GEOMETRY OF SURFACES

B3a course 2013

Nigel Hitchin
hitchin@maths.ox.ac.uk

1
Contents
1 Introduction 4

2 The topology of surfaces 9

2.1 The definition of a surface . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Planar models and connected sums . . . . . . . . . . . . . . . . . . . 14
2.3 The classification of surfaces . . . . . . . . . . . . . . . . . . . . . . . 22
2.4 Orientability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 The Euler characteristic . . . . . . . . . . . . . . . . . . . . . . . . . 28

3 Riemann surfaces 32
3.1 Definitions and examples . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2 Meromorphic functions . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3 A new look at the torus . . . . . . . . . . . . . . . . . . . . . . . . . 40

4 Surfaces in R3 45
4.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 The first fundamental form . . . . . . . . . . . . . . . . . . . . . . . . 51
4.3 Isometric surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.4 The second fundamental form . . . . . . . . . . . . . . . . . . . . . . 60
4.5 The Gaussian curvature . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.6 The Gauss-Bonnet theorem . . . . . . . . . . . . . . . . . . . . . . . 65
4.7 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.8 Gaussian curvature revisited . . . . . . . . . . . . . . . . . . . . . . . 76

5 The hyperbolic plane 80

5.1 Isometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.2 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.3 Angles and distances . . . . . . . . . . . . . . . . . . . . . . . . . . 83

2
5.4 Hyperbolic triangles . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.5 Non-Euclidean geometry . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.6 Complex analysis and the hyperbolic plane . . . . . . . . . . . . . . . 91

3
1 Introduction
This is a course on surfaces. Your mental image of a surface should be something like
this:

or this
However we are also going to try and consider surfaces intrinsically, or abstractly, and
not necessarily embedded in three-dimensional Euclidean space like the two above.
In fact lots of them simply cant be embedded, the most notable being the projective
plane. This is just the set of lines through a point in R3 and is as firmly connected
with familiar Euclidean geometry as anything. It is a surface but it doesnt sit in
Euclidean space.
If you insist on looking at it, then it maps to Euclidean space like this

called Boys surface. This is not one-to-one but it does intersect itself reasonably
cleanly.

4
A better way to think of this space is to note that each line through 0 intersects the
unit sphere in two opposite points. So we cut the sphere in half and then just have
to identify opposite points on the equator:

... and this gives you the projective plane.

Many other surfaces appear naturally by taking something familiar and perform-
ing identifications. A doubly periodic function like f (x, y) = sin 2x cos 2y can
be thought of as a function on a surface. Since its value at (x, y) is the same as at
(x+m, y+n) it is determined by its value on the unit square but since f (x, 0) = f (x, 1)
and f (0, y) = f (1, y) it is really a continuous function on the space got by identifying
opposite sides:

and this is a torus:

5
We shall first consider surfaces as topological spaces. The remarkable thing here is
that they are completely classified up to homeomorphism. Each surface belongs to
two classes the orientable ones and the non-orientable ones and within each class
there is a non-zero integer which determines the surface. The orientable ones are the
ones you see sitting in Euclidean space and the integer is the number of holes. The
non-orientable ones are the one-sided surfaces those that contain a Mobius strip
and projective space is just such a surface. If we take the hemisphere above and
flatten it to a disc, then projective space is obtained by identifying opposite points
on the boundary:

Now cut out a strip:

and the identification on the strip gives the Mobius band:

6
As for the integer invariant, it is given by the Euler characteristic if we subdivide
a surface A into V vertices, E edges and F faces then the Euler characteristic (A)
is defined by
(A) = V E + F.
For a surface in Euclidean space with g holes, (A) = 2 2g. The invariant has
the wonderful property, like counting the points in a set, that

(A B) = (A) + (B) (A B)

and this means that we can calculate it by cutting up the surface into pieces, and
without having to imagine the holes.

One place where the study of surfaces appears is in complex analysis. We know that
log z is not a single valued function as we continue around the origin it comes back
to its original value with 2i added on. We can think of log z as a single valued
function on a surface which covers the non-zero complex numbers:

7
The Euclidean picture above is in this case a reasonable one, using the third coordinate
to give the imaginary part of log z: the surface consists of the points (rei , )
3
C
p R = R and log z = log r + i is single-valued. But if you do the same to
z(z 1) you get

a surface with self-intersections, a picture which is not very helpful.

p The way out is
to leave R3 behind and construct an abstract surface on which z(z 1) is single-
valued. This is an p example of a Riemann surface. pRiemann surfaces are always
orientable, and for z(z 1) we get a sphere. For z(z 1)(z a) it is a torus,
which amongst other things is the reason that you cant evaluate
dx
Z
p
x(x 1)(x a)
using elementary functions. In general, given a multi-valued meromorphic function,
the Euler characteristic of the Riemann surface on which it is defined can be found
by a formula called the Riemann-Hurwitz formula.

We can look at a smooth surface in Euclidean space in many ways as a topological

space as above, or also as a Riemannian manifold. By this we mean that, using the
Euclidean metric on R3 , we can measure the lengths of curves on the surface.

8
If our surface is not sitting in Euclidean space we can consider the same idea, which
is called a Riemannian metric. For example, if we think of the torus by identifying
the sides of a square, then the ordinary length of a curve in the plane can be used to
measure the length of a curve on the torus:

A Riemannian metric enables you to do much more than measure lengths of curves: in
particular you can define areas, curvature and geodesics. The most important notion
of curvature for us is the Gaussian curvature which measures the deviation of formulas
for triangles from the Euclidean ones. It allows us to relate the differential geometry
of the surface to its topology: we can find the Euler characteristic by integrating the
Gauss curvature over the surface. This is called the Gauss-Bonnet theorem. There
are other analytical ways of getting the Euler characteristic one is to count the
critical points of a differentiable function.

Surfaces with constant Gaussian curvature have a special role to play. If this curvature
is zero then locally we are looking at the Euclidean plane, if positive it is the round
sphere, but the negative case is the important area of hyperbolic geometry. This has
a long history, but we shall consider the concrete model of the upper half-plane as a
surface with a Riemannian metric, and show how its geodesics and isometries provide
the axiomatic properties of non-Euclidean geometry and also link up with complex
analysis. The hyperbolic plane is a surface as concrete as one can imagine, but is an
abstract one in the sense that it is not in R3 .

2 The topology of surfaces

2.1 The definition of a surface

We are first going to consider surfaces as topological spaces, so lets recall some basic
properties:

9
Definition 1 A topological space is a set X together with a collection T of subsets
of X (called the open subsets of X) such that

T and X T ;

if U, V T then U V T ;
S
if Ui T i I then iI Ui T .

X is called Hausdorff if whenever x, y X and x 6= y there are open subsets

U, V of X such that x U and y V and U V = .

A map f : X Y between topological spaces X and Y is called continuous if

f 1 (V ) is an open subset of X whenever V is an open subset of Y .

f : X Y is called a homeomorphism if it is a bijection and both f : X Y

and its inverse f 1 : Y X are continuous. Then we say that X is homeomor-
phic to Y .

X is called compact if every open cover of X has a finite subcover.

Subsets of Rn are Hausdorff topological spaces where the open sets are just the
intersections with open sets in Rn . A surface has the property that near any point
it looks like Euclidean space just like the surface of the spherical Earth. More
precisely:

Definition 2 A topological surface (sometimes just called a surface) is a Hausdorff

topological space X such that each point x of X is contained in an open subset U
which is homeomorphic to an open subset V of R2 .
X is called a closed surface if it is compact.

A surface is also sometimes called a 2-manifold or a manifold of dimension 2. For any

natural number n a topological n-manifold is a Hausdorff topological space X which
is locally homeomorphic to Rn .

Remark: (i) The Heine-Borel theorem tells us that a subset of Rn is compact if

and only if it is closed (contains all its limit points) and bounded. Thus the use of the
terminology closed surface for a compact surface is a little perverse: there are plenty
of surfaces which are closed subsets of R3 , for example, but which are not closed
surfaces.

10
(ii) Remember that the image of a compact space under a continuous map is always
compact, and that a bijective continuous map from a compact space to a Hausdorff
space is a homeomorphism.

Example: The sphere. The most popular way to see that this is a surface according
to the definition is stereographic projection:

Here one open set U is the complement of the South Pole and projection identifies it
with R2 , the tangent plane at the North Pole. With another open set the complement
of the North Pole we see that all points are in a neighbourhood homeomorphic to R2 .

We constructed other surfaces by identification at the boundary of a planar figure.

Any subset of the plane has a topology but we need to define one on the space
obtained by identifying points. The key to this is to regard identification as an
equivalence relation. For example, in constructing the torus from the square we
define (x, 0) (x, 1) and (0, y) (1, y) and every other equivalence is an equality.
The torus is the set of equivalence classes and we give this a topology as follows:

Definition 3 Let be an equivalence relation on a topological space X. If x X

let [x] = {y X : y x} be the equivalence class of x and let

X/= {[x] : x X}

be the set of equivalence classes. Let : X X/ be the quotient map which sends
an element of X to its equivalence class. Then the quotient topology on X/ is given
by
{V X/: 1 (V ) is an open subset of X}.

11
In other words a subset V of X/ is an open subset of X/ (for the quotient topology)
if and only if its inverse image

1 (V ) = {x X : [x] V }

is an open subset of X.

So why does the equivalence relation on the square give a surface? If a point lies
inside the square we can take an open disc around it still in the interior of the square.
There is no identification here so this neighbourhood is homeomorphic to an open disc
in R2 . If the chosen point lies on the boundary, then it is contained in two half-discs
DL , DR on the left and right:

We need to prove that the quotient topology on these two half-discs is homeomorphic
to a full disc. First take the closed half-discs and set B = DL DR . The map
x 7 x + 1 on DL and x 7 x on DR is a continuous map from B (with its topology
from R2 ) to a single disc D. Moreover equivalent points go to the same point so it is
a composition
B B/ D.
The definition of the quotient topology tells us that B/ D is continuous. It is also
bijective and B/, the continuous image of the compact space B, is compact so this
is a homeomorphism. Restrict now to the interior and this gives a homeomorphism
from a neighbourhood of a point on the boundary of the square to an open disc.
If the point is a corner, we do a similar argument with quadrants.
Thus the torus defined by identification is a surface. Moreover it is closed, since it is
the quotient of the unit square which is compact.

Here are more examples by identification of a square:

12
 The sphere

Projective space

The Klein bottle

13
 The Mobius band

The Mobius band is not closed, as the dotted lines suggest. Here is its rigorous
definition:

Definition 4 A Mobius band (or Mobius strip) is a surface which is homeomorphic

to
(0, 1) [0, 1]/
with the quotient topology, where is the equivalence relation given by

(x, y) (s, t) iff (x = s and y = t) or (x = 1 s and {y, t} = {0, 1}).

2.2 Planar models and connected sums

The examples above are obtained by identifying edges of a square but we can use
any polygon in the plane with an even number of sides to construct a closed surface
so long as we prescribe the way to identify the sides in pairs. Drawing arrows then
becomes tiresome so we describe the identification more systematically: going round
clockwise we give each side a letter a say, and when we encounter the side to be
identified we call it a if the arrow is in the same clockwise direction and a1 if it is
the opposite. For example, instead of

14
we call the top side a and the bottom b and get

aa1 bb1 .

This is the sphere. Projective space is then abab, the Klein bottle abab1 and the
torus aba1 b1 . Obviously the cyclic order is not important. There are lots of planar
models which define the same surface. The sphere for example can be defined not
just from the square but also by aa1 , a 2-sided polygon:

and similarly the projective plane is aa.

Can we get new surfaces by taking more sides? Certainly, but first lets consider
another construction of surfaces. If X and Y are two closed surfaces, remove a small
open disc from each. Then take a homeomorphism from the boundary of one disc
to the boundary of the other. The topological space formed by identifying the two
circles is also a surface called the connected sum X#Y . We can also think of it as
joining the two by a cylinder:

15
The picture shows that we can get a surface with two holes from the connected sum
of two tori. Lets look at this now from the planar point of view.
First remove a disc whose boundary passes through a vertex but otherwise misses the
sides:

Now open it out:

and paste two copies together:

16
This gives an octagon, and the identification is given by the string of letters:

aba1 b1 cdc1 d1 .

Lets look more closely at what we have done here. The open disc we removed from
the square had no points on the boundary where the equivalence plays a role, so we
removed an open disc from the torus. On the other hand opening it out means
that we also removed the vertex and then replaced it by two points A and B.
CORRECTIONS
A
B
p 15 open disc
p16 There are no CORRECTIONS
equivalences to be applied to the small disc.
A you may object that opening it up means removing not just the in
of that
p 15 open disc disc, but the vertex it is emanating from, and then replacin
vertex by two points. But remember that the vertex belongs to a
p16 There are no equivalences
equivalence class which intothebefirst
applied toconsists
picture the small disc.
of all four vertices,
equivalence class of the three remaining vertices is still there. Moreo
you may object that opening it up means removing not just the int
the identifications of the second picture there is again a single equiv
of that class
disc, ofbut the vertex
vertices. So all it
weishave
emanating
done is tofrom, and
remove an then replacing
open disc.
vertex by two points. But remember that the vertex belongs to a s
p18 too
equivalence many
class whichNowsin the first picture consists of all four vertices, so
p21 update
equivalence class ofthe
thereference to other approaches
three remaining vertices isto still
triangulations
there. Moreov
the identifications of Euler
p28 define the the second pictureforthere
characteristic is again
objects a single
more general equival
than surfa
class of vertices. So all we have done is to remove an open disc.
p30 replaced by .... by
p18 too p31
many Nows the Riemann surfaces first: conformal mapping rather
17
motivate
functions
p21 update the reference to other approaches to triangulations
p33the
p28 define emphasize holomorphic functions
Euler characteristic for objects more general than surfac
p35 form a field this algebraic approach
But all four vertices of the square belonged to the same equivalence class, so the
removed vertex is still represented on the torus by the equivalence class of the other
three. Putting in A, B in the picture above and performing the indicated identifica-
tions turns all of the vertices to a single point. So it is a planar model which represents
a torus with an open disc removed. The boundary circle of the disc is now the edge
AB with end points identified. So the octagon really is a model for the connected
sum of the two tori.
Its not hard to see that this is the general pattern: a connected sum can be repre-
sented by placing the second string of letters after the first. So in particular

a1 b1 a1 1 1 1 1 1
1 b1 a2 b2 a2 b2 . . . ag bg ag bg

describes a surface in R3 with g holes.

