Papers on Group Theory and Topology

Max Dehn

Papers on Group Theory

and Topology
Translated and Introduced by
John Stillwell

With 151 Illustrations

Springer-Verlag
New York Berlin Heidelberg
London Paris Tokyo
John Stillwell
Department of Mathematics
Monash University, Clayton
Victoria 3168, Australia

AMS Classification: 01-A75, 55-03, 20 F 05,20 F 32

Library of Congress Cataloging in Publication Data

Dehn, Max, 1878-1952.
Papers on group theory and topology.
Includes bibliographies.
I. Combinatorial group theory-Collected works.
2. Topology-Collected works. I. Title.
QA17I.D393A25 1987 512'.22 87-317

© 1987 by Springer-Verlag New York Inc.

Softcover reprint of the hardcover 1st edition 1987
All rights reserved. This work may not be translated or copied in whole or in part without the written
permission of the publisher (Springer-Verlag, 175 Fifth Avenue, New York, New York 10010, USA),
except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed is forbidden.
The use of general descriptive names, trade names, trademarks, etc. in this publication, even if the former
are not especially identified, is not to be taken as a sign that such names, as understood by the Trade Marks
and Merchandise Marks Act, may accordingly be used freely by anyone.

Printed and bounded by R.R. Donnelley & Sons, Harrisonburg, Virginia.

Printed in the United States of America

987654321

ISBN-13: 978-1-4612-9107-7 e-ISBN-13: 978-1-4612-4668-8

DOl: 10.1007/978-1-4612-4668-8
 PREFACE

The work of Max Dehn (1878-1952) has been quietly influential in

mathematics since the beginning of the 20th century. In 1900 he became

the first to solve one of the famous Hilbert problems (the third, on the

decomposition of polyhedra), in 1907 he collaborated with Heegaard to produce

the first survey of topology, and in 1910 he began publishing his own

investigations in topology and combinatorial group theory. His influence

is apparent in the terms Dehn's algorithm, Dehn's lemma and Dehn surgery

(and Dehnsche Gruppenbilder, generally known in English as Cayley diagrams),

but direct access to his work has been difficult. No edition of his

works has been produced, and some of his most important results were never

published, at least not by him.

The present volume is a modest attempt to bring Dehn's work to a

wider audience, particularly topologists and group theorists curious about

the origins of their subject and interested in mining the sources for new

ideas. It consists of English translations of eight works : five of

Dehn's major papers in topology and combinatorial group theory, and three

unpublished works which illuminate the published papers and contain some

results not available elsewhere. In addition, I have written a short

introduction to each work, summarising its contents and trying to establish

its place among related works of Dehn and others, and I have added an

appendix on the Dehn-Nielsen theorem (often known simply as Nielsen's

theorem) . The latter theorem was never published by Dehn, though Nielsen

gives him credit for it, and Dehn's approach to the theorem became forgotten

when Nielsen (1927) buried it in a l80-page paper. In fact, Dehn's

approach can be made to work quite simply, as the appendix demonstrates.

In surface topology and the associated theory of fuchsian groups,

Dehn's work fills a gap between the works of Poincare and Nielsen.

The three together are the main sources of contemporary work in this

field, by Thurston in particular. with the recent publication of my

translations of poincare's Papers on Fuchsian Functions (Springer-Verlag

1985) and Nielsen's Collected Mathematical Papers (Birkhauser 1986),

these sources are now available in what I hope is a convenient form.

The present volume would not have been possible without the generosity

and encouragement of Wilhelm Magnus. He provided me with copies of

important unpublished manuscripts of Dehn and put me in contact with

Dehn's widow, Mrs. Toni Dehn, who very kindly gave permission for trans-

lations of these manuscripts to be published. Thanks are also due to

B.G. Teubner and Co. for permission to publish the papers which originally

appeared in Mathematische Annalen, to the Institut Mittag-Leffler for

permission to publish the paper from Acta Mathematica, and to the Mathe-

matisches Seminar of the University of Hamburg for permission to publish the

Otto Schreier paper whose translation is included as an appendix to

Paper 6.

The various papers in this book were typed by Anne-Marie Vandenberg

and Joan Williams, at a time when publication was not foreseen. It is

a tribute to the quality of their work that the book could be photographed

directly from their typescripts. Wilhelm Magnus and Dave Johnson read

large portions of the book and saved me from many errors. To all these

friends and colleagues I offer my sincere thanks.

 CONTENTS

Translator's Introduction 1 1

Paper 1 : Lectures on group theory 5

Translator's Introduction 2 47

Paper 2 : Lectures on surface topology 52

Translator's Introduction 3 86

Paper 3 : On the topology of three-dimensional space

(Uber die Topologie des dreidimensionalen Raumes.

Math. Ann. 69, (1910), 137-168) 92

Translator's Introduction 4 127

Paper 4 : On infinite discontinuous groups

(Uber unendliche diskontinuierliche Gruppen.

Math. Ann. 71 (1912), 116-144) 133

Translator's Introduction 5 179

Paper 5 : Transformation of curves on two-sided surfaces

(Transformation der Kurven auf zweiseitigen Flachen.

Math. Ann. 72 (1912), 413-421) 183

Translator's Introduction 6 200

Paper 6 : The two trefoil knots

(Die beiden Kleeblattschlingen.

Math. Ann. 75 (1914), 402-413) 203

Appendix to Paper 6

(Uber die Gruppen AaBb = 1, by Otto Schreier,

Abh. Math. Sem. Univ. Hamburg 3 (1924), 167-169) 224

Translator's Introduction 7 229

Paper 7 : On curve systems on two-sided surfaces,

with application to the mapping problem.

(Uber Kurvensysteme auf zweiseitigen Flachen

mit Anwendung auf das Abbildungsproblem.
Vortrag (erganzt) im math. Kolloquium,
Breslau 11/2/1922) 234
Translator's Introduction 8 253

Paper 8 : The group of mapping classes

(Die Gruppe der Abbildungsklassen.

Acta Math. 69 (1938), 135-206) 256

Appendix The Dehn-Nielsen theorem 363

 TRANSLATOR'S INTRODUCTION

The following article is part of the first chapter of Dehn's

lectures on group theory, probably written in 1909 or 1910. The date
can be guessed from the next chapter of the notes, which includes an
outline of some results from the famous paper, Dehn [19l0),mentioning
that details will appear in the Mathematische Annalen, which indeed
they did in 1910.

The article is of interest mainly as Dehn's first discussion of

the diagrams which he called "Gruppenbilder", and which are now known
as Cayley diagrams. His first published work on these, Dehn [1910),
concedes that Cayley had the idea first (Cayley [18'Z8)), but it is
clear from the present article that Dehn was not at all influenced by
Cayley. He was, however, heavily influenced by the geometric represent-
ations of groups which arise in function theory the Poincar~ [1882)
definition of fuchs ian groups by tessellations of the hyperbolic plane,
and the more general theory of groups defined by tessellations in
Dyck [1882).

Following Dyck and POincart, a tessellation which is mapped onto

itself by all elements of a group G can be viewed as a "picture" of
G when an arbitrary cell, Rl , is associated with the identity element
1 of G, and the cell R to which Rl is sent by g 6 G is
g
associated with g. The particular elements which send
to adjacent cells can be taken as generators of G, since can
be sent to any Rg by repeatedly moving from one cell to an adjacent
cell. Dehn's idea was simply to "dualise" this diagram: choose a
point, called a vertex, vg in each cell R
g
to represent g E G,
draw edges from each vertex to the vertices in adjacent cells, and
orient and label these edges with the corresponding generators o

An immediate advantage is that the relations of G are now highly

visible, as closed edge paths in the diagram, and the relation corres-
ponding to a given closed path can be immediately written down by
reading the sequence of labels on its edges o In a Dyck style diagram,
on the other hand, relations correspond to closed chains of cells, which
are considerably harder to read o
 2

Another advantage of Dehn's diagrams is that they can be defined

without reference to the plane or other surfaces. The diagram D of
G
G is just a labelled, directed, graph, with a vertex V for each
g
g E G and, for each generator g. E G, an edge labelled
gi directed
~

from V to V Dehn does not actually use this definition;

g ggi
perhaps because it is of no help in actually constructing diagrams.
In practice, G is given by defining relations rl =r2 00, = rn 1,
and the whole problem is to decide what the vertex set of DG is, ioe.
to decide when two words in the generators are equal as a consequence
of the relations.

There is no general way to do this, because of the unsolvability

of the word problem (Novikov [1955]), but one can build "pieces" of
DG, each consisting of a vertex with closed paths r l , ,r attached, .00
m
and then hope that these pieces can be fitted together o Dehn's definition
of the group diagram (section 1 below) reflects this state of affairso
At each vertex one has
(1) One incoming, and one outgoing, edge for each generator;
(2) A closed path for each defining relation ("basic path") •
A graph satisfying these conditions will in general represent only a
quotient of G, since it may contain closed paths (relations) which are
not put together (consequences) from the basic paths (defining relations) •
One therefore needs Dehn's third condition.:
(3) Each closed path in the diagram can be put together from basic
paths 0

Dehn does not define "put together" (zusammensetzen), but a suitable

definition is evident when one considers the graphical counterpart of the
process of inserting a defining relator in a word. Insertion of relator
r between the subwords u,v of w = uv is achieved by
multiply by v-lrv -1 cancel vv- 1
w = uv '>' uv v rv ;. urv, the graphical
counterpart of which is Fig o 1

-1
multiply by v rv
>

cancel vv-1
)
 3

In other words "putting together" should mean forming products with

paths of the form v-lrv, where r is a basic path and v is arbitrary,
and allowing trivial cancellation. One is then enabled to pull a path
from one side to the other across any basic path.

Dehn's property 3 obviously holds for any planar or spherical

group diagram, since the diagram then partitions the surface into cells
whose boundaries are the basic paths, and any closed path can be reduced
to the identity by pulling it across one cell at a time. Fortunately,
many familiar and interesting groups are of this type, so Dehn is able
to confine most of his exposition to this intuitively simple case.
Most of the results he discusses were in fact already known, though of
course he is putting them in a new and more unified setting.

The lectures are handwritten (not by Dehn) in a note book which

for many years was in the possession of Dehn's student, Ruth Moufang.
After her death it passed to Wilhelm Magnus, who very kindly provided me
with a photocopy of the most interesting sections. The notebook is now
with the rest of the Dehn Nachlass in the Humanities Library, Austin,
Texas. The diagrams in the notebook are hand drawn and rather too small
to reproduce well, so I have redrawn most of them, hoping that they will
now be clearer, though still faithful to the originals. For extremely
accurate and beautiful group diagrams one should gonsult the works of
Fricke and Klein, or the commentary on them by Magnus [1974]. Some
lesser known sources with splendid group diagrams are Burnside [1911],
Threlfall [1932] and Coxeter [1939] (particular pp. 126-139) and [1979].
In a sense the first non-trivial group diagram is in Schwarz [1872],
though Schwarz does not make the group explicit. His diagram is also
noteworthy because it does not hide the construction lines, something
which all other authors, apart from Coxeter, have done.

REFERENCES

W. Burnside [1911] Theory of Groups of Finite Order

2nd ed., Cambridge University Press

A. Cayley [1878] On the theory of groups

Proc. Lond. Math. Soc. 9, 126-133

H.S.M. Coxeter [1939] : The abstract groups Gm,n,p

Trans. Amer. Math. Soc. 45, 73-150
[1979] The non-euclidean symmetry of Escher's
 4

picture 'Circle Limit III'

Leonardo 12, 19-25

M. Dehn [1910] Uber die Topologie des dreidimensionalen

Raumes.
Math. Ann. 69, 137-168.

W. Dyck [1882] Gruppentheoretische Studien

Math. Ann. 20, 1-45

W. Magnus [1974] Noneuclidean Tesselations and Their Groups.

Academic Press.

P.S. Novikov [1955] On the algorithmic unsolvability of the word

problem in group theory (Russian) Trudy Mat.
Inst. Steklov 44, 143 pp.

H.A. Schwarz [1872] Ueber diejenige FaIle, in welchem die Gaussische

hypergeometrische Reihe eine algebraische Function
ihres vierten Elementes darstellt.
J. reine u. angew. Math. 75, 292 - 335.

W. Threlfall [1932] Gruppenbilder

Abh. sachs. Akad. Wiss. Math-phys. Kl. 41, I-59.
 LECTURES ON GROUP THEORY

GENERATION AND GEOMETRIC REPRESENTATION

1. Generation and geometric representation of L3!

We begin our considerations with the generation and geometric represent-

ation of the group L3! of permutations of three thingso We have
already shown that this group may be generated by a transposition sl'
which exchanges the first and second terms, and a cyclic permutation
s2 of all three terms, which replaces the first element by the second,
the second by the third, and then the third by the first. When we apply
the transposition sl twice, or the cycle s2 three times, then in each
case we come back to the identity, hence

1 and

However, there must be another relation between these generating

elements and s2' otherwise infinitely many elements could be composed
from themo This relation could say, e.g o, that a certain power of the
product s l s2 equals the identity. If the initial ordering of the three
things is 123, then the operation sl leads us to the ordering 213,
and if we apply the operation s2 to this we come to the ordering 132;
this ordering becomes 312 after application of sl' whence we return by
s2 to the identity 123. Thus we have found that

Now we shall consider whether the group E3! is completely defined by the
generators and and the relations

between them. The answer to this question is given to us by the

geometric representation of a group, or the group diagram.

We take an arbitrary point [vertex], which will represent the identity,

and then two more vertices which will represent the orderings obtained from
the identity by the generators and The line segments [edges]
from the identity to these two points then represent the generators sl and
s2 and we therefore label them by sl and s2' or 1 and 2 for short,
and indicate their directions by arrows. When we agai.n apply sl to the
 6

vertex just obtained by the substitution sl' we shall return to the

identity since sl2 = 1. Thus the opposite direction on the edge 1
also corresponds to the element sl' so that we can omit the arrow on
this edgeo However, this is not the case on the edge 2, because if
we reapply the operation s2 we do not come back to the identity, but
to a new vertex o If we now apply the operation s2 to this new vertex,
3
then we do return to the identity, since s2 = 1. Finally, we apply the
operation s2 to the vertex reached by the operation sl; we then reach
a new vertex, from which we must return to the identity by subsequent
2
application of the substitutions sl and s2' since (sls2) = 1. The
vertex at which we arrive by the

Fig. 1
1

substitution must coincide with the vertex for the substitution

2
s2 ' because from both places we return to the identity by application of
the operation When we represent all the different substitutions
obtainable from and s2 by composition, we come to a figure which is
equivalent to the group generated by sl and s20 This figure has the
following properties :

Fig. 2

10) If we label each vertex with the operation by which it is reached

from the identity, then one sees that the ~eighbouringvertices of each vertex
-1 -1
are precisely those obtainable by attachment of sl,sl ,s2,s2 0

2.) At an arbitrary vertex, regarded as the identity, the three rel-

ations are satisfied, since each vertex is in a quadrilateral, a
triangle and 2-gon, i.e. a [double] line, and each relation between the
generators corresponds to such a closed polygono
 7

3) Each relation in the figure, i.eo each closed edge path, can
be put together from the three basic relations, since each closed edge
path in the figure can be put together from basic polygons. E.g.

It follows from this property that our construction makes two different
elements of the qroup correspond to two different vertices, and conversely,
so that the figure has exactly as many vertices as the group has plements.
Now, since our figure has six vertices, the group defined by sl and
and the relations between them must be identical with L 3 !. Because a
group which has additional relations must necessarily have fewer elements,
since elements which were different from the identity in the original
group now become equal to ito

The circumstances of one-to-one correspondence between group elements

and vertices, and the display of all possible relations between them,
enti tle one to call such figures "group diagrams". We shall now see the
great advantages of such group diagrams in further exampleso

2. The group diagram of L4! • We now address ourselves to the problem

of constructing the group diagram for the group of permutations of four
things 0 Here we can choose the generating elements to be a transposition
sl and a four-termed cycle s2' where the transposition always exchanges
the first and second elements and the cycle sends the first element to
the last place. Between and s2 we then have the easily verified
relations

2
The relation sl 1 corresponds to a 2-gon in the group diagram, which
we represent by an edge without an arrow, the relation s4 = 1 corresponds
3 2
to a quadrilateral, and the relation (s l s2) = 1 corresponds to a
hexagon. Thus, if the three relations are to be satisfied, we have to
draw the group diagram so that the three polygons in question meet at each
vertex. We then obtain a figure with 24 vertices, which assures us that
the group defined by sl and s2 and the three relations is identical with
the symmetric group L4 !. One could also show this by mere calculation but

the process would be extraordinarily inconvenient. Our group diagram gives

us the result immediately; it is to a certain extent a formula or
 8

2 2 2 1 2 Fig. 3

computational scheme, and provides us with an example from the border

zone between formula and figure.

3. Group diagrams for the alternating groups A12 and AGO

Next we shall sketch the group diagram fur the alternating group A12 •
Here we can proceed in different ways. On the one hand we can consider
two three termed cycles as generators, say

sl = (123) and s2 (234)

for which we have the relations

3 3 2
sl = s2 = (s l s2) 1;

on the other hand, we can also take a double transposition with a three
termed cycle, e.g.

sl = (13) (24) and (123) ,

for which the relations are then

In the first case two triangles and a quadrilateral meet at each point,
in the second case there is a 2-gon, a triangle and a hexagon.
 9

\
Fig. 4

Now we shall similarly draw the group diagram for the alternating group
A60 of five symbols, which, as we know, consists of all even order
permutations in ~ " Since each odd cycle can be generated by an even
5.
number of transpositions, we choose the generators of our group to be
a three-termed and five-termed cycle, say

sl = (123) and s2 = (24153).

These two elements do in fact generate the alternating group A60 ,

since the group generated by them must be transitive and primitive: the
first fact follows because it contains a five term cycle, the second
because 5 is a prime number, because the permutations composed from
and can only be of even order, and finally because a three-
termed cycle appears in the group. Now the generators satisfy the
relations

so that there is a triangle, a quadrilateral and a pentagon at each vertex o

 10

In order to give the figure the clearest possible form, we construct it

from the dodecahedral net by replacing each vertex by a triangle whose
sides correspond to the cycle sl' and connect the vertices of the
triangles in a suitable way by edges which correspond to the cycle s2.
The presentation of the alternating group A60 , which we have obtained
in this way, was first given by Hamilton~

4. The icosahedral and dodecahedral groupo The alternating group A60

coincides with the icosahedral and dodecahedral group since, as we shall
now show, the latter have the same generators and relations as A600
The three-termed cycle sl obviously corresponds to the robation of the
icosahedron about a face normal, i.e. about the line passing through the
centres of two opposite faces, and the five-termed cycle corresponds to the
rotation about a main diagonal, which connects two opposite vertices.

Now we have to establish the relations which hold between

and If we consider one triangle of the icosahedron, say the
triangle ABC, then sl' the rotation about a face normal, sends
A to B, B to C and C to A. If we then apply the operation s2
Le. rotate about the main diagonal BB' then the new point A remains
fixed, while the new point B goes to the new point C and the point
C goes to the point V [in the neighbour of the original triangle].
Thus the product sls2 sends the original points A,B,C to the points
B,A,V respectively. We see from the icosahedron figure that this
change of positions can be brought about by an 180 rotation about
the line connecting the midpoints of the edges AB and A'B'. If
we now carry out a second such rotation, then we arrive back at the
identity. Thus we have shown without special calculation that
and that a rotation about a line connecting midpoints of opposite edges
can be obtained from the generators and Now, since generators
and and the three relations.

= s52=1
(s s )2 = 1
2
define a group with 60 elements, and since only 60 different rotations
are possible for the icosahedron, we have in fact found the icosahedral
group. We see from this that our generators and defining relations
establish an isomorphism between the alternating group A60 and the
icosahedral and dodecahedral groupo

* Memorandum respecting a new system of roots of unityo Phil M.ag 0 12 (1856)

446 (Translator I s note).
 11

5. Extension of the icosahedral group by reflections

We shall now derive, from our icosahedral group, a new group 0120
of order 120, by combining the rotations of the icosahedron with
its reflection in its midpoint. Such a reflection sends each vertex
of the icosahedron to the diametrically opposite vertex. One sees
without difficulty that such an exchange of vertices is not obtainable
by rotation. Since one can apply this reflection to each of 60 positions
obtainable by pure rotation of the icosahedron, the extension of the
icosahedral group by reflection is a group with 120 elements. In
order to be able to construct its group diagram we have to add a third
generator 53' representing the reflection in the midpoint, to the
generators sl and s2 of the icosahedral group. Since reflection
in a point has period 2, i.e. two applications of the same reflection
return each point to its original position, 53 satisfies the equation
2
s3 = 1.

The relations between and the two rotations and are also
easy to determine. When we reflect the vertices of a triangle on the
icosahedron, say AB~ then we obtain the diametrically opposite triangle
and indeed A goes to A' , B to B ' , and C goes to C '. If we then
take the operation sl' and thus rotate the triangle A'B'C' about the
normal through its midpoint, then A' goes8' , 8' to C' and C'
to
to A'. Finally, when we reflect the triangle A'B'C', in its new
position, the point A' goes to the old point 5, B' goes to the old point
C, and C' goes to the old point A. Thus the operation has
sent the points A,B,C -to B,e,A respectively, as also occurs with the
-1
rotation alone. Recalling that s3 = s3' we then have the relation

between and One can show similarly that and satisfy

the same relation

Thus both the generators sl and s3 are sent to themselves by trans-

formation [conjugation] with s3' We shall now show, with the help of the
group diagram, that the group is in fact completely defined by the
°120
operations and the six conditions

1, 1,
 12

We construct the group diagram from the group diagram of A' 60 by

erecting perpendiculars of equal length, representing the generator
s3' at each vertex. We connect the endpoints of these perpendiculars
so as to make another diagram of A60 • We then obtain a figure
consisting of 120 vertices, whose top and bottom levels are both the
352
figure for A60 • Thus the three relations sl = s2 (s l s2) = 1
are certainly satisfied at each vertex. It is also obvious that
2 -1-1
s3 = 1, and the other two relations s3 s l s3 = sl and s3 s2s 3 s2
are satisfied at each point as well, as one sees immediately by
consideration of the small rectangles

Fig. 6

formed by corresponding edges on the top and bottom levels and the
connecting perpendiculars. since our figure has the property that the
number of edges issuing from each vertex is the same as the number of
generators which can be applied, and the six relations between the
generators are satisfied at each vertex, it is the diagram of the group
defined by the generators sl,s2,s3 and the six equations

It only has to be proved that all possible relations between sl,s2,s3

in our group diagram can be put together from the six relations just
mentioned. This holds not only for the top and bottom levels, where
it is immediate since each is the diagram for A60 , but also for the
whole figure, because each closed path in it bounds a simply connected
piece of surface made up of pieces corresponding to relations. As in the
well known Stokes' Theorem of integral calculus, a circuit round
individual pieces is equivalent to a circuit of the boundary polygon of
the whole surface, since inner edges are traversed twice, and in
opposite senses, so that they cancel out. We have therefore proved
completely that the six given relations between the generators sl,s2,s3
 13

completely define the group G120 , and that it is represented by our

spatial group diagram.

6. The group diagram of ES!o The extension of the icosahedral group

by reflection is by no means identical with the symmetric group ES!'
even though they have the same number of elements. This will be shown
by investigation of the latter group and its group diagram. As we know,
the symmetric group ES! results from the alternating group A60 when
we add a transposition which satisfies the equation

1.

Now we have only to establish the relations between s3 and the other
two generators sl and s2' which are connected by the relation
2
(s l s2) = 1, sl being the three-termed cycle (123) and s2 the
five-termed cycle (24153)0 If we choose s3 to be the transposition
(12), then simple calculation yields the easily verified relations

(132) and (14253) 0

If the group G120 were isomorphic to the symmetric group on five

symbols then the reflection would necessarily correspond to a transposition
or a double transposition; because among the permutations of five symbols
the only elements whose second powers equal the identity are the
transpositions and double transpositions. On the other hand, the operation
s2 of G120 must correspond to an element of fifth order in the symmetric
group , and in E120 this can only be a five-termed cycleo But now since
we always have

in E120 when t is a transposition or double transposition and Cs is

a five-termed cycle (in fact, transformation by t leaves at least one
term, say 1, unmoved, and then the term to which 1 is sent by Cs is unmoved
-1
also, if t cst is to equal cs ' and so on, so that t is the identity).
It follows that G120 and E120 cannot be isomorphic groupso

We shall now use construction of the group diagram to show that the
symmetric group E120 is completely defined by the generators

(123) , s2 = (24153), s3 = (12)

 14

with the relations above. We set up the group diagram in a similar way
to that for G120 ' by choosing the group diagram of the alternating
group A60 as top and bottom figure in a spatial structure whose
outline is that of a five-sided prism with parallel top and bottom
faces. The relations satisfied by and alone are then
satisfied at each of the 120 vertices of this figure. To satisfy the
remaining relations at each point, we connect each vertex of the bottom
figure with the points of the top figure associated with it by
-1 -1 -1 2 2 2
s3 s l s3 = sl and s3 s2s 3 s l s2s 1 s 2' and label the connecting
edges with s3· It may then be shown that edges corresponding to
-1 -1 -1
each of the five operations sl,sl s2,s2 , s3 s3 issue from
each vertex in the figure, also that the six given relations are
satisfied at each point and that each possible relation between the
generators in E120 can be expressed by these six basic relations.
This proves that our figure is in fact the group diagram for the symm-
etric group E5 !.

7. Generation of the group El20 by two operations.

In connection with the symmetric group El20 , we now take a somewhat
general view of the generation and geometric representation of groups.
Namely we pose the question whether the symmetric group L 120 can be
generated by two generators. This is certainly possible, because we
shall show that a five-termed cycle s2 and a transposition s3 alone
suffice to bring each of the five symbols 1,2,3,4,5 to an arbitrary
position. We choose the five-termed cycle so that it brings the symbol
in the last place to the first place, first to second, etc., and the
transposition so that it exchanges first and second places. To
demonstrate our claim it suffices to prove that repeated application
of suffices to bring an arbitrary symbol to a given
position, symbol 1 to the third place, without changing the order
of the other symbols.
Thus we have the original ordering 12345
Order after application of s3 21345
-1
s2 13452

s3 31452

J' s2 23145
 15

This example discloses how we can generate the whole symmetric group
E120 by application of 52 and alone.

But now we still have to establish the relations for and

First, we certainly have
2
1 and 53 1
and the scheme :

12345

53 21345
52 52134
53 25134
52 42513
53 24513
52 32451
53 23451
52 12345

yields the relation

4
(5 35 2' 1

between and However, the three equations found so far

are by no means sufficient to completely define the group E120 • This
is certainly probable, in view of the fact that just now we needed six
relations between the three generators, but nevertheless a rigorous
proof is required. We do this by construction of the group diagram

FiB. 7
 16

When we draw the diagram of a group with generators s2 and s3 and

5 2 4
relations s = s = (s3 s 2' = 1, a pentagon and two octagons must
2 3
meet at each point, where the edges of the pentagons are denoted by
s2 and the edges common to two octagons are denoted by s30 In
our figure the s2 edges are provided with arrows which give the
direction of the operation s2; the other edges need no indication of
-1
direction since s3 = s3 • (Here we are speaking only of the heavily
drawn polygons; the lightly superimposed net of triangles will be
explained latero' We now see from the constructed part of the group
diagram that the group defined by s2 and and the three relations
certainly has more than 120 elements, because our figure already
shows more than 120 vertices, from each of which issue three edges
-1 -1
corresponding to the operations s2,s2 ,s3 = s3 and at each of which
the three relations are satisfied o It may now be shown that the process
of constructing two octagons and a pentagon at each vertex so as to
satisfy the above relations, used for the polygons constructed so far,
may be continued indefinitelyo A rigorous proof of this can be given,
either by elementary theorems of topology and induction, or else by
methods of non-euclidean geometryo At any rate, it follows from this fact
that the group defined by the generators s2 and and the relations
5 2 4
s2 = s3 = (s3s 2' =1 is of infinite order.

8. Excursus into non-euclidean geometry. Here we come for the first time
to an infinite groupo In the construction of group diagrams for infinite
groups, a knowledge of non-euclidean geometry is indispensable. We
shall therefore assemble the theorems of this discipline which are most
important from our present point of view.

There are two different methods of representing thenon-euclidean plane,

one of which was developed by Poincare in connection with the theory
of automorphic functions, and the other of which was found directly by
Cayley and Klein. Poincare takes the flm~uclidean plane to be the upper
half of the ordinary euclidean plane bounded below by the so-called
"Poincare line~. The "non-euclidean lines" in this half plane are the
semicircles with midpoint on the Poincare line, so the Poincare line is
cut orthogonally by all non-euclidean lines.
 17

Fig. 8

Poincare line

The justification of the term "line" for these semicircles is that such
a semicircle is uniquely determined by any two of its points, and two
such semicircles or non-euclidean lines have only one pmint of inter-
section [at mostl, since the lower half plane is omitted. The
"angle" between two intersecting non-euclidean lines is taken by
Poincare to be the ordinary euclidean angle between the semicircles.

We shall now prove the following fundamental theorem of noneuclidean

geometry :

In a noneuclidean triangle, the angle sum is less than two right

angles.

Since inversion, or transformation by reciprocal radii, is known to

leave angles invariant, it suffices to prove our theorem for a right-
angled triangle whose short sides are segments of a perpendicular to the
Poincare line and a semicircle bisected by this perpendicular. Thus we
have to prove our theorem for the triangle ABC, with right angle C,
shown in the accompanying figure. It will then follow immediately for the
general right-angled triangle because we can always map the latter onto
one of our special right-angled triangles by inversion in one short side.
To prove the theorem we describe concentric circles around the foot of
the perpendicular. Then the angle between these circles and the hypotenuse
circle of our triangle increases until it takes its maximum value,
8 8max ' at the point where the hypotenuse meets the perpendicular.
The most convenient way to see this is to replace the angle between the
circles by the angle between their radii. If we let ~ denote the angle
 18

c
Fig. 9

,~
81 '8 1
,"~ I \

-
/ 1 \ :' \
/ 'I
/ El..:'8 1)'
' I
~ r
":," \'\ ,
I
/
;' "" ',,'
() ~,
",
1 '\r
"\ \
/ \

