Differential Geometry, Gabriel-Paternain

DIFFERENTIAL GEOMETRY, D COURSE
Gabriel P. Paternain
Department of Pure Mathematics and Mathematical Statistics,
University of Cambridge, Cambridge CB3 0WB, England
E-mail address: [email protected]
Contents
Chapter 1. Smooth manifolds and smooth maps 1
1.1. Denitions 1
1.2. Tangent spaces and derivatives 2
1.3. Regular values and Sards theorem 3
1.4. Transversality 5
1.5. Manifolds with boundary 6
1.6. Degree modulo 2 7
1.7. Abstract manifolds and Whitneys embedding theorem 11
Chapter 2. Length, area and curvature 13
2.1. Arc-length, curvature and torsion of curves 13
2.2. The isoperimetric inequality in the plane 14
2.3. First fundamental form and Area 16
2.4. The Gauss map 19
2.5. The second fundamental form in local coordinates 20
2.6. Theorema Egregium 21
Chapter 3. Critical points of length and area 23
3.1. Geodesics 23
3.2. Covariant derivative and parallel transport 24
3.3. Minimal surfaces 25
3.4. The Weierstrass representation 26
3.5. The meaning of g in the Weierstrass representation 29
Chapter 4. Global Riemannian geometry 31
4.1. The exponential map and geodesic polar coordinates 31
4.2. The Gauss-Bonnet theorem 33
4.3. Applications 38
4.4. Fenchels theorem 39
4.5. The Fary-Milnor theorem 40
3
CHAPTER 1
Smooth manifolds and smooth maps
1.1. Denitions
Definition 1.1. Let U be an open set of R
n
. A map f : U R
m
is called
smooth if it has continuous partial derivatives of all orders. A map f : X R
m
dened on an arbitrary subset X of R
n
is called smooth if for each x X there is
an open set U R
n
containing x and a smooth map F : U R
m
such that F
equals f on U X.
We will use the shorthand term local when we wish to refer to behaviour only
in a neighbourhood of a point. Smoothness is a local property. (The term global
refers to the whole space X.)
Definition 1.2. A smooth map f : X Y between subsets of Euclidean
space is a dieomorphism if it is a bijection, and if the inverse f
1
: Y X is also
smooth. X and Y are dieomorphic if such an f exists.
Dierential Topology is about properties of a set X R
n
which are invariant
under dieomorphisms.
Arbitrary sets of R
n
may be too wild. We would like to have a class of sets on
which can do locally the same as in Euclidean space.
Definition 1.3. Let X be a subset of R
N
. We say that X is a k-dimensional
manifold if each point possesses a neighbourhood V in X which is dieomorphic to
an open set of R
k
. A dieomorphism : U V , where U is an open set of R
k
,
is called a parametrization of the neighbourhood V . The inverse dieomorphism
1
: V U is called a coordinate system or a chart on V .
If we write
1
= (x
1
, . . . , x
k
) the k smooth functions x
1
, . . . , x
k
on V are
called coordinate functions. The dimension k of X is written as dim X.
Example 1.4. The unit sphere S
2
given by all (x, y, z) R
3
with x
2
+y
2
+z
2
=
1 is a smooth manifold of dimension 2. The dieomorphism
(x, y) (x, y,
_
1 x
2
y
2
),
for x
2
+y
2
< 1, parametrizes the region z > 0 of S
2
. By interchanging the roles of x,
y and z and changing the sign of the variables, we obtain similar parametrizations
of the regions x > 0, y > 0, x < 0, y < 0 and z < 0. Since these cover S
2
, it follows
that S
2
is a smooth manifold. Similarly S
n
:= x R
n+1
: [x[ = 1 is a smooth
manifold of dimension n.
Exercise 1.5. If X and Y are manifolds, so is X Y , and dim X Y =
dim X + dim Y .
1
2 1. SMOOTH MANIFOLDS AND SMOOTH MAPS
Definition 1.6. If X and Z are both manifolds in R
N
and Z X, then Z is
a submanifold of X. In particular, X itself is a submanifold of R
N
. Any open set
of X is a submanifold of X. The codimension of Z in X is dim X dim Z.
1.2. Tangent spaces and derivatives
For any open set U R
n
and x U, the tangent space to U at x (denoted T
x
U)
is dened to be R
n
. Recall from Analysis II that for any smooth map f : U R
m
,
the derivative of f at x U is the linear map
df
x
: R
n
R
m
given by
df
x
(h) = lim
t0
f(x +th) f(x)
t
for h R
n
.
Now let us dene the tangent space T
x
X of an arbitrary smooth manifold X
R
N
. Choose a parametrization : U X around x where U is an open set of R
k
.
Without loss of generality assume that (0) = x. Think of as a map from U to
R
N
, so d
0
: R
k
R
N
is dened. Set
T
x
X := d
0
(R
k
).
Lemma 1.7. T
x
X does not depend on and dim T
x
X = k.
Proof. Suppose : V X is another choice, with (0) = x. By shrinking
both U and V , we may assume that (U) = (V ). Then the map h =
1
:
U V is a dieomorphism. Now we write = h and we dierentiate. Using
the chain rule we have: d
0
= d
0
dh
0
. Since dh
0
is an invertible linear map, it
follows at once that d
0
and d
0
have the same image.
Since
1
: (U) U is a smooth map, we can choose an open set W in R
N
containing x and a smooth map : W R
k
that extends
1
. Then is the
identity map of U, so the chain rule implies that the composition of linear maps
R
k
d0
T
x
X
dx
R
k
is the identity map of R
k
. Thus d
0
: R
k
T
x
X is an isomorphism and dim T
x
X =
k.
We can now dene the derivative of a smooth map f : X Y between arbitrary

manifolds. Let x be a point in X and set y = f(x). The derivative must be a linear
map df
x
: T
x
X T
y
Y which gives us back the usual derivative when X and Y are
open sets in Euclidean space and it must also satisfy the chain rule.
Keeping this in mind, let : U X and : V Y be parametrizations
around x and y respectively, where U is an open set of R
k
and V is an open set of
R
l
. As before suppose (0) = x and (0) = y. If U is small enough we have the
diagram:
X
f
Y
U
h=
1
f
V
1.3. REGULAR VALUES AND SARDS THEOREM 3
Hence our denition of df
x
must be so that the following diagram commutes:
T
x
X
dfx
T
y
Y
d0
d0
R
k
dh0
R
l
Since d
0
is an isomorphism, we must have
df
x
:= d
0
dh
0
d
1
0
.
We must verify that this denition does not depend on the choices of and .
Exercise 1.8. Prove that this is indeed the case.
Chain Rule. If X
f
Y
g
Z are smooth maps of manifolds, then
d(g f)
x
= dg
f(x)
df
x
.
1.2.1. The inverse function theorem. Let f : X Y be a smooth map
between manifolds. We say that f is a local dieomorphism at x if f maps a
neighbourhood of x dieomorphically onto a neighbourhood of f(x).
Theorem 1.9 (The inverse function theorem). Suppose that f : X Y is a
smooth map whose derivative df
x
at the point x is an isomorphism. Then f is a
local dieomorphism at x.
Exercise 1.10. Prove the theorem assuming the inverse function theorem for
smooth functions between open sets of Euclidean space, as you have seen it in
Analysis II.
1.3. Regular values and Sards theorem
Let f : X Y be a smooth map between manifolds. Let C be the set of all
points x X such that df
x
: T
x
X T
f(x)
Y is not surjective.
Definition 1.11. A point in C will be called a critical point. A point in f(C)
will be called a critical value. A point in the complement of f(C) will be called a
regular value.
Remark 1.12. Note that if dim X < dim Y , then C = X and the preimage of
a regular value is the empty set.
Theorem 1.13 (Preimage theorem). Let y be a regular value of f : X Y with
dim X dim Y . Then the set f
1
(y) is a submanifold of X with dim f
1
(y) =
dim X dim Y .
Proof. Let x f
1
(y). Since y is a regular value, the derivative df
x
maps
T
x
X onto T
y
Y . The kernel of df
x
is a subspace K of T
x
X of dimension p :=
dim X dim Y . Suppose X R
N
and let T : R
N
R
p
be any linear map such
that Ker(T) K = 0. Consider the map F : X Y R
p
given by
F(z) = (f(z), T(z)).
The derivative of F is given by
dF
x
(v) = (df
x
(v), T(v))
which is clearly nonsingular by our choice of T. By the inverse function theorem,
F is a local dieomorphism at x, i.e. F maps some neighbourhood U of x dif-
feomorphically onto a neighbourhood V of (y, T(x)). Hence F maps f
1
(y) U
dieomorphically onto (y R
p
) V which proves that f
1
(y) is a manifold with
dim f
1
(y) = p.
The case dim X = dim Y is particularly important. The theorem says that if
y is a regular value of f, then f
1
(y) is a 0-dimensional manifold, i.e. collection of
points. If X is compact, this collection must be nite (why?) so we obtain:
Corollary 1.14. Let f : X Y be a smooth map between manifolds of the
same dimension. If X is compact and y is a regular value, f
1
(y) consists of a
nite set of points.
We can actually say a bit more:
Theorem 1.15 (Stack of records theorem). Let f : X Y be a smooth map
between manifolds of the same dimension with X compact. Let y be a regular value
of f and write f
1
(y) = x
1
, . . . , x
k
. Then there exists a neighbourhood U of
y in Y such that f
1
(U) is a disjoint union V
1
V
k
, where V
i
is an open
neighbourhood of x
i
and f maps each V
i
dieomorphically onto U.
Proof. By the inverse function theorem we can pick disjoint neighbourhoods
W
i
of x
i
such that f maps W
i
dieomorphically onto a neighbourhood of y. Observe
that f(X
i
W
i
) is a compact set which does not contain y. Now take
U :=
i
f(W
i
) f(X
i
W
i
).
If we let #f
1
(y) be the cardinality of f
1
(y), the theorem implies that the
function y #f
1
(y) is locally constant as y ranges over regular values of f.
The Preimage theorem gives a particularly nice way of producing manifolds.
For example, if we consider the map f : R
n+1
R given by f(x) = [x[
2
we
can check easily that 1 is a regular value of f (do it!). Hence S
n
= f
1
(1) is a
smooth manifold of dimension n. By switching our mental wavelength a little we
can produce quite interesting examples as follows:
Example 1.16 (Orthogonal group). Let O(n) be the group of orthogonal ma-
trices of size n n, i.e., a matrix A O(n) if and only if AA
t
= I, where I is the
identity matrix. The space of all n n
matrices M(n) is just R
n
2
, so we can think of O(n) as living inside R
n
2
. We
will show that O(n) is a manifold of dimension
n(n1)
2
.
Let S(n) M(n) be the space of all symmetric matrices. Since it is a vector
space, is clearly a submanifold of M(n) and it has dimension
n(n+1)
2
.
Let f : M(n) S(n) be the smooth map f(A) = AA
t
. Since O(n) = f
1
(I)
it suces to show that I is a regular value of f. We compute:
df
A
(H) = lim
s0
f(A+s H) f(A)
s
= lim
s0
(A+s H)(A+s H)
t
AA
t
s
= H A
t
+AH
t
.
1.4. TRANSVERSALITY 5
Fix A with AA
t
= I. We must prove that given any C S(n) = T
I
S(n) there is
H M(n) = T
A
M(n) such that
df
A
(H) = H A
t
+AH
t
= C.
Since C is symmetric, we can write C =
1
2
C+
1
2
C
t
so if we can solve H A
t
=
1
2
C we
are done. Multiplying by A on the right and using AA
t
= I we obtain H =
1
2
C A
which solves df
A
(H) = C. Thus I is a regular value of f.
Note that O(n) is both a group and a manifold. In fact, the group operations are
smooth, that is, the maps (A, B) AB and A A
1
= A
t
are smooth. (Why?)
A group that is manifold, and whose group operations are smooth, is called a Lie
group.
1.3.1. Sards theorem. The Preimage theorem raises the question: is it easy
to nd regular values? How are abundant are they?
Recall that an arbitrary set A in R
n
has measure zero if it can be covered by a
countable number of rectangular solids with arbitrary small total volume. In other
words given > 0, there exists a countable collection R
1
, R
2
, . . . of rectangular
solids in R
n
, such that A is contained in
i
R
i
and
i
Vol(R
i
) < .
Let X be manifold. An arbitrary subset A X has measure zero if, for every local
parametrization of X,
1
(A) has measure zero in Euclidean space.
The following deep theorem, tells us that there are plenty of regular values.
Theorem 1.17 (Sards theorem, 1942). The set of critical values of a smooth
map f : X Y has measure zero.
Since a set of measure zero cannot contain a non-empty open set we obtain:
Corollary 1.18. The regular values of any smooth map f : X Y are dense
in Y .
1.3.2. Morse functions. See the rst example sheet.
1.4. Transversality
We know that the solutions of the equation f(x) = y form a smooth manifold,
provided that y is a regular value of f : X Y . Suppose now that we replace y
by a submanifold Z Y and we ask, when is f
1
(Z) a submanifold of X?
Definition 1.19. A smooth map f : X Y is said to be transversal to a
submanifold Z Y if for every x f
1
(Z) we have
Image(df
x
) +T
f(x)
Z = T
f(x)
Y.
We write f Z.
Note that if Z = y, the notion of transversality reduces to the notion of
regular value.
With this denition, we can now state a more general version of the Preimage
theorem.
Theorem 1.20. If the smooth map f : X Y is transversal to a submanifold
Z Y , then f
1
(Z) is submanifold of X. Moreover, f
1
(Z) and Z have the same
codimension.
Proof. (Non-examinable, sketch only.) It is not hard to see that Z can be
written in a neighbourhood of a point y = f(x) as the zero set of a collection of
functions h
1
, . . . , h
r
, where r is the codimension of Z in Y . Let H := (h
1
, . . . , h
r
).
Then near x, f
1
(Z) is the zero set of the function H f. Thus if 0 R
r
is a
regular value of H f we are done. But dH
y
df
x
is surjective if and only if
Image(df
x
) +T
f(x)
Z = T
f(x)
Y
since dH
y
: T
y
Y R
r
is onto with kernel T
y
Z.
Example 1.21. Consider f : R R
2
given by f(t) = (0, t) and let Z be the
x axis in R
2
. Then f is transversal to Z, but for example the map h(t) = (t, t
2
) is
not.
An important special case occurs when f is the inclusion of a submanifold X of
Y and Z is another submanifold of Y . In this case the condition of transversality
reduces to
T
x
X +T
x
Z = T
x
Y
for every x X Z. This condition is quite easy to visualize and when it holds we
say that X and Z are transversal (we write X Z). We now have:
Theorem 1.22. The intersection of two transversal submanifolds of Y is a
submanifold of codimension given by
codim(X Z) = codimX + codimZ.
One of the main virtues of the notion of transversality is its stability i.e. it
survives after small perturbations of the map f. You can convince yourself of this
property by looking at pictures of transversal submanifolds of Euclidean space.
Another very important virtue is genericity: smooth maps may be deformed by
arbitrary small amounts into a map that is transversal to Z. (This is non-obvious
and we refer to Chapter 2 in the book by Guillemin and Pollack for details.)
1.5. Manifolds with boundary
Consider the closed half-space
H
k
:= (x
1
, . . . , x
k
) R
k
: x
k
0.
The boundary H
n
is dened to be the hyperplane x
k
= 0 in R
k
.
Definition 1.23. A subset X R
N
is called a smooth k-manifold with bound-
ary if each x X has a neighbourhood dieomorphic to an open set in H
k
. As
before, such a dieomorphism is called a chart on X. The boundary of X, denoted
X, is given by the set of points that belong to the image of H
k
under some local
parametrization. Its complement is called the interior of X, Int(X) = X X.
Remark 1.24. Warning: do not confuse the boundary or interior of X as
dened above with the topological notions of interior and boundary as a subset of
R
N
.
The tangent space is dened as before, so that T
x
X is a k-dimensional vector
space even for points x X. The interior of X is a k-manifold without boundary
and X is a manifold without boundary of dimension k 1 (this requires a proof!).
Here is an easy way of generating examples.
1.6. DEGREE MODULO 2 7
Lemma 1.25. Let X be a manifold without boundary and let f : X R be a
smooth function with 0 as a regular value. Then the subset x X : f(x) 0 is
a smooth manifold with boundary equal to f
1
(0).
Proof. The set where f > 0 is open in X and is therefore a submanifold of
the same dimension as X. For a point x X with f(x) = 0, the same proof of
the Preimage Theorem 1.13 shows that x has a neighbourhood dieomorphic to a
neighbourhood of a point in H
k
.
As an easy application of the lemma, consider the unit ball B