Note that when we defined a torus from a square, all four vertices are equivalent and
this persists when we take the connected sum as above. The picture of the surface
one should have then is 2g closed curves emanating from a single point, and the
complement of those curves is homeomorphic to an open disc the interior of the
polygon.

If S is a sphere, then removing a disc just leaves another disc so connected sum with
S takes out a disc and replaces it. Thus

X#S = X.

Connected sum with the projective plane P is sometimes called attaching a cross-cap.
In fact, removing a disc from P gives the Mobius band

18
so we are just pasting the boundary circle of the Mobius band to the boundary of the
disc. It is easy to see then that the connected sum P #P is the Klein bottle.
You cant necessarily cancel the connected sum though: it is not true that X#A =
Y #A implies X = Y . Here is an important example:

Proposition 2.1 The connected sum of a torus T and the projective plane P is
homeomorphic to the connected sum of three projective planes.

Proof: From the remark above it is sufficent to prove that P #T = P #K where

K is the Klein bottle. Since P can be described by a 2-gon with relation aa and the
Klein bottle is bcbc1 , P #K is defined by a hexagon and the relation aabcbc1 .

Now P #T is aabcb1 c1 :

19
Cut along the dotted line...

... detach the triangle and turn it over...

... reattach...

20
... cut down the middle...

... turn the left hand quadrilateral over and paste together again...

...and this is aabcbc1 .

21
 2

2.3 The classification of surfaces

The planar models allow us to classify surfaces. We shall prove the following

Theorem 2.2 A closed, connected surface is either homeomorphic to the sphere, or

to a connected sum of tori, or to a connected sum of projective planes.

We sketch the proof below (this is not examinable) and refer to [2] or [1] for more
details. We have to start somewhere, and the topological definition of a surface is
quite general, so we need to invoke a theorem beyond the scope of this course: any
closed surface X has a triangulation: it is homeomorphic to a space formed from the
disjoint union of finitely many triangles in R2 with edges glued together in pairs.
For a Riemann surface (see next section), we can directly find a triangulation so long
as we have a meromorphic function, and that is also a significant theorem. If you do
a bit more of the differential geometry of surfaces than we do here then the study
of geodesics leads to the notion of convex neighbourhoods and you can use geodesic
triangles. But both of these use structure beyond the topological definition. Take a
look at http://mathoverflow.net/questions/17578/triangulating-surfaces if you want
to see an accessible proof.
We shall proceed by using a planar model.

Now take one triangle on the surface, and choose a homeomorphism to a planar
triangle. Take an adjacent one and the common edge and choose a homeomorphism
to another plane triangle and so on... Since the surface is connected the triangles form
a polygon and thus X can be obtained from this polygon with edges glued together in
pairs. It remains to systematically reduce this, without changing the homeomorphism
type, to a standard form.

Step 1: Adjacent edges occurring in the form aa1 or a1 a can be eliminated.

22
Step 2: We can assume that all vertices must be identified with each other. To see
this, suppose Step 1 has been done, and we have two adjacent vertices in different
equivalence classes: red and yellow. Because of Step 1 the other side going through
the yellow vertex is paired with a side elsewhere on the polygon. Cut off the triangle
and glue it onto that side:

The result is the same number of sides but one less yellow and one more red vertex.
Eventually, applying Step 1 again, we get to a single equivalence class.
Step 3: We can assume that any pair of the form a and a are adjacent, by cutting
and pasting:

We now have a single equivalence class of vertices and all the pairs a, a are adjacent.
What about a pair a, a1 ? If they are adjacent, Step 1 gets rid of them, if not we
have this:

23
If all the sides on the top part have their partners in the top part, then their vertices
will never be equivalent to a vertex in the bottom part. But Step 2 gave us one
equivalence class, so there is a b in the top half paired with something in the bottom.
It cant be b because Step 3 put them adjacent, so it must be b1 .
Step 4: We can reduce this to something of the form cdc1 d1 like this. First cut
off the top and paste it to the bottom.

Now cut away from the left and paste it to the right.

Finally our surface is described by a string of terms of the form aa or bcb1 c1 : a

connected sum of projective planes and tori. However, if there is at least one projective
plane we can use Proposition 2.1 which says that P #T = P #P #P to get rid of the
tori.

2.4 Orientability
Given a surface, we need to be able to decide what connected sum it is in the Classi-
fication Theorem without cutting it into pieces. Fortunately there are two concepts,
which are invariant under homeomorphism, which do this. The first concerns orien-
tation:

24
Definition 5 A surface X is orientable if it contains no open subset homeomorphic
to a Mobius band.

From the definition it is clear that if X is orientable, any surface homeomorphic to

X is too.
We saw that taking the connected sum with the projective plane means attaching a
Mobius band, so the surfaces which are connected sums of P are non-orientable. We
need to show that connected sums of tori are orientable. For this, we observe that
the connected sum operation works for tori in R3 embedded in the standard way:

so a connected sum of tori can also be embedded in R3 . The sketch proof below
assumes our surfaces are differentiable we shall deal with these in more detail later.

Suppose for a contradiction that X is a non-orientable compact smooth surface in

R3 . Then X has an open subset which is homeomorphic to a Mobius band, which
means that we can find a loop (i.e. a closed path) in X such that the normal to
X, when transported around the loop in a continuous fashion, comes back with the
opposite direction. By considering a point on the normal a small distance from X,
moving it around the loop and then connecting along the normal from one side of X
to the other, we can construct a closed path : [0, 1] R3 in R3 which meets X
at exactly one point and is transversal to X at this point (i.e. the tangent to at x
is not tangent to X). It is a general fact about the topology of R3 that any closed
differentiable path : [0, 1] R3 can be filled in with a disc; more precisely there
is a differentiable map f : D R3 , where D = {(x, y) R2 |x2 + y 2 1}, such that

(t) = f (cos 2t, sin 2t)

for all t [0, 1]. Now we can perturb f a little bit, without changing or the values
of f on the boundary of D, to make f transversal to X (i.e. the image of f is not
tangent to X at any point of intersection with X). But once f is transversal to X
it can be shown that the inverse image f 1 (X) of X in D is very well behaved: it
consists of a disjoint union of simple closed paths in the interior of D, together with
paths meeting the boundary of D in exactly their endpoints (which are two distinct
points on the boundary of D). Thus f 1 (X) contains an even number of points on

25
the boundary of D, which contradicts our construction in which f 1 (X) has exactly
one point on the boundary of D. The surface must therefore be orientable.

This argument shows why the projective plane in particular cant be embedded in
R3 . Here is an amusing corollary:

Proposition 2.3 Any simple closed curve in the plane contains an inscribed rectan-
gle.

Proof: The closed curve C is homeomorphic to the circle. Consider the set of pairs
of points (x, y) in C. This is the product of two circles: a torus. We now want to
consider the set X of unordered pairs, so consider the planar model of the torus. We
identify (x, y) with (y, x), which is reflection about the diagonal. The top side then
gets identified with the right hand side, and under the torus identification with the
left hand side.

The set of unordered points is therefore obtained by identification on the top triangle:

26
and this is the projective plane with a disc removed (the Mobius band):

Now define a map f : X R3 as follows:

1
(x, y) 7 ( (x + y), |x y|) R2 R
2
The first term is the midpoint of the line xy and the last is the distance between
x and y. Both are clearly independent of the order and so the map is well-defined.
When x = y, which is the boundary circle of the Mobius band, the map is

x 7 (x, 0)

which is the curve C in the plane x3 = 0. Since the curve bounds a disc we can
extend f to the surface obtained by pasting the disc to X and extending f to be
the inclusion of the disc into the plane x3 = 0. This is a continuous map (it can be
perturbed to be differentiable if necessary) of the projective plane P to R3 . Since P
is unorientable it cant be an embedding so we have at least two pairs (x1 , y1 ), (x2 , y2 )
with the same centre and the same separation. These are the vertices of the required
rectangle. 2

27
2.5 The Euler characteristic
It is a familar fact (already known to Descartes in 1639) that if you divide up the
surface of a sphere into polygons and count the number of vertices, edges and faces
then

V E + F = 2.

This number is the Euler characteristic, and we shall define it for any surface. First
we have to define our terms:

Definition 6 A subdivision of a compact surface X is a partition of X into

i) vertices (these are finitely many points of X),
ii) edges ( finitely many disjoint subsets of X each homeomorphic to the open interval
(0, 1)), and
iii) faces ( finitely many disjoint open subsets of X each homeomorphic to the open
disc {(x, y) R2 : x2 + y 2 < 1} in R2 ,
such that
a) the faces are the connected components of X \ {vertices and edges},
b) no edge contains a vertex, and
c) each edge begins and ends in a vertex (either the same vertex or different vertices),
or more precisely, if e is an edge then there are vertices v0 and v1 (not necessarily
distinct) and a continuous map
f : [0, 1] e {v0 , v1 }
which restricts to a homeomorphism from (0, 1) to e and satisfies f (0) = v0 and
f (1) = v1 .

Definition 7 The Euler characteristic (or Euler number) of a compact surface X

with a subdivision is
(X) = V E + F

28
where V is the number of vertices, E is the number of edges and F is the number of
faces in the subdivision.

The fact that a closed surface has a subdivision follows from the existence of a trian-
gulation. The most important fact is

Theorem 2.4 The Euler characteristic of a compact surface is independent of the

subdivision

which we shall sketch a proof of later. Note that we can define a subdivision for more
general topological spaces than closed surfaces, for example a triangle has one face,
3 vertices and 3 edges and hence Euler characteristic equal to 1.

A planar model provides a subdivision of a surface. We have one face the interior of
the polygon and if there are 2n sides to the polygon, these get identified in pairs so
there are n edges. For the vertices we have to count the number of equivalence classes,
but in the normal form of the classification theorem, we created a single equivalence
class. In that case, the Euler characteristic is

1 n + 1 = 2 n.

The connected sum of g tori had 4g sides in the standard model a1 b1 a1 1 1 1

1 b1 . . . ag bg ag bg
so in that case (X) = 2 2g. The connected sum of g projective planes has 2g sides
so we have (X) = 2 g. We then obtain:

Theorem 2.5 A closed surface is determined up to homeomorphism by its orientabil-

ity and its Euler characteristic.

This is a very strong result: nothing like this happens in higher dimensions.

To calculate the Euler characteristic of a given surface we dont necessarily have to

go to the classification. Suppose a surface is made up of the union of two spaces X
and Y , such that the intersection X Y has a subdivision which is a subset of the
subdivisions for X and for Y . Then since V, E and F are just counting the number
of elements in a set, we have immediately that

(X Y ) = (X) + (Y ) (X Y ).

We can deal with a connected sum this way. Take a closed surface X and remove a
disc D to get a space X o . The disc has Euler characteristic 1 (a polygon has one face,

29
n vertices and n sides) and the boundary circle has Euler characteristic 0 (no face).
So applying the formula,

(X) = (X o D) = (X o ) + (D) (X o D) = (X o ) + 1.

To get the connected sum we paste X o to Y o along the boundary circle so

(X#Y ) = (X o )+(Y o )(X o Y o ) = (X)1+(Y )10 = (X)+(Y )2.

In particular, (X#T ) = (X) 2 so this again gives the value 2 2g for the
connected sum of g tori.

To make all this work we finally need:

Theorem 2.6 The Euler characteristic (X) of a compact surface X is a topological

invariant.

We give a sketch proof (which is not examinable).

Proof:
The idea is to give a different definition of (X) which makes it clear that it is a
topological invariant, and then prove that the Euler characteristic of any subdivision
of X is equal to (X) defined in this new way.
For each continuous path f : [0, 1] X define its boundary f to be the formal
linear combination of points f (0) + f (1). If g is another map and g(0) = f (1) then,
with coefficients in Z/2, we have

f + g = f (0) + 2f (1) + g(1) = f (0) + g(1)

which is the boundary of the path obtained by sticking these two together. Let C0 be
the vector space of finite linear combinations of points with coefficients in Z/2 and C1
the linear combinations of paths, then : C1 C0 is a linear map. If X is connected
then any two points can be joined by a path, so that x C0 is in the image of if
and only if it has an even number of terms.
Now look at continuous maps of a triangle ABC = to X and the space C2 of all
linear combinations of these. The boundary of F : X is the sum of the three
paths which are the restrictions of F to the sides of the triangle. Then

F = (F (A) + F (B)) + (F (B) + F (C)) + (F (C) + F (A)) = 0

30
so that the image of : C2 C1 is contained in the kernel of : C1 C0 . We define
H1 (X) to be the quotient space. This is clearly a topological invariant because we
only used the notion of continuous functions to define it.

If we take X to be a surface with a subdivision, one can show that because each face
is homeomorphic to a disc, any element in the kernel of : C1 C0 can be replaced
by a linear combination of edges of the subdivision upon adding something in C2 :

Now we let V, E and F be vector spaces over Z/2 with bases given by the sets of
vertices, edges and faces of the subdivision, then define boundary maps in the same
way
: E V and : F E.
Then
ker( : E V)
H1 (X)
= .
im( : F E)
By the rank-nullity formula we get
dim H1 (X) = dim E rk( : E V) dim F + dim ker( : F E).
Because X is connected the image of : E V consists of sums of an even number
of vertices so that
dim V = 1 + rk( : E V).
Also ker( : F E) is clearly spanned by the sum of the faces, hence
dim ker( : F E) = 1
so
dim H1 (X) = 2 V + E F.
This shows that V E + F is a topological invariant. 2

31
3 Riemann surfaces

3.1 Definitions and examples

From the definition of a surface, each point has a neighbourhood U and a homeomor-
phism U from U to an open set V in R2 . If two such neighbourhoods U, U 0 intersect,
then
U 0 1 0 0
U : U (U U ) U 0 (U U )

is a homeomorphism from one open set of R2 to another.

U
U

If we identify R2 with the complex numbers C then we can define:

Definition 8 A Riemann surface is a surface with a class of homeomorphisms U

such that each map U 0 1
U is a holomorphic (or analytic) homeomorphism.