I
':",",
I
..! , I
I' ,
',r" \ \
' \'
\

\ \
" I....... ... ',\ \
I \ 1/ '\..F, '\' \ ,
,,',<+,
\ ~ ,
"',.,\,
~~~
I
v
a
between the Poincar~ line and the variable radius vector to the
hypotenuse, let r be the radius of the hypotenuse circle and let
a be the distance from the centre of this circle to the foot of
the perpendicular, then the sine law of ordinary trigonometry gives

sin 8 =~ sin ~ =c sin ~,

where c is a constant. Thus 9 takes its maximum value for ~ = 90 0 ,

and hence for the circle through the intersection of the hypotenuse and
the perpendicular. If we let n denote the angle in our right-angled
triangle at this point, then

n + 9max n + e = 90 0

and consequently we must have the inequality

n + 9 < 90 0

for our right angled triangle. This proves the assertion that the
angle sum of triangle ABC is less than two right angles.

Now, to prove this theorem for the angle sum of a general non-
euclidean triangle, we shall show that each triangle is decomposable
into two right-angled triangles. For this, we need only the following
lemma

A trian~e can have no more than one obtuse angle

For convenience' sake we again transform our triangle so the side

on which the hypothetical two obtuse angles lie is a segment of a
 19

perpendicular ~ to the Poincare line. Now any circle whose lower

left angle with the perpendicular is obtuse must have its centre to
the left of t, and hence it cannot cut the circle k in such a way
that the angles on t of the resulting triangle are both obtuse.

Fig. 10

Now that we have shown that a triangle can have at most one obtuse
angle, we transform a given triangle so that the side opposite the
obtuse angle ACB goes onto a perpendicular to the Poincare line,
and then take the foot of this perpendicular as the centre of a semi-
circle which goes through C
Fig. 11

Poincare line

since this circle must always meet the base of our triangle, we have
divided the given triangle into two right-angled triangles, with
hypotenuses AC and BC. But now in each of these two right-angled
triangles the two angles at the hypotenuse sum to less than 90 0 ,
hence the angle sum of the given non-euclidean triangle is less than two
right-angles, as was to be proved.
 20

We can now go easily from the Poincare half plane by a conformal

mapping to the Klein-Cayley representation of the non-€uclidean plane.
"Conformal" here means only that the Poincare angles are sent to
angles with the same value in the Cayley angle measure. We first map
the Poincar~ half plane onto a sphera whose south pole touches the
half plane on the Poincare line, by the inverse stereographic project-
ion from the North Pole. This carries the Poincare line into the
vertical meridian circle coplanar with it, and the Poincare half plane
into the corresponding hemisphere. Since stereographic projection is
conformal, the circles orthogonal to the Poincare line became circles
orthogonal to the main meridiano These circles go to straight lines
under parallel projection orthogonal to the plane of the main meridian,
since our orthogonal circles are parallel to the direction of projection.
The interior of the circle, onto which the hemisphere is projected,
represents the non-euclidean plane according to Klein's method.

Fig. 12

Poincare line

Each straight line in the latter disc corresponds to a circle orthogonal

to the main meridian on the hemisphere and then to a circle orthogonal
to the Poincare line in the half plane. For lines through the midpoint
of our absolute figure, which correspond to great circles on the sphere,
the angles between them are preserved by the orthogonal parallel projec-
tion, though the angle in general is changed o Each conformal circle-to-
 21

circle mapping in the Poincare half plane corresponds to a collineation

of the disc into itself. This enables us to carryover the measure of
angle between two non-euclidean lines to the Klein representation.
Namely, we can apply a collineation to send the two given lines to
lines through the midpoint, and then define the angle between the
original lines to be the angle between their transforms at the midpoint,
which equals the angle between the corresponding semicircles in the
Poincare half plane.

It follows, first, that angle measures in both representations

of the non-euclidean plane agree. In particular, the Klein representation
also admits the important theorem that the angle sum of a non-euclidean
triangle is less than two right angles. But the collineations of the
disc into itself also yield a precise definition of equality of segments
in the non-euclidean plane. Namely, if we are given two segments AB
and A'B' on the lines XY and X'Y' then we can use two separate
collineations

------ Y

Fig. 13
x

X'

to carry these lines, separately, to lines through the midpoint, then

bring the latter lines to the same position by rotation, and finally
bring the initial points of our segments into coincidence by a displacement
of one line along the other, when these three collineations bring the
segments AS and A' B' into coincidence, we say they are equaL The
condition for this is equality of the two cross ratios (ASXY) and
(A' B' X, Y').
 22

We now conclude our few details of non-euclidean geometry with

a remark of special importance for the construction of group diagrams.
A regular polygon in the non-euclidean plane can be continuously
changed, without i~ceasing to be regular, into one whose vertices
lie on the boundary circle. One sees this immediately by consideration
of polygons whose midpoints are at the centre of the circle, since
they are simultaneously regular as both euclidean and non-euclidean
figures.

Fig. 14

But two lines meeting on the boundary circle enclose an angle of zero
non-euclidean measure, since the corresponding semi-circles in the Poincare
half plane meet at the Poincare line and therefore touch, so that the
angle between them is zero. It follows from this, together with the
theorem on the angle sum of a triangle, that in the non-euclidean plane
the angle sum of a closed polygon with n sides can take any value between
o and (n - 2)~.

9. The group diagram of an infinite group. Our last remark now provides
us with a simple construction of the group diagram for the infinite group
defined by generators with the relations

In this construction we apply the process of reflection in a line or circle,

whereby a point A is replaced by its harmonic point A' with respect to
G and S, where G is the pole of the line of reflection and S is the
intersection of GA with this line. This reflection, which always maps
the interior of the circle onto itself, is characterised by the fact that
 23

Fig. 15

the cross ratio to S A A' 1 = -10 Reflection obviously sends each

closed polygon to a closed polygon, and each regular polygon to a
regular polygono

The group diagram of the above mentioned group must be set up so

that a pentagon and two octagons meet at each vertex o Thus if we want
to choose regular polygons in the ordinary euclidean plane, the rays
-1 -1
oorr~ponding to the operations s2' s2' s3 = s3 out of a given
vertex will be such that the first two make an angle 3rr/5, while each
makes an angle 3rr/4 with the thirdo But this is impossible in the
ordinary plane, since Srr/5 + 2 • 3rr/4 > 2rr. In the non-euclidean plane,
however, we can make the angle of regular polygon take any value between
o and n
by suitably choosing the side length, and hence we can
also arrange for the sum of the angles of the pentagon and two octagons
meeting at a vertex to be 2rr. When this is done, the figure consisting
of the regular pentagon and the two regular octagons immediately yields
the whole group diagram, by reflecting the octagon in the edges corr-
esponding to s3.

Since this reflection preserves non-euclidean angles, five octagons

always come to enclose a regular pentagon. As is easily proved, this
process may be continued indefinitely without resulting in overlapping
angles or incompatible labelling of edges. One sees immediately from
the group diagram thus obtained that the group in question is of infinite
order, as claimedo We also remark that the group diagram drawn by us was
not obtained in the manner just described, however it gives an approximate
conception of the actual figure produced by reflectiono
 24

10 0 Investigation of a certain class of groups

In connection with the infinite group just treated, we shall now
investigate an groups defined by generators and and
relations

where a and S are any integers. We shall show that this group
has infinitely many elements when the angle sum

a-2 213-2
--Tf + 2 • 2"i3 Tf ;;;. 2Tf.
a

Namely, in this case the group diagram has a non-euclidean regular

a-gon and two non-euclidean regular 2S-gons at each vertex. Thus
we construct the group diagram by choosing the side length of these
regular polygons so that the sum of the three angles round a vertex
equals 2Tf. By continued reflection of 2S-gon in those of its sides
which correspond to the operation s2' we obtain a group diagram of
infinite extent, so that our groups are infinite in the cases mentioned.

On the other hand, when the above relation, which we can bring into
the form,

2 (a + 13) .;; as,

is not satisfied, then the elements and and relations

a 2 13
sl s2 s
(s2 1) = 1 define only finite groups. If CI. = 2 or S = 2
then 2(a + 13) > as always and the groups are finite : if a = 2 the
group diagram is a polygon with 213 edges, denoted alternatively by
sl and s2' and hence we have a group with
213 elementso The groups with
13 = 2 are isomorphic to these, since one has only to set s2 s 1 = 0 1 to
obtain the relations
2 2 a
01 s2 (s2 0 1 ) 1

For a = 3 or 13 = 3 we have infinite groups when the inequality

6.;; 13 resp. 6';; a is satisfied, but we have to investigate the cases of
13 resp. a equal to 3,4,50 When a = 4 or 13 = 4 we get the inequality
4 .;; 13 respo 4';; a, consequently we have only to investigate the case
13 = 3 resp. a = 3 further. The same thing happens for a = 5 or S = 5,
since the inequality then takes the form 10';; 3S(a)0 On the other hand,
for a = 6 or 13 = 6 our inequality is satisfied for all values a > 2,
S > 2, so that only infinite groups result.
 25

Thus the only finite groups defined by generators and

and relations

are those for which one of the integers a,S has the value 2, in which
case finiteness is easily shown by construction of the group diagram,
and the groups with the pairs of values

a = 3, S = 3; a = 3, S 4, or a 4, S 3

a = 3, S = 5 or a 5, S = 3.
As we have already seen,

defines the alternating group A12 , which is isomorphic to the tetra-

hedral group. The pairs of values a = 3, S = 4 and a = 4, S 3
both give the same group, because if we introduce s2s1 as a new
generator si' then s s' =
s2 s 2s 1 ; and hence the relations
2 1
4
sl =
2
s2 = (s2sl)
3
=1 = s2 = (s2 s 4 l.
may be replaced by s,3 i)
1 2
The group obtained here is the octahedral group, which indeed is iso-
morphic to the symmetric group ~24 on four symbols. The pairs of
values a = 3, S = 5 and = 5, S = 3,
which again represent a
a
523
single group because of the replaceability of sl = s2 = (s2 s 1) 1
by s,3 = s2 = (s2si) 5 = I, lead to the icosahedral group, or altern-
1 2
ating group A60 on five symbols.

11. Group diagrams of the second kind. By a kind of duality we

can now move from the group diagrams previously discussed to "group
diagrams of the second kind", which are very useful in certain circum-
stances. Again we begin with the example of the infinite group
2 5 4
defined by the relations sl s2 = (sl s2) = 1, whose group diagram
has a pentagon and two octagons meeting at each vertex. In order to
obtain the second kind of diagram from the diagram of this group prev-
iously constructed, we mark the midpoint in each of the regular polygons
from which the group diagram was constructed, and connect the midpoints
of neighbouring polygons by straight lines. All these connecting lines,
which bisect the polygon sides, form in our case the lightly drawn net
of triangles which we have spread over the old group diagram [Fig.7J.
We obtain triangles here because three polygons meet at each vertex
in the group diagram of the first kind in general, when n different
 26

edges issue from each vertex of a group diagram of the first kind, then
the group diagram of the second kind consists of a net of n-gons o
Each vertex of the group diagram of the first kind is enclosed by such
an n-gon, and each polygon midpoint in the group diagram of the first
kind is the origin of as many edges in the diagram of the second kind
as the polygcn has sideso

In our example, each vertex of the basic figure formed by the

non-euclidean regular pentagon and octagons is enclosed by a triangleo
Since eight such triangles must meet at the midpoint of each octagon, and
five must meet at the centre of each pentagon, the group diagram of the
second kind consists of a net of non-euclidean isosc8£s triangles in
21f
which the angle at the pentagon midpoint is S = '3 and the angles at
21f
the two octagon midpoints are a = EI The triangles are constructible
0

only in non-euclidean geometry, since the angle sum

2a + S = 2" 281f + 2S1f :0 1f is less than 1f Using such a non-
0

euclidean isosceles triangle , we can easily generate the whole net of

triangles, by repeated reflection in its sideso The resulting net of
triangles, which covers the whole non-euclidean plane, gives us the group
diagram of the second kind for our infinite group. The group is now
represented by associating each triangle with the element corresponding
to the vertex it encloses in the group diagram of the first kind. Then,
as one easily sees, the operation sl is represented by reflection in

the base, the operation s2 by reflection in the right side, and s;l
by reflection in the left side, of the triangle. One realises immed-
iately, by consideration of the figure, that reflecting twice in the
base returns one to the initial position, but that five reflections
in the righ~ side are required to do the same thing. The accompanying
fig. 16 shows the way in which individual triangles correspond to
group elements. One starts from an element denoted by 1 and writes
in each triangle the corresponding element. The operation has
been indicated by an arrow crossing the side where the corresponding
-1
reflection occurs; since sl = sl we omit arrows indicating reflection
in the base.

It may now be asked how one can use such a net of n-gons, uniquely
determined from the generators and defining relations, for the represent-
ation of finite [quotient] groups. To do this it is necessary to know
representations of the individual elements [coset representatives] of
the [quotient] group in terms of the generators. We then extend the
27

Fie. 16
 28

polygon net just as far as is needed to associate each element with

exactly one polygon. Since the number of [quotient] group elements
is finite, the associated polygons form a finite surface, possibly
of higher connectivity, ioe. possessing holes, with a boundary
consisting of closed curves. We call this finite surface, which con-
tains each element of the [quotient] group once and only once, the
fundamental domain of the [quotient] group*. Thus the fundamental
domain is a finite piece of an infinitely extended polygon net we shall
call an "infinite basis" for the finite [quotient] groupo Now let
F be such a fundamental domain and let R be the boundary of this
finite surface F; for the sake of simplicity we assume that F is
made of triangleso Now consider a triangle a, one edge of which lies
on the boundary R of the surface F, and apply to it the operation s
corresponding to passage across this edge. We then arrive at a triangle
a. on the surface [in the coset of as], one edge of which necessarily
l.
lies on the boundary R. Because i f this were not the case, application
-1
of the operation s would yield a triangle a' of F not coinciding
with the triangle a, whereas the element a. s-l = ass- l must be ao
l.
The same element of the group would then be represented by two different
triangles in the fundamental domain, contrary to hypothesis. Consequently:
in a fundamental domain the boundary edges correspond in pairs.

Fig. 17

*Nowadays we would say it is the fundamental domain of the normal subgroup

by which the original group is divided in order to obtain the quotient

[~ranslators note].
 29

In order to realise this correspondence geometrically, we shall

pull the fundamental domain together so that corresponding boundary
edges are unitedo We then obtain a closed surface in space, on which
each element of the group is represented by a polygon. For the six
elements making up the symmetric group E3! on three symbols, for
example, we have a fundamental domain of the form shown in the follow-
ing two figures

The arrows denote the operation s2' for which we must have s3 = 1,
2 2
while the operation sl' for which sl = 1, corresponds to passage
across the unmarked edges. The corresponding boundary edges are in
each case those cut by the same arrow. In order to reflect this corr-
espondence, we have drawn together the fundamental domain so as to unite
corresponding boundary edges thus obtaining the group diagram of E3!
in the form of a double pyramid with triangular base, or alternatively
the subdivision of the circumscribing sphere into six spherical
triangles. In general, of course, the surface in space will be much
more complicated than this simple example and usually of higher conn-
ectivity. It is obvious that in each case one can work backwards from
such a group diagram of the second kind to one of the first kind, by
the dualisation process we used in the first place.

13. Sufficient relations for the generators. Using the example of the
symmetric group E24 on four symbols, we shall now answer an important
 30

question concerning the definition of groups by generators o We already

know that each symmetric group can be generated by all the transpositions
with a fixed term. Consequently, we can choose the three transpositions

sl = (12), s2 = (13), s3 = (14),

which satisfy the equations

1,

as generators for our group L 24 0 In order to construct the group diagram,

we must set up a system of sufficient relations for these generators o
We first find the relations

However, these relations are still not sufficient, since they define a
group with infinitely many elements. This is easily shown by the group
diagram of the second kind, which is very simple here. We shall not go
into this, but we recommend proving this assertion as an exercise in the
theory. What we are really interested in is the problem of finding a
necessary and sufficient system of relations for the generators of a given
finite groupo

In order to be able to solve this problem more easily in general, we

first consider the special problem of finding sufficient relations between
the generators sl,s2,s3 of the symmetric group L240 We learned
previously that the symmetric group £24 is also generated by a trans-
position tl and a four-termed cycle t 2 , and that the relations
243
tl = t2 = (t l t 2 ) = 1

are necessary and sufficient to define it. Thus if we express tl and

t2 in terms of the new generators

(14) ,
for which
1,
then we have

and we necessarily have the following relations for sl,s2,s3

However, as we have remarked above, these relations are certainly not

 31

sufficient. This is already clear on the grounds that s2 and s3

as yet appear together only as s2s3' so that nothing has been said
about them individually. In order to find sufficient relations, we
shall represent sl,s2,s3 in terms of t l ,t 2 ; we then have

as one easily checks by calculation. If we now replace t l ,t 2 by their

expressions in sl,s2,s3' then the latter equations become

These relations, in conjunction with the previous ones, now constitute

a necessary and sufficient system for the generators sl,s2,s3. We
shall prove this assertion in a more general form o

Thus, let

be generators of a group, which we know to be completely defined by

generators

and p relations

t~ = ~~(sl,s2,s3,.oo,sn) (~ = 1,2, •.• ,m)

then the relations between t 1 ,.oo,tm necessarily imply the relations

between If these relations are not yet sufficient, we

represent in terms of obtaining

If we replace the t l ,t 2 ,oootm here by their expressions in sl,s2, ••• ,sn'

then we obtain the following additional relations for our new generators:

The p+n relations

now give us a system of necessary and sufficient relations for the generators
 32

To prove this we need only show that each relation possible

between sl,···,sn in the given group follows from our system. Suppose

Cl l Sl 01 Cl 2
S2 O2 Cl S 0
r r r
R = sl s2 s
n sl s2 s
n sl s2 s
n
=1
is any relation between sl,s2' ••• 's in our group. If we replace
n * *
sl,s2, ••• ,sn by their expressions f 2 (sl,·.·,sn) , ••• ,fn(sl, ••• ,sn)'
then we obtain
*Cl l
R = fl

01
1jI n (<p 1 ' 0 •• , <Pm)
o
Wnr(<Pl,<P2,···,<Pm)

Cl l Sl 01
ljIl (t l ,···,tm)1jI2 (tl,···,tm)··· ljIn (tl,···,tm)
o
ljInr (t l ,t 2 ,··o,tm)

But now, all relations in the group between t l ,t 2 ,oo.,tm are con-
sequences of the p relations grr(t l ,t 2 ,.o.,tm) = 1.
We therefore have
Cl l 0
1jJl (tl,··,tm) .•• 1jJn r (tlP.otm)

and our assertion is proved.

13. Groups with more than two generators

A few remarks remain to be made about the smalle&number of generators of

a group. We have been able to generate all the finite groups considered
up to now by two elements, but it is easy to give examples of finite
groups for which at least three generators are
required. These include, in particular, commutative or abelian groups.
As generators of such an abelian group we choose three elements sl,s2,s3
and subject them to the six relations
 33

This group obviously cannot be generated by less than three elements,

because if we choose two elements as generators, then either one
generator is missing entirely, or else two of the elements sl,s2,s3
appear only in combination, from which they cannot be determined
individually.

Now our abelian group also has the special property, because of the
exchangeability of the generators, of being decomposable. In general,
a "decomposable group" is defined as the set of all products from
two or more groups, where each element of one group commutes with
each element of any other group. In our example the group decomposes
into three groups, with the elements 1, sl' resp. 1, s2' resp. 1, s3'
each of which is of order 2 and generated by ~ element. We have
the important theorem that : when a groul2 G is decom)22sable into k
groups Gl , G2 ,··· ,G k , then G cannot have fewer than n1 + n 2 + ... +nk
generators*, and it has order °102··· 0k' where n.1. is the minimum
number of generators of G. and 0. is its order, i = 1,2, .•• ,k.
1. 1.

The group diagram of our abelian group of eighth order is a cube,

whose eight vertices correspond to the eight elements, and the group diag-
ram of the second kind is a regular octahedron, whose faces correspond
to the individual elements.

2 2

3 3

We mention in passing that Holder has given an example, in the

Mathematischen Annalen of 1893, of an indecomposable finite group which
cannot be generated by two elements.

While there are finite groups which require more than two generators,
they are nevertheless always subgroups of a group with two generators,

* This is false. E.g. ~2 x ~3 =~6 can be generated by one element.

(Translator's note.)
 34

namely the symmetric group. We can prove the theorem that

Each finite group Gn of order n is isomorphic to a subgroup

of the symmetric group r nl of order n!

We denote the elements of Gn by

and introduce the notation

for the product of two of these elements. The element a. can now be
~

associated with the substitution

which replaces the indices of

(~i
2
2. 3 ..
~
3

~ :J
a l ,·· .a n by those of alai' a 2a i ,··· ,ana i
respectively, where the latter sequence results from multiplication by
ai and hence also contains each element of Gn • In the same way, the
element a. is associated with the permutation
J

(~j
2
2. 3.
)

To prove our theorem, we now have only to show that

3

) :J
ei
2
2. 3.
~
3

~
. n )
• ni 3.
Cj 2j J
2 3

:J C 2 3
1 tij 12 (ij) 3 (ij)

or, since the left hand side equals

("i:' )
2 3 • n
(2i 1 j (3.) . . (n.l .
~ J ~ J

that
(k.l. k(ijl
~ J
But now (ki)j is the index of (akai)a j and k(ij) is the index of
ak(aia j ), and hence the associative law implies

(a a.la. or
k ~ J
whence the correctness of our assertion, which has been extensively used
by Frobenius and Netto in a series of important investigations.
 35

A situation similar to that with finite groups, where the subgroup

may need more generators· than the group itself, also can be observed in
infinite groups. As an example, we consider the infinite group with
generators sl and s2' and no relations between them. In the diagram
of this group which consists of infinitely many vertices, from each of
-1 -1
which issue four edges, corresponding to the operations sl,s2,sl ,s2 '
there is no closed edge path. As subgroup of this infinite group we
take the set of all elements resulting from sl by composition and
transformation [conjugation] by arbitrary elements of the group itself.
The elements of this subgroup, which is normal, cannot be generated by
finitely many elements. We shall simply indicate the route to a simple
proof of this assertion by means of an example. If we choose, e.g.,
-1 -2 2
s2 sl s2 and s2 sl s2 as generators, then they can generate only
elements of the form
-1 Al -2 2 ~l
(s2 sl s2) (s2 s l s2 l

so that appears to no exponent greater than two. The element

-a a
s2 s l s2' where a > 2, can therefore not be generated.

14. Hurwitz' investigations. To conclude this chapter we shall now

give, in a form somewhat modified to suit our purpose, a brief survey of
the results published by Hurwitz in the Mathematische Annalen*.
Hurwitz gives a quite specific rule for the construction of the fundamental
dornaJn of a finite group from the infinite basis**. If

are the generators of the finite group, then Hurwitz introduces a new
generator sn+l by
-1 -1
= 1 or sn+l = sn sn_l
so that the product of all generators is equal to 1. When the product
s l s2 •.• sn is already 1, then of course it is unnecessary to add
sn+l. In any case, since our generators yield a finite group, equations
of the form
rn
n
s 1
n

*Vol.41(1893) 403-442 **This idea is actually due to Dyck, Math. Ann 20(lR82)
1-45 (Translators note ).
 36

must hold, where ml,m2, •• ,mn,mn+l are finite integers, in addition to

the relation
sls2 ••• snsn+l = 1

If the operations sl,s2, ••• ,sn,sn+1 have to satisfy only these

equations then they generate an infinite group, whose group diagram of
the first kind may be very easily constructed. At each vertex we need an
m1-gon with edges 51' an m2-gon with edges s2"'" an mn+1 -gon with
edges sn+1 and between any two of these polygons in succession, an
(n+1)-gon with edges sl,s2, ••. ,sn+1' whose two edges at the given vertex
are shared with the neighbouring polygons.

Fig.20

In our figure we have replaced these polygons by arcs which return to the
initial point, and inside each arc we have inscribed the number of edges
of the polygon. If these polygons are to be regular then of course they
are only attainable in non-euclidean geometry, apart from the simplest
cases.

If we now consider the point B, at which we arrive from A by

means of the operation sl' then an ml-gon and an (n+l)-gon are
already attached, next to each other, at this point. We must therefore
fill the remaining angle around B with (n+1)-gons which alternate with
an m2-gon, and m3- gon, .•• , an mn+l-gon, i.e. a polygon net exactly
the same as the one which fills the angle around A. The situation is
similar at all other vertices, and it is clear how to continue the
construction.

We can now proceed from this group diagram of the first kind to the
group diagram of the second kind, by connecting midpoints of neighbouring
 37

polygons,whence the non-euclidean plane becomes covered with a net of

(2n+2) - gons. If further relations are then given between the generators,
and if the resulting group is finite, then we know how to mark off in this
net a fundamental domain and join together its corresponding boundary
sides. We then obtain the group diagram of the second kind for the finite
group in the form of a closed surface, which in general will have higher
genus. By dualisation we can convert this group diagram of the second
kind into one of the first kind on the surface.

We now derive a simple relation between the order of the group and
the genus of the surface resulting from the "Hurwitz basis", which
carries the group diagram. Our starting point is the Euler polyhedron
formula, already known from school :

If is the number of vertices, "1 the nomber of edges, and

a2 the number of faces of a convex polygon, then

However, this formula holds only for polyhedra which can be continuously
deformed into a sphere. For surfaces of higher connectivity, which have
several holes, continuous distortion into a sphere is no longer possible.
In this case we can obtain a simply connected surface by p cuts through
the holes [and closing off the resulting perforations]. The cut open
surface can be continuously deformed into a sphere, and hence the Euler
polyhedron formula holds for it. But in cutting we have increased the
number of faces by 2p, while changing the numbers of vertices and edges
by the same amount, so that the number remains the same.
Hence for a surface with p holes we have

2 - 2p,

assuming that the formula is correct for the sphere. In order to prove
the latter, we show that the left hand side of this equation is independent
of the subdivision of the surface into polygons. We can obviously convert
one subdivision into another by either dividing edges in two [by new
vertices] or by dividing faces in two by new edges. In the first case we
we have one edge and one vertex more, in the second case one face and one
edge more. The alternating sum
 38

Cl O - Cl l + Cl 2 therefore remains unaltered in both cases. Now for

the simplest subdivision of the sphere, say by 2 caps (Ci O = Cl l , Cl 2 2)
the formula Cl O - Cl l + ct 2 = 2 is immediate, and hence it also holds
in general.
The important formula

2 - 2p,

which relates the numbers of vertices, edges and faces of a subdivision

to the genus of surface resulting from continuous deformation, now easily
yields the Hurwitz theorem on the order of a group. If we consider
the subdivision as the group diagram of the group generated by the
operations sl,s2, .•. ,sn+l' then the number of vertices, Cl O' is the
number of elements or the order I~ of the group, since 2 (n + II edges
issue from each vertex, the number of edges, Cl l , is N(n + 1). Finally,
the number of faces is obviously equal to the sum
n+l
N N
ml
+
I.J
m2
+ ... + + .!:!.-. (n +
n+l
1) Nel + L
1
m.
l.
mn+ l i=l l.

Our polyhedron formula therefore gives us the equation

n+l
N - N(n + 1) + N(l + L .l..) 2 - 2p
m.
i=l l.
or
n+l 1
N{ L (1 - -} - 2} = 2p - 2.
i=l mi
This formula now permits all kinds of conclusions to be drawn concerning
the order of the group, when we investigate individual surfaces in terms
of their genus.

Suppose first that the group diagram is represented on a surface

of genus p > 1, so that when n > 3 we have the equation
n+l
n-3 ;;. M
2p - 2 N{ L (1 - .l.. I - 2} ;;. I.J ,
m. 2 2
i=l l.

since 2 is the smallest possible value of m .• But it then follows that

l.

IJ
4

and hence the genus of the surface increasffiwith the order of the group.
On the other hand, if we have n = 3 then not all the generators
 39

sl,s2,s3,s4 can have order 2, since p is to be > 1. If we take

one of them to Be of order 3 and the others of order 2, then we obtain
the equation
2 3 N
2p - 2 N( "3 + "2 - 2)
6
or
p = 1 +
N
12'
and hence for all possiBle values of ml ,m 2 ,m 3 ,m 4 the genus of the
surface must s·atisfy
Ji..
12
Finally, we still have to consider the case n = 2, for which we have
the equation

since p > 1, no two of the three numbers ml ,m 2 ,m 3 can have the

value 2, otherwise the left hand side would be negative. Also, they
cannot all be 3, since p is to be greater than 1. For the same reason,
when one of the three numI:>ers, say ml , is 2, we must still exclude
the case ml +...l- ;;.~, which occurs for pairs satisfying m2 ';;; 4,
2 m3
m3 .;;; 4 or m2 ';;; 3, m3 ';;; 6. We have therefore established that when
P > 1 and n = 2 the orders cannot be such that two of them
have the value 2, or such that one has the value 2 and the others the
values 3 and 3, or 3 and 4, or 4 and 4, or 3 and 5, or 3 and 6, or
finally such that all three have the value 3. The smallest possible
values of ml ,m 2 ,m3 for p > 1 and n = 2 are therefore

7 2) ml = 2, m2 = 4, m3 = 5 ;

4 for which the corresponding values of p

are 1 + 1L
40
1 + 1L
24
Thus, if P > 1 then it also satisfies
the inequality
p;;' 1
N
+ 84'
which gives us the Hurwitz result on the order of the group and the genus
of the group diagram.

Now that we have dealt with the surfaces of higher connectivity,

we once again consider the case p = 0, in which the polygonal sub-
division is on a convex surface deformable into a sphere. Here, our
general formula yields the equation
 40

n+l 1 n+l 1
2
,Ii ( Z (1 - - ) - 2) -2 or L (1 - - ) 2
i=l mi i=l mi IJ

Since m. 2 is the smallest possible value for the order of a

~

generator, we certainly have

n+l
2 - 2
= E (1 -~) ;;. n+l or
4
n';; 3 - -< 3
,Ii m. 2 IJ
i=l ~

Thus the only possible values of n are

n = 1 and n 2.

If n = 1 we have
1 1 2 1 1 ~
2 2 or - + -
ml ffi2 N ffi l m2 N

whence
2m l ffi 2
N =
ml + ffi2

Now suppose ffi2 is the greater of the two numbers ffi l and ffi 2 , so
certainly

2 2
and
N N

whence
ffi l .;; M';; ffi2 •

On the other hand, since the order of the group itself ffiUst be greater
2
than or equal to ffi 2 , since it contains the m2 elements s2,s2' .•.
we necessarily have

The equation 2 then yields

so that in this case we have

Consequently, our group is defined by

N M
sl s2 s l s2 1 ,

where the relation is in any case immediate froffi since

s~ = s11.1 = 1. The finite group obtained is therefore the "~.lic group"

well known to us, or the "polygon group", whose group diagram can certainly
be drawn on a sphere
 41

Fig. 21

Now we have to look at the case p o and n 2.

Here
2
2 1 + l
N N

In order to investigate what cases are possible, we suppose that the ord-
ering of ml ,m 2 ,m 3 is

We then have the following inequality for ml

;;. 1 + l or m < 3,
N 1

since N has to be a positive integer. We therefore have 2 and

hence
1 1 1 2
and - + - = - + -
m2 m3 2 N
From the inequality

1.2 + 2
N
 42

we find that
4N
m2 ";;; N+4 < 4,

so that only the values 2 and 3 come into consideration for

The equation
2
1 + -
N
N
then enables m3 to be calculated easily, and indeed m3 when
2
6,IJ 72 6
ml m2 2 and m3 = N+l2 6 6 - -,::r- when
N+l2
12+1
ml 2 and m2 3. In the latter case m3 is an integer only i f

~ + 1 equals 2,3 or 6, since i~ certainly has to be greater than

zero. These three possibilities correspond to the values m3 = 3,4,5,
for which the ,J values are N 12, 24, 60 respectively. Thus p = 0
the only possibilities for a group with three generators 5 1 ,5 2 ,5 3
connected by the relation
5 1 5 25 3 1
are
2 2 m3
1) 51 52 53 1

2 3 3
2) 51 52 53 1 N 12

3) 2 3 4 1 N 24
51 52 53
2 3 5
4) 51 52 53 1 N 60

The first case is the dihedral, or double pyramid, group which is isomorphic
to the group of rotations of a double pyramid whose base is a regular
m3 - gon. The operations 5 1 ,5 2 ,5 3 respectively represent the two rotations
which take one apex to the other and the rotation about the axis between
the top and bottom apex. The remaining three cases lead us back to the
frequently mentioned groups of the regular polyhedra, namely the tetrah~dral,

octahedral and icosahedral groups.

Thus we have completely settled the case p = 0; it has led us

only to finite groups. The remaining case p = 1 gives only infinite groups
as "bases", whose group diagrams are easily constructed, since they involve
certain regular subdivisions of the euclidean plane.
 43

In the case p lour polyhedron formula becomes

n+l 1 n+l 1
L (1 - --) - 2 = 0 or E (1 - ;- l = 2,
i=l mi i=l i
the interesting aspect of which is that the order N of the group drops
out entirely. Since 2 is the smallest possible value of the mi , the
number n satisfies the inequality

n+l .;;; 2 or n .;;; 3,

2
As one sees immediately, the equality sign here is only possible when

we shall look at the corresponding group again shortly. In the case n = 2

we have the equation
1 1 1 111
3 ---- 2 or --+--t--= 1
ml m2 m3 ml m2 m3
whose only solutions in integers ml ,m 2 ,m 3 are the triples
 44

as is easily checked. A detailed treatment of the corresponding group

given in the twelfth chapter of Burnside's Theory of Groups.

The minimal value of p for which the group diagram can be

drawn on a closed surface of genus p is called the "genus of the group".
Here there is a certain arbitrariness in the Hurwitz basis, since for
certain groups one can find a representation on the sphere, whereas the
Hurwitz method gives a genus p > O. We shall now show a simple example
of this.

We construct the group diagram for the infinite group with the
four generators and the relations

The group diagram consists simply of rectangles which cover the whole plane.
Since each vertex is the origin of as many edges as each cell has sides,
our figure can be viewed simultaneously as a group diagram of the first
and second kind. The first point of view takes the numbers on the edges as

3 1 3 1

2 4 2 4
1 3 1 3 2

4 2
4 2 4
3 1 3 1

2
4 2 4
2
1 3 1 3

4 2 4 2
4
3 1 3 1
Fig. 23

operations by which one proceeds from one vertex to another, the second
takes the operation to be the process of going from the cellon one side
of the edge to the cellon the other. This diagram is the Hurwitz basis
of the the finite abelian group GS given by the generators sl,s2,s3
and the relations
 45

1.

Because if we introduce the generator s4 by the equation

then in fact
2
s4 1

so that a fundamental domain for the abelian group Ga may be found in

our group diagram. This fundamental domain is made up of eight rectangles,
and the corresponding edges on its boundary are those with the same labels.

3 1 3

Fig. 24

By uniting the corresponding boundary edges one obtains the group diagram
on a closed surface which has the form of a ring. While this

Fig. 25

gives the genus p = 1 for the group Ga , we have seen earlier that we can
represent it in a much simpler way by a cube, and hence by a surface of
genus p = o.
 46

We have now covered the most important aspects of the geometric

representation of groups. Previously, these representations were used
chiefly for intuitive support, but now they prove to be almost in-
dispensable in the treatment of infinite groups, and consequently of
increased significance.
 TRANSLATOR'S INTRODUCTION 2

The following chapter of Dehn's lecture notes deals mainly with

surface topology, presenting several of the results which were

eventually published in Dehn [1912a). The account here is different

however, giving more details on non-orientab1e surfaces, and giving

different proofs of his main result, the solution of the conjugacy

problem (or "transformation problem" as Dehn calls it). These proofs

greatly illuminate the curious solution given in Dehn [1912a), which is a

combinatorial algorithm but justified by appeal to the hyperbolic metric,

revealing it as a transitional stage between the metric solutions and

the purely combinatorial solution given in Dehn [1912b).

The second proof given in the lectures is actually simpler than

the first, being an easy consequence of some ideas in Poincar~ [1904),

a paper to which Dehn refers in [1912 a,b). Poincar~ takes the standard

universal covering F of a surface F of genus p > 1 by a net of

4p-gons in the hyperbolic plane, and lifts any non-contactib1e curve c

on F to a chain of curves ••• , c(-l) , c(O) , c(l) •..• on F, where

the origin is the initial point of ~(O) and final point of ~(i) =

initial point of ~(i+l). The endpoints R,S of this chain on the

boundary circle aF are the fixed points of the motion, corresponding

~(i+1)
to c, which maps e(i) onto c , and the hyperbolic line

connecting R,S is the axis A(c) of the motion (Fig. 1).

(Instead of the projective model used by Dehn in the previous paper,

I am now using the more familiar conformal model, in which "lines" are

circular arcs orthogonal to the boundary.)

 48

F
aF

Fig. 1

~(O)
Any lift in the chain, say c , can be deformed onto a

displacement segment o(c) of the axis while keeping the position of

its ends equivalent, thus there is a corresponding transformation of

c on F to the projection a(c) of o(c). Moreover, any curve

transformable into c is transformable into a(c), and conversely,

hence a criterion for the transformability of one curve c into another,

d, is that a(c) = a(d). The curve a(c) is called the geodesic

representative of the (free) homotopy class of c.

The important point is that a(c) is in a reasonable sense

computable from c, so that one has an effective algorithm for solving

the transformation problem. To compute a(c) one simply collects the

segments of 8(c) from different polygons in the net and assembles

their equivalents in a single polygon, e.g. as in Fig. 2.

 49

Fig. 2

The curve aCc) then corresponds to a sequence of directed arcs in a

polygon, and is determined by the endpoints of these arcs on the

polygon edges, which in turn are computable from the ends R,S of

the axis ACc) and the position of the polygon net. Such calculations

are not difficult in principle, nor do they seem to be complex in the

sense in which one measures computational complexity nowadays, but it

seems that the first and only time they were done was in the thesis of

Dehn's student Gieseking in 1912. For more information on this work, and

some explicit matrices for generating the coordinates of the net, see

Magnus [1974].

Incidentally, the geodesic representative aCc) also gives a

criterion for the transformabi1ity of c into a simple curve; namely,

this is the case if and only if aCc) is simple. This was Poincar~'s

reason for looking at aCc).

 50

Dehn's other solution of the transformation problem uses

concepts from hyperbolic geometry, which also explain how to

determine the ends R,S of the axis A(c). The motion with fixed

points R,S which maps c(i) onto c(i+1) and A(c) onto itself

is a displacement in the sense of hyperbolic geometry and therefore

it maps each curve at a constant hyperbolic distance from A(c) (or

distance curve of A(c» into itself. Conversely, any sequence of

points ••• P- 1 , PO' P1 , ••• such that Pi is sent to Pi +1 by the

displacement must lie along a distance curve. In particular, this

holds if

Pi final point of c(i)

initial point of c(i+1) ,

hence by taking three such points in succession we can construct the

distance curve through them, and hence find R,S as its intersections

Dehn then decides whether two curves are transformable into

each other by comparing distance curves rather than axes. This seems

awkward and roundabout, particularly since he still needs the axes for

his calculations, but it points the way to the next step, in Dehn [1912aj,

where axes and distance curves themselves are no longer needed, only

certain properties they induce in the net of polygons, and finally to

Dehn [1912bj, where all metric notions are completely eliminated.

REFERENCES

M. Dehn [1912aj: Uber unend1iche diskontinuier1iche Gruppen.

Math. Ann. 71, 116-144.

[1912bj: Transformation der Kurven auf zweiseitigen F1~chen.

Math. Ann. 72, 413-421.

 51

H. Poincar~ [1904]: Cinquienne complement a l'ana1ysis situs.

Rend. eire. mat. Palermo 18, 45-110.

W. Magnus [1974]: Noneuclidean Tesselations and Their Groups

Academic Press.
 LECTURES ON SURFACE TOPOLOGY

In the previous section we have discussed the generation of groups

by given operations, and we have repeatedly come up against infinite

groups thereby. In this chapter we shall now deal exclusively with such

groups, and use them to make a series of important applications to

analysis situs or topology.

§1. Closed surfaces

1. Preparatory theorems from topology. Before we treat the infinite

groups themselves we must first get to know some preparatory theorems

