18.

950 (Differential Geometry) Lecture Notes

Evan Chen
Fall 2015

This is MIT’s undergraduate 18.950, instructed by Xin Zhou. The formal

name for this class is “Differential Geometry”.
The permanent URL for this document is http://web.evanchen.cc/
coursework.html, along with all my other course notes.

Contents

1 September 10, 2015 3

1.1 Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Parametrized Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 September 15, 2015 5

2.1 Regular curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Arc Length Parametrization . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Change of Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4 Vector Product in R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.5 OH GOD NO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.6 Concrete examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3 September 17, 2015 9

3.1 Always Cauchy-Schwarz . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.4 Frenet formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.5 Convenient Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

4 September 22, 2015 12

4.1 Signed Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.2 Evolute of a Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.3 Isoperimetric Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

5 September 24, 2015 15

5.1 Regular Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.2 Special Cases of Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 15

6 September 29, 2015 and October 1, 2015 17

6.1 Local Parametrizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
6.2 Differentiability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
6.3 The Tangent Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1
Evan Chen (Fall 2015) Contents

7 October 6, 2015 19
7.1 Local Diffeomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7.2 Normals to Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7.3 The First Fundamental Form . . . . . . . . . . . . . . . . . . . . . . . . . 19
7.4 Relating curves by the first fundamental form . . . . . . . . . . . . . . . . 21
7.5 Areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

8 October 8, 2015 and October 15, 2015 23

8.1 Differentiable Field of Normal Unit Vectors . . . . . . . . . . . . . . . . . 23
8.2 Second Fundamental Form . . . . . . . . . . . . . . . . . . . . . . . . . . 23

9 October 20, 2015 25

9.1 Self-Adjoint Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
9.2 Line of Curvatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
9.3 Gauss curvature and Mean Curvature . . . . . . . . . . . . . . . . . . . . 25
9.4 Gauss Map under Local Coordinates . . . . . . . . . . . . . . . . . . . . . 26

10 October 22, 2015 27

10.1 Completing Equations of Weingarten . . . . . . . . . . . . . . . . . . . . . 27
10.2 An Example Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
10.3 Elliptic and Hyperbolic Points . . . . . . . . . . . . . . . . . . . . . . . . 28
10.4 Principal Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

11 October 29, 2015 and November 3, 2015 30

11.1 Isometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
11.2 The Christoffel Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
11.3 Gauss Equations and Mainardi-Codazzi . . . . . . . . . . . . . . . . . . . 31
11.4 Bonnet Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

12 November 5, 10, 12, 17, 19 2015 32

12.1 The Covariant Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
12.2 Parallel Transports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
12.3 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

13 November 24 and 26, 2015 34

13.1 Area Differential Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
13.2 Global Gauss-Bonnet Theorem . . . . . . . . . . . . . . . . . . . . . . . . 34

14 December 1 and December 3, 2015 35

14.1 The Exponential Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
14.2 Local Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

15 December 8 and December 10, 2015 36

15.1 Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
15.2 Variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
15.3 The Second Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
15.4 Proof of Bonnet Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

A Examples 40
A.1 Sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
A.2 Torus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
A.3 Cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

2
Evan Chen (Fall 2015) 1 September 10, 2015

§1 September 10, 2015

Here is an intuitive sketch of the results that we will cover.

§1.1 Curves
Consider a curve µ : [0, `] → R3 in R3 ; by adjusting the time scale appropriately, it
is kösher to assume kµ0 (t)k = 1 for every t in the interval. We may also consider the
acceleration, which is µ00 (t); the curvature is then kµ00 (t)k denoted k(t).
Now, since we’ve assumed kµ0 (t)k = 1 we have dt is a “length parameter” and we can
consider Z
k(t) dt.
µ

Theorem 1.1 (Fenchel)

For any closed curve we have the inequality
Z
k(t) ≥ 2π.
µ

with equality if and only if µ is a convex plane curve.

Here the assumption that kµ0 (t)k = 1 is necessary. (Ed note: in fact, this appeared as
the last problem on the final exam, where it was assumed that µ bounded a region with
mean curvature zero everywhere that was also homeomorphic to a disk.)
R
Remark 1.2. The quantity µ k(t) is invariant under obvious scalings; one can easily
check this by computation.
Some curves are more complicated, like a trefoil. In fact, geometry can detect topology,
e.g.

Theorem 1.3 (Fory-Milnor)

R
If µ is knotted then µ k(t) dt > 4π.

§1.2 Surfaces
We now discuss the topology of surfaces.
Roughly, the genus of a surface is the number of handles. For example, the sphere S 2
has genus 0.
In R3 it turns out that genus is the only topological invariant: two connected surfaces
are homeomorphic if and only if they have the same genus. So we want to see if we can
find a geometric quantity to detect this genus.
We define the Euler characteristic χ(Σ) = 2(1 − g(Σ)) for a surface Σ. It turns out
that this becomes equal to χ = V − E + F given a triangulation of the surface.
Define the Gauss curvature to a point p ∈ Σ: for a normal vector v, we consider Bp
the tangent plane, and we consider
AreaS 2 (µ(Σ ∩ Bp )
k(P ) = lim .
r→0 AreaΣ (Σ ∩ Bp )

where µ is the Gauss map; areas are oriented.

3
Evan Chen (Fall 2015) 1 September 10, 2015

Theorem 1.4 (Gauss-Bonnet Theorem)

R
Σ k(p)dp = 2πχ(Σ).

This relates a geometric quantity (left) to a topological quantity (right).

Example 1.5
The Gauss map is the identity map for S 2 , so k(p) = 1 for every p on S 2 .

§1.3 Parametrized Curves

Let’s do some actual calculus now.

Definition 1.6. A parametrized curve is a differentiable map α : I → R3 where

I = [a, b] ⊂ R.

4
Evan Chen (Fall 2015) 2 September 15, 2015

§2 September 15, 2015

Recall last time we considered a differential curve in space, i.e. a differentiable α : [a, b] →
R3 . Given t ∈ I, if α0 (t) 6= 0 the tangent line is the set

α(t) + sα0 (t) | s ∈ R .



§2.1 Regular curves

If t ∈ [a, b] has α0 (t) = 0, then we say t is a singular point of α. In this class we will
focus on regular curves which do not contain any singularities. In this case,
Given a regular curve α : I → R3 we define the arc length from a fixed parameter
t0 ∈ I is Z t
def  0 
s(t) = α (t) dt.
t0

Example 2.1 (Actual Coordinates)

Suppose we equip R3 with its standard inner product and put α(t) = (x(t), y(t), z(t)).
Then Z tp
s(t) = x0 (t)2 + y 0 (t)2 + z 0 (t)2 dt.
t0

The integral captures the ideal that we can approximate a curve by “close” straight line
segments..
t2

t1

t0 = a

tn = b

To be more precise, we consider a partition

P : a = t0 < t1 < · · · < tn = b

and consider X
`(α, P ) = kα(ti ) − α(ti−1 )k
i

and as the mesh of P (defined as maxi (ti − ti−1 ) approaches zero).

5
Evan Chen (Fall 2015) 2 September 15, 2015

Proposition 2.2 (Riemann Integral for Arc Length)

As the mesh of the partition approaches zero,
Z b
α0 (t) dt.

`(a, P ) →
a

Proof. Because [a, b] is compact, there is a uniform C1 such that kα00 (t)k < C. In what
follows all big O estimates depend only on α.
Then we can write for any t > ti−1 the inequality
Z Z
0 t t
α (t) − α0 (ti−1 ) = α00 (r) dr ≤ α00 (r) dr ≤ C [t − ti−1 ] .

ti−1 ti−1

Thus, we conclude that

Z ti  
α (t) − α0 (ti−1 ) = O [ti − ti−1 ]2 .
0
(1)
ti−1

Therefore, we can use (1) to obtain

Z ti
α0 (ti−1 ) + α0 (t) − α0 (ti−1 ) dt
 
α(ti ) − α(ti−1 ) =
ti−1
Z ti
0
 0
α (t) − α0 (ti−1 ) dt

= (ti − ti−1 )α (ti−1 ) +
ti−1
 
=⇒ (ti − ti−1 ) α (ti−1 ) = kα(ti ) − α(ti−1 )k + O [ti − ti−1 ]2 .
0

Thus,
Z ti 0
Z ti 0
α (t) − α0 (ti−1 ) + α0 (ti−1 ) dt

α (t) dt =
ti−1 ti−1
Z ti 0
α (t) − α0 (ti−1 ) dt + (ti − ti−1 ) α0 (ti−1 )

≤
ti−1

And applying the above result with (1) gives

  h  i
= O [ti − ti−1 ]2 + kα(ti ) − α(ti−1 )k + O [ti − ti−1 ]2
 
= kα(ti ) − α(ti−1 )k + O [ti − ti−1 ]2 .