k
given by all
x R
k
such that [x[ 1. By considering the function f(x) = 1 [x[
2
it follows
that B
k
is a smooth manifold with boundary S
k1
.
Theorem 1.26. Let f : X Y be a smooth map from an m-manifold with
boundary to an n-manifold, where m > n. If y is a regular value, both for f and for
the restriction of f to X, then f
1
(y) is a smooth (mn)-manifold with boundary
equal to f
1
(y) X.
Proof. Recall that being a submanifold is a local property so without loss of
generality we can suppose that f : H
m
R
n
with y R
n
a regular value. Consider
z f
1
(y). If z belongs to the interior of H
m
, then as in the Preimage theorem
1.13, f
1
(y) is a smooth manifold near z.
Let z be now in H
m
. Since f is smooth, there is a neighbourhood U of z in
R
m
and a smooth map F : U R
n
such that F restricted to U H
m
is f. By
shrinking U if necessary we can assume that F has no critical points in U (why?).
Hence F
1
(y) is a smooth manifold of dimension mn.
Let : F
1
(y) R be the projection (x
1
, . . . , x
m
) x
m
. Observe that the
tangent space of F
1
(y) at a point x
1
(0) is equal to the kernel of
dF
x
= df
x
: R
m
R
n
.
Hence 0 must be a regular value of since we are assuming that y is a regular value
of f restricted to H
m
.
But F
1
(y) H
m
= f
1
(y) U is the set of all x F
1
(y) with (x) 0 and
by Lemma 1.25, is a smooth manifold with boundary equal to
1
(0).
It is not hard to guess what is the appropriate version of this theorem for the
more general case of a map f : X Y and a submanifold Z of Y . Suppose X
has boundary, but Y and Z are boundaryless. The next theorem is stated without
proof.
Theorem 1.27. Suppose that both f : X Y and f [
X
: X Y are
transversal to Z. Then f
1
(Z) is a manifold with boundary given by f
1
(Z) X
and codimension equal to the codimension of Z.
1.6. Degree modulo 2
Let X be a smooth boundaryless manifold. Then X [0, 1] is a manifold with
boundary X = X 0 X 1.
Definition 1.28. Two maps f, g : X Y are called smoothly homotopic if
there exists a smooth map F : X [0, 1] Y with F(x, 0) = f(x) and F(x, 1) =
g(x) for all x X. The map F is called a smooth homotopy between f and g.
The relation of smooth homotopy is an equivalence relation. (Why?) The
equivalence class to which a map belongs is its homotopy class. Let f
t
: X Y be
the 1-parameter family of maps given by f
t
(x) = F(x, t).
Definition 1.29. The dieomorphism f is smoothly isotopic to g if there exists
a smooth homotopy F : X [0, 1] Y from f to g such that for each t [0, 1], f
t
maps X dieomorphically onto Y .
Lemma 1.30 (Homotopy Lemma). Let f, g : X Y be smooth maps which
are smoothly homotopic. Suppose X is compact, has the same dimension as Y and
X = . If y is a regular value for both f and g, then
#f
1
(y) = #g
1
(y) (mod 2).
Proof. The proof relies on the following important fact.
Theorem 1.31 (Classication of 1-manifolds). Every compact connected 1-
manifold is dieomorphic to [0, 1] or S
1
.
As every compact manifold is the disjoint union of nitely many compact con-
nected manifolds we have:
Corollary 1.32. The boundary of any compact 1-manifold consists of an even
number of points.
Let F : X [0, 1] Y be a smooth homotopy between f and g. Assume for
the time being that y is also a regular value for F. Then F
1
(y) is a compact
1-manifold with boundary equal to
F
1
(y) (X 0 X 1) = f
1
(y) 0 g
1
(y) 1.
Thus the cardinality of the boundary of F
1
(y) is just #f
1
(y) + #g
1
(y). By
the corollary above
#f
1
(y) = #g
1
(y) (mod 2).
If y is not a regular value of F we proceed as follows. From the Stack of records
theorem 1.15 we know that w #f
1
(w), w #g
1
(w) are locally constant
as w ranges over regular values. Thus there are neighbourhoods V and W of y,
consisting of regular values of f and g respectively for which
#f
1
(w) = #f
1
(y)
for all w V , and
#g
1
(w) = #g
1
(y)
for all w W. By Sards theorem we can choose a regular value z of F in V W.
Then
#f
1
(y) = #f
1
(z) = #g
1
(z) = #g
1
(y) (mod 2)
as desired.
Lemma 1.33 (Homogeneity Lemma). Let X be a smooth connected manifold,

possibly with boundary. Let y and z be points in Int(X). Then there exists a
dieomorphism h : X X smoothly isotopic to the identity such that h(y) = z.
1.6. DEGREE MODULO 2 9
Proof. Let us call two points y and z isotopic if there exists a dieomor-
phism h isotopic to the identity that maps y to z. It is evident that this is an
equivalence relation. Since X is connected, it suces to show that each equivalence
class is a open set.
To prove that equivalence classes are open we will construct an isotopy h
t
of
R
k
such that h
0
is the identity, each h
t
is the identity outside the open unit ball
around the origin, and h
1
(0) is any specied point in the open unit ball. This will
suce, since we can now parametrize a small neighbourhood of y in Int(X) and use
h
t
to construct an isotopy that will move y to any point nearby.
Let : R
k
R be a smooth function such that (x) > 0 for [x[ < 1 and
(x) = 0 for [x[ 1 (why do they exist?). Given a unit vector u R
k
consider the
ordinary dierential equation in R
k
given by
dx
dt
= u(x).
Let F
t
: R
k
R
k
be the ow of this dierential equation, i.e., for each x R
k
, the
curve t F
t
(x) is the unique solution passing through x. Standard theorems on
dierential equations tell us that:
(1) F
t
is dened for all x R
k
and t R and smooth;
(2) F
0
is the identity;
(3) F
t+s
= F
t
F
s
.
Clearly for each t, F
t
leaves all points outside the unit ball xed. For appropriate
choices of u and t, F
t
will map the origin to any point in the open unit ball.
(I may replace this proof by the one given by Guillemin and Pollack on page
142 of their book which does not use ows.)
In what follows suppose that X is compact and without boundary and Y is

connected and with the same dimension as X. Let f : X Y be a smooth map.
Theorem 1.34 (Degree mod 2). If y and z are regular values of f then
#f
1
(y) = #f
1
(z) (mod 2).
This common residue class is called the degree mod 2 of f (denoted deg
2
(f)) and
only depends on the homotopy class of f.
Proof. Given y and z, let h be the dieomorphism smoothly isotopic to the
identity such that h(y) = z given by the Homogeneity Lemma. Observe that z is
also a regular value of h f. Since h f is homotopic to f, the Homotopy Lemma
tells us that
#(h f)
1
(z) = #f
1
(z) (mod 2).
But (h f)
1
(z) = f
1
(h
1
(z)) = f
1
(y) and therefore
#f
1
(y) = #f
1
(z) (mod 2)
as desired. Let g be smoothly homotopic to f. By Sards theorem, there is a point
y Y which is a regular value for both f and g (why?). By the Homotopy Lemma
deg
2
(f) = #f
1
(y) (mod2) = #g
1
(y) (mod2) = deg
2
(g)
which completes the proof.