from topology. As we have already mentioned in the introduction, we can

cut the simple ring [torus] along two curves a and b, one of which runs

around the inner hole and the other across it, so that the torus surface

can be spread out as a rectangle. The edges of this rectangle ABeD

correspond to the circuit cuts a and b, in fact each pair of opposite

edges corresponds to one of the two cuts a, b. If we give each of the

cuts a, b a definite direction, and denote the corresponding

-1 -1
opposites by a b , then a complete circuit of the rectangle ABeD
corresponds to traversal of the cuts of the torus surface in the order
-1 -1
a,b,a ,b • We must therefore denote the four edges AB, BC, eD, VA
-1 -1
of the rectangle by a,b,a ,b respectively.

Since our rectangle can obviously be regarded as a map of the

torus surface, each curve r on the to~us can be mapped onto the rectangle.

At the places where r cuts the curves a and b its image will stop
 53

at a point on an edge of the rectangle, then continue at the

corresponding ("homologous") point on the opposite edge. Consequently,

a continuous curve on the torus appears on the rectangle divided

into several pieces, whose endpoints lie on the edge in such a way

that the final point of one piece and the initial point of the next

are homologous points on opposite boundary edges. Conversely, when

there is a system of curve segments given in the rectangle, pasting

the rectangle together to form a torus brings homologous points on

opposite edges together and the result is a continuous curve on the

torus. This curve is then closed on the torus when its initial and

final point coincide in the rectangle or, if they lie on the boundary,

when they are at homologous points on opposite edges.

Fig.2

b- l

-1
a
a

b 6

Fig.3
 54

The mapping of curves on the torus becomes somewhat clearer

when we no longer represent them by pieces, but instead return to

continuous curves. To do this, a curve which meets an edge of the

rectangle is continued, not at the homologous point on the opposite

edge, but in a new rectangle on the other side of the edge being

crossed. In this way we obtain a net of infinitely many rectangles,

in which the individual pieces join together again into a connected

curved path. This net of rectangles covers the plane simply and

without gaps; four rectangles meet at each of its vertices, each

-1 -1
rectangle has the four sides a,b,a ,b ,in the same order in each

rectangle, and each vertex of the net is the origin of four segments
-1 -1
a,b,a ,b • Geometrically, the attachment of new rectangles is a

multiple covering of the torus, the individual sheets of which meet along

the curves a and b. If we are given a curve in the net of rectangles,

then we can immediately find the corresponding curve on the torus by

pasting together. The latter curve is then closed when the initial and

final point of the image curve in the net correspond. This

correspondence can indeed be set up arbitrarily in the construction of

the net, but generally it is natural, for reasons of convenience, to let

geometrically homologous points correspond.

We shall now similarly consider curves on the double torus. We

cut this surface in the way already known to us along the four circuits

a,b,c,d which have the positions shown in the figure. By spreading out

the cut surface we then obtain an octagon as elementary surface piece,

and we draw it as a regular octagon. With suitable choice of direction

 55

-1
a

c a

Fig.S

-1
c
 56

for the cuts a,b,c,d, the sides of the elementary octagon become the
-1 -1 -1-1
sequence a,b,a ,b ,c,d,c ,d .

One finds this orientation when one traverses the cuts a,b,c,d

on the surface in such a way as to avoid crossing the point O. In

general, one sees that a closed surface of genus p can be spread out as

a plane elementary surface piece after cutting along 2p circuits

aI' b l , a 2 , b 2 ,···, a p ' bp ' with a common point 0 and running alternately

through the holes of the surface and around them. This elementary

surface piece can be presented as a regular polygon of 4p sides, denoted

-1 -1 -1 -1
in order by aI' b l , aI' b l , ... ,ap ' bp ' a p ' bp ' For this purpose

one chooses the orientation of the curves a. and bi' where a. runs
~ ~

through the ith hole and b.~ runs around it, so that a circuit of the

polygon corresponds to a traversal of the cuts ai' b.~ on the surface

without crossing other cuts.

Each curve on the double torus which cuts none of the curves

a, b, c, d corresponds to a curve on the elementary surface piece. An

arbitrary curve on the double ring is represented by separate curve

pieces in the octagon, bounded by the points where the curve meets the

circuit cuts a,b,c,d. The boundary points are such that the final

point of one piece and the initial point of its successor are homologous
-1 -1
points on one of the edge pairs a and b ,c and c
-1
d and d • One can again replace these separate curve pieces by a

connected curve if one places a new octagon alongside each edge of the

original, with the same labelling and orientation. In the resulting net of
 57

octagons, eight polygons meet at each vertex. We must therefore set

things up so that octagon angle equals ~n = 45°. It follows that our

net of octagons can only be represented in the non-euclidean plane,

because there one can let the angle of a polygon take an arbitrarily

small value. The accompanying figure shows how the eight octagons lie

round a particular vertex in this non-euclidean net; the curved lines

are to be replaced by inscribed octagons.

Fig. 6

This net of octagons admits a representation of curves on the double

torus similar to the representation of curves on the torus in the net

of rectangles. Each continuous curve on the surface maps to a continuous

curve on the net. Conversely, each curve on the net may be carried to

a curve on the double torus. The result is a closed curve on the

surface when the corresponding curve on the octagon net has homologous

initial and final point. Thus the net representation gives the means

of immediately deciding between closed and non-closed curves on the

surface. Naturally curves which are closed on both surface and net play
 58

a special relle. These closed curves have the property that they may

be continuously contracted to a point. Certainly, such a curve on the

net may be reduced to a point, but then each continuous movement on the

net corresponds to such a movement on the surface. Now, genera1ising

these considerations from the double torus to the surfaces of genus p

and the corresponding 4p-sided polygons which are their elementary

surface pieces, we obtain the following topological theorems:

1) Each curve on the surface which meets none of the circuit cuts

or corresponds to a curve within the elementary surface piece

which represents the surface. An arbitrary curve r divides into

segments bounded by the points where r meets the ai' b i ; each such
segment corresponds to a segment on the elementary surface piece

connecting two boundary points, in such a way that the final point of one

segment and the initial point of its successor are homologous points on

segments

2) If one constructs a net in the non-euclidean plane consisting

21T
of 4p-sided regular polygons with angles 4p , and labels the edges of

these polygons, which cover the plane simply and without gaps, in the same

way and orientation as the elementary surface piece, as is always

possible, then each continuous curve r on the surface corresponds to a

continuous curve G on the net. If the curve r on the surface is closed

then the endpoints of the corresponding curve G are ho10mogous points

of the net, and conversely, when the endpoints of the net curve G

have homologous positions, the corresponding curve r on the surface is

closed. A closed curve on the net corresponds to a closed curve on the

surface which is contractible to a point.

 59

The latter theorem may also be extended somewhat:

Each continuous transformation of a closed surface curve into another

may be assembled from separate operations in which at least one point

of the curve remains fixed.

Because, if r is a given closed curve on the surface we can

fix a point A on it and then continuously deform it into another

curve which has this point A in common with r and which also goes

through an arbitrary point B of the surface. In the same way, the new

curve can then be deformed into another while the point B remains fixed.

In our polygon net the curve r corresponds to a curve G = AIAZ whose

end points Al and AZ are homologous net points which correspond to

the point A The curve G is then deformed, first into another curve G'
which has initial and final points Al and AZ in common with G
and which goes through the point Bl [corresponding to the point B
(Translator's note)). If we choose B as the fixed point of the curve,

then we must replace the piece AlBl of G' by the homologous piece

AZB Z beginning at AZ' and we then have G' = BlB Z' The curve G'
is then continuously deformed into the curve G* which

Fig. 7
 60

has the endpoints Bl and B2 in common with G'. The net curves

GandG * t h en represent two curves r and r * on the surface which

are continuously deformable into each other.

2. Fundamental group of the surface of genus p

After these somewhat brief topological preparations, we now go to the

fundamental groups of surfaces. The generators of the fundamental group

of the surface of genus p will be called

a , b
p p

for which a single relation holds:

The group diagram of the group so defined is immediately seen to be

equivalent to the net subdivision of the non-euclidean plane by

4p-sided regular polygons, or when p = 1, to the corresponding square

net subdivision of the euclidean plane. The edges of an individual

-1 -1 -1 -1
polygon are labelled by al,bl,a l ,b l , ..• ,ap,bp,ap ,b p , and in this

order for all polygons. Issuing from each vertex of the polygon we have

constructed are 4p edges, with the labels

-1 -1
a ,b ,a ,b . Also, the single relation between the generators is
p p p p
satisfied at each vertex of the net, and each other r6lation resulting

from the net is a consequence of this single relation, since each closed

net curve may be put together from single polygons. Thus our polygon

net is indeed the group diagram of the fundamental group of our surface.
 61

With this group diagram we have already gained a great deal.

First, we can immediately dec.ide whether two given elements A and

8 of our group are identical or not. They are identical, i.e.,

AB-1 = 1, if and only if the points corresponding to them in the group

diagram coincide. Thus the construction of the group diagram solves

the most important problem, which we shall call the identity problem

[now known as the word problem (translator's note)]; this is of great

significance, since the equality of two elements of the fundamental

group, given as products of generators, cannot be determined from the

relation between ai,b i without extra work. Now, since our group

diagram coincides with the mapping of our surface as a regular polygon

net in the non-euclidean plane, we can also regard the vertices

corresponding to elements of the group as the endpoints of net curves.

If now the elements A and B are identical, or if AB- l 1, then

the edge paths leading to them from the identity vertex form a closed

curve in the net. Thus the identity problem is equivalent to the

question of "null reduction", Le., to the question of whether a closed

curve on the surface may be continuously contracted to a point.

It is significantly more difficult to solve the transformation

problem; its solution answers the question of whether two given elements

A and B of the group can be transformed into each other by a

third element C of the group, 1. e. A = C-lBC [hence the current

term, conjugacy problem (translator's note)]. This question proves to

be equivalent to the deformability of closed curves rand r' on

 62

the surface into each other. In order to see this, we first deform

the curves rand r' into curves which go through a point 0

of the surface, and then replace them, by an inessential displacement,

by curves which follow the 2p circuit cuts. The possibility of such

continuous transformation of all closed curves on the surface follows

immediately from our last theorem in the topological preparations.

Now when we choose the point 0 as initial point, our new curves
correspond to certain edge paths in the polygon net , with their endpoints

at vertices. Now if G = AlA Z is such an edge path, with endpoints

Al and AZ' then when we choose another vertex Bl as the initial

point of G we must attach to the point AZ a piece AZBZ homologous

-1
to AlB l , and we then obtain the edge path Bl BZ = H GH, where

H= AlB l . Thus by choosing a new initial point we have arrived at a

transformed edge path, which we now have to reattach to the initial

point AI' We then obtain the two edge paths G = AlA Z and K = AlBi'
where G is transformed into K as a result of the equation

A1=B'1 Fig. 8

B'z
 63

Now we can regard the vertex Al as the identity element in the

group diagram of our fundamental group, then the point A2

corresponds to an element A and the point Bl to an element C,

,
and finally the point B2 corresponds to an element B, so we have

Now since the curve K results from G by continuous transformation,

we have shown that the transformation problem for the elements of

our group corresponds to the question of when two closed curves on

the surface are deformable into each other. Here we again notice

the close connection between the fundamental group and the surface,

to each element of the group corresponds a curve on the surface, so

that we can almost represent the group by the surface itself.

3. Solution of the transformation problem. While the identity problem

is immediately solved by the construction of the group diagram, a

complete solution of the transformation problem requires a deeper

investigation of the group diagram. The simplest diagram of a

fundamental group that we have seen is the square net in the

euclidean plane. Here we have only two generating operations, a and b,

with the relation

-1 -1
aba b = 1.

We notice immediately from the validity of the commutative law

ab = ba that our group is abelian and consequently representable

symbolically by
 64

in which m and n are positive or negative integers.

Now let

be two elements which are transformable into each other by the

element

then we have
~~
a b

or, because of the validity of the commutative law,

But then it follows by consideration of the diagram that

and

Thus, in the fundamental group of the torus, two elements are

transformable into each other if and only if they are identical.

Thus the transformation problem here is trivial and without

beauty.

Our problem becomes much more interesting in the general case.

Here the relations between the points A2 and B2 (Fig. 8) have to

be investigated. We shall now denote the polygonal path AIA2

by A, and the polygonal path AlB; by B, and we shall represent

both by straight lines for the sake of simplicity. If we now move

 65

the net onto itself in such a way that the point Al goes to

A2 and each edge goes to a like-labelled edge, then the point

A2 goes to A3 ' A3 to A4 ' A4 to AS' .•. , AO to AI'

A_I to AO ' ..• and all these points are at distance A from

each other; they constitute a chain of equal sides, which we

shall call the A-chain for short. The angles between successive

sides in this A-chain are equal, since the motion of the net

carries the point Ai to Ai +l and hence brings the triangle

Ai _l Ai AHI into coincidence with Ai Ai+! AH2 . But i t

follows from this that these points Ai all lie on a non-euclidean

circle, whose midpoint M can be found, say, by bisecting the

angles at two points A1..• This midpoint

A
A

proves to be ideal, i.e. it lies outside the non-euclidean plane.

If it were real, then the motion of the net, which is obviously a

rotation about M, would bring a netpoint sufficiently close to

M onto an arbitrarily close point. However, this is impossible,

since a non-identity motion of the net sends each point inside a

polygon to a homologous point in another polygon, and two such

homologous points are always at a finite distance from each other

[greater, e.g., than the length of a net edge (translator's note»).

For curves on the surface, this means that there are no infinitely
 66

small closed curves which cannot be contracted to a point. Since

the circle in which the A-chain is inscribed has an imaginary or

ideal midpoint, it must meet the absolute, i.e. the boundary circle

of the non-euclidean plane, in two distinct points Rand S.

This implies, by the way, the important property that no power of

A can equal 1, without A itself being equal to 1; of course such

a property is impossible in a finite group. The line connecting

these two infinitely distant points will be called the "axis of the

motion".

This axis is nothing but the polar of the point M with

respect to the absolute. For all points A of an ellipse which

touches the absolute at Rand S, the cross-ratio of the four

points M,P,A,Q is constant, as we know from analytic geometry.

We are therefore justified in

regarding such ellipses as

non-euclidean circles with

ideal midpoint M. All

such non-euclidean circles

with the same ideal mid-

point have the same real

axis in common.

If r is the
M
geometric connection of
Figure 10
the point Al to the

point Bl which corresponds to the element C, then the net motion

 67

represented by the rotation about the ideal midpoint M sends the

The points B.1. thereby have distance r from the points Ai and

-1
which constitute a r A r-chain, therefore also lie on a non-euclidean

circle which is concentric with the A-circle and has the same axis

RS. Moreover, the ideal angle of rotation about M is the same

-1
for pairs of successive points on the A-chain and r A r-chain

respectively.

We can now briefly summarise our various results in the

following theorems:

1. In a motion of the net which brings like-labelled edges

into coincidence and carries the vertex Al to A2 , all vertices

of the resulting A-chain lie on a circle, with an ideal midpoint,

which touches the absolute at two distinct points, whose connecting

line is called the axis of the motion.

2. Under this motion, all points of a r- l A r-chain lie on

a circle which has the same midpoint, axis and angle of rotation

as the A-chain.

With the help of these theorems the transformation problem can

now be solved without too much difficulty. Let A and B be the

elements under investigation, whose straight line connections with

the identity element, which is represented by the point Al = Bi '

will be denoted by A = AlA2 and B = BiB;. First we construct

the A-chain and B-chain and their circumscribing circles, which

 68

both have ideal midpoints and angles of rotation, but real axes.

Then we describe the non-euclidean circle with the same midpoint

as the A-chain and the same radius as the B-chain, so that the

axis of the [new] B-circle coincides with that of the A-circle.

However, since these circles have imaginary midpoints, we must

construct the new circle by determining all the points which have

the same distances ~ from the A-axis as the points of the [old]

B-circle have from the B-axis. [Actually, db will be constant,

as a non-euclidean distance (Translator's note).]

new B-circle

B'4
A
',db
, db , 3
, I
,

B-axis

A-axis

Figure 11

Now if A is transformable into B, the new B-circ1e must

satisfy two requirements vis a vis the axis of the A-circle: in

the first place, a circle K of radius ill around one of its

3
vertices must contain at least one vertex Y such that the edge
 69

path B issuing from Y ends at a vertex H on the new B-circle.

Secondly, when this necessary condition is satisfied, the ideal

angle of rotation of the motion Y+ H must agree in sense and

magnitude with the ideal angle of rotation of the motion Al + A2 .

In this way the identity and transformation problem can be

solved by geometric operations. The analytic execution of this

process is still called for, and could for that reason be the

subject of a longer work.

4. Second method for the solution of the transformation problem.

We shall now give a second general method for the solution of the

identity and transformation problems, which will also prove to be

useful sometimes. If A is an element of the fundamental group,

represented by an edge path AlA2' then this uniquely determines

a motion of the net which carries Al to A2 • Because if there

were a second motion, other than the rotation of the A-chain about

the ideal midpoint, it would have to be a rotation about a real

midpoint A', and we have shown that such a motion of the net is

impossible. Now suppose that

the element A is transformed

[conjugated] by an arbitrary

element C. We then find the

transformed element B = C- l AC

by attaching to Al and A2

the homologous edge paths

AlC l and A2C2 which

correspond to the element C, A'1

and then shifting the initial

Figure 12
 70

Since AlC l and AZC Z are homologous edge paths, the

motion of the net corresponding to the element A simultaneously

sends Al to AZ and Cl to Cz . But Al is sent to AZ by

a translation V and Cl is sent to Cz by a translation

a
Vb = V-I V V
cae
, where Vc denotes the translation which sends

Al to Cl . Moreover, the translation Vb obviously sends the

point Al to BZ • We therefore have

(1) Two elements A and B of the fundamental group are

transformable into each other [conjugate] when the corresponding

translations Va and Vb are transformable into each other, i.e.

when

V = v-I V V
b c a c

Two mutually transformable [conjugate] translations Va and Vb

are distinguished by the fact that they represent the same motion

of the net, i.e. they have the same ideal angle of rotation, and

homologous ideal midpoints.

This theorem enables us to give a simple solution to the

transformation problem for two translations. If XI,X Z and

YI,Y Z are pairs of points on the axis of the A-chain which
correspond under rotations with the same angle, then XIX Z = ~z .

Further, if Zl and Zz are two corresponding points on the axis

of the B-chain, then ZIZZ = XIX Z ' since both segments correspond

to the same angle and one axis is sent to the other by the motion
-1
Vc . If we now denote the distance between two points which

correspond under the rotation through the given angle as the

 71

"displacement length cr of the motion", then we obtain the

a
theorem:

A necessary condition for the trans formability of two dis-

placements Va and Vb into each other [conjugacy] is the

equality of their displacement lengths, Le. crb = cr a .

However, this condition is not completely sufficient. In

order to find other conditions, we recall that our rotations about

homologous ideal points must carry the net onto itself. Thus if

we construct the image of the displacement segment v in

a
a single polygon, which is well known to be a sequence of individual

straight line segments, then the endpoints must coincide or else be

corresponding points on homologous boundary edges or

Now the motion carries the axis segment XIX Z to XZX3 ,

Figure 13

but since the net is unchanged in the process, XZX3 must map back

onto the same sequence of line segments in a polygon as Xl X2 , and

 72

indeed with the same initial and final point. Since the same is

true of the segments X3X4,X4XS"",XOXl,X_lXO"'" we see that

the axis runs homologously through the different polygons. Thus

if Y-l'YO,Yl , ..• is another series of corresponding points on

the axis of the same motion,

always map onto the same sequence of segments in each polygon, as

do the segments ""X_lXO,XOXl,XlX2,X2X3"'" except that the two

images in general have different initial and final points. But

this means that the image of the segment vb = Zl Z2 must also

coincide with the image of the segment v Xl X2 , up to a cyclic

a
rearrangement of the sequence of segments, because the position of
-1
this image is not altered by an arbitrary motion V • Corres-
c
ponding to each motion of the net we therefore have a uniquely deter-

mined cyclic sequence of line segments in a polygon, which we call

the "axis image of the motion". It then follows immediately that:

The necessary and sufficient condition for the mutual transforma-

bility [conjugacy] of two translations Va and Vb is the

equality of their axis images.

We have therefore obtained a method for deciding, in a finite

number of steps, whether two motions, or the corresponding group

elements, are mutually transformable. Here one can also express

the condition for transformability analytically, and then arrive

at an analytic solution of the conjugacy problem.

S. One-sided surfaces. This enables us to conclude our investi-

gat ion of two-sided surfaces, and we shall now treat the one-sided

surfaces in a similar way. The best known example of such a sur-

face, on which one can no longer distinguish two sides, is of

 73

course the "Mobius band" which one can easily construct by pasting

together the two ends of a paper strip in such a way that one side

continues into the other. However, this figure is not the simplest

one-sided surface of all, because it still has a boundary curve.

In order to obtain a one-sided surface without boundary we divide

the boundary of a circular disc into four equal parts by four

points A,B,C,V, and identify the parts in such a way that A

coincides with Band C and V, in other words, AC is

defined as equal to BV, and CB as equal to VA. This diametric

relationship between the B

boundary parts may be easily

visualised by suitably

joining together the identi-

fied parts. The result is

V
a closed one-sided surface

of the form Tl . The in-

dex of T denotes the

number k of circuit cuts

required to reduce a closed Figure 14

surface Tk to a disc; thus for two-sided surfaces k has the value

2p, i. e. twice the genus, and one also has the equation

where a O,a l ,a 2 denote the numbers of vertices, edges and faces

respectively in a polyhedron obtainable by continuous deformation

of the surface. In order to obtain the circuit cut for our

simplest one-sided surface TI' we connect two homologous points,

say A and B of the disc boundary [Fig.14]. By the construction

 74

of the surface, this connection is a closed curve. All closed

curves on our one-sided surface are reducible either to a

point or to this one curve. The fundamental group of TI ,

i.e. the group of curves on this surface, has only a single

2
generator, which we will denote by s, and the relation s I,

so that

are the elements of this fundamental group.

A one-sided surface of higher connectivity can be produced

by cutting an ordinary two-sided ring [torus] or tube and pushing

one end through to the other in such a way that the outer side of

the tube joins up with the inner

Figure 15

The resulting one-sided ring surface [Klein bottle] may be reduced

to a disc by two circuit cuts a and b, and hence it is a

surface of the form T2 . If the cut ring is spread out with a

definite direction of the cuts displayed then the result is a

rectangle ABCV whose sides AB,BC,CV,VA are denoted in turn by

-1
a,b,a,b . In this single rectangle, as in the case of the two-
 75

sided ring, curves on the surface

a
are represented by curve segments

with endpoints on the boundary of

the rectangle. In order to obtain

a
an unbroken image of a continuous

curve on the one-sided ring, we Figure 16

again replace the single rectangle by a net of rectangles which cover

the whole plane simply and without gaps, in which like-positioned or

homologous points of individual rectangles correspond to each other.

We can then regard this rectangular net as an infinite-sheeted

covering of the ring, the individual sheets of which meet along the

curves a and b. Now the rectangular net just constructed is at

the same time the group diagram of the fundamental group of the one-

sided ring, which is determined by the two generators a and b

and the relation

abab -1 1.

a a a a

b b b b b
a a a a

b b b b b
a a a a

b b b b b
a a , a a

b b b b b
a a a a

b b b b b
a a a a

Figure 17
 76
-1 -1
Namely, we have the four segments a,b,a ,b issuing from each

vertex 0f h
tenet, ·
t h e re1 atlon abab- 1 -- 1 is satisfied, and each

relation of the group, i.e. each closed edge path of our net, is

composed from this relation. Thus the construction of the group

diagram gives an immediate solution of the identity problem. Two

elements of the group are identical if and only if they are

represented by edge paths with the same initial and final point;

topologically, this means that a closed curve on the ring is

contractible to a point if and only if its image in the net is

closed.

While the identity problem is thus settled exactly as in the

case of the simple two-sided ring, the solution of the transfor-

mation problem here takes a rather different form. This is because

we have two different types of motion of the net. There are

parallel displacements along the segments a and b, but a dis-

placement along an odd number of b segments has to be composed

with a reflection in the vertical. The first operation maps the

horizontal lines into themselves, while the second, if it in-

volves reflection, sends a horizontal line to one which is an odd

number of horizontal strips away. Now since two group elements are

mutually transformable if and only if the associated displacements

can be transformed into each other, and two mutually transformable

displacements are related via the same motion of the net, our

problem is to bring the parallel displacements and reflections into

a canonical form. Two pure parallel displacements P and Pc"

c
associated with the elements c and c', are obviously trans-

formable into each other if and only if they have the same length
 77

and direction, i.e. when they are identical. Thus the first con-

dition for the mutual transformability of two elements c and c'

is in the theorem: two elements of the fundamental group are trans-

formable into each other [conjugate] if b appears to an even

power in both elements, and the exponent sums of a and b in both

elements are respectively equal.

Two parallel displacements through an odd number of horizontal

strips, related via reflection in a vertical, are obviously trans-

formable into each other only if they involve the same number of

horizontal strips; the reflection of the net can take place either

in a b-line or in the line connecting midpoints of opposite a-

edges. We thereby obtain a second condition for the mutual trans-

formability of two elements in the theorem: two elements of the

fundamental group are transformable into each other [conjugate]

when both elements contain b to an odd power, in fact to the

same power in both, and the power of a is either even in both

or odd in both.

Thus, for example, the elements aab and b must be

transformable into each other; likewise the elements ab and

-1
a b. In fact, as one finds immediately with the help of the

group diagram,

-1
a(b)a a(aba)a- l aab

and
-1
a b,

using the relation aba = b each time. Our two theorems now

yield the following canonical representation for the elements of

the group:
 78

each element of the group is transformable into a unique element of

this form. This implies the theorem that each curve on the surface

is uniquely reducible to one of these three types. It may be re-

marked, incidentally, that the fundamental group of the one-sided

-1 -1
ring is not abelian, because, e.g., aba b is not equal to 1,
2
but to a . However, our group does indeed have a non-trivial

abelian subgroup, namely the subgroup of all elements in which the

total power of b is even, and in fact this subgroup is isomorphic

to the group of the two-sided ring.

We obtain the next one-sided surface T3 by cutting a

circular hole in the one-sided ring T2 and identifying points

on the boundary of the hole diametrically, as with Tl . By

cutting along suitable circuits a,b,c and spreading it out,

-1 -1
we can reduce this surface to a regular hexagon with a,b,c,b ,c ,a

as its sequence of boundary sides. We now construct a net of

regular hexagons in the non-euclidean plane, with six such hexagons

meeting at each vertex, so that the hexagon angle is ~rr = 60 0

then we can label the sides of each hexagon in the sequence

-1 -1
a,b,c,b ,c ,a. This net of hexagons then represents, on

Fig.18
 79

the one hand, the infinite-sheeted covering of our one-sided surface

T3 , but on the other hand it can also be regarded as the diagram of

the fundamental group of our surface, which is given by the three

generators a,b,c and the relation

The identity problem is again immediately solved by this group

diagram, in the same way as in the earlier examples.

The transformation problem requires us to determine the

mutually conjugate motions, and consequently to bring the motions

of the net into canonical form. Here we again have to distinguish

between two kinds of net motions. The first kind, which includes,

e.g., the motions corresponding to the elements band c, are

the rotations about an ideal midpoint, already familiar to us.

The second kind include, e.g. the motion determined by the

element a, consisting of a reflection in the line associated

with the rotation part of a, whose midpoint in our example is

the centre of the polygon and hence finite. Our problem then is

to find a canonical form for the second kind of motion. For this

purpose we will determine a line in the net which is fixed by this

kind of motion. If we drop the perpendicular MF from the

rotation centre M to the reflection line and construct half the
w
rotation angle, on each side of MF from M, then our
2'
motion which consists of rotation about M through the angle w
followed by reflection in the reflection line, obviously leaves

invariant the perpendicular through F to the initial leg of

the angle w. We therefore call this the reflection axis of

our motion. The motion of the net determines a certain length of

 80

displacement along the

reflection axis, called its

"displacement length". It

may now be easily shown that

the two motions are trans-

formable into each other

M
only if the corresponding

reflection axes have

homologous positions on the

net polygons and equal Figure 19

displacement lengths. We have then also solved the transformation

problem for the fundamental group of the T3 •

Finally, we shall briefly treat the one-sided double ring T4 •

Here, the fundamental group is given by four generators a l ,b l ,a 2 ,b 2

with the relation

In this case the group diagram is a net of regular octagons in the

non-euclidean plane, in which each octagon angle has the value

211 11
8="4' The identity problem is settled with the construction of

the group diagram, and the transformation problem is solved by

bringing the motions of the net into a canonical form.

6. Decomposable groups. In the groups treated previously we had to

satisfy only one relation, in which each generator appeared twice.

We then found that the group diagram was represented by a net of

regular polygons in the non-euclidean plane. The question now

 81

arises, whether each group with a single defining relation,

containing each generator twice, has a group diagram in the

form of a non-euclidean polygonal net. As we shall now see

in a few examples this question must be answered in the

negative.

Namely, let a,b,c be the generators of a group with the

relation

~c~c 1.

We shall set abc = d and then take a,b,d as generators. The

single relation which must then be satisfied is d 2 = 1, so that

a and b are completely arbitrary. In order to find the group

diagram for the group so determined, we only need five segments to

issue from each vertex, corresponding to the operations

-1 -1 -1
a,a ,b,b ,d = d ,and the whole diagram contains no closed

edge path. But this also shows that the diagram of our group
-1
a

-1
a

Fig.20
 82

cannot possibly be reduced to a polygon net. One does not see

quite so immediately that the group with generators a,b,c,d,e

and relation

-1 -1_ -1 2
a cdc odb ae 1

contains arbitrary elements. If we substitute in this

-1 -1
a c f and a b = g,

then

Hence if we consider b,d,e,f,g as generators of our group, then

we now have to satisfy the relation

so that the operation b is completely arbitrary. On this basis,

the group diagram here can never be a net of polygons. We call a

group which has a set of generators in which one or more are

arbitrary a "decomposable group". For such decomposable groups,

when there is only one relation and each generator appears twice,

the group diagram is easy to construct, and the identity and

transformation problems can then be solved.

7. Analytic remarks. For all indecomposable groups with only one

relation, in which each generator appears twice, we have completely

solved the identity and transformation problem, assuming that the

group diagram is represented as a simple regular polygon net in

the euclidean or non-euclidean plane. It may now be shown that

these groups are isomorphic either to groups of linear nonhomo-

geneous transformations of two variables, as is the case, e.g.,

 83

with the fundamental group of the two-sided and one-sided ring,

which is represented in the euclidean plane, or to groups of

linear homogeneous transformations of three variables leaving

the equations invariant, when the group

diagram is a regular polygon net in the non-euclidean plane.

If we take the Poincare representation of the non-euclidean plane

as a basis, then the motions which come into consideration as

group elements are analytically equivalent to the linear

transformations

z' = az + b
cz + d

of the complex variable z = x + iy, where a,b,c,d are real

numbers and the determinant ad - bc > 0 • The proof of this

assertion follows from the fact that our motions are circle-

preserving mappings of the upper complex half plane, since the

half-discs bounded by Poincare lines, i.e. the non-euclidean

half planes, are mapped onto each other and angles are preserved.

Poincare calls the groups of such real linear transformations

of the complex variable "fuchs ian groups". To obtain the

fundamental groups of the one-sided surfaces, whose motions also

include reflections, we must add to the above transformations

the analytic expression

z' -z = -x + iy

for a reflection. We may call groups consisting of such

transformations "extended fuchs ian groups". Thus the fundamental

groups of our surfaces are isomorphic to such extended fuchs ian

groups.
 84

To round off this information we shall add a short remark

on higher groups. As PoincRr~ hRs proved, the linear transfor-

mations of the complex variable z ~ x + iy with complex co-

efficients a,S,y,o are representable by motions of non-

euclidean space. Poincare's image of this space is a half space

determined by the z-plane, the so-called "Poincare half space",

in which the non-euclidean planes are represented by spheres

which cut the z-plane orthogonally. Since linear transformations

of the z-plane send circles to circles, and hence at the same time

can be considered to send orthogonal spheres to orthogonal spheres,

a linear transformation of the z-plane corresponds to a motion

of the non-euclidean space and conversely. The group of motions

of non-euclidean space is identical with the group of linear

transformations of z. This suggests going on to group diagrams

in this non-euclidean space, for example in the case of the

group with the generators a and b and the relation

Here we obtain the dual group diagram when we represent the four
-1 -1
operations a,a ,b,b by the reflections of a regular non-

euclidean tetrahedron in its four faces. As a result of the

relation between the generators, six tetrahedra must come

together at each edge. The edge angle of the regular tetra-

hedron must therefore have the value 60°. But this is the case

for the regular tetrahedron whose vertices lie on the absolute

sphere which bounds the non-euclidean space in the Klein

representation, because ordinary euclidean geometry holds at

the points of the absolute sphere. By repeated reflection in

 85

these tetrahedra we can produce

the whole group diagram. Each

a -1
a
element of the group may then
be associated with a particular

motion of the tetrahedral net.

Fig.2l
Then by again bringing these

motions of the non-euclidean

space into a canonical form we obtain the solution of the identity

and transformation problems. Analytically, the motions are

equivalent to complex linear transformations of the complex

variable z = x + iy, because, as transformations of the Poincare

half space, these motions have the property of sending spheres

orthogonal to the boundary plane to other such spheres, so that

the motion sends circles to circles in this boundary plane and

angles are preserved. It follows that here we are in fact dealing

with linear transformations of the complex variable z in the

boundary plane. Poincare calls the groups of such transformations

"Kleinian groups" because Klein stimulated the development of these

ideas with his investigations of modular functions and automorphic

functions.
 TRANSLATOR'S INTRODUCTION 3

In his first major publication in topology, Dehn uses his ex-

perience in surface topology and group theory to formulate and
attack the major problems of 3-dimensional topology. Together
with Poincar~ [1904], the paper which follows is the foundation
for almost all subsequent work in this field. A brief summary of
the contents will support this claim: statement of the word and
conjugacy problems for finitely presented groups, realisation of
such groups as fundamental groups of surface complexes, Dehn's
lemma, the diagram of the trefoil knot group, equivalence between
triviality of a knot and commutativity of its group, construction
of homology spheres by surgery, and proof that one of them (actually
Poincare's homology sphere) has finite fundamental group; all making
their first appearance in print.

In Chapter I, §l, the word and conjugacy problems are formulated, and
Dehn describes a process which determines the diagram of a finitely
presented group. He points out that the process is non-constructive,
however, being equivalent to solving the word problem, for which an
algorithm was then unknown (and now known to be non-existent). In
practice, Dehn only constructs diagrams whose correctness is immediate
from the defining relations, and his illustration is of the icosahedral
group, no doubt recalled from his lectures.

The short §2 of I represents finitely presented groups by

surface complexes; the generators by particular paths in a graph and
relations by discs spanning paths which represent the defining
relators. Later (in Chapter III, §l), Dehn observes that such a
representation of the fundamental group of a 3-manifold arises
naturally from its representation as a polyhedron with identified
faces. If one deletes the interior of the polyhedron and then
identifies the faces, the result is a surface complex (nowadays
called a spine of the manifold) with the same fundamental group.
The surface complex is useful in its own right for studying groups.
In particular, Reidemeister [1932] shows that subgroups correspond
to covering complexes, a result which can be used, e.g., to give an
 87

elegant proof of the Kurosh subgroup theorem (Baer &Levi [1936]).

Chapter I, §3 states, but does not completely prove, the famous

Dehn's Zemma. A gap in the proof was found by Hellmuth Kneser, and
he informed Dehn of it in a letter dated 22 April 1929. The in-
teresting correspondence which ensued is preserved in the Humanities
Library, Austin, Texas; it includes Kneser's elegant representation
of nenn's homology sphere (see later, and also Threlfall & Seifert
[1930], p.66) as a dodecahedron with suitable identification of
opposite faces. However, Dehn and Kneser were unable to fix the
proof of the lemma, and a correct proof did not appear until
Papakyriakopoulos [1957]. It may be significant that Papakyriakopoulos
had previously undertaken a lengthy study of surface complexes