Summing up gives the conclusion.

§2.2 Arc Length Parametrization

Suppose α : [a, b] → R3 is regular. Then we have
Z t
0
s(t) = α (t) dt.
a

Observe that
ds(t)
= α0 (t) 6= 0.

dt

6
Evan Chen (Fall 2015) 2 September 15, 2015

Thus the Inverse Function Theorem implies we can write t = t(s) as a function of s;
thus
def
α
e(s) = α (t(s))
is the arc length parametrization of s.
Hence, in what follows we can always assume that our curves have been re-parametrized
in this way.

§2.3 Change of Orientation

Given α : (a, b) → R3 we can reverse its orientation by setting β : (−b, −a) → R3 by
β(s) = α(−s).

§2.4 Vector Product in R3

Given two sets of ordered bases {ei } and {fi }, we can find a matrix A taking the first
to the second. We say that these bases are equivalent if det(A) > 0. Each of these two
equivalence classes is called an orientation of Rn .

Example 2.3 (Orientations when n = 1 and n = 3)

(a) Given R and a single basis 0 6= e ∈ R, then the orientation is just whether
e > 0.

(b) Given R3 the orientation can be determined by the right-hand rule.

§2.5 OH GOD NO
We define the cross product in R3 using the identification with the Hodge star:
 
e1 e2 e3
~u × ~v = det  u1 u2 u3  .
v1 v2 v3

(The lecturer uses ×, but this is misleading since it is not an exterior power.) This is the
vector perpendicular to ~u and ~v satisfying

k~u × ~v k = k~uk k~v k sin θ.

This is anticommutative, bilinear, and obeys

 
u1 u2 u3
~ = det  v1 v2 v3  .
(~u × ~v ) · (w)
w1 w2 w3

Also,  
~u · ~x ~v · ~x
(~u × ~v ) · (~x × ~y ) = det
~u · ~y ~v · ~y
which one can check by using linearity.
More properties:

• ~u × ~v = 0 ⇐⇒ ~u k ~v .

• d
dt (~
u × ~v )(t) = ~u0 (t) × ~v (t) + ~u(t) × ~v 0 (t).

7
Evan Chen (Fall 2015) 2 September 15, 2015

§2.6 Concrete examples

Consider the helix
α(t) = (a cos(t), a sin(t), b) .
Note that
α (t) = a2 + b2 def
0 p
= r.
Hence the arc length parametrization is just α
e(s) = α(s/r).

8
Evan Chen (Fall 2015) 3 September 17, 2015

§3 September 17, 2015

§3.1 Always Cauchy-Schwarz
Pointing out explicitly from last lecture:

Proposition 3.1 (Triangle Inequality in the Spirit of L1 )

For any f : I → Rn we have
Z ` Z `


f (t) dt
≤ kf (t)k dt.
0 0

Proof. By components.

§3.2 Curvature
Assume we have a curve α : I → R3 with kα0 (s)k = 1 meaning that α0 (s) actually
interprets an angle. Then we can consider α00 (s), which measures the change in the angle.

Definition 3.2. The curvature k(s) is kα00 (s)k.

Remark 3.3. Note that k(s) ≡ 0 if and only if α is a straight line.

Remark 3.4. The curvature remains unchanged by change of orientation.

Recall now that kα0 (s)k = 1 ⇐⇒ α00 (s) ⊥ α0 (s). Thus, we can define

α00 (s)
n(s) = .
kα00 (s)k

and call this the normal vector. At this point we define t(s) as the tangent vector
α0 (s). The plane spanned by α0 (s) and α00 (s) is called the osculating plane.

Definition 3.5. If k(s) = 0 at some point we say s is a singular point of order 1.

§3.3 Torsion
Definition 3.6. The binormal vector b(s) is defined by t(s) × n(s).

Proposition 3.7
In R3 we have that b0 (s) is parallel to n(s).

Proof. By expanding with Leibniz rule we can deduce

b0 (s) = t(s) × n0 (s).

Now, kb(s)k = 1 =⇒ b0 (s) ⊥ b(s). Also, b0 (s) ⊥ t(s). This is enough to imply the
conclusion.

Definition 3.8. In light of this, we can define b0 (s) = τ (s)n(s), and we call τ (s) the
torsion.

9
Evan Chen (Fall 2015) 3 September 17, 2015

Proposition 3.9 (Plane Curve ⇐⇒ Zero Torsion)

The curve α is a plane curve only if it has zero torsion (meaning τ (s) ≡ 0 everywhere).

Proof. τ (s) = 0 ⇐⇒ b0 (s) = 0 ⇐⇒ b(s) = b0 for a fixed vector b0 . For one direction,
note that if b(s) = b0 we have

d
[α(s) · b0 ] = α0 (s) · b0 = t(s) · b0 = 0
ds
the last step following from the fact that b(s) is defined to be normal to the tangent
vector. Thus α(s) · b0 is constant, so we can take s0 such that

b0 · (α(s) − α(s0 )) = 0

holds identically in s. The other direction is similar (take the derivative of the lpane
equation).

Remark 3.10. The binormal vector remains unchanged by change of orientation.

Example 3.11 (Helix)

Consider again the helix
 s s s
α(s) = a cos , a sin , b .
r r r
We leave to the reader to compute
 
b b
b(s) = sin(s/r), − cos(s/r), a/r .
r r

Thus  
0 b b
b (s) = cos(s/r), 2 sin(s/r), 0 .
r2 r
One can also compute

α00 (s)
n(s) = = (− cos(s/r), − sin(s/r), 0) .
k(s)

Thus one can obtain the torsion τ (s) = −b/r2 .

§3.4 Frenet formulas

Let α : I → R3 be parametrized by arc length. So far, we’ve associated three orthogonal
unit vectors

t(s) = α0 (s)
α00 (s)
n(s) =
kα00 (s)k
b(s) = t(s) × n(s)

which induce a positive orientation of R3 . We now relate them by:

10
Evan Chen (Fall 2015) 3 September 17, 2015

Theorem 3.12 (Frenet Formulas)

We always have

t0 (s) = k(s)n(s)
n0 (s) = −k(s)t(s) − τ (s)b(s)
b0 (s) = τ (s)n(s)

Proof. The first and third equation follow by definition. For the third, write n = b × t
and differentiate both sides, applying the product rule.

So it seems like we can recover t, n, b from k, τ by solving differential equations. In fact,

this is true:

Theorem 3.13 (Fundamental Theorem of the Local Theory of Curves)

Given differentiable functions k(s) > 0 and τ (s) on an interval I = (a, b), there
exists a regular parametrized curve α : I → R3 with that curvature and torsion.
Moreover, this curve is unique up to a rigid motion, in the sense that if α and α
both satisfy the condition, then there is an orthogonal linear map ρ : R3 → R3 with
det ρ > 0 such that α = p ◦ α + c. Intuitively, this means “uniqueness up to rotation
and translation”.

The uniqueness part is not difficult; the existence is an ordinary differential equation.

§3.5 Convenient Formulas

The curvature of an α not parametrized by arc length is

|α0 × α00 |
κ= .
|α0 |3

The torsion is
(α0 × α00 ) × α000
τ =− .
|α0 × α00 |2

11
Evan Chen (Fall 2015) 4 September 22, 2015

§4 September 22, 2015

§4.1 Signed Curvature
We wish to describe signed curvature for a plane curve.
Assume that α : I → R2 , so we have a tangent vector t(s). Then we can give a unique
normal vector n(s) such that (t, n) forms a positive basis.
~n(s)

~t(s)

Thus we can define the signed curvature according to

t0 (s) = k(s)n(s)
which may now be either positive or negative.