We call each function U a holomorphic coordinate.

In your course on complex analysis you used holomorphic functions in two ways: one
involved adding, multiplying, differentiating and taking contour integrals; the other
concerned conformal mappings, taking one domain to another, generally in order to
simplify a contour integral. It is this second viewpoint which we use in this definition.

Examples:
1. Let X be the extended complex plane X = C {}. Let U = C with U (z) =
z C. Now take
U 0 = C\{0} {}

32
and define z 0 = U 0 (z) = z 1 C if z 6= and U 0 () = 0. Then
U (U U 0 ) = C\{0}
and
U 1
U 0 (z) = z
1

which is holomorphic.
In the right coordinates this is the sphere, with the North Pole and the coordinate
maps given by stereographic projection. For this reason it is sometimes called the
Riemann sphere.

2. Let 1 , 2 C be two complex numbers which are linearly independent over the
reals, and define an equivalence relation on C by z1 z2 if there are integers m, n such
that z1 z2 = m1 + n2 . Let X be the set of equivalence classes (with the quotient
topology). A small enough disc V around z C has at most one representative in
each equivalence class, so this gives a local homeomorphism to its projection U in X.
If U and U 0 intersect, then the two coordinates are related by a map
z 7 z + m1 + n2
which is holomorphic.
This surface is topologically described by noting that every z is equivalent to one
inside the closed parallelogram whose vertices are 0, 1 , 2 , 1 + 2 , but that points
on the boundary are identified:

We thus get a torus this way. Another way of describing the points of the torus is as
orbits of the action of the group Z Z on C by (m, n) z = z + m1 + n2 .

3. The parallelograms in Example 2 fit together to tile the plane. There are groups
of holomorphic maps of the unit disc into itself for which the interior of a polygon

33
plays the same role as the interior of the parallelogram in the plane, and we get a
surface X by taking the orbits of the group action. Now we get a tiling of the disc:

In this example the polygon has eight sides and the surface is homeomorphic by the
classification theorem to the connected sum of two tori.

4. A complex algebraic curve X in C2 is given by

X = {(z, w) C2 : f (z, w) = 0}
where f is a polynomial in two variables with complex coefficients. If (f /z)(z, w) 6=
0 or (f /w)(z, w) 6= 0 for every (z, w) X, then using the implicit function theorem
(see Appendix A) X can be shown to be a Riemann surface with local homeomor-
phisms given by
(z, w) 7 w where (f /z)(z, w) 6= 0
and
(z, w) 7 z where (f /w)(z, w) 6= 0.

Definition 9 A holomorphic map between Riemann surfaces X and Y is a continu-

ous map f : X Y such that for each holomorphic coordinate U on U containing
x on X and W defined in a neighbourhood of f (x) on Y , the composition
W f 1
U

is holomorphic.

In particular if we take Y = C, we can define holomorphic functions on X, and then

we can use the ring structure of C to add and multiply these functions.

Before proceeding, recall some basic facts about holomorphic functions (see [3]):

34
 A holomorphic function has a convergent power series expansion in a neigh-
bourhood of each point at which it is defined:

f (z) = a0 + a1 (z c) + a2 (z c)2 + . . .

If f vanishes at c then

f (z) = (z c)m (c0 + c1 (z c) + . . .)

where c0 6= 0. In particular zeros are isolated.

If f is non-constant it maps open sets to open sets.

|f | cannot attain a maximum at an interior point of a disc (maximum modulus

principle).

f : C 7 C preserves angles between differentiable curves, both in magnitude

and sense.

This last property shows:

Proposition 3.1 A Riemann surface is orientable.

Proof: Assume X contains a Mobius band, and take a smooth curve down the
centre: : [0, 1] X. In each small coordinate neighbourhood of a point on the
curve U is a curve in a disc in C, and rotating the tangent vector 0 by 90 or 90
defines an upper and lower half:
Identification on an overlapping neighbourhood is by a map which preserves angles,
and in particular the sense anticlockwise or clockwise so the two upper halves
agree on the overlap, and as we pass around the closed curve the strip is separated
into two halves. But removing the central curve of a Mobius strip leaves it connected:
which gives a contradiction. 2

From the classification of surfaces we see that a closed, connected Riemann surface
is homeomorphic to a connected sum of tori.

35
3.2 Meromorphic functions
Recall that on a closed (i.e. compact) surface X, any continuous real function achieves
its maximum at some point x. Let X be a Riemann surface and f a holomorphic
function, then |f | is continuous, so assume it has its maximum at x. Since f 1U is a
holomorphic function on an open set in C containing U (x), and has its maximum
modulus there, the maximum modulus principle says that f must be a constant c in
a neighbourhood of x. If X is connected, it follows that f = c everywhere.
Though there are no holomorphic functions, there do exist meromorphic functions:

Definition 10 A meromorphic function f on a Riemann surface X is a holomorphic

map to the Riemann sphere S = C {}.

This means that if we remove f 1 (), then f is just a holomorphic function F with
values in C. If f (x) = , and U is a coordinate neighbourhood of x, then using
the coordinate z 0 , f 1
U is holomorphic. But z = 1/z if z 6= 0 which means that
(F U1 )1 is holomorphic. Since it also vanishes,
a0
F 1
U = + ...
zm
which is usually what we mean by a meromorphic function.

Example: A rational function

p(z)
f (z) =
q(z)

where p and q are polynomials is a meromorphic function on the Riemann sphere S.

The definition above is a geometrical one a map from one surface to another. On the
other hand, if we think of it as a function with singularities, we can add and multiply
meromorphic functions form a field which is the algebraic approach. These two
viewpoints can be very valuable. The second one allows us to manipulate with ease:
here is an example using the algebraic approach of a meromorphic function on the
torus in Example 2.
Define  
1 X 1 1
(z) = 2 + 2
2
z 6=0
(z )

36
where the sum is over all non-zero = m1 + n2 . Since for 2|z| < ||
 
 1 1  |z|
 (z )2 2  10 ||3


this converges uniformly on compact sets so long as

X 1
< .
6=0
||3

But m1 + n2 is never zero if m, n are real so we have an estimate

|m1 + n2 | k m2 + n2

so by the integral test we have convergence. Because the sum is essentially over all
equivalence classes
(z + m1 + n2 ) = (z)
so that this is a meromorphic function on the surface X. It is called the Weierstrass
P-function.

It is a quite deep result that any closed Riemann surface has meromorphic functions.
We are now going to consider them in more detail from the geometric point of view.
So let
f :XS
be a meromorphic function. If the inverse image of a S is infinite, then it has
a limit point x by compactness of X. In a holomorphic coordinate around x with
z(x) = 0, f is defined by a holomorphic function F = f 1 U with a sequence of points
zn 0 for which F (zn ) a = 0. But the zeros of a holomorphic function are isolated,
so we deduce that f 1 (a) is a finite set. By a similar argument the points at which
the derivative F 0 vanishes are finite in number (check using the chain rule that this
condition is independent of the holomorphic coordinate). The points of X at which
F 0 = 0 are called ramification points.
The word ramification means branching. We defined it here analytically through
the vanishing of a derivative, but we need to understand its geometric meaning. The
simplest example is the map f (z) = z 2 from C to C so that z = 0 is a ramification
point. In a neighbourhood of zero there
is no single-valued inverse to f in complex
analysis we say that the square root w has two branches. The origin has one inverse
image, any other point has two, but we cant distinguish between them because as w
goes around a circle surrounding the origin one square root extends continuously to
its negative. A similar phenomenon holds for the map z 7 z n .

37
In fact, if f is any holomorphic function on C such that f 0 (0) = 0, we have

f (z) = z n (a0 + a1 z + . . .)

with a0 6= 0. We can expand

1/n
(a0 + a1 z + . . .)1/n = a0 (1 + b1 z + . . .)

in a power series and define

1/n
w = a0 z(1 + b1 z + . . .).

Since w0 (0) 6= 0 we can think of w as a new coordinate and then the map becomes
simply
w 7 wn .
So, thinking geometrically of C as a Riemann surface where we are allowed to change
coordinates, a ramification point is a map of the form z 7 z n . The integer n is its
multiplicity.

So now return to a holomorphic map f : X S. There are two types of points: if

F 0 (x) 6= 0, then the inverse function theorem tells us that f maps a neighbourhood
Ux of x X homeomorphically to a neighbourhood Vx of f (x) S. Define V to be
the intersection of the Vx as x runs over the finite set of points such that f (x) = a,
then f 1 V consists of a finite number d of open sets, each mapped homeomorphically
onto V by f :

V
f
If F 0 (x) = 0 then the map looks like w 7 wn . The inverse image of f (x) = a is then
a disjoint union of open sets, on some of which the map might map homeomorphicall
to a disc, but where on at least one (containing x) the map is of the form z 7 z n .

Removing the finite number of images under f of ramification points we get a sphere
minus a finite number of points. This is connected. The number of points in the
inverse image of a point in this punctured sphere is integer-valued and continuous,
hence constant. It is called the degree d of the meromorphic function f .

With this we can determine the Euler characteristic of the Riemann surface S from
the meromorphic function:

38
Theorem 3.2 (Riemann-Hurwitz) Let f : X S be a meromorphic function of
degree d on a closed connected Riemann surface X, and suppose it has ramification
points x1 , . . . , xn where the local form of f (x) f (xk ) is a holomorphic function with
a zero of multiplicity mk . Then
n
X
(X) = 2d (mk 1)
k=1

Proof: The idea is to take a triangulation of the sphere S such that the image of the
ramification points are vertices. This is straighforward. Now take a finite subcovering
of S by open sets of the form V above where the map f is either a homeomorphism
or of the form z 7 z m . Subdivide the triangulation into smaller triangles such that
each one is contained in one of the sets V . Then the inverse images of the vertices
and edges of S form the vertices and edges of a triangulation of X.
If the triangulation of S has V vertices, E edges and F faces, then clearly the tri-
angulation of X has dE edges and dF faces. It has fewer vertices, though in a
neighbourhood where f is of the form w 7 wm the origin is a single vertex instead
of m of them. For each ramification point of order mk we therefore have one vertex
instead of mk . The count of vertices is therefore
n
X
dV (mk 1).
k=1

Thus n n
X X
(X) = d(V E + F ) (mk 1) = 2d (mk 1)
k=1 k=1

using (S) = 2. 2

Clearly the argument works just the same for a holomorphic map f : X Y and
then n
X
(X) = d(Y ) (mk 1).
k=1

As an example, consider the Weierstrass P-function : T S:

1 X 1 1

(z) = 2 + .
z 6=0
(z )2 2

39
We constructed this by adding and multiplying but we want to know geometrically
what the map from a torus to a sphere looks like.

Firstly, has degree 2 since (z) = only at z = 0 and there it has multiplicity 2.
The multiplicity of a ramification point cannot be bigger than this beacuse then it will
look like z 7 z n and a non-zero point will have at least n inverse images. Thus the
only possible value at the ramification points here is mk = 2. The Riemann-Hurwitz
formula gives:
0=4n
so there must be exactly 4 ramification points. In fact we can see them directly,
because (z) is an even function, so the derivative vanishes if z = z. Of course
at z = 0, (z) = so we should use the other coordinate on S: 1/ has a zero of
multiplicity 2 at z = 0. To find the other points recall that is doubly periodic so
0 vanishes where
z = z + m1 + n2
for some integers m, n, and these are the four points
0, 1 /2, 2 /2, (1 + 2 )/2 :

The geometric Riemann-Hurwitz formula has helped us here in the analysis by show-
ing us that the only zeros of 0 are the obvious ones.

3.3 A new look at the torus

Viewing the torus the points z C moduli z z + m1 + n2 is not the only way to
see it as a Riemann surface. We shall now look at it, via the P-function, in a manner
which will show us how to construct other Riemann surfaces.

40
So consider again the P-function, thought of as a degree 2 map : T S. It has 4
ramification points, whose images are and the three finite points e1 , e2 , e3 where

e1 = (1 /2), e2 = (2 /2), e3 = ((1 + 2 )/2).

So its derivative 0 (z) vanishes only at three points, each with multiplicity 1. At each
of these points has the local form

(z) = e1 + (z 1 /2)2 (a0 + . . .)

and so
1
((z) e1 )((z) e2 )((z) e3 )
0 (z)2
is a well-defined holomorphic function on T away from z = 0. But (z) z 2 near
z = 0, and so 0 (z) 2z 3 so this function is finite at z = 0 with value 1/4. By the
maximum argument, since T is compact, the function is a constant, namely 1/4, and

0 (z)2 = 4((z) e1 )((z) e2 )((z) e3 ) (1)

When we introduced the torus as Example 2, then z or z + c were local holomorphic

coordinates. But now if z 6= 0, (z) is finite and if 0 (z) 6= 0 this gives us another
local coordinate. From (1), 20 00 = 4q 0 ()0 where q(x) = (x e1 )(x e2 )(x e3 )
and so
1
00 (ei ) = q 0 (ei )
2
which is non-zero since e1 , e2 , e3 are distinct. This means that 0 (z) is a local coordi-
nate near z = i /2.
Finally, near z = 0, 0
0 (z)

1
= = 2z + . . .
(z) 2 (z)
which means that 0 (z)/2 (z) is a coordinate near z = 0.

We now see the torus rather differently. Consider

C = {(x, y) C2 : y 2 = q(x) = 4(x e1 )(x e2 )(x e3 )}.

Now : T S is surjective (otherwise the degree would be zero!), so (z) takes every
value in C = S \ {}. Moreover, since (z) = (z), we have 0 (z) = 0 (z)
hence for each value of x there is a value of z for which for 0 (z) takes each of the two
values of y. Thus (x, y) = ((z), 0 (z)) defines a homeomorphism from T \ {0} to C.
So we have T = C {} with local coordinates:

41
 x near a if q(a) 6= 0

y near a if q(a) = 0

y/x2 near

Now generalize this to the case

C = {(x, y) C2 : y 2 = q(x) = 4(x e1 )(x e2 ) . . . (x en )}

where the ei are distinct. As above we define a local coordinate x near a if q(a) 6= 0.
Then  Z x 0 
1 q (z)
y = exp dz
2 c q(z)
defines y as a locally invertible holomorphic function of x.
Near x = ei , q 0 (x) 6= 0 so by the inverse function theorem x = f (q(x)) and x = f (y 2 )
is holomorphic, so y is a local coordinate.
Near x = put z = 1/x and if n = 2m, w = y/xm then

w2 = (1 e1 z)(1 e2 z) . . . (1 e2m z).