(Papakyriakopoulos [1943]), which included proving the Hauptvermutung
and solving the homeomorphism problem for them.

The lack of a proof of Dehn's lemma in 1929 was as big a dis-

appointment to Kneser as to Dehn. Kneser's son, Martin, informed
me in a letter that his father had planned to write a book on 3-
dimensional topology based on Dehn's lemma. The collapse of Dehn's
proof may have been instrumental in his moving from topology to
several complex variables. Nevertheless, Dehn's "switchover"
(Umschaltung) technique proved to be of fundamental importance when
the theory of 3-manifolds finally took off in the 1960's, and
Kneser's application of it, Kneser [1929], was then recognised as
a decisive contribution (see e.g. Hempel [1976]).

Chapter II, on knots and their groups, begins in §l by defining

a knot to be non-trivial ("knotted") if it cannot be spanned by a
non-singular disc. §2 encloses the knot by a tubular neighbourhood
and observes that, even if the knot is non-trivial, there is always
a curve on the tube surface which is null homologous in the comple-
ment of the tube. It follows that the homology of the knot comple-
ment can be killed by attaching a disc along this curve, and this is
the basis of Dehn's homology sphere construction later. §3 arrives
 88

at a presentation of the knot group, similar to the Wirtinger

presentation first explained in Tietze [1908], but by a different
route. Dehn appeals extensively to surface topology in his proof,
though this is not really nece55ary,

The first major result of Chapter II is in §4. Dehn's lemma

is used to show that a knot is trivial if its group is abelian
(the converse is obvious). He points out that one can then decide
whether a given knot K is trivial by construction of its group
diagram, since this in turn will decide whether the group is abelian.
-1 -1
(E.g., by checking whether a,a,a, a, = 1 for each pair of
1 J 1 J
generators a, ,a, .) However, this is a little misleading, since
1 J
one needs a uniform algorithm which constructs the group diagram
from K, i. e. a uniform algorithm which solves the word problem
for knot groups, and such an algorithm was first given by Waldhausen
[1968]. By this time, a geometric algorithm for recognising the
trivial knot had already been given, by Haken [1961], Haken's method
uses none other than Dehn's switchovers, in a very elaborate way, and
furthermore, Waldhausen's work is based on Haken's. One can there-
fore say that Dehn was on the right track, though he had a long way
to go.

Chapter II, §5 contains the diagram of the trefoil knot group,

and thereby solves its word problem. The diagram is non-planar and
quite unlike any group diagram previously constructed, though sur-
prisingly simple and beautiful. Dehn uses it to construct the group
diagram for the homology sphere obtained by killing the homology of
the trefoil knot complement in an operation now known as Dehn surgery,
showing that for the simplest surgery the group is of order 120 and
in fact the extended icosahedral group he had studied in his lec-
tures. It was shown by Weber & Seifert [1933] that Dehn's homology
sphere is the same as the first one discovered, by Poincare [1904],
A simple pictorial proof is given in Rolfsen [1976]. Poincare did
not show that its group was finite, only that it had the icosahedral
group as homomorphic image. This homology sphere is still the only
 89

one known which has a finite non-trivial fundamental group.

It should be mentioned that Dehn had in fact introduced

surgery earlier, in the abstract Dehn [1907], for a different
homology sphere construction which is short enough to quote in
full:

A very clear example of such a noteworthy manifold

can be constructed as follows: a knotted torus T2 in
ordinary space divides the latter into a solid torus
T3 and a part K3 not homeomorphic to it. Suppose that
curves C resp. r are non-separating on T2 and
bounding in resp. K3 . One joins K3 to a homeo-
morphic body which is bounded by Ti (with the
curves C' and r') in such a way that T2 is identi-
fied with Ti, C' with r, and C with r'. In the
resulting closed M3 , each curve is bounding when taken
once. However, it is not homeomorphic to the ordinary
space, since this M3 is divided by the torus T2 into
two parts K3 and K;, neither of which is homeomorphic
to a solid torus.

The geometric theorem used here by Dehn, that a torus in $3

bounds a solid torus on at least one side, was not proved until
much later, by Alexander [1924]. As I pointed out in Stillwell
[1979], one can instead justify the construction by a group
theoretic argument and Dehn's lemma. It is plausible that Dehn first
began thinking about the lemma in the context of this example. Nowa-
days, Dehn surgery is not used only to kill homology. It can in
fact be used to construct any orientab1e 3-manifold from $3
(Wallace [1960], Lickorish [1962]).

The final Chapter III derives representations of 3-manifolds by

polyhedral schemata and Heegaard decomposition, which of course
were already known, and then tries to see what light they throw on
the problems of recognising whether a manifold is homeomorphic to
 90

$3, and settling the Poincare [1904] conjecture that a simply

connected closed 3-manifold is 53. Since these problems are
still open, 70 years later, it is not surprising that Dehn's
discussion of them is rather inconclusive.

For the sake of readability I have made the slightly

anachronistic translations of "ElementarflachenstUck" by "disc"
or "2-cell" and "ElementarraumstUck" by "3-cell" rather than the
lengthy literal translations. I have also renumbered the first
seven figures, which occur out of order in the original paper.

REFERENCES
.T.W. Alexander [1924]: On the subdivision of 3-space by a polyhedron.
Proc. Nat. Acad. Sci. 10, 6-8.
R. Baer & F. Levi [1936]: Freie Produkte und ihre Untergruppen.
Compositio Math. 3, 391-398.
M. Dehn [1907]: Berichtigender Zusatz zu III AB3 Analysis situs.
Jber. Deutsch. Math. Verein. 16, 573.
W. Haken [1961]: Theorie der Normalflachen.Acta Math. 105, 245-375.
J. Hempel [1976]: 3-manifoZds. Ann. of Hath. Studies 86, Princeton
University Press.
H. Kneser [1929]: Gesch10ssene Flachen in dreidimensionalen
Mannigfaltigkeiten. Jber. Deutsch. Math. Verein. 38, 248-260.
W.B.R. Lickorish [1962]: A representation of orientable combina-
torial 3-manifolds. Ann. Math. 76, 531-540.
C.D. Papakyriakopoulos [1943]: A new proof of the invariance of
the homology groups of a complex (Greek). Bull. Soc. Math.
Grece 22, 1-154.
[1957]: On Dehn's lemma and the asphericity of knots. Ann.
Math. 66, 1-26.
H. Poincare [1904]: Cinquieme complement a l'analysis situs
Rend. circ. mat. Palermo 18, 45-110.
K. Reidemeister [1932]: EinfUhrung in die kombinatorische TopoZogie.
Teubner, Leipzig.
 91

D. Rolfsen [1976]: Knots and Links. Publish or Perish, Inc.

J.C. Stillwell [1979]: Letter to the Editor. Math. Intelligencer
1, 192.
W. Threlfall & H. Seifert [1930]: Topologische Untersuchungen der
Diskontinuit~tsbereiche end1iche Bewegungsgruppen des
dreidimensionales sph~rischen Raumes I. l1ath. Ann. 104, 1-70.
H. Tietze [1908]: Uber die topologischen Invarianten
mehrdimensionaler Mannigfaltigkeiten. Monatsh. Math. Phys.
19, 1-118.
F. Waldhausen [1968]: The word problem in fundamental groups of
sufficiently large irreducible 3-manifolds. Ann. Math. 88,
27~-280.

A.H. Wallace [1960]: Modifications and cobounding manifolds. Can.

J. Math. 12, 503-528.
C. Weber & H. Seifert [1933]: Die beiden Dodekaederr~ume. Math.
Zeit. 37, 237-253.
 ON THE TOPOLOGY OF THREE-DIMENSIONAL SPACE

In the present work, we topologically investigate closed curves

(knots) in ordinary (i.e. hyperspherical) space (Chapter II), as
well as three-dimensional space in general (Chapter III). Preceding
these discussions are the necessary general considerations (Chapter
I). In fact in §l we consider groups of discrete operations, namely
those constructed from a finite number of generating operations,
among which a finite number of relations are given.

A diagram is constructed for such a group, a regular line segment

complex, whose construction, in contrast to previous representations
of general infinite groups, disposes of important problems connected
with the group. At the same time this construction also involves
some difficulties. In the course of this work group diagrams will
be given for known (p.1OO) as well as for previously unknown finite
and infinite groups (p.ll6). They are all connected with regular
tessellations of euclidean or non-euclidean planes.

In §2 the construction of the fundamental group of an arbitrary

surface complex follows. By means of the group diagram, each surface
complex is associated with a regular line segment complex, infinite
in general. In the case of closed two sided surfaces it is known
that these give regular nets of 4n-sided cells in the hyperbolic
plane, with 4n cells meeting at each vertex.

In §3 a general theorem on surface complexes is derived, the

IlellUlla" which is often used in the later investigations. It is the
2-dimensional generalisation of the (trivial) theorem that one can
connect any two points of a line segment complex by a polygonal
path free of singularities.
 93

In chapter II we first settle the definition of knot (§l) and

in §2 introduce the enveloping tube of a knot. In §3 we construct

th~ fUftdA!!fttdl ~roup of the knot, the Jelinit!on of which is

derived immediately in the simplest way from the plane projection
of the knot. If this group is abeZian then the given curve is un-
knotted (§4, Theorem 2). Further, a method is given by Theorem 1
for deciding whether a curve is knotted or not, by construction of
the diagram of the fundamental group of the knot.

In §5, special knots are handled, in particular the trefoil

knot and its relatives. The associated group diagrams are con-
structed, and in addition Poincar~ spaces are derived from them,
i.e. three dimensional manifolds which, despite being simply
connected*and without torsion, are not homeomorphic to the usual
space. The simplest one found here has a finite fundamental group
and indeed it is the icosahedral group with reflections. The others
have infinite groups. The group diagram is established for all of
them; this settles the question whether a given curve of the manifold
under consideration can be contracted to a point or not. This
question had not previously been answered for such manifolds, nor
did one know whether the associated fundamental group was finite or
infinite. These results have the particular value of yielding a very
simple method for the construction of infinitely many Poincare spaces,
of which the discoverer had constructed only one, in a complicated way.**

In §l of chapter III different, easily visualised, methods for

generating the three dimensional manifolds are given. For example,
all two sided manifolds are divided into two pieces by a closed two
sided surface, and the pieces are homeomorphic to the simplest part
of space which can be bounded by such a surface. These methods of
generation also lead to the construction of the fundamental group of
the given manifold.

§2 deals with the important problem of the topological charac-

terisation of ordinary space, without resolving the problem however.
It treats the question of how ordinary space is to be topologically
defined through the properties of its closed curves, and how to make

*In terms of homology (Translator's note).

** See Dehn, D. Math. Ver. 1907 p.573.
 94

it possible to decide whether a given space is homeomorphic to

ordinary space or not. The history of this problem began when
first Heegard (Diss. Copenhagen 1898) and then Poincare (Pal. Rend.
v.13 and Lond. M.S. v.32) pointed out that in order to characterise
ordinary space it does not suffice to assume that each curve bounds,
possibly when multiply traversed. Indeed the manifolds with torsion
exhibit this. Then Poincare proved in Pal. Rend. 1904, by con-
struction of a "Poincare space" that it is even insufficient for
each curve to bound when traversed once.

It now is natural to investigate whether it suffices to suppose

that each curve of the space bounds on elementary piece of surface
[disc]. This is also suggested at the end of Poincare's work.
However the reduction of the problem given in the foregoing work does
not appear to lead directly to a solution. A deeper investigation of
the fundamental groups of two sided closed surfaces seems to be
unavoidRhle.

CHAPTER I Preliminaries
§l Group theoretic aid (the group diagram)
In topology one is frequently led to problems of the following
form:
A finite series of operations al.aZ •..•• a n are given as
generating operations of a group. The group G is then completely
determined by a finite number of reZations between the defining
operations. say of the form

A +1 or -1)

(m)
E.
n a ~ S 1
i k~m) m
~

The group G is in fact completely determined by A. For if two

operations Sand T of the group are given as products of the
ai it is completely determined whether S = T follows from the
 95

relations A or not, i.e. whether S = T is the case in the group

with generating operations and relations A , or not.

Our problems now run as follows:

1. A method is sought for deciding in a finite number of steps
whether two products of the given operations a i of G are equal
or not, in particular whether one such product operation is equal to
the identity.
2. A method is sought for deciding in a finite number of steps,
given two substitutions Sand T of G, whether there is a third,
U, such that

S = UTU- 1

Le. whether S may be shown by A to be a conjugate of T.

These two problems are completely solved for the special groups
G given as follows:
p

G
P

G is none other than the fundamental group of curves on a two-sided

P
sUl'face Fp of genus p. Our problem here is equivalent to the
problem: when are two closed curves on F reducible to one
p
another (with or without fixing a point)? The solution is obtained
by a mapping of G onto a regular tessellation of the hyperbolic
plane. We shall attempt the solution of the general problem through a
generalisation suggested by this mapping.

We construct the following line segment complex corres-

ponding to a group G. Suppose

Then through a point Z we draw a circle (closed curve) K' with

R, vertices Z,P l ,· .. ,PR,-l and label ZP l with +ak or
1
-ak ,PIP z with +ak or -a etc. , according as £1'£2'" . are
kZ
1 2
 96

equal to +1 or -1. Then we draw a second circle K2 through Z

whose segments, starting from Z, are labelled by +a k or
2
-ak , ... ,+ak or -a (the signs again decided by the values Ei ).
2 ~ k~
We proceed in this way until we have exhausted all cyclic trans for-
mations of 81 . There are then ~ circles attached to Z, whose
segments, taken in a suitable sense, have the given labelling. If
they were traversed in the reverse sense, this labelling would be
taken with the opposite signs. Now if two segments ZP and ZQ
emanating from Z, running to P and Q respectively, carry the
same labelling then P and Q are made to coincide, together with
the segments themselves, and the process will only be continued as
long as different segments emanate from Z. We also carry out
the same process for the other vertices of the circle, again only
as long as all segments emanating from a vertex are different. We
-1
call the line segment complex so constructed Cl , and call Z

its centre.

-1
We now prove the following property of Cl : If a closed curve
-1
of Cl , traversed in a particular sense, consists of the segments
dl,dZ, •.• ,dq (all equal to ±a i ) then by virtue of the relation
81 = 1 the operation dldZ ... dq is likewise equal to 1: it results
from the operation 51 by conjugation and composition.

In fact: our assertion is correct for the initial state of Cl ,

where all closed curves of the complex go through Z and comprise
curves of the form:

But each closed path whose initial and final point are Z is
representable by a composition of these expressions. Any other
choice of initial point however represents a cyclic interchange,
and hence is obtainable by conjugation. We must now convince our-
selves that this property cannot be lost with the coincidence of
two segments originating from the same point. If d = PQ = PR is
such a segment we can assume that the curve in question goes through
Q, since all other closed curves would also be closed before the
 97

alteration and have the property claimed, by hypothesis. So suppose

k is a closed curve which goes through Q. Then the curve
-1
k' ~ dkd Wag alrnndy ~l~§~A b~f6~~ the ~lteration, and since by
assumption it results from Sl by conjugation and composition, the
same is true of k, since it is a conj ugate of k'.

We now introduce in succession the other relations S2"",Sm'

where we can assume that each generating operation ai appears in
at least one of them, or in Sl' Otherwise we add to A the trivial
-1
relation aia i = 1, without altering the group. We now apply again
the process described above until we come to a complex ci in which
no vertex has two identically labelled segments originating from it,
-1 -1
and each of the edges a l , ... ,an' a l , ... ,an emanates from Z.

It is completely determined which of the curves Sl'

-E l El
CL kl Sl CL kl '"''S2'''''Sn''' goes through a point P of as

initial point. We now add to Cl circles at each point P,

1
labelled in such a way that now aZZ curves of the collection
-E l El
Sl,a k Sl a k , ... ,S , .. emanate from P. This is done for all
1 1 n
1
vertices of Cl . We now modify the resulting complex by
identifying all edges which originate from a common pOint and are
2
identically labelled, thus obtaining a complex Cl • Then we
construct c3 from c2 as we have constructed
1 1
c2 from CIl • I
In this way we obtain a sequence of line segment complexes which is
in general unbounded, though in certain circumstances it is bounded
(the series breaks off when all curves of the system Sl""'Sn""
i
already emanate from each point of a complex Cl ).

(a) It follows as previously that if the segments of a

closed curve on ci traverse dl ,d 2 , .•• ,d q in order then the
operation

is generated from Sl'S2"",Sn by conjugation and composition,

i.e. S = 1 in G.
 98

(b) We shall now show: Given an operation S, there exists

a number r such that, when i > r, there is a curve in Ci with
I
Z as initial point which represents S and is closed if S results
from the Si by conjugation and composition.

For let S = dldZ .•. df where the representation on the right

hand side is chosen so that the generation of S by composition
and conjugation of the Si arises immediately, i.e. the right hand
side consists of factors which are either identically Si or terms
which result from composition and conjugation of the Si' One can
then find r such that the curve corresponding to this representation
(which is therefore comprised of the segments dl,dZ, •.. ,d f )
originates from Z and exists in each c~ (i > r). For a factor of
the curve of the form originating from a point of C~ exists by
construction in the c~ (i > r) and returns to the same point.
However, if a factor T of the curve is closed, then so too is the
factor of the form UTU- l , which lies in C~, because U and U- l
are coincident but oppositely directed paths in ci (i > r), by
our construction.

If S is originally given another form, then this results from

the above by repeated deletions of two successive terms of the form
e and e- l But such a deletion transforms a closed curve into
another closed curve, as again follows immediately from the proper-
ties of our construction.

This yields: two distinct vertices P and U in a can

be brought into coincidence by extension of this complex. However,
there is in any case a number r. such that any two distinct
l. r.l.
i
vertices of CI ' which remain distinct in CI ' remain distinct
R-
r.
in any CI : each curve of Ci which is unclosed in C l. remains
I I
that way as the construction expands, and consequently represents
an element of the group different from the identity. We shall
i r..
denote the permanent form of CI ' as exhibited by cll., by [C~],
and construct the infinite line segment complex dJ, the group
diagram of the group G, whose vertices and edges are obtained as
vertices and edges of [Ci] when we give i the successive values
I
 99

1,Z, ••. ,n. We associate the centre Z with the identity, then
each element S = dldZ ... d f is associated with the endpoint of
j polYBOllul puth. Tuo digti~~~ ~l~m~nt~ receive d15tlnct vertlces
of the diagram, and conversely, distinct vertices of the diagram
represent distinct elements of the group. The first fundamentaZ
probZem is soZved by eonstruetion of the group diagram c~.
However the above gives a proof of the existence of the diagram,
not a way of obtaining this diagram in a finite number of steps.
We can express this more precisely as follows: in order to decide
whether two elements Sand T are equal we construct a ci in
which both elements are represented by polygonal paths originating
at Z. If the endpoints of these paths are also distinct in
r.1
Cl ' then Sand T are distinct elements of the group. We have
certainly proved the existence of r i , but we know no general
method for really finding it. In the following section we shall
derive a whole series of such group diagrams, which will be of the
greatest use to us in the investigation of the corresponding groups
and topological figures. Here we shall show that the group diagrams
for the fundamental groups (defined at the beginning) of closed two-
sided surfaces are quite simple for abelian groups, but this has no
significance, because the solution of the fundamental problem for
these groups is trivial.

As an example, the group diagram for the icosahedral group is

given in Figure 1. The group is generated by the substitutions
and with the three relations

1 .

In the figure the line segments, with sign indicated by the arrows,
are numbered with the subscripts of the corresponding generating
substitutions. One sees clearly how the 60 elements of the group
are obtained from an initial element by and

*The relation between this complex and the icosahedral group is

known (see Maschke Am. Journ. 1896). Furthermore, the group
diagram is not strictly new for finite groups. It is closely
related to the Cayley "Colour diagram".
 100

1
Fig. L

The group diagram for a finite group is a finite line segment

complex (see the group diagram for the icosahedral group).

We shall draw a general conclusion from our construction: we

saw that each closed curve of the diagram represented an operation
equal to the identity because it resulted from the Si by compo-
sition and conjugation. On the other hand, each operation equal to
the identity is represented by a closed curve. Thus we have the
theorem:

Each operation equal to the identity in a group given by

generators and defining relations 8 1 = 8 2 = ... = 8 m = 1,
results from the 8i by composition and conjugation.

§2. The fundamental group of a surface complex.

In order to employ the properties of infinite groups in
topology, we need the following observations.

1. In a line segment complex C1 with aO vertices and a 1

edges one can draw a 1 - ao + 1 = ~ circles (closed polygonal paths)
a1,a2, ... ,a~ which go through a point 0 of C1 which is chosen
 101

so that each other circle of C1 can be continuously transformed

in C1 into a circle k' which consists of circles
8u~bab2y ordered and dIrected.

We write

,)
£. = +1 or -1
k k' (
k: = 1,Z, ... or

to denote that k' results from running through first a k ' then
1
a k , ... , finally a k , and in positive or negative
Z n
direction according as £l' •.• '£n are +1 or =1.

The proof of the theorem comes immediately from known theorems

on the reduction of curves on closed two sided surfaces when one
considers the two-side surface of genus p = ~ which results from
Cl as follows: one replaces each n-tuple node by a sheet with
n - 1 holes [i.e. with n boundary curves (Translator's note»)
and each line segment by a pair of such holes [boundary curves)
and a tube connecting them.

Z. We convert Cl into a surface complex Cz by putting

discs through m circles kl , k Z' ... , km .

Let kl,k Z' ... have the representation

t
(1) ,/1)
£1 £2 (1) nl
kl a (1) a (1) •.• a (1)
kl kZ k
nl

(m) £ (rn) ~
£1 nl

J
k a
m k(m) ak(m)
1 n
m

Then we construct the group whose operations are al, ... ,a~

and whose relations come from replacing the left hand side of A
by l's. This group is called the fundamental group Gc of the
2
 102

surface complex C2 . Each circle k of c2 corresponds to an

operation of this group through its representation by the a.
and conversely, each operation of the group is a circle of C2 '" •

When one notes (a) that each operation of GC that is equal

2
to 1 results from the elements of the group corresponding to
kl, ..• ,km by composition and conjugation, (b) that each of
collection of surface pieces comprising a disc includes at
least one whose removal leaves a disc remaining, it follows
easily that:

The necessary and sUfficient condition for a circle k in

C2 to bound a disc is that the operation of the fundamental g1'OUp
c01'1'esponding to k be equal to 1.

§3. A topological aid (the "Lemma").

We shall often need the following theorem from the topology of
surface complexes, which we shall refer to as the lemma, because of
its important place in this work.

Let C2 be a surface complex in the interior of a homogeneous

manifold Mn (n > 2). Let k be a curve in C2 which bounds a
(singular) disc E2 . If E2 has no singularities on its boundary,
then k also bounds a completely non-singular disc in Mn'

This theorem is obvious when the M has more than three

n
dimensions, for in such a manifold each two dimensional figure
has a neighbour without singularities. We can therefore assume
n = 3.

lea) Let AB be a multiply counted edge (A I B), at which

a number of sheets meet, i.e. AB bounds some pairs of 2-cells
whose preimages form adjacent pairs of 2-cells in the non-singular
preimage
o of E1 .
E2 Since we are inside a homogeneous M3 , by
2
hypothesis, the neighbourhood of AB in M3 is a 3-cell, and we
can therefore speak of the sheets intersecting (cutting) or touching
along AB.
 103

We now assume that all the sheets along AB touch, so that

there is a sheet with no others on one side of it, and we can re-
place this sheet by another which lies on this side and contains
no interior point of the edge AB, without creating any new inter-
sections with the other sheets. We can then proceed in the sdmg
way with the other sheets along AB, until sheets no longer touch
along AB J so that AB ceases to be a singular edge, and no new
singularities appear.

(b) Let A be a mul tiply counted vertex at which a number of

sheets meet, with the latter having no common edge emanating from
A. Since the neighbourhood of A in M3 is again a 3-cell, we
can find a sheet which has no others to one side of it, and there-
fore replace it by a sheet on this side which no longer goes
through A, without introducing any new intersections with the
other sheets. Proceeding in this way, the singularity at A can
be removed, without introducing new singularities.

We see from (a) and (b) that we can replace an E'

with
2
singularities by a non-singular E2 , provided that only touching
sheets occur at singular edges. Our theorem will therefore be
proved when we give a process which transforms a given E' into
2
one whose singularities are all of the above type.

2(a) First, one can remove all n-tuply (n > 2) counted

singular edges by replacing individual sheets by neighbours in
the way one resolves on n-tuple point of a curve into double points.

(b) By modifications of type (1) we remove all touching of

sheets at vertices or edges. This can result in other singular
points, at which two sheets multiply intersect (i.e. there can
be lines of intersection with corresponding multiple points).
Such occurrences are removed by replacing these sheets by neighbours.
Finally, by the same process, we can resolve each n-tuple point
(n > 3) into triple points. These transformations make each
singular point: either a general point on a singular double line,
or an ordinary triple point through which three sheets pass, like
the three cartesian coordinate planes at the origin ("branch points"
 104

are not singular points in our terms, Gince only one [self-inter-
secting] she€t goes through such a point). The singular> lines on
Ei are only: 1. Pairs of singular segments with common endpoints,
which constitute unclosed double lines ending at simple branch
points, 2. closed lines which constitute closed double lines,
either cyclically or pairwise. No single edge (of the non-singular
preimage) can end at a single point of Ei unless another singular
edge emanates from it. Otherwise this point would lie on the
boundary of Ei contrary to the hypothesis: the boundar>y would
meet an inter>ior> point of E2.

(A) We shall first deal with unclosed double lines. Let

R,' and R," be two edges on the non-singular preimage EO of
2
E' with the same initial and final vertices A and B The
2
lines 2' and coincide in E' and constitute a double line
2
2. A and B are the corresponding branch points. 2' and R,"
can have double points and also intersect each other. Let e be
such a point of intersection, so that e = e' on R,' corresponds
to say e" on R,", and we can assume e' and e" are distinct,
otherwise e could be regarded as the end and branch point of the
double line (see Figure 2).

r
18AI:
c·c.
r
8

Figure 2 Figure 3

Now let AF' and AF" (see Figure 3) be two corresponding

non-singular segments of R,' and 2" without a common point; then
we can make a IIswitchover>II along the corresponding segment AF of
2, i.e. join the half sheets of the sheets which meet along AF
 105

in another way, indeed, so that the new sheets touch along AF

and the neighbourhood of A is two-sheeted. This can be done
by connecting each half sheet with that half of the other sheet,
which is not separated from it in E~ by either of the lines
AF' or AF" (see Figure 3, where this joining is indicated by
different types of arrows). As a result, the point A, which
was previously counted singly, becomes double; the double point
F becomes single. Thus the number a O of vertices of E Z
remains unchanged, and hence also its characteristic, and since
the new surface does not fall apart, as e.g. a glance at the figure
shows, it remains simpLy connected. From the point F there is
another unclosed double line k, which however is not necessarily
identical to a part of ~. Since the intersection points of the
segments constituting the double line do not themselves correspond,
we can repeat this process until only a segment DE = {(DE)',(DE)"}
of the double line remains, on which there is no intersection point
of the pieces (DE)' and (DE)" and no double point of them.
(DE)' and (DE)" are thus non-singular segments without common
points.

"()0
It is now easy to see that we can cut sheets along this double
line, by switchover of sheet halves, so that the neighbourhoods of
D and E remain one-sheeted. The
sheets to be joined are indicated
by like shading in Figure 4. Again
the number of vertices is unchanged,
the new surface
is connected, and hence simply
connected. Thus if we remove the Figure 4
touching along AB, using (I), we have removed the singularity of
~ without introducing new singularities or changing the connecti-
vity. In this way we can replace Ei by a singular disc E2 with
the same boundary curve, in which the only singularities are closed
double lines.

(B) It still remains to remove closed double lines. We collect

them into groups, where a group consists of the segments in a double
line" together with all segments of the singular lines through points
 106

of the first double line, plus all segments of singular lines through
singular points of the latter family of lines, etc •. The segments of
a group thus constitute a series of closed curves on the non-singular
preimage of the disc. Thus if we have any two points of the disc which
do not lie on these lines, then all their connecting segments have
either an even or odd number of points in common with each line. We
now take any pair of corresponding segments of the lines in a group
and switch over the half sheets along them in such a way that half
sheets are connected when any two of their points have a connecting
segment which meets the lines of the group in an odd number of
points. We now claim the following: when we continue each switch-
over sufficiently far along one side of the segment, we come to all
segments of the group, with the exception of a pair of corresponding
segments on which no singular point lies. The disc becomes a two-
or one-sided surface with the number p = 1 resp. k = 2, on which
the remaining segments constitute a closed, non-singular, non-
separating, two-sided curve. If we cut the surface along it and
join the segments of the two boundary curves together, then the
switchover is completed for all lines of the group, i.e. without
introducing new singularities we have converted the disc into a
new, simply connected manifold in which the singularities along
the lines of the group can be removed by 1. Thus the lemma will
be completely proved by a proof of the above assertion.

1. We show that continuation of the switchover leads to all

segments of the group: if we continue the switchover through the
next singular point, then we obviously obtain a connected manifold,
in which the outgoing singular line from the endpoint of the
switchover has changed: before the switchover it ran a second time
through the singular point. Then, as a result of the switchover,
the two branches through the
singular point other than the
switchover segment are them-
selves switched over, as is
shown in Figure 5 by the dotted
lines. Thus if we now come
from the endpoint of the
switchover segment back to
the singular point we must Figure 5
 107

again remove ourselves from it on the other branch. This other

branch be10ngs to a second double line, belonging to the same
group. By traversing the latter we again come back to the
singular point (N.B. the line is indeed closed) and depart from
the point on the segment still remaining, which belongs to the
first singular line. If we follow this further, we come back to
the initial point of the switchover segment. Thus we see: the
singular line running from the initial point of the switchover
segment consists of the remaining part of the first singular line,
and the other line which emanates from the singular point, and
for this new line the point in question is no longer singular.
Proceeding in this way, we see that by continuation of the
switchover and running through all lines of the group we finally
obtain the singular line emanating from the endpoint of the
switchover as a segment, ending at the initial point, which has
more singular points. New singular segments do not occur.

2. The surface obviously remains connected under the switch-

over. The multiplicity of all vertices remains the same, except
at the initial and final point of the switchover, which were double
and have now become single. Thus the new surface has the number
p = I resp. k = 2

3. Suppose A'B' and A"B"

(see Figure 6) are the non-
singular segments along which we
have begun the switchover. But
the latter is directed in such a
way that we can connect two
points p' and p" on the
sides of B'A' and B"A" which
go into each other by a poly-
gonal path IT which has an odd
number of points in common with Figure 6.
the edges of the group which remain singular. If we now continue
the switchover from the next singular vertex S say to (e' ,e") ,
then we can find a curve IT' which also has an odd number of
 108

points in common with the remaining edges of the group; we need

only replace each segment ¥:i of II cutting B'C' or B"C"
by the segment pair YC',C'X resp. YC", C"X, which has three
points in common with each segment. The result of this process
on II is again a cZosed curve with the desired property. Pro-
ceeding in this way we finally are left with just two non-singu-
lar segments AZ of the group, and we can find a closed curve
lID which has an odd number of points in common with both these
segments. These therefore constitute a non-separating curve on
the surface.

4. We have finally to prove that the curve consisting of

the two segments AZ is two-sided. We consider the curve con-
sisting of the two segments A'B' and A"B" with which we began
the switchover. This curve is two-sided, its two sides consist
of the two sides of A' B' resp. A"B" . * Thus nothing is changed
when we continue the switchover to the point C over a singular
point. But the curve (A' B' ,A"B") resp. (A' C' ,A"C") is non-
separating, because there is a closed curve II, resp. II', which
meets it in only a single point.
If we continue the switchover
up to the point Z, we there-
fore obtain a two-sided, non-
separating simple curve
which is cut by the curve
formed by the two segments
AZ in two points, namely
A and Z. Consequently A

this latter curve is also

two-sided, which completes
our proof. Figure 7

COROLLARY. We shall use the lemma in the following form:

K aZso bounds a non-singuZar disc under hypotheses which are the
same as above except that singuZarities do Zie upon K in the
E2 given originaZZy, but K does not intersect the interior of
E2 (i.e. does not pass from one side of E2 to the other).
* Thetwo singular segments emanating from A lie on different
sides of the curve (A'B' ,A"B").
 109

In fact this hypothesis suffices for what is necessary to

the proof ~see p,104}- ~b. lrnp055ibility of ~inyular ~~gmQntg
which terminate in the interior of E~ without being attached
to another singular segment •

CHAPTER II Knots and groups

§l Definition
If two figures G and G' of ordinary space transform
into each other by a continuous deformation of the whole space
then in the encyclopedia article G and G' are called iso-
topic in relation to the space. Two non-singular line segments,
discs, spheres or 3-cells [in ordinary space] are isotopic to each
other. On the other hand there are homeomorphic, but non-isotopic,
closed curves, bounded or closed surfaces of higher genus: let
U be the boundary curve of a non-singular disc, K another closed
space curve without singularities. Then K is isotopic to U iff
there is a non-singular disc bounded by K. The two are not iso-
topic in general, namely when K is a knotted curve in the usual
sense.

Definition: A cZosed (non-singuZar) curve K is caZZed un-

knotted iff it is isotopic to the boundary of a non-singuZar disc.

In the third paragraph of this chapter we shall give a method

of deciding, for each topologically defined space curve, whether it
is knotted or not.

While the use of infinite groups is necessary later, we shall

first of all develop purely geometric properties of knots, which
are also important in what follows.

§2 The knotted tube

The boundary of the neighbourhood of a closed space curve K
is a ring-surface R (a tube). We find it as the boundary of a
I 2 I 2
domain J consisting of two 3-cells E3 and E3: E3 and E3
2
have two discs in common, and E2 , and the space curve K

consists of two segments which connect two points of and

2
through the interiors of and ~. We notice that each non-
 110

self-intersecting curve on R, which intersects the boundary of

E~ and E; only once (it does not separate R) bounds, to-
gether with K, a singularity-free strip in J and hence is
isotopic to K in J. Thus if K is knotted, so is each such
curve. The domain which, together with the interior J of the
tube, makes up the whole space, will be called the complement
I
space A; further, the boundary of
E2 , which by construction
bounds a simply connected piece of surface in J, will be denoted
by ~ ~ We shall show that among the curves which cut ~ ~

once there is one which is bounding in A (null homologous). In

fact: if we add to A the 3-cell E; then A becomes a 3-cell,
in which each closed curve is null homologous. Thus each closed
curve of A is homologous to a number of curves on the boundary
of EI3 . However, since only the boundaries of E~ and E; are
not bounding in A, and both are homologous to::J , i t follows
that each curve in A and on its boundary is null homologous in
conjunction with a multiple of ~ Thus if C is an arbitrary
curve on the tube cutting ~ once, then in A

C- n.s"
so C-n'J-O.

But one can always find a singularity-free curve

~. ~
~ ,cutting
~ once, which is homologous to C- n~. It follows, then,
that

v. 0

in the complement space A, as claimed. Indeed, the piece of

surface bounded by tx in A can be simply connected iff K
is unknotted. For in this case, by addition of the singularity-
free strip described above, we can obtain a disc bounded by K
without boundary singularities and hence, by the lemma, without
singularities at all, so that K is unknotted. We shall come
back to this in the next paragraph.
 111

§3 Construction of the fundamental group of a knot

Each line segment complex yields (see Chapter I, §2) a
multiply connected surface: the segrn~n~~ i~~ ~urrounded by
tubes which connect the vertices. We shall consider these sur-
faces specially for the plane projection of a knot. E.g. the
simplest projection of the simplest knot, the trefoil knot,
yields a surface of genus 4. In general a projection with
n - I crossings yields a surface F of genus p =n .

In order to have a definite intuition as a basis for the

following discussion we think of the surface F symmetrically
in relation to the plane of projection. In addition it can be
placed so that it has n + I curves, in common with the plane
of projection: the outer contour, and the contours of the n
compartments. We transform the latter curves on F by double
segments, from a common point P on (say) the upper side of
the surface to points on the inner contours and back again, which
are disjoint from each other and the contours except at these
points. We denote the n curves so constructed through P on
the surface by Cn+I,Cn+2"",C2n' We put another n curves
CI"",C n through P so that Ci outs the curve Cn+i in P
(i.e. goes from one side of Cn+i to the other), Ci does not
cut the curve Cn+h (h ~ i) and two curves Ci and Ck have no
point in common outside of P. Then the surface is modified into
a polygon by cutting along CI , ... ,C 2n , the boundary of which con-
-1 -1
sists of the curves CI'Cn+I'C I ,Cn+l , ••. , the notation indicating the
given direction of passage. Each curve L on F is equivalent to
a number of curves Ci in a definite order, as shown by an expression

E
Cm
~
m

If we construct the infinite (fuchs ian) group with the generating

operations CI ,c 2 "",C 2n and the single relation
 112

then each element of the group corresponds to a curve on F and

conversely; further, each element of the group capable of being
proved =1 from the relation corresponds to a curve bounding a
simply connected piece of the surface and conversely.

Now if we construct a new group with the same generating

operations and

Cn+l = Cn+2 = ••• = C2n = 1

then the above relation follows and further, each curve on F

whose corresponding substitution in the new group = 1 bounds a
disc in the complement space of F, since this is the case for
Cn+l , ••. ,C 2n . Conversely also, each surface curve L which
bounds a disc in the complement space corresponds to a substitution
of the group which is equal to 1. In order to see this one con-
structs a non-singular disc in the complement space bounded by
Cn+l ,··· ,C 2n . If this is taken double, then together with F it
constitutes a sphere. A disc E, whose boundary is L, and which
we can assume singularity-free by the lemma, intersects this sphere,
if at all, in closed curves. Each of the pieces into which E is
decomposed may be replaced by a piece of the sphere and thus
results in a disc bounded by L which is comprised merely of
pieces of a sphere. However these are bounded by the curves
Cn+l ,···,C 2n or t h e curves C1 Cn+l C-l
1 C-1 f h· h ~t
n+l .•. , rom w ~c •
follows that the element of the group corresponding to L is
equal to 1. Since we can omit generators which are equal to 1
from a group, we obtain the curves on the surface F in relation
to the complement space as those generated by Cl '· •• ,Cn ' between
which there are no relations.

We shall now add to the com-

plement space of F n - 1 3-
cells, corresponding to the n - 1
crossings of the projection figure:
Let ABCD (see Figure 8) be a