Example 1.35. The identity map of a compact boundaryless manifold X has
deg
2
= 1 and the constant map has deg
2
= 0. Therefore they are never homotopic.
When X = S
n
, this implies that there is no smooth map f : B
k+1
S
k
which
restricts to the identity on S
k
(i.e. there is no retraction). Indeed, if such a map
exists it would give rise to a homotopy F : S
k
[0, 1] S
k
, F(x, t) = f(tx) which
is a homotopy between the constant map and the identity.
The previous example yields the smooth version of a famous xed point theo-
rem.
Theorem 1.36 (Smooth Brouwer xed point theorem). Any smooth map f :
B
k
B
k
has a xed point.
Proof. Suppose f has no xed point. We will construct a map g : B
k
S
k1
which restricts to the identity on S
k1
which contradicts Example 1.35. For x B
k
,
let g(x) S
k1
be the point where the line segment starting at f(x) and passing
through x hits the boundary. One can write a formula for g to show smoothness.
In fact, any continuous map f : B
k
B
k
has a xed point (Brouwer xed
point theorem). One can prove this using the smooth version approximating f
by polynomials. Indeed, by the Weierstrass approximation theorem, given > 0,
there is a polynomial P such that [f(x) P(x)[ < for all x B
k
. Set Q(x) =
P(x)/(1 + ). Now Q maps B
k
into B
k
and [f(x) Q(x)[ < 2 for all x B
k
.
Suppose f(x) ,= x for all x. Then the continuous function [f(x) x[ must take a
positive minimum on B
k
. If now let = /2 the Q from above does not have a
xed point, which contradicts the smooth Brouwer xed point theorem.
1.6.1. Intersection numbers modulo 2. What we said above about degree
modulo 2 can be generalized to the case of a smooth map f : X Y and Z a
submanifold of Y . Suppose that:
(1) X is compact without boundary;
(2) Z is closed and without boundary;
(3) f Z;
(4) dim X + dim Z = dim Y .
Under these conditions f
1
(Z) is a closed 0-dimensional submanifold of X and
hence it consists of nitely many points. Dene the mod 2 intersection number of
the map f with Z, I
2
(f, Z), to be the cardinality of f
1
(Z) modulo 2. An analogue
of the Homotopy Lemma (for intersection numbers) also holds here: if f
0
and f
1
are maps transversal to Z and homotopic, then I
2
(f
0
, Z) = I
2
(f
1
, Z). In fact, we
can dene I
2
(f, Z) for a map f which is not necessarily transversal to Z. Using
that transversality is generic, we can nd a map g homotopic to f such that g Z
and we now set I
2
(f, Z) := I
2
(g, Z). It does not matter which g we choose as long
as g is homotopic to f, thanks to the Homotopy Lemma for intersection numbers.
Let us have a closer look at the important special case in which X itself is a
compact submanifold of Y and Z a closed submanifold of complementary dimension.
In this case f is the inclusion map X Y , and if X Z, I
2
(X, Z) := I
2
(f, Z) is
just #(XZ) mod2. If I
2
(X, Z) ,= 0, it means that no matter how X is deformed,
we cannot move it away completely from Z.
Example 1.37. Let Y be the 2-torus S
1
S
1
. Let X = S
1
1 and Z =
1 S
1
. Then I
2
(X, Z) = 1 and we cannot smoothly pull the circles apart.
1.7. ABSTRACT MANIFOLDS AND WHITNEYS EMBEDDING THEOREM 11
When dim X =
1
2
dim Y , we can consider I
2
(X, X) the self-intersection number
modulo 2. If X is the central curve in a Mobius band, then I
2
(X, X) = 1.
1.7. Abstract manifolds and Whitneys embedding theorem
One can actually dene manifolds without making any reference to the ambient
space R
N
. You will study manifolds in this more abstract setting in Part III and
in the Riemann Surfaces course. In any case, here is the denition:
Definition 1.38. An n-dimensional smooth manifold is a second countable
Hausdor space X together with a collection of maps called charts such that:
(1) a chart is a homeomorphism : U (U), where U is open in X and
(U) is open in R
n
;
(2) each point x X belongs to the domain of some chart;
(3) for charts : U (U) R
n
and : V (V ) R
n
, the map

1
: (U V ) (U V ) is smooth;
(4) the collection of charts is maximal with respect to the properties above.
A set of charts satisfying the rst three properties is called an atlas. An atlas
can always be enlarged uniquely to give a maximal atlas as in the denition. If in
the denition we require the maps
1
just to be of class C
k
(k 1) then we
say that we have a manifold of class C
k
. (Recall that a map is of class C
k
if it has
continuous partial derivatives up to order k.)
The denition is set up so that it is plain how to dene a smooth map between
manifolds. An immersion is a smooth map f : X Y such that df
x
is injective for
all x X. A submersion is a smooth map f : X Y such that df
x
is surjective
for all x X. An embedding is a smooth map f : X Y which is an immersion
and a homeomorphism onto its image.
The next theorem tells us that we did not lose much by restricting out attention
to manifolds as subsets of Euclidean space.
Theorem 1.39 (Whitneys embedding theorem). A smooth n-manifold X can
be embedded into R
2n+1
.
The proof of this theorem can be found in most books about manifolds. In
fact, Whitney proved the much harder result that X can be embedded in R
2n
.
CHAPTER 2
Length, area and curvature
2.1. Arc-length, curvature and torsion of curves
Definition 2.1. Let I R be an interval and let X be a manifold. A curve in
X is a smooth map : I X. The curve is said to be regular if is an immersion,
i.e., if the velocity vector (t) := d
t
(1) T
(t)
X is never zero.
If X = R
2
, R
3
(or any Riemannian manifold) we can talk about the arc-length
of a regular curve. By denition, given t I, the arc-length of : I R
3
from the
point t
0
is given by
s(t) :=
_
t
t0
[ ()[ d.
If the interval I has endpoints a and b, a < b, the length of is
() :=
_
b
a
[ (t)[ dt.
The curve is said to be parametrized by arc-length if [ (t)[ = 1 for all t
I. Since the curve is regular, the function t s(t) is strictly increasing, and
therefore there exists a smooth inverse t = t(s). The new curve s (t(s)) is now
parametrized by arc-length and has the same image and the same length as .
From now on we will assume that curves are parametrized by arc-length.
Definition 2.2. Let : I R
3
be a curve parametrized by arc-length. The
curvature of at s I is the number k(s) := [ (s)[. If k(s) ,= 0, the unit vector
n(s) in the direction of (s) is well dened by the equation (s) = k(s) n(s).
Dierentiating [ [
2
= 1 we see that , ) = 0 and thus n(s) is normal to (s) and
is called the normal vector at s. The plane determined by (s) and n(s) is called
the osculating plane at s.
Traditionally one denotes by t(s). The unit vector b(s) := t(s) n(s), where
denotes the usual cross product in R
3
, is normal to the osculating plane and is
called the binormal vector at s. Note
b(s) =

t(s) n(s) +t(s) n(s) = t(s) n(s),
thus we may write
b(s) = (s) n(s)

for some function (s). (Note that b,

b) = 0 since b has unit norm.)
Definition 2.3. The number (s) is called the torsion of the curve at s.
For each point s for which k(s) ,= 0, we have three orthonormal vectors t(s), n(s)
and b(s) which form the so called Frenet trihedron at s. Observe that since n = bt,
13
14 2. LENGTH, AREA AND CURVATURE
dierentiating with respect to s we obtain:
n(s) =

b(s) t(s) +b(s)

t(s) = (s)b(s) k(s)t(s).
Thus we have obtained the following three equations, called the Frenet formulas:
t = k n,
n = k t b,
b = n.
We can think of a curve in R
3
as being obtained from a straight line by bending
(curvature) and twisting (torsion). It is not hard to guess that curvature and torsion
determine the curve locally. Indeed we have:
Theorem 2.4 (Fundamental theorem of the local theory of curves). Given
smooth functions k(s) > 0 and (s), s I, there exists a regular curve : I R
3
such that s is arc-length, k(s) is the curvature, and (s) is the torsion of . More-
over any other curve , satisfying the same conditions, diers from by an isom-
etry; that is, there exists an orthogonal linear map T, with positive determinant,
and a vector a such that = T +a.
You are invited to give a proof by using the Frenet formulas and the theorem
on existence and uniqueness of solutions of ordinary dierential equations that you
saw in Analysis II.
Remark 2.5. If is a plane curve, that is, (I) is contained in a plane, then
0. In this case the proof of the fundamental theorem is actually quite simple.
For plane curves : I R
2
, one can give a sign to the curvature.
Let e
1
, e
2
be the standard basis of R
2
and dene the normal vector n(s),
s I, by requiring the basis t(s), n(s) to have the same orientation as e
1
, e
2
.
The curvature k is then dened by the equation:
t(s) = k(s) n(s)

which might be either positive or negative. The absolute value of k coincides with
the previous denition.
2.2. The isoperimetric inequality in the plane
There are many proofs of the isoperimetric inequality. The one we present
here uses a bit of Fourier analysis and is due to A. Hurwitz (1901). This is our
rst example of a global result, and its extensions and ramications are of central
importance in Dierential Geometry.
Let R
2
be a domain, that is, a connected open set. We will assume that
has compact closure and that its boundary is a connected 1-manifold of class
C
1
. We will denote by A() the area of .
Theorem 2.6 (The isoperimetric inequality in the plane). Let be as above.
Then
2
() 4 A(),
with equality if and only if is a disk.
We will need a preliminary lemma, which is an exercise in Fourier series.
2.2. THE ISOPERIMETRIC INEQUALITY IN THE PLANE 15
Lemma 2.7 (Wirtingers Inequality). Let f : R R be a C
1
function which is
periodic with period L. Suppose
_
L
0
f(t) dt = 0.
Then
_
L
0
[f
[
2
(t) dt
4
2
L
2
_
L
0
[f[
2
(t) dt,
with equality if and only if there exist constants a
1
and a
1
such that
f(t) = a
1
e
2it/L
+a
1
e
2it/L
.
Proof. Consider the Fourier series expansions of f and f
:
f(t) =
k=
a
k
e
2ikt/L
,
f
(t) =
k=
b
k
e
2ikt/L
where
a
k
=
1
L
_
L
0
f(t)e
2ikt/L
dt
b
k
=
1
L
_
L
0
f
(t)e
2ikt/L
dt.
The hypothesis implies a
0
= 0. Also note that
b
0
=
1
L
_
L
0
f
(t) dt =
1
L
(f(L) f(0)) = 0.
Integration by parts gives
b
k
=
2ik
L
a
k
for all [k[ 1. Parsevals identity gives:
_
L
0
[f
[
2
dt = L
k=0
[b
k
[
2
=
4
2
L
k=0
k
2
[a
k
[
2
4
2
L
k=0
[a
k
[
2
=
4
2
L
2
_
L
0
[f[
2
dt.
Equality holds if and only if a
k
vanishes for all k with [k[ > 1.
Proof of Theorem 2.6. By translating if necessary, we can assume that

