ212 − Γ111 ) − N Γ211 .
This is the first Codazzi–Mainardi equation, and the proof of the other one is similar, starting
with the equation (σ uv )v = (σ vv )u .
For the Gauß equations, we express Nu and Nv in terms of σ u and σ v using the fact
that  
−1 a c
FI FII = ,
b d
where
−Nu = aσ u + bσ v , −Nv = cσ u + dσ v ,
a result from week 6. Equating coefficients of σ u above, we get
(Γ111 )v − (Γ112 )u = Γ212 Γ112 − Γ211 Γ122 − Lc + M a,
where
M F − LG NF − MG
a= , c= .
EG − F 2 EG − F 2
Thus we have
−L(N F − M G) + M (M F − LG) F (LN − M 2 )
−Lc + M a = = − = −F K
EG − F 2 EG − F 2
thanks to the formula
LN − M 2
K=
EG − F 2
from week 6. This implies the second of the Gauß equations. The other three are proven
similarly. 
 2. THE THEOREMA EGREGIUM 93

The Codazzi–Mainardi and Gauß equations are in fact the only relations between the two
fundamental forms, in the following sense: given any E, F, G, L, M, N satisfying the two sets
of equations, there exists a surface patch whose first fundamental form is E du2 + 2F du dv +
G dv 2 and whose second fundamental form is L du2 + 2M du dv + N dv 2 . Furthermore, any
two surface patches with the same two forms are isometric via an orientation-preserving
isometry. We will not prove these facts here: the uniqueness part is not so complicated,
as it can be deduced from uniqueness of solutions to ODEs. The existence part is more
complicated, as it relies on existence theorems for families of partial differential equations.
However, this is much more subtle, as partial differential equations do not always admit
solutions; see problems 5–7.
2. The Theorema Egregium
Let’s take a closer look at Proposition 1.2. The Christoffel symbols on the right side
only depend on the first fundamental form, not on the second fundamental form. It follows
that K only depends on the first fundamental form. Now, recall what the first fundamental
form tells us: it’s about lengths of curves on the surface. Thus it is invariant under local
isometries. We have now deduced Gauß’s Theorema Egregium, or Remarkable Theorem:
Theorem 2.1 (Theorema Egregium). The Gaussian curvature of a surface is preserved
by local isometries.
Let’s start with an amusing application of the Theorema Egregium to everyday life before
we get to the deeper stuff. If you pick up a piece of pizza from the outer edge, then the tip
may flop over, leaving toppings all over the place. However, if you gently roll the edge up,
then you have forced one of the principal curvatures to be nonzero. Since bending the slice
is an isometry, the Gaussian curvature must remain the same as it was before (i.e. zero),
so the other principal curvature, in the orthogonal direction, must be zero. It follows that
after rolling the edge, the tip cannot flop over, as then the other principal curvature would
be nonzero, so the Gaussian curvature would be nonzero.
The Gauß equations actually allow us to get a formula for the Gaussian curvature in
terms of the first fundamental form. Here it is:
Corollary 2.2. We have
 1   
− 2 Evv + Fuv − 12 Guu 1
E
2 u
Fu − 12 Ev 0 1
2
E v
1
2
G u
det  Fv − 12 Gu E F  − det  1 Ev E
2
F 
1 1
G
2 v
F G G
2 u
F G
K= 2 2
.
(EG − F )
Naturally, checking this will be an annoying computation, so we omit it. Here are some
nicer special cases:
Corollary 2.3.
(1) If F = 0, we have
 
1 ∂ Gu ∂ Ev
K=− √ √ + √ .
2 EG ∂u EG ∂v EG
(2) If E = 1 and F = 0, we have
√
1 ∂2 G
K = −√ .
G ∂u2
94 8. THE THEOREMA EGREGIUM

Example. Consider the surface of revolution σ(u, v) = (f (u) cos v, f (u) sin v, g(u)), with
the usual conditions f > 0 and f 02 + g 02 = 1. Then we have E = 1, F = 0, and G = f (u)2 ,
so we have √
1 ∂2 G f 00
K=− √ =− ,
G ∂u2 f
the same answer that we got before.
One other important consequence of the Theorema Egregium is that any map of any
region on the Earth’s surface must distort distance, because the sphere is not locally iso-
metric to the plane. Cartographers have come up with various projects to preserve some
relevant features, but it’s not possible to have everything. In the most famous projection,
the Mercator projection, angles are preserved (i.e. the projection is conformal), but distances
and areas are distorted. In particular, regions near the poles are drastically stretched out. It
is also possible to draw a map that preserves areas but not angles. In fact, we have already
seen this projection: it’s Archimedes’s projection from the sphere to the cylinder, followed
by unwrapping the cylinder into a rectangle. But we cannot preserve both angles and area
simultaneously.

3. Surfaces of Constant Gaussian Curvature

Theorem 3.1. Any point of a surface of constant Gaussian curvature is contained in a
patch that is isometric to an open subset of a plane, sphere, or pseudosphere.
Proof. Let p be a point on a surface S of constant Gaussian curvature K. After scaling,
we may restrict to the cases K = 0, K = 1, and K = −1. √
Let σ be a geodesic patch of S with σ(u, v) = p. Let g = G. Then the first fundamental
form is du2 + g(u, v)2 dv 2 . We have
∂ 2g
+ Kg = 0.
∂u2
We have g(0, v) = 1 and gu (0, v) = 0 by the properties of geodesic patches.
Suppose that K = 0. Then we have g(u, v) = αu + β for some α, β that are smooth
functions of v alone. The initial conditions give α = 0 and β = 1 for all v, so g = 1, and the
first fundamental form is du2 + dv 2 . This is the same as the first fundamental form for the
plane, so σ is isometric to an open subset of the plane.
If K = 1, then we get g(u, v) = α cos u + β sin u, where α and β only depend on v.
The boundary conditions give α = 1 and β = 0, so the first fundamental form is du2 +
cos2 u dv 2 . This is the first fundamental form of S2 with u and v being latitude and longitude,
respectively. Thus σ is isometric to an open subset of the sphere.
If K = −1, then we find in the same way that the first fundamental form of σ is du2 +
cosh2 u dv 2 . This is not the usual first fundamental form for the pseudosphere. However, we
can reparametrize it by setting V = ev tanh u and W = ev sech u. Then the first fundamental
form becomes
dV 2 + dW 2
.
W2
This is the first fundamental form of one parametrization of the pseudosphere. 
We can be more precise in the case of surfaces of constant negative curvature.
 3. SURFACES OF CONSTANT GAUSSIAN CURVATURE 95