curve around one of the crossing
points on the surface, and indeed
let the points A,B,C,D lie in
such compartments as correspond to
a branch through the under side of Figure 8
 113

the crossing. Further, the segments AB and CD shall lie on

the upper side of F, BC and AD on the lower side. Let
A'B'C'D' be a neighbouring curve of ABCD on F. Then we
remove from the interior of F a 3-cell bounded by the strip
between them on F and two discs bounded by them inside F
and add it to the complement space. We proceed similarly with
each other crossing. We then obtain a complement space bounded
by a ring surface R knotted in exactly the same way as the
original curve K.

The n - 1 curves ABCD etc. are in the complement space

of F, and hence reducible to curves of the form

1,2, ... , or n)

as shown above. We shall denote the n - 1 curves in this form

by Sl"",Sn_l'

In relation to the complement space A of R the curves

of R then have the group:

generating
GK {
relations:

This group GK wiZZ be caZZed the fundamentaZ group of the

cZosed space curve K.

We shall now explain how one may obtain the n - 1 relations

in the simplest way directly
from the projection figure: let
(Figure 9) ml ,m 2 ,m3 ,m4 be the
numbers of the compartments
which meet at the crossing
point, in the order ml m2m3m4
around it. Suppose now that
ml and m4 , likewise m2
and m3 , meet along the lower
branch, then the reZation
corresponding to the crossing m Figure 9
 114

S
m

when the sense of the Ci is suitably chosen. For the crossings

which occur at the outer region the term corresponding to this
region is simply left out. The proof of this assertion is easy
with the help of theorems on the reduction of curves on surfaces.

All relations of the group GK are thus three- or four-te~ed.

In this way we obtain the fundamental group

of the trefoil knot (Figure 10):

generating elements: Cl ,C 2 ,C 3 ,C4

{ -1 -1
Cl C4 C2 C2 C4 C3
relations {
-1
C3 C4 Cl 1

Figure 10

§4 Unknotted space curves

An unknotted space curve may be characterised as the boundary
of a non-singular disc. Since the ring surface R is knotted like
the given curve K, the curve 1.X on R which is bounding in the
complement space is knotted or unknotted according as K is. If
K is unknotted then <1 bounds a non-singular disc in the com-
plement space A. Thus if SiX is the substitution representing
iX in the group GK , Su must = 1. Conversely: if S1); = 1
then in the complement space 'ZX bounds a disc which, since t(
lies on the boundary of A, can have no essential boundary singu-
larities. Consequently 1). also bounds a non-singular disc in A,
by the lemma. Then we have the theorem.

THEOREM 1: For a space CUI'Ve K to be unknotted it is necessary

and sufficient that a certain substitution S~ equal 1 in the
group GK corresponding to K.
 115

In the case that K is unknotted, and hence also the tube

R, all curves on R are reducible in the complement space to
powers of a single curve which loops around the bounding curve
,....,
,J' on R. So in this case GK is isomorphic to the (abelian)
group, generated by a single operation, without relations. Con-
versely, if GK is an abelian group, it follows that K is un-
knotted. For since -v:
is null homologous in the complement
space it follows that

n n-l
U - L \I,
1-
C, -
1-
L A,
1 J
s,
J
1

where Sj stands in place of the curve on R corresponding to Sj

But it follows from this congruence, if changing the order of sub-
stitutions is allowed, that

SiJ: = 1

so that the sufficient condition of Theorem 1 has been reached.

We therefore have

THEOREM 2: K is un7<Ytotted iff the fundamental group of K is

abelian.

If a knot group is abelian, then it is isomorphic to the

group {Sa} .

One can apply yet another method for recognising unknotted

curves: let Jt be any singularity-free curve on the ring surface
R satisfying the homology

(Each curve 11
......
which cuts it only once satisfies a homology of
this type.) "s (the curve bounding in the interior) can be represented
by substitution C, (where i is the index of a compartment which
1-
has a segment in common with the outer boundary). Let 1l be
represented by the substitution Sp.: of GK . Now if GK is
abelian, then
 116

S'V: 1 and S}l = Ci .

If one now adds to A a slice of J bounded by a strip lying along

~, then A becomes a 3-cell for any knot. However the group of
the curves on R relative to this 3-cell consists of
the additional relation Ci = 1. But since all curves on
bound a disc in this 3-cell the group which results from
addition of Ci = 1 must be the identity. Consequently GK must
also become the identity, when K is unknotted, when one adds
S~ = 1. However if K is knotted, this will not be the case in
general, as we shall show in the next paragraph. On the contrary,
knotted tubes give rise to manifolds of the kind discovered by
Poincare (Pal. Rend. 1904), and which we shall therefore call
Poincar~ spaces. These are [homologically] simply connected,
closed M3 without torsion, i.e. each closed curve is bounding
in M3 when traversed once. But they are in general not homeo-
morphic to ordinary space. Such a manifold cannot, in this general
case, have the identity as fundamental group, as will be proved in
the last chapter. Consequently we have (see construction of the
fundamental group in M3 , Chapter III, §3).

K is knotted if addition of anyone of certain reZations

S1t = 1 to the fundamentaZ group GK does not result in the
identity.

§5 Special knots and Poincare spaces

(a) The trefoil knot
As we showed above, the fundamental group of the trefoil knot is:

generating substitutions:
{
. -1
relat10ns: Cl C4 Cz = Cz 1

The group diagram (see Chapter I, §l) is placed on

a strip (Figure 11) bounded by lines consisting of
segments C4 • The two parallel chains of seg-
ments C4 are now connected by a polygonal path
C3CZClC3CZCl etc. so that the vertices lie al-
ternately on the two C4-chains. In this way we
obtain three kinds of vertex on the strip:

Pig. 11.
 117

-1
where Cl and go out, where C2 and C3 go out, and finally

where C3 and go out. We now attach to the boundary of this

strip two more strips, running in the same direction, so that all
three types of vertex come together at each vertex of the common
boundary. We only need to make sure that this is the case for a
single vertex, in order to obtain it for all vertices. At each of
thB thrgg frgg hou~dari~s of this triple strip we attach two further
strips in the same way, and so on. We easily see that the resulting
infinite line segment complex is the group diagram of the given
group. For emanating from each point we have eight segments

and hanging from each of these segments are the two resp. three
circles which correspond to the two resp. three relations in which
the corresponding substitution appears.

We now see immediately that the group is not isomorphic to the

group {Sa}, so that it is not abelian. For example the polygonal
-1 -1
path Cl C4 Cl C4 is not closed.

The curve tr on R corresponds to the substitution

-1 -1 -1
St( = Cl
C2 C3 C2 C2 C2 . We see immediately from the group
diagram that S~ is represented by an open polygonal path, so
tx does not bound a disc in the complement space, but only a
surface of higher connectivity, as it must, since K is knotted.

We shall now consider the Poincar~ space of the trefoiZ knot

(see p.116) : a curve:n of the ring surface which cuts the curves
<.X andCl once is represented by the substitution
-1 -1 --11-
Sil. = Cl C2 C3 C2 C2 . If we add a disc along J~ to the comple-
ment space we obtain Poincar~ space ~ bounded by a spherical
surface. Since all curves of the complement space are reducible to
curves of the ring surface, the group of this space is equal to the
group of curves on the ring surface relative to ~, i. e . it is
the following:
 118

relations:

We shall construct the group diagram: for this purpose we

consider the following
dodecahedral Figure 12.
The arrows and numbers
associated with seg-
ments mean that the
segments, taken in the
direction of the arrows,
represent the genera-
ting operation with the
index corresponding to
the number. The figure
consists of nine penta-
gons, one heptagon and
one decagon and the
boundary polygon. Each Figure 12
polygon represents, when traversed in a definite sense, either the
left hand side of the fourth relation or also this expression trans-
formed by one of the first three relations. We note that the four
relations yield the relation

(which corresponds to the boundary polygon). On the other hand, by

transformation of the fourth relation with the help of the three
others we obtain:

from which it follows that

7) 1.
 119

Before we use these relations we shall derive something else

in conjunction with the assertion that the relations corresponding
to Figure 12, as well as 6), follow from 4) by means of 1), 2), 3).
For this purpose we construct a part of the group diagram of the
group of the trefoil knot in which the
expression on the left hand side of
4) is represented. This part con-
sists of five strips (see Figure
13). It follows from the re-
lation 4) that the two free
boundaries coincide, and in-
deed in such a way that the
end points of the polygons
representing expression 4)
coincide, and correspon-
dingly the other points
of both boundaries. It
then follows by direct Figure 13
consideration of the figure that the other relations hold as
claimed: the polygons representing them end in corresponding
points of the two boundaries. (In Figure 13 the label of a seg-
ment is given on the segment which results from it by horizontal
displacement. Of the segments C4 only those on the boundary are
represented, and the direction of only one of these is given.
Relation 4) is represented in the figure by a heavily drawn line,
those relations corresponding to polygons of the dodecahedral figure
are represented by dotted and half-dotted lines respectively, while
that of relation 6) is represented by a wavy line.)

Using relation 7) we now obtain the following construction of

the group diagram: we construct a ring surface from the five strips
represented in Figure 13, in which we bring the two free boundaries
into coincidence, corresponding to relation 4); the relation C~ = 1
further corresponds to any chain of six segments C4 being closed.
In this way we obtain a ring surface. Each polygon of the
dodecahedral figure is a non-separating closed curve on this ring
surface; we can therefore pass such a ring surface through each of
 120

the polygons. Since three segments meet at every vertex of the

dodecahedral figure, and the segments correspond to three different
vertices, then in the totality of points and strips on the 12 ring
surfaces which meet pairwise along the length of a strip and triply
at a corner we have the desired group diagram. The group of our
Poincar~ space is therefore finite, it consists of 120 substitutions,
for there are 30 vertices on each ring surface, but each vertex lies
on three ring surfaces, so we obtain the number of substitutions as
12'30 = 120. This number suggests the conjecture that the group is
3
isomorphic to the icosahedral group extended by reflection, a con-
jecture which is easy to verify. In fact cl ,C 2 ,C 3 may be regarded
as rotations about three vertices of an icosahedron combined with
reflection in the midpoint, C4 as the rotation about the midpoint
of the icosahedron combined with reflection in the midpoint.

The fundamental group of the Poincare space corresponding to

the trefoil knot is isomorphic to the order 120 group of the ico-
sahedron extended by reflection.

It is easy to find the groups of other Poincare spaces corres-

ponding to the trefoil knot. They are, with the natural exception
of the trivial case of ordinary space, all infinite. The simplest
case is obtained when one adds to the GK of the trefoil knot the
relation:

Geometrically, this means that one adds to the complement space an

elementary piece of space along a curve cutting ~ twice and -[l
twice. The resulting Poincare space has the group which is given
by the generating operations with the relations 1), 2), 3), 8).
With the help of the relations we can transform 8) (analogously
to the previous transformation) into

We see that the group diagram is constructed as follows: We let

the boundaries of a sheet of 11 strips coincide and thus obtain a
cylinder. Further, we take a regular net in the Lobatchevsky plane,
consisting of ll-gons which meet triply at each vertex. We put
 121

cylinders through the interstices of the net, having strips pairwise

in common and vertices triply in common. The cylinders are positioned
in such a way that the three strips at a corner coincide just as they
did in the group diagram of the trefoil knot. It is not difficult to
see that this is possible. The vertices and edges on these cylinders
give the diagram of the given group, which is therefore infinite.
Similar statements are true for the Poincare groups generated by the
curves of the ring which cuts ~ n times and <X once. We have
the result:

All Poincare spaces corresponding to the trefoil knot, with

the exception of ordinary space and the space studied above have
infinite groups, whose diagrams are generated by regular tessella-
tions of the non-euclidean plane in which three (6n-1)-gons meet
at each vertex (n > 1).

In order to present the meaning of the group diagram more

strongly, we remark that each closed curve on the diagram
corresponds to a substitution which is =1, and to a curve on the
manifold contractible to a point, i.e. one which bounds a simply
connected piece of surface. The above arrangement of the group
diagram not only helps to prove that space curves are knotted, or
that the spaces are not homeomorphic to ordinary space, but it
also solves the problem of deciding, for· a given curve of the com-
plement space or Poincare space respectively, whether it is con-
tractible to a point or not.

(b) Other knots. Here we shall only deal with the groups
of a few especially simple knots, closely connected with the
trefoil knot. The first two of the series are shown in the
figures 14, 15, the others are obtained by proceeding analogously.

Figure 14 Figure 15
 122

They have 5,7, ••• ,3+Zn crossings and go over to the trefoil knot
by 1,Z, ••• ,n switches of a crossing point. One easily sees from
the rule given in the above paragraph that for n =1 the group
is given by

generating operations:
{
-I
relations: { Cl C6 Cz

The group diagram consists of similar strips

as for the trefoil knot (see Figure 16), any
five strips meet at a C6 chain. Similarly
for n > 1. If we add to these relations,
in the case n I, the new relation

then the group represents a Poincar~ space

corresponding to the knot. One sees that
the group diagram is generated by cylinders Figure 16
made from nine strips, which meet five-fold along the C6 chains.
The cross section of the cylinders corresponds to a regular tessel-
lation of the hyperbolic non-euclidean plane in which five 9-gons
meet at each vertex. A similar result holds for n > I. Thus we
have:

The Roinaa~ spaaes aOPresponding to the above knots have

simiZar infinite groups, ~hiah are generated by reguZar tesseZ-
Zations of the hyperboZia pZane.

Outside of the icosahedral group and the identity we have

thus obtained only infinite groups for Poincar~ spaces. For
manifolds with torsion the only known finite groups are cyclic.

Further investigation of knot groups, in particular as well

as general, i.e. the investigation of properties characteristic
of all knot groups, will be left to later works.
 123

CHAPTER III Three dimensional manifolds

§l General (construction of the seam complex and the fundamental group)
Let M3 be a closed three-dimensional manifold, S~ an arbitrary
constituent piece of the same, TI~ the spherical surface bounding
1 2
S3' Let S3 be a constituent piece of space that has at least one
1
disc in common with S3' Using Sl3 and S32 we construct a 3-cell

~~ in which we join and together along one common disc.

The spherical surface bounding ~; consists only of faces which

belong to M3 , but a pair of such faces on TI~ may correspond to
only a single face in M3 , namely, when S~ and S~ have more
than one disc in common. We now attach to ~; a third piece of
space of M3 , which has at least one disc in common with

and in fact we attach along such a face.~2 and S3 together con-

3 3
3 3
stitute ~3' with boundary TI 2 . We proceed in this way until we
obtain a 3-cell ~3 which contains all the constituent pieces of
space of M3 , and is bounded by the spherical surface TI 2 • Each
face of TI2 also appears in M3 , but a pair of faces of TI2
correspond to one and the same surface in M3 , since M3 is
closed by hypothesis. We call TI2 a "aut surface of M3"·
M3
is therefore homeomorphic to a 3-ceZZ bounded by a cut surface when
the faces of the Zatter are identified in the way resuZting from
the above construction (first general method of generating M3 ).
The relationship of the faces of il Z to one another also gives
their orientation. From the definition of two resp. one-sidedness
of multidimensional manifolds we immediately have the result: if
two corresponding surfaces of TI2 always have different orientations
in relation to TI Z given by the above procedure then M3 is two-
sided, otherwise one-sided. So the TI2 gives us a convenient
means of deciding this first question about the character of M3 .
We now let corresponding faces of the cut surface TI2 coincide
and obtain a surface complex N2 , which we call the seam surface
of M3 . This notation indicates that one obtains the original M3
from a spherical space by stitching together along N2 . In the
 124

case of two dimensional manifolds a cut system corresponds to

Cz ' by means of which MZ may be transformed into a simply
connected piece of surface. We therefore have the following
result:

(I) Each line segment complex and each surface complex

of is homotopic to a line segment resp. surface complex
of the seam surface (i.e. it may be continuously deformed into
one, allowing self-intersection), because each line segment or
surface complex not in the seam surface NZ lies in the
spherical space E3 and thus is homotopic to a complex on the
boundary lIZ and hence to one such on NZ • Hence if a curve
(and also a curve complex) bounds a disc resp. a piece of surface
of higher connectivity in M3 , then a corresponding curve (or
corresponding curve complex) of NZ bounds a disc resp. surface
of higher connectivity on NZ ' and conversely. If two curve
systems on NZ are not continuously deformable into each other
this is also the case for M3 ; in particular if a curve on NZ
is not contractible to a point (i.e. bounds no disc of NZ )'
then it is also not contractible to a point in M3 . For this
reason we define the fundamentaZ group of N2 to be the funda-
mentaZ group of M3: GN = GM . The construction of GN
2 3 Z
through representation of bounding circles by a suitably chosen
system of curves through a point on N2 is developed in Chapter I,
§2.

Underlying NZ we have a line segment complex which we

shall denote by Nl The neighbourhood of this complex, in
the case where M3 is two-sided, is bounded by a two-sided
closed surface of genus p = ~ = vI - Vo + 1 (VI: number of
segments, vO: number of vertices of Nl ) (see Chapter I, §Z).
This neighbourhood J3 consists of a 3-cell with p further
pieces of space ("handles") attached. Thus it is representable
in our ordinary space. We can call it the ordinary two-si-
ded muZtipZe ring space. Likewise, the part of M3 which
remains when J3 is taken away, the "complement space" A3
 125

of Nl , will be such an ordinary multiple ring space, as follows

immediately from the construction. In the case that ~ is one-
sided, J 3 and A3 are bounded by a one-sided surface of even
characteristic. J3 and A3 constitute ordinary one-sided
mUltiple ring spaces. Thus we have a second method of generation:

The general homogeneous closed M3 is obtained by fusing the

surfaces of two ordinary multiple ring spaces.

It follows from this that

Each homogeneous closed M3 is decomposable into four non-

singular 3-cells.

Each curve of the space is isotopic to a curve on the

surface F which separates A3 and J 3 • Then without going
any further one can derive the Poincare theorems on the relation
between curves and surface connectivity, which we will not pursue
further here. It need only be remarked that these considerations
may be immediately extended to any number of dimensions, so that,
among other things, it follows that any homogeneous Mn may be
decomposed into four n-cells, so there is also a simpler proof of
the corresponding Poincare theorems for an Mn (n > 3).

§z Ordinary space
One can see immediately that the fundamental group of ordinary
space is the identity, since each curve in space bounds a disc
(possibly with singularities), so that the same is true of each
curve on NZ ' from which it follows by Chapter I, §Z, that the
fundamental group of NZ ' and hence M3 , is the identity.

If conversely the fundamental group GN of NZ is the

Z
identity then each curve on NZ is bounding. Under this hypothesis
NZ contains no closed two-sided surface, from which it follows that
no open segment of NZ alone bounds a disc. Then if Nl contains
an open segment we can remove this by pulling together, i.e. if we
insert discs in the modified Nl by suitable circles, then these,
taken doubly, again constitute a sphere, and if we identify the
 126

two halves of the corresponding 3-cell boundary in the manifold,

then we obtain a homeomorph of the given M3 • In this way we
can remove all open segments of NI : the latter then consists
of pure circles with one common point O. Now if
GN is the
identity, each of these circles bounds a disc, whose only singular
lines are the circles through O. If we could now show that none
of these circles had essential boundary singularities (see
Chapter I, §3), then it would follow immediately from the lemma
that there is a system of non-singular discs which each had one
of the circles as boundary and no point in common apart from O.
However, the neighbourhood of such a surface complex in a homo-
geneous M3 is a spherical space. Thus we can embed NI in
i
a spherical space E3 . But, the boundary KZ of this
spherical space lies in a suitably chosen complement space A3
~
and in this bounds a 3-cell E3 , by the argument of the previous
paragraphs. Because each closed surface is bounding in the
ordinary multiple ring space, in particular a spherical surface
bounds a 3-cell and indeed this holds whether or not we use the
assumption that the ring space is two-sided. However the E~
i
and E3 , when fused along the bounding spherical surface K,
result in the given M3 , and this is consequently homeomorphic
to ordinary space. However our assumption of the non-existence
of essential boundary singularities is not yet established.
Therefore it is necessary to study further the cases in which GN
is the identity.
 TRANSLATOR'S INTRODUCTION 4

This famous and influential paper is a memorable intro-

duction to topology, combinatorial group theory and non-

euclidean geometry. To see Dehn weave these threads into a

coherent pattern is an unforgettable demonstration of the

unity of mathematics. Of course, there is a historical basis

for this coherence in the work of Klein and Poincare in the

1880's, and the relation between surface topology and hyperbolic

geometry had already been used with considerable sophistication

by Poincare (1904), but Dehn's paper is the first, I believe,

to show equal sophistication in the treatment of combinatorial

group theory. In particular, it is the first in which a purely

combinatorial algorithm is justified by appeal to the hyperbolic

metric.

The result in question is Dehn's first combinatorial

algorithm for the conjugacy (transformation) problem, and his

method of proof is probably its most important part. The

algorithm itself seems, if anything, less efficient than the

analytic algorithm he wants to supplant, and in any case he

found a much better algorithm later (Dehn [l9l2b)), with a

purely combinatorial proof.

Suppose that c, d are conjugate elements of the fundamental

 128

group ~l (i) of i, 50 that

c = ede- l

for some element e £ ~l (i) . We then have a quadrilateral

(Fig. 1) in the group diagram

d

Fig. 1
e

of ~l (i), the sides of which are edge paths (not, in general,

single edges) whose labels spell out words for c, d, e as

indicated. In general we call a path whose labels spell a word

for g £ ~l (i) a g-path, and a sequence of such paths laid end

to end a g-chain. We know that the endpoints of successive

g-paths in a g-chain lie on a distance curve, hence in the

situation of Fig. 1 we get a c-chain and d-chain running along

distance curves with the same ends R, S on at, as in Fig. 2

(Cf. Fig. 1 in the introduction to Dehn's 1910 lectures on

surface topology. The present c-chain corresponds to the

lift chain we called ... c(-l)c(O)c(l) ... there.)

 129

Figure 2

Now if we are given any c, d as words in the generators

of wI (F), and if we want to know whether there is an element e

as in Fig. 1, then hyperbolic geometry suggests that we can

decide this by searching within a bounded hyperbolic distance of

any c-path. Namely, it is a property of the displacement which

maps a c-chain onto itself that all distance curves through its

ends on aF are also mapped onto themselves and that the length
of displacement along a given distance curve increases beyond

all bounds with the distance from the axis. Therefore, if we

go far enough to either side of the c-chain we shall reach a

distance at which the displacement exceeds the length between

 130

the endpoints of d, beyond which it is useless to search for

a d-path connecting ends of identically labelled paths issuing

from the ends of a c-path as in Fig. 1. This is the first idea

underlying Dehn's proof.

The second idea is that the word distance between vertices

A, B in the group diagram (i.e. the minimum number of edges in

a path connecting A and B, which is the number of letters in

the word labelling the path) is roughly comparable to the hyper-

bolic distance between A, B. This is formalised by Dehn's

lemma which states that, for suitably chosen unit of hyperbolic

length,

word distance < 4p(p+l) x hyperbolic distance

where p is the genus of F. It follows that we can use the

lengths of the words c, d, rather than the hyperbolic metric,

to compute a bound on the (word) distance we need to travel on

either side of the c chain in search of a d-chain tied to the

c-chain bye-paths as in Figure 2. The search is conducted

by emitting identically labelled paths, e', from the ends of a

c-path. If the e'-paths grow together, one edge at a time,

then they will hit the d-chain simultaneously, at which moment

 131

there will be a d'-path connecting their free ends, where

d' is a cyclic permutation of d. We can therefore observe

contact with the d-chain by checking at each stage whether

a d'-path joins the free ends of the e'-paths. If contact

does not occur within the computed distance, then the required

d-chain does not exist.

The computational catch in this algorithm is that the

e'-paths are found essentially by trying all paths out of the

ends of a c-path until one reaches a vertex whose displacement

under the motion of the c-chain is sufficiently large. Since

the number of such paths grows exponentially with the length

of the word c, the algorithm is not computationally feasible

as it stands. Nevertheless, the proof captures a very

interesting relation between analytic and combinatorial facts

the rough comparability of the word metric and the hyperbolic

metric - which might well have been lost had Dehn thought of

his purely combinatorial proof sooner. This relationship

remains important and has been taken up again recently, e.g.

by Floyd [1980], Cannon [1984] and Gromov [1981].

REFERENCES

J. Cannon [1984] : The combinatorial structure of cocompact

discrete hyperbolic groups. Geom. Dedicata, 123-148.

 132

M. Dehn (1912b] : Transformation der Kurven auf zweiseitigen

Flachen. Math. Ann. 72, 413-421.

w. Floyd [1980) : Group completions and limit sets of Kleinian

groups. Inv. Math., 57, 205-218.

M. Gromov [1981) : Hyperbolic manifolds, groups and actions.

Riemann Surfaces and Related Topics (Eds. I. Kra & B. Maskit),

Princeton Univ. Press, 183-213.

H. Poincare [1904) : Cinquieme complement a l'analysis situs.

Rend. eire. mat. Palermo 18, 45-110.

 ON INFINITE DISCONTINUOUS GROUPS

INTRODUCTION

The three fundamental problems for infinite discontinuous

The general discontinuous group is given by n generators and

m relations between them

RI (S. . .. ) 1
~l

as first defined by Dyck (Math. Ann., 20 and 22). The results of

those works, however, relate essentially to finite groups. The general

theory of groups defined in this way at present appears very undeveloped

in the infinite case. Here there are above all three fundamental

problems whose solution is very difficult and which will not be possible

without a penetrating study of the subject.

1. The identity problem [word problem, trans.] : An element of

the group is given as a product of generators. One is required to give

a method whereby it may be decided in a finite number of steps whether

this element is the identity or not.

2. The transformation problem [conjugacy problem, trans.].

Any two elements Sand T of the group are given. A method is

 134

sought for deciding the question whether Sand T can be trans-

formed into each other, i.e. whether there is an element U of the

group satisfying the relation

-1
S UTU

3. The isomorphism problem: Given two groups, one is to decide

whether they are isomorphic or not (and further, whether a given

correspondence between the generators of one group and elements of

the other group is an isomorphism or not).

These three problems have very different degrees of difficulty.

Problem 1 is a special case of problem 2. The first obviously

concerns the character of one element of a group, the second the

relation between two elements of one group, the third the relation

between the elements of two groups. It may be presumed that the

solution of these three problems would constitute a natural foundation

for the methodical presentation of the theory of infinite groups.

However, one is already led to them by necessity with work in topology.

Each knotted space curve, in order to be completely understood topo-

logically, demands the solution of the three above problems in a

special case : each curve K is unknotted iff GK is an abelian group,

which leads to an identity problem*) . Any other curve in space in

relation to K corresponds to a definite element of GK. Two

space curves are then transformable into each other without cutting K

iff the corresponding elements of GK are transformable into each

*)see Math. Ann. 69.

 135

other. Finally, the question whether a given space curve K' can

be transformed into K without self-intersection requires a solution

of the third problem for GK, and GK.

The appearance in topology also dispels certain objections which

might be raised against the restriction to infinite groups which can be

defined in the above manner. In fact, such objections can arise when

one considers that in general, each infinite group has subgroups which

cannot be generated by a finite number of generators and relations between

them. [Actually, there are infinitely many exceptions, such as ~n - trans. 1

Consider, say, the group which is given by two generators 51 and 52

without defining relations, and collect all those elements which result

from 51 by composition and transformation in the group, e.g.

These elements constitute a group, and indeed

a normal subgroup of the given group, whose elements cannot be generated

by any finite number of them. This group therefore does not come under

the heading of the groups defined above.

Each of the three problems for finite groups is solvable without

further effort and therefore meaningless. If one confines oneself

to analytic groups, i.e. groups of transformations in an n-dimensional

real or complex number manifold, in the case where the generators are

given directly by certain transformations, the first problem is solved

in most cases without further consideration. The first problem

may likewise be solved by constructing the group diagram

 136

corresponding to the given group, i.e. the infinite line segment

complex which was introduced in the work in Math. Ann. 69 and also

made the basis of that work.

If the group is given as a proper discontinuous group of motions

in euclidean or non-euclidean space, then the second problem is

settled by construction of the fundamental domain associated with the

group. Likewise we obtain the solution of the problem in the general

case by canonical representation of the self-motions of the group

diagram.

The analytic representation is less useful for the solution of

the third problem. Topological investigation of the group diagram

appears to be of essential importance here.

In the present work the three problems are settled for the case

where each generator appears only twice altogether in the defining

relations (Chap. II). In order to attain this, we must begin with

an exact study of the simplest class of such groups, namely the

fundamental groups of closed surfaces (Chap. I). Finally (Chap. III),

a few theorems on the most general groups are given, and a few

examples of higher groups are treated. Among other things, the trans-

formation problem for some knot groups is settled here, the ones

for which the identity problem has already been solved in Math. Ann. 69

by setting up the group diagram.

 137

Chapter I

The fundamental groups of closed surfaces

Strictly speaking. one cannot say that pure group theory is of

great service to topology in the case of fundamental groups of

closed surfaces. On the contrary. the solution of the topological

problem bears fruit for group theory. for in certain simple cases

that solution immediately settles the three fundamental problems of

group theory. and indicates a way that leads to the same goal in many

other cases.

Thus we must first explain topological matters. and in fact these

concern the continuous transformations of closed curves on surfaces.

Here we have to be somewhat more detailed than would be necessary had

this theory. whose important results are known*. at least for two-sided

surfaces. been carried out anywhere else.

§l. Two-sided surfaces

1. By cutting along 2p circuits which all go through a point

and fall into successive pairs of intersecting curves a l .b l .a 2 .b 2 •...•

a p ' b p ' the closed two-sided surface of genus p is transformed into

a 4p-gon. whose edges are identified with each other in pairs. If

-1
one denotes by a. resp. a. etc. the curve a. with its two
~ ~ ~

directions. then the edges of the 4p-gon give the sequence

*)poincare. Pal. Rend. 1905 p. 14 ff. The great work of C. Jordan

(J. d. Math. 1866). which is entirely devoted to this problem. leads
to an absolutely incorrect result. One also sees clearly here the
substantial simplification brought about by the application of non-
euclidean geometry.
 138

when one traverses the boundary in a definite direction. Each

closed curve on the surface becomes, in the 4p-gon, either a closed

curve or curve pieces 6 62' ...

l' each bounded by two boundary

points of the 4p-gon. The final !Dint of 6i and the initial point

of 6+i l are homologous points on corresponding edges or else both

vertices of the 4p-gon.

2. As one easily sees, one can construct an infinite net of the

following character without making any metric assumptions, i.e. purely

topologically 4p line segments emanate from each vertex, each cell

is a 4p-gon. The edges of the net can be denoted in such a way that

in each cell the series of edges follow each other in the same order

Then at each vertex the line segments taken in the same order are

Of course a segment which is denoted in one sense by a. will be

l
-1
denoted by a. when traversed in the opposite sense. It is
l

suitable for what follows to take the net for p ~ 1 to be a net of

squares in the euclidean plane, and for p >1 to take the net of

regular 4p-gons in the hyperbolic plane.

3. These nets can be viewed more familiarly as mappings of the

 139

two-sided surface covered infinitely often, where any two covering

sheets are connected along one circuit cut. Points of different

sheets which correspond to the same point of the surface will

appropriately determine homologous points, in the metric sense,

of different cells of the net. A connected curve on the surface

determines a likewise connected curve of the net, which enters new

cells as often as the original curve cuts circuits on the surface.

The initial and final point of the image of a closed curve are homologous

points of the cells concerned. Each closed curve of the surface

determines,by its image in the net, a motion of the net into

itself, in which each net segment goes into one with the same

denotation, and the initial point of the image goes into the final point.

Here a definite sense is associated with the curve on the surface.

The net motion which carries the final point to the initial point

corresponds to the curve traversed in the opposite direction.

4. I f the final point coincides with the initial point then the

corresponding net motion is the identity ; for when the denotation

and one point of the net are fixed there is no motion of the

net onto itself except the identity; correspondingly, given that a

point a shall go to a homologous point a', the motion is

uniquely determined.

If the image of a curve on the surface is a closed net curve,

then the surface curve is contractible to a point; for as the

mapping shows, the surface curve bounds a simply connected piece

 140

which in general will cover a piece of the surface with a certain multi-

plicity. Conversely, the image of a surface curve contractible to

a point is a closed curve of the net. For such a curve may be

continuously ("isotopically") transformed on the surface into an

arbitrary curve contractible to a point, the image of which may be

chosen to lie in a single cell and hence must be a closed curve.

But since continuous transformations on the surface correspond to

continuous transformations on the net, the image of the original

curve must also be closed. Since we can decide whether a curve on

the net is closed or not without metric help, the question whether

a given curve on the surface is reducible to a point is directly

answered in this way by pure topology. It is assumed that the surface

curve is given by the sequence of its intersections with the circuit

cuts, and correspondingly the image curve on the net is given in a

similar way by successive points on the edges of the net.

5. The image of a surface curve not contractible to a point

has infinitely many forms depending on the choice of initial point.

If one chooses a certain initial point for the surface curve, but

different (homologous) points as images,then the surface curve corres-

ponds to different net motions, but ones which transform

into each other* by motions of the net. If we choose another

point as the initial point of the surface curve, but retain the

original image of the original final ]Xlint, then the associated motion

of net remains unaltered. If one fixes the initial point of the

* I.e. are conjugate (Translator's note).

 141

surface curve and continuously deforms the latter on the surface,

then the associated initial and final point of the net curve are retained.

The associated net motion is therefore likewise unaltered.

Conversely, all curves bounded by two fixed net points are images

of homotopic surface curves. Now since any continuous transfor-

mation of a surface curve can be assembled from transformations in

which a point of the curve remains fixed, we have the result

Two surface curves are continuously transformable into each other

on the surface iff any pair of net motions associated with the

surface curves are transformable into each other.

Deciding the question whether two surface curves are trans-

formable into each other or not is therefore equivalent to the solution

of a problem concerning the euclidean resp. non-euclidean metric.

Without going further we cannot give a purely topological method for

settling the general problem. This is only possible in the case

p = 1, i.e. in the case where the net lies in the euclidean plane.

Here all motions result from parallel translations in the

directions of the a and b sides of the net respectively. The

motion group of the net is abelian, and two motions are

transformable into each other if they are composed of equal numbers

of a- and b-translations. This leads in an obvious way to a

simple topological resolution of the question.

In the case p >1 we have to deal with motions (without reflection) in the

hyperbolic plane. A single such motion is equivalent to a

 142

rotation about a real or ideal midpoint. In our case it must be

ideal, since there is no motion different from the identity

with a fixed real point, as already remarked. Thus each

motion under consideration is a translation along a real axis.

Then, in order for two such motions to be transformable into

each other it is necessary and sufficient that the corresponding

displacement segments along the axes be composed of homologous

pieces in the same way. Thus if we represent the displacement

segments by segments which each lie in a single cell, then the

motions are transformable into each other iff these unions of

segments are identical. In this way our problem is reduced to an

algebraic problem which is to be settled in a finite number of steps:

in fact, 1. the coordinates of the net points are roots of algebraic

equations; 2. the net motions are representable as linear

transformations with coordinates which are known algebraic functions

of the coordinates of the net points; 3. therefore, the

coefficients of the equation of the displacement axis are known algebraic

functions of those quantities, and the same holds for the coordinates of

the intersection point of the axis with the net sides; 4. the

solution of our problem then comes down to deciding whether certain

roots of given algebraic equations are identical with each other,

which can be settled in a finite number of steps by known theorems.

While the foregoing general remarks are already bringing the

conviction that our problem may be settled in a finite number of

steps, it is still unsatisfactory that the method implied is too

 143

intricate to be of practical value in particular cases. In this

connection the following practical method, based on the foregoing

considerations, is of significance. Its presentation imposes on

us the pleasant duty of exploring more deeply the geometric

properties of polygonal nets.

*)
6. Relative to a vertex A, all vertices of the net can be

divided into classes as follows: the first class contains A, the

second class contains all vertices of the 4p net edges emanating

from A, the third the vertices of the edges emanating from points
th
of the second class, etc., the points of the m class being the

vertices of edges emanating from points of the (m_l)th class, as

long as these do not belong to preceding classes. A point of the

mth class, B, say, can be connected to A by an (m-l)-step polygon

edge path, but by no shorter path. We call (m-l)s, where s

is the length of the net edge, the net separation of A and B

(m-l)s.

Further, E(A,B) shall denote the straight line separation

between A and B.

Lemma

We obtain the proof in two stages.

I. The distance between two points on the boundary of a net

*)In this section we fix p > 1.

 144

s
polygon which lie on two non-adjacent edges is greater than
2P
a) We connect the midpoints of two adjacent edges by a circular

arc with centre at the common vertex. Since the lines connecting

the midpoint of the polygon with the midpoints of the edges are

perpendicular to the edges and lie entirely inside the polygon, the

circular arc also lies entirely inside the polygon. The distance

of a vertex of the polygon from a point on a non-adjacent edge is

therefore

b) If e and f (see Fig. 1) are two non-adjacent polygon