_
X ds = 0
where X(x, y) = (x, y) R
2
. Let us apply the 2-dimensional divergence theorem
to the vector eld X in the domain . Let n be the outward unit normal vector
eld along . The divergence theorem says
_
divX dA =
_
X, n) ds.
Since divX = 2, we have
2 A() =
_
X, n) ds.
The Cauchy-Schwarz inequality for the inner product of R
2
gives
X, n) [X[
and thus
(2.1) 2 A()
_
[X[ ds.
Now the integral Cauchy-Schwarz inequality yields
(2.2) 2 A()
__
[X[
2
ds
_
1/2
__
1
2
ds
_
1/2
=
1/2
()
__
[X[
2
ds
_
1/2
.
Since we are parametrizing by arc-length, the components of X(s) = (x(s), y(s))
along are C
1
periodic functions of period L = (). By Wirtingers inequality
2.7 (applied to each component of X(s)) we have
(2.3)
__
[X[
2
ds
_
1/2
2
()
4
2
_
[X
[
2
ds
_
1/2
=
3/2
()/2
Combining (2.2) and (2.3) yields:
2 A()
2
()/2,
with equality if and only if we have equality in (2.1), (2.2) and (2.3). But if we have
equality in (2.2), the characterization of the equality case in the Cauchy-Schwarz
inequality implies that s [X(s)[ is constant, i.e. must be a disk.
2.3. First fundamental form and Area

A surface is a 2-dimensional manifold S. In this section we begin the study of
surfaces which live inside R
3
. As in the case of curves, the inner product of R
3
will
induce an important geometric structure on the surface: a Riemannian metric. As
we move along you will be able to guess that many of the ideas that we present
actually hold with much greater generality. A good grasp of these ideas will make
Part III courses in Geometry/Topology much easier to understand. The old fashion
term for Riemannian metric in this context is First fundamental form.
Definition 2.8. Let S R
3
be a surface. The quadratic form I
p
on T
p
S given
by
I
p
(w) := w, w) = [w[
2
is called the rst fundamental form of the surface at p.
2.3. FIRST FUNDAMENTAL FORM AND AREA 17
In general, if X is a manifold, a Riemannian metric on X, is a smooth map that
asigns to each p X an inner product g
p
in T
p
X. If f : X Y is an immersion and
Y has a Riemannian metric g, then using df we can endow X with a Riemannian
metric h dened by
h
p
(w) := g
f(p)
(df
p
(w)).
The rst fundamental form is the Riemannian metric on S = X determined by the
inner product in R
3
= Y , where f is the inclusion map S R
3
.
Definition 2.9. Two surfaces S
1
and S
2
are said to be isometric if there exists
a dieomorphism f : S
1
S
2
such that for all p S
1
, df
p
is a linear isometry
between T
p
S
1
and T
f(p)
S
2
.
A Riemannian metric will allow us to do various measurements on the surface,
like length and area. Before we go into that, let us express a Riemannian metric
locally.
Let : U R
2
S R
3
be a parametrization of a neighbourhood of a point
p S. We will denote by (u, v) points in U and let
u
(u, v) :=

u
(u, v) T
(u,v)
S,
v
(u, v) :=

v
(u, v) T
(u,v)
S.
(These vectors are linearly independent, why?) Set
E(u, v) : =
u
,
u
)
(u,v)
,
F(u, v) : =
u
,
v
)
(u,v)
,
G(u, v) : =
v
,
v
)
(u,v)
.
Since a tangent vector w T
p
S is the tangent vector of a curve (t) = (u(t), v(t)),
t (, ), with p = (0) = (u
0
, v
0
) we have
I
p
( (0)) = (0), (0))
p
=
u
,
u
)
p
( u)
2
+ 2
u
,
v
)
p
u v +
v
,
v
)
p
( v)
2
= E( u)
2
+ 2 F u v +G( v)
2
.
Example 2.10. Let S be the torus obtained by rotating a circle of radius r
about a straight line belonging to the plane of the circle and at a distance a > r
away from the centre of the circle. A parametrization which covers all of the torus
except for a meridian and a parallel is given by
(u, v) = ((a +r cos u) cos v, (a +r cos u) sinv, r sinu)
where u, v (0, 2). Now we can easily compute E, F and G:
E = r
2
, F = 0, G = (r cos u +a)
2
.
The local coecients E, F and G are useful when we wish to compute the
length of a curve in S. If we can express the curve in local coordinates we just have
to integrate
_
E( u)
2
+ 2 F u v +G( v)
2
.
Exercise 2.11. Show that
[
u

v
[ =
_
EGF
2
.
We now dene the area of a bounded domain in a surface. A domain of S is
an open and connected subset of S. It is bounded if it is contained in some ball in
R
3
. Suppose that is contained in the image of a parametrization : U S.
Lemma 2.12. The integral
_
1
()
[
u

v
[ dudv
does not depend on the parametrization.
Proof. We have to show that if :

U S is another parametrization with
(
U), then
_
1
()
[
u

v
[ dudv =
_
1
()
[
u

v
[ d ud v.
Let J( u, v) be the Jacobian of h :=
1
. Since
[
u

v
[ = [J[ [
u

v
[ h
the lemma follows from the formula for the change of variables of multiple integrals
that you have seen in the Vector Calculus course (h is the change of variables).
Definition 2.13. Let S be a bounded domain contained in the image of
a parametrization : U S. The positive number
A() :=
_
1
()
[
u

v
[ dudv
is called the area of .
It is possible to dene the area of open sets which are not contained in the image
of a parametrization. This is done using a partition of unity, which is technical
gadget designed to glue smooth objects in a manifold. You will see this in Part
III. For practical purposes one never uses partitions of unity to compute area.
Usually a clever choice of parametrization covers the set that we are interesed in
except for some curves (i.e. sets of measure zero) which do not contribute to the
area. In fact, using the exponential map it is possible to produce always such
parametrizations. Once you have the notion of area for open sets, you can extend
it to a measure dened on the Borel sigma algebra generated by the open sets. In
this form we obtain the Riemannian measure. Dierent Riemannian metrics give
rise to dierent Riemannian measures, but they all have the same measure zero
sets. We will denote the Riemannian measure by dA and we will have the chance
to use it when we discuss the Gauss-Bonnet theorem. If f : S R is a continuous
function and : U S is a parametrization that covers S up to a set of measure
zero, then
_
S
f dA =
_
U
f(u, v)
_
EGF
2
dudv.
Example 2.14. Let us compute the area A of the torus from Example 2.10.
Recall that
[
u

v
[ =
_
EGF
2
.
For our example
_
EGF
2
= r(r cos u +a)
2.4. THE GAUSS MAP 19
thus
A =
_
(0,2)(0,2)
r(r cos u +a) dudv = 4
2
ra.
2.4. The Gauss map
Given a parametrization : U R
2
S R
3
around a point p S, we can
choose a unit normal vector at each point of (U) by setting
N(q) :=

u

v
[
u

v
[
(q).
Thus we have smooth map N : (U) R
3
, which in fact takes values in S
2
R
3
.
Definition 2.15. A smooth eld of unit normal vectors on a surface S is a
smooth map N : S S
2
R
3
such that for every p S, N(p) is orthogonal
to T
p
S. A surface S R
3
is orientable if it admits a smooth eld of unit normal
vectors. The choice of such a eld is called an orientation.
The Mobius band is an example of a non-orientable surface (why?). Orientabil-
ity is denitely a global property. Observe that an orientation as we have dened
it induces an orientation on every tangent plane T
p
S by declaring a basis u, v
positively oriented if u, v, N is a positively oriented basis of R
3
.
Definition 2.16. Let S be an oriented surface and N : S S
2
the smooth
eld of unit normal vectors dening the orientation. The map N is called the Gauss
map of S.
The Gauss map contains lots of geometric information about S. To start un-
packing it we look at its derivative dN
p
: T
p
S T
N(p)
S
2
. Since T
p
S and T
N(p)
S
2
are parallel planes, we will regard dN
p
as a linear map dN
p
: T
p
S T
p
S. We state
and prove at once the key property of the Gauss map:
Proposition 2.17. The linear map dN
p
: T
p
S T
p
S is selfadjoint.
Proof. Let : U S be a parametrization around p. If (t) = (u(t), v(t))
is a curve in (U) with (0) = p we have
dN
p
( (0)) = dN
p
( u(0)
u
+ v(0)
v
)
=
d
dt
[
t=0
N(u(t), v(t))
= u(0)N
u
+ v(0)N
v
.
In particular dN
p
(
u
) = N
u
and dN
p
(
v
) = N
v
and since
u
,
v
is a basis of the
tangent plane, we only have to prove that
N
u
,
v
) = N
v
,
u
).
To prove the last equality, observe that N,
u
) = N,
v
) = 0. Take derivatives
with respect to v and u to obtain:
N
v
,
u
) +N,
uv
) = 0,
N
u
,
v
) +N,
vu
) = 0,
which gives the desired equality.

Definition 2.18. The quadratic form dened on T
p
S by II
p
(w) := dN
p
(w), w)
is called the second fundamental form of S at p.
Let : (, ) S be a curve parametrized by arc-length with (0) = p. If we
let N(s) be N we have N(s), (s)) = 0 for all s. If we dierentiate we obtain:
N(s), (s)) =