Proposition 3.2. Let S be a surface of constant Gaussian curvature −1. If p ∈ S,

there is a surface patch of S containing p whose first and second fundamental forms are
du2 + 2 cos θ du dv + dv 2 and 2 sin θ du dv,
respectively, where θ(u, v) is a smooth function such that 0 < θ < π for all u, v.
Proof. Since K < 0, S has no umbilics. Thus there is a surface patch whose first and
second fundamental forms are
E dU 2 + G dV 2 and L dU 2 + N dV 2 ,
respectively, as we saw in week 6. The principal curvatures are then
L N
κ1 = , κ2 = .
E G
Since κ1 κ2 = −1, there are two cases:
(1) κ1 > 0 and κ2 < 0 for all U, V ,
(2) κ1 < 0 and κ2 > 0 for all U, V .
We’ll focus on case (1); some minor changes can be used to take care of case (2).
There is a smooth function ω(U, V ) such that 0 < ω < π2 and
κ1 = tan ω, κ2 = − cot ω.
The first Codazzi–Mainardi equation is
 
1 L N
LV = EV + .
2 E G
Thus we have
1
(E tan ω)V = EV (tan ω − cot ω),
2
i.e.
EV
ωV = − cot ω.
2E
Thus we have
   
cos ω ωV sin ω EV cos ω sin ω EV
√ =− √ − =−√ ωV + cot ω = 0.
E V E 2E 3/2 E 2E
cos
√ ω
Thus E
only depends on U , i.e. there is some positive smooth function e(U ) such that
E = e(U ) cos2 ω.
Similarly, the second Codazzi–Mainardi equation implies that there is a positive smooth
function g(V ) such that
G = g(V ) sin2 ω.
Now let U e (U ) and Ve (V ) be smooth functions such that

dU
e p dVe p
= e(U ), = g(V ),
dU dV
and let σ
e (U
e , Ve ) be the reparametrization of σ(U, V ) in terms of U
e , Ve . Then the first fun-
damental form of σ e is
cos2 ω dU
e 2 + sin2 ω dVe 2 ,
96 8. THE THEOREMA EGREGIUM

and the second fundamental form is

L e2 N e 2 e 2 − dVe 2 ).
dU + dV = sin ω cos ω(dU
e g
Letting u = U +2 V and v = U −2 V gives a reparametrization of σ e whose first and second
e e e e

fundamental forms are as stated in the proposition, with θ = 2ω. 

The function θ(u, v) cannot just be anything. So far, we have only used the Codazzi–
Mainardi equations, so the Gauß equations must also be satisfied. These equations are all
equivalent to each other, so we only have to check one of them. The Christoffel symbols are
Γ111 = θu cot θ,
Γ211 = −θu csc θ,
Γ112 = 0,
Γ212 = 0,
Γ122 = −θv csc θ,
Γ222 = θv cot θ.
Thus the first Gauß equation becomes
−1 = (−θu csc θ)v − θu θv csc θ cot θ,
i.e.
θuv = sin θ.
This is called the sine-Gordon equation.
Theorem 3.3. Let θ(u, v) be a smooth solution of the sine-Gordon equation such that
0 < θ < π. Then there exists a surface, unique up to an orientation-preserving isometry of
R3 , of constant Gaussian curvature −1 and first and second fundamental forms as in Propo-
sition 3.2. Conversely, any surface of constant Gaussian curvature −1 has a parametrization
with first and second fundamental forms as in Proposition 3.2, in which θ is a solution of
the sine-Gordon equation.
We can also handle the case of constant positive curvature, when the surface is compact.
Theorem 3.4. Every connected compact surface with constant Gaussian curvature is a
sphere.
Note that this forces the curvature to be positive. We begin with a Lemma.
Lemma 3.5. Let σ : U → R3 be a surface patch containing a point p that is not an
umbilic. Let κ1 ≥ κ2 be the principal curvatures of σ, and suppose that κ1 has a local
maximum at p, while κ2 has a local minimum there. Then the Gaussian curvature of σ at
p is nonpositive.
Proof. Since p is not an umbilic, κ1 > κ2 at p. By shrinking U if necessary, we
may therefore suppose that κ1 > κ2 on all of U . Now, suppose that the first and second
fundamental forms of σ are
E du2 + G dv 2 and L du2 + N dv 2 ,
 4. GEODESIC MAPPINGS 97

respectively. Then by problem 4 we have

2E 2G
Ev = − (κ1 )v , Gu = (κ2 )u ,
κ1 − κ2 κ1 − κ2
and the Gaussian curvature is
 
1 ∂ Gu ∂ Ev
K=− √ √ + √ .
2 EG ∂u EG ∂v EG
Since p is a local extremum of κ1 and κ2 , we have (κ1 )v = (κ2 )u = 0 at p, which implies
that Ev = Gu = 0 at p. Thus at p, we have
 