Since z = 0 is not a root of the polynomial, this is like the first case above and z is a
coordinate: we have then defined a Riemann surface structure on

X = C {, }

where the two extra points are given by (w = 1, z = 0).

If n = 2m + 1, put w = y/xm+1 and then

w2 = z(1 ze1 ) . . . (1 ze2m+1 )

and now since w = 0 is a root, we use w as a coordinate and define a Riemann surface
structure on
X = C {}
where the single extra point is given by (w = 0, z = 0).

The Riemann-Hurwitz formula enables us to calculate the Euler characteristic of X.

We can use x as a meromorphic function.
Firstly, given x, y 2 = q(x) has two solutions in general so the degree of the map is
two. The ramification points are where x0 = 0 but since x is a coordinate where

42
q(x) 6= 0 this can only occur at x = ei or x = . At x = ei , y is a coordinate and
y = 0. Since x = f (y 2 ), we see that x0 = 0. Because the map is of degree two the
multiplicty can only be two. At infinity we have a ramification point if n is odd and
not if n is even. Thus, applying Riemann-Hurwitz (X) = 4 n if n is even and
4 (n + 1) if n is odd.
This
p type of Riemann surface is called hyperelliptic. Since the two values of y =
q(x) only differ by a sign, we can think of (y, x) 7 (y, x) as being a holomorphic
homeomorphism from X to X, and then x is a coordinate on the space of orbits.
Topologically we can cut the surface in two an upper and lower half and
identify on the points on the boundary to get a sphere:

It is common also to view this downstairs on the Riemann sphere and insert cuts
between pairs of zeros of the polynomial p(z):

Hyperelliptic Riemann surfaces occcur in a number of dynamical problems where one

needs to integrate
du
Z
p .
(u e1 )(u e2 ) . . . (u en )
The simplest example is the pendulum:

43
 00 = (g/`) sin
which integrates once to
02 = 2(g/`) cos + c.
Substituting v = ei we get
p
v0 = i 2(g/`)(v 3 + v) + cv 2 .

By changing variables this can be brought into this form

dx
2cdt = p
(x e1 )(x e2 )(x e3 )

which can be solved with x = (ct). So time becomes (the real part of) the parameter
z on C. In the torus this is a circle, so (no surprise here!) the solutions to the
pendulum equation are periodic.

44
4 Surfaces in R3

4.1 Definitions
At this point we return to surfaces embedded in Euclidean space, and consider the
differential geometry of these:

We shall not forget the idea of an abstract surface though, and as we meet objects
which we call intrinsic we shall show how to define them on a surface which is not
sitting in R3 . These remarks are printed in a smaller typeface.

Definition 11 A smooth surface in R3 is a subset X R3 such that each point has

a neighbourhood U X and a map r : V R3 from an open set V R2 such that

r : V U is a homeomorphism

r(u, v) = (x(u, v), y(u, v), z(u, v)) has derivatives of all orders

at each point ru = r/u and rv = r/v are linearly independent.

Already in the definition we see that X is a topological surface as in Defini-

tion 2, since r defines a homeomorphism U : U V . The last two conditions
make sense if we use the implicit function theorem (see Appendix 1). This tells
us that a local invertible change of variables in R3 straightens out the sur-
face: it can be locally defined by x3 = 0 where (x1 , x2 , x3 ) are (nonlinear) local
coordinates on R3 . For any two open sets U, U 0 , we get a smooth invertible
map from an open set of R3 to another which takes x3 = 0 to x03 = 0. This
means that each map U 0 1 U is a smooth invertible homeomorphism. This
motivates the definition of an abstract smooth surface:

45
 Definition 12 A smooth surface is a surface with a class of homeomorphisms
U such that each map U 0 1
U is a smoothly invertible homeomorphism.

Clearly, since a holomorphic function has partial derivatives of all orders in

x, y, a Riemann surface is an example of an abstract smooth surface. Similarly,
we have

Definition 13 A smooth map between smooth surfaces X and Y is a contin-

uous map f : X Y such that for each smooth coordinate system U on
U containing x on X and W defined in a neighbourhood of f (x) on Y , the
composition
W f 1
U

is smooth.

We now return to surfaces in R3 :

Examples:
1) A sphere:
r(u, v) = a sin u sin v i + a cos u sin v j + a cos v k

2) A torus:
r(u, v) = (a + b cos u)(cos v i + sin v j) + b sin u k

46
These are the only compact surfaces it is easy to write down, but the following non-
compact ones are good for local discussions:

Examples:
1) A plane:
r(u, v) = a + ub + vc
for constant vectors a, b, c where b, c are linearly independent.

47
2) A cylinder:
r(u, v) = a(cos v i + sin v j) + uk

3) A cone:
r(u, v) = au cos v i + au sin v j + uk

4) A helicoid:
r(u, v) = au cos v i + au sin v j + vk

48
5) A surface of revolution:

r(u, v) = f (u)(cos v i + sin v j) + uk

6) A developable surface: take a curve (u) parametrized by arc length and set

r(u, v) = (u) + v 0 (u)

This is the surface formed by bending a piece of paper:

49
A change of parametrization of a surface is the composition

r f : V 0 R3

where f : V 0 V is a diffeomorphism an invertible map such that f and f 1 have

derivatives of all orders. Note that if

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

then by the chain rule

(r f )x = ru ux + rv vx
(r f )y = ru uy + rv vy

so     
(r f )x ux vx ru
= .
(r f )y uy vy rv
Since f has a differentiable inverse, the Jacobian matrix is invertible, so (r f )x and
(r f )y are linearly independent if ru , rv are.

Example: The (x, y) plane

r(x, y) = xi + yj
has a different parametrization in polar coordinates

r f (r, ) = r cos i + r sin j.

We have to consider changes of parametrizations when we pass from one open set V
to a neighbouring one V 0 .

Definition 14 The tangent plane (or tangent space) of a surface at the point a is
the vector space spanned by ru (a), rv (a).

50
Note that this space is independent of parametrization. One should think of the
origin of the vector space as the point a.

Definition 15 The vectors

ru rv

|ru rv |
are the two unit normals (inward and outward) to the surface at (u, v).

4.2 The first fundamental form

Definition 16 A smooth curve lying in the surface is a map t 7 (u(t), v(t)) with
derivatives of all orders such that (t) = r(u(t), v(t)) is a parametrized curve in R3 .

A parametrized curve means that u(t), v(t) have derivatives of all orders and 0 =
ru u0 +rv v 0 6= 0. The definition of a surface implies that ru , rv are linearly independent,
so this condition is equivalent to (u0 , v 0 ) 6= 0.

The arc length of such a curve from t = a to t = b is:

Z b Z bp
0
| (t)|dt = 0 0 dt
a a
Z bp
= (ru u0 + rv v 0 ) (ru u0 + rv v 0 )dt
a
Z b
= Eu02 + 2F u0 v 0 + Gv 02 dt
a

where
E = ru ru , F = ru rv , F = rv rv .

51
Definition 17 The first fundamental form of a surface in R3 is the expression
Edu2 + 2F dudv + Gdv 2
where E = ru ru , F = ru rv , G = rv rv .

The first fundamental form is just the quadratic form

Q(v, v) = v v
on the tangent space written in terms of the basis ru , rv . It is represented in this basis
by the symmetric matrix
 
E F
.
F G

So why do we write it as Edu2 + 2F dudv + Gdv 2 ? At this stage it is not worth

worrying about what exactly du2 is, instead lets see how the terminology helps to
manipulate the formulas.
For example, to find the length of a curve u(t), v(t) on the surface, we calculate
s  
2  2
du du dv dv
Z
E + 2F +G dt
dt dt dt dt
divide the first fundamental form by dt2 and multiply its square root by dt.
Furthermore if we change the parametrization of the surface via u(x, y), v(x, y) and
try to find the length of the curve (x(t), y(t)) then from first principles we would
calculate
u0 = ux x0 + uy y 0 v 0 = vx x0 + vy y 0
by the chain rule and then
Eu02 + 2F u0 v 0 + Gv 02 = E(ux x0 + uy y 0 )2 + 2F (ux x0 + uy y 0 )(vx x0 + . . .
= (Eu2x + 2F ux vx + Gvx2 )x02 + . . .
which is heavy going. Instead, using du, dv etc. we just write
du = ux dx + uy dy
dv = vx dx + vy dy
and substitute in Edu2 + 2F dudv + Gdv 2 to get E 0 dx2 + 2F 0 dxdy + G0 dy 2 . Using
matrices, we can write this transformation as
  0
E F0
   
ux uy E F ux vx
=
vx vy F G uy vy F 0 G0

52
Example: For the plane
r(x, y) = xi + yj
we have rx = i, ry = j and so the first fundamental form is

dx2 + dy 2 .

Now change to polar coordinates x = r cos , y = r sin . We have

dx = dr cos r sin d
dy = dr sin + r cos d

so that

dx2 + dy 2 = (dr cos r sin d)2 + (dr sin + r cos d)2 = dr2 + r2 d2

Here are some examples of first fundamental forms:

Examples:
1. The cylinder
r(u, v) = a(cos v i + sin v j) + uk.
We get
ru = k, rv = a( sin v i + cos v j)
so
E = ru ru = 1, F = ru rv = 0, G = rv rv = a2
giving

du2 + a2 dv 2

2. The cone
r(u, v) = a(u cos v i + u sin v j) + uk.
Here
ru = a(cos v i + sin v j) + k, rv = a(u sin v i + u cos v)
so
E = ru ru = 1 + a2 , F = ru rv = 0, G = rv rv = a2 u2
giving

53
 (1 + a2 )du2 + a2 u2 dv 2

3. The sphere
r(u, v) = a sin u sin v i + a cos u sin v j + a cos v k
gives

ru = a cos u sin v i a sin u sin v j, rv = a sin u cos v i + a cos u cos v j a sin v k

so that
E = ru ru = a2 sin2 v, F = ru rv = 0, G = rv rv = a2
and so we get the first fundamental form

a2 dv 2 + a2 sin2 vdu2

4. A surface of revolution

r(u, v) = f (u)(cos v i + sin v j) + uk

has
ru = f 0 (u)(cos v i + sin v j) + k, rv = f (u)( sin v i + cos v j)
so that

E = ru ru = 1 + f 0 (u)2 , F = ru rv = 0, G = rv rv = f (u)2

gives

(1 + f (u)02 )du2 + f (u)2 dv 2

5. A developable surface
r(u, v) = (u) + vt(u).
here the curve is parametrized by arc length u = s so that

ru = t(u) + vt0 (u) = t + vn, rv = t

where n is the normal to the curve and its curvature. This gives

(1 + v 2 2 )du2 + 2dudv + dv 2

54
 The analogue of the first fundamental form on an abstract smooth surface
X is called a Riemannian metric. On each open set U with coordinates (u, v)
we ask for smooth functions E, F, G with E > 0, G > 0, EG F 2 > 0 and on an
overlapping neighbourhood with coordinates (x, y) smooth functions E 0 , F 0 , G0
with the same properties and the transformation law:
  0
E F0
   
ux uy E F ux vx
=
vx vy F G uy vy F 0 G0
A smooth curve on X is defined to be a map : [a, b] X such that U
is smooth for each coordinate neighbourhood U on the image. The length of
such a curve is well-defined by a Riemannian metric.

Examples:
1. The torus as a Riemann surface has the metric

dzdz = dx2 + dy 2

as the local holomorphic coordinates are z and z + m1 + n2 so that the

Jacobian matrix is the identity. We could also multiply this by any positive
smooth doubly-periodic function.
2. The hyperelliptic Riemann surface w2 = p(z) where p(z) is of degree 2m has
Riemannian metrics given by
1
(a0 + a1 |z|2 + . . . + am2 |z|2(m2) )dzdz
|w|2

where the ai are positive constants.

3. The upper half-space {x + iy C : y > 0} has the metric

dx2 + dy 2
.
y2

None of these have anything to do with the first fundamental form of the surface
embedded in R3 .

We introduced the first fundamental form to measure lengths of curves on a surface

but it does more besides. Firstly if two curves 1 , 2 on the surface intersect, the
angle between them is given by

01 02
cos = (2)
| 01 || 02 |

55
But 0i = ru u0i + rv vi0 so
0i 0j = (ru u0i + rv vi0 ) (ru u0j + rv vj0 )
= Eu0i u0j + F (u0i vj0 + u0j vi0 ) + Gvi0 vj0
and each term in (2) can be expressed in terms of the curves and the coefficients of
the first fundamental form.

We can also define area using the first fundamental form:

Definition 18 The area of the domain r(U ) R3 in a surface is defined by

Z Z
|ru rv |dudv = EG F 2 dudv.
U U

The second form of the formula comes from the identity

|ru rv |2 = (ru ru )(rv rv ) (ru rv )2 = EG F 2 .

Note that the definition of area is independent of parametrization for if

rx = ru ux + rv vx , ry = ru uy + rv vy
then
rx ry = (ux vy vx uy )ru rv
so that
Z Z Z
|rx ry |dxdy = |ru rv ||ux vy vx uy |dxdy = |ru rv |dudv
U U U
using the formula for change of variables in multiple integration.

Example: Consider a surface of revolution

(1 + f 0 (u)2 )du2 + f (u)2 dv 2
and the area between u = a, u = b. We have
EG F 2 = f (u)2 (1 + f 0 (u)2 )
so the area is
Z b p Z b p
0 2
f (u) 1 + f (u) dudv = 2 f (u) 1 + f 0 (u)2 du.
a a

If a closed surface X is triangulated so that each face lies in a coordinate neighbour-

hood, then we can define the area of X as the sum of the areas of the faces by the
formula above. It is independent of the choice of triangulation.

56
4.3 Isometric surfaces
Definition 19 Two surfaces X, X 0 are isometric if there is a smooth homeomorphism
f : X X 0 which maps curves in X to curves in X 0 of the same length.

