Classification of Surfaces

Corrin Clarkson
REU Project
August 17, 2006

One of the simplest topological spaces is that of the surface. There are
many reasons why surfaces are nice objects to study. Our natural intuition
about space can easily be adopted to this study. Many surfaces can be
modeled in three space and so are things we can literally get our hands
on. For the most part the surfaces that we can understand intuitively are
compact, meaning they don’t go off to infinity. Probably the nicest property
of compact surfaces is the fact that they can be classified. This is very useful,
because it means we know exactly how many different types of compact
surfaces there are. The goal of this paper is to present the proof of the
classification of compact connected surfaces without boundary as well as
some of the key underlying proofs and definitions.

1 A Few Definitions
Definition 1.1 (Homeomorphism). A homeomorphism is a continuous in-
vertible function mapping one topological space to another. The inverse of
a homeomorphism is also continuous. Two Spaces are said to be homeomor-
phic, topologically equivalent, if there exists a homeomorphism mapping one
to the other. We write A ∼ B, if A is homeomorphic to B.

Definition 1.2 (Surface). A surface or 2-manifold is a topological space X

with the following characteristics:

1. X is Hausdorff

2. ∀ x ∈ X ∃ a neighborhood of x that is homeomorphic to R2

1
 Figure 1: The torus

After the first pair of edges is associated the square looks like a cylinder. When the
second pair is associated we get the torus. Hence the torus can be thought of as the
surface of a doughnut.

3. X has a countable basis

The following three surfaces are very important. Our goal will be to prove
that they are in fact the building blocks of all compact surfaces.

Sphere S 2 = {x ∈ R3 | |x| = 1} with the topology induced by the standard

topology of R3

Torus A torus or T 2 is the quotient space of the unit square obtained by

the equivalence relations (x, 0) ∼ (x, 1) and (0, y) ∼ (1, y). See Figure
1. It is also R2 /Z2 . It’s topology comes from the standard topology of
R2

Projective plane A projective plane or RP 2 is the quotient space obtained

by associating each point on the sphere with its polar opposite. It can
also be thought of as a 2-gon with its edges associated with opposite
orientation. See Figure 2

Definition 1.3 (Connected Sum). The connected sum of two surfaces is the
space obtained when an open disk is removed from each surface, and the
resulting boundaries are mapped together via a continuous function. We
write A#B for the connected sum A and B. See Figures 3. The connected
sum is a well defined operation.

Definition 1.4 (Triangulation). A triangulation of a surface is a finite set

of subsets that cover the surface with the following properties. Each subset
is homeomorphic to a triangle in the Euclidean plane. Any two subsets are

2
 Figure 2: The projective plane

A sphere with points associated to their polar opposites is homeomorphic to a

hemisphere with points on the boundary associated to their polar opposites. This in turn
is homeomorphic to a 2-gon with its edges associated with opposite orientation.

Figure 3: The connected sum of two tori

disjoint or have a single vertex or a single edge in common. (Here the terms
vertex and edge refer to those points that map respectively to vertices and
edges of a triangle under the homeomorphism.) See Figure 4 for examples
for illegal intersections.

Definition 1.5 (Euler Characteristic). The Euler Characteristic of a trian-

gulation is the defined by the equation X(M ) = V − E + F , where V is the
number of vertices in a triangulation of M , E the number of edges, and F
the number of faces or triangles. The Euler Characteristic is well defined for
all surfaces that can be triangulated. This can be rigorously proven, but due
to time constraints the reader is asked to take it as a matter of faith.