1 1 2G 2E
K=− (Guu + Evv ) = − (κ2 )uu − (κ1 )vv .
2EG 2EG κ1 − κ2 κ1 − κ2
Since κ1 has a local maximum at p, we have (κ1 )vv ≤ 0 there, and since κ2 has a local
minimum at p, we have (κ2 )uu ≥ 0. Thus K ≤ 0 at p. 
Proof of Theorem 3.4. Let J = (κ1 − κ2 )2 , where κ1 and κ2 are the principal cur-
vatures. Note that this does not depend on the choice of which principal curvature is κ1
and which is κ2 , or their signs (if we change both of them together), so J is a well-defined
function on S, independent of any choices. We will prove that on a connected compact
surface with constant Gaussian curvature, J is identically zero, so every point is an umbilic.
It then follows from a result from week 6 that since K > 0, S is an open subset of a sphere,
say S. In fact, S must be all of S because it is compact, hence a closed subset of S. But it
is also an open subset of S, and the only nonempty subset of S that is both open and closed
is all of S.
Suppose that J is not identically zero on S. Since S is compact, J attains a maximum
value at some point p ∈ S, and we have J(p) > 0. Choose a patch σ : U → R3 of
S containing p, and let κ1 , κ2 be its principal curvatures. Since κ1 κ2 = K > 0, we can
assume by reparametrization if necessary that κ1 , κ2 > 0. Suppose that κ1 > κ2 and p.
Then, by shrinking U if necessary, we can assume that κ1 > κ2 everywhere on U . Since
K > 0 is constant, the function (x − Kx )2 increases with x as long as x > Kx > 0. Since
κ1 > κ2 = κK1 > 0, the function is increasing at x = κ1 , so κ1 must have a local maximum at
p, so κ2 = κK1 has a local minimum there. Thus K ≤ 0 at p by Lemma 3.5. Thus contradicts
our assumption that K > 0. 

4. Geodesic Mappings
Definition 4.1. A pre-geodesic is a reparametrization of a geodesic.
Given two surfaces S and S, e a smooth map F : S → S, e and a curve γ on S, we get a
curve F ◦ γ on S,e defined by (F ◦ γ)(t) = F (γ(t)). If γ is a geodesic or a pre-geodesic, we
can ask if its image F ◦ γ is also a pre-geodesic.
Definition 4.2. Let S and Se be surfaces. A local diffeomorphism F : S → Se is said to
be geodesic if it takes every pre-geodesic on S to a pre-geodesic on S.
e

There are some obvious examples of geodesic local diffeomorphisms, given in the following
proposition.
Proposition 4.3. The following are geodesic local diffeomorphisms:
98 8. THE THEOREMA EGREGIUM

(1) Local isometries.

(2) Dilations.
(3) Compositions of geodesic local diffeomorphisms.
The first two types take geodesics to geodesics. But there are other geodesic local diffeo-
morphisms that only take geodesics to pre-geodesics.
Example. Let S be the (open) lower hemisphere of the unit sphere centered at the
origin, and let Se be the plane z = −1. We let F : S → Se be the projection from the origin.
We claim that F is a geodesic local diffeomorphism. One way to see this is to note that any
(pre-)geodesic on S is the intersection of S with a plane P passing through the center of the
sphere. Its image under F is the intersection of P with Se i.e. a line, which is a pre-geodesic
on S.
e However, the length of a geodesic γ on S is finite (in fact, π), whereas the length of
its image on Se is infinite, so the parametrization of F ◦ γ cannot be constant-speed, hence
not a geodesic but only a pre-geodesic.
For the local diffeomorphism F in the example, its inverse F −1 provides a parametrization
of the hemisphere with the property that geodesics in the hemisphere correspond to straight
lines in the parameter plane. The following theorem of Beltrami tells us which surfaces have
the property that there exists such a parametrization.
Theorem 4.4 (Beltrami). Let S be a connected surface. If there is a geodesic local
diffeomorphism from S to a plane, then S has constant Gaussian curvature. Conversely,
if S has constant Gaussian curvature, then for any point p ∈ S, there is a surface patch
σ : U → S of S such that p ∈ σ(U ) and σ −1 : σ(U ) → U is a geodesic diffeomorphism.
Proof. Suppose that σ is a geodesic local diffeomorphism from an open subset of the
plane to S, and let u and v be parameters. Consider a line of the form v = λu + µ, where λ
and µ are constants. Let t 7→ (u(t), v(t)) be a parametrization such that γ(t) = σ(u(t), v(t))
is a unit-speed geodesic on S. Then by a theorem from week 7, we have
u00 + (Γ111 + 2λΓ112 + λ2 Γ122 )u02 = 0,
λu00 + (Γ211 + 2λΓ212 + λ2 Γ222 )u02 = 0.
Eliminating u00 and noting that u0 cannot be zero (or else v 0 = 0 as well, contradicting the
assumption that γ is unit speed), we get
λ(Γ111 + 2λΓ112 + λ2 Γ122 ) = Γ211 + 2λΓ212 + λ2 Γ222 .
Since every line in the uv-plane corresponds to a geodesic γ, this equation must hold for all
λ. Thus we have
Γ211 = Γ122 = 0, Γ111 = 2Γ212 , Γ222 = 2Γ112 .
Thus the Gauß equations become
KE = (Γ212 )2 − (Γ212 )u ,
KF = (Γ112 )u − 2(Γ212 )v + Γ212 Γ112 ,
KF = (Γ212 )v − 2(Γ112 )u + Γ112 Γ212 ,
KG = (Γ112 )2 − (Γ112 )v .
Subtracting the middle two equations gives
(Γ112 )u = (Γ212 )v ,
 4. GEODESIC MAPPINGS 99

so the four equations become

(4.1) KE = (Γ212 )2 − (Γ212 )u ,

(4.2) KF = −(Γ212 )v + Γ112 Γ212 ,
(4.3) KF = −(Γ112 )u + Γ112 Γ212 ,
(4.4) KG = (Γ112 )2 − (Γ112 )v .