A practical example of this is to take a piece of paper and bend it: the lengths of
curves in the paper do not change. The cone and a subset of the plane are isometric
this way:

Analytically this is how to tell if two surfaces are isometric:

Theorem 4.1 The coordinate patches of surfaces U and U 0 are isometric if and only
if there exist parametrizations r : V R3 and r0 : V R3 with the same first
fundamental form.

Proof: Suppose such a parametrization exists, then the identity map is an isometry
since the first fundamental form determines the length of curves.

Conversely, suppose X, X 0 are isometric using the function f : V V 0 . Then

r 0 f : V R3 , r : V R3

are parametrizations using the same open set V , so the first fundamental forms are

Edu2 + 2F dudv + Gdv 2 , Edu2 + 2F dudv + Gdv 2

and since f is an isometry

Z p Z
Eu02 + 2F u0 v 0 + Gv 02 dt = Eu02 + 2F u0 v 0 + Gv 02 dt
I I

57
for all curves t 7 (u(t), v(t)) and all intervals. Since

d a+t
Z
h(u)du = h(a)
dt a
this means that
p
Eu02 + 2F u0 v 0 + Gv 02 = Eu02 + 2F u0 v 0 + Gv 02

for all u(t), v(t). So, choosing u, v appropriately:

u = t, v = a E = E
u = a, v = t G = G
u = t, v = t F = F

and we have the same first fundamental form as required. 2

Example:
The cone has first fundamental form

(1 + a2 )du2 + a2 u2 dv 2 .

Put
r= 1 + a2 u
then we get
a2
 
2
dr + r2 dv 2
1 + a2
and now put r
a2
= v
1 + a2
to get the plane in polar coordinates

dr2 + r2 d2 .

Note that as 0 v 2, 0 where

r
a2
= 2 < 2
1 + a2
as in the picture.

58
 Example: Consider the unit disc D = {x + iy C|x2 + y 2 < 1} with first
fundamental form
4(dx2 + dy 2 )
(1 x2 y 2 )2
and the upper half plane H = {u + iv C|v > 0} with the first fundamental
form
du2 + dv 2
.
v2
We shall show that there is an isometry from H to D given by
wi
w 7 z =
w+i
where w = u + iv H and z = x + iy D.
We write |dz|2 = dx2 + dy 2 and |dw|2 = du2 + dv 2 . If w = f (z) where
f : D H is holomorphic then

f 0 (z) = ux + ivx = vy iuy

and so

|f 0 (z)|2 |dz|2 = (u2x +vx2 )(dx2 +dy 2 ) = (ux dx+uy dy)2 +(vx dx+vy dy)2 = du2 +dv 2 = |dw|2 .

Thus we can substitute  2

 dw 
2
|dw| =   |dz|2 (3)
dz
to calculate how the first fundamental form is transformed by such a map.
The Mobius transformation
wi
w 7 z = (4)
w+i
restricts to a smooth bijection from H to D because w H if and only if
|w i| < |w + i|, and its inverse is also a Mobius transformation and hence is
also smooth. Substituting (4) and (3) with

dw 1 (w i) 2i
= =
dz w + i (w + i)2 (w + i)2

into v 2 |dw|2 gives 4(1 |z|2 )2 |dz|2 , so this Mobius transformation gives us
an isometry from H to D as required.

59
4.4 The second fundamental form
The first fundamental form describes the intrinsic geometry of a surface the expe-
rience of an insect crawling around it. It is this that we can generalize to abstract
surfaces. The second fundamental form relates to the way the surface sits in R3 ,
though as we shall see, it is not independent of the first fundamental form.
First take a surface r(u, v) and push it inwards a distance t along its normal to get a
one-parameter family of surfaces:

R(u, v, t) = r(u, v) tn(u, v)

with
Ru = ru tnu , Rv = rv tnv .
We now have a first fundamental form Edu2 + 2F dudv + Gdv 2 depending on t and
we calculate
1
(Edu2 + 2F dudv + Gdv 2 )|t=0 = (ru nu du2 + (ru nv + rv nu )dudv + rv nv dv 2 ).
2 t
The right hand side is the second fundamental form. From this point of view it is
clearly the same type of object as the first fundamental form a quadratic form on
the tangent space.
In fact it is useful to give a slightly different expression. Since n is orthogonal to ru
and rv ,
0 = (ru n)u = ruu n + ru nu
and similarly
ruv n + ru nv = 0, rvu n + rv nu = 0
and since ruv = rvu we have ru nv = rv nu . We then define:

Definition 20 The second fundamental form of a surface is the expression

Ldu2 + 2M dudv + N dv 2

where L = ruu n, M = ruv n, N = rvv n.

Examples:
1) The plane
r(u, v) = a + ub + vc
has ruu = ruv = rvv = 0 so the second fundamental form vanishes.

60
2) The sphere of radius a: here with the origin at the centre, r = an so

ru nu = a1 ru ru , ru nv = a1 ru rv , rv nv = a1 rv rv

and
Ldu2 + 2M dudv + N dv 2 = a1 (Edu2 + 2F dudv + Gdv 2 ).

The plane is characterised by the vanishing of the second fundamental form:

Proposition 4.2 If the second fundamental form of a surface vanishes, it is part of

a plane.

Proof: If the second fundamental form vanishes,

ru nu = 0 = rv nu = ru nv = rv nv

so that
nu = nv = 0
since nu , nv are orthogonal to n and hence linear combinations of ru , rv . Thus n is
constant. This means

(r n)u = ru n = 0, (r n)v = rv n = 0

and so
r n = const
which is the equation of a plane. 2

Consider now a surface given as the graph of a function z = f (x, y):

r(x, y) = xi + yj + f (x, y)k.

Here
rx = i + fx k, r y = j + fy k
and so
rxx = fxx k, rxy = fxy k, ryy = fyy k.
At a critical point of f , fx = fy = 0 and so the normal is k. The second fundamental
form is then the Hessian of the function at this point:
   
L M fxx fxy
= .
M N fxy fyy

61
We can use this to qualitatively describe the behaviour of the second fundamental
form at different points on the surface. For any point P parametrize the surface by
its projection on the tangent plane and then f (x, y) is the height above the plane.
Now use the theory of critical points of functions of two variables.
2
If fxx fyy fxy > 0 then the critical point is a local maximum if the matrix is negative
definite and a local minimum if it is positive definite. For the surface the difference
is only in the choice of normal so the local picture of the surface is like the sphere
it lies on one side of the tangent plane at the point P .

2
If on the other hand fxx fyy fxy < 0 we have a saddle point and the surface lies on
both sides of the tangent plane:

62
In fact any closed surface X in R3 , not just rabbit-shaped ones, must have points of
the first type.

Proposition 4.3 Any closed surface X in R3 has points at which the second funda-
mental form is negative definite.

Proof: Since X is compact, it is bounded and so can be surrounded by a large

sphere centre the origin. Gradually deflate the sphere until at radius R it touches X
at a point. Let this be the direction of the z-axis and describe X locally as the graph
of a function f as above. Then X lies below the sphere so
p
f R 2 x2 y 2 0
with f (0) = R and fx (0) = fy (0) = 0. Hence
1 1 2
(fxx x2 + 2fxy xy + fyy y 2 ) + (x + y 2 ) 0
2 2R
so
1
Lx2 + 2M xy + N y 2 (x2 + y 2 ).
R
2

It is easy to understand qualitatively the behaviour of a surface from whether LN M 2

is positive or not. In fact there is a closely related function called the Gaussian
curvature which we shall study next.

4.5 The Gaussian curvature

Definition 21 The Gaussian curvature of a surface in R3 is the function
LN M 2
K=
EG F 2
Note that under a coordinate change
E0 F0
     
ux uy E F ux vx
=
vx vy F G uy vy F0 G0
so taking determinants
(ux vy uy vx )2 (EG F 2 ) = (E 0 G0 F 02 ).
Since the second fundamental form is a quadratic form on the tangent space just like
the first, it undergoes the same transformation, so the ratio (LN M 2 )/(EG F 2 )
is independent of the choice of coordinates.

63
Examples:
1. For a plane, L = M = N = 0 so K = 0
2. For a sphere of radius a, the second fundamental form is a1 times the first so that
K = a2 .

We defined K in terms of the second fundamental form which we said describes

the extrinsic geometry of the surface. In fact it only depends on E, F, G and its
derivatives, and so is intrinsic our insect crawling on the surface could in principle
work it out. It was Gauss who showed this in 1828, a result he was particularly
pleased with.

What it means is that if two surfaces are locally isometric, then the isometry maps
the Gaussian curvature of one to the Gaussian curvature of the other for example
the Gaussian curvature of a bent piece of paper is zero because it is isometric to the
plane. Also, we can define Gaussian curvature for an abstract Riemannian surface.
We prove Gausss egregious theorem, as he proudly called it, by a calculation. We
consider locally a smooth family of tangent vectors

a = f ru + grv

where f and g are functions of u, v. If we differentiate with respect to u or v this is

no longer necessarily tangential, but we can remove its normal component to make it
so, and call this the tangential derivative:

u a = au (n au )n
= au + (nu a)n

since a and n are orthogonal.

64
The important thing to note is that this tangential derivative only depends on E, F, G
and their derivatives, because we are taking a tangent vector like ru , differentiating
it to get ruu and ruv and then projecting back onto the tangent plane which involves
taking dot products like ruu ru = (ru ru )u /2 = Eu /2 etc.
Now differentiate u a tangentially with respect to v:

v u a = avu (n avu )n + v ((nu a)n).

But since we are taking the tangential component, we can forget about differentiating
the coefficient of n. Moreover, since n is a unit vector, nv is already tangential, so
we get:
v u a = avu (n avu )n + (nu a)nv
Interchanging the roles of u and v and using the symmetry of the second derivative
auv = avu we get

v u a u v a = (nu a)nv (nv a)nu = (nu nv )a.

Now
nu nv = n (5)
so we see that v u u v acting on a rotates it in the tangent plane by 90 and
multiplies by , where is intrinsic. Now from (5),

n ru rv = (nu nv ) (ru rv ) = (nu ru )(nv rv ) (nu rv )(nv ru ) = LN M 2

but also
n ru rv = EG F 2
which gives
= (LN M 2 )/ EG F 2 . (6)
It follows that LN M 2 and hence K depends only on the first fundamental form.

4.6 The Gauss-Bonnet theorem

One of the beautiful features of the Gaussian curvature is that it can be used to de-
termine the topology of a closed orientable surface more precisely we can determine
the Euler characteristic by integrating K over the surface. We shall do this by using
a triangulation and summing the integrals over the triangles, but the boundary terms
involve another intrinsic invariant of a curve in a surface:

65
Definition 22 The geodesic curvature g of a smooth curve in X is defined by

g = t0 (nt)

where t is the unit tangent vector of the curve, which is parametrized by arc length.

This is the tangential derivative of the unit tangent vector t and so is intrinsic.
The first version of Gauss-Bonnet is:

Theorem 4.4 Let be a smooth simple closed curve on a coordinate neighbourhood

of a surface X enclosing a region R, then
Z Z
g ds = 2 KdA
R

where g is the geodesic curvature of , ds is the element of arc-length of , K is the

Gaussian curvature of X and dA the element of area of X.

Proof: Recall Stokes theorem in R3 :

Z Z
a ds = curl a dS
C S

for a curve C spanning a surface S. In the xy plane with a = (P, Q, 0) this becomes
Greens formula Z Z
0 0
(P u + Qv )dt = (Qu Pv )dudv (7)
R

Now choose a unit length tangent vector field, for example e = ru / E. Then e, ne
is an orthonormal basis for each tangent space. Since e has unit length, u e is
tangential and orthogonal to e so there are functions P, Q such that

u e = P ne, v e = Q ne.

In Greens formula, take a = (P, Q, 0) then the left hand side of (7) is
Z Z
(u u e + v v e) (ne)ds = e0 (ne)ds
0 0
(8)

Let t be the unit tangent to , and write it relative to the orthonormal basis

t = cos e + sin ne.

66
So
t0 (ne) = cos e0 (ne) + cos 0 .
The geodesic curvature of is defined by g = t0 (nt) so

t0 = n + g nt = n + g (cos ne sin e)

and so
g = e0 (ne) + 0 .
We can therefore write (8) as Z
(g 0 )ds

and as changes by 2 on going round the curve, this is

Z
g ds 2.

To compute the right hand side of (7), note that

v u e = v (P ne) = Pv ne + P nv e = Pv ne + P Q n(ne)

since nv e is normal. Interchanging the roles of u and v and subtracting we obtain

(v u u v )e = (Pv Qu )ne

and from (6) this is equal to K EG F 2 .

Applying Greens theorem and using dA = EG F 2 dudv gives the result. 2

Note that the extrinsic normal was only used to define ne which is one of
the two unit tangent vectors to X orthogonal to e. If the surface is orientable
we can systematically make a choice and then the proof is intrinsic.

If the curve is piecewise smooth a curvilinear polygon then jumps by the

external angle i at each vertex, so the integral of 0 which is 2 in the theorem is
replaced by Z X X
0 ds = 2 i = i (n 2)
i i

where i are the internal angles. The Gauss-Bonnet theorem gives in particular:

67
Theorem 4.5 The sum of the angles of a curvilinear triangle is
Z Z
+ KdA + g ds.
R

Examples:
1. In the plane, a line has constant unit tangent vector and so g = 0. Since the
Gaussian curvature is zero too this says that the sum of the angles of a triangle is .
2. A great circle on the unit sphere also has g zero, for example if (s) = (cos s, sin s, 0),
then t = ( sin s, cos s, 0) and t0 = (cos s, sin s, 0) which is normal to the sphere.
Since here K = 1, we have, for the triangle with angles A, B, C

+ + = + Area(ABC).