Figure 4: Illegal intersections

3
 Figure 5: A oriented pair of triangles

Definition 1.6 (Orientable). A surface is said to be orientable if all triangu-

lations can be oriented. A triangulation is said to be oriented if an ordering
of vertices of each face can be chosen such that the orderings of any two
triangles with a common edge induce opposite orderings on the shared edge.
See Figure 5

2 Some Theorems
For the purposes of this paper let M be an arbitrary, compact, connected
2-manifold without boundary.

Theorem 2.1. M can be triangulated.1

Proof. Let {Bi } be an irreducible finite cover of M consisting of closed balls,
B1 , B2 , . . . , Bn . Such a cover exists due to the fact that M is compact. Let
n
[
C= ∂Bn
i=1

Also let A be the set of all points in C that do not have Euclidean neighbor-
hoods.
If A is empty, then M is the sphere. The proof of this goes as follows.
A = ∅ implies one of three things: {Bi } consists of only a single closed
ball, none of the boundaries of the balls intersect, or the only boundary
intersections involve balls having their entire boundary in common. The
first case is impossible due to the fact that M has no boundary and so can
not be covered by a single closed ball.
1
This theorem was originally proven by Rado [6] . A shorter version is given by Doyle
and Moran [1]

4
 In general the fact that M has no boundary implies
[
∂Bm ⊂ B ∀1≤m≤n
B∈{Bi }\Bm

This means that in case two the boundary of any Bi is completely contained
in the interior of some Bj where i 6= j. Because the cover is irreducible,
we know that Bi 6= Bj . We have that ∂Bi ⊂ int(Bj ) ⇒ ∂Bj ⊂ int(Bi )
This in turn gives us Bi ∪ Bj homeomorphic to S 2 . The sphere can be
covered by two disks, one slightly larger than the upper hemisphere and
the other slightly larger than the lower hemisphere. The boundary of each is
completely contained in the other. This covering is homeomorphic to Bi ∪Bj .
This gives us a subset of M homeomorphic to S 2 . M is a 2-manifold so ∀ Bm
st Bm ∩ (Bi ∪ Bj ) 6= ∅ we have either Bm ⊂ (Bi ∪ Bj ). There are however
no such Bm , because the cover is irreducible. M is connected ⇒ @ Bm st
Bm ∩ (Bi ∪ Bj ) = ∅. It follows that M is homeomorphic to S 2 .
Similarly in the third case, two balls with common boundary are homeo-
morphic to the covering of the sphere that consists simply of the upper and
lower hemispheres. This leads to M being homeomorphic to the sphere in
this case as well. Therefore A = ∅ ⇒ M ∼ S 2 .
The sphere is easily triangulated. An example of a triangulation of the
sphere is the tetrahedron. Therefore if A is empty, M can be triangulated.
If A is nonempty, we cover it with a closed 2-cell, D. We can do this,
because A is compact and totally disconnected. M 0 = M/D ∼ M , because
modding out by a 2-cell doesn’t change the topology. Let R be the image of
C under the quotient map. This implies that M 0 \ R ∼ M \ (C ∪ D), because
C ∩ D 6= ∅ by definition of D. R is the one-point union of countably many
simple closed curves. We also have that all points in R except for the one
that corresponds to the image of D have Euclidean neighborhoods. Let p be
the point corresponding to the image of D.
Now let V be a 2-cell neighborhood of p. V covers all but finitely many
simple closed curves in R, by the compactness of R. Therefore we can assume
R is the one point union of finitely many simple closed curves, because we
can always mod out by V if necessary. This in turn implies that there exists a
2-cell neighborhood of p, V 0 that meets each curve in R in exactly two points.
By the Jordan-Schoenflies Theorem the complement of R ∪ V 0 is composed of
finitely many open 2-cells. If we thicken each of the arcs of R ∩ V 0 we obtain
a surface which we can triangulate. This triangulation is then extended to
the open 2-cells of the complement. Thus M can be triangulated.

5
 Figure 6: Triangulation proof on the torus

(A) A torus covered by finitely many closed balls. (B) A two cell that covers the points
with non-Euclidean neighborhoods. (C) Modding out by the 2-cell (D) The
neighborhood of p (E) The union of the boundaries with the neighbor hood of p (F) The
set on a torus, the open 2-cells of the complement

Corollary 2.2. M is homeomorphic to a polygon, P with edges “glued”

together in pairs.
Proof. By Theorem 2.1 we have shown that M can be triangulated. Let
{Ti } be a triangulation of M consisting of the triangles T1 , T2 , . . . , Tn . Let
the triangles be ordered such that Ti shares an edge ei with some Tj where
1 ≤ j < i ∀ i. We can do this, because M is connected. Each edge in {Ti }
belongs to exactly two triangles. This is due to M being a 2-manifold without
boundary. If there were and edge that belonged to only one triangle, it would
be a boundary component of M . If there were an edge that belonged to three