N(s), (s)).
But II
p
( (0)) =

N(0), (0)) and thus
II
p
( (0)) = N(0), (0)) = N, k n)(p)
where k is the curvature of and n its unit normal. The expression N, k n)(p) is
called the normal curvature of at p and denoted by k
n
(p). Observe that k
n
(p)
only depends on the tangent vector (0).
Recall from Linear Maths that any selfadjoint linear map can be diagonalized
in an orthonormal basis. Hence there exist an orthonormal basis e
1
, e
2
of T
p
S
and real numbers k
1
and k
2
with k
1
k
2
such that dN
p
(e
i
) = k
i
e
i
for i = 1, 2.
Moreover, the numbers k
1
and k
2
are the maximum and minimum of II
p
restricted
to the set of unit vectors on T
p
S. (They are extreme values of the normal curvature
at p.)
Definition 2.19. The numbers k
1
and k
2
are called the principal curvatures
at p and e
1
and e
2
are called the principal directions at p.
Definition 2.20. The determinant of dN
p
is the Gaussian curvature K(p) of
S at p. Minus half of the trace of dN
p
is called the mean curvature H(p) of S at p.
Clearly K = k
1
k
2
and H =
k1+k2
2
. A point p S of a surface is called elliptic
if K(p) > 0, hyperbolic if K(p) < 0, parabolic if K(p) = 0 and dN
p
,= 0, and planar
if dN
p
= 0.
Example 2.21. The points of a round sphere are elliptic points. The point
(0, 0, 0) of the paraboloid z = x
2
+ y
2
is also elliptic. The point (0, 0, 0) of the
hyperboloid z = x
2
y
2
is a hyperbolic point. The points of a cylinder are parabolic
points and the points of a plane are planar.
A point p S is called umbilical if k
1
= k
2
.
Proposition 2.22. If all points on a connected surface are umbilical, then S
is either contained in a sphere or a plane.
Proof. See Example sheet 2.
2.5. The second fundamental form in local coordinates
Let : U S be a parametrization around a point p S. Let us express the
second fundamental form in the basis
u
,
v
. Since N,
u
) = N,
v
) = 0 we
have
e : = N
u
,
u
) = N,
uu
),
f : = N
v
,
u
) = N,
uv
) = N
u
,
v
),
g : = N
v
,
v
) = N,
vv
).
The coecients e, f and g are sometimes also denoted by L, M and N respectively.
If is a curve passing at t = 0 through p we can write:
II
p
( (0)) = dN
p
( (0)), (0)) = e( u)
2
+ 2 f u v +g( v)
2
.
2.6. THEOREMA EGREGIUM 21
It is now a simple exercise in Linear Maths to express K and H in terms of the
local coecients E, F, G and e, f, g. Write:
dN
p
(
u
) = N
u
= a
11
u
+a
21
v
,
dN
p
(
v
) = N
v
= a
12
u
+a
22
v
.
Taking inner products of these equations with
u
and
v
we obtain:
(2.4)
_
E F
F G
__
a
11
a
12
a
21
a
22
_
=
_
e f
f g
_
.
Check now that:
(2.5) K =
eg f
2
EGF
2
,
(2.6) H =
eG2fF +gE
2(EGF
2
)
.
The principal curvatures are the roots of the characteristic polynomial:
k
2
2Hk +K = 0.
Example 2.23. Let us compute the Gaussian curvature of the torus from Ex-
ample 2.10. We have already calculated E, F and G. To compute e, f and g we use
the formulas above to obtain
e = r, f = 0, g = cos u(a +r cos u).
Thus
K =
cos u
r(a +r cos u)
.
We see that K = 0 along the parallels u = /2 and u = 3/2 (parabolic points).
In the region /2 < u < 3/2, K is negative (hyperbolic points) and in the region
0 < u < /2 or 3/2 < u < 2, K is positive (elliptic points).
2.6. Theorema Egregium
Theorem 2.24 (Theorema Egregium, Gauss 1827). The Gaussian curvature
K of a surface is invariant under isometries.
Proof. Since by denition isometries preserve the rst fundamental form it
suces to prove that K can be expressed in local coordinates purely in terms of
the coecients E, F and G of the rst fundamental form and their derivatives.
Let : U S be a parametrization. On each point of (U) we have a basis of
R
3
given by
u
,
v
, N.
We can now express the derivatives of the vectors
u
,
v
in this basis to obtain
(recall the denitions of the coecients e, f and g):
uu
=
1
11
u
+
2
11
v
+eN, (2.7)
uv
=
1
12
u
+
2
12
v
+fN, (2.8)
vu
=
1
21
u
+
2
21
v
+fN, (2.9)
vv
=
1
22
u
+
2
22
v
+gN, (2.10)
The coecients
k
ij
are called the Christoel symbols. Since
uv
=
vu
we have
1
12
=
1
21
and
2
12
=
2
21
.
If in the rst equation we take inner products with
u
and
v
we obtain:
1
11
E +
2
11
F =
uu
,
u
) =
1
2
E
u
, (2.11)
1
11
F +
2
11
G =
uu
,
v
) = F
u

1
2
E
v
. (2.12)
Since EG F
2
,= 0 we can solve for
1
11
and
2
11
and arguing in a similar fashion
with the other equations we conclude that we can express the Christoel symbols
in terms of E, F and G and their rst derivatives.
Consider the identity:
uuv
=
uvu
.
Using (2.7) and (2.8) we get:
1
11
uv
+
2
11
vv
+eN
v
+
_
1
11
_
v

u
+
_
2
11
_
v

v
+e
v
N
=
1
12
uu
+
2
12
vu
+fN
u
+
_
1
12
_
u
u
+
_
2
12
_
u
v
+f
u
N.
Using equations (2.7), (2.8) and (2.10) and equating the coecients of
v
one
obtains after some manipulations using (2.4) from the previous section:
_
2
12
_
u

_
2
11
_
v
+
1
12
2
11
+
2
12
2
12

2
11
2
22

1
11
2
12
(2.13) = fa
21
+ea
22
= E
eg f
2
EGF
2
= EK.
This formula shows that K can be expressed solely in terms of the coecients of
the rst fundamental form and their derivatives.
Remark 2.25. Formula (2.13) is known as Gauss formula. If is an orthogonal

parametrization, i.e. F = 0, calculations simplify quite a bit and the Gauss formula
yields:
(2.14) K =
1
2
EG
__
E
v
EG
_
v
+
_
G
u
EG
_
u
_
.
Definition 2.26. A parametrization is said to be isothermal if E = G =
2
(u, v) and F = 0.
For isothermal parametrizations equation (2.14) simplies even further to give:
(2.15) K =
1
2
(log ),
where is the Laplacian in (u, v)-coordinates.
CHAPTER 3
Critical points of length and area
In this chapter we consider the extremals of the functionals length and area. As
you have seen in the Methods course the extremals of these functionals will satisfy
the corresponding Euler-Lagrange equations. The solutions of these equations will
be geodesics (length) and minimal surfaces (area). In the case of geodesics, the
Euler-Lagrange equations are just ODEs and existence and uniqueness of solutions
will be a fairly straightforward consequence of the results you have seen in Analysis
II. In the case of minimal surfaces what we obtain is a PDE and existence and
uniqueness is a fairly delicate problem.
3.1. Geodesics
Let S R
3
be a surface and p, q two points in S. Let (p, q) be the set of
all curves : [0, 1] S with (0) = p and (1) = q. Let : (p, q) R be the
functional given by the length of , i.e.
() =
_
1
0
[ [ dt.
A closely related functional, which is quite useful for variational purposes, is the
energy E : (p, q) R given by
E() =
1
2
_
1
0
[ [
2
dt.
One of the advantages of energy over length is that energy is sensitive to reparametriza-
tions. The next simple exercise, which is a consequence of the Cauchy-Schwarz
inequality, describes the relationship between the two functionals.
Exercise 3.1. ()
_
2E() with equality if and only if is parametrized
by a parameter proportional to arc-length.
Thus a curve minimizing E will have parameter proportional to arc-length.
Let us try to nd the extremals of E. For this consider a smooth 1-parameter
family of curves
s
(p, q) for s (, ) with
0
= . Let E(s) := E(
s
). We
have
dE
ds
s=0
=
_
1
0
_

s
s
t
,

s
t
_
dt.
Integrating by parts we nd:
dE
ds
s=0
= W(1), (1)) W(0), (0))
_
1
0
W(t) , (t)) dt
where
W(t) :=

s
(t)
s
s=0
.
23
24 3. CRITICAL POINTS OF LENGTH AND AREA
Since
s
(p, q), W(0) = W(1) = 0 and thus:
dE
ds
s=0
=
_
1
0
W(t) , (t)) dt.
Note that for each t [0, 1], W(t) T
(t)
S, so if has the property that (t) is
orthogonal to T
(t)
S at the point (t) for all t, then is an extremal point of E.
This motivates the following:
Definition 3.2. A curve : I S is said to be a geodesic if for all t I, (t)
is orthogonal to T
(t)
S at the point (t).
In other words we can say that is a geodesic if its acceleration has no compo-
nent in the direction of the surface. Thus, in physical terms, a geodesic describes
the motion of a free particle on the surface and what makes the motion interesting
is the bending of the surface.
3.2. Covariant derivative and parallel transport
Let : I S be a curve. A vector eld V along is a smooth map V : I R
3
such that for all t, V (t) T
(t)
S.
Definition 3.3. The vector obtained by the normal projection of
dV
dt
(t) onto
the plane T
(t)
S is called the covariant derivative of V at t and is denoted by
DV
dt
(t).
We can now say that a geodesic is a curve for which
D
dt
= 0.
Definition 3.4. A vector eld V along a curve : I S is said to be parallel
if
DV
dt
(t) = 0 for every t I.
Proposition 3.5. Let V and W be parallel vector elds along : I S. Then
W(t), V (t)) is constant.
Proof.
d
dt
W(t), V (t)) =
_
dW
dt
(t), V (t)
_
+
_
W(t),
dV
dt
(t)
_
But the last two terms are zero because of the denition of parallel vector eld and
the fact that V and W are tangent to S.
This proposition implies that if is a geodesic, then [ [ is constant, so geodesics
are parametrized with parameter proportional to arc-length.
Let : U S be a parametrization and let : I S be a curve such that
(I) (U). For some functions u(t) and v(t) we can write (t) = (u(t), v(t)).
Let V be a vector eld along . For some functions a(t) and b(t) we can write
V (t) = a(t)
u
+b(t)
v
. We have
dV
dt
= a(
uu
u +
uv
v) +b(
vu
u +
vv
v) + a
u
+

b
v
.
Using equations (2.7), (2.8) and (2.10) and the denition of covariant derivative
(just drop the normal component) we obtain:
DV
dt
= ( a +
1
11
a u +
1
12
a v +
1
12
b u +
1
22
b v)
u
(3.1) + (
b +
2
11
a u +
2
12
a v +
2
12
b u +
2
22
b v)
v
.
3.3. MINIMAL SURFACES 25
The last expression shows that the covariant derivative only depends on the rst
fundamental form, even though it was dened using the ambient space R
3
. From
(3.1) one can also obtain easily the equations of geodesics in local coordinates. It
suces to set a = u and b = v which gives the pair of equations:
(3.2) u +
1
11
( u)
2
+ 2
1
12
u v +
1
22
( v)
2
= 0;
(3.3) v +
2
11
( u)
2
+ 2
2
12
u v +
2
22
( v)
2
= 0.
Note the nonlinearity of (3.2) and (3.3).
If we regard (3.1) as a linear ODE in a and b, we see that given v
0
T
(t0)
S,
t
0
I, there exists a unique parallel vector eld V (t) along (t) with V (t
0
) = v
0
.
The vector V (t
1
), t
1
I, is called the parallel transport of v
0
along at the point
t
1
.
Let (p, q). Denote by P : T
p
S T
q
S the map that assigns to each
v T
p
S its parallel transport along at q. The map P is a linear map (why?) and
by Proposition 3.5, P is an isometry.
Finally (3.2) and (3.3) imply:
Proposition 3.6. Given a point p S and a vector v T
p
S, there exists an
> 0 and a unique geodesic : (, ) S such that (0) = p and (0) = v.
3.3. Minimal surfaces
Definition 3.7. A surface S R
3
is said to be minimal if its mean curvature
vanishes everywhere.
Let : U R
2
S be a parametrization and let D U be a bounded domain
with closure D U. Let h : D R be a smooth function.
Definition 3.8. The normal variation of (D) determined by h is the map
: D (, ) R
3
given by
(u, v, t) = (u, v) +th(u, v)N(u, v).
Remark 3.9. If the closure of D has the property that its boundary is a nice
set, e.g. a piecewise smooth 1-manifold, then one often demands the boundary to
be xed in the variation. This can be achieved by selecting h so that it vanishes on
the boundary.
For each xed t (, ), consider the map
t
: D R
3
given by
t
(u, v) :=
(u, v, t). Since
0
= , for small enough
t
(D) is a smooth surface in R
3
(this
follows from
t
u
and
t
v
being linearly independent). Clearly
t
u
=
u
+thN
u
+th
u
N,
t
v
=
v
+thN
v
+th
v
N.
If we let E
t
, F
t
and G
t
be the coecients of the rst fundamental form of
t
(D)
we get:
E
t
= E + 2th
u
, N
u
) +t
2
h
2
N
u
, N
u
) +t
2
h
u
h
u
,
F
t
= F + 2th
u
, N
v
) +t
2
h
2
N
u
, N
v
) +t
2
h
u
h
v
,
G
t
= G+ 2th
v
, N
v
) +t
2
h
2
N
v
, N
v
) +t
2
h
v
h
v
.
Using the denition of e, f, g and (2.6) we obtain
E
t
G
t
(F
t
)
2
= EGF
2
2th(Eg 2Ff +Ge) +r
= (EGF
2
)(1 4thH) +r
where lim
t0
r/t = 0. If we let A(t) be the area of
t
(D) we have:
A(t) =
_
D
_
E
t
G
t
(F
t
)
2
dudv
=
_
D
1 4thH + r
_
EGF
2
dudv,
where r = r/(EGF
2
). Clearly t A(t) is smooth and
(3.4) A
(0) =
_
D
2hH
_
EGF
2
dudv.
Proposition 3.10. (U) is minimal if and only if A
(0) = 0 for all bounded

domains D U and all normal variations of (D).
Proof. If H vanishes, (3.4) immediately gives A
(0) = 0. Suppose H(q) ,= 0

for some q D. Let h : D R be H. Then by (3.4), A
(0) < 0 for the variation

determined by such h.
Remark 3.11. The proposition explains the use of the term minimal for
surfaces with vanishing mean curvature. It should be noted that the critical point
of A may not actually be a minimum.
The mean curvature vector is H := HN. A normal variation in the direction
of HN always has A
(0) < 0 (provided H does not vanish).