edges, then there is always a line segment within the polygon perpen-

dicular to both edges and hence equal

to the shortest distance between them.

Then if we connect the ends of e and f

71,.1. by two non-intersecting chords, a quadri-

lateral with four acute angles results

of which any two on a chord are equal, since e and f are equal

chords in a circle. (The angles are smaller than the polygon angle,

therefore acute, since we assume this of the polygons.) The

feet of the common perpendicular lie at the midpoints of the edges

when these are opposite each other, otherwise nearer to the end-

points of the shorter chord, since the perpendicular bisectors of

the edges cut the longer chord.

If AB and AC are two edges of the polygon, Ml and M2

 145

their midpoints and M the midpoint of the polygon (see Fig. 2)

then it follows from the congruence of

H
A Jl1 B
the triangles M2AMl and M2MMl that

1. AM1M2 = M2M1M !. and that

4
MM =AM =~ Now Ml , M2 etc.
112
Jf
PI,. ,. describe a regular 4p-gon inscribed in

the circle with radius MM l , so that

2pM1M2 is greater than 2MM 2 , the diameter of the circle, and

thus

Now if 0 is a segment of shortest length between two edges

which are separated by at least two edges, then 0 cuts off a piece

of the circle with centre M, radius MM l , which is not greater than

a half circle and in which one chord M1M2 is completely contained

s
(Fig. 2) ; consequently o > M1M2 > 2p . Finally let DE be a

segment of shortest length between two edges which are separated

only by a single edge (see Fig. 3),

f~
and connect Ml and M2 , Ml and M,

Ml and M3 by lines which cut DE

in F, G, H respectively. It
Fi,. s.
follows from the congruence of the

triangles M1GF and M2DF (the three angles are respectively

equal) that DF = FG, M2F = M1F. Since now, as was shown earlier,

1. M2M1M3 = %' it follows that

DE 2FH > 2FMl M M > ~
1 2 2p·
 146

With this the first objective is reached.

II. A segment connecting two vertices of the net can

s
be divided into pieces which by the foregoing are > 2p :
a) diagonals of a polygon, b) lines connecting a vertex and a point

on a non-adjacent edge of the same polygon, c) lines connecting

two points of the same polygon which are on two non-adjacent edges,

d) segments of the form XY shown in Fig. 4 : the number of

edges emanating from A met by

XY is arbitrary. YC belongs

to the same polygon as AB, but need

not be adjacent to AB at B.

•.,... However, Y can also coincide with

Band Z. ZE' is the mirror

image of ZE with respect to AB. I t follows that:

YZ + ZE' > YE' > s

2p

Thus also XY > s

2p

In the cases a), b) and c) one can connect the endpoints of a

subsegment by a path of successive polygon edges,

the length of which is ~ 2ps, in case d) by a path whose length

(2p+2)s
is ~ (2p+2)s. The ratio is therefore !S s
for each
2p
subsegment, and hence for the whole segment

EN ~ 4p(p+l)E, as claimed.
*

* In the original, Dehn erroneously reverses the last two inequalities.

(Translator's note)
 147

A motion yields a mapping of net vertices, and is determined

by the correspondence between two vertices, If goes to

under a motion then we shall denote the point which goes into

Xo under the same motion by X_ l , the point into which Xl

goes by X2 , etc, If now AO,A l and BO,B l are any two vertex pairs

of the net and it is to be decided whether the motion

can be transformed into the motion

connects BO with Bl by an edge

Further, let the points which correspond to ao'So'Yo"" under the

By suitable trials (in which

one successively investigates the points of the 2 nd , 3 rd "" classes

of vertices with respect to AO and Al respectively) one obtains

on both sides of the edge

corresponding poille" XO'Xl and YO'Yl whose net separationS dre

greater than 4p(p+l)m, *

One connects and respectively

by edge paths

Through the vertices SO,SO' "" and nO,n O' "" one lays all

edge

BO j~ Bl is transformable into AO j~ Al iff there is such a

*Here Dehn uses the fact that the distance a point is displaced by
a hyperbolic translation increases indefinitely with its distance
from the axis (Translator's note),
 148

~ (i)
path beginning in or which ends in or
1

respectively.

It is clear that this condition is sufficient because any two

paths with the same labels can be carried into each other by a motion.

The necessity of the condition results as follows: by our lemma the

it follows that outside the region between

between corresponding points is always greater than E(XO'X l ) resp.

E(YO'Y l )' so their net separation is also greater than the net

Then if there is

a motion, carrying BOBl into a point pair BO'Bl' which consists of

two points which correspond under the motion and

al , aO and Sl"" would go into pairs of points a O' ,a l ' ;

... which correspond according to AO ~ Al , and so the whole

and ... Y-lYOY l ... and hence meet the edge path Xo ... AO ... YO

in at least one point, from which the necessity of our condition

follows immediately.

§2. One-sided surfaces

If one cuts a closed one-sided surface of characteristic k

 149

(= the maximal number of closed curves on the surface which are

jointly not bounding) along k suitably chosen circuits running

through a point. a l .a 2 •...• a k • then one obtains a simply connected

piece of surface whose boundary, when traversed in a certain sense.

consists of the segments al.ar.a2.a2 •...• ak.ak. If. in the

case k = 1. we construct a sphere covered by two 2-gons ; in

the case k = 2 a square net in the euclidean plane ; in the case

k > 2 a net of regular 2k-gons in the hyperbolic plane. where 2k

of them meet at each vertex. then we can in each case denote the

edges of the net in such a way that the boundary of each cell.

traversed in a suitable sense. consists of the above al.al •...• ak.a k .

Then just as in the case of two-sided surfaces we can map each

closed curve to a connected curve of the net-bearing plane (resp.

sphere). with homologous endpoints. Corresponding to surface

curves which are continuously transformable into each other we have

motions of the net which are transformable into each other.

In the case k > 2 these motions are rotations about an ideal

midpoint with or without reflection. In fact. with any motion

different from the identity no real point remains at rest. Two

motions are then transformable into each other iff they are both

translations with reflection or both translations without reflection.

and further. the representatives of the displacement segments along

the axes of both translations are identical in a net cell. The

same considerations hold here as those presented for two-sided

surfaces in paragraphs 4. 5. 6. These lead us to a method of

 150

deciding the question whether two net motions are trans-

formable into each other in the case k > 2, or, what comes to

the same thing, whether two surface curves are continuously trans-

formable into each other.

In the case k = 1 we have only one motion different

from the identity, namely rotation of the sphere through 1800 ,

and correspcndingly only one kind of surface curve not contractible

to a pcint.

The case k = 2 still remains. The surface is developed

on a square net in the euclidean plane. The edges of a cell form

the series a, a, b, b. The motions of the net into itself

are either parallel displacements or reflections combined with a

parallel displacement along the axis of reflection. The axis of

reflection connects either the midpoints of edges denoted a, or

else the midpoints of edges denoted b. It is then easy to decide

whether two motions are transformable into each other: two

motions transformable into each other are either parallel

displacements of equal magnitude and direction, or reflections in

homologous axes combined with equal parallel displacements along

the axis of reflection.

§3. Analytic expression for surface curves

Each motion of the net into itself can be generated by

 151

the correspondence between two net vertices. Thus each surface

curve is transformable into one whose net image consists purely

of net edges. Such a curve, or the corresponding net motion,

will be denoted by the names of the net edges taken in the order

corresponding to traversal of the curve from initial to final point

If A is any expression denoting a path of net edges, and B

any other such expression, then

-1
B AB C

is an expression denoting the most general path transformable into

A. The net motions resp. surface curves corresponding to

the edge paths A and C are transformable into each other

and conversely : if A and C are two edge paths which

correspond to net motions transformable into each other,

then one can always find an edge path B such that the above

relation holds between A, B, C. These remarks hold for all

developments of surface complexes on nets. I f we apply this

specially to the two-sided and one-sided ring surface, it follows

that in the first case all paths of net edges are transformable

into an expression

n
where nand m are integers and a naturally denotes the

edge path consisting of n segments a. Two such expressions

 152

are then transformable into each other iff their nand m values

coincide. The situation with the one-sided ring is somewhat more

complicated. We note for future use that in any expression we can

delete a sequence aabb, since the corresponding four segments

-1
constitute a closed path. We can anywhere replace b by a ab
-1
and b- l by (ab) a (where (ab)-l denotes the edge path ab

traversed in the opposite sense). ab will be denoted c for

simplicity. We then have:

-1
ac C a.

Then in each edge path which contains only c and a we can

collect all c's and a's together and thus obtain (without trans-

formation) an expression of the form

n m mn
a c or c a

expressions convertible into each other by transformation. If n

is even this expression represents a parallel displacement, and it

easily follows that no transformation can alter the numbers m or

n. So for two such expressions with even n to be transformable

into each other, their numbers m and n must coincide. If,

however, n is odd and m = ml + m2 then by the above

mn
C a

ml +m 2 n
So by transformation, c a goes into Thus any
 153

expression with an odd exponent for a goes into one of the forms

n n
c or ca

by transformation. Summing up we have : two expressions are

transformable into each other iff: 1. the algebraic sums of the

exponents of a are equal, and 2. a) if the sum is even, then the

algebraic sums of the c exponents are also equal, b) if the sum is

odd, the algebraic sums of the c exponents are either both even

or both odd. With this, the theory of transformations of curves

on the one-sided ring is completely settled.

§4. Amphidrome surface curves

On one-sided surfaces with k: 1 or 2 there are curves

not contractible to a point which are transformable into themselves

reversed. Indeed, for k: 1 there is only a single type of

surface curve not contractible to a point. The curves, twice

traversed, bound a simply connected piece of surface, and thus such

a curve is transformable into itself reversed, a point of the curve

remaining fixed in the process. For k: 2 the curves expressed

by ab, ba respectively in the above notation are transformable into

each other. However,

abba 1

so ab
 154

which says that the curve corresponding to the expression ab is

-1
transformable into the curve corresponding to the expression (bal ,

continuously and with a point fixed. Consequently the curves

corresponding to the expressions ba and (bal- l , which are the

same except for sense, are transformable into each other. We shall

call such curves amphidrome and show that there are no amphidrome

curves on other closed surfaces.

We first consider the one-sided surfaces with k > 2. The

corresponding net motions are rotations about an ideal midpoint

in the hyperbolic plane with or without reflection in the real axis.

Since no real point can remain at rest under a net motion,

a pure reflection without simultaneous rotation is not possible.

No line of the plane apart from the axis goes into itself under the

motion. Thus in order that a motion other than the identity go into

the inverse motion as a result of a second motion, the second motion

must have the same axis as the first. But if this is the case,

the two motions commute. It follows from this that the first motion

performed twice in succession yields the identity. But this is

impossible, since each motion is a rotation about an ideal centre.

The same conclusion also holds for two-sided surfaces when p > 1.

If p; 1 all motions commute with each other and in fact are parallel

displacements in the euclidean plane. No motion performed twice in

succession can result in the identity, and consequently no motion

can be transformed into its inverse. Amphidrome surface curves

 155

correspond, however, to net motions which are transformable into

Thus we have the theorem:

Amphidrome curves exist only on one-sided surfaces with k 1

and 2.

For k = 2 it is easy to ascertain all amphidrome curve

types. They correspond to parallel displacements of the net perpen-

dicular to the axis of reflection and hence correspond to expressions

of the form (ab)m or curves transformable into such.

§5. Introduction of the fundamental group

The motions of the net, on which we have developed a closed

surface as above, constitute a group. It is clear

that we can take as generators of this group the 2p resp. k

motions corresponding to the 2p resp. k closed surface curves

along which the surface is cut in order to be developed. If we

denote each motion by the name of the corresponding curve then

according to the above notation we have as generators of the group

of motions

a single relation, namely

-1 -1
a b a b 1
p P P P

resp. 1.

In fact each relation between motions of the net may be derived from
 156

the defining relation, because the closed curve corresponding to

the given relation bounds a region consisting of cells, and the

defining relation says that a cell is bounded by a closed curve.

The net is the group diagram*) of the group of motions of

the net into itself. This group is called the fundamental group

of the corresponding surface. Each closed surface curve running

through a fixed point of the surface corresponds to an element of

the group, two surface curves transformable into each other corres-

pond to two group elements transformable into each other, in particular

surface curves contractible to a point correspond to the identity.

If A and B are surface curves corresponding to elements a and

b of the fundamental group, then the element ab corresponds to

the surface curve which consists of A followed by B. The identity

and transformation problems for these fundamental groups are

completely solved by the considerations of §1-3.

The fundamental groups so obtained are isomorphic, as is easily

seen, to groups with the same number of generators but a different

relation between them. E.g. the group

generators : a l ,a 2 ;
{
-1
relation : a l a 2a l a2 = 1

is isomorphic to a fundamental group of the above form

{ gener~tors :
__________~_~r~e.lat10n : 1.

*)see Math. Ann. 69 (Oehn's "On the topology of 3-dimensional space").

 157

We shall present a condition which is sufficient to ensure that a

group with one defining relation is isomorphic to the fundamental

group of a closed surface. We shall see later that this condition

*)
is also necessary If we construct the group diagram of the

group given by the relation, in which we allow polygons corresponding

to the relation and all its cyclic permutations to emanate from each

point, then the given group is isomorphic to the fundamental group

of a closed surface when these polygons, joined along like labelled

sides emanating from the initial point, close to a single sheet.

In fact, in this case each generator must appear twice in the

relation. A closed surface results when I bend the piece of surface

bounded by the polygon corresponding to the relation and identify

the sides with the same denotation. Each side becomes a closed

surface curve, and it is easy to see that the fundamental group of

this surface is the group given, so our assertion is proved.

§6. Topological properties of the group diagrams

of fundamental groups of closed surfaces

We shall now derive a theorem on the group diagram of the

most general group isomorphic to the fundamental group of a closed

surface. The diagram of such a group in "canonical" form is a

net in the euclidean or non-euclidean plane. It is clear that we

*)when each generator appears at most twice in the relation.

 158

can connect any two net vertices* by a net edge path which has no point

in common with the vertex that represents the identity. This

carries over in a certain form to the most general representation

of this group; let Sl'S2' ... 'Sn be the generators of the group

in canonical form, Tl ,T 2 , .•. ,Tm the generators of the general

representation, then by hypothesis the Si may be expressed in

terms of the T .. Let h be the greatest number of T. which

~ ~

appear in any of these expressions, then in the general represen-

tation we consider the totality of all those net vertices whose "net

separation" (see p.143) from the identity is not greater than h.

We also consider the image B of this collection in the canonical

net. All vertices of the net with the exception of a finite number

may be connected by edge paths having no point in common with the

vertices of B. In the general representation all these elements

are vertices which may be connected by edge paths having no

vertices in common with the identity. In fact : two of these vertices

may be connected, by the above, by an edge path consisting of

terms which are expressions for the S. in terms of the T. and

~ ~

which connect points which have a separation from the identity greater

than h. Consequently, none of these terms touches the identity,

SO we have the theorem:

In each representation of the fundamental group of a closed surface

one can connect any two elements, with a finite number of exceptions,

by an edge path which does not contain the identity.

*different from the identity (Translator's note).

 159

It follows from this by an easy generalization that:

Given a set* of elements in any representation of the funda-

mental group of a closed surface, there are only a finite number

of elements which one cannot connect by edge paths having no

point in common with the given set.

Further:

In each representation of the fundamental group of a closed

surface, each closed edge path satisfying a certain inequality

has the property that a ring-shaped band surrounding it divides all other

elements into two classes, inner and outer. Any two outer or

any two inner elements can be connected by an edge path which

has no elements in common with the given closed path. Each

edge path which connects an inner and an outer element ha·s at

least one element in common with the closed path.

Thus we see that in a certain sense the simple connection of

the elements in the canonical net is preserved in the general

representation. Similar considerations evidently apply to any kind

of group : certain connectivity properties of the group diagram

appear as "invariants" of the group, independent of the representation.

To conclude our examination of the fundamental group we prove

the theorem:

Two fundamental groups of closed surfaces are isomorphic iff

they correspond to homeomorphic surfaces.

* Of course, the set has to be finite (Translator's note) .

 160

. *)
TLetze has proved the following theorem on infinite groups:

If one allows exchange of terms in the fundamental relations, then

in the resulting relations between generators some may be expressed in

terms of an arbitrary r of the remainder. This number r is

the same for two isomorphic groups - for a surface of genus p,

r is equal to 2p, for a one-sided surface of characteristic k

r is equal to k-l. If our theorem were incorrect, then on account

of the Tietze theorem the fundamental group of a two-sided surface

of genus p would be isomorphic to the fundamental group of a

one-sided surface with characteristic 2p+l. Now Tietze has

also proved that two surface complexes which have isomorphic groups

have the same torsion numbers. On the other hand, two-sided

surfaces always have torsion numbers equal to unity. However, each

one-sided surface has curves which first bound when traversed twice,

so such a surface has a torsion number equal to 2, from which the

desired result follows.

Chapter II

Groups in which each generator appears at most

twice in the defining relations

Our task is to use the investigations of the preceding chapter

to bring the groups referred to in the title into a canonical form

*)
Wien. Ber. 1907. (Also, Monatsh. Math. Phys. 19 (1908), 1-118.
Translator's note.)
 161

and thereby solve, first the isomorphism problem, then the trans-

formation problem.

Let G be a group with generators Sl,S2, ... ,Sm and the

relations Rl R2 = ... = Rn 1, in which each generator appears

at most twice altogether in the relations.

1. If a generator Sl appears in a relation, say Rl = 1,

only once, i.e. if Rl is equal to SlT, perhaps oy cyclic

interchange of factors, where Sl does not appear in T, then G

is isomorphic to a group G' with generators S2, ... ,Sm and

-1
relations R2 ... = Rn = 1, in which Sl is replaced by T

S2, ... ,sm then appear at most twice in each relation. Thus by

this reduction one can reach the stage where each generator appears

in at most one relation, and in this one twice. We can therefore

begin by confining ourselves to groups having only one relation,

in which each generator appears twice.

2. Let G be a group with a single relation R = 1 in which

the generators Sl'S2' ... 'Sm each appear twice. If we then

construct the group diagram, in which we attach to a point all the polygons

corresponding to cyclic interchanges of R, then in general we do not

find the same simple situation as we did for the fundamental groups

of one- and two-sided surfaces. On the contrary, the polygons

will come together in several sheets.

 162

E.g. if we take the group

G f generators Sl'£2'£3'£4

l relation: 8281818384848283 1

then the eight associated polygons fall into two classes. The

four polygons which join the same-sensed pairs of edges

-1 -1 -1 -1
Sl 8 1 ' 8 1S 3 , 8 3 8 2 on the point constitute one sheet, while

those for the edge pairs constitute

the other. If we now transform this group by introducing a new

generator <1

{
generators Ll ,8 2 ,S3,8 4
G'
-1
relation : 8 2 <1 8 3 LlS4848283 1

The polygons of the group diagram attached to a point again fall

into two classes : those with the edge pairs

-1 -1
S3 8 2 resp. If one introduces

one obtains the

group

generators <1' <2' <3' 8 4 ;

{
-1
relation: <2<3 S4 8 4 L2 L3 = 1.

One sees that the generator Ll does not appear in the relation at

all, and the six polygons at a point in the group diagram corres-

ponding to the relation constitute a single sheet. In fact the

series of edge pairs is

 163

Thus if we omit the generator EI we obtain a group whose diagram

is a net of regular 6-gons in the hyperbolic plane.

The method explored in this special case may be made general,

as we shall show in the following.

Suppose that

are ~ edge pairs belonging to polygons which constitute a

single sheet~. Each of the generators SI""'S~ appears among

the pairs either twice or four times. We shall first consider

the two extreme cases :

a) The generators SI""'S~ all appear four times in the

edge pairs of the sheet 15 : since each generator appears in

the relation at most twice, by hypothesis, then it belongs to at most

four edge pairs. Thus none of the generators SI""'S~ forms

an edge pair with a generator different from any of these. But

since there is only one relation by hypothesis, and hence only one

simple polygon in the group diagram, in which each generator

appearing in the relation is represented by edges , then no generators

appear in the relation. These generators

alone, in the presence of the relation, determine a group whose

diagram is a net of regular n-gons, where n = 2~.

 164

b) All generators Sl' ... 'S~ appear twice in the edge

pairs of a sheet~ : we choose as new generators

whence

All generators Sl' ... 'S~ may be expressed in terms of Sl' L2, ... ,L~

Thus the given group G is isomorphic to a group r in which

according to the

above formulas. By hypothesis each generator S. (i = 1, ... ,~)

1

appears in the relation only twice, once in the context Si_lSi'

-1
the second time in the context Si Si+l' but by the above these
contexts are identical with LiLi+l' and for i = ~, with

Thus by introduction of the new generators, Sl

no longer appears in the relation. Thus by this transformation

we have reduced the number of generators appearing in the relation

by one (also, as is easy to see, the number of sheets formed by the

polygons at a point of the group diagram has also become one smaller).

In addition, after the transformation each generator appears in the

relation either twice or not at all.

 165

General case : Some generators appear four times and some

twice in the sheet-forming pairs. Then there is a pair consisting

of generators which appear four times and twice respectively, since

any two successive pairs have a generator in common. We may

assume that this is the case for the first pair SlS2 and in fact

that Sl appears four times, S2 twice. If we step again through

the series of pairs, then before the end we shall reach a pair in

which Sl appears again, and in fact the first such pair has the
s-ls -1 -1
form for i f i t had the form Sr Sl the first r pairs
r 1
would already constitute a sheet and Sl would occur only twice

among the pairs. The series therefore runs

and contains t+l pairs. We now set

and introduce Ll as a new generator in place of Sl. We now obtain

from the pairs the series

and this yields the t sheet-forming pairs

Here Ll appears only twice, and we have thus converted our sheet

into one with a pair less. The relation between the new generators,
 166

when SlS2 is replaced by El , again contains each generator either

twice or not at all. For El appears twice in place of S15 2

-1
as well as the substitute edge E1S 2 for 51 where it appears

in the relation for the second time. Similarly 52 appears twice,

namely in this substitute edge ,as well as the place where it

appeared for the second time originally. It is worth remarking

that the above conclusion is no longer correct when S2 also appears

four times among the original pairs; for if the original relation ran,

say,

-1 -1
then after insertion of El the pair 5 t El would follow
-1 -1 -1
St 52 ' after which E1 5 2 and 5 25 u etc. would follow, i.e. we

would again have ~+l sheet-forming pairs, as before the trans-

formation. One can easily be convinced of this phenomenon by an

example.

Using this process one can successively diminish the number of

pairs of a sheet until one comes to the extreme case b), i. e. until

there are the same number of pairs as generators corresponding to

the sheet. But then we can use the separate reduction procedures

above, by means of which one sheet disappears, and one generator no

longer appears in the relation. Then we admit more relations again

and repeat until we arrive at the following normal form of the group

among the generators are w generators which appear in no relation

("free generators") and v classes of other generators; the generators

 167

of each class appear in only one relation, and in this relation, twice.

The generators of each class,together with the class relation,

determine a group whose diagram is a regular net in the euclidean or

non-euclidean plane (see p.156) .

If the given group is already determined by a single relation

then indeed arbitrarily many further free generators can be given,

however, only one sheet -forming class. In this case, as is easy

to see, the group is the fundamental group of curves on a surface

complex consisting of a closed surface and w non-bounding closed

curves attached to a point of the latter. On the other hand, the

original form of the group relation leads to surfaces with self-contact

points. In place of free curves we can also introduce "horns"

(Fig. 5), i.e. spheres with a

self-contact point. The different

forms of surface corresponding to the

Fi,. .5.
group then correspond to the fact:

Each closed surface with w self-contact points is homeomorphic

to a singularity-free surface with w horns attached.

In the general case, i.e. when the group is determined by

arbitrarily many relations, the group in the reduced form is the

fundamental group of v closed surfaces and w free closed curves

attached to the same point. The group in its original form corres-

ponds to the system in which some of the free curves are transformed
 168

into self-contact points of the surface.

For each group of the kind considered, i.e. in which each

generator appears at most twice in the defining relations, we thus

have a characteristic series of numbers:

Here PI' ... 'P denote the genus numbers of the two-sided surfaces
vI
corresponding to the generator classes of the given group, kl, ... ,k
v2
the characteristics of the associated one-sided surfaces, and

finally w is the number of free curves. Since homeomorphic

surface complexes have isomorphic fundamental groups we have the

theorem:

If the characteristic numbers of two groups coincide, then

these groups are isomorphic.

The converse also holds.

If two groups of the kind in question are isomorphic, then their

characteristic numbers coincide (i.e. one can bring such a group to

reduced form in only one way) .

In fact let G and G' be the two isomorphic groups, each in

reduced form, let M2 be any closed manifold whose fundamental group

corresponds to a generator class of G. We shall show that the

elements of G' corresponding to this fundamental group are again

all the elements of the fundamental group of a surface, or else

 169

result from the elements of such a group by conjugation with a

certain element of G'. For this purpose we remark that the group

diagram for the fundamental group of a complex of two or more surfaces

consists of infinitely many sheets, with two or more meeting at

every vertex. Now let T be any element of the subgroup of G'

in question, then the different powers of T will only lie on the

same sheet if T either lies on a single sheet, or is transformable

into such an element. Further: if T, U are two elements of the

subgroup, then the different powers of T, U and TU will only

belong to the same sheet if T and U are both conjugates of elements

on the same sheet. Now since this subgroup of G' is isomorphic

to the fundamental group of a closed surface then by the theorem

proved on P.158 its representation in the group G' has a connectivity

of the same number of sheets. This is not the case, as is easily

seen, if the different powers of T, U and TU belong to infinitely

many different sheets. Thus the subgroup in question is itself a

subgroup of a fundamental group of a surface corresponding to G'

and since we can apply our conclusion just as well to the image of

a fundamental group corresponding to G' in G, it follows that

each fundamental group of a surface associated with G corresponds

to a fundamental group of a surface associated with G', possibly

transformed by an element of G'. Then it follows from the earlier

theorem that only the fundamental groups of homeomorphic surfaces

can be isomorphic, that the numbers

 170

are the same for G and G'. Finally, it follows from the

Tietze theorem that the horn number w must also be the same for

G and G'.

After we have brought the groups in question into a canonical

form it is easy to give a solution to the transformation problem.

Let 51 and 52 be two elements of one of these groups, Kl and

K2 the corresponding curves on the associated surface complex,

which in the canonical form we have developed consists of one- and

two-sided surfaces, plus horns, which all meet at a single point.

51 and 52 are then transformable into each other iff Kl and K2

are continuously transformable into each other on the surface complex.

(1) (2) (m l )
Kl may be divided into successive closed curves Ll ' Ll , ... ,L l '

where we assume that none of the curves is contractible to

L(i) and (i+l)

a point and also that no two successive curves Ll or
1
(m l )
Ll and L(l) lie on the same surface of the complex. Likewise when
1
L (1) (2) (m 2 )
we divide 2 ' L2 , ... ,L 2 Then i f Kl

are to be transformable into each other, and must be equal.

Further, l"f ml = m2 = 1 ' L(l)

1 and must lie on the same

surface and be continuously transformable into each other there, and

if > 1, then with possibly a suitable cyclic rearrangement, the

1 (1) 1 2 (2) -1 ml (m 2 )_1
curves Ll (L 2 ) - , Ll (L 2 ) , ... , Ll (L 2 ) must be

contractible to a point. The conditions are also sufficient, as

well as necessary, as is easily seen.

 171

Chapter III

Higher groups

§l. General Remarks

For each given infinite group one can find a four-dimensional

(homogeneous) manifold whose fundamental group is isomorphic to

the given group : each surface complex can be represented without

'3ingulari ties in a four-dimensional space. If the surface complex

has the given group as fundamental group then the four-dimensional

neighbourhood (domain) of the complex likewise has the given group

as fundamental group. One can also find a closed four-dimensional

manifold with these properties, and indeed in the four-dimensional

neighbourhood of the singularity-free representation of the surface

complex in five dimensions. Two M4 with the same fundamental group

need not be homeomorphic. For example: the four-dimensional

neighbourhood of an elementary piece of surface and a spherical

surface in five-dimensional space both have the identity as fundamental

group, but they are certainly not homeomorphic. For the second

Betti number is equal to 1 for the first manifold, 3 for the

second.

In the preceding chapter we have treated those groups in which

each generator appears at most twice in the defining relations.

It is easy to show that each group can be brought into a form where

each generator appears at most three times in the defining relations.

 172

In fact : let 8 be a generator which appears h times (h > 3)

in the defining relations, then we introduce the relation

where 81 is a new generator and replace 8 twice in the relations

Then 8 appears (h-l) times in the new relations, and

8 1 three times. If h > 4 then we also take the relation

replace 8 twice by 8 2 , and so on, until, with the aid of h-3

new generators 8 1 ,8 2 , ... ,8 h _ 3 the given group is brought into a

form where there is one fewer generator which occurs more than three

times in the relations. In this way our assertion is proved.

The next simplest class of groups after that we have already

treated is that comprising all groups in which the defining relations

include only one generator which appears three times. We shall

deal with this in a later work.

§2. Knot groups

As we have shown earlier*) the group of the trefoil knot is

presentable in the following form:

generators : 8 1 ,8 2 ,83'8 4
{
-1 -1
relations: 8 18 4 8 2 8 28 4 8 3 1.

*)
Math. Ann. 69.
 173

In the same place we presented the group diagram of this group

and thereby solved the identity problem. Here we shall show that

the group diagram can be represented by a regular net in a non-

euclidean space. In fact the group diagram found consisted of

parallel strips (see Fig. 11 of the work cited), three of which

meet at each boundary. Now we construct a system of segments

in the hyperbolic plane with the following properties: it contains

no closed polygon, and emanating from each vertex there are three
2~
equal segments at equal angles 3i. This is easy to obtain if one

takes the lengths of the sides to be the base of an isosceles

triangle with angle o at the top and base angles

3
~
It then

follows that a path of segments of this length and corner angles

2~
does not close or cut itself (for the vertices lie
3i
on a horocycle ), and hence also a path of segments of this
2~ 4~
length and corner angles does not end or
3i m 3i
cut itself. Consider for example the hyperbolic metric generated

by a fundamental circle in the euclidean plane. Over this circle

we erect a right cylinder, and likewise vertically over each segment

a parallel strip, and place points on the boundaries of the latter

at equal distances so a point on one side of the boundary is

perpendicularly opposite the mid-point of two successive points on

the other side. We then connect each point on the boundary with

the two nearest to it on the opposite boundary and we then have

in this collection of segments on the parallel strips the same

 174

infinite line segment complex which, when correctly labelled, is

the diagram for the group of the trefoil knot. As a result we

have shown that this group is isomorphic to a (properly discontinuous)

group of linear transformations of space, which transforms the

cylindrical surface into itself.

The transformation problem for the group in question is also

easy to solve. For this purpose it is unnecessary to bring the group of

motions of the non-euclidean space with the cylinder as fundamental

figure into a canonical form : the edge path corresponding to

an element of the group traverses a series of parallel strips.

This is then only diminished by transformation if the projection of

the edge path on the non-euclidean plane contains doubled segments

running in opposite directions, including the case when the last term

is placed at the beginning. One can always convert the edge

path by transformation into one which traverses the minimal number

of parallel strips. Two edge paths transformed into this form

are then transformable into each other iff they have the same initial

andf~ ~ts,possibly after cyclic interchange of terms. In this

way we have established a method for deciding the transformability of

two elements in a finite number of steps.

By quite analogous considerations one solves the transformation

problem for the knot groups corresponding to the curves closely

 175

related to the trefoil knot, which are presented on p. 164 of

Math. Ann. 69 * The basis for the diagrams of these groups

again comes from infinite regular line segment complexes in the

hyperbolic plane without closed curves, of the type where five

resp. seven segments meet at each vertex. The solution of the

transformation problem follows exactly as for the trefoil knot.

It is also possible to give the solution of the transformation

problem for the fundamental groups of the Poincare spaces associated

with these knots. Among these are infinitely many corresponding

to the trefoil knot. The transformation problem for these is

solvable without further investigation. The group diagrams of

the others admit representations on a complex of prisms, the cross-

section of which is a regular polygon net in the hyperbolic plane.

Thus the groups may be represented by real linear transformations

of space in which a cylindrical surface is mapped into itself.

A motion of the group diagram into itself corresponds to a motion

of the non-euclidean plane determined by two corresponding points

and two corresponding polygons attached to these points. Any

two points A and Al of the group diagram generate two distinct

such mo tions , namely those which carry A into Al and

Al into A. Now let S be an element for which we

want to decide whether it is transformable into an element T or not.

An arbitrary point 0 of the group diagram may be sent to A by S, to

B by T. With these two motions a point 0 of the cross-section

*p. 121 this volume (Translator's note) .

 176

net goes into Cl resp.)O and at the same time the net polygon

P1} attached to fj goes into the net polygon P", resp. p~)
If Sand T are transformable into each other these two net

motions must be congruent and indeed transformable into each other

by a net motion corresponding to a motion of the group diagram

into itself. This may be decided in a finite number of steps by

investigation using the methods developed in Chapter I. However.

this condition is not yet sufficient. Suppose U is an element whose

corresponding net motion carries the net motion corresponding to S

into that of T. The edge path corresponding to the element

u-lsu. whose associated net motion is identical with the motion

{ () 1+ '6 . p~ 1+ Ph} may start from a point 01 and end at

a point Bl . (0. 01 and 0 and similarly B. Bl and)j each

lie on a straight line perpendicular to the cross-section net.)

Then S is only transformable into T when the segments 00 1 and

BBl are equal and in the same direction.

In this way we have brought the necessary and sufficient

conditions for transformability of two elements of the group considered

into such a form that one can decide in a finite number of steps

whether or not they are satisfied. At the same time we have made

it possible to settle the question whether two given curves of the

Poincare spaces in question are continuously transformable into each

other (with self-intersections) .

 177

§3. Groups with two generators

To solve the three fundamental problems for all groups with

two generators (say a and b) still appears to be very difficult.

This class includes all knot groups,* as is easily seen by geometric

considerations. It is simple to settle the problem for those

groups in which a appears in the defining relations three times

and b twice. Difficulties already appear when a and b both

appear three times in the defining relations. Here we shall only

say a little about the group given by

We construct the group diagram from "cubic hexagons", i.e. spatial

hexagons whose edges are edges of a cube such that each face of the

cube contri.Wtes two edges. We denote the edges of a cubic hexagon ,

with a given sense, by a, a, a, b, b, b. In order to construct

the group diagram we associate the cubes corresponding to the cubic

hexagons with prisms. Each prism is cut by a second prism in each

of its cubes. Going through two successive cubes of a prism there

are two further prisms, whose axes are perpendicular. This

completely determines the group diagram. However, it is important to

notice that identical points of the group diagram are only those

which are so as a result of the relations between the prisms, not

those whose coincidence results from the special metric properties

of euclidean space. E.g. Two parallel prisms Pl and P2

* This is incorrect (Translator's note).

 178

go through two cubes WI and W2 of a prism which are separated by

an odd number of cubes. If W' and W' are two cubes on

1 2
resp. P2 which lie on homologous sides of WI and W2 separated

by the same odd number of cubes, then in euclidean space the two

prisms cutting resp. in W' and W' naturally coincide.

1 2
However, in our group diagram they have no point in common. Thus

our gro~p is not isomorphic to a group of motions in euclidean

space, but one can show on the other hand that it is isomorphic

to a group of motions in hyperbolic space, admittedly not properly

discontinuous.

With the exhibition of the group diagram the identity problem

is immediately solved. The transformation problem may alsc be

settled with the help of general principles which will be developed

in a later work.
 TRANSLATOR'S INTRODUCTION 5

The next paper contains Dehn's last word on the conjugacy

problem for surface groups, as well as a very simple solution

of the word problem. The solutions, by what is known as

Dehn's algorithm, are based on a detailed combinatorial study

of closed paths in the group diagram. This study shows, in

effect, that if a word for an element of the fundamental group

can be reduced in length, then it can be reduced monotonically,

namely by replacing any subword which is more than half the

defining relator by its complement, and by trivial cancellations.

(When only a conjugate of the original element is sought, the

word is regarded as circular in applying these operations.)