§4.2 Evolute of a Curve

Given a plane curve α : I → R2 , we define the evolute of α to be the curve
1
β(s) = α(s) + n(s).
k(s)

Proposition 4.1 (Properties of the Evolute)

Let α(s) be a curve parametrized by arc length and β its evolute. Then

(a) β is normal to α, and

(b) Let s1 and s2 be points, and consider the intersection p of their normal vectors
to α. As s1 , s2 → s, p → β(s).

We call β(s) the center of curvature of α(s).

s2
s1

12
Evan Chen (Fall 2015) 4 September 22, 2015

Proof. Part (a) is easy, since the Frenet equations imply n0 (s) = −kt (as the torsion
vanishes), and hence
1 1 0 k0 1 k0
β 0 (s) = α0 (s) − k 0
(s)n(s) + n (s) = t − n + (−kt) = − n(s).
k(s)2 k k2 k k2
Part (b) involves more arithmetic. Let α(s) = (x(s), y(s)) so t(s) = (x0 (s), y 0 (s)). Then
n(s) = (−y 0 (s), x0 (s)). Now, given s1 and s2 we seek the point
p = α(s1 ) + λ1 n(s1 ) = α(s2 ) + λ2 n(s2 )
for some λ1 and λ2 . One can check after work that
x0 (s2 )(x(s2 ) − x(s1 )) + y 0 (s2 )(y(s2 ) − y(s1 ))
λ1 = .
y 0 (s2 )x0 (s1 ) − x0 (s2 )y 0 (s1 )
In the limit, one can check that this becomes
x0 (s)2 + y 0 (s)2 1 1
lim λ1 = = 00 = .
s1 ,s2 →s y 00 (s)x0 (s) − x00 (s)y 0 (s) y (s)x0 (s) − x00 (s)y 0 (s) k(s)
Thus, α(s) + λ(s)n(s) = β(s) by definition.

§4.3 Isoperimetric Inequality

Problem 4.2. Of all simple closed curves in the plane with a given length `, which one
bounds the largest area?
In this situation we have α : [a, b] → R3 a regular curve. It is closed if α(a) = α(b)
and α(k) a = α(k) b for all k ∈ N. It is simple if it is not self-intersecting. We will quote
without proof that

Theorem 4.3 (Jordan Curve Theorem)

A closed simple regular curve bounds a region in R2 .

Thus one can assign the curve a positive orientation, namely the counterclockwise direction
(formally, n(s) points into the bounded region D described by the Jordan curve theorem).
We will now give:

Theorem 4.4 (Isoperimetric Inequality)

Let α be a simple closed plane curve of length ` and let A be the area of the bounded
region. Then
`2 ≥ 4πA
with equality if and only if α gives a circle.

To bring areas into the picture, we also quote:

Theorem 4.5 (Special Case of Green’s Theorem)

Given closed simple regular α(t) = (x(t), y(t)) the area bounded is
b b b
1
Z Z Z
A=− yx0 = xy 0 = (xy 0 − yx0 ) dt
a a 2 a

which can be found e.g. by using Green’s Theorem.

13
Evan Chen (Fall 2015) 4 September 22, 2015

Proof of Isometric Inequality. Parametrize the original curve α = (x, y) by arc length.
Thus α : [a, b] → R2 .
Suppose that x runs from a maximal right side coordinate xmax and a minimal xmin ;
we let 2r = xmax − xmin , and now draw a circle S with radius r, centered at the origin.
Now we parametrize S by β but not by arc length: rather, we parametrize β so that the
x-coordinates of β and α always coincide. So β : [a, b] → S. Let β = (x, y).
Let A be the area of the bounded region. Then by our area form
Z b
2
A + πr = (xy 0 − yx0 ) ds
a
Z bq
≤ (x2 + y 2 )(x02 + y 02 ) ds
a
Z bp
= x2 + y 2 ds
a
Z b
= r ds
a
= `r

Now by AM-GM we have √

`r = A + πr2 ≥ 2r Aπ
and this gives the desired bound on A.

14
Evan Chen (Fall 2015) 5 September 24, 2015

§5 September 24, 2015

We now begin the study of surfaces.

§5.1 Regular Surfaces

Definition 5.1. A regular surface is a subset S ⊆ R3 such that for every point p,
there is a neighborhood p ∈ V ⊆ R3 which is diffeomorphic to an open subset of R2 .
By diffeomorophic, we mean that there is a map
x : R2 ⊃ U → V ∩ S ⊂ R3
such that x is differentiable and bijective, with continuous inverse, and moreover the dif-
ferential dx : R2 → R3 is injective (i.e. it has full rank). We call the x a parametrization
at x and the ordered triple (x, U, V ∩ S) a coordinate chart.
(We don’t yet require the inverse to be differentiable because at the moment our set S
only has a topological structure, and don’t yet have a differential structure on S.)
In practice to check that the differential is injective one simply computes the associated
matrix and evaluates the rank (say by finding a 2 × 2 minor with full rank).

Example 5.2 (Projection of Sphere)

For a sphere S 2 = {(x, y, z) | x2 + y 2 + z 2 = 1}, we can take a projection onto the
xy, yz, zx planes for a total of six charts. For example on of the charts is
p
R2 ⊃ {u2 + v 2 < 1} → R3 by (u, v) 7→ (u, v, 1 − u2 − v 2 ).

Let’s verify this actually works. First, it is clear the map is differentiable. To see
that the differential has full rank, we simply use the first two coordinates and put
 
∂(u, v) 1 0
= det = 1.
∂(u, v) 0 1

Finally, we need to construct a continuous inverse. We just take the projection

(x, y, z) 7→ (x, y). This is clearly continuous.

§5.2 Special Cases of Manifolds

As the above example shows it does take some time to verify the (rather long) definition
of a manifold. So we would like some way to characterize regular surfaces more rapidly.
One obvious way to do this is:

Theorem 5.3 (Graphs are Regular Surfaces)

Let U ⊂ R2 and f : U → R be differentiable. Then the graph of f , i.e. the set

{(x, y, f (x, y)) | (x, y) ∈ U }

is a regular surface.

Proof. Repeat the work we did when we checked the sphere formed a regular surface.
Note that in this case we actually have a global parametrization, namely U → R3 by
(x, y) 7→ (x, y, f (x, y)).

15
Evan Chen (Fall 2015) 5 September 24, 2015

Here is a second way to do this. First, recall that

Definition 5.4. Given a differential map F : U → Rm (where U ⊂ Rn is open) a critical

point p ∈ U of F is one such that (dF )p : Rn → Rm is not surjective. The value of F (p)
is called critical value.
Values in the image Rm which are not critical are called regular values.

In particular, if F : U → R then p is a critical value if and only if (DF )p vanishes, i.e.

the gradient is zero.
Now we claim that

Theorem 5.5 (Level Images are Surfaces)

Let f : R3 ⊃ U → R is differentiable and c be a regular value. Then the pre-image

(x, y, z) ∈ R3 | f (x, y, z) = c


is a regular surface.

The proof of this is essentially the inverse function theorem.

16
Evan Chen (Fall 2015) 6 September 29, 2015 and October 1, 2015

§6 September 29, 2015 and October 1, 2015

I was absent this week, so these notes may not be a faithful representation of the lecture.

§6.1 Local Parametrizations

Regular surfaces are nice because we have local coordinates at every point.

Proposition 6.1 (Local Parametrization as Graph)

Let S ⊂ R3 and p ∈ S. Then there exists a neighborhood V of p ∈ S such that V
is the graph of a differentiable function in one of the following forms: x = f (y, z),
y = f (z, x), z = f (x, y).

Sketch of Proof. By definition of regular surface, we can take a parametrization at p.

One of the Jacobian determinants is nonzero, by hypothesis.

Finally, we state a lemma that says that if we know S is regular already, and we have
a candidate x for a parametrization x : U → R3 for U ⊂ R2 , then we need not check x−1
is continuous if x is injective.

Proposition 6.2
Let p ∈ S for a regular surface S. Let x : U → R3 be differentiable with image in
p such that the differential dx is injective everywhere. Assuming further that x is
injective then x−1 is continuous.