Differentiating (4.1) with respect to v and (4.2) with respect to u and equating (Γ212 )uv =
(Γ212 )vu gives

EKv − F Ku = K(Fu − Ev ) + 2Γ212 (Γ212 )v − Γ212 (Γ112 )u − Γ112 (Γ212 )u

= K(Fu − Ev ) + 2Γ212 (Γ112 Γ212 − KF ) − Γ212 (Γ112 Γ212 − KF ) − Γ112 ((Γ212 )2 − KE)
= K(Fu − Ev ) + K(EΓ112 − F Γ212 ),

where we used (4.1)–(4.3) to go from the first line to the second.

Using the definition of the Christoffel symbols, we find that
(EG + F 2 )Ev − 2EF Gu
EΓ112 − F Γ212 = .
2(EG − F 2 )
Since Γ211 = 0 and Γ111 = 2Γ212 , we get

EEv = 2EFu − F Eu , 3F Ev = 2EGu + 2F Fu − EGu

from the formula for Christoffel symbols. Thus we have

(EG + F 2 )Ev − 2EF Gu = 2(EG − F 2 )Ev + (3F 2 − EG)Ev − 2EF Gu

= 2(EG − F 2 )Ev + F (2EGu + 2F Fu − GEu ) − G(2EFu − F Eu ) − 2EF Gu
= 2(EG − F 2 )(Ev − Fu ).

This gives
EΓ112 − F Γ212 = Ev − Fu
and
EKv − F Ku = 0.
A similar calculation shows that
F Kv − GKu = 0.
Thus we have
(EG − F 2 )Kv = GF Ku − F GKu = 0,
so Kv = 0. Similarly, Ku = 0. Thus K is constant.
For the converse, if S has constant Gaussian curvature, then each point of S is contained
in a surface patch that is isometric to an open subset of a plane, sphere, or pseudosphere.
We have already dealt with the plane and sphere, and the case of the pseudosphere was
handled in a problem from week 7. 
100 8. THE THEOREMA EGREGIUM

5. Problems
(1) A surface patch has first and second fundamental forms
cos2 v du2 + dv 2 and − cos2 v du2 − dv 2 ,
respectively. Show that this surface is an open subset of a sphere of radius 1. Write
down a parametrization of S2 with these fundamental forms.
(2) Show that there is no surface patch with first and second fundamental forms
du2 + cos2 u dv 2 and cos2 u du2 + dv 2 .
(3) Suppose that a surface patch σ(v, w) has first and second fundamental forms
dv 2 + dw2
and L dv 2 + N dw2 ,
w2
where w > 0.
(a) Show that L and N do not depend on v.
(b) Show that LN = − w14 .
(c) Show that
dL
Lw5 = 1 − L2 w 4 .
dw
(d) Solve this equation for L and conclude that σ cannot be defined in the entire
half-plane w > 0.
(4) Suppose that the first and second fundamental forms of σ are
E du2 + G dv 2 and L du2 + N dv 2 ,
respectively.
(a) Write down the Codazzi–Mainiardi equations for this surface patch.
(b) Show that κ1 = EL and κ2 = N G
are the principal curvatures.
(c) Show that
Ev Gu
(κ1 )v = (κ2 − κ1 ) and (κ2 )u = (κ1 − κ2 ).
2E 2G
(5) A partial differential equation of the form
∂ 2u ∂ 2u ∂u
2
− 2 = f (x, t), u(x, 0) = g(x), (x, 0) = h(x)
∂t ∂x ∂t
is called a wave equation. Show that if u : U = R × (0, ∞) → R satisfies the
wave equation, then it also satisfies the weak wave equation: if φ : U → R is
an infinitely differentiable function with compact support (i.e. it is zero outside a
bounded region), then
Z  2
∂ φ ∂ 2φ
Z Z   
∂φ
φf dx dt + φ(x, 0)h(x) − (x, 0)g(x) dx = − 2 u dx dt.
U R ∂t U ∂t2 ∂x
A solution to the weak wave equation is called a weak solution to the wave equation,
whereas a solution to the original wave equation is called a classical solution. Note
that a classical solution must be twice differentiable, whereas a weak solution only
has to be integrable!
(6) Find a weak solution of the wave equation with f (x, t) = 0, g(x) = |x|, and h(x) = 0
that is not a classical solution.
 5. PROBLEMS 101

(7) Unlike ordinary differential equation equations, partial differential equations do not
necessarily have solutions. Prove Lewy’s Theorem: If u(t, z) is a function on R × C
satisfying a differential equation
∂u ∂u
− iz = φ0 (t)
∂ z̄ ∂t
for some continuously differentiable function φ. Then φ must be real-analytic (i.e.
have a convergent power series) in a neighborhood of the origin. Thus φ is continu-
ously differentiable but not real analytic, then the partial differential equation has
no solutions. One can show that there are no weak solutions either.
(8) Let S be a surface patch, and let p ∈ S. The geodesic circle of radius r is the set
of points in S at a distance of r from p. Let Cr and Ar denote the circumference
and area, respectively, of the geodesic circle of radius r centered at p.
(a) Show that
 
K(p) 2
Cr = 2πr 1 − r + remainder ,
6
where remainder
r2
tends to 0 as r → 0.
(b) Show that
 
2 K(p) 2
Ar = πr 1 − r + remainder ,
12
where remainder
r2
tends to 0 as r → 0.
(9) One parametrization of a Möbius strip is given by the surface patch
    