A word equals 1 if and only if its reduced form is

identical to I, and two words are conjugate if and only if

their reduced forms are identical as circular words (apart from

some simple exceptional cases) . Thus, just as Dehn's previous

algorithm depended on the existence of the hyperbolic metric

but used only the combinatorial structure of the group diagram,

the new algorithm depends on the existence of the group diagram

but uses only the structure of words, thereby achieving great

computational efficiency.

The geometrical interpretation of the algorithm for the

 180

word problem is particularly simple : a closed edge path

in the group diagram can be contLacted to a point by trivial

cancellation and the operation of pulling a subpath which

traverses more than half a polygon boundary to the other side

of the polygon in question. To prove this one only has to

show that a closed path which admits no cancellation contains

a subpath traversing more than half a polygon boundary, and

Dehn's argument for this is extremely simple.

His argument for the conjugacy problem is more complicated,

but along the same elementary lines. More sophisticated

proofs, using the Euler characteristic, are in Reidemeister

[1932] and Lyndon & Schupp [1977].

Dehn's algorithm for the conjugacy problem and its proof,

being completely combinatorial, avoid the axis and distance

curves which were central to the earlier proofs. But

surprisingly, the ghost of the axis is still there. Nielsen

[1927, §7g] points out that applying Dehn's algorithm to a

word c amounts to transforming a c-chain (cf. introduction

to the previous paper) into a chain which lies as close as

possible to the axis of the motion which maps the c-chain

onto itself. This observation clarifies the exceptional

 181

cases in Dehn's proof - they correspond to different chains

which lie equally close to the axis - and suggests refinements

of the algorithm reducing the number of exceptions which have

to be considered (Nielsen [1927, §7g]).

It is also suggested, though Nielsen does not say so,

that we re-examine the original application of the axis

(Poincare [1904]), which was to the simple curve problem;

deciding whether a given curve is homotopic to a simple curve.

Recall, from the introduction to Dehn's lecture notes on surface

topology (especially Fig. 2 there), that deciding this question

reduces to finding the course of the axis through the cells of

the group diagram. The approximation of the axis by an edge

path determines the course of the axis up to a finite (though

possibly large) number of topologically distinct possibilities.

These possibilities can then be examined in turn to see whether

any of them yields a simple curve. This may explain the claim

in Dehn [1922] that Poincare's algorithm for the simple curve

problem can be greatly simplified. Recently, Birman and

Series have used Nielsen's methods to develop this idea into

an efficient algorithm for the simple curve problem.

 182

REFERENCES

M. Dehn [1922] : tiber Kurvensysteme auf zweiseitigen Flachen,

mit Anwendung auf das Abbildungsproblem (Breslau lecture

notes) .

R.C. Lyndon and P.E. Schupp [1977] Combinatorial Group Theory,

Springer-Verlag.

J. Nielsen [1927] : Untersuchungen zur Topologie der

geschlossenen zweiseitigen Flachen, Acta Math., 50, 189-358.

H. Poincare [1904] : Cinquieme complement a l'analysis situs,

Rend. circ. mat. Palermo 18, 45-110.

K. Reidemeister [1932] : Einflihrung in die kombinatorische

Topologie. Teubner, Leipzig.

 TRANSFORMATION OF CURVES ON TWO-SIDED SURFACES

The problem we shall deal with in what follows is one of

the simplest of topology : given two closed curves on a closed

two-sided surface, to decide whether one may be "transformed"

into the other by a continuous deformation. The solution of

this problem for surfaces of genus p > 1 by means of "polygon

groups" and hence on the basis of the metric of the hyperbolic

plane is evident, and is indicated e.g. by Poincare (Rend.

Circ. Mat. Pal. 1904), also developed more precisely by me

in Math. Ann. 71*. In the same work I have given a method for

deciding the question purely topologically without the help

of the metric. However, in the foundation of this method I

have made essential use of properties of figures in the hyper-

bolic plane. For surfaces of genus p = 0 and p 1 the

solution of the problem is very simple in the first case

all curves are transformable into each other, in the second

case the fundamental group is abelian, and each curve is

transformable into one which traverses a fixed curve C m times

and a fixed curve r ~ times, and these numbers m and ~

are independent of the particular transformation, so that the

*Cf. the analytic exposition in the dissertation of Gieseking,

soon to appear.
 184

transformation problem is solved.

It is now a curious fact, and characteristic of the

imperfect nature of topological investigation, that the solution

of the transformation problem for p > 1 was previously

unknown, despite its simplicity in the case p = 1 : if one

dissects the surface by 2p cuts into a

simply connected surface piece, the boundary of which is formed

-1 -1 -1
by the series of curves al,bl,a l ,b l ,a 2 ' ... ,bp , then each

closed curve L on the surface may be transformed into one

"generated" by these curves, say

a a'
bPI _ A.
P al

If there is, then, possibly after a cyclic interchange

of terms, a part of the expression A which contains more than

2p, say q, terms of the relation

b -1 1
p

in order or reverse order, then one replaces this part by the

equal expression consisting of the 4p-q remaining terms.

-1
Furthermore, one cancels any successive terms c and C which

result, possibly after cyclic interchange, and continues these

two processes as long as possible. Then A is finally

 185

converted into an expression K which we shall call a

reduced expression. K represents a curve into which L is

transformable. Then we have

Main Theorem : Apart from some (easily settled) exceptional

cases, each curve determines a unique reduced expression, up

to cyclic interchange.

Apart from the exceptional cases, two curves are trans-

formable into each other if and only if their reduced expressions

are identical up to cyclic interchange.

A few exceptional cases are very obvious, e.g. two

expressions consisting of half the terms of the relation are

transformable into each other, for p = 2

and

Less trivial exceptional cases will be dealt with later when

the occasion arises.

In what follows we assume a knowledge of the simplest

properties of the mapping of the two-sided surface of genus

p > 1 onto a net of 4p-gons, 4p of which meet at each vertex.

This may be found in Chapter I, §l, paragraphs 1-5 and §5 of

 186

my work in Math. Ann. 71. I recall here only that this net

is the diagram of an infinite group, the fundamental group of

the surface, generated by with relation

b -1 1.
P

All curves on the surface which are transformable into each

other correspond to conjugate elements of the group. In particular,

curves which are contractible to a point, the identity element

of the group, are represented by closed polygonal paths of

the net.

§l. The identity problem

a '
1
Theorem 1 : If the element ... a l ... of

the fundamental group equals the identity, then this expression

is reducible. I.e. there is either a sequence of q terms

(q > 2p) in the expression which also occurs in the defining

relation, and hence may be replaced by the remaining 4p-q

terms, or else two successive terms sand s-l occur in the

expression and therefore may be cancelled.

To prove this we have merely to show that each closed

polygonal path in the group diagram of 4p-gons has more than

 187

2p edges in common with a net polygon, or else has two

oppositely directed edges occurring in succession.

We consider first the complex Gl consisting of a

single net polygon, then the complex G2 of net polygons

which have a point in common with the boundary of Gl , then

G3 , which results from G2 by adding all polygons which have

a point in common with the boundary of G2 , etc . . Each closed

polygonal path in Gl certainly has the required property,

but so also does each closed path in G2 . In fact, each vertex

of G2 which does not belong to Gl is a boundary point of

the simply connected complex G2 , and at most one edge from it

goes to the interior, in which case there are two reducible

polygonal paths emanating from the point which belong wholly

to the boundary. Thus a closed polygonal path which possesses

no successive oppositely directed edges is either wholly

contained in Gl or necessarily contains a reducible polygonal

path on the boundary of G2 . Exactly the same argument allows

us to carry the proof from G2 to G3 , G4 , etc., whence our

theorem is proved.

Our proof makes essential use of the properties of the

group diagram, whereas the theorem itself yields a process

 188

for deciding directly whether an expression represents the

identity element. without construction of the group diagram.

This is an advantage over the method in which one first con-

structs the group diagram and then uses it to decide whether

an expression yields a closed polygonal path or not.

For what follows it is necessary to find further properties

of closed polygonal paths : we shall first confine ourselves

to singularity-free paths. A series of net polygons

Pl.P2 •...• where each has an edge in common with its successor.

but no two polygons have a common element apart from this. will

be called a chain. We begin our considerations. for simplicity.

with a polygon contained in the region r bounded by the path

IT in question and having only a single open polygonal path in

common with IT. Using the above notation. we denote this

polygon by Gl . If IT does not merely consist of the boundary

of a polygon. then polygons of G2 must be contained in r. for

otherwise IT would have double edges. However. if only one

polygon of G2 is in then this polygon and the first

constitute a chain on the boundary of which there are two

reducible paths of 4p-l segments; while if several polygons

of G2 are in r. the boundary of the region composed of Gl

 189

and these polygons has at least three reducible paths of at

least 4p-2 segments. Now since a polygon of G3 never

has two edges in common with the boundary of G2 , otherwise

two.polygons of G3 would simultaneously have a segment in

common with each other and with G2 , none of the reducible

paths of r can be destroyed without an equal number taking

their place. By continuing this argument we arrive at the

result:

A closed singularity-free curve of the net has at least

three reducible paths of at least 4p-2 edges with the single

exception of the case where it constitutes the boundary of a

chain or of a single polygon.

We now consider curves with singularities, firstly those

with only one. For each such curve IT we can begin at the

singular point and traverse the curve IT in such a way as to

describe a simple closed curve (a loop) . This loop then has

no more than two reducible polygonal paths if and only if it

is the boundary of a chain; because of the fact that IT may not

end at the singular point, as the loop does, this means at

most one reducible polygonal path of the loop does not lie on

IT itself. But now each curve with singularities has at least

 190

two loops without a common edge. Thus in order for IT to

have less than three reducible paths of at least 4p-2 edges,

IT must have two loops without a common edge, and these must

each bound a chain. But it then follows also that neither of

the two loops can have singular points. For if I consider

a loop ABA, where the point B is a singular point of IT,

then two cases are possible :

(1) (Fig. 1) The two paths from

A other than the loop run to points

Band C on the loop without them-

selves forming loops, i.e. without

double points; let these paths be,

Fig. 1.
say, AXB and AYC. IfAXB

together with BA bounds a chain, then this has BA in common

with the loop ABCA, so that BA consists of a single edge.

Thus among the reducible paths on AXBA at least one with at

least 4p-l segments remains for the path AXB on IT. But

ifAXB together with BA bounds no chain, then indeed BA

need not be a single edge, but in return AXBA has at least

three reducible paths of 4p-2 edges, of which at least one

lies on the path AXB or on the path AB of IT. By considering

the other path AYC, it follows that there is at least one

 191

reducible path of at least 4p-2 edges on it, or on AC.

Then when we bear in mind that the loop AVCYA ~g wall ii

the loop AXCBA has at least one reducible path of IT with

4p-2 edges, it follows immediately that in this case there

are again three reducible paths of at least 4p-2 edges on IT.

2. (Fig. 2) One of the paths on IT

from A other than the loop ABA

describes a loop CYC before

returning to the singular point on x

ABA. Then if IT is not to have

three reducible paths, the other

path from A must run singularity-

Fig. 2.
free to the singular point B of

the loop ABA (otherwise one would have at least three

separated loops); however, its presence results in at least

two reducible paths of 4p-2 edges on IT, as above, so that

together with the path yielded by the loop CYC we again have

three such paths for IT.

Thus the existence of a singular point B on the loop

necessarily has the consequence that IT has at least three

reducible polygonal paths of at least 4p-2 edges. If this

 ;92

is not the case. then IT divides the plane into compartments

IT l • IT 2 •...• ITm each of which has only a point in common with

its successor. and is either a chain or a single net polygon.

If we also admit double edges. then nothing essential is

changed in our argument. Summing uP. we obtain:

Theorem 2. A closed curve IT of the net has at least

three reducible polygonal paths of at least 4p-2 edges.

provided it is not the boundary curve of a chain. or a series

of chains (or net polygons) each of which has either a single

point in common with its successor. or else is connected to it

by a double edge of IT.

Thus if IT does not have three reducible polygonal paths

of at least 4p-2 edges. and if it is not the boundary of a

single net polygon. then IT has at least two reducible polygonal

paths of at least 4p-l edges.

§2. The transformation problem

With the help of Theorem 2 it is now easy to solve the

transformation problem. i.e. the problem of reduction of

closed curves on a surface of genus p > 1. in the way intended.

 193

Let U and V be two reduced expressions in the notation

of the introduction, i.e. after cyclic interchange U and V

-1
contain no reducible polygonal paths or paths of the form SS

(This reduction may be made without use of the group diagram,

by directly applying the fundamental relation to the given

expression.) Now let U and V be transformable into each

other, i.e. by suitable choice of a third expression T the

polygonal path

is closed. By first making cyclic interchanges in the terms

of U and V we can arrange that IT contains no path of

-1
the form SS ; furthermore, by possible reduction of T and
-1
T we can ensure that no reducible polygonal paths lie on

these paths.

We assume now that IT is not the boundary of a chain or a

series of chains (see §l). Then by §l, IT has at least three

reducible polygonal paths of not less than 4p-2 edges.

Since the paths U, V and T themselves contain no such paths,

there must be two of these paths which have a part in common

with U but none with V, or else two which have a part in

 194

corrunon with V but none with u. We take, say, the first

possibility, and let tu and t'u' be the two reducible

-1
polygonal paths, where t and t' belong to T and T

respectively, and u and u' belong to u. Since t and

t' are two final sections of the same path T, one of them

must be only ~ edge long. For tu and t'u' must belong

to the boundary of a net polygon, however two edges appear in

the same order only once in the boundary of a polygon. Thus

if t and t' both have more than one edge, u and u' must

have the same beginning. But then the first and last edge

of U are the same but oppositely directed, and U can be

reduced by cancellation of a term ss-l after cyclic inter-

change, contrary to hypothesis. However, if t has only

one edge, u must have at least 4p-3 edges because tu has

at least 4p-2. Thus U has a path of 4p-3 edges, which

is reducible because p > 1, and we have a contradiction to

the hypothesis that U is a reduced expression.

Consequently, we have the result:

Two reduced paths which are transformable into each other

always constitute the boundary ofa chain or series of chains

when a suitable transforming expression is added.

 195

For a general element, however, it is impossible to

-1 -1
find elements T and V such that TUT V IT is the

boundary of a proper chain, on the contrary IT will consist

of a double path traversing the same route in opposite directions.

In fact, if the boundary is not to be degenerate, each

of U and V must, apart from double paths in IT, decompose

intb polygonal paths of not less than 2p-2 edges which lie

on the boundary of the same net polygon : in this case we

have two reducible paths of at least 4p-l edges on IT.

These may both lie simultaneously on U and V, in which case

it follows as above that T consists of a single edge at most.

-1 -1
Or else one lies, say, on TU and the other on TV.

Let one be denoted by tu, the other by t'v', in the sense

-1
used above. Then one can replace t in T by cu (where

c is a generator) or u- l respectively, according as tu

consists of 4p-l or 4p edges. But then we can leave out

-1
the term u in the transforming expression by carrying out

the corresponding cyclic interchange of terms in U, and obtain

a new transforming expression T which has at least 2p-2

terms fewer than T. This can be continued as long as T does

not form a reducible polygonal path in combination with U

and V. Thus we obtain the result

 196

Corollary to the Main Theorem: If U and v, together

with a suitable transforming expression, form the boundary of

a chain etc., then by suitable cyclic interchange of the

terms in U and V one can find that U equals V or else

U goes into V by transformation with a single generator.

This result suffices to settle the transformation problem

in the exceptional case by reducing it to the identity problem

of the previous paragraph. In fact, if U and V are trans-

formable into each other we must have

cuc-lV- l 1

---1
or UV 1

where c is anyone of the 2p generators and U, V are

expressions resulting from U, V respectively by cyclic

interchange. Thus we have now completely settled the trans-

formation problem. The case we have called the exceptional

case is in fact, as remarked above, quite special : in order

for it to occur it is necessary that paths of length at least

2p-2 edges from a net polygon appear in U as well as V.

Accordingly, we can immediately conclude e.g. : two expressions

l 2
of the form c l m c 2m Cv
mv where c l ,c 2 ' ... are any
 197

generators and Imll ,lm21 , ... , Im~1 are all greater than 1,

are transformable into each other only when they result from each

other by cyclic interchange of terms, likewise two expressions

of the form

Cl l Cl 2 Cl
a a a q
nl n2 n
q
or
81 82 8q
b b bg,
g,l g,2 q

in which only one or other half of the generators appear

are transformable into each other if and only if they result

from each other by cyclic interchange of terms. These are

results which could have been derived only with great difficulty

with the previous methods.

We give yet another example for the exceptional case

let p = 2, then
-1 -1 -1
albla l al bl a2 u
-1 -1 -1
and a 2b 2 a2 al b 2a 2 v

are transformable into each other, for

 198

because the expression on the left-hand side is, in detail,

Here the last three and the first four terms constitute a

path of 4p-l = 7 edges of the fundamental polygon, and

are thus replaceable by one edge, namely bl . There then

remain eight terms which exactly describe the boundary of

a fundamental polygon.

The method for deciding transformability, as well as its

proof, is quite free of metric elements. Indeed, as far as

the special case of the identity problem goes, as well as

Theorem 2, the particular labelling of the net edges is not

used, so that these results hold for all groups whose diagrams

are nets of 4p-gons, 4p of which meet at each vertex.

It is easy to see that even this much need not be assumed,

but merely that the polygons are at least seven-sided, and

that at least four polygons meet at each vertex.

In the solution of the transformation problem the special

labelling of the net edges is used at one place, namely the

property that a [labelled] polygon path of more than one edge

 199

occurs at most once in a given net polygon. However, this

property of the labelling follows from the topological property

of the net, that there is no vertex where only two polygons

meet.
 TRANSLATOR'S INTRODUCTION 6

Of all Dehn's results, perhaps the one most understandable

to a non-mathematician is the following: a left-hand trefoil

knot cannot be deformed into a right-hand trefoil knot. The

statement concerns everyday objects, and only the proof is

difficult. Dehn's proof is based on the observation that a

deformation of one trefoil knot to the other would induce an

automorphism of the knot group which associates certain geometrically

significant elements, namely "latitude" and "longitude" curves

on the neighbourhood torus of the knot, which determine an

orientation of the ambient space.

Dehn sticks to the presentation and group diagram he

developed in Dehn [1910], which is far from easy to handle,

and only his virtuosity in hyperbolic geometry carries him through.

A much easier solution of the basic group theoretic problem,

which I have included as an appendix below, was obtained

algebraically by Schreier [1924], based on the presentation

<A,B ; A2 = B3>. This presentation is immediate from Dehn's

relation
-2
a4 a l a 4a l = 1,
-3 2
i.e. a4 (a 4 a l ' = 1,
 201

A, a 4 = B. Schreier's argument in fact

a b
covers all groups <A,B ; A B 1> where a, b are integers,

and hence all torus knot groups. (For a beautiful proof that

all torus knot groups are of this form, see Seifert & Threlfall

[1934] .) Moreover, for these groups Schreier solves the problem

posed by Dehn at the end of I below, to find a presentation of

the automorphism group of a given finitely presented group.

Dehn's pregnant remark, that he would attack this problem for

surface groups "in a work to appear shortly", had a long gestation

period. His results eventually appeared in the 1920's and 1930's,

partly in his own work and partly as reworked by Nielsen, Baer

and Goeritz. See the next two papers in this volume, and their

introductions, for more information.

The brief investigation of the automorphisms of the figure

eight knot group in III was completed by Magnus [1931], who

showed that the two automorphisms given by Dehn do in fact

generate the outer automorphism group. According to Magnus

[1979], the Kiel student who constructed the diagram of the

figure eight knot group was named Fritz Klein. It seems likely

that he died in World War I, and the diagram of the group has

never appeared anywhere. It may have been based on the 1912

Munster dissertation of Dehn's student Hugo Gieseking (the relevant

 202

part of this thesis is expounded in Magnus [1974]). Gieseking

arrives at the same group as a discontinuous group of motions

of hyperbolic 3-space. This remarkable discovery became more

understandable when Thurston [1977b] showed directly that the

figure eight knot complement could be given a hyperbolic

structure, and that this phenomenon is fairly typical of knot

complements.

REFERENCES

M. Dehn [1910] : tiber die Topologie des dreidimensionales Raumes.

Math. Ann. 69, 137-168.

w. Magnus [1931] : Untersuchungen liber einige unendliche dis-

kontinuierliche Gruppen. Math. Ann. 105, 52-74.

[1974] : Noneuclidean Tesselations and Their Groups.

Academic Press.

[1979] : Max Dehn. Math. Intelligencer 1, 132-143.

o. Schreier [1924] : tiber die Gruppen AaBb = 1. Abh. Math.

Sem. Univ. Hamburg 3, 167-169.

H. Seifert & w. Threlfall [1934] : Lehrbuch der Topologie.

Teubner, Leipzig.

w. Thurston [1977b] : Geometry and topology of 3-manifolds.

Lecture notes, Princeton Univ., Mathematics Department.

 THE TWO TREFOIL KNOTS

In the following I wish to show that investigation of the

groups associated with interwoven space curves (knots) puts us

in a position to treat basic topological problems simply and

rigorously. For this purpose I have chosen a seemingly special

example, which on the one hand is of general interest, while

on the other hand being so simple that no special difficulties

stand in the way of the method.

The trefoil knot (Fig. 1) is the simplest knot, i.e. the

simplest interwoven space curve which cannot be continuously

transformed in space into a circle. It is

the only knot with a plane projection having

only three double points. In space one can

distinguish two different kinds of trefoil knot,

which result from each other by reflection, say,

Fig. 1. in a plane. Both kinds are shown in Fig. 1.

Our problem is to find a way of distinguishing

* i.e. to
the two kinds of trefoil knot in space topologically,

show that the left knot is not convertible into the right by a

* And in fact also in the closed spherical space, contrary to a

remark in the Enzyklopadie.
 204

continuous space transformation. Of course this fact is

already known to those who have gone into the subject of knots

and visualized space curves, say, as threads. But this knowledge,

while perhaps adequate practically, is very far from being a

** ***
real proof In any case, the phenomenon is very remarkable

The space with a left trefoil knot and the space with a right

trefoil knot are homeomorphic, i.e. so composed that the knots

occupy corresponding regions in the two spaces. Despite this

the two knots in the same space are not convertible into each

other by a continuous transformation of the space. Thus we

have before us a kind of topological symmetry. In two-dimensional

space (on the disc) symmetric regions not convertible into each

other are far less beautiful, as we may note incidentally.

Fig. 2 shows two of them, chosen more

or less arbitrarily.

In what follows I shall now

lay the foundations of the proof.

The route will be clearer if one

Fig. 2
first considers the continuous

transformations 6 of a three-dimensional manifold M3 into

**
This was also recognized by H. Tietze, Monats. f. Math. & Phys.
19, p. 97.
***
This phenomenon and its opposite number, the "amphicheirality" of
knots, were first pointed out by Listing.
 205

itself under which two curves KI and K2 lying in M3 go

into each other (in our case M3 is either a simply connected

piece of ordinary space or the whole three-dimensional spherical

space) .

~1 __~a=n~d~K=2 __~g~o~i=n~t~o-=e~a~c=h-=o~t~h~e~r-=u~n~d~e~r__~~~~t~h~e~n~t~h=e

*
~g~r~o~u~p~s__~o~f__~KI ___
a_nd
___K_:2 can be mapped isomorphically onto each

other in such a way that all curves which correspond under ~

also correspond under the isomorphism.

This "main condition" follows immediately from the circum-

stance that any curve which bounds a disc in the complement

space of K goes into a curve which bounds a disc in the

complement space of K2 under the continuous transformation

Thus if a l ... an are elements of the group of KI

for which

then for the corresponding elements of the group of K2 we

have

a ' a ' I
1 n

where and a ' correspond to curves which go into each

1

* See the remark below.

 206

other under (1. Thus the correspondence a. t+ a. t is an

1- 1-

isomorphic mapping of the group onto itself. If we surround

a point of a curve K by a sufficiently small 3-cell E3 in

M3 then K will cut the spherical surface bounding E3 in

two points A and B a curve which separates A and B

on the boundary of E3 will be called a curve of latitude.

If we surround K by a sufficiently narrow torus inside M3

then this torus will have no point in cornmon with K. The

torus will contain a certain curve of latitude which does not

separate the surface. A curve on the surface which cuts S

just once will be called a curve of longitude A. Then as a

special case of the main condition we have:

a) The isomorphism between the groups of KI and K2

can be chosen in such a way that each curve of latitude Sl

of KI goes into a curve of latitude S2 of K2 , and likewise

a curve of longitude Al of KI goes into a curve of longitude

In fact, ~ takes a point of K, resp. spherical neighbourhood

of a point of KI resp. torus around KI into a point of K2

resp. spherical neighbourhood of a point of K2 resp. torus

* See the remark at the end of this work.

 207

around K2 , and K2 , like Kl has two points in common with

the spherical neighbourhood, none in common with the torus, etc.

This condition a) holds for all two-sided and one-sided

manifolds M3 . ~3 is two-sided we can sharpen condition

a) still further : if we take any pair 8 and A for a curve

K and give 8 and A a sense, then 8 gives, first to the

3-cell E3 around a point P of K, and then to the whole mani-

fold M3 , an indicatrix. For example, one can associate with

E3 the screw determined by the sense of

8 and the sense of K when traversed in

the same direction as A (see Fig. 3).

Now since the indicatrix of a two-sided

M3 is unaltered under continuous transformation

into itself, we have the following sharpening

Fig. 3
of the condition a) .

a') The isomorphism between the groups of Kl and K2~

be chosen in such a way that when a pair of sensed curves 81

and Al ~l goes into a pair of sensed curves 82 and A2

of K2 the indicatrix of M3 determined by 81 and Al is the

same as that determined by 82 and A2 .

In our case M3 is ordinary space. The group GK~ of the

 208

trefoil knot in this space is already known from earlier

investigations ** In the next section we investigate the

isomorphisms of GK~ onto itself, the automorphism group of

GK~' It will be shown that these may be very simply described

apart from the obvious "inner" automorphisms there is essentially

only one "outer" automorphism. In this way we cover all

possibilities of associating curves relative to the left trefoil

knot with curves relative to the right trefoil knot, in such a

way as to induce an isomorphism between the groups of these

curves. It turns out that the condition a') is never satis-

fied, so that our goal is achieved.

I have sought a further clarification of the problem by

briefly treating an "amphicheiral" knot in the third section,

i.e. an intertwined space curve not transformable into a circle,

but convertible into its own mirror image by a continuous space

transformation. The knot I deal with is the simplest after

the trefoil knot, namely, the one which has only four double

points under suitable projection. We shall see that the amphi-

cheiral character of this knot corresponds to the more complicated

nature of its automorphism group.

** See Math. Ann. 1910 and 1911 as well as the MUnster dissertation
of Gieseking. [The subscript K~ is for "Kleeblattschling",
German for "trefoil knot" (Translator's note).J
 209

Thus our topological problem leads to questions of auto-

morphisms of an infinite group, and to the very general problem:

given a presentation of a group by generators and defining

relations, find a presentation of its automorphism group of

the same type. In the case of the fundamental groups of closed

surfaces we shall attack this problem in a work to appear shortly.

II.

The group GK~ of the trefoil knot, as has been established

several times * , can be defined in the following way:

Generators:

Relations

1.

The associated group diagram consists of strips (see Fig. 4)

which meet in threes at each boundary. If one represents the

group diagram by a regular net in a non-euclidean space (see

Math. Ann. 1911), the boundary of which is a cylinder with sides

parallel to the strip edges, i.e. the generator a 4 , then each

element of the group corresponds to a certain motion of this

geometry, which carries the group diagram, together with its

*See note H on the previous page.

 210

labelling, into itself. In particular, the generator a4

n
resp. a4 corresponds to a rotation about an edge through
2'11 2n'll
the origin by resp. combined with a parallel displace-
3 3
ment along this edge through c resp. nc, where c is the

length of a4 . In the plane perpendicular to the axis we

have the usual hyperbolic geometry. If we intersect the

group diagram with such a plane through the origin 0, then the

cross-sections of the strips give an infinite line segment

complex r (see Fig. 5) : three segments emanate from each

point of r at angles of 120 0 • r is a tree, i.e. there is

no closed polygon in r. An element of induces a motion

of r into itself. The generator a4 and its powers are the

only elements which leave the origin 0 fixed.

Fig. 4 Fig. 5 Fig. 6

 211

1. Under an automorphism of GK~ (i.e. an isomorphic

3 3
mapping onto itself) a4 goes into a4 or In fact,

an element which commutes with all elements of GK~ (a

distinguished element of GK~) must be sent to another such

element. But and its powers are the only distinguished

elements of GK for which this holds, consequently the group

{(a!)n} goes into the group {(a!)n} under the automorphism.

But the group {(a!)n} may be generated by ~ element, which

3 -3 3
may be chosen as either (a 4 ) or (a 4 ), so a4 must go into
3 -3
either a4 or a4 However, we must still prove that (a!)n

are the only distinguished elements of GK~

Suppose now that S is any distinguished element, then in

particular

and thus causes a rotation about O. Now let 0 go into P,

Pinto 0, under the motion of corresponding to S (see Fig. 6).

Then under the motion sa 4 the point 0 goes into P and the

point P, say, into Q, where ~ OPQ = 2~ and OP = PQ.

3
-1
Consequently 0 goes into Q under the motion sa 4 s ,because

the line segment S, the projection S' of which on the plane

of r is OP, is carried by the rotation a4 about P into a

segment with direction QP on the plane of r. Thus the

 212

element only yields a rotation about o when p

coincides with O. Consequently S must,be a power of a4

to yield a rotation about 0, and all distinguished elements

of are of the form But

-1
a 4a l a 4 a2
2 -2
a 4a l a 4 a3

+1 +2
so neither a- or a- are distinguished elements, since
4 4
and are different. On the other hand

3 -3
a4a l a 4 al
3 -3
a 4a 2a 4 a2
3 -3
a 4 a 3a 4 a3

as the figure shows, from which it follows immediately that

(a!)n are the only distinguished elements, as was to be proved.

2. Under an automorphism of GK~' ~4 goes into

-1 -1
or sa 4 S

If a4 goes into a 4 , then the third power of a4 must

3 -3
be either a4 or a 4 , by the above. Thus the motion which

corresponds to (a- 4 ) 3 must leave r fixed. Thus a4 itself

must cause a rotation of r of 1200 about a real midpoint M

of the plane of r. We now observe that r can be put together

 213

from regular infinite polygons with angles of 1200 , and with

vertices, after suitable choice of the length of side (or length

of the projection of al etc.), lying on a limit circle (i.e.

a circle lying on the boundary cylinder) . We shall call these

infinite polygons fundamental polygons for short. The funda-

mental polygons now go into one another by rotation about M.

If M lies in the interior of a fundamental polygon, then the

latter must go into itself by the rotation about M. For M

lies in the interior of only a single polygon, and M cannot

leave the interior of this polygon when it is rotated about M.

But the fundamental polygon cannot go into itself by a rotation

about a real point through an angle different from 2~. Further-

more, this is only possible when the midpoint for the rotation

coincides with the midpoint of the polygon, which is ideal in

any case (and lies on the boundary cylinder under a suitable

choice of segment length for r). Thus M cannot lie in the

interior of a polygon. Consequently M must lie on at least

one line segment of r. But if M lies on ~ segment, then

under rotation through 1200 about M this goes into a different

segment which has only the point M in common with the original.

Thus M coincides with a vertex of r. Consequently a4

causes a rotation of 1200 about a vertex M of r, and thus

 214

yields or by conjugation with a suitable element s.

In fact, if S carries 01 into 0, and 0 into M (see Fig.7) ,

then o goes into itself by

is a rotation about 0 and

~
o S' J/
S But n can only be +1
Ut 1200 S
- 3 +3 -3
or -1, since (a 4 ) a4 or a4 and
Q,
- 3 n -1 3 3 3n -3 ; a 3n
(a 4 ) (sa 4 S ) S a4 s 4
Fig. 7

3. ~4 goes into a4 under an automorphism, then al

goes into

The group GKt can be generated by and alone,

for the other generators and are the conjugates of

by and respectively. is defined by one

relation between and aI' namely

which one obtains by elimination of and from the three

original relations.

Now suppose we have an automorphism which sends a4 into

into Then according to the relation we must have

 215

so that this is a rotation about O. Now let (Fig. 8) OPP l

be the projection of the polygonal

path ala l on the plane of f. Then

OPQl is the projection of the polygonal

~)

and

Fig. 8 OP.
2
a 4 , and thus a rotation about

0, when Ql coincides with 0, so that ~ OPP l 120 0 (and

in fact the rotation about P from 0 to Pl must be right-

handed when a4 causes a left-hand rotation of 120 0 and

conversely) . We now use the "projection

-a , of a to
l l

construct an infinite line segment complex f which has origin

0 and again three segments meeting at each vertex at angles

of 120 0 . Since the length of al ' is no less than the length

of a ' (the segment length of f), f is also a tree. It

1

contains only vertices which are also vertices of r, but not

all the vertices of unless the length of

- ,
al equals that

of a l , All vertices of the group diagram which can be

obtained from a4 and al project on to the vertices of r,

-
but since al and a 4 generate the whole group, a l , must
 216

have the segment length of r, and thus must equal

+l
or a- But
3

for i 1, 2, 3.

Thus or a 3 , i.e.

Since and are the conjugates of and

respectively, it follows that a2 and are correspondingly

associated with

4. It follows quite similarly that: ~4 goes into

-1 CI -1 -CI
~4 __~t_h~e~n~~al __~g~0_e~s__i_n~t~0__~a'4~1 ~4 . Then it follows from

the relation

that

(taking inverses), and by conjugation

so that, because commutes with all elements,

1.

Thus the same conclusions hold for and as we

 217

found under 3. for a4 The mapping

-1
a4 ~ a4 represents an outer automorphism of GK~' For

is not conjugate to as one sees most simply by

making GK~ abelian by addition of suitable relations. For

then GK~ goes into the ordinary infinite cyclic groups

n -1
{a l }, in which al is not a conjugate of al otherwise
2
al 1 would follow.

5. I f a4 goes into then there is an auto-

morphism (namely the ordinary inner one) under which al goes

into and it follows from 4. that under the automor-

phism which sends a4 into is sent into

a -l-a
a 4sa l s a 4 ,for by introduction of the automorphism

we obtain an automorphism from the one given, in which a4

goes into a 4 , and hence al must go into Likewise

-1 -1
goes into when goes into Sa 4 S

But since by 2 there are no elements paired with a4 other

than the foregoing, we have the result:

All automorphisms of GK~are given by the following

correspondences:
 218

-1
a. 1+ Sa. S
1 1
(i 1,4)
and a .... sa.-ls- l
1 1

respectively. Thus for and we obtain the corres-

pondences
-1 -1
a2 :+ sa 2 s , a 3 ,+ sa 3s
-1 -1 -1 -1
and a2 1+ sa 3 S , a 3 ... Sa 2 S

Using suitable inner automorphisms, all outer auto-

morphisms can be brought into the form

6. Now if we have the two trefoil knots K~A and K~ ,

P
then the group of space curves relative to K~A is isomorphically

related to the group of space curves relative to K~ when one

P
associates the generating curves al and a4 with the corresponding

curves on the right in the opposite sense (see Fig. 9) (in

which the latter curves result from those on the left by

reflection in the plane of projection) .

 219

Fig. 9a Fig. 9b

. . 3
al lS a curve of latltude, a 4 represents a curve of

longitude on both knots (see Fig.) (one results from the other

by reflection in the plane of projection) . One sees

immediately that the indicatrix determined by and

in the case of K~A is the opposite of that determined by

3
al and a4 for K~ Thus this correspondence between the
p

curves of K~A and K~ cannot correspond to a continuous

p

space transformation in which the curves go into each other.

But any other possible association of and with curves

of K~p corresponds to an automorphism of GK~. We know that

all automorphisms of GK~ may be reduced to the identity

 220

or the correspondence

by means of an inner automorphism (transformation). As we

have seen, the first correspondence gives no possibility of

a continuous conversion of K~A into K~p' but neither does

the second, for and yield the same indicatrix

as and

Now to complete our proof we have only to show that

sals -1 and sa 4 3s- l as latitude resp. longitude curves yield

the same indicatrix as and However, we can easily

show even more, namely, that the indicatrix is not altered

when one arbitrarily transforms (i.e. continuously deforms

in the complement space of U) the latitude curve and

the longitude curve In fact, in order to determine the

indicatrix it suffices to first determine the indicatrix given

by al and the knot line with an assigned sense, say

...K~, then

compare the sense of with that of

...
KL But in the first

place the indicatrix with respect to

...
K~ and al is not

altered by any continuous deformation of al in the complement

3
space of K~, and in the second place a4 cannot be continuously

deformed into a longitude curve of the opposite sense in relation

to
...
KR.. We can give the latter the form This is
 221

because all longitude curves can firstly be given the form

a 4 ±3 aI'
n Lof necessary b y trans f orrnatLon
°
on th e rLng
°
sur f ace

surrounding K~. Also, all these longitude curves trans for-

mabIe into each other on the ring surface have the same sense

a4 al n al n
+ 3 -3
relative to K~, and has the opposite sense to a4
+ 3 -3 n
in relation to U. But i f a4 can be transformed into a4 al

in the complement space then n must equal 12, as follows

immediately from the defining relations of GK~ when one allows

all elements to commute. It follows from

that

because commutes with all elements. Thus we must have

a relation which is certainly not satisfied, as a glance at

the group diagram shows. Thus the two curves which result from

transformation of and yield the same indicatrix as

3
and a 4 ' and the proof is complete.

III.

As we said in the beginning, we shall conclude by briefly

treating an amphicheiral knot, i.e. a knot which is trans for-

mabIe into its own mirror irnage.* The group is obtained

*It was studied at my suggestion by a Kiel student. He has

succeeded in obtaining the interesting group diagram for the knot.
 222

directly by the process developed

00
in Math. Ann. 1910 by con-

sideration of the four crossing

4

points where the five inner 1 3

2
regions meet:
Fi,.lo.
Generators:

Relations

1.

We consider the following correspondence between the

generators of the group GL of our knot and the other elements:

-1
al >+ a 2a 5 bl

-1
a 2 .... a 3a 5 b2

-1
a3 1+ a 4a 5 b3

-1
a4 1+ al a5 b4

-1
as .... a 5 bs ·

It is easy to see that this correspondence is an automorphism, in

which 1) all defining relations take the same form for the b.
1.

as for the ai' 2) the ai are expressible in terms of the bi ,

as follows immediately from the above relations.

A latitude curve S and a longitude curve A for L are

represented by the elements:

 223

(A can be best understood by proceeding from the point A

in the direction of the arrow). Under the isomorphism S