§6.2 Differentiability
We want a notion of differentiability for a function f : V → R where V is an open subset
of a surface S. This is the expected:

Definition 6.3. Let f : V ⊂ S → R be a function defined in an open subset V of

a regular surface S. Then f is differentiable at p ∈ V if for some parametrization
x : U ⊂ R2 → S, the composition
x f
U−
→V −
→ R.

is differentiable at x−1 (p) (giving x−1 (p) 7→ p 7→ f (p)).

One can quickly show this is independent of the choice of parametrization, so “for
some” can be replaced by “for all”.

§6.3 The Tangent Map

Given our embedding in to ambient space, we can define the tangent space as follows.

Definition 6.4. A tangent vector to p ∈ S is the tangent vector α0 (0) of a smooth

curve α : (−ε, ε) → S with α(0) = p.

The “type” of the “tangent vector” is actually vector in this case.

17
Evan Chen (Fall 2015) 6 September 29, 2015 and October 1, 2015

Proposition 6.5
Let x : U → S be a parametrization (U ⊂ R2 ). For q ∈ U , the vector subspace

dxq (R2 ) ⊂ R3

coincides with the set of tangent vectors of S at q.

In particular, dxq (R2 ) doesn’t depend on the parametrization x, so we denote it now by

Tp (S), the tangent space at S.
This lets us talk about the notion of a differential of a smooth map between surfaces.
Specifically, suppose
φ
S1 ⊃ V −
→ S2 .
Then given a map α : (−ε, ε) → V we get a composed path β

(−ε, ε)

α
β
-

?
V - S
φ

Proposition 6.6
Consider the situation above. The value of β 0 (0) depends only on α0 (0), so we obtain
a map
(dφ)p : Tp (S1 ) → Tφp (S2 ).
Moreover, this map is linear; actually,
 ∂φ1 ∂φ1   0 
0 ∂u ∂v u (0)
β (0) = dφp (w) = ∂φ2
∂u
∂φ2
∂v
v 0 (0)

where φ = (φ1 , φ2 ) in local coordinates.

18
Evan Chen (Fall 2015) 7 October 6, 2015

§7 October 6, 2015
§7.1 Local Diffeomorphism
Definition 7.1. A map φ : U ⊂ S1 → S2 is a local diffeomorphism at p ∈ U if there
exists a neighborhood p ∈ V ⊂ U such that φ is a diffeomorphism of V onto its image in
S2 .

Example 7.2 (Local 6= Global)

The map S 1 → S 2 by z 7→ z 2 is a local diffeomorphism but not globally (since it is
not injective).

We can check whether functions are local diffeomorphisms by just their differen-
tials.

Proposition 7.3 (Inverse Function Theorem)

Let S1 and S2 be regular surfaces, φ : U ⊂ S1 → S2 differentiable. If the differential

(dφ)p : Tp S1 → Tφ(p) S2

is a linear isomorphism (full rank), then φ is a local diffeomorphism at p.

Proof. The point is that at p, φ(p), S1 and S2 look locally like Euclidean space. Using
the parametrizations x and x respectively, we induce a map U → U whose differential is
an isomorphism:
φ -
S1 S2
6 6
x x
? ?
U - U

The classical inverse function theorem gives us an inverse map, which we then lift to a
map of the surfaces.

§7.2 Normals to Surfaces

Let p ∈ S be a point, and let x : U → S be a local parametrization of S at p. Then we
define the normal vector to the surface as follows: observe that the tangent space Tp (S)
has a basis x1 , x2 corresponding to the canonical orientation of R2 (as inherited by U ).
Then we define
x1 × x2
N (p) = ± .
|x1 × x2 |
Then, if two surfaces intersect at p, the angle between the surfaces is the angle between
the normal vectors.

§7.3 The First Fundamental Form

Let S ,→ R3 be a regular surface. Then for any p ∈ S, the tangent plane Tp S inherits an
inner product from R3 , because it is a subsace of R3 .

19
Evan Chen (Fall 2015) 7 October 6, 2015

Thus, we have a map

Ip : Tp S → R by w 7→ hw, wi = kwk2 .

Note that by abuse, we use h−, −i despite the fact that this form depends on p

Definition 7.4. The map Ip is called the first fundamental form of S at p ∈ S.

The power of Ip is that it lets us forget about the ambient space; we can talk about
the geometry of the surface by looking just at Ip . In fact, it was shown very recently
by John Nash that the intrinsic and extrinsic notion of geometry coincide: roughly, any
such surface with a fundamental form of this sort can be embedded in Rn .
To repeat: we will see that the first fundamental form allows us to make
measurements on the surface without referring back to the ambient space
R3 .
One can compute this explicitly given local coordinates. Suppose we parametrize
x : U ⊂ R2 → S and take p ∈ U . Then if we denote R2 = uR ⊕ vR, then Tp S has a basis
∂x
{xu , xv }, where xu = ∂u (p) and xu = ∂x
∂v (p).
Now, let w ∈ Tp (S). Let α : (−ε, ε) → S such that α(t) = x(u(t), v(t)) and

w = α0 (0) = u0 (0)xu + v 0 (0)xv .

Then by expanding, we obtain

Ip (w) = hw, wip

= u0 (0)xu + v 0 (0)xv , u0 (0)xu + v 0 (0)xv



= kxu k2 u0 (0)2 + 2 hxu , xv i u0 (0)v 0 (0) + kxv k2 v 0 (0)2 .

We write this as

= E · u0 (0)2 + 2F · u0 (0)v 0 (0) + G · v 0 (0)2 .

where

E = hxu , xu i
F = hxu , xv i
G = hxu , xv i .

Example 7.5 (Plane)

Let P ⊂ R3 be the plane parametrized globally by a single chart

x(u, v) = p0 + uw1 + vw2

for some orthonormal vectors w! and w2 . Then, (E, F, G) = (1, 0, 1).

Then given any tangent vector w = aw1 + bw2 at an arbitrary point p we get
Ip (w) = a2 + b2 ; we just get the length of the vector.

20
Evan Chen (Fall 2015) 7 October 6, 2015

Example 7.6 (Cylinder)

Consider the cylinder
C = (x, y, z) | x2 + y 2 = 1 .


We now take the chart

x(u, v) = (cos u, sin u, v)
where the domain is U = (0, 2π) × R; this covers most of the cylinder C (other
than a straight line). We notice that the tangent plane is exhibited by the basis
xu = (− sin u, cos u, 0) and xv = (0, 0, 1). Thus E = sin2 u + cos2 u = 1, F = 0,
G = 1.
So the cylinder and the plane have the same first fundamental form. This is the
geometric manifestation of the fact that if we cut a cylinder along an edge, then we
get a plane.

Example 7.7 (Stuff Depends on Parametrization)

Suppose now we use the parametrization on the cylinder

x(u, v) = (cos u, sin u, 5v)

Then the tangent plane is exhibited by the basis xu = (− sin u, cos u, 0) and xv =
(0, 0, 5). Thus E = sin2 u + cos2 u = 1, F = 0, G = 5.

§7.4 Relating curves by the first fundamental form

Let α : I → S be a regular curve. We already knew that the arc length is given by
Z t
0
s(t) = α (t) dt.
0

If the curve is contained entirely within a chart α(t) = x(u(t), v(t)) then we may rewrite
this as Z tq Z tp
0
Iα(t) α (t) dt = E(u0 )2 + 2F u0 v 0 + G(v 0 )2 dt.
0 0

Colloquially, we can write this as ds2 = E(du)2 + 2F dudv + G(dv)2 .

Given α, β : I → S now with α(t0 ) = β(t0 ), the angle θ between the two tangent curves
is
hα0 (t0 ), β 0 (t0 )i
cos θ = .=
kα0 (t0 )k kβ 0 (t0 )k
As a special case, suppose α and β are coordinate curves, meaning α0 (t) = xu and
β 0 (t) = xv for all α, β. (Think rulings on a surface.) The angle between them xu and xv
is thus given by
hxu , xv i F
cos φ = =√ .
kxu k kxv k EG
Example application: loxodromes.