θ θ θ
σ(t, θ) = 1 − t sin cos θ, 1 − t sin sin θ, t cos .
2 2 2
Show that this Möbius strip cannot be obtained by gluing together a piece of paper
with a half-twist.
(10) Show that the surfaces
σ(u, v) = (u cos v, u sin v, log u), σ
e (u, v) = (u cos v, u sin v, v)
e ◦σ −1 is not an isometry.
have equal Gaussian curvatures at (u, v), but that the map σ
Conclude that the converse to the Theorema Egregium is false.
(11) Show that a compact surface with positive Gaussian curvature and constant mean
curvature is a sphere.
(12) Show that the solution
θ(u, v) = 2 arctan(sinh(u − v + c))
of the sine-Gordon equation corresponds to a pseudosphere.
(13) Show that a local diffeomorphism between surfaces taking unit-speed geodesics to
unit-speed geodesics is a local isometry.
(14) A composition of a dilation with a local isometry takes geodesics to geodesics. Are
there any other local diffeomorphisms between surfaces that also take all geodesics
to geodesics?
 CHAPTER 9

The Gauß–Bonnet Theorem

1. Introduction to the Gauß–Bonnet Theorem

Here we are at last, at the most important and beautiful theorem in differential geometry:
the Gauß–Bonnet Theorem! This theorem relates R the curvature of a surface to its global
topology: it says if S is a compact surface, then S K dA is a topological invariant, related
to the Euler characteristic of the surface, which is a measure of the number of holes in
the surface (e.g. 0 for a sphere, 1 for a torus, 2 for a double torus, and so forth). If you
have studied complex analysis, you should consider the Gauß–Bonnet Theorem to be the
differential geometry analogue of the Cauchy integral formula.
Before we get to the full Gauß–Bonnet Theorem, we have to do some simpler cases first.

2. The Gauß–Bonnet Theorem for Simple Closed Curves

Definition 2.1. A curve γ : (−∞, ∞) → R2 is said to be a simple closed curve with
period T if γ(t1 ) = γ(t2 ) if and only if t1 − t2 is a multiple of T .
Definition 2.2. A curve γ(t) = σ(u(t), v(t)) on a surface patch σ : U → R3 is said
to be a simple closed curve with period T if π(t) = (u(t), v(t)) is a simple closed curve in
R2 with period T such that the region int(π) enclosed by π is entirely contained in U . The
curve γ is said to be positively oriented if π is positively oriented. The image of int(π) under
σ is called the interior of γ and is denoted int(γ).
Theorem 2.3 (Gauß–Bonnet Theorem for simple closed curves). Let γ(s) be a unit-speed
simple closed curve on a surface patch σ of length `(γ), and assume that γ is positively
oriented. Then
Z `(γ) Z
κg ds = 2π − K dAσ ,
0 int(γ)
where κg is the geodesic curvature of γ, K is the Gaussian curvature, and dAσ is the area
element of σ.
We call the integral on the right side the total curvature of the region int(γ).
Let us slowly build up the proof of the Gauß–Bonnet Theorem for simple closed curves.
We first compute the geodesic curvature. Let {e0 , e00 } be a smoothly varying orthonormal
basis of the tangent plane of σ at each point. Then {e0 , e00 , N} is an orthonormal basis of
R3 , and we shall assume that it is a right-handed basis, i.e. N = e0 × e00 . Let θ(s) be the
oriented angle between e0 and γ̇. Thus we have
γ̇ = cos θ e0 + sin θ e00 ,
and
N × γ̇ = − sin θ e0 + cos θ e00
103
104 9. THE GAUSS–BONNET THEOREM

and
γ̈ = cos θ ė0 + sin θ ė00 + θ0 (− sin θ e0 + cos θ e00 ).
Thus the geodesic curvature is
κg = (N × γ̇) · γ̈
= θ0 (− sin θ e0 + cos θ e00 ) · (− sin θ e0 + cos θ e00 ) + (− sin θ e0 + cos θ e00 ) · (cos θ ė0 + sin θ ė00 )
= θ0 + cos2 θ (ė0 · e00 ) − sin2 θ (ė00 · e0 ) + sin θ cos θ (ė00 · e00 − ė0 · e0 )
= θ0 − e0 · ė00 ,
using the fact that e0 and e00 are perpendicular unit vectors, which implies that
e0 · ė0 = e00 · ė00 = ė0 · e00 + e0 · ė00 = 0.
So, for our next step, we have to compute the integrals of θ0 and e0 · ė00 . We begin with
the integral of e0 · ė00 .
Lemma 2.4. Let
E du2 + 2F du dv + G dv 2 and L du2 + 2M du dv + N dv 2
be the first and second fundamental forms of σ, respectively. Then we have
LN − M 2
e0u · e00v − e00u · e0v = √ .
EG − F 2
Proof. Since derivatives of e0 are perpendicular to e0 , and similarly for e00 , there exist
scalars α, β, α0 , β 0 , λ0 , µ0 , λ00 , µ00 such that
e0u = αe00 + λ0 N,
e0v = βe00 + µ0 N
e00u = −α0 e0 + λ00 N,
e00v = −β 0 e0 + µ00 N.
Differentiating the equation e0 · e00 = 0 with respect to u, we find that e0u · e00 = −e0 · e00u , so
α0 = α, and similarly β 0 = β. Thus we have
e0u = αe00 + λ0 N,
e0v = βe00 + µ0 N
e00u = −αe0 + λ00 N,
e00v = −βe0 + µ00 N.
Thus
e0u · e00v − e00u · e0v = λ0 µ00 − λ00 µ0 .
Now, we also have
σu × σv √
Nu × Nv = K σ u × σ v , N= , kσ u × σ v k = EG − F 2 .
kσ u × σ v k
Combining all of these, we find that
LN − M 2
Nu × Nv = √ N,
EG − F 2
so
LN − M 2
(Nu × Nv ) · N = √ .
EG − F 2
 2. THE GAUSS–BONNET THEOREM FOR SIMPLE CLOSED CURVES 105