Here is the most interesting version of Gauss-Bonnet:

Theorem 4.6 If X is a smooth orientable closed surface with a Riemannian metric,

then Z
KdA = 2(X)
X

Proof: Take a smooth triangulation so that each triangle is inside a coordinate

neighbourhood and apply Theorem 4.5 and add. The integrals of g on the edges
cancel because the orientation on the edge from adjacent triangles is opposite (this is
for Greens theorem we use the anticlockwise orientation on ). The theorem gives
the total sum of internal angles as
Z
F + KdA.
X

But around each vertex the internal angles add to 2 so we have

Z
2V = F + KdA
X

and as our faces are triangles whose sides meet in pairs there are 3F/2 edges. Hence
Z Z
2(X) = 2(V E + F ) = F + KdA 3F + 2F = KdA.
X X

68
The Gauss-Bonnet theorem and its method of proof give another formula for the
Euler characteristic, involving smooth real-valued functions f : X R on a closed
surface X. Since X is compact, f certainly has a maximum and a minimum, but may
have other critical points too. Think of a surface in R3 and the function f given by
its height above a plane:

This has 2 maxima, 2 minima and 6 saddle points. We shall be able to calculate the
Euler characteristic from these numbers.
First recall that a smooth function f (u, v) has a critical point at a if
fu (a) = fv (a) = 0.
Because of the chain rule, this condition is independent of coordinates: if u =
u(x, y), v = v(x, y) then
fx = fu ux + fv vx , fy = fu uy + fv vy
so fu and fv vanish if and only if fx and fy vanish. This means we can unambiguously
talk about the critical points of a smooth function on a surface X.
The Hessian matrix  
fuu fuv
fuv fvv
at a critical point transforms like
     
ux uy fxx fxy ux vx fuu fuv
=
vx vy fxy fyy uy vy fuv fvv
and so
2
(fuu fvv fuv ) = (ux vy uy vx )2 (fuu fvv fuv
2
)
therefore to say that the determinant of the Hessian is non-zero, or positive or nega-
tive, is again independent of the choice of coordinate.

69
Definition 23 A function f on a surface X has a nondegenerate critical point at
a X if its Hessian at a is invertible.

2
We know from calculus that if fuu fvv fuv > 0 and fuu > 0 we have a local min-
2
imimum, if fuu < 0 a local maximum and if fuu fvv fuv < 0 a saddle point. The
theorem is the following:

Theorem 4.7 Let f be a smooth function on a closed surface X with nondegenerate

critical points, then the Euler characteristic (X) is the number of local maxima and
minima minus the number of saddle points.

In the picture, we have (X) = 4 6 = 2 which is correct for the connected sum of
two tori. If we turn it on its side we get one maximum, one minimum and 4 saddle
points again giving the same value: 2 4 = 2.

Proof: Given a function f on X we can define its gradient vector field:

1
a= [(Gfu F fv )ru + (Efv F fu )rv ]
EG F 2
which is normal to the contour lines of f . Away from the critical points we can
normalize it to get a unit vector field e. Surround each critical point by a small

70
closed curve i enclosing a disc Ri . Let Y be the complement of the discs, then from
the argument of Theorem 4.4
Z XZ
KdA = e0 (ne)ds
Y i i

using the negative sign because Y is outside Ri .

Inside Ri we choose a unit vector field f and then we get
Z Z
KdA = f 0 (nf )ds
Ri i

so adding gives Z XZ
KdA = [f 0 (nf ) e0 (ne)]ds.
X i i

From the proof of the theorem we had

g = e0 (ne) + 0 = f 0 (nf ) + 0

where is the angle between 0 and e and between 0 and f . So the contribution is
just the change in angle between the vector field e and a fixed one f which extends.
This is an integer multiple of 2 so we can evaluate it by deforming to the standard
Euclidean case. A local minimum is f = x2 + y 2 which gives

e = (cos , sin )

and contributes +1, as does the local minimum (cos , sin ). For a saddle point
f = x2 y 2 which gives

e = (cos , sin ) = (cos(), sin())

and contributes 1. 2

4.7 Geodesics
Geodesics on a surface are curves which are the analogues of straight lines in the
plane. Lines can be thought of in two ways:

shortest curves

straightest curves

71
The first point of view says that a straight line minimizes the distance between any
two of its points. Conceptually this leads to the idea of stretching a string between two
points on a surface until it tightens, and this certainly is one approach to geodesics.
The second approach is however generally easier. A line is straightest because its
tangent vector doesnt change it is constant along the line. We generalize this to
a curve on a surface by insisting that the component of t0 tangential to the surface
should vanish. Or....

Definition 24 A geodesic on a surface X is a curve (s) on X such that t0 is normal

to the surface.

From Definition 22 this is the same as saying that the geodesic curvature vanishes.
The general problem of finding geodesics on a surface is very complicated. The case
of the ellipsoid p
is a famous example, needing hyperelliptic functions to solve it
integrals of dz/ p(z) where p(z) is a polynomial of degree 6. But there are cheap
ways to find some of them, as in these examples:

Examples:
1) The normal to a curve in the plane is parallel to the plane, so the condition that
t0 is normal to the plane means t0 = 0 which integrates to r = sa + c, the equation
of a straight line. Geodesics in the plane really are straight lines, then.
2) Take the unit sphere and a plane section through the origin. We saw earlier that
g = 0 here.
3) Similarly, any plane of symmetry intersects a surface in a geodesic, because the
normal to the surface at such a point must be invariant under reflection in the plane
of symmetry and hence lie in that plane. It is orthogonal to the tangent vector of the
curve of intersection and so t0 points normally.
A useful class of examples is provided by a surface of revolution

r(u, v) = f (u)(cos v i + sin v j) + uk

The reflection (x, y, z) 7 (x, y, z) maps the surface to itself, as, by symmetry, does
any reflection in a plane containing the z-axis. So the meridians v = const. are
geodesics:

72
To find the geodesics in general we need to solve a nonlinear system of ordinary
differential equations:

Proposition 4.8 A curve (s) = (u(s), v(s)) on a surface parametrized by arc length
is a geodesic if and only if
d 1
(Eu0 + F v 0 ) = (Eu u02 + 2Fu u0 v 0 + Gu v 02 )
ds 2
d 1
(F u0 + Gv 0 ) = (Ev u02 + 2Fv u0 v 0 + Gv v 02 )
ds 2
Proof: We have for the curve
t = ru u0 + rv v 0
and it is a geodesic if and only if t0 is normal i.e.
t0 ru = t0 rv = 0.
Now
t0 ru = (t ru )0 t r0u
so the first equation is
(t ru )0 = t r0u .
The left hand side is
d d
((ru u0 + rv v 0 ) ru ) = (Eu0 + F v 0 )
ds ds
an the right hand side is
t (ruu u0 + ruv v 0 ) = ru ruu u02 + (rv ruu + ru ruv )u0 v 0 + rv ruv v 02
1 1
= Eu u02 + (rv ru )u u0 v 0 + Gu v 02
2 2
1
= (Eu u02 + 2Fu u0 v 0 + Gu v 02 )
2
The other equation follows similarly. 2

73
 It is clear from 4.8 that geodesics only depend on the first fundamental
form, so that geodesics can be defined for abstract surfaces and moreover an
isometry takes geodesics to geodesics.

Examples:
1) The plane: E = 1, F = 0, G = 1 in Cartesian coordinates, so the geodesic equations
are
x00 = 0 = y 00
which gives straight lines

x = 1 s + 1 , y = 2 s + 2 .

2) The cylinder
r(u, v) = a(cos v i + sin v j) + uk
has first fundamental form

du2 + a2 dv 2 = du2 + d(av)2 .

This is isometric to the plane so the geodesics are of the form

u = 1 s + 1 , v = 2 s + 2

which gives a helix

= a(cos(2 s + 2 ) i + sin(2 s + 2 ) j) + (1 s + 1 )k

The differential equation for geodesics gives us the following general fact:

74
Proposition 4.9 Through each point P on a surface and in each direction at P there
passes a unique geodesic.

Proof: We are solving a differential equation of the form

u00 = a(u, v, u0 , v 0 ), v 00 = b(u, v, u0 , v 0 )

or equivalently a first order system

u0 = p
v0 = q
p0 = a(u, v, p, q)
q0 = b(u, v, p, q)

and the Cauchy existence theorem (see Appendix B) gives a unique solution with
initial conditions (u, v, p, q), namely the point of origin and the direction. 2

Example: Given a point a on the unit sphere and a tangential direction b the span
of a, b is a plane through the origin which meets the sphere in a great circle through
a with tangent b. Thus every geodesic is a great circle.

There is one case a surface of revolution where the geodesic equations can be
solved, or anyway, reduced to a single integration. We have

E = 1 + f 0 (u)2 , F = 0, G = f (u)2

and the equations become

d
((1 + f 02 )u0 ) = f 0 (f 00 u02 + f v 02 )
ds
d 2 0
(f v ) = 0
ds
We ignore the first equation it is equivalent to a more obvious fact below. The
second says that
f 2v0 = c (9)
where c is a constant. Now use the fact that the curve is parametrized by arc length
(this is an integral of the equations), and we get

(1 + f 02 )u02 + f 2 v 02 = 1 (10)

75
Substitute for v 0 from (9) in (10) to get

c2
(1 + f 02 )u02 + =1
f2
and then s
1 + f 02
Z
s= f du
f 2 c2
which is only an integration. Having solved this by u = h(s), v can be determined
by a further integration from (9):

c
Z
v(s) = dt.
f (h(t))2

If we are only interested in the curve and not its arclength parametrization, then (9)
and (10) give
 2
0 2 du f (u)4
(1 + f (u) ) + f (u)2 =
dv c2
which reduces to the single integration
s
c 1 + f 0 (u)2
Z
v= du.
f (u) f (u)2 c2

4.8 Gaussian curvature revisited

We may not be able to solve the geodesic equations explicitly, but existence of
geodesics through a given point and in a given direction give rise to various nat-
ural coordinate systems, modelled on Cartesian coordinates. Here is one: choose a
geodesic parametrized by arc length. Through the point (v) take the geodesic
v (s) which intersects orthogonally, and define

r(u, v) = v (u).

76


Since ru and rv are orthogonal at u = 0 they are linearly independent in a neighbour-

hood and so are good coordinates.
Now the curves v = const. are parametrized by arc length, so E = 1. These curves
are also geodesics and u is arc length so in the second geodesic equation
d 1
(F u0 + Gv 0 ) = (Ev u02 + 2Fv u0 v 0 + Gv v 02 )
ds 2
we put v = const. and u = s which, with E = 1, gives Fu = 0. But F vanishes
at u = 0 because the two geodesics are orthogonal there, hence F = 0 and the first
fundamental form is
du2 + G(u, v)dv 2 .

In this form the Gaussian curvature is simple:

Proposition 4.10 The Gaussian curvature of the metric du2 + G(u, v)dv 2 is

K = G1/2 (G1/2 )uu

Examples:
1. For the plane dx2 + dy 2 , G = 1 and K = 0.
2. For the unit sphere with first fundamental form du2 + sin2 udv 2 , G = sin2 u so
1 1
K= (sin u)uu = sin u = 1.
sin u sin u

77
3. For the upper half-space with metric (dx2 + dy 2 )/y 2 put u = log y and v = x and
then we have du2 + e2u dv 2 , so that

K = eu (eu )uu = eu eu = 1.

Proof: Recall the tangential derivative : the tangential component of the ordinary
derivative. Then since by construction ru is the unit tangent vector of a geodesic, by
the definition of a geodesic its u-derivative is normal so u ru = 0.
Consider now v ru = Aru + Brv . The dot product with ru gives

Ev /2 = rvu ru = A

but E = 1 so A = 0.
Using E = 1 and F = 0 the product with rv gives

Gu /2 = rv rvu = BG.

Now from (6)

(v u u v )ru = K EG F 2 nru = KG1/2 (rv G1/2 ) = Krv

But the left hand side (using u rv = v ru which follows from ruv = rvu ) is

u (Gu /2G)rv = ((Gu /2G)u + (Gu /2G)2 )rv

which gives the result. 2

With this coordinate system we can characterize surfaces with constant Gaussian
curvature:

Theorem 4.11 A surface with K = 0 is locally isometric to the plane, with K = 1

locally isometric to the unit sphere and with K = 1 locally isometric to the upper
half space with metric (dx2 + dy 2 )/y 2 .

Proof: Use the form du2 + Gdv 2 .

i) If K = 0 then (G1/2 )uu = 0 so G1/2 = A(v)u + B(v). But at u = 0, ru and rv are
unit so B(v) = 1. Also, the curve u = 0 is a geodesic the initial curve with v
arc length. So the geodesic equation
d 1
(Eu0 + F v 0 ) = (Eu u02 + 2Fu u0 v 0 + Gu v 02 )
ds 2
78
gives 0 = Gu (0, v)/2 and this means in our case A(v) = 0. The first fundamental
form is therefore du2 + dv 2 and by 4.1 this is isometric to the plane.
ii) If K = 1, the equation for G1/2 is

(G1/2 )uu + G1/2 = 0

which is solved by G1/2 = A(v) sin u + B(v) cos u. The boundary conditions give
G = cos2 u and the metric du2 + cos2 u dv 2 the sphere.
iii) If K = 1 we have du2 + cosh2 u dv 2 . The substitution x = ev tanh u, y = ev sech u
takes this into (dx2 + dy 2 )/y 2 . 2

79
5 The hyperbolic plane

5.1 Isometries
We just saw that a metric of constant negative curvature is modelled on the upper
half space H with metric
dx2 + dy 2
y2
which is called the hyperbolic plane. This is an abstract surface in the sense that we
are not considering a first fundamental form coming from an embedding in R3 , and
yet it is concrete enough to be able to write down and see everything explicitly. First
we consider the isometries from H to itself.
If a, b, c, d R and ad bc > 0 then the Mobius transformation
az + b
z 7 w = (11)
cz + d
restricts to a smooth bijection from H to H with smooth inverse
dw b
w 7 z = .
cw + a
If we substitute
 
az + b a c(az + b) (ad bc)
w= and dw = dz = dz
cz + d cz + d (cz + d)2 (cz + d)2
into
du2 + dv 2 4|dw|2
=
v2 |w w|2
we get
4(ad bc)2 |dz|2 4(ad bc)2 |dz|2 4|dz|2 dx2 + dy 2
= = = .
|(az + b)(cz + d) (az + b)(cz + d)|2 |(ad bc)(z z)|2 |z z|2 y2
Thus this Mobius transformation is an isometry from H to H. So is the transformation
z 7 z, and hence the composition
b az
z 7 (12)
d cz
is also an isometry from H to H. In fact (11) and (12) give all the isometries of H,
as we shall see later.