21
Evan Chen (Fall 2015) 7 October 6, 2015

§7.5 Areas
Let R ⊂ S be a closed and bounded region of a regular surface contained in the
neighborhood of a parametrization x : U ⊂ R2 → S. Let Q = x−1 (R). The positive
number Z
kxu × xv k du dv
Q

does not depend on the choice of x and

√ is defined as the area of R.
In local coordinates, kxu × xv k = EG − F 2 .

22
Evan Chen (Fall 2015) 8 October 8, 2015 and October 15, 2015

§8 October 8, 2015 and October 15, 2015

I am now attending roughly only every third class or so. Here are my notes from my
own reading from the content I missed.

§8.1 Differentiable Field of Normal Unit Vectors

Let N : S → S 2 ⊆ R3 be a map which gives a normal unit vector at each point.
For example, if x : U → S (where U ⊂ R2 ) is a parametrization, then we define for
each q ∈ x(U ) by
xu × xv
N (q) = at q.
|xu × xv |
By abuse, we denote this Nq .
More generally, if V ⊂ S is open, N : V → S 2 ⊂ R3 is a differentiable map, then we
say that N is a differentiable field of normal unit vectors
Remark 8.1. Not all surfaces admit such a field of unit vectors (Möbius strip, for
example). Of course, most things (like spheres) do.
This N is called the Gauss map.

Example 8.2 (Cylinder)

Let S be the cylinder {(x, y, z) | x2 + y 2 = 1}. We claim that N = (−x, −y, 0) are
unit normal vectors. A tangent vector is given by α0 (0) = (x0 (0), y 0 (0), z 0 (0)) where
α = (x, y, z) has α(0) = p. Since x(t)2 + y(t)2 , we see that along this curve, we see
that N · α0 = 0, as desired.

§8.2 Second Fundamental Form

In this way, N provides an orientation on S. We see in this case, for any point p we have
a map
'
dNp : Tp S → TN (p) S 2 −
→ Tp S

Example 8.3
Let S be the cylinder {(x, y, z) | x2 + y 2 = 1}. Let us see what dNp does to tangent
vectors. Using the setup with α as before (so α(0) = p), the tangent N (t) to α(t) is
(−x(t), −y(t), z(t)) and thus

dNp x0 (0), y 0 (0), z 0 (0) = −x0 (0), −y 0 (0), 0 .

In other words, dNp is projection onto the xy-plane. In particular, the tangents to
the cylinder parallel to the xy-plane are −1 eigenvectors while the tangents to the
cylinder parallel to the z-axis are all mapped to zero (in the kernel).

Lemma 8.4 (dNp is self-adjoint)

As defined, dNp : Tp S → Tp S is self-adjoint, in the sense that

hdNp (v), wi = hv, dNp (w)i .

23
Evan Chen (Fall 2015) 8 October 8, 2015 and October 15, 2015

Proof. Let α(0) = p, α(t) = x(u(t), v(t)), α0 (0) = u0 (0)xu + v 0 (0)xv . Write
 
0 d
dNp (α (0)) = (N (u(t), v(t))) = Nu u0 (0) + Nv0 v 0 (0).
dt t=0

So, it suffices to show that

hNu , xv i = hxv , Nu i .
Take the derivatives of hN, xu i = 0 and hN, xv i = 0 relative to u and v to see that both
are equal to − hN, xuv i = − hN, xvu i.

Definition 8.5. Let Ip be the quadratic form defined in Tp (S) by

Ip (v) = − hdNp (v), vi = − hv, dNp (v)i .

This is called the second fundamental form of S at p.

Now, let us think about curves for a moment.

Definition 8.6. Let C be a regular curve in S, and p ∈ S a point on it. Let k be the
curvature of C at p, and set cos θ = hn, N i, where n is the normal to C and N is the
normal vector to S at p. The number kn = k cos θ is then called the normal curvature
of C ⊂ S at p.

In other words, kn is the length of the projection of the vector kn = α0 over the normal
to the surface at p, with sign given by orientation.

Theorem 8.7 (Meusnier)

The normal curvature of a curve C at p depends only on the tangent line at the
curve.

Proof.

Thus, it make sense to talk about the curvature at a direction to the surface.

24
Evan Chen (Fall 2015) 9 October 20, 2015

§9 October 20, 2015

§9.1 Self-Adjoint Maps
Since dNp is self-adjoint, there exists an orthonormal basis {e1 , e2 } of Tp S such that

dNp (e1 ) = −k1 e1 and dNp (e2 ) = −k2 e2 .

Definition 9.1. The constant k1 and k2 are called the principal curvature and the
basis e1 and e2 are called the principal directions.

Given w ∈ Tp S, say w = cos θe1 + sin θe2 , then the value of the normal curvature to
kn along w is

− hdNp (cos θe1 + sin θe2 ), cos θe1 + sin θe2 i = hk1 cos θe1 + k2 sin θe2 , cos θe1 + sin θe2 i
= k1 cos2 θ + k2 sin2 θ.

So we can express any normal curvature in terms of the principal ones.

§9.2 Line of Curvatures

Definition 9.2. If a regular curve C on S has the property that for all p ∈ C, the
tangent line of C at p is a principal direction at p, then C is called a line of curvature
of S.

Example 9.3 (Examples of Lines of Curvatures)

(a) In a plane, all curves are lines of curvature (trivially). (dNp = 0, so everything
is an eigenvalue.)

(b) All curves on a sphere are also lines of curvature. (dNp = id, so everything is
an eigenvalue.)

(c) On a cylinder, the vertical lines and the meridians are lines of curvatures of S.

Proposition 9.4 (Testing for Line of Curvature)

Let C be a regular curve. Then C is a line of curvature if and only if

dNp (α0 (t)) = λ(t)α0 (t)

where α(t) is the arc length parametrization, and λ(t) ∈ R.

Proof. “Harmless exercise”. (A better word might be “trivial”.)

§9.3 Gauss curvature and Mean Curvature

 
k1 0
In the orthonormal basis of p, we have dNp = − . Of course, we define
0 k2
Definition 9.5. The quantity det dNp = k1 k2 is called the Gauss curvature. The
quantity Tr(dNp ) = −(k1 + k2 ) is more finicky. We instead define H = 12 (k1 + k2 ) =
− 21 Tr(dNp ) the mean curvature.

25
Evan Chen (Fall 2015) 9 October 20, 2015

Remark 9.6. One can view H as the “average” of all curvatures of curves passing
through H.

Definition 9.7. A point p ∈ S is called

• elliptic if det(dNp ) > 0, meaning k1 and k2 have the same sign.

• hyperbolic if det(dNp ) < 0,

• parabolic if det(dNp ) = 0, but dNp 6= 0, and

• planar if dNp is the zero matrix.

It’s worth noting that planes are not the only example where planar points. For example,
if z = x4 , and we wrote this around the z-axis to get a surface (i.e. z = (x2 + y 2 )2 ) then
the origin is also a planar point.

Definition 9.8. An umbilical point of S is a point p ∈ S at which k1 = k2 , meaning

dNp is a constant times the identity.

Theorem 9.9
All connected surfaces S which consist entirely of umbilical points are subsets of the
sphere or a plane.

Proof. Prove it locally first. Then for any fixed p, for other q ∈ S take a path from p to
q, and use compactness.

§9.4 Gauss Map under Local Coordinates

Let p ∈ S where S is an regular surface with an orientation N : S → S 2 ⊂ R3 . For p ∈ S
and x : U → S a parametrization near p, we can set

α(t) = x(u(t), v(t)) ∈ S

such that α(0) = p. So,

α0 = u0 xu + v 0 xv
where xu and xv are the usual basis of Tp S (i.e. the partial derivatives of x with respect
to u and v). Thus, it follows that
d
dNp (α0 ) = N (u(t), v(t)) = u0 Nu + v 0 Nv .
dt
So, we simply need to compute Nu and Nv . If we sidestep this for now and say
Nu = a11 xu + a21 xv , and Nv = a12 xu + a22 xv , then this reads
 0 
a11 u0 + a12 v 0
    0
u a11 a12 u
dNp 0 = 0 0 = .
v a21 u + a22 v a21 a22 v0
 
a11 a12
Note that the matrix need not be symmetric unless xu and xv are orthogonal.
a21 a22
(Self-adjoint implies symmetric only in this case.)
One can compute the aij , but since I really don’t want to. The result is called the
equations of Weingarten.