Remark 3.12. Minimal surfaces are often associated with soap lms. A phys-
ical argument shows that at regular points the mean curvature of the lm must be
zero.
3.4. The Weierstrass representation
Proposition 3.13. Let : U R
3
be an isothermal parametrization (cf.
Denition 2.26). Then
uu
+
vv
= 2
2
H.
Proof. Since is isothermal E = G =
2
and F = 0. By dierentiating
u
,
u
) =
v
,
v
) with respect to u we get
uu
,
u
) =
vu
,
v
)
and since
vu
,
v
) =
u
,
vv
) we obtain
uu
+
vv
,
u
) = 0.
Similarly
uu
+
vv
,
v
) = 0.
It follows that
uu
+
vv
is parallel to N. Using (2.6) we have
H =
g +e
2
2
.
Thus
2
2
H = g +e = N,
uu
+
vv
);
3.4. THE WEIERSTRASS REPRESENTATION 27
hence
uu
+
vv
= 2
2
H as desired.
The proposition immmediately implies:

Corollary 3.14. Let : U R
3
be an isothermal parametrization. Then
(U) is minimal if and only if = 0, where is the Laplacian in (u, v)-
coordinates.
Example 3.15. The catenoid is given by
(u, v) = (a coshv cos u, a coshv sinu, av),
for u (0, 2) and v R (together with an obvious additional parametrization).
This surface is obtained by rotating the catenary y = a cosh(z/a) about the z-axis.
An easy calculation shows that E = G = a
2
cosh
2
v and F = 0, so is isothermal.
Moreover = 0 and so the catenoid is a minimal surface. In fact, it is the only
surface of revolution which is minimal (Can you prove this?).
Example 3.16. The helicoid is given by
(u, v) = (a sinhv cos u, a sinhv sinu, au),
for u R and v R. As in the previous example an easy calculation shows that
E = G = a
2
cosh
2
v and F = 0, so is isothermal and = 0. Thus the helicoid
is a minimal surface. In fact, it is only minimal surface, besides the plane, which is
ruled, i.e. it can be obtained from a straight line sliding smoothly along a curve.
J.B. Meusnier in 1776 discovered that the helicoid and the catenoid were min-
imal surfaces. For a long time they were the only examples. The next proposition
establishes a connection between minimal surfaces and holomorphic functions that
will allow us to nd more examples. Write (u, v) = (x(u, v), y(u, v), z(u, v)).
Proposition 3.17. Consider the complex valued functions
1
:= x
u
ix
v

2
:= y
u
iy
v

3
:= z
u
iz
v
.
Then is isothermal if and only if
2
1
+
2
2
+
2
3
0. Suppose is isothermal.
Then is minimal if and only if
1
,
2
and
3
are holomorphic functions.
Proof. An easy calculation gives:
2
1
+
2
2
+
2
3
= E G2iF,
which immediately implies the rst claim in the proposition. To see the second
claim, consider the Cauchy-Riemann equations of
1
:
x
uv
= x
vu
,
x
uu
= x
vv
.
The rst equation is trivially satised ( is smooth) and the second one is just
x = 0.
The next lemma will tell us the solutions of

(3.5)
2
1
+
2
2
+
2
3
0.
Lemma 3.18. Let D be a domain in the complex plane, g() an arbitrary mero-
morphic function in D and f() a holomorphic function in D having the property
that at each point where g() has a pole of order k, f() has a zero of order at least
2k. Then the functions
(3.6)
1
=
1
2
f(1 g
2
),
2
=
i
2
f(1 +g
2
),
3
= fg
will be holomorphic in D and satisfy (3.5). Conversely, every triple of holomorphic
functions satisfying (3.5) may be represented in the form (3.6), except for
1
i
2
,
3
0.
Proof. A calculation shows that the functions given by (3.6) satisfy (3.5).
Conversely, given a solution of (3.5), set
(3.7) f :=
1
i
2
, g :=

3
1
i
2
.
We now write (3.5) in the form
(3.8) (
1
i
2
)(
1
+i
2
) =
2
3
to obtain:
(3.9)
1
+i
2
=

2
3
1
i
2
= fg
2
.
Combining (3.7) and (3.9) gives (3.6). The condition relating the zeros of f with the
poles of g must hold, otherwise by (3.9),
1
+i
2
would fail to be holomorphic. This
representation can only fail if the denominator in the formula for g in (3.7) vanishes
identically. In this case, by (3.8) we have that
3
0, which is the exceptional case
mentioned.
Let
0
be a xed point in D, and let be a path in D connecting
0
to D.
If D is simply connected and
i
is holomorphic, the integrals
_
i
() d
do not depend on the path and dene holomorphic functions whose real parts are
x, y and z. Hence if we take functions f and g as in Lemma 3.18 and D is simply
connected, then the equations
x(u, v) = '
_
_
(u,v)
0
1
2
f()(1 g
2
()) d
_
, (3.10)
y(u, v) = '
_
_
(u,v)
0
i
2
f()(1 +g
2
()) d
_
, (3.11)
z(u, v) = '
_
_
(u,v)
0
f()g() d
_
(3.12)
dene a parametrization which gives rise to a minimal surface. The equations
(3.10)-(3.12) are called the Weierstrass representation of the surface. Of course one
has to worry about whether is an immersion or an embedding.
3.5. THE MEANING OF g IN THE WEIERSTRASS REPRESENTATION 29
Exercise 3.19. Show that if is dened by the Weierstrass representation,
then is an immersion if and only f vanishes only at the poles of g and the order
of its zero at such a point is exactly twice the order of the pole of g.
The converse is also true: any simply connected minimal surface in R
3
can be
represented by formulas (3.10)(3.12).
By appropriate choices of f and g one obtains various celebrated examples of
minimal surfaces. More importantly, the Weierstrass representation allows one to
obtain general theorems about minimal surfaces by using complex analysis.
Example 3.20 (Ennepers surface 1864). The simplest choice of D, f and g is
D = C, f() = 1 and g() = . It gives the minimal immersion : C R
3
(u, v) =
1
2
(u u
3
/3 +uv
2
, v +v
3
/3 u
2
v, u
2
v
2
).
Its Gaussian curvature is
K =
16
(1+ [ [
2
)
4
,
where = u +iv.
Example 3.21 (Scherks surface 1835). Let D be the open unit disk in the
complex plane, f() = 4/(1
4
) and g() = . This gives the immersion : D
R
3
(u, v) =
_
arg
+i
i
, arg
+ 1
1
, log
2
+ 1
2
1
_
.
From the expressions of x, y and z it is easily seen that
z = log
cos y
cos x
.
Exercise 3.22. Find D, f and g for the catenoid and the helicoid.
There are great web sites about minimal surfaces. Take a look at this one. It
contains beautiful pictures of all the surfaces we have seen and much more.
3.5. The meaning of g in the Weierstrass representation
Let : S
2
C be the stereographic projection. Recall that
(3.13)
1
(z) =
_
2'z
1+ [ z [
2
,
2z
1+ [ z [
2
,
[ z [
2
1
[ z [
2
+1
_
.
Let be a parametrization given by formulas (3.10)-(3.12). Clearly
u
i
v
= (
1
,
2
,
3
)
and hence
2
= E = G =
1
2
[
k
[
2
=
_
[ f [ (1+ [ g [
2
)
2
_
2
.
Furthermore
u

v
= (
2
3
,
3
1
,
1
2
)
and using the expressions of
i
in terms of f and g we derive:
u

v
=
[ f [
2
(1+ [ g [
2
)
4
(2'g, 2g, [ g [
2
1).
Note also that
[
u

v
[=
_
[ f [ (1+ [ g [
2
)
2
_
2
=
2
.
Hence, if N : D S
2
is the Gauss map, we obtain:
(3.14) N =
_
2'g
1+ [ g [
2
,
2g
1+ [ g [
2
,
[ g [
2
1
[ g [
2
+1
_
.
From (3.13) and (3.14) we see that
(3.15) N = g.
If you are attending the Riemann surfaces course you will notice that (3.15) implies
right away the following lemma:
Lemma 3.23. The Gauss map N : D S
2
is a holomorphic map.
Exercise 3.24. Show that the Gaussian curvature is given by
K =
_
4 [ g
[
[ f [ (1+ [ g [
2
)
2
_
2
.
Show that either K 0 or its zeros are isolated.
Here is a beautiful application of (3.15):
Theorem 3.25. If D = C, then either the surface lies on a plane, or else the
Gauss map takes all directions with a most two exceptions.
Proof. To the surface we associate the function g() which fails to be dened
only if
1
i
2
,
3
0. But in this case z is constant and the surface lies in a
plane. Otherwise g() is a meromorphic function on the entire complex plane, and
by Picards theorem it either takes all the values with at most two exceptions, or
else is constant. By (3.15) the same alternative applies to the Gauss map, and in
the latter case the surface lies on a plane.
In the catenoid, the Gauss map misses exactly two points. The following re-
markable result is stated without proof (its proof is a more elaborate application
of (3.15):
Theorem 3.26 (Osserman 1960). Let S be a minimal surface in R
3
which is
not a plane and is a closed subset of R
3
. Then the image of the Gauss map is dense
in S
2
.
In fact, subsequent work (Xavier 1981, Fujimoto 1988), shows that if S is a
minimal surface in R
3
which is not a plane and is a closed subset of R
3
, then the
image of the Gauss map can omit at most four directions.
CHAPTER 4
Global Riemannian geometry
4.1. The exponential map and geodesic polar coordinates
Let S R
3
. By Proposition (3.6) we know that given a point p S and a
vector v T
p
S, there exists an > 0 and a unique geodesic : (, ) S such
that (0) = p and (0) = v. To indicate the dependence on the vector v, let us
write
v
= . Here is a simple exercise:
Exercise 4.1. Prove that if the geodesic
v
(t) is dened for t (, ), then
the geodesic
v
(t), R, ,= 0, is dened for t (/, /), and
v
(t) =
v
(t).
Definition 4.2. If v T
p
S, v ,= 0, is such that
v/|v|
([v[) =
v
(1) is dened,
we set
exp
p
(v) =
v
(1) and exp
p
(0) = p.
The important point is that exp
p
is always dened and dierentiable in some
ball B
with center at the origin in T

p
S. We will not prove this result here, but you
should know that it is a fairly straightforward application of a well known theorem
in ODEs about smooth dependence of solutions on initial conditions (as far as I
know this result is not proved in the Tripos). The map exp
p
: B
S is called
the exponential map at p. If the surface S is a closed subset of R
3
, then one can
show that exp
p
is dened on all of T
p
S and geodesics are dened for all t R. The
exponential map also depends smoothly on p.
We can actually prove the following important complement:
Proposition 4.3. exp
p
: B
S is a dieomorphism onto its image in a