80
In 4.3 we saw that the unit disc D with the metric
du2 + dv 2
(1 u2 v 2 )2
is isometric to H, so any statements about H transfer also to D. Sometimes the
picture is easier in one model or the other. The isometries f : D D of the unit
disc model of the hyperbolic plane are also Mobius transformations, if they preserve
orientations, or compositions of Mobius transformations with z 7 z if they reverse
orientations. The Mobius transformations which map D to itself are those of the form
 
i za
z 7 w = e
1 az
where a D and R. They are isometries because substituting for w and
(1 |a|2 )
dw = ei dz
(1 az)2
in 4(1 |w|2 )2 |dw|2 gives 4(1 |z|2 )2 |dz|2 .

Notice that the group Isom(H) of isometries of H acts transitively on H because if

a + ib H then b > 0 so the transformation
z 7 bz + a
is an isometry of H which takes i to a + ib. Similarly the group Isom(D) of isometries
of D acts transitively on D since if a D then the isometry
za
z 7
1 az
maps a to 0. Notice also that the subgroup of Isom(D) consisting of those isometries
which fix 0 contains all the rotations
z 7 ei z
about 0 as well as z 7 z.

5.2 Geodesics
The hyperbolic plane is a case where the geodesic equations can be easily solved:
since E = G = 1/y 2 and F = 0, and these are independent of x, the first geodesic
equation
d 1
(Eu0 + F v 0 ) = (Eu u02 + 2Fu u0 v 0 + Gu v 02 )
ds 2
81
becomes
d x0
( )=0
ds y 2
and so
x0 = cy 2 . (13)
We also know that parametrization is by arc length in these equations so
x02 + y 02
=1 (14)
y2
If c = 0 we get x = const., which is a vertical line. Suppose c 6= 0, then from (13)
and (14) we have s
dy y 2 c2 y 4
=
dx c2 y 4
or
cydy
p = dx
1 c2 y 2
which integrates directly to
p
c1 1 c2 y 2 = x a
or
(x a)2 + y 2 = 1/c2
which is a semicircle centred on the real axis.

The isometry from H to D given by

wi
w 7 z =
w+i

82
takes geodesics to geodesics (since it is an isometry) and it is the restriction to H of
a Mobius transformation C {} C {} which takes circles and lines to circles
and lines, preserves angles and maps the real axis to the unit circle in C. It therefore
follows that the geodesics in D are the circles and lines in D which meet the unit
circle at right angles.

Using geodesics we can now show that any isometry is a Mobius transformation as
above. So suppose that F : D D is an isometry. Take a Mobius isometry G taking
F (0) to 0, then we need to prove that GF is Mobius. This is an isometry fixing 0, so it
takes geodesics through 0 to geodesics through 0. It preserves angles, so acts on those
geodesics by a rotation or reflection. It also preserves distance so it takes a point on a
geodesic a distance r from the origin to another point at the same distance. However,
as we noted above, each rotation R : z 7 ei z is a Mobius isometry, so composing
with this we see that RGF = 1 and F = (RG)1 is a Mobius isometry.

5.3 Angles and distances

Hyperbolic angles in H and in D are the same as Euclidean angles, since their first
fundamental forms satisfy E = G and F = 0. Distances between points are given
by the lengths of geodesics joining the points. Since the interval (1, 1) is a geodesic
in the unit disc D, the distance from 0 to any x (0, 1) is given by the hyperbolic
length of the line segment [0, x], which is
Z xp Z x
dt
0 2 0 0
Eu + 2F u v + Gv dt =0 2
2
= 2 tanh1 x
0 0 1t

83
where u(t) = t and v(t) = 0 and E = G = (1 u2 v 2 )2 and F = 0. Given any
a, b D we can choose R such that
 
b a  b a 
ei =  
1 ab 1 ab 
is real and positive, so its distance from 0 is
 
1  b a 
 
2 tanh  .
1 ab 
Since the isometry
za
z 7 ei
1 az
preserves distances and takes a to 0 and b to ei (b a)/(1 ab), it follows that the
hyperbolic distance from a to b in D is
 
1  b a 
 
dD (a, b) = 2 tanh  .
1 ab 
We can work out hyperbolic distances in H in a similar way by first calculating the
distance from i to i for [1, ) as the length of the geodesic from i to i given
by the imaginary axis, which is
Z
dt
= log ,
1 t
and then given a, b H finding an isometry of H which takes a to i and b to i for
some [1, ). Alternatively, since we have an isometry from H to D given by
wi
w 7 z = ,
w+i
the hyperbolic distance between points a, b H is equal to the hyperbolic distance
between the corresponding points (a i)/(a + i) and (b i)/(b + i) in D, which is
   
ai bi (b i)(a + i) (a i)(b + i) b a
) = 2tanh1   = 2tanh1 
   
dH (a, b) = dD ( ,  b a  .

a+i b+i (a + i)(b + i) (a i)(b i) 

5.4 Hyperbolic triangles

A hyperbolic triangle is given by three distinct points in H or D joined by geodesics.
We see immediately from Gauss-Bonnet that the sum of the angles of a triangle is
given by
A + B + C = Area().

84
We can also consider hyperbolic triangles which have one or more vertices at infinity,
i.e. on the boundary of H or D. These triangles are called asymptotic, doubly (or
bi-) asymptotic and triply (or tri-) asymptotic, according to the number of vertices at
infinity. The angle at a vertex at infinity is always 0, since all geodesics in H or D
meet the boundary at right angles.

Theorem 5.1 (The cosine rule for hyperbolic triangles) If is a hyperbolic triangle
in D with vertices at a, b, c and
= dD (b, c), = dD (a, c) and = dD (a, b)
then
cosh = cosh cosh sinh sinh cos
where is the internal angle of at c.

Proof: Because the group of isometries of D acts transitively on D we can assume

that c = 0. Moreover, since the rotations z 7 ei z are isometries which fix 0, we can
also assume that a is real and positive. Then = 2tanh1 (a) so
a = tanh(/2)
and similarly
b = ei tanh(/2)
while  
 ba 
tanh(/2) =  .
1 ab 

85
Recall that
1 + tanh2 (/2)
= cosh()
1 tanh2 (/2)
so
|1 ab|2 + |b a|2 (1 + |a|2 )(1 + |b|2 ) 2(ab + ab)
cosh() = = .
|1 ab|2 |b a|2 (1 |a|2 )(1 |b|2 )
Now
1 + |a|2 1 + tanh2 (/2)
= = cosh
1 |a|2 1 tanh2 (/2)
as above, and similarly
1 + |b|2
= cosh
1 |b|2
while
2(ab + ab) 2tanh(/2)tanh(/2)(ei + ei )
= = sinh sinh cos.
(1 |a|2 )(1 |b|2 ) sech2 (/2)sech2 (/2)

This completes the proof. 2

Theorem 5.2 (The sine rule for hyperbolic triangles) Let be a hyperbolic triangle
in D with internal angles A, B, C at vertices a, b, c and

= dD (b, c), = dD (a, c) and = dD (a, b).

Then
sinA sinB sinC
= = .
sinh sinh sinh

Proof: Two alternatives approaches:

1) Use the cosine rule to find an expression for sinh2 sinh2 sin2 C in terms of cosh,
cosh and cosh which is symmetric in , and , and deduce that

sinh2 sinh2 sin2 C = sinh2 sinh2 sin2 B = sinh2 sinh2 sin2 A.

2) First prove that if C = /2 then sinA sinh = sinh by applying the cosine
rule to in two different ways. Then deduce the result in general by dropping a
perpendicular from one vertex of to the opposite side. 2

86
Gauss-Bonnet and its limits give the following:

Theorem 5.3 (Areas of hyperbolic triangles)

(i) The area of a triply asymptotic hyperbolic triangle is .
(ii) The area of a doubly asymptotic hyperbolic triangle with internal angle is
.
(iii) The area of an asymptotic hyperbolic triangle with internal angles and is
.
(iv) The area of a hyperbolic triangle with internal angles , and is .

5.5 Non-Euclidean geometry

As we see above, the analogy between Euclidean geometry and its theorems and the
geometry of the hyperbolic plane is very close, so long as we replace lines by geodesics,
and Euclidean isometries (translations, rotations and reflections) by the isometries of
H or D. In fact it played an important historical role.
For centuries, Euclids deduction of geometrical theorems from self-evident common
notions and postulates was thought not only to represent a model of the physical space
in which we live, but also some absolute logical structure. One postulate caused some
problems though was it really self-evident? Did it follow from the other axioms?
This is how Euclid phrased it:

That if a straight line falling on two straight lines makes the interior angle on the
same side less than two right angles, the two straight lines if produced indefinitely,
meet on that side on which the angles are less than two right angles.

Some early commentators of Euclids Elements, like Posidonius (1st Century BC),
Geminus (1st Century BC), Ptolemy (2nd Century AD), Proclus (410 - 485) all felt
that the parallel postulate was not sufficiently evident to accept without proof.
Here is a page from a medieval edition of Euclid dating from the year 888. It is
handwritten in Greek. The manuscript, contained in the Bodleian Library, is one of
the earliest surviving editions of Euclid.

87
The controversy went on and on with Greek and Islamic mathematicians puzzling
over it. In 1621 Sir Henry Savile, Warden of Merton College, called attention to two
blemishes in Euclidean geometry: the theory of parallels and the theory of proportion
(nevertheless you can see Euclid, wearing an unsuitable hat, standing next to him on
the memorial in Merton College Chapel). Johann Lambert (1728-1777) realized that
if the parallel postulate did not hold then the angles of a triangle add up to less than
180 , and that the deficit was the area. He found this worrying in many ways, not
least because it says that there is an absolute scale no distinction between similar
and congruent triangles. Finally Janos Bolyai (1802-1860) and Nikolai Lobachevsky
(1793-1856) discovered non-Euclidean geometry simultaneously. It satisfies all of
Euclids axioms except the parallel postulate, and we shall see that it is the geometry
of H or D that we have been studying.
Bolyai became interested in the theory of parallel lines under the influence of his
father Farkas, who devoted considerable energy towards finding a proof of the parallel
postulate without success. He even wrote to his son:

I entreat you, leave the doctrine of parallel lines alone; you should fear it like a
sensual passion; it will deprive you of health, leisure and peace it will destroy all joy
in your life.

Another relevant figure in the discovery was Carl Friedrich Gauss (1777-1855), who
as we have seen developed the differential geometry of surfaces. He was the first
to consider the possibility of a geometry denying the parallel postulate. However,
for fear of being ridiculed he kept his work unpublished, or maybe he never made
the connection with the curvature of real world surfaces and the Platonic ideal of
axiomatic geometry. Anyway, when he read Janos Bolyais work he wrote to Janoss
father:

88
If I commenced by saying that I must not praise this work you would certainly be
surprised for a moment. But I cannot say otherwise. To praise it, would be to praise
myself. Indeed the whole contents of the work, the path taken by your son, the results
to which he is led, coincide almost entirely with my meditations, which have occupied
my mind partly for the last thirty or thirty-five years.

Euclids axioms were made rigorous by Hilbert. They begin with undefined concepts
of

point

line

lie on ( a point lies on a line)

betweenness

congruence of pairs of points

congruence of pairs of angles.

Euclidean geometry is then determined by logical deduction from the following ax-
ioms:

EUCLIDS AXIOMS
I. AXIOMS OF INCIDENCE
1. Two points have one and only one straight line in common.
2. Every straight line contains a least two points.
3. There are at least three points not lying on the same straight line.
II. AXIOMS OF ORDER
1. Of any three points on a straight line, one and only one lies between the other two.
2. If A and B are two points there is at least one point C such that B lies between
A and C.
3. Any straight line intersecting a side of a triangle either passes through the opposite
vertex or intersects a second side.
III. AXIOMS OF CONGRUENCE

89
1. On a straight line a given segment can be laid off on either side of a given point
(the segment thus constructed is congruent to the give segment).
2. If two segments are congruent to a third segment, then they are congruent to each
other.
3. If AB and A0 B 0 are two congruent segments and if the points C and C 0 lying on
AB and A0 B 0 respectively are such that one of the segments into which AB is divided
by C is congruent to one of the segments into which A0 B 0 is divided by C 0 , then the
other segment of AB is also congruent to the other segment of A0 B 0 .
4. A given angle can be laid off in one and only one way on either side of a given
half-line; (the angle thus drawn is congruent to the given angle).
5. If two sides of a triangle are equal respectively to two sides of another triangle,
and if the included angles are equal, the triangles are congruent.
IV. AXIOM OF PARALLELS
Through any point not lying on a straight line there passes one and only one straight
line that does not intersect the given line.
V. AXIOM OF CONTINUITY
1. If AB and CD are any two segments, then there exists on the line AB a number
of points A1 , . . . , An such that the segments AA1 , A1 A2 , . . . , An1 An are congruent to
CD and such that B lies between A and An

Clearly H does not satisfy the Axiom of Parallels:

The fact that hyperbolic geometry satisfies all the axioms except the parallel postulate
is now only of historic significance and the reader is invited to do all the checking.

90
Often one model is easier than another. Congruence should be defined through the
action of the group of isometries.

5.6 Complex analysis and the hyperbolic plane

The intricate metric structure of the hyperbolic plane geodesics, triangles and all
is actually determined purely by the holomorphic functions on it, so we could also
think of hyperbolic geometry as a branch of complex analysis. Here is the theorem
that makes it work:

Theorem 5.4 Any holomorphic homeomorphism f : D D is an isometry of the

hyperbolic metric.

Proof: The argument follows Schwartzs lemma. By applying an isometry we can

assume that f (0) = 0, and since the image of f is D, we have |f (z)| < 1 if z D. Now
since f (0) = 0, f1 (z) = f (z)/z is holomorphic and applying the maximum principle
to a disc of radius r < 1 we get
1
|f1 (z)|
r
and in the limit as r 1, |f1 (z)| 1 or equivalently

|f (z)| |z|.

Since f is a homeomorphism, its inverse satisfies the same inequality so

|z| |f (z)|

and |f1 (z)| = 1 everywhere. Since this is true at an interior point the function must
be a constant c so f (z) = cz and since |f (z)| = |z|,

f (z) = ei z

which is an isometry. 2

In fact a similar result holds for C:

Theorem 5.5 Any holomorphic homeomorphism f : C C is of the form f (z) =

az + b with a 6= 0.