26
Evan Chen (Fall 2015) 10 October 22, 2015

§10 October 22, 2015

§10.1 Completing Equations of Weingarten
Let me finish the computation from last time. Take α(t) = x(u(t), v(t)), α(0) = p, Recall
the we have written

Nu = a11 xu + a21 xv
Nv = a11 xu + a21 xv

(again these subscripts are partial derivatives at the point p). The point is that

dNp : Tp S → Tp S by u0 xu + v 0 xv 7→ u0 Nu + v 0 Nv

where Nu , Nv ∈ Tp S as well. Thus, We have that

 def
Ip (α0 ) = Nu u0 + Nv v 0 , xu u0 + xv v 0 = e(u0 )2 + 2f u0 v 0 + g(v 0 )2

for e = − hNu , xu i = hN, xuu i and similarly for others. If we use the first equation, then
putting in the aij we derive
    
e f a11 a12 E F
− = .
f g a21 a22 F G
| {z }
dNp

In particular, we have
eg − f 2
k1 k2 = det dNp = .
EG − F 2
Through pain and suffering, we also can compute
1 1 eG − 2f F + gE
H = (k1 + k2 ) = .
2 2 EG − F 2

§10.2 An Example Calculation

We consider a torus, parametrized by

x(u, v) = ((a + r cos u) cos v, (a + r cos u) sin v, r sin u) .

The partial derivatives are

xu = (−r sin u cos v, −r sin u sin v, r cos u)

xv = (−(a + r cos u) sin v, (a + r cos u) cos v, 0) .

Thus, we can extract

E = hx, xu i = r2
F = hxu , xv i = 0
G = hxv , xv i = (a + r cos u)2 .

27
Evan Chen (Fall 2015) 10 October 22, 2015

xu ×xv
√
Let N = |xu ×xv | , noting that the denominator is EG − F 2 . We define the constants

e = hNu , xu i
= hN, xuu i
 
−r sin u cos v −r sin u sin v r cos u
det −(a + r cos u) sin v (a + r cos u) cos v 0 
−r cos u cos v −r cos u sin v −r sin u
= √
EG − F 2
2
r (a + r cos u)
= = r.
r(a + r cos u)

Similarly,

f = hN, xuv i = · · · = 0
g = hN, xvv i = · · · = cos u(a + r cos u).

Thus, we have
cos u
K= .
r(a + r cos u)

§10.3 Elliptic and Hyperbolic Points

Proposition 10.1
Let p ∈ S with S a regular surface.

• If p ∈ S is elliptic, then all points in some neighborhood of S are on one side

of the tangent plane Tp S

• If p ∈ S is hyperbolic, for any neighborhood of p there is a point on both sides

of Tp S.

Proof. Let x(u, v) be a parametrization near p, with x(0, 0) = p. We’re interested in the
signs of the function
f (u, v) = hx(u, v) − x(0, 0), N i
with Np the normal at p. Locally, we have

1
xuu u2 + 2xuv uv + xvv v 2 + R

x(u, v) = x(0, 0) + xu (0, 0)u + xv (0, 0)v +
2
with R an error term with u2 R
+v 2
→ 0 as (u, v) → 0. Now, hxu (0, 0), N i = hxv (0, 0), N i =
0. If we expand and go through the computation, denoting R = hR, N i one can compute

u2 + v 2
   
1 (u, v) R
f (u, v) = Ip (uxu + vxv ) + R = Ip √ + 2 .
2 2 u2 + v 2 u + v2
 
Suppose first that k1 ≥ k2 > 0. Then Ip √(u,v) 2
u +v 2
= 12 (k1 cos2 θ + k2 sin2 θ) ≥
1
2min{k1 , k2 } is bounded away from zero. So for u2 + v 2 small enough, f (u, v) ≥ 0.
An analogous argument works if k1 > 0 > k2 .

28
Evan Chen (Fall 2015) 10 October 22, 2015

§10.4 Principal Directions

Theorem 10.2 (Criteria for Principal Directions)

Fix a point p ∈ S. Then u0 xu + v 0 xv is an eigenvector of (dN )p if and only if
 0 2
(v ) −u0 v 0 (u0 )2


0 = det  E F G .
e f g

29
Evan Chen (Fall 2015) 11 October 29, 2015 and November 3, 2015

§11 October 29, 2015 and November 3, 2015

§11.1 Isometries
Let S1 and S2 be surfaces.
Definition 11.1. A diffeomorphism φ : S → S 0 is an isometry if for all w1 and w2 in
Tp (S), we have
hw1 , w2 ip = h(dφ)p (w1 ), (dφ)p (w2 )iφ(p) .
In that case, we say S1 and S2 are isometric.
Naturally, this means that the first fundamental form Ip is preserved. In fact, by
considering hv + w, v + wi − hv, vi − hw, wi we see that isometries are equivalent to
preserving just the first fundamental form.
Definition 11.2. Let p ∈ V ⊂ S 0 , where S 0 is a surface. A map φ : V → S, where V
is a neighborhood of some surface, is a local isometry at p if it is an isometry onto a
neighborhood of φ(p)
We say S locally isometric to S 0 if one can find a local isometry into S 0 at every
point p ∈ S. If the other direction holds too, we say the surfaces are locally isometric.
Properties that are preserved under local isometry are said to be intrinsic. for example,
the first fundamental form is intrinsic.

Example 11.3 (Cylinder and Plane are Locally Isometric)

The cylinder is locally isometric to a plane (0, 2π) × R, but not globally (since the
spaces are not homeomorphic, their fundamental groups differ).

Example 11.4 (Half-Cylinder and Plane are Globally Isometric)

Half a cylinder is globally isometric to the plane (0, π) × R. Note that the second
fundamental form of a cylinder is nonzero (unlike the plane),

Thus, second fundamental form is not intrinsic. In fact, we will later see that the mean
curvature is not intrinsic, but the global curvature is.
For aid with local coordinates:

Proposition 11.5
Suppose x : U → S and x0 : U → S 0 are parametrizations such that we have an
equality E = E 0 , F = F 0 and G = G0 of the coefficients of the first fundamental
form. Then x0 ◦ x−1 is a local isometry.

§11.2 The Christoffel Symbols

This is some very long calculations. The idea is that xu , xv , N form a triple of orthogonal
vectors at every point p of the surface. Then, we can let
xuu = Γ111 xu + Γ211 xv + eN
xuv = Γ112 xu + Γ212 xv + f N
xvu = Γ121 xu + Γ221 xv + f N
xvv = Γ122 xu + Γ222 xv + gN.

30
Evan Chen (Fall 2015) 11 October 29, 2015 and November 3, 2015

Here the e, f , g coefficients were realized by taking the inner form of each equation
with N . Also, note that xuv = xvu , so Γi12 = Γi21 for i = 1, 2. We say the Γijk are the
Christoffel symbols.

§11.3 Gauss Equations and Mainardi-Codazzi

If we take dot products of all relations with xu and xv , we obtain a system of equations
that lets us solve for everything in terms of E, F , G and their derivatives with respect to
u, v. (Note e, f , g are not present!)
There are some more relations we can derive. For example, note that

(xuu )v = (xuv )u .

It turns out that by expanding this and equating the xu coefficients, we obtain that

−KE = (Γ212 )u − (Γ211 )v + Γ112 Γ211 + Γ212 Γ212 − Γ211 Γ222 − Γ111 Γ212

where K is the Gauss curvature; this is the Gauss equation (Equating the xv coefficients
turns out to give the same equation.) The philosophical point is that it implies:

Theorem 11.6 (Theorema Egregium, due to Gauss)

The Gaussian curvature K is intrinsic, i.e. it is invariant under local isometries.

On the other hand, equating the N coefficients gives

eu − fu = eΓ112 + f (Γ212 − Γ111 ) − gΓ211 .

Doing a similar thing gives

fu − gu = eΓ122 + f (Γ222 − Γ112 ) − gΓ212 .

These two relations are called the Mainardi-Codazzi equations.

§11.4 Bonnet Theorem

We may wonder if we can get more relations. The answer is no.