Since N = e0 × e00 , we get

(Nu × Nv ) · N = (Nu × Nv ) · (e0 × e00 )
= (Nu · e0 )(Nv · e00 ) − (Nu · e00 )(Nv · e0 )
= (N · e0u )(N · e00v ) − (N · e00u )(N · e0v )
= λ0 µ00 − λ00 µ0 ,
where we used the equation
Nu · e0 = −N · e0u
and other related formulae, obtained from differentiating the equation N · e0 = 0. This
completes the proof. 
Now, we compute
Z `(γ) Z `(γ) Z
0 00 0
e · ė ds = e · (u0 e00u + v 0 e00v ) ds = [(e0 · e00u ) du + (e0 · e00v ) dv].
0 0 π
By Green’s Theorem, we have
Z `(γ) Z
0 00
e · e ds = [(e0 · e00v )u − (e0 · e00u )v ] du dv
0 int(π)
Z
= [(e0u · e00v ) − (e0v · e00u )] du dv
int(π)

LN − M 2
Z
= √ du dv
int(π) EG − F 2
LN − M 2 √
Z
= EG − F 2 du dv
int(π) EG − F 2
Z
= K dAσ .
int(π)

So, that takes care of one term on each side in the Gauß–Bonnet Theorem for simple
closed curves. It remains to show that
Z `(γ)
θ0 ds = 2π.
0
This is a version of an important theorem known as Hopf ’s Umlaufsatz. We cannot give a
complete proof of the Umlaufsatz, but here’s a heuristic for it, which can be made precise
with some ideas from topology. The idea is that we suppose we have another simple closed
curve γ
e , contained in the interior of γ. Then there is a smooth family γ τ of simple closed
curves, such that γ 0 = γ and γ 1 = γe . This entire intermediate family must be contained in
U . The reason that we need int(π) to be contained in U is to avoid a situation like the one
shown in Figure 1.
Now, the integral
Z `(γ τ )
θτ0 ds
0
should be a continuous function of τ , where θτ denotes the θ function for the curve γ τ . Since
γ τ and e0 both return to their starting values as we go once around γ τ , the integral must
be a multiple of 2π. A continuous function on a closed interval that always takes on values
106 9. THE GAUSS–BONNET THEOREM

Figure 1. There is no family of curves connecting the red and blue curves
while staying entirely in the gray region.

that are multiples of 2π must be a constant function, so it suffices to check the result for a
convenient curve γ, such as a circle, where it’s easy to check that the Umlaufsatz holds.

3. Gauß–Bonnet for Curvilinear Polygons

So, we’ve done Gauß–Bonnet for smooth curves. Now it’s time to move on to piecewise
smooth curves, i.e. curves that are smooth except at finitely many corners.
Definition 3.1. A curvilinear polygon in R2 is a continuous map π : R → R2 such that,
for some real number T > 0 and points 0 = t0 < t1 < · · · < tn = T , the following properties
hold:
• π(t) = π(t0 ) if and only if t − t0 is a multiple of T .
• π is smooth on each of the open intervals (t0 , t1 ), (t1 , t2 ), . . . , (tn−1 , tn ).
• The one-sided derivatives
π(t) − π(ti ) π(t) − π(ti )
π̇ − (ti ) = lim− and π̇ + (ti ) = lim+
t→ti t − ti t→ti t − ti
exist for i = 1, . . . , n, are nonzero, and not parallel.
The points π(ti ) are called the vertices of the curvilinear polygon, and the images of the
intervals (ti , ti+1 ) are called the edges.
As in the case of smooth curves, we can talk about curvilinear polygons being positively
oriented if the interior of the polygon is on the left of the boundary as we travel around the
boundary by increasing t.
Now, if we have a curvilinear polygon π in R2 and a surface patch σ : U → R3 of a
surface S, then we can compose them to get a curvilinear polygon γ = σ ◦ π : R → R3 on S.
We use all the same terminology for curvilinear polygons on surfaces as we do for curvilinear
polygons on R2 . We call a curvilinear polygon γ unit-speed if kγ̇k = 1 whenever γ(t) is not
a vertex of γ; the derivative isn’t defined at the vertices, so we have to exclude those from
our definition. We also define the angles θt± to be the angles between γ̇ ± (ti ) and e0 , where
notation is as before. Let δt = θt+ − θt− , so δt = 0 away from the vertices, and it’s equal to
the angle at each vertex. Let αi = π − δti . All angles are defined only up to a multiple of
2π, so let’s assume that 0 < αi < 2π.
Theorem 3.2 (Gauß–Bonnet for Curvilinear Polygons). Let γ be a positively oriented
unit-speed curvilinear polygon with n edges on a surface patch σ, and let α1 , . . . , αn be its
 3. GAUSS–BONNET FOR CURVILINEAR POLYGONS 107

interior angles at its vertices. Then

Z `(γ) n
X Z
κg ds = αi − (n − 2)π − K dAσ .
0 i=1 int(γ)

Proof. The same proof as for the smooth case shows that
Z `(γ) Z `(γ) Z
0
κg ds = θ ds − K dAσ .
0 0 int(γ)

So we must prove that

Z `(γ) n
X
0
θ ds = 2π − δi ,
0 i=1

which implies the result.