91
If |a| = 1 this is an isometry of the Euclidean metric dx2 + dy 2 . The extra scaling
z 7 z is what gives rise in classical geometrical terms to similar but non-congruent
triangles.

Proof: For |z| > R consider the function g(z) = f (1/z). Suppose g has an essential
singularity at z = 0. Then the Casorati-Weierstrass theorem (Exercise 17.5 in [3])
tells us that g(z) gets arbitrarily close to any complex number if z is small enough,
and in particular to values in the image of {z : |z| R} under f . But we assumed f
was bijective, which is a contradiction. It follows that g has at most a pole at infinity
and so f (z) must be a polynomial of some degree k.
However the equation f (z) = c then has k solutions for most values of c, and again
since f is bijective we must have k = 1 and
f (z) = az + b.
2

For completeness, we add the following

Theorem 5.6 Any holomorphic homeomorphism f of the Riemann sphere to itself

is a Mobius transformation z 7 (az + b)/(cz + d).

Proof: By using a Mobius transformation we can assume that f () = and then

the previous theorem tells us that f (z) = az + b. 2

These results are all about the complex plane and its subsets. In fact hyperbolic
geometry has an important role to play in the study of compact Riemann surfaces.
Recall that local holomorphic coordinates on a Riemann surface are related by holo-
morphic transformations and these preserve angles. Given two smooth curves on a
Riemann surface, it makes good sense to define their angle of intersection and this is
called a conformal structure. A metric also defines angles so we can consider metrics
compatible with the conformal structure of a Riemann surface. In a local coordinate
z such a metric is of the form
f dzdz = f (dx2 + dy 2 ).
The remarkable result is the following uniformization theorem:

Theorem 5.7 Every closed Riemann surface X has a metric of constant Gaussian
curvature compatible with its conformal structure.

92
Note that by the Gauss-Bonnet theorem K > 0 implies (X) > 0, i.e. X is a sphere,
K = 0 implies (X) = 0, i.e. X is a torus, and K < 0 gives (X) < 0.

Proof: The proof is a corollary of a difficult theorem called the Riemann mapping
theorem. Recall that a space is simply-connected if it is connected and every closed
path can be shrunk to a point. The Riemann mapping theorem (proved by Poincare
and Koebe) says that every simply-connected Riemann surface is holomorphically
homeomorphic to either the Riemann sphere, C or H.
If X is any reasonable topological space, one can form its universal covering space X
(see [2]) which is simply connected and has

a projection p : X X
every point x X has a neighbourhood V such that p1 (V ) consists of a disjoint
union of open sets each of which is homeomorphic to V by p
there is a group of homeomorphisms of X such that p(gy) = p(y), so that
permutes the different sheets in p1 (V ).
no element of apart from the identity has a fixed point
X can be identified with the space of orbits of acting on X.

The standard example of this is X = S 1 , X = R, p(t) = eit and = Z acting by

t 7 t + 2n. It is easy to see that the universal covering of a Riemann surface is a
Riemann surface, so applying the Riemann mapping theorem we see that X is either
the Riemann sphere, C or H.

So consider the cases:

If X is the sphere S, it is compact and so p : X 7 X has only a finite number k

of sheets. By counting vertices, edges and faces it is clear that (X) = k(X).
Since (S) = 2, we must have k = 1 or 2, but if the latter (X) = 1 which is
not of the allowable form 2 2g for an orientable surface and a Riemann surface
is orientable. So it is only the Riemann sphere in this case.
If X = C, we appeal to Theorem 5.5. The group of covering transformations
is holomorphic and so each element is of the form z 7 az + b. But has
no fixed points, so az + b = z has no solution which means that a = 1. the
transformations z 7 z + b are just translations and are isometries of the metric
dx2 + dy 2 which has K = 0.

93
 If X = H, then from Theorem 5.4, the action of preserves the hyperbolic
metric.

So we see that these abstract metrics have a role to play in the study of Riemann
surfaces a long long way from surfaces in R3 .

94
6 APPENDIX: Technical results

6.1 A: The inverse function theorem

Lemma 6.1 (Contraction mapping principle) Let M be a complete metric space and
suppose T : M M is a map such that

d(T x, T y) kd(x, y)

where k < 1. Then T has a unique fixed point.

Proof: Choose any point x0 , then

d(T m x0 , T n x0 ) k m d(x0 , T nm x0 ) for n m

k (d(x0 , T x0 ) + d(T x0 , T 2 x0 ) + . . . + d(T nm1 x0 , T nm x0 ))
m

k m (1 + k + . . . + k nm1 )d(x0 , T x0 )
km
d(x0 , T x0 )
1k
This is a Cauchy sequence, so completeness of M implies that it converges to x. Thus
x = lim T n x0 and so by continuity of T ,

T x = lim T n+1 x0 = x

For uniqueness, if T x = x and T y = y, then

d(x, y) = d(T x, T y) kd(x, y)

and so k < 1 implies d(x, y) = 0. 2

Theorem 6.2 (Inverse function theorem) Let U Rn be an open set and f : U

Rn a C function such that Dfa is invertible at a U . Then there exist neighbour-
hoods V, W of a and f (a) respectively such that f (V ) = W and f has a C inverse
on W .

Proof: By an affine transformation x 7 Ax + b we can assume that a = 0 and

Dfa = I. Now consider g(x) = x f (x). By construction Dg0 = 0 so by continuity
there exists r > 0 such that if kxk < 2r,
1
kDgx k <
2
95
It follows from the mean value theorem that
1
kg(x)k kxk
2
and so g maps the closed ball B(0, r) to B(0, r/2). Now consider
gy (x) = y + x f (x)
(The choice of gy is made so that a fixed point gy (x) = x solves f (x) = y).

If now kyk r/2 and kxk r, then

1 1 1
kgy (x)k r + kg(x)k r + r = r
2 2 2
so gy maps the complete metric space M = B(0, r) to itself. Moreover
1
kgy (x1 ) gy (x2 )k = kg(x1 ) g(x2 )k kx1 x2 k
2
if x1 , x2 B(0, r), and so gy is a contraction mapping. Applying Lemma 1 we have a
unique fixed point and hence an inverse = f 1 .

We need to show first that is continuous and secondly that it has derivatives of all
orders. From the definition of g and the mean value theorem,
kx1 x2 k kf (x1 ) f (x2 )k + kg(x1 ) g(x2 )k
1
kf (x1 ) f (x2 )k + kx1 x2 k
2
so
kx1 x2 k 2kf (x1 ) f (x2 )k
which is continuity for . It follows also from this inequality that if y1 = f (x1 ) and
y2 = f (x2 ) where y1 , y2 B(0, r/2) then x1 , x2 B(0, r), and so
k(y1 ) (y2 ) (Dfx2 )1 (y1 y2 )k = kx1 x2 (Dfx2 )1 (f (x1 ) f (x2 ))k
k(Dfx2 )1 kkDfx2 (x1 x2 ) f (x1 ) + f (x2 )k
Akx1 x2 kR
where A is a bound on k(Dfx2 )1 k and the function kx1 x2 kR is the remainder
term in the definition of differentiability of f . But kx1 x2 k 2ky1 y2 k so as
y1 y2 , x1 x2 and hence R 0, so is differentiable and moreover its derivative
is (Df )1 .

Now we know the derivative of :

D = (Df )1
so we see that it is continuous and has as many derivatives as f itself, so is C . 2

96
6.2 B: Existence of solutions of ordinary differential equa-
tions
Lemma 6.3 Let M be a complete metric space and T : M M a map. If T n is a
contraction mapping, then T has a unique fixed point.

Proof: By the contraction mapping principle, T n has a unique fixed point x. We

also have
T n (T x) = T n+1 x = T (T n x) = T x
so T x is also a fixed point of T n . By uniqueness T x = x. 2

Theorem 6.4 Let f (t, x) be a continuous function on |t t0 | a, kx x0 k b and

suppose f satisfies a Lipschitz condition

kf (t, x1 ) f (t, x2 )k kx1 x2 k.

If M = sup |f (t, x)| and h = min(a, b/M ), then the differential equation

dx
= f (t, x), x(t0 ) = x0
dt
has a unique solution for |t t0 | h.

Proof: Let Z t
(T x)(t) = x0 + f (s, x(s))ds
t0

Then T x is differentiable since f and x are continuous and if T x = x, x satisfies the

differential equation (differentiate the definition). We use the metric space

X = {x C([t0 h, t0 + h], Rn ) : kx(t) x0 k M h}

with the uniform metric

d(x1 , x2 ) = sup kx1 (t) x2 (t)k

|tt0 |h

which makes it complete. If x M , then T x M and we claim

ck
|T k x1 (t) T k x2 (t)k |t t0 |k d(x1 , x2 )
k!

97
For k = 0 this is clear, and in general we use induction to establish:
Z t
k k
kT x1 (t) T x2 (t)k kf (s, T k1 x1 (s) f (s, T k1 x2 (s)kds
t0
Z t
c kT k1 x1 (s) T k1 x2 (s)kds
t0
Z t
k
(c /(k 1)!) |s t0 |k1 ds d(x1 , x2 )
t0
k k
(c /k!)|t t0 | d(x1 , x2 )
So T n is a contraction mapping for large enough N , and the result follows. 2

Theorem 6.5 The solution above depends continuously on the initial data x0 .

Proof: Take h1 h and > 0 such that M h + b, and let

Y = {y C([t0 h1 , t0 + h1 ] B(x0 , ); Rn : ky(t, x) xk M h, y(t0 , x) = x}
which is a complete metric space as before. Now set
Z t
(T y)(t, x) = x + f (s, y(s, x))ds
t0

Since M h1 + b, T maps Y to Y and just as before T n is a contraction mapping

with a unique fixed point which satisfies
y
= f (t, y), y(t0 , x) = x
t
Since y is continuous in t and x this is what we need. 2

If f (t, x) is smooth then we need more work to prove that the solution to the equation
is smooth and smoothly dependent on parameters.

6.3 B: Smooth dependence

Lemma 6.6 Let A(t, x), B(t, x) be continuous matrix-valued functions and take M
supt,x kBk. The solutions of the linear differential equations
d(t, x)
= A(t, x)(t, x), (t0 , x) = a(x)
dt
d(t, x)
= B(t, x)(t, x), (t0 , x) = b(x)
dt
98
satisfy
eM |tt0 | 1
sup k(t, x) (t, x)k CkA Bk + ka bkeM |tt0 |
x M
where C is a constant depending only on A and a.

Proof: By the existence theorem we know how to find solutions as limits of n , n

where
Z t
k = a + Ak1 ds
t0
Z t
k = b + Bk1 ds
t0

Let gk (t) = supx kk (t, x) k (t, x)k and C = supk,x,t kk k. Then

Z t
gn (t) ka bk + CkA Bk|t t0 | + M gn1 (s)ds
t0

Now define fn by f0 (t) = ka bk and then inductively by

Z t
fn (t) = ka bk + CkA Bk|t t0 | + M fn1 (s)ds
t0

Comparing these two we see that fn gn . This is a contraction mapping, so that

fn f with Z t
f (t) = ka bk + CkA Bk|t t0 | + M f (s)ds
t0
and solving the corresponding differential equation we get
eM |tt0 | 1
f (t) = ka bkeM |tt0 | + CkA Bk
M
As gn (t) fn (t),
sup kn (t, x) n (t, x)k fn (t)
x
and the theorem follows by letting n . 2

Theorem 6.7 If f is C k and

d
(t, x) = f (t, (t, x)), (0, x) = x
dt
then is also C k .

99
Proof: The hardest bit is k = 1. Assume f is C 1 so that f /t and f /xi exist
and are continuous. We must show that is C 1 in all variables. If that were true,
then the matrix valued function where (i = /xi ) would be the solution of the
differential equation
d
= Dx f (t, ) (15)
dt
so we shall solve this equation by the existence theorem and prove that the solution
is the derivative of . Let F (s) = f (t, a + s(b a)). Then
dF
= Dx f (t, a + s(b a))(b a)
ds
so Z 1
f (t, b) f (t, a) = Dx f (t, a + s(b a))(b a)ds
0
But then
d
((t, x + y) (t, x)) = f (t, (t, x + y)) f (t, (t, x))
dt Z 1
= Dx f (t, (t, x) + s((t, x + y) (t, x)))((t, x + y) (t, x))ds
0

Let A(t, x) = Dx f (t, (t, x)) and (t, x) = (t, x)y and
Z 1
By (t, x) = Dx f (t, (t, x)+s((t, x+y)(t, x)))ds, y (t, x) = (t, x+y)(t, x).
0
The mean value theorem for a function g gives
kg(x + h) g(x) Dg(x)hk Kkhk sup kDg(x + y) Dg(x)k.
kykkhk

Use this estimate for g(x) = f (t, (t, x)), apply the previous lemma and use the
continuity of the derivative of f and we get
sup k(t, x)y ((t, x + y) (x))k = o(kyk)
|t|

and so Dx = , which is continuous in (t, x). Since also d/dt = f (t, ) this means
that is C 1 in all variables.
To continue, suppose inductively that the theorem is true for k 1, and f is C k .
Then A(t, x) = Dx f (t, (t, x)) is C k1 but since
d
= A
dt
we have is C k1 . Now Dx = so the xi -derivatives of are C k1 . But also
d/dt = f (t, ) is C k1 too, so is C k . 2

100
References
[1] P A Firby and C F Gardiner, Surface topology. Second edition. Ellis Horwood
Series: Mathematics and its Applications. Ellis Horwood, New York distributed
by Prentice Hall, Inc., Englewood Cliffs, NJ, 1991, ISBN 0-13-855321-1

[2] W S Massey, A basic course in algebraic topology. Graduate Texts in Mathe-

matics, 127. Springer-Verlag, New York, 1991. ISBN 0-387-97430-X

[3] H A Priestley, Introduction to complex analysis. Revised second edition. Oxford

University Press, Oxford, 2003. ISBN 0-19-852562-1

[4] P M H Wilson, Curved spaces, Cambridge University Press, Cambridge, 2008.

ISBN 978-0-521-71390-0.

101