Theorem 11.7 (Bonnet)

Let E, F , G, e, f , g be differentiable functions V → R, with V ⊂ R2 . Assume
E, G > 0, that Gauss and Codazzi holds, and moreover EG − F 2 > 0. Then:

(a) for every q ∈ V there is a neighborhood U of V such that a diffeomorphism

x : U → R3 such that x(U ) is a surface having the above coefficients.

(b) this x is unique up to translation and orthogonal rotation.

31
Evan Chen (Fall 2015) 12 November 5, 10, 12, 17, 19 2015

§12 November 5, 10, 12, 17, 19 2015

In what follows, a vector field w on a surface S will mean a differentiable vector field of
tangents; i.e. w(p) ∈ Tp S for every p ∈ S.

§12.1 The Covariant Derivative

Definition 12.1. Let w be a vector field, and let y ∈ Tp S be a tangent vector, realized by
y = α0 (0) for α : (−ε, ε) → S parametrized. The covariant derivative is the projection
of (w ◦ α)0 (0) onto the tangent plane, and denoted (Dy w)(p). We also denoted it by
(Dw/dt)(0) if α is given.

This of course does not depend on α. We can apply an analogous definition if w is

only defined along points of a given path α.
In particular, this is zero if and only if (w ◦ α)0 is parallel to the normal vector of Tp (S).
One standard example is the case w = α0 .

§12.2 Parallel Transports

Definition 12.2. Let α be a parametrized curve and w a vector field defined on it. We
say α is parallel if Dw/dt is zero everywhere on α.

In other words, this occurs if for every α(t) = p and y = α0 (t) we have (Dy w)(p) is
parallel to the normal vector of Tp (S). In particular if S is the plane this is equivalent to
w being constant.

Proposition 12.3
Let w and v be parallel vector fields along α. Then hw(t), v(t)i is fixed.

Thus parallel vector fields preserve angles and have constant lengths.
By theory of differential equations, we have

Theorem 12.4 (Unique Parallel Vector Field)

Let α be a parametrized curve in S and pick a fixed initial vector w0 ∈ Tα(0) S. Then
there exists a unique parallel vector field w along α such that w(0) = w0 .

So in this way we can define the parallel transport along a parametrized curve: given
α joining p to q, for any vector w ∈ Tp (S) we can follow along the parallel vector field q
to get a unique tangent vector in Tq (S).

§12.3 Geodesics
Most important special case of parallelism:

Definition 12.5. If α(s) is a regular parametrized curve and α0 is parallel, then α is

said to be a geodesic.

In this case, it means that the normal vectors ~n for α coincide with normal vectors to
the surface itself.

32
Evan Chen (Fall 2015) 12 November 5, 10, 12, 17, 19 2015

Theorem 12.6 (Unique Geodesic in Any Direction)

Let p be a point of a regular surface S and 0 6= w ∈ Tp (S). There is a geodesic
α : (−ε, ε) → S with α(0) = p, and α0 (0) = w, with the usual uniqueness up to
domain.

Example 12.7 (Examples of Geodesics)

(a) Straight lines are always geodesics, because their normal vector vanishes
altogether.

(b) In particular, on a plane, the straight lines are the only geodesics.

(c) On a sphere, the great circles are the unique geodesics.

(d) On a right cylinder, geodesics are helixes.

More generally, let α be a regular parametrized curve on an oriented surface S. At a

point p, consider the tangent vector α0 = ~t and then the vector α00 . We can project it
onto the Tp (S) as before, arriving at Dα0 /ds; its signed length (with respect to N × ~t) is
the geodesic curvature kg . Thus geodesics are those curves with kg = 0.

Remark 12.8. Since kn is the other projection, k 2 = kn2 + kg2 .

33
Evan Chen (Fall 2015) 13 November 24 and 26, 2015

§13 November 24 and 26, 2015

The main point of this section is to present the Gauss-Bonnet Theorem.

§13.1 Area Differential Form

This is technically review, but anyways. Let x : U → R3 be a parametrization. We can
perform integrals of the form
Z Z p
f (u, v) |xu ∧ xv | = f (u, v) EG − F 2 du ∧ dv
R x−1 (R)

which is the integral of f over the region U . This is the pullback of the differential form
xu ∧ xv on the surface. For convenience, we will abbreviate this to
Z
f dσ.
R

§13.2 Global Gauss-Bonnet Theorem

We then have the following.

Theorem 13.1 (Gauss-Bonnet)

Let R ⊂ S be a regular region of an oriented surface S and let C1 , . . . , Cn be

positively oriented curves forming the boundary of R. Let θ1 , . . . , θp be the p = n2
external angles formed by the Ci . Then
n Z
X Z p
X
kg (s) ds + K dσ + θi = 2πχ(R)
i=1 Ci R i=1

where χ is the Euler characteristic.

Corollaries:

Corollary 13.2 (Gauss-Bonnet for Simple Regions)

If R is a simple region, then χ(R) = 1 and thus
n Z
X Z p
X
kg (s) ds + K dσ + θi = 2π.
i=1 Ci R i=1

Corollary 13.3 (Gauss-Bonnet for Orientable Compact Surfaces)

If S is a orientable compact surface then
Z
K dσ = 2πχ(S).
S

Proof. Take R = S, and use an empty boundary.

34
Evan Chen (Fall 2015) 14 December 1 and December 3, 2015

§14 December 1 and December 3, 2015

§14.1 The Exponential Map
Let S be a regular surface and p ∈ S a point. We consider the tangent plane Tp (S). For
a vector v ∈ Tp (S), we can consider the geodesic γv with γv0 (0) = v. In the event that
γ(1) is defined, we define the exponential map by

expp (v) = γv (1)

and set expp (0) = p.

Example 14.1 (Examples of the Exponential Map)

Let S = S 2 be a sphere, and p ∈ S.

(a) expp (v) is defined for each v ∈ Tp (S). If |v| = (2k + 1)π, then expp (v) = −p,
while if |v| = 2kπ then expp (v) = p.

(b) If we delete −p from S then expp is only defined in an open disk of radius π.

Proposition 14.2 (expp Defined Locally)

There exists a neigborhood U of 0 ∈ Tp (S) such that expp is defined, meaning we
have a map
expp : U → S,
and moreover expp is a diffeomorphism.

We say that V ⊂ S is a normal neighborhood of p if V is the diffeomorphic image

under expp of some neighborhood U of Tp (S).

§14.2 Local Coordinates

We can impose coordinates on normal neighborhoods of a point via the exponential map,
since it’s a diffeomorphism.
If we put rectangular coordinates (u, v), then the geodesics of S correspond to straight
lines through the origin.
Now consider a polar coordinate system with 0 < θ < 2π and radius ρ, so pairs (ρ, θ).
Note this omits the half-line ` corresponding to θ = 0. The images of circles at U
(centered at zero) in S are then called the geodesic circles of V while the images of lines
through 0 are called the radial geodesics; these correspond to setting ρ and θ constant.

Theorem 14.3 (First Fundamental Form in Geodesic Polar Coordinates)

Let x : U \ ` → V \ L be a system of geodesic polar coordinates (ρ, θ). Then the
first fundamental form obeys

K(p) 3
E = 1, F = 0, G(ρ, θ) = ρ − ρ + o(ρ3 ).
6

The fact that F = 0 reflects the so-called Gauss lemma: radial geodesics are orthogonal
to the geodesic circles.

35
Evan Chen (Fall 2015) 15 December 8 and December 10, 2015

§15 December 8 and December 10, 2015

Our goal is to prove the Bonnet Theorem:

Theorem 15.1 (Bonnet’s Theorme)

Let S be a complete surface. Suppose there exists a δ > 0 such that the Gaussian
curvature K is bounded below by δ. Then S is compact and any two points are
joined by a geodesic with length at most √πδ .

This is “optimal” by taking a sphere. (Definition of “complete” in next section.)

§15.1 Completeness
Definition 15.2. A regular connected surface S is complete if for any point p ∈ S the
exponential map expp : Tp (S) → S is defined for every v ∈ Tp (S).

Theorem 15.3 (Complete Surfaces are Nonextendible)

A complete surface S is nonextendible, i.e. it is not a proper subset of any regular
connected surface.