and A go into

The first curve again represents a latitude curve S'

+
which, together with the knot sensed as shown, L, gives the

reverse indicatrix to S. The second curve is again a curve

of longitude, A' (as one can see best by proceeding from

B in the direction of the arrow). Since A' has the same

+
sense as A relative to L, the indicatrix (S,A) is carried

into the reverse indicatrix (S',A') by the automorphism.

This corresponds to the possibility of the knot being amphicheiral.

In fact, this isomorphism of GL corresponds to a transformation

of L into its mirror image. One sees this easily when one

thinks of L spread over a spherical surface and converts the

outer region into the inner region 5. Then L goes into its

own mirror image (say, relative to the spherical surface).

It is natural to seek further automorphisms of GL • One

finds a second one of the form:

 224

This is also outer, like the first. If we compose these two

automorphisms then we obtain, when we always associate an

inner automorphism with the identity, a group of eight auto-

morphisms which coincides with the eight-termed dihedral group.

Whether this group yields all automorphisms by conjugation re-

quires further investigation.

These automorphisms all have the property of carrying

latitude and longitude curves into latitude and longitude

curves again. Investigating whether this is the case for all

automorphisms of knot groups will be an important step in

mastering the general problem of trans formability of knots

into each other.

APPENDIX

ON THE GROUPS Otto Schreier

In what follows I derive a few properties of the groups

Ga,b generated by the elements A,B and defined by a relation

AaBb = 1. The numbers a and b which may be assumed non-

negative, and which I assume to be > 1 in order to exclude

trivial cases. These groups include the fundamental groups of

all knots which can be embedded in the torus, e.g. G3,2 is the
group of the trefoil knot. Theorem II can be used to simplify

l)I have presented these results at a topological seminar

organized by Herr REIDEMEISTER in Vienna.
 225

DEHN's proof of the distinctness of the two trefoil knots 2 ) ,

it avoids construction of the group diagram and applies word

for word to arbitrary knots on the torus 3 ) •

I. The numbers a,b are uniquely determined, up to order,

by the group Ga, b.

We first construct the subgroup C of G generated by

a,b
Aa is an invariant element* of G therefore
a,b
C is a normal subgroup of Ga, b. Let F = Ga, b/C. F is
generated by the elements IA = CA, E = Cs, and defined by the

relations fAa = 1, Eb = 1. As is easy to see, each element of

F has a unique normal form

where

0';; xl < a, 0 < x 2 < a, ••• , •.• 0 < yp-l < b, 0 .;; yp < b.

Lemma 1. F contains no invariant element other than 1.

x y
Namely, if IA 1 •. JB P is an invariant element in normal

form, then its commutativity with IA and the uniqueness of the

normal form implies yp = ° and p = 1, similarly, commutativity

with E implies xl 0, and thus our element is lAO EO = 1.

2) Math.Ann. 75 p.402.

3) As Herr DEHN has pointed out to me, his proof is also appli-
cable to all these knots.

* Element which commute~with all others (Trans.).

 226

Lemma 2. Each element of F of finite order is conjugate to

a power of fA or E.

Xl YP
Let r be an element of order c and let /A ••• E

be its normal form. Then at least one of the numbers

xl'yp = 0, otherwise r C ~ 1 because of the uniqueness of

the normal form. This proves the assertion in case p = 1.

Suppose it is true for all numbers < p. If say

YP
0, then
X -x x -x Yp - l
p = IA 1 P ..• E
"PrlA and the assertion holds for
x -xp
IAPrlA , hence also for r. (And analogously when x = 0.)
1

It follows immediately from Lemma 1 that C is the centre

of G b. Lemma 2 now shows that max(a,b) is the greatest
a,
finite order for an element of the quotient of G by its
a,b
centre, and it is therefore uniquely determined by Ga, b. Now
if we add commutativity of the generators to the defining

relations of F then we obtain an abelian group of order ab;

thus ab is the order of the abelianised quotient of Ga,b by

its centre, and thus uniquely determined by Ga, b. Thus

Theorem I is proved.

In order to present the automorphism group of Ga, b' we prove

Lemma 3: When a ~ b all the automorphisms of F are given by

VA'" T-l/Ar T, 13 1+ T-l:BsT} where r is prime to a and s is

prime to band T denotes an arbitrary element of F; when

 227

a =b there are the additional automorphisms

VA t+ T- I ErT, lB t+ T-l/AsTL

The given substitutions clearly generate automorphisms of F.

We have to show that there are no others. Now each automorphism

must send fA and B to elements /A' and B' of orders a

and b respectively. lA' and lB' cannot be conjugate to

powers of the same generator of F, otherwise the other generator

would have exponent sum 0 in each element generated by /A';S'

and thus would not appear in the group they generate. Thus when

a ~ b we must have lA' are

elements of F and rand s have the same meaning as above.

If {AI. t+ A' , B t+ E'} determines an automorphism of F, so does

VA t+ I1- I/Ar Il , E I+E s } where PE- I = II. But since IA must be

generated from I1- l lAr rr and BS, II can only have the form

AX BY. Now if we set T = /A-xp = BYE, we have /A' = T-l/ArT,

B' = T- I BST. (Analogously for a = b.)

II. All the automorphisms of Ga,b for a ~ b are given by

element of Ga, b: when a = b one must add the automorphism

{A t+ T-lB~T, B t+ T-IA~T}.

Since C is a characteristic subgroup of Ga,b i t is

mapped into itself by each automorphism, and each automorphism

of Ga,b uniquely determines automorphisms of C and F. If

{A t+ A', B ,+ B'} determines an automorphism of Ga, b' then we

'a ~a -~b '-b
must have A A = B = B , and by Lemma 3, for a ~ b,

T-IBS+kbT, where h,k are integers. It

 228

follows that r + ha s + kb. (Analogously for a = b.)

III. The automorphism group of G

a,b
, for a t b, is generated

by 3 elements I,J,K and defined by the relations Ia = 1,

Jb = 1, K2 = 1, (KI) 2 = 1, (KJ) 2 =1 as an abstract group; for

a =b there is an additional generator L with the relations

L2 = 1, KL = LK, LI JL.

In order to see this one has merely to denote the auto-

morphisms {A ~ A, B ~ A-1BA}, {A ~ B-1AB, B.+ B},

of Ga,b by I,J,K, and for a = b denote

the automorphism determined by {A ~ B, B ~ A} by L; and to

observe that C is the centre of G .

a,b
 TRANSLATOR'S INTRODUCTION 7

The next paper is an important unpublished work of Dehn

notes of a lecture to the Breslau Mathematics Colloquium on

February 22, 1922. The paper is mentioned a few times in

works of Baer and Goeritz (see below) but it exists only in the

form of a few mimeographed copies of handwritten notes, and

would probably have been lost if not for the careful preservation

of the Dehn Nachlass by Wilhelm Magnus. The remaining mimeo-

graphed copies are mostly in the Humanities Library, Austin,

Texas. In translating the notes I have also redrawn the

diagrams.

On the first two, introductory, pages there are some

interesting remarks on other lost proofs. Dehn cites the

process of Poincare [1904] (not 1905, as given by Dehn) for

deciding the number of topologically necessary intersections

of curves on a surface, and claims that the process can be

greatly simplified. As I have mentioned in connection with

the simple curve problem (i.e., the special case of self-

intersection), this simplification may have been via Dehn's

algorithm for the conjugacy problem. At any rate, Dehn

observes that such a process yields an algorithm for deciding

whether a given replacement of the generators, a k * a k ',

bk ~ b k '. in the fundamental group corresponds to a mapping class.

He is thereby assuming that any automorphism of the surface

 230

group can be induced by a homeomorphism of the surface,

a result which he also claims, but for proof cites only a set

of lecture notes, Nielsen [1921].

A copy of this set has recently been found by Fenchel,

and my translation of it will appear in the forthcoming edition

of Nielsen's works. Unfortunately, the proof is quite specific

to the genus 2 case, and the only clue it gives to the general

case is its use of the invariance of axis intersections

under automorphisms, a result which is proved generally in

Nielsen [1927, §9] and there credited in essence to Dehn.

A rather simple route from intersection invariance to a proof

that all automorphisms can be induced by homeomorphisms is

given in Marden [1976, p. 70]. After I spoke on the Dehn-

Nielsen theorem in Copenhagen in June 1983, Professor Borge

Jessen recovered a letter he had received from Nielsen, dated

11 December 1931, pointing out that the proof in Nielsen [1927]

can be replaced by a much simpler one. Nielsen then sketches

a proof essentially the same as the one given 45 years later

by Marden.

Returning now to the paper before us: it is not very

clearly written, but one can see at least the following revolu-

tionary ideas : decomposition of the surface into 3-holed

spheres, determination of simple curve systems by intersection

 231

and twist parameters at the boundaries of the 3-holed

spheres, i.e. by (6p-6)-dimensional integer vectors, where p

is the genus, and representation of mapping classes by trans-

formations of this (6p-6)-dimensional space which, at least

in the special cases considered, are linear* transformations.

Twist mappings appear, perhaps for the first time, and their

general significance is suggested.

Certain aspects of the curve representation and its

application to mappings were followed up in Baer [1927], [1928]

and Goeritz [1933a], [1933b], and the representation of mapping

classes was later investigated more thoroughly in Dehn [1938],

where it is shown that all mapping classes are represented by

linear transformations, generated by a finite number which

represent "twists". (For more on this, see the introduction

to the translation of Dehn [1938] below.) However, Dehn's

ideas did not begin to fulfil their potential until they were

rediscovered and given a new lease of life from the highly

original viewpoint of Thurston [1976] (see also Poenaru,

Laudenbach & Fathi [1979]). Part of the reason for Thurston's

success is that he reunites the combinatorial ideas with hyper-

bolic geometry, reversing Dehn's tendency to eliminate metric

proofs in favour of combinatorial ones.

* Though in certain cases the mapping breaks into different linear

pieces in cones with their apex at the origin (cf. p. 251).
 232

REFERENCES

R. Baer [1927] Kurventypen auf F1achen. J. reine u. angew.

Math. 156, 231-246.

[1928] Die Abbi1dungstypen der orientierbaren, gesch10ssenen

F1achen von Gesch1echt 2. J. reine u. angew. Math. 160,

1-25.

M. Dehn [1938] Die Gruppe der Abbi1dungsk1assen.

Acta Math. 69, 135-206.

L. Goeritz [1933a] : Norma1formen der Systeme einfacher Kurven

auf orientierbaren F1achen. Abh. Math. Sem.

Univ. Hamburg 9, 223-243.

[1933b] : Die Abbi1dungen der Breze1f1achen und Vo11breze1

vom Gesch1echt 2. Abh. Math. Sem. Univ. Hamburg 9,

244-259.

A. Marden [1976] : Isomorphisms between fuchs ian groups. Advances

in Complex Function Theory (Eds. W. Kirwan, L. Zalcman).

Springer-Verlag, 56-78.

J. Nielsen [1921] : Die Abbildungstypen geschlossener Flachen und

ihre Beziehungen zu unendlichen Gruppen. (Notes of

Breslau lectures, 9 and 11 of March 1921) .

[1927] : Untersuchungen zur Topologie der geschlossenen

zweiseitigen Flachen. Acta Math. 50, 189-358.

V. Poenaru, F. Laudenbach and A. Fathi [1979] : Travaux de Thurston

sur les surfaces. Asterisque, 66-67.

 233

H. Poincare [1904] : Cinquieme complement a l'analysis situs.

Rend. eire. mat. Palermo 18, 45-110.

W. Thurston [1976] : On the geometry and dynamics of diffeomorphisms

of surfaces I (pre print).

 ON CURVE SYSTEMS ON TWO-SIDED SURFACES,
WITH APPLICATION TO THE MAPPING PROBLEM

Lecture (supplemented) to the math. colloquium,

Breslau 11-2-1922 by Max Dehn

I Survey of previous results on the mapping problem: the fundamental

group of the closed surface has always been the starting point.

1) p 1. Ordinary ring surface (torus)

Mapping class S'

Ct l
S T
Sl
where I Ct l Sl r =
Ct 2 62
± 1

Ct 2 S2
T' = S T

we therefore get the extended modular group.

T Fundamental group

Generators S, T
{
Relation ST = TS

connected with the period transformation of elliptic functions.

2) Two-sided surface with arbitrary p:

Fundamental group
G'o~a"''" :' ••..•
{

Relation IT (a.
i=l ~

a) Necessary and sufficient conditions for the transformations

r

j a'
k Fl,k (a l ,··· , a P ; b l ,··· , bP )

l
A

-
b'
k FZ,k (a l ,··· , a p; b l ,··· , b p )
 235

to correspond to a mapping class are that, with fixed basepoint, a k and bk

have no irreducible double points, a k and bk have one irreducible point of

intersection, and otherwise no 2 of the 2p curves (with fixture of the

basepoint) have an irreducible point of intersection. It can be decided

whether the system (a k, bk) has this property following Poincare (Pal. Rend.

1905), with the help of the diagram of the fundamental group in the hyperbolic

plane. [The general process indicated by Poincare can be greatly simplified.l

Similarly, it can be decided by this method whether 2 mappings belong to the

same class (i.e. whether one can be continuously carried into the other).

(b) Each mapping class corresponds to an isomorphism of the

fundamental group with the same system A of transformations (immediately

clear). Conversely: each isomorphism of the fundamental group corresponds

to a mapping class with the same system t of transformations. (Dehn, see

autogr. Vortrag of Nielsen, Breslau 9 and 11-3-21). Thus the mapping problem

reduces to the isomorphism problem for the fundamental group. The proof uses

essentially "topological" properties of the diagram of the fundamental group.

3) Two-sided double ring (p = 2). The group of isomorphisms of the

fundamental group is generated by 3 elements. (Nielsen loco cit.) The

difficult proof depends on considerations from abstract group theory,

essentially on the diagram of the fundamental group.

It turns out that the group of 4 x 4 matrices with determinant +1 is

only a quotient of the group of isomorphisms. Only the positive value + 1

of the determinant comes into consideration here, because in a mapping which

carries the fundamental curves into themselves, any reversal of orientation

must occur in 2 fundamental curves simultaneously. A direct definition

of the isomorphism group (without the detour of 2a) has not been possible

as yet, because no means is at hand to present the relations between the

 236

generators of the isomorphism group.

II New approach (without use of the fundamental group)

The element of the fundamental group, i.e. the most general closed

oriented curve with a fixed initial point, proves to be too general in

one direction, and too special in the other, for the description of

topological relations on the surface of genus p > 1. The suitable object

of investigation for this purpose seems to be the system of finitely many

curves without double points and intersection points, and not including

curves contractible to a point. We shall consider two such systems to be

the same when one can be continuously deformed into the other on the

surface (i.e. when they belong to the same class).

1) Torus

The curves of the system meet S in the

points 1, 2, ••• , m, which lie on S in

that order. We assume that no curve,

after leaving r, returns to S on the

T
same side at r + 1 or r - 1 (as can always

be arranged by continuous deformation). Then there is no curve at all which

returns to S on the same side from which it left. Thus the curves of the

system determine a permutation of the numbers 1, ••• , m, which because of the

absence of intersection points, is cyclic. To a number r there corresponds

the number r' at which the curve leaving S at r returns to S on the

(arbitrarily chosen) positive side. We have r' =r + V (mod m), where v

is the total number of intersection points the curves of the system have with

T, each counted positively or negatively according as the curve crosses T

from the negative to the positive, or from the positive to the negative side
 237

(the positive side is the side of T on which the point 1 lies).

This construction uni~ely associates a number pair (~) with the curve

system, as we have defined it, and conversely. (We have first assumed m > o.
We also set (~) = (=~) and associate (~) with the system of Ivi curves

parallel to S. Then the scheme is complete.)

The number of closed curves of the system (all belonging to the same

class) is equal to the number of cycles of the permutation (r : v) and thus

equal to (m, V). The transformations(~) + (~) and (~) + (V : m) (which do

not change the number of cycles) correspond to mappingcof the torus onto itself.

Thus in particular, each pair (~), where (m, V) = 1, can be reduced to (~) or (~)
From this it follows that, using the mappings which correspond to these trans-

formations, each simple closed curve U can be mapped onto the curve S. Under

this mapping, any simple closed curve V which meets U once goes into a

curve with the symbol (~). and under further transformati::ms which leave
S = (~) fixed, into (~), i.e. it is mapped onto T. We therefore have : any

two simple curves which meet in one point can be mapped onto Sand T by

the mappings corresponding to the two symbol transformations. In order to

separate the mapping classes completely from each other it is necessary to take

account of orientation and to reflect this in the Symbols. For this purpose

we make the convention that when m > 0 the curves corresponding to (~) run

towards the positive side of S, and also that when V > 0 the curves

corresponding to have the same sense as S1 finally, the curves

corresponding to the symbol \(--~\

~ shall have the opposite sense to those

denoted by (~). This uniquely determines the orientation in all cases.

From (~) we obtain the following by successive transformations:

(~) + (~) + (m VV) + (m ~ V) + (m : V) + (m m V) (_~)

+

Similarly: (~) + (~) + (_~) + (-~)

 238

Consequently, each pair (~~) and (~:) of symbols which represent

2 simple curves u and V with a single point of intersection can be carried

into the 4 pairs of symbols (±~) and (±~). Each can be obtained by a mapping,

using the mappings which correspond to the transformations. Thus some already

known results are derived in a new way.

2) Determination of a curve system on the surface of genus p by

3p - 3 pairs of numbers.

a) p 2

upper half

The system has mi intersections

with a i . By continuous deformation

we can always arrange that no curve

which leaves a. at r returns at

l.

r ± I so that the intervening piece

lower half of curve together with the piece between

rand r ± I on a i is contractible to a point. The ml + m2 + m3 curve pieces

unite in pairs on the "upper and "lower halves of the double ring (m l + m2 + m3

must therefore be even). Such a half is a sphere with 3 holes a i . If we extend

the holes along the pieces of the curve system connecting them, until the

latter are replaced by direct contact of the ai' then, taking all properties

of the curve system into consideration, we obtain only the following possibilitie

of combination, corresponding to the 3 topologically different ways of covering

a sphere by 3 discs.

Here we let ~l be the number of pieces leaving a2 which unite on either

half with pieces coming from a3. Analogously for ~2 and ~3
 239

The corresponding "seam diagram"

is

Seam diagram

~l (~3) is the number of pieces eoanating from a 2 which unite in each half

with pieces from a 3 (a l ) - ~2 is the number of pieces emanating from a 2 which

unite in each half with the others from a 2 •

Seam diagram

)<1

~
i

The ~i and hence the seam diagram are thus completely determined by the

numbe~mi' For this reason the lower half always has the same seam diagram
ml + m2 + m3
as the upper half. On each of the seam diagrams there are ----~2----~ points,

which lie above and below in the same way on the 3 (resp. 2) segments of the

seam diagram. We can speak of corresponding points on the two figures. We shall
 240

give each point a double designation, corresponding to the two pieces which

unite in this point, say 1, 1 to 1, ml ; 2, 1 to 2, m2 ; 3, 1 to 3, m3 • The

point with the designation i,r k,s is where piece number r of the ith

division unites with piece number s of the kth division. The pieces

themselves are numbered in the same way as their intersection points with

the a i • If we leave one of the mi pieces of the ith division from the upper

seam diagram, then we again return from the lower to the upper figure, but

in general not at the same place, because the piece emanating from the point

i,r does not meet the lower figure at the point i,r but at the point i,r',

where r' _ r + V (mod m.).

L

Thus the curve system generates a substitution of the numbers of the

upper seam diagram (the numbers r. are determined up to multiples of m.

L L

by the remainder). In order to fix the Vi geometrically, we start with

the simplest case, in which the same 2 curve pieces are joined together on

the lower seam figure as on the upper. The system then consists of

ml + m2 + m3 closed curves.
2

Type a) Type b)

The simplest case also has the property that for type a)the system has no

intersection points with the curves b. (see fig.), i.e. the curves lie
L

wholly on the front or wholly on the back of the surface, and that for type b)
 241

the system has only m£ - mi - mk such intersection points with the curve b£

We now fix the points on the seam figure and construct the most general system

by "twisting" the m. curve pieces of the 3 divisions. The number

~

is then the number of intersection points of the m. curve pieces with bk

~

as well as with b£. The sign of Vi is taken to be positive, say, when the

mi curve pieces of the ith division constitute a right-handed screw system.

We denote the curve system by One can also follow the course

of the curves of the system on the seam diagram by inscribed polygons with the

displacement segments V. (cf. in the torus case the role of the m-gon within
~

the m-gon, with the displacement segment V).

5
Example ( 42 2, ]12 1, ]13 3
3

Divides into 2 closed curves

Seam diagram
..... 4 • •
f- '. •

Schema of the correspondence

i,r
I
upper
><
lower
,
k,r'

upper

seam diagram
 242

The two substitutions for the upper seam diagram:

(1,3) (1,1 3,1 2,2 2,4 3,3)

(2,3) (2,1 1,2 3,2 1,4 2,5)

The 2nd substitution is always the inverse of the first, when symbols for

coincident points are regarded as the same. The number of cycles of the

substitution equals the number of closed curves of the system. Each curve

system on the double ring (of simple, disjoint, non-contractible curves)

is uniquely determined by 3 pairs of numbers. The 3 second terms

VI, \)2, \)3 are arbitrary, but the first terms of

and their sum must be even. We shall understand

system that consists of

(~ and curves parallel to

b) P >2

mi intersection points of the

curve system with a i (under

the assumption, which is always

achievable, that no curve bounds

a disc in conjunction with a

piece of ai'. Emi must be even. For p> 2 it is no longer convenient, as

it was with p 2, to depend upon the seam diagrams of the upper and lower

halves. As one quickly realizes, even for p = 3, there are 5 different

types of seam diagram, and the type is by no means uniquely determined by

the mi , as was the case for p = 2. Moreover, for each system of m.

~
there

are infinitely many seam diagrams not continuously deformable into each other

on the surface. However, the following construction enables the case p >2

to be carried back to the case p = 2 by a kind of reduction.

 243

Suppose, say, that p = 4. Then in the upper and lower halves we add

one of the curves resp. c 2 meeting the curve b l5 and b 23 each

once, similarly on both halves, one of the curves c3 resp. c4 meeting

of the curves b l5 and b 34 each once (see fig.) Then the upper and

lower halves both divide into 3 spheres, each with 3 holes. Let R.. be
J.

the number of points of intersection of the curves of the system with c i '
(again under the hypothesis that no curve bounds a disc in conjunction with

a piece of ci). Then by II, 2a the kind of connection on each one of the

spheres of the 6 spheres is uniquely determined by the mi and b i • From any

(normal) system with given mi and R.i we obtain the most general one with the

same numbers by twists through the "angle" ~

m.
\!.
J.
resp. 2"K
R..
I)
i
along the
J. J.
boundary lines ai and ci between the spheres. A convenient way to

construct the normal system is to begin with one of the 3-holed spheres and

then construct normal systems of connections for all the spheres in a fixed

order, following the construction of the normal system (simplest case) for

p = 2. Thus the most general curve system on the surface of genus p =4

is uniquely determined by 5 + 4 number pairs. The second terms of these

number pairs are any integer;; the first terms are all ~O and satisfy the

following conditions:

ml + m2 + R. I , ml + m2 +R. 2 , m3 + ~l + ~3' m3 + ~2 + R.4 , ~3 + m4 + m5 ,

R.4 + m4 + m5 are even

In general we have:

The most general system of disjoint, simple, non-contractible curves

on a closed surface of genus p is uniquely determined by 3p - 3 number

pairs
ml R. R. m
( 2p-5 2p-4 p
\\ 1)2p_5 1)2p-4 \!
P
 244

The numbers ml + m2 + ,Q,l' ml + m2 + ,Q,2' ...

,Q,2k-5 + ~ + ,Q,2k-3' ,Q,2k-4 + mk + ,Q,2k-2'

,Q, m + mp +l ' ,Q, + m + mpH
2p-5 + p 2p-4 P are even

The numbers in the upper row are cardinalities, namely the numbers of

intersection points of the curves of the given system with the 3p-3 curves

a i , •.• , a p +l ' c I "'" c 2p _4 ' by means of which the surface is divided into

2 (p-l) three-holed spheres. The numbers in the lower row are arbitrary

positive or negative integers and they denote the twists which convert a

normal system with the same upper row into the given system.

Since for p =I the class is determined by ~pair, and for p =0

no distinct classes exist at all, we have an exact analogue to Riemann's

theorem on classes of algebraic functions of the same genus. (Connection?)

3) Transformation, reduction of systems of number pairs

The association of number pairs with curve systems is important first

of all because it is related in a simple, natural way to the course of the

curves in the system, so that one may expect that significant geometric

properties of the system are expressed in simple arithmetic properties of

the system of number pairs. This expectation is fulfilled.

a) p = 2

We consider the following 4 continuous mappings of the surface onto

itself.

I Rotation about the triple

21f
symmetry axis by
3

II Reflection in the

equatorial plane (through b l , b 2 , b 3 ).

(lxis
 245

III Exchange of the outer and inner spaces (or upper and lower,
back and front)

IV Twisting the tube neighbouring a l through 2rr

I, II and III extend to continuous mappings of the whole
space onto itself, and therefore carry curve systems
into ones which are isotopic relative to all of
m.
I corresponds to cyclic interchange ( ~
mi+l

II corresponds to

IV corresponds to

III is somewhat more difficult to express arithmetically.

For type a) (m i and Vi ~ 0, III corresponds to tne arithmetic

transformation: mk + m/! nL~ )

2

Example
5 IV-; I ~ (4 5 3)
3 ,-2 -2 -1

II ., (4 5
n
,
2 2

(
~)
III 3 3
2 1

I, IV -1 , I:; (
3 3 4\

., ,
1 2 1)

III ( 3
2
2
2 i)
IV-l,~
( 3
2
2
0 i)
---~
III
( 1 3
~)
,
1 2

I,IV -1 ,II ( 1
~)
3
..... 0 1

III
i ( 2 1
i)
,
2 0

I,IV- l (
~)
;.
2 1
0 0

III
1 ( 0
0
0
1 ~)
 246

5
and thus the system ( 42 3
3
2) consists of two non-separating curves
not belonging to the same class. (cf. fig. p. 242)

The general arithmetic form of the transformation III will be given

here only for the mi , which is sufficient for our purpose:

m. t+ IVk + vl!,l+IvkHvl!,1 IVkl+lvl!,I-lvk+ vl!,1

l.
2 + I 2 - l1i \ - l1i

where we set
Imi + ml!, - mkl+lm i + mk - ml!,1
11·l. = ~ + ml!, - 2
2

The transformations carry the types into each other. One sees without

difficulty that each number pair triple can be converted, by means of

the 4 transformations, into a triple in which each pair contains at

least ~ zero, and indeed into the following triples:

1) ( VI0 V02 ~) 1)

2) ( VIo m02 V30 ) 2)

3)
( 0 112+m3 m3)
VI 0 °
3)
( ml 112+ml O )
4) o 0 V3

To prove reducibility it sufficeo ~o snow tnat, by suitable transfor-

mations, the maximum of the m. can always be reduced, until one of

l.

the 3 minimal systems above is reached. Now let (~~~; ~;) be the
 247

given system, ml ~ m2 ~ m3 , then, if we do not have ml = m2 = m3 ,

ml is transformed by III into mi < mI' Namely, for type a)
m2+m 3 m2+m 3
mi = ---2--- < ml ' for type b) mi = ---2--- + ml - m2 - m3 < mI'
ml ml m2
Then one uses II and IV to make IVll ..; -2" ' then mi ..; 2 + 2 ..; ml ,
m iii
mj ..; ...!. + 2 < ml , so that fewer numbers attain maximal value ml after
2 2
the transformation than before. Only the case ml = m2 = m3 remains. In
m.
case we do not have O<V l = 2'
J.
one can begin as before by reducing one

of the mtwithout making the others larger. However, if v. = mi one

J. -
2
uses II to replace VI by -VI' and then obtains the desired reduction

similarly.

(3 ~ ) .... l~) )]
20 (9 16 9 16 9
Example: 1 -10 \10 -7 .... [( 1 -7 1

.... ( 6
8
2
-3
6
8 ).... [( 6
2
2
1
6
2 )--] .... ( 3
1
4
5
3
1 )
.... [( 3 1
4
1
3
1 ) .... ].... ( 2
2
2
1
2
2 ) .... [( ~ 2
1
2
0 ).... ]
.... ( 1
1
0
1
1
1 ) .... [( 1
0
0
1
1
0 ) .... ] .... (\ 0
1
1
0 ~)

b) p = 3

Normal system

We base our discussion on a decomposition of the p =3 surface

into 4 three-holed spheres, so placed as to exhibit tetrahedral symmetry

(see fig. Correspondingly, we denote curves denoted ai and c.

J.

above by al to a 6 , and the numbers mi and R.i by mi ,· •• , m6 , and

the numbers Vi and 0i by V l ' •.• 'V 6 • To determine a normal system

we shall associate each of the curves a. with one of the four curves
J.

b I , b II , b IV ' bv intersecting it in the external space, say:

 248

For the sake of simplicity we think of all the intersection points

of the curves lying on the front side of the surface. We can now

place the connecting segments in such a way that none of them are inter-

sec ted by the curves b I , b II , b IV ' bv except those which return to

the same a i from which they depart. These will be placed in such a

way that they have an intersection point with the ai on the piece

of the sphere in question lying opposite b7 , and an intersection point

with the bk associated with the ai' cf. the example shown,

2 3 I 0
o 0 o 0

This determines a curve system

with the numbers mi uniquely

up to continuous deformation.

If we take, say, the right

handed screw system as positive,

then it becomes uniquely deter-

mined which system is represented

by the symbol (ml m2 m3 m4 mS m6)

VI V 2 V3 V 4 Vs V6

Transformations

I a) Rotation through T about the 3-fold symmetry axis through the

2

"midpoint" 0IV of the "triangle" I 3 3 V II I

b) Rotation through ~jl about the 3-fold symmetry axis through the
"midpoint" 0I of the "triangle" 3 4 S IV II V
 249

II Reflection in the equatorial plane through I, II, IV, V.

III In order to fix a particular kind of exchange of internal and

external space, we introduce (see fig.) the curve b II on the front

side enclosing b I and b II , and the

curve bVI on the back side,

enclosing b I and bv . Then a

mapping of the surface which

exchanges the internal and

external spaces is determined

by the following exchange rules

between the curves a i and b k :

1 ++ I (with orientation), 2 ++ II, 3 ++ III, 4 ++ IV, 5 ++ V, 6 ++ VI.

This is our transformation III.

IV Twist of the neighbourhood of al through 2IT.

Ia corresponds to the substitution

and

and

II corresponds to (_VV~.· )
~

IV corresponds to (Vl )
Vl+ml

III, as for the case p 2, is complicated, and will not be investigated

further here.

Whether the 5 transformations I-IV suffice for a complete reduction,

as do the 4 transformations I-IV in the case p=2, is doubtful.

 250

4. Continuous (homeomorphic) mappings of the p 2 surface onto

itself.

The minimal systems attainable by the transformations are in

part "equivalent" to each other, i.e. convertible into each other by

continuous surface mappings. (See Enzyklopadie, Analysis situs B7),

and indeed obviously

0 0 0 0
( \)1 \1 2 ~) is equivalent to ( \1 3 \1 2 ~), likewise

0 m 0 m 0
( \11 02 ~) to ( \1 3 02 \11 ), and

( o ]12+m3
( m30
m3 ]12-Hn 3
\
\. \1 1 0 0 )
to
0 ~),
for they are mapped into each other when the left and right halves of

the double ring are exchanged. We add these transformations, as V, to

those previously given for p 2. Further, it is easy to see that

0 2]12-Hn3 m3\ (0 2]12 0 \

(
\1 1 0 0) is equivalent to \110m3) - it is exchange of latitude

and longitude ln one (the right) half of the double ring, whereby the

\1 1 curves parallel to a l and the ]12 separating curves go into each

other, and the m3 curves parallel to bl go into m3 curves parallel

to a 3 • We add such a transformation, as VI, to the above, and note that

we can construct the transformation III by combining VI and V. Thus it

follows that :

I, II, IV, V, VI generate each class of continuous mappings of the

p = 2 surface onto itself. For with these 5 transformations we can first

transform a system of 3 curves W is separating and

non-contractible, Ul and U2 non-separating, and

 251

disjoint from W and each other, into the system (a l , a3' d 2 ) =

(~
2
0 n, where d
2
= (~
2
0 ~)
As a result, any 2 curves which meet

Ul , U2 in single points but do not

meet W, go into curves e 1

\)1 \)2 ~) and (~ \)2 1\)3 ). and by further mappings
1

(~ 0 ~) and (~ 0 1
1 1
corresponding to IV and I, into o)' .I..e. I.nto
. b 3 and

bl . But our mappings, namely II (reflection in the vertical plane through

the b'S) transformed with the mapping III, also enable a l and b 3 to be

mapped into themselves with reversal of orientation, where d 2 , a 3 and b l

go into themselves (by the mapping II VI (II)-l (VI)-l, which sends al

-1 -1
and b 3 into themselves, a 3 and b l into a 3 and b l resp.) Thus we have

shown chat each mapping can be composed from these 5 mappings, for we have

already remarked in the introduction that, when the orientation of al , a3

and b 3 is given, the orientation of b l is also determined. But a mapping

which carries a l , a 3 , b l , b 3 into themselves without change of orientation

belongs to the identity class.

This derives the result of Nielsen in a direct, elementary way. That

5 generators were used here matters little, for naturally one can show

that the transformations I, II, IV, V, VI can be composed from the 3

Nielsen transformations. We have also seen that these transformations can

be represented by linear transformations of the 6 dimensional (~i~iL

number lattice, and indeed those corresponding to I, II, IV, VI are the

same for all lattice points. On the other hand V is different in

different sectors with the zero of the lattice at the vertex. This

holds incidentally, for the 5 transformations I-IV for p=3.

 252

Thus the way is opened for a precise study of the mapping class

group for p=2. Moreover, it is to be expected that the solution of the

problem for p>2 will involve no further serious difficulties. It should

be noted that the peculiarly non-analytic character of the set of linear

transformations is not to be avoided. On the contrary, a greater

symmetry is perhaps attainable by introduction of a second seam diagram,

corresponding to the transformation III.

5. Knots and links

Our representation brings the gain to knot and link theory that one

is able to represent the general knot or link uniquely by a series of

number pairs. Further: the transformations I, II, III for p=2 and p=3

transform knots and links on the surface into others which are spatially

equivalent (isotopic). The same holds for the special case of IV.

Thus for the first time we have the possibility of systematically treating

previously unassailable problems of the reduction of knots and links.

However, the 4 transformations certainly do not suffice for a complete

explanation in the case p=2 or p=3.

Example. Trefoil knot on the double ring

--------------,
I ,_ _ _ _ _ _ _ _ _ _ _ •
, ( 2 3 1
) I • ~"
I 0 2 0 I'
I
I -tIl 'I' "' ..I
,•
I
I I ____
...,
j I
( 2 0 2
)
I 1- _ _ _ _ _ _ _ _ - .
1..---------- __
1 2
'------_ ... 0

It therefore lies on the simple ring and has the symbol there Less

trivial examples may of course be given for p>2.

 TRANSLATOR'S INTRODUCTION 8

The final paper in this series is in effect a vindication

of Dehn's 1922 Breslau lecture, although it did not appear until

1938. In it Dehn studies mapping classes by their action on

simple curve systems, for orientable surfaces of arbitrary genus p,

finding that they are represented by linear transformations of

the space of (6p-6)-dimensional integer vectors, and giving

a finite set of generators, which represent twist maps (now known

as Dehn twists) . The proof is by induction on the complexity

of the surface and the mapping, with a very complicated base step

which requires detailed analysis of spheres with up to 5 holes.

Perhaps because of its very demanding proof, the result went

unnoticed until it was rediscovered independently by Lickorish

[1962]. His proof is similar in concept to Dehn's but much

shorter, using twists to reduce complexity until a very simple

base step is reached. Some further simplifications are made

in Birman [1977]. An impressive aspect of Lickorish's redis-

covery is that he made it in conjunction with a rediscovery of

Dehn surgery, when Dehn himself had not connected these two ideas!

The work of Lickorish undoubtedly enables us to skip many

of the details of Dehn's proof, but it does not replace the whole

paper. Dehn had another idea which also went unnoticed, even

when it was rediscovered in the unpublished, but famous, Thurston

[1976]. This was the idea of studying the mapping class group

by its action on the space of simple curve systems. As mentioned

above, Dehn took this idea far enough to represent mapping classes

by linear transformations of a (6p-6)-dimensional space of integer

 254

vectors, but not far enough to understand the geometric meaning

of this space.

Thurston [1976] shows how simple curve systems can be

interpreted as "rational points" in the boundary of a space of

surfaces of genus p, known as Teichmliller space T T is

p P
defined as the space of hyperbolic structures on a surface of

genus p, and thus requires a return to the hyperbolic geometry

Dehn had abandoned in his quest for purely combinatorial proofs.

Fricke and Klein [1897] found that Tp has dimension 6p-6,

while Teichmliller [1939] found a natural metric on Tp , under

which the mapping class group acts on Tp as a group of iso-

metries. The action extends continuously to Thurston's boundary

points, which arise from degeneration of the hyperbolic structure,

and the simplest type of degeneration yields simple curve systems.

Thus Dehn's action of the mapping class group on simple curve

systems can be viewed as the restriction of its action on the

whole of Tp and its boundary. What Dehn did not know is

that mapping classes are much easier to understand when they act

on this large space, because Tp is in fact a topological ball.

Dehn was not able to give a finite set of defining relations

for the mapping class group of arbitrary genus p. This was

first done by McCool [1975], using algebraic methods, while a

more geometric proof was given by Hatcher and Thurston [1980].

 255

REFERENCES

J. Birman [1977] The algebraic structure of surface mapping

class groups. Discrete Groups and Automorphic

Functions (Ed. W.J. Harvey), Academic Press, 163-198.

R. Fricke & F. Klein [1897] : Vorlesungen tiber die Theorie der

Automorphen Funktionen I, Teubner, Leipzig.

A. Hatcher & W. Thurston [1980] : A presentation for the mapping

class group of a closed orientable surface.

Topology 19, 221-237.

W.B.R. Lickorish [1962] : A representation of orientable, combinatorial

3-manifolds. Ann. Math. 76, 531-540.

J. McCool [1975] Some finitely presented subgroups of the auto-

morphism group of a free group. J. Algebra 35,

205-213.

O. Teichmtiller [1939] : Extremale quasikonforme Abbildungen und

quadratische Differentiale. Abh. Preuss. Akad.

Wiss., math.-naturw. Kl. 22, 1-197. Collected

Papers, Springer-Verlag, 1982, 335-531.

W. Thurston [1976] : On the geometry and dynamics of diffeomorphisms

of surfaces I (preprint).
 THE GROUP OF MAPPING CLASSES

(The arithmetic field on surfaces)

CONTENTS

Introduction 259

§l. Generalities on self-mappings of surfaces 261

a) The group of mappings and the group of mapping classes

b) Different types of mappings

c) Homotopy with fixed and movable boundary

d) Indicatrix

§2. The self-mappings of the one-, two-, and three-holed spheres 264

a) One-holed sphere

b) Two-holed sphere

c) Three-holed sphere with movable and fixed boundaries

§ 3. Special mappings of the four-holed sphere 273

§4. Self-mappings of the torus and the one-holed torus 276

a) Torus

b) One-holed torus
 257

§S. The arithmetic field on the two- and three-holed spheres

and the torus 278

a) General

b) Two-holed sphere

c) Torus. Orientation of curves

d) Transformations of the arithmetic field on the torus by

mapping classes. Generation and relations.

e) Three-holed sphere. Arithmetic field. Invariance under

homotopic transformations

§6. The arithmetic field or the curve systems on the one-holed

torus 298

a) Introduction of the field

b) Transformation by mappings. Examples.

c) One-holed torus with fixed boundary. Connection with the

trefoil knot

§7. Arithmetic field on the four-holed sphere 311

a) System of closed curves on the four-holed sphere

b) Mappings of the four-holed sphere and the action on the

arithmetic field

c) Derivation of invariants of a curve system from the arithmetic

presentation

d) Orientation. Examples

e) Geometric presentation of the symbols. View of higher cases

f) Curve systems on the four-holed sphere with endpoints on a

boundary
 258

1) Normal form

2) Arithmetic field

g) Four-holed sphere with fixed boundaries

1) Derivation of a relation

2) Application to the two-holed torus with fixed boundaries

3) Twists along the boundary of singly-bounded surfaces

4) Twists along separating curves on closed or singly-

bounded surfaces

§8. Five-holed sphere 335

a) Coordinate systems

b) Presentation of a system of closed curves

c) Reduction of symbols. Generation of mappings

§9. Generation of the mapping classes for the sphere with n holes 342

1) - 5) Lemmas

6) Generation with the help of complete induction

7) The five-holed sphere as an example

8) Direct exhibition of generators on the basis of

a cyclic ordering of boundaries. The number of generators

§lO. Generation of the mapping classes for every orientable surface 353

1) and 2) Lemmas

3) Generation with the help of complete induction

4) Double and triple torus as examples

5) Direct exhibition of generators on the basis of a normal

representation of the surfaces. The number of generators.

 259

6) Arithmetic field in the general case.

Introduction

In combinatorial topology, topological concepts are represented

by arithmetic concepts. Thus, in principle, all problems of

combinatorial topology are reduced to arithmetic problems. However,

this reduction is of no use for the resolution of the problems in

most cases, because the corresponding arithmetic problems have

little relation to known results or methods. This is especially

true of all problems in which homotopic transformations are considered

non-trivial or, as one can also say, in which the exceedingly

numerous and hard to visualize constructions of simply connected

polyhedra of different dimensions corne into play. In an earlier

work* I have attempted to represent this construcuion arithmetically

in an understandable way for two-dimensional polyhedra. In doing so,

I showed that homotopy problems yield arithmetic problems which fall

outside the extensive