6
 Figure 7: The normal form for the sphere

or more triangles, the points belonging to it would not have neighborhoods

that were homeomorphic to an open disk. Therefore if an edge belonged to
three or more triangles M would not be a manifold.
This implies that the space obtained by associating all the ei is a polygon
with its edges paired. This can be easily seen if you look at the process step
by step. First start with T1 and then add T2 at edge e2 . The resulting space
is homeomorphic to a disk. The space remains homeomorphic to the disk
as each triangle is added, because there are never more than two triangles
meeting at a single edge. The space that results when all the ei have been
associated is homeomorphic to a disk and so must be a polygon. Each edge
of this polygon is an edge of a triangle in the triangulation and so must be
associated with exactly one other edge. Therefore M is homeomorphic to a
polygon with its edges associated in pairs.

Remark 2.3. From this point on when referring to “pair” of edges will mean
only a set of two edges that are identified with one another.
For the classification theorem we will need a normal form for the sphere,
the projective plane and the torus. We will also need a normal form for the
connected sum of these surfaces. The normal form for the torus and the
projective plane have already been mentioned. That of the torus is a square
with opposite sides associated with the same orientation. The normal form
of the projective plane is the 2-gon with its edges associated with opposite
orientation, as shown in Figure 2. The normal form for the sphere is similar
to that of the projective plane. It is a 2-gon with its edges associated with
the same orientation. See Figure 7.
The normal form for the connected sums is obtained by summing the
normal forms as shown in Figures 8 and 9. We assign a letter to each pair
of edges. If an edge is oriented counterclockwise we write the letter as its
inverse. Thus we can now represent each normal form with a sequence of
letters, for example the normal form for a torus would be written aba−1 b−1 .

7
 Figure 8: The normal form for the connected sum of two tori

Figure 9: The normal form for the connected sum of two RP 2

Similarly the normal forms for the sphere and the projective plane are written
aa−1 and aa respectively.
It is fairly easy to see that there is a connection between the normal forms
of surfaces and that of their connected sums. We would write aabbcc for the
connected sum of three RP 2 and aba−1 b−1 cdc−1 d−1 for the connected sum of
two tori. One would like to infer from this that we could write aabcb−1 c−1 for
the connected sum of a torus and a projective plane. In turns out however
that this is the same as the connected sum of three RP 2 and so should have
the same normal form.
In general we will call a polygon of normal form if it is the normal form
for one of the following figures, S 2 , T 2 , RP 2 , the connected sum of n tori,
or the connected sum of n projective planes. Our goal is to show that these
are in fact all connected compact surfaces without boundary.

Lemma 2.4. The connected sum of three RP 2 is homeomorphic to the con-

nected sum of RP 2 and T 2 .
Proof. The proof of this lemma is easily accomplished by taking the normal
form of the connected sum of three RP 2 and cutting/gluing it so that it is in
the “normal form” for RP 2 #T 2 . The process is laid out in Figure 10.

8
 Figure 10: The connected sum of three RP 2 ∼ (RP 2 #T 2 )

The dotted lies show where cuts are to occur and the grey lines show where edges have
been glued.

Theorem 2.5. M is homeomorphic to a polyhedron of normal form.2

Proof. By Corollary 2.2 M is homeomorphic to a polyhedron P with edges
identified in pairs. Below is a step by step method for transforming P into a
normal form while preserving its topological properties.

1. If P has ≥ 4 edges, then eliminate any pair that consists of adjacent

edges of opposite orientation.