neighbourhood U B
of the origin in T
p
S.
Proof. By the inverse function theorem it suces to show that the dierential
d(exp
p
)
0
is nonsingular. Let (t) := tv, v T
p
S. The curve (exp
p
)(t) =
exp
p
(tv) =
v
(t) has at t = 0 the tangent vector v. Thus d(exp
p
)
0
(v) = v.
The last proposition implies that the exponential map exp
p
when restricted to
a small ball is actually a parametrization of S around p. The set V := exp
p
(U) is
called a normal neighbourhood of p.
Since exp
p
maps lines through the origin to geodesics in S, it gives in fact a very
useful and geometrically friendly parametrization. For example, if we choose in
T
p
S cartesian coordinates, and we look at the coecients of the rst fundamental
form in these coordinates, then we see right away that E(p) = G(p) = 1 and
F(p) = 0, so the metric at p looks like the Euclidean metric.
If we now choose polar coordinates (r, ) on T
p
S we obtain the so called geodesic
polar coordinates on S. Recall that polar coordinates in the plane are not dened
in the closed half line that corresponds to = 0. Set L = exp
p
(). Since
31
32 4. GLOBAL RIEMANNIAN GEOMETRY
exp
p
: U V L is still a dieomorphism, we may parametrize V L by the
coordinates (r, ). However for any (r, ) we may still dene
(4.1) (r, ) := exp
p
(r(cos e
1
+ sin e
2
)) = exp
p
(rv()) =
v()
(r),
where v() = cos e
1
+ sin e
2
.
Definition 4.4. The images under : U V of circles in U centered at the
origin are called geodesic circles and the images under of the lines through the
origin are called radial geodesics.
The coecients of the rst fundamental form and the Gaussian curvature take
a very simple form in geodesic polar coordinates as the next proposition shows.
Proposition 4.5. The coecients E(r, ), F(r, ) and G(r, ) satisfy:
E = 1, F = 0, G(0, ) = 0, (
G)
r
(0, ) = 1.
Moreover, the Gaussian curvature K can be written as:
K =
(
G)
rr
G
.
Proof. From (4.1) we see that
r
=
v()
(r) and thus E = 1 since v() has
norm one and geodesics travel at constant speed. Let w =
dv
d
. Also from (4.1) we
see that
= d(exp
p
)
rv
(rw) = rd(exp
p
)
rv
(w).
Hence
F = r
v
(r), d(exp
p
)
rv
(w)),
G = r
2
[d(exp
p
)
rv
(w)[
2
.
Obviously F(0, ) = 0 and from the last equality we see that
(
G)
r
(0, ) = [d(exp
p
)
0
(w)[ = [w[ = 1.
We now compute
F
r
=
rr
,
) +
r
,
r
)
=
r
,
r
) =
1
2
r
,
r
)
=
1
2
E
= 0,
where we used that
r
=
v
is a geodesic. It follows that F = 0 everywhere since
F(0, ) = 0.
Finally the expression for K follows right away from (2.14).
4.2. THE GAUSS-BONNET THEOREM 33

4.2. The Gauss-Bonnet theorem
4.2.1. Geodesic curvature.
Definition 4.6. Let W be a dierentiable eld of unit vectors along a curve
: I S on an oriented surface S. Since W(t) has norm 1, (dW/dt)(t) is normal
to W(t) and therefore
DW
dt
= (N W).
The real number = (t), denoted by [DW/dt], is called the algebraic value of the
covariant derivative of W at t.
Remark 4.7. Note that the sign of [DW/dt] depends on the orientation of S.
Also note that
_
DW
dt
_
=
_
dW
dt
, N W
_
.
Definition 4.8. Let : I S be a regular curve parametrized by arc-length.
The algebraic value of the covariant derivative
k
g
(s) :=
_
D
dt
_
= , N )
is called the geodesic curvature of at (s).
Remark 4.9. Note that k
g
changes sign if we change the orientation of S. Also
note that is a geodesic if and only if its geodesic curvature is zero.
Let V and W be two dierentiable unit vector elds along a curve : I S.
Let iV be the unique unit vector eld along such that for every t I, V (t), iV (t)
is a positively oriented orthonormal basis of T
(t)
S. Thus we can write
W(t) = a(t)V (t) +b(t)iV (t)
where a and b are smooth functions in I and a
2
+b
2
= 1.
Exercise 4.10. Let a and b be smooth functions on I with a
2
+b
2
= 1 and
0
be such that a(t
0
) = cos
0
, b(t
0
) = sin
0
. Then, the smooth function
(t) :=
0
+
_
t
t0
(a
b b a) dt
is such that cos (t) = a(t), sin(t) = b(t), t I, and (t
0
) =
0
.
The smooth (t) given by the exercise will be called a a smooth determination
of the angle from V to W.
Proposition 4.11. Let V and W be two smooth unit vector elds along a curve
: I S. Then
_
DW
dt
_
_
DV
dt
_
=
d
dt
,
where is a smooth determination of the angle from V to W.
Proof. By denition
_
DW
dt
_
=

W, N W),
_
DV
dt
_
=
V , N V ) =
V , iV ).
Write W = cos V + siniV and dierentiate to obtain:
W = (sinV + cos iV ) + cos

V + sin
d
dt
(iV ).
Also N W = cos iV sinV . Thus
_
DW
dt
_
= +sinV + cos iV, cos

V + sin
d
dt
(iV ))
= + cos
2
iV,

V ) sin
2
V,
d
dt
(iV ))
= +
_
DV
dt
_
,
where we used that iV,

V ) +V,
d
dt
(iV )) = 0 (dierentiate V, iV ) = 0).
Remark 4.12. Let : I S be a curve parametrized by arc-length and let

V (s) be a parallel unit vector eld along . Let be a smooth determination of
the angle from V to . Then by the proposition
k
g
(s) =
d
ds
.
If the curve is a plane curve, we can take as V (t) a xed direction in the plane and
we see that k
g
reduces to the usual curvature of .
Proposition 4.13. Let (u, v) be an orthogonal parametrization (i.e. F = 0)
of an oriented surface S which is compatible with the orientation. Let W(t) be a
smooth unit vector eld along the curve (u(t), v(t)). Then
_
DW
dt
_
=
1
2
EG
G
u
v E
v
u +
d
dt
,
where is the angle from
u
to W in the given orientation. In particular, if
: I (U) is a curve parametrized by arc-length,
k
g
(s) =
1
2
EG
_
G
u
dv
ds
E
v
du
ds
_
+
d
ds
,
u
to in the given orientation.
Proof. Let e
1
=
u
/
E and e
2
=
v
/
G. e
1
, e
2
is a positively oriented
orthonormal basis of the tangent plane. By the previous proposition
_
DW
dt
_
=
_
De
1
dt
_
+ .
We compute
_
De
1
dt
_
= e
1
, N e
1
) = e
1
, e
2
)
= (e
1
)
u
u + (e
1
)
v
v, e
2
)
= (e
1
)
u
, e
2
) u +(e
1
)
v
, e
2
) v.
Note (F = 0):
(e
1
)
u
, e
2
) = (
u
/
E)
u
,
v
/
G) =
1
EG
uu
,
v
).
Also F = 0 implies
uu
,
v
) =
u
,
uv
) = E
v
/2. Thus
(e
1
)
u
, e
2
) =
E
v
2
EG
.
Similarly
(e
1
)
v
, e
2
) =
G
u
2
EG
.
4.2.2. Gauss theorem for small geodesic triangles.

Theorem 4.14 (Gauss 1827). Let T be a geodesic triangle (i.e. its sides are
segments of geodesics) on a surface S. Suppose T is small enough so that it is
contained in a normal neighbourhood of one of its vertices. Then
_
T
K dA =
_
3
i=1
i
_
,
where K is the Gaussian curvature of S, and 0 <
i
< , i = 1, 2, 3, are the
internal angles of the geodesic triangle T.
Proof. This was Gauss original proof. Consider geodesic polar coordinates
(r, ) centered at one of the vertices of T. Suppose also that one of the sides of T
corresponds to = 0. Let the two other sides be given by =
0
and a geodesic
segment . Observe that is given in these coordinates by r = h() (why?). Then
_
T
K dA =
_
T
K
Gdrd =
_
0
0
_
lim
0
_
h()
Gdr
_
d.
Since (cf. Proposition 4.5) K
G = (
G)
rr
and lim
r0
(
G)
r
= 1, we have that
_
lim
0
_
h()
Gdr
_
= 1 (
G)
r
(h(), ).
Suppose the geodesic (r(s), (s)) makes an angle (s) with the curves =const.
in the orientation determined by the coordinates (r, ). Then, since is a geodesic,
Proposition 4.13 tell us that
(
G)
r
d
ds
+
d
ds
= 0.
Hence
_
T
K dA =
_
0
0
d +
_
(s0)
(0)
d =
_
3
i=1
i
_
.
Let us dene the excess of T as e(T) :=

_
3
i=1
i
_
. Note that if K is
never zero in T,
_
T
K dA is, up to a sign, the area of the image N(T) of T by the
Gauss map N : S S
2
. Thus we can rephrase the last result by saying that the
excess of a geodesic triangle T equals the area of its spherical image N(T) with the
appropriate sign. This was the way Gauss himself phrased his theorem.
4.2.3. Triangulations and Euler characteristic.
Definition 4.15. Let S be a compact surface. A triangulation of S consists
of a nite number of closed subsets T
1
, . . . , T
n
that cover S, such that each T
i
is homeomorphic to an Euclidean triangle in the plane. The subsets T
i
are called
triangles. Moreover, we require that any two distinct triangles either be disjoint,
have a single vertex in common, or have an entire edge in common.
It is a theorem that every compact surface (or compact manifold) admits a
triangulation. In fact one can choose the T
i
s to be dieomorphic to Euclidean
triangles. By considering barycentric subdivisions of the triangles we can produce
triangulations with very small triangles. Smooth triangulations exist even if S
has piecewise smooth boundary. Finally, one can actually choose the edges to be
geodesic segments, so that the triangles will be geodesic triangles.
Given a triangulation of S, let F be the number of triangles (faces), let E be
the number of edges and let V be the number of vertices. The number
(S) := F E +V
is called the Euler characteristic of the surface. It is an important topological
result that (S) does not depend on the triangulation, so that (S) is a topological
invariant of the surface. You will meet this invariant in the Algebraic Topology
course and the Riemann Surfaces course.
Compact orientable surfaces can be completely classied using . Any such S
will be dieomorphic to a sphere with g handles and (S) = 22g. The sphere has
= 2 (no handle), the torus (sphere with one handle) has = 0 and the double
torus (2 handles) has = 2. The number g is called the genus of S.
Remark 4.16. Compact boundaryless surfaces in R
3
are orientable.
4.2.4. The global Gauss-Bonnet theorem.
Theorem 4.17 (Global Gauss-Bonnet theorem). Let S be a compact surface
with empty boundary; then
_
S
K dA = 2(S).
Proof. Consider a triangulation with geodesic triangles T
i
, i = 1, . . . , F. We
will assume that each T
i
is small enough so that it is contained in a normal neigh-
bourhood of one of its vertices.
Let
i
,
i
,
i
be the interior angles of T
i
. By Theorem 4.14
_
Ti
K dA =
i
+
i
+
i
.
Summation over i yields
_
S
K dA =
F
i=1
(
i
+
i
+
i
) F.
Note that the sum of the angles at every vertex of the triangulation is 2, hence
F
i=1
(
i
+
i
+
i
) = 2V.
On the other hand, every edge of the triangulation belongs to two triangles, which
implies
2E = 3F.
Putting everything together we obtain:
_
S
K dA = (2V F) = (2V 2E + 2F) = 2(S).
Now we would like to generalize Theorem 4.14 to allow for boundary curves
which are not necessarily geodesics. This will bring in an extra term involving the
geodesic curvature of the boundary curves (this was actually Bonnets contribution).
Theorem 4.18 (Gauss-Bonnet Theorem (local)). Let : U S be an orthog-
onal parametrization of an oriented surface S, where U is a disk in R
2
and is
compatible with the orientation of S. Let : I (U) be a smooth simple closed
curve enclosing a domain R. Assume that is positively oriented and parametrized
by arc-length. Then
_
I
k
g
(s) ds +
_
R
K dA = 2,
where k
g
is the geodesic curvature of .
Proof. By Proposition 4.13 we may write
k
g
(s) =
1
2
EG
_
G
u
dv
ds
E
v
du
ds
_
+
d
ds
,
u
to in the given orientation. Set I = [0, l]. Integrating
the equality we obtain
_
I
k
g
(s) ds =
_
I
1
2
EG
_
G
u
dv
ds
E
v
du
ds
_
ds +(l) (0).
By Greens theorem
_
I
k
g
(s) ds =
_
1
(R)
__
G
u
2
EG
_
u
+
_
E
v
2
EG
_
v
_
dudv +(l) (0).
Using (2.14) gives:
_
I
k
g
(s) ds =
_
1
(R)
K
EGdudv +(l) (0).