To prove this, imagine smoothing each vertex of γ, as shown in Figure 2. We’ll call the
resulting smooth curve γe . Then we have
Z `(eγ )
θe0 ds = 2π.
0

Since γ and γ
e only differ near the vertices of γ, the difference
Z `(eγ ) Z `(γ)
0
θ ds −
e θ0 ds
0 0

is a sum of the contributions at all n vertices. Near a vertex γ(si ), γ and γ e are equal except
0 00
when s belongs to a small interval, say (si , si ) containing si , so the contribution at γ(si ) is
Z s00
i
Z si Z s00
i
θe0 ds − 0
θ ds − θ0 ds.
s0i s0i si

The first integral is the angle between γe 0 (s00i ) and γ

e 0 (s0i ), which tends to δi as s0i and s00i tend
to si . The last two integrals tend to 0 as s0i and s00i tend to si , because θ0 is smooth on those
intervals. Thus the contribution from the ith vertex is δi . This completes the proof. 

Corollary 3.3. If γ is a curvilinear polygon with n edges, each of which is an arc of a

geodesic, and the internal angles are α1 , . . . , αn , then we have
n
X Z
αi = (n − 2)π + K dAσ .
i=1 int(γ)

On a plane, this gives the usual formula that the sum of the angles of an n-gon is (n−2)π.
On a unit sphere, K = 1, so this tells us that the sum of the angles of an n-gon is (n − 2)π
plus the area of the polygon, a result that we have seen before. On a unit pseudosphere,
K = −1, so the sum of the angles is (n − 2)π minus the area of the polygon.
108 9. THE GAUSS–BONNET THEOREM

Figure 2. Smoothing the vertices of a curvilinear polygon.

4. Gauss–Bonnet for Compact Surfaces

And now, we finally reach the pinnacle of the differential geometry of surfaces: the global
Gauß–Bonnet Theorem! For this we need a bit of surface theory from topology.
Definition 4.1. Let S be a compact surface, with an atlas consisting of surface patches
σ i : Ui → R3 . A polygonization of S is a collection of curvilinear polygons, each of which is
contained, together with its interior, in one of the σ i (Ui )’s, such that
• Every point of S lies in at least one of the curvilinear polygons.
• Two curvilinear polygons are either disjoint, or their intersection is a vertex or an
edge of each.
• Each edge is an edge of exactly two polygons.
We will use the following result without proof:
Theorem 4.2. Every compact surface has a polygonization with only finitely many poly-
gons.
See Figure 3 for a picture of a polygonization of a torus into quadrilaterals.
From now on, we’ll assume that all polygonizations have only finitely many polygons,
without specifying this again.
While a surface can have many different polygonizations, there is a topological invariant
that we can extract from a polygonization, that is independent of the choice of polygonization
(but that does depend on the surface).
Theorem 4.3. Suppose we have a polygonization of a surface S with finitely many poly-
gons. Let V , E, and F be the number of vertices, edges, and faces, respectively, of the
polygonization. The Euler characteristic of S is defined to be
χ(S) = V − E + F.
 5. HOLONOMY 109

Figure 3. A polygonization of a torus into quadrilaterals.

It is true, but not that obvious, that the Euler characteristic is independent of the choice
of polygonization. (See problem 4.) One of many ways of seeing this is as a consequence of
the Gauß–Bonnet Theorem, which relates the Euler characteristic to the Gaussian curvature.
Theorem 4.4 (Gauß–Bonnet). Let S be a compact surface. Then
Z
K dA = 2πχ(S).
S

Proof. Fix a polygonization with triangular polygons Pi , each of which is contained in

a patch σ i : Ui → R3 , say with Pi = σ i (Ri ), where each Ri ⊆ Ui . Then by the Gauß–Bonnet
Theorem for curvilinear polygons, we have
Z Z `(γ i )
K dAσi = ∠i − π + κg ds,
Pi 0

where ∠i is the sum of the angles of Pi , γ i is the curvilinear polygon forming the boundary
of Pi , and `(γ i ) is its length. The left side of the Gauß–Bonnet Theorem is the sum of the
integrals on the left, overP all the triangles in the polygonization.
First, let’s look at i ∠i . At each vertex, P several polygons meet, and the sum of all the
angles at each vertex P is 2π. Thus we have i ∠i = 2πV .
Next, we have i π, which is clearly πF . Since P all the faces are triangular, we have
3F = 2E, so we can write F = 2E − 2F , which gives i π = 2πE − 2πF .
Finally, we claim that
X Z `(γ i )
κg ds = 0.
i 0

This is because we integrate twice along each edge, once in each direction. The geodesic
curvature changes sign when we traverse the edge in the opposite direction, so the two
contributions along each edge cancel out. This completes the proof of the Gauß–Bonnet
Theorem. 

5. Holonomy
Let’s now see an application of the Gauß–Bonnet Theorem. This one will be about
Gauß–Bonnet for simple closed curves. Recall the notion of holonomy: if we take a point p
on a surface S, a closed curve γ on S starting and ending at p, and a unit tangent vector
110 9. THE GAUSS–BONNET THEOREM