2. We will call two vertices equivalent if they are identified with the same
point. Our goal is to make all the vertices of P equivalent. The method
to accomplish this is the following.
Assume step 1 has been completed. If P has 2 edges it is either the
projective plane or the sphere, both of which are normal forms. If it
has ≥ 4 edges, then we can assume there are no pairs of consisting of
adjacent edges of opposite orientation. We will now further examine
the case where P has ≥ 4 edges.
If all the vertices are not equivalent, then there must be a pair A, B
such that A and B are not equivalent , and they are separated by
exactly one edge, x. Let y be the edge on the other side of A . If
x and y form a pair and have the same orientation, then A must be
equivalent to B. If they were a pair of opposite orientation, then they
would have been eliminated by step 1. Therefore the two edges x and
y do not form a pair. There must then be some other edge y 0 of P that
is identified with y.
2
This proof can be found in both Massy [3] and Seifert-Threlfall [5]

9
 Now cut along a line from B to the vertex of y that is not A, and then
“glue” y to y 0 . The resulting polygon is still homeomorphic to M and
has all its edges paired. It also has one less vertex in the equivalence
class of A and one more in that of B. We can continue this process,
returning to step 1 whenever possible, until all the vertices are in the
same equivalence class or P has been reduced to a sphere.
3. As the normal form requires that pairs of the same orientation be ad-
jacent, we are in need of a way to bring such a pair into this position.
The following procedure is an example of how this can be done.
If P has a pair of nonadjacent edges a, a0 of the same orientation, then
cut along the line, d, connecting the first point of a to the first point
of a0 . Now all that remains is to glue a and a0 together. The resulting
polygon has a pair d and d0 that are adjacent. As this proceeder does
not separate any adjacent pairs, it can be repeated until all pairs of the
same orientation are adjacent. As before repeat step 1 when possible.
If at this point there are no pairs of opposite orientation, P is in a
normal form.
4. We will now work through the case where there are pairs of opposite
orientation. Given one pair of opposite orientation a, a−1 , there must
be a second pair of opposite orientation b, b−1 such that these four
edges are in the following arrangement, a . . . b . . . a−1 . . . b. The proof
of this statement follows from the simplification accomplished by the
previous steps.
From step 1 we have that there must be at least one edge between a
and a−1 . As a and a−1 are of opposite orientation, the edges of P can
be divided into two groups, those that lie between the fist points of
a and a−1 and those that lie between the last. From step 3 we have
that any pair of the same orientation can not be split between these
two groups, but must be fully contained in one. Therefore if there is
no pair of opposite orientation that is split between the two groups, all
pairs must be fully contained in one group or the other. This implies
however that the first point of a can not be identified with the last
point of a, a contradiction of step 2. Therefore there must be a pair
b, b−1 of the given arrangement.
In order to achieve the normal form aba−1 b−1 cut along the line, c,
connecting the first points of a and a−1 and then glue b and b−1 together.

10
 After this we cut along the line, d, connecting the first points of c and c0
and glue a and a−1 together. The resulting polygon is homeomorphic to
M and has edges in the following arrangement cdc0 d0 . As this procedure
doesn’t disrupt the arrangements of other such pairs it can be repeated
until all pairs of opposite orientation are in this arrangement. Once
again we repeat step 1 whenever possible. If at this point there are no
pairs of the same orientation P is in a normal form.

5. After completion of step 4 P is reduced to the connect sum of projective

planes and tori. From Theorem 2.4 we have that the connect sum
of three RP 2 is homeomorphic to the connect sum of a torus and a
projective plane. Therefore if P has pairs of the same orientation, it
can be simplified to a normal form by converting the tori into pairs of
projective planes.

After completion of the above procedure P is in the normal form. As

each step preserves the topology of P , it follows that M is homeomorphic to
a polyhedron in normal form.

Lemma 2.6. The Euler Characteristic of the connect sum of two compact
connected surfaces without boundary, L and N , is two less than the sum of
the Euler Characteristics of L and N , ie X(L#N ) = X(L) + X(N ) − 2
Proof. Let L and N be triangulated. Let TL and TN be triangles in the tri-
angulations of L and N respectively. Both TL and TN are homeomorphic to
triangles in the plane by definition. This implies that they are both home-
omorphic to closed disks. We now remove the interior of TL and that of TN
from L and N respectively. Lastly associate the edges of the two triangles in
pairs. The resulting space is L#N . Its Euler Characteristic is

X(L#N ) = (V (L) + V (N ) − 3) − (E(L) + E(N ) − 3) + (F (L) + F (N ) − 2)

= V (L) + V (N ) − E(L) − E(N ) + F (L) + F (N ) − 2
= X(L) + X(N ) − 2

Because both the Euler Characteristic and the connected sum are well de-
fined, this is enough to show that the lemma always holds.