Since (l) (0) = 2 (Theorem of the turning tangents) we obtain :
_
I
k
g
(s) ds =
_
R
K dA+ 2.
Remark 4.19. If is piecewise smooth, then jumps by the external angle
i
at each vertex, so the integral of
d
ds
which is 2 in the proof of the theorem is
replaced by
2
i
,
so one gets:
_
I
k
g
(s) ds +
_
R
K dA =
i
(p 2)
where
i
are the internal angles at the vertices and p is the number of vertices. If
there are 3 vertices and is a geodesic on each smooth portion, we recover Theorem
4.14.
Remark 4.20. A very pretty proof of the theorem of the turning tangents due
to Hopf (1935) may be found in do Carmos book or in Volume II of Spivak. It is
based on the invariance under homotopies of the degree of a continuous map from
S
1
to S
1
.
Now we can state:
Theorem 4.21 (Global Gauss-Bonnet Theorem including boundary terms).
Let R S be a domain of an oriented surface. Suppose R has compact closure
and its boundary consists of n piecewise smooth simple closed curves
i
: I
i
S,
i = 1, . . . , n whose images do not intersect. Suppose the
i
s are parametrized by
arc length and are positively oriented. Let
i
, i = 1, . . . , p, be the external angles of
the vertices of these curves. Then
n
i=1
_
Ii
k
g
(s) ds +
_
R
K dA+
p
i=1
i
= 2(R).
Proof. I will not do this proof in lectures. The idea is the same as the proof of
Theorem 4.17: consider a triangulation, apply the local theorem and put everything
together. An important point to note is that each interior edge of the triangula-
tion is described twice in opposite orientations, hence from all the contributions of
the integral of k
g
only the boundary ones survive. Writing a full detailed proof is
an excellent exercise (if you get stuck pick on do Carmos book).
4.3. Applications
Theorem 4.22. A compact oriented surface S with positive Gaussian curvature
is dieomorphic to S
2
. Moreover, if there exist two simple closed geodesics, they
must intersect.
Proof. The Global Gauss-Bonnet tells us right away that (S) > 0 and since
the sphere is the only orientable surface with positive Euler characteristic the rst
claim in the theorem follows.
To see the second claim, suppose that there are two simple closed geodesics
1
and
2
whose images do not intersect. Then the set formed by their images is
the boundary of a domain R with (R) = 0. By the Gauss-Bonnet theorem with
boundary terms
_
R
K dA = 0,
which is a contradiction since K > 0.
In fact, the rst claim in the theorem can be proved without using Gauss-
Bonnet and we can actually give an explicit dieomorphism. If K > 0 everywhere,
every point in S
2
is a regular value of the Gauss map N : S S
2
(why?). Hence
4.4. FENCHELS THEOREM 39
by the Stacks of records theorem 1.15 N is a covering map (making contact again
with Algebraic Topology). Since S
2
is simply connected, N must be a bijection
and therefore a dieomorphism.
Theorem 4.23. Let S be a surface homeomorphic to a cylinder. Suppose S
has negative Gaussian curvature. Then S has at most one simple closed geodesic.
Proof. Example sheet 4.
4.4. Fenchels theorem
Definition 4.24. A plane regular closed curve : [0, L] R
2
is convex if,
for each t [0, L], the curve lies in one of the closed half-planes determined by the
tangent line at t.
Definition 4.25. Let : [0, L] R
3
be a curve parametrized by arc-length
and let k(s) 0 be its curvature. The total curvature of is the number
_
L
0
k(s) ds.
Theorem 4.26 (Fenchels theorem). The total curvature of a simple closed
curve is 2, and equality holds if and only if the curve is a plane convex curve.
We will not prove the convexity statement and we will assume from now on that
the curvature is positive everywhere (or equivalently, that it is nowhere vanishing).
The proof of this theorem (and the next) will be given as a sequence of guided
exercises with hints. The two main ingredients are the tube T around and the
area formula.
Recall from the example sheets that the tube is given by:
(s, v) = (s) +r(ncos v +b sinv),
where r is suciently small, n is the normal to , b the binormal and s [0, L],
v [0, 2] ( is parametrized by arc-length and has length L).
The area formula applied to the Gauss map N : S S
2
of a compact surface
S (perhaps with non-empty boundary) states that
(4.2)
_
S
[K[ dA =
_
S
2
#N
1
(y) dy.
Try to prove it using the Stack of records theorem 1.15; it is a very challenging
exercise!
We now start the proof.
(1) We orient the tube T using . Show that the Gauss map is given by
N = (ncos v +b sinv).
(2) Show that for the tube we have EGF
2
= r
2
(1 rk cos v)
2
and
K =
k cos v
r(1 kr cos v)
.
Conclude that K = 0 i the line through b is orthogonal to the tube.
(3) Let R := p T : K(p) 0. Show that
_
R
K dA = 2
_
L
0
k(s) ds.
(4) Show that the Gauss map N : R S
2
is surjective.
(5) Use the previous item and the area formula (4.2) applied to R to conclude
that
2
_
L
0
k(s) ds =
_
R
K dA =
_
S
2
#(N[
R
)
1
(y) dy 4.
(6) We now need to show that if
(4.3)
_
R
K dA = 4,
then is a plane curve. (If you get stuck with this part look at Spivaks
Volume III, page 289.) Show rst that if equality in (4.3) holds then
given a point p T with K(p) > 0, the tube T must lie on one side of
the ane tangent plane at p. Now x a circle s = s
0
and show that T lies
between the parallel ane tangent planes P
1
and P
2
determined by the
two points on the circle s = s
0
where the Gaussian curvature vanishes.
The nal step consists in showing that is completely contained in the
plane midway between P
1
and P
2
. For this argue by contradiction and
consider a point (s
1
) on which is at a maximum distance from P (but
not in P). Use that the tangent vector (s
1
) must be parallel to P to
derive the contradiction.
(7) Complete the proof of Fenchels theorem by showing that if is a plane
curve, then its total curvature is 2 (apply Gauss-Bonnet to the at region
enclosed by ).
4.5. The Fary-Milnor theorem
Think of S
1
as the standard unit circle in R
2
. Let : S
1
R
3
be a continuous
simple curve (i.e. is injective). We say that is unknotted if there exists a
homotopy H : S
1
[0, 1] R
3
such that if we let
t
(s) = H(s, t), then
0
(s) =
(s, 0) R
2
R = R
3
,
1
= and
t
is a simple closed curve for all t [0, 1]. Such
a homotopy is called an isotopy.
Alternatively we can say that is unknotted if (S
1
) is the boundary of an
embedded disk.
If is not unknotted, then it is said to be knotted.
Theorem 4.27 (Fary-Milnor). The total curvature of a knotted simple closed
curve is greater than 4.
Proof. We will prove that the total curvature is 4 if is knotted (Fary,
1949). Milnor (1950) proved in fact the sharper statement that the total curvature
is > 4 and that the inmum over all simple closed curves is 4. As in the case of
Fenchels theorem we will assume that the curvature k is nowhere vanishing (so we
can use the tube as dened).
(1) Use the area formula (4.2) to show that
4
_
L
0
k(s) ds =
_
S
2
#N
1
(y) dy.
(2) Suppose that
_
L
0
k(s) ds < 4. Using that b([0, L]) S
2
and b([0, L])
S
2
have measure zero show that there exists y S
2
b([0, L]) for which
#N
1
(y) 3.
4.5. THE F
ARY-MILNOR THEOREM 41
(3) For y as in the previous item, let h(s) := (s), y). Show that h has at
most 3 critical points which must all be strict local maxima or minima.
Conclude that h has only two critical points which are the global maximum
and minimum of h.
(4) Without loss of generality we may suppose y = (0, 0, 1). Think of h as a
height function. Show that consists of two arcs joining the lowest and
highest points, each arc having monotonically increasing heights.
(5) Each plane orthogonal to y and with height between the lowest and highest
points intersects in two points. Join each such pair of points by a line
segment to obtain an embedded disk whose boundary is . Conclude that
is unknotted.
If you are interested in seeing how the hypothesis of having nowhere vanishing
curvature may be removed in Fenchels theorem and the Fary-Milnor theorem take
a look at Spivaks volume III.
In 1953 Milnor (Math. Scand. 1 (1953) 289296) proved the inequality
_
L
0
_
k
2
+
2
ds 4
for closed space curves with nowhere zero torsion . Much later Burt Totaro re-
discovered this result (Internat. J. Math. 1 (1990) 109117). (I thank Burt for
pointing this out to me.)

Differential Geometry, Gabriel-Paternain

Uploaded by

Copyright:

Available Formats

Differential Geometry, Gabriel-Paternain

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Differential Geometry, Gabriel-Paternain

Uploaded by

Copyright:

Available Formats

DIFFERENTIAL GEOMETRY, D COURSE

We can now dene the derivative of a smooth map f : X Y between arbitrary

As an easy application of the lemma, consider the unit ball B

Lemma 1.33 (Homogeneity Lemma). Let X be a smooth connected manifold,

In what follows suppose that X is compact and without boundary and Y is

10 1. SMOOTH MANIFOLDS AND SMOOTH MAPS

b(s) = (s) n(s)

t(s) = k(s) n(s)

Proof of Theorem 2.6. By translating if necessary, we can assume that

2.3. First fundamental form and Area

20 2. LENGTH, AREA AND CURVATURE

Remark 2.25. Formula (2.13) is known as Gauss formula. If is an orthogonal

(0) = 0 for all bounded

(0) = 0. Suppose H(q) ,= 0

(0) < 0 for the variation

(0) < 0 (provided H does not vanish).

The proposition immmediately implies:

The next lemma will tell us the solutions of

with center at the origin in T

S is a dieomorphism onto its image in a

4.2. THE GAUSS-BONNET THEOREM 33

W = (sinV + cos iV ) + cos

Remark 4.12. Let : I S be a curve parametrized by arc-length and let

4.2.2. Gauss theorem for small geodesic triangles.

Let us dene the excess of T as e(T) :=

EGdudv +(l) (0).

Remark 4.19. If is piecewise smooth, then jumps by the external angle

You might also like