v ∈ Tp S, then we can parallel transport v along γ, and it ends at some v

e ∈ Tp S. However,
v
e is not necessarily equal to v.
Proposition 5.1. Let γ be a unit-speed curve on a surface patch σ, and let v be a
nonzero parallel vector field along γ. Let φ be the oriented angle between γ̇ and v, chosen to
be a continuous function of s. Then we have
dφ
κg = − .
ds
Proof. Since the length of v is constant for any parallel vector field, we can scale if
necessary and assume that kvk = 1. Write
v = cos φ t + sin φ t1 ,
where t = γ̇ and t1 = N × t. Since v is parallel along γ, v̇ is parallel to N. Thus we have
0 = t · v̇ = t · ((cos φ ṫ + sin φ ṫ1 ) + φ0 (− sin φ t + cos φ t1 )) = sin φ(t · ṫ1 − φ0 )
since t · ṫ1 = t · t1 = 0. Similarly, t1 · v̇ = 0, so we have
0 = cos φ (t1 · ṫ + φ0 ).
Now,
κg = γ̈ · (N × γ̇) = ṫ · (N × t) = ṫ · t1 ,
and since t · t1 = 0, we have
t · ṫ1 = −ṫ · t1 = −κg .
Hence we get
(φ0 + κg ) sin φ = 0 = (φ0 + κg ) cos φ.
Thus κg = −φ0 , as claimed. 
Now suppose that γ is a unit-speed closed curve. If we go around γ once, then φ increases
by
Z `(γ) Z `(γ)
dφ
ds = − κg ds,
0 ds 0
where `(γ) is the length of γ. If γ is a simple closed curve, then γ̇ rotates by 2π going once
around γ by Hopf’s Umlaufsatz. Thus we have the following:
Proposition 5.2. Let γ be a positively oriented unit-speed simple closed curve on a
surface σ, let κg be the geodesic curvature of γ, and let v be a nonzero parallel vector field
along γ. Then, on going around γ once, v rotates by an angle of
Z `(γ)
2π − κg ds.
0

Definition 5.3. If γ is a unit-speed closed curve on a surface S, then the angle

Z `(γ)
2π − κg ds
0
is called the holonomy around γ and is denoted by hγ .
We can combine what we have done to give an elegant formula for the holonomy of a
simple closed curve:
 6. PROBLEMS 111

Theorem 5.4. With notation as above, we have

Z
hγ = K dAσ .
int(γ)

This theorem allows us to express the holonomy in terms of the Gaussian curvature.
However, we can also go backward and express the Gaussian curvature in terms of the
holonomy: if γ is a sufficiently small curve circling around a point p ∈ S, then the Gaussian
curvature at p is approximately
hγ
.
Area(int(γ))
Proposition 5.5. Suppose that a surface S has the property that, for any two points
p, q ∈ S, the parallel transport Πpq
γ is independent of the choice of curve γ connecting p and
q. Then S is flat.
Proof. Letting p = q, we find that the holonomy around any simple closed curve is 0.
This implies that K = 0 everywhere, for if K 6= 0 somewhere, say K > 0, then there would
be an open set U with K > 0 everywhere on U . But then the holonomy would be positive
for any positively oriented simple closed curve contained in U . 

6. Problems
(1) Suppose that a surface patch σ has Gaussian curvature ≤ 0 everywhere. Show that
there are no simple closed geodesics on σ. How is this consistent with the fact that
parallels on a cylinder are geodesics?
(2) Let γ be a unit-speed curve in R3 with curvature κ 6= 0 everywhere and torsion τ .
Let n be the principal normal of γ, considered as a curve on S2 .
(a) Show that the geodesic curvature of n is
d τ
± arctan .
ds κ
(b) Show that if n is a simple closed curve on S2 , then the interior and exterior of
n are regions of equal area.
(3) Let σ be the surface of revolution
σ(u, v) = (f (u) cos v, f (u) sin v, g(u)),
with the usual hypotheses. Let u1 < u2 be constants, let γ 1 and γ 2 be the parallels
u = u1 and u = u2 , respectively, on σ, and let R be the region on σ with u1 ≤ u ≤
u2 . Compare the integrals
Z `(γ 1 ) Z `(γ 2 ) Z
κg ds, κg ds, K dσ A .
0 0 R
(4) In this exercise, we will show that χ(S) is independent of the choice of polygoniza-
tion, directly from the definition in terms of vertices, edges and faces.
(a) Show that dividing all the polygons into triangles does not change the value of
V − E + F . Thus from now on we can assume that all faces are triangular, and
we call the polygonzation a triangulation.
(b) Show that if we add one vertex to a triangulation, thus subdividing one or more
of the triangles, the quantity V − E + F does not change.
112 9. THE GAUSS–BONNET THEOREM

(c) Given two triangulations T1 and T2 of S, we say that T2 is a refinement of T1

if every triangle in T1 is a union of triangles in T2 . Show that if V1 , E1 , F1 and
V2 , E2 , F2 are the numbers of vertices, edges, and faces of T1 and T2 , respectively,
and T2 is a refinement of T1 , then V1 − E1 + F1 = V2 − E2 + F2 .
(d) Show that if T1 and T2 are two triangulations of S, then, after a very small
deformation of T1 or T2 , there exists a triangulation T3 that is a refinement
of both T1 and T2 . Conclude that, with the notation as in part (b), we have
V1 − E1 + F1 = V2 − E2 + F2 , so the Euler characteristic does not depend on
the choice of triangulation.
(5) Let T be the torus
σ(θ, φ) = ((a + b cos θ) cos φ, (a + b cos θ) sin φ, b sin θ),
where 0 < b < a.
(a) Describe the image of the Gauß map.
(b) Show directly, without using the Gauß–Bonnet Theorem, that
Z
K dAσ = 0.
T
(6) Let S be a compact connected surface with K > 0 everywhere. Prove that every
two simple closed geodesics in S must intersect.
(7) Does there exist a surface S that is diffeomorphic to S2 such that the region with
K ≤ 0 has greater area than the region where K ≥ 0?
(8) Does there exist a surface S that is diffeomorphic to a 2-holed torus such that the
region with K ≥ 0 has greater area than the region where K ≤ 0?
(9) Compute the holonomy of around the circle with latitude φ on S2 .
(10) Let
σ(θ, φ) = ((a + b cos θ) cos φ, (a + b cos θ) sin φ, b sin θ)
be a torus, where 0 < b < a. Compute the holonomy around the parameter circles
θ = θ0 and φ = φ0 .
(11) Compute the holonomy around the parameter circle v = 1 on the cone σ(u, v) =
(v cos u, v sin u, v).