11
Lemma 2.7. Each orientable “normal form” has a different Euler Charac-
teristic. Also each non-orientable normal form has a different Euler Char-
acteristic.
Proof. The connected sum of tori is always orientable, because the torus is
orientable. By Lemma 2.6 the Euler Characteristic of the connected sum of
n tori is n · X(T ) − 2n, and hence is different for all n.
Similarly the connected sum of projective planes is never orientable, be-
cause the projective plane is not orientable. We have that the Euler Char-
acteristic of the connected sum of n projective planes is n · X(RP 2 ) − 2n,
by Lemma 2.6. Therefore the Euler Characteristic of each non-orientable
normal form is different.

Theorem 2.8. If two compact connected surfaces without boundary are home-
omorphic, then they have the same Euler Characteristic.
Proof. Let L and N be homeomorphic compact connected surfaces without
boundary. By Theorem 2.1 L can be triangulated. Let ψ : L → N be
the homeomorphism form L to N . ψ is subjective by the definition of a
homeomorphism. This implies that the image of the triangulation of L under
ψ covers N . The images of the triangles must be homeomorphic to triangles
in R2 , because ∼ is a transitive relation. Similarly ψ must map vertices to
vertices and edges to edges. ψ is also injective by definition. This implies
that the number of triangles in the image is the same as the number in the
triangulation. Similarly the number of edges and faces must be the same.
Therefore L and N must have the same Euler Characteristic.

Corollary 2.9. M is homeomorphic to exactly one surface of “normal form”.

Proof. An orientable surface can not be homeomorphic to a non-orientable
surface. This is do to the fact that for any triangulation of a surface there ex-
ists a corresponding triangulation for any homeomorphic surface, as shown in
the proof of Theorem 2.8. Therefore if you can orient every triangulation of
a surface you can do the same for any homeomorphic surface. By Lemma 2.7
all the orientable normal forms have different Euler Characteristic. By The-
orem 2.8 homeomorphic surfaces have the same Euler Characteristic. There-
fore each orientable normal form is unique. Similarly each non-orientable
normal form is unique. By Theorem 2.5 we have that M is homeomorphic to
a surface of normal form. This implies that M is homeomorphic to exactly

12
one surface of normal form, because we have just shown that all such surfaces
are unique.

We have obtained a full classification of compact, connected surfaces with-

out boundary. By Corollary 2.9 we have that all such surfaces are either the
sphere or the connected sum of n tori or the connected sum of n projective
planes. This classification can easily be extended to compact surfaces with-
out boundary that are not connected, as such surfaces must simply be unions
of finitely many compact connected surfaces without boundary. It turns out
that compact surfaces with boundary also lone themselves to classification.
The ease with which 2-manifolds can be classified might lead one to hope
that such a classification is possible for all n-manifolds. This is however not
the case, with higher dimension the classification theorem get increasingly
difficult. Advances in the classification of 3-manifolds and 4-manifolds aare
still in progress. .

References
[1] P. H. Doyle and D. A. Moran “A short proof that compact 2-manifolds
can be triangulated” Inventiones Mathematicae 5:160-162, Springer,
Berlin 1968.

[2] Klaus Jänich “Topology” trans. Silvio Levy, Springer-Verlag, Berlin

1980.

[3] William S. Massey “Algebraic Topology: An Introduction” Springer-

Verlag, Berlin 1967.

[4] James R. Munkres “Topology” 2nd Ed., Prentice Hall, Upper Saddle
River, NJ 2000.

[5] H. Seifert and W. Threlfall “Lehrbuch der Topologie” Chelsea Publish-

ing Company, New York 1980.

[6] Radó, T “Über den Begriff der Riemannschen Fläche” Acts. Litt. Sci.
Szeged. 2, 101-121 (1925)

13