The classification of closed connected surfaces is one of the problems which gave
rise to modern topology, and will be the highlight of Math. 655. Recall that a
closed surface is a space which in the vicinity of each point looks like the plane
and which satisfies the finiteness condition: *the surface can be cut up into a
finite number of pieces each homeorphic to a disk*. (To simplify the
classification problem we are avoiding consideration of surfaces with boundary,
like a cylinder or Möbius strip.)

There is a very nice answer to this problem. First of all, there are the orientable
surfaces:

These consist of the sphere and finite connected sums of tori. The

More problematic are the unorientable surfaces. To cut a long story short, all the closed connected unorientable surfaces can be constructed by cutting out a finite number of holes from a sphere and then abstractly gluing in Möbius strips along the boundary of each hole. Note that the boundary of a Möbius strip is a simple closed curve, and thus can be mated (abstractly) to the boundary of a hole. Unfortunately the boundary of the Möbius strip is a knotted simple closed curve. (For the standard picture of the Möbius strip the boundary is a trefoil knot.) Thus it is difficult to visualize this construction.

While it is impossible to faithfully represent closed unorientable surfaces as surfaces in 3-dimensional Euclidean space, it is possible to do so if we allow selfintersections, and this gives us some idea of what the surface looks like. (We have already seen this with the Klein bottle.)

First we need a new representation of the Möbius strip:

This representation of the Möbius strip is called the

and the following animation shows what the result looks like:

The line of selfintersection between the green and red triangles in the center consists of double points, except for the two endpoints. Thus in the actual crosscap (ie. the Möbius strip) there are two distinct edges, one in the green triangle and one in the red triangle, and the two edges are glued to each other at the endpoints to form a simple closed curve. The following abstract gluing picture of the crosscap

may help you to see that it is homeomorphic to the Möbius strip. Note that the diagonally opposite corners of the inner square are glued together, and that the two edges referred to above are

The one advantage that the crosscap has over the more usual nonselfintersecting
representation of the Möbius strip in 3-dimensional space is that its boundary
is unknotted. Hence crosscaps can be physically glued into holes in a sphere, thus
yielding (selfintersecting) pictures of the closed unorientable surfaces in
3-dimensional space:

The first surface shown above is called

We can now state the classification theorem for surfaces precisely:

**Classification Theorem for Surfaces** *Any closed connected
surface is homeomorphic to exactly one of the following surfaces: a sphere, a finite
connected sum of tori, or a sphere with a finite number of disjoint discs removed and
with crosscaps glued in their place. The sphere and connected sums of tori are
orientable surfaces, whereas surfaces with crosscaps are unorientable.*

Möbius was the first to attempt the classification of surfaces. In an 1870 paper he proved the above theorem for orientable surfaces smoothly imbedded in 3-dimensional Euclidean space. The classification of unorientable surfaces was first announced by W. von Dyck in 1888, but his proof was incomplete. (Among other problems, at that time there was no satisfactory concept of an abstract surface, not imbedded in Euclidean space.) The first essentially rigorous proof of the classification theorem was given by M. Dehn and P. Heegard in 1907, under the assumption that surfaces can be triangulated (ie. cut up into a finite number of (curved) triangles intersecting with each other along (curved) edges or vertices.) The triangulability of surfaces was first proved by T. Rado in 1925, thus completing the proof of the classification theorem.

A key ingredient in the proof of Rado's theorem is (a strong
form of) the Jordan curve theorem:
any simple closed curve in the plane separates the plane into two regions, which
was proved by
A. Schönflies in 1906. (His proof contained some errors which were fixed by
L E J Brouwer in 1909.) It is easy to see that the
Jordan curve theorem for the plane is equivalent to the Jordan curve theorem for
the sphere:

The Jordan curve theorem is not true for the other closed surfaces. There are simple
closed surfaces on these surfaces which do not separate the surface. The appropriate
generalization of the Jordan curve theorem for arbitrary closed surfaces is given
below. It is stated in terms of *genus* of a surface, a concept which we
define as follows:

A sphere is defined to have genus 0, the connected sum of *g* toris is defined
to have genus *g* and an unorientable surface with *g* crosscaps is
defined to have genus *g-1*.

**Jordan Curve Theorem for Surfaces** *The maximum number of disjoint
simple closed curves which can be cut from an orientable surface of genus g
without disconnecting it is g. The maximum number of disjoint simple closed
curves which can be cut from an unorientable surface of genus g
without disconnecting it is g+1.*

The following pictures show maximal nonseparating systems of disjoint simple closed
curves on the connected sum of two tori (genus 2):

and on the Klein bottle (genus 1)