Theorem 15.4 (Closed Surfaces Are Complete)

Any closed surface S ⊆ R3 is complete. In particular, any compact surface is
complete.

Complete surfaces have a nice notion of “distance”.

Theorem 15.5 (Hopf-Rinow)

If S is a complete surface then for any p, q ∈ S there is a minimal geodesic joining p
to q.

Thus we may define the distance between two points on a complete surface by considering
the minimal geodesic.

§15.2 Variations
In what follows, S is regular and connected but not necessarily complete.
To do this, consider a curve γ : [0, `] → S parametrized by arc length. A variation of
γ is a differentiable map
h : [0, `] × (−ε, ε) → S.
such that h(s, 0) = γ(s) for each s. We assume also that our variation h is proper,
meaning h(0, t) = γ(0) and h(`, t) = γ(1). (In other words, h is a path homotopy.)
This then gives rise to a vector field
 
∂
V (s) = h(s, t) .
∂t t=0

36
Evan Chen (Fall 2015) 15 December 8 and December 10, 2015

which we call the variational vector field of h. Since h is proper it follows that
V (0) = V (`) = 0.
Define L : (−ε, ε) → R by letting L(t) be the arc length of h(−, t). Formally,
Z `
∂h
L(t) = (s, t) ds.
∂s
9

Thus L(t) measures the arc length of the “nearby” curves specified by the variation h.

Theorem 15.6 (The First Variation)

We have Z `
0
L (0) = − hA(s), V (s)i ds
0
where  
D ∂h
A(s) = (s, 0) .
∂s ∂s

Proof. Some direct computation. You need the fact that V (`) = V (0) = 0.

Remark 15.7. The vector A(s) is the acceleration vector of α and kA(S)k is the
geodesic curvature of α.

Now we can formally write down that geodesics are “locally minimal”.

Theorem 15.8 (Geodesics Are Minimal For Any Variation)

Given α : [0, `] → S parametrized by arc length we have α is a geodesic if and only
if for any proper variation h of α, we have L0 (0) = 0.

Proof. If α is a geodesic, then A(s) ≡ 0 and the result is immediate.

Otherwise, support V (s) = f (s)A(s) for some nonnegative “weight” function f sup-
ported on (0, `) where f (0) = f (`) = 0. By differential equations, we can construct h
with this variation. Now according to the previous result,
Z ` Z `
0
L (0) = − hA(s), A(s)f (s)i ds = − f (s) kA(s)k2 ds ≤ 0
0 0

which can only be zero if A(s) ≡ 0.

§15.3 The Second Variation

From now on, assume γ : [0, `] → S is arc length, V (s) is proper, and V (0) = V (`) = 0,
and moreover that we have an orthogonality relation

V (s), γ 0 (s) = 0.



In other words we assume V is an orthogonal variation.

We aim to compute L00 (0).

37
Evan Chen (Fall 2015) 15 December 8 and December 10, 2015

Lemma 15.9 (Partials Don’t Commute)

Let x : U → S be a parametrization at p ∈ S. Let K be the Gaussian curvature at
p. Then
D D D D
xu − xu = K(xu × xv ) × xu .
∂v ∂u ∂u ∂v

Proof. More direct computation with Christoffel symbols.

Remark 15.10. Philosophically, “curvature is something you can see by changing the
orders of derivatives”.

Lemma 15.11
∂h
Let h : [0, `] × (−ε, ε) → S, and let W (s, t) = ∂t (s, t) be a differential vector field
along h. Then we have
 
DD DD ∂h ∂h
W− W = K(s, t) × × W.
∂t ∂s ∂s ∂t ∂s ∂t

Theorem 15.12 (The Second Variation)

Suppose γ is a geodesic parametrized by arc length, and h : [0, `] × (−ε, ε) → S be a
proper orthogonal variation of of γ. Letting V be its variational vector field,
Z ` 2 !
00
D 2
L (0) = V (s) − K(s) kV (s)k
∂s ds
0

where K(s) is the Gaussian curvature at h(s, 0) = γ(s).

Proof. We take the derivative of the formula

Z `
D ∂h ∂h

0 ∂s ∂t (s, t), ∂s (s, t)
L (t) =
∂h  12
0 ∂h
∂s (s, t), ∂s (s, t),

that we previously obtained. Move the differential operator through the integral sign.
This gives us, by repeatedly applying Product Rule,
Z ` "  2  #
D D ∂h ∂h D D ∂h
L00 (0) =

(s, 0), (s, 0) +
∂s V (s) − ∂s V (s), ∂s (s, 0)
ds.
0 ∂t ∂s ∂t ∂s

Do some further computation now.

§15.4 Proof of Bonnet Theorem

Now that we have L0 (0) and L00 (0) we can prove Bonnet’s Theorem.
Assume S is complete, and K ≥ δ at all points of S. Consider two points p and q such
that the distance p, q is maximized, and let γ : [0, `] → S be the minimal geodesic which
joins them (permissible by the Hopf-Rinow Theorem). Now assume for contradiction
` > √πδ Thus we have L0 (0) = 0. But if we can show L00 (0) < 0, then we will have shown

38
Evan Chen (Fall 2015) 15 December 8 and December 10, 2015

γ is in fact a local maximum and this will be a contradiction. (To see this: imagine the
long geodesic joining two points on a sphere. While locally the shortest path, globally it
is longer than any perturbation.)
Consider proper variations V ; we have
Z ` " 2 #
D
V (s) − K(s) kV (s)k2 ds.
L00 (0) =

∂s
0

Pick a tangent vector w0 ∈ Tp (S) such that w0 ⊥ γ 0 (0) and kw0 k = 1, and let w(t) be a
parallel transport of w0 along γ. Then we define V (s) again by
π 
V (s) = sin s w(s).
`
As V (0) = V (`) = 0, we again can construct an h corresponding to it. Then direct
calculation gives
D π π π D
V (s) = − cos( s)w(s) + sin( s) w(s)
∂s ` ` | ` {z∂s }
=0
2 2
D
V (s) = π cos2 ( π s)

∂s `2 `
π
kV (s)k2 = sin2 ( s).
`
π2
If K ≥ δ > `2
, then we have
Z ` Z `
00 π2 π  π  π2 π 
L (0) < − 2 cos ( s) − sin2
2
s ds < − 2 cos(2 s) ds = 0
` 0 ` ` ` 0 `

which is the desired contradiction.

The choice of sin is designed for the ending. If we had used an arbitrary function f as
before, we would have hf 00 (s) + k(s)f (s), f (s)i. So we choose a function f for which this
becomes a zero integral.

39
Evan Chen (Fall 2015) A Examples

§A Examples
Here is a collection of facts about some common examples. Recall for convenience that
1 eG − 2F f + gE
K = EG − F 2 and H = .
2 EG − F 2

§A.1 Sphere
Parametrize the unit sphere S 2 in R3 by x(u, v) = (R cos u cos v, R cos u cos v, R sin v).

E = R2 , F = 0, G = (R cos u)2

e = R, f = 0, g = R cos2 u.
The Gauss curvature at all points is the constant K = R12 . The mean curvature for this
parametrization is H = R1 . The geodesics are the great circles.
The genus of S 2 is zero, so its Euler characteristic is 2. This surface is a compact
oriented manifold without boundary.

§A.2 Torus
Parametrize a torus by x(u, v) = ((a + r cos u cos v), (a + r cos u sin v), r sin u).

E = r2 , F = 0, G = (a + r cos u)2

e = r, f = 0, g = cos u(a + r cos u).

The Gauss curvature is
cos u
K= .
r(a + r cos u)
The mean curvature is
a + 2r cos u
H= .
2r(a + r cos u)
The genus of S 1 × S 1 is 1, so its Euler characteristic is 0. This surface is a compact
oriented manifold without boundary.

§A.3 Cylinder
Parametrize the cylinder by x(u, v) = (R cos u, R sin u, v). Then

E = R2 , F = 0, G = 1.

e = −R, f = 0, g = 0.
The Gauss curvature is K = 0 everywhere. The mean curvature at all points is − R1 . The
geodesics are helixes, including meridians and vertical lines.
This surface is not compact, but it is orientable, with empty boundary. It has genus 1
and Euler characteristic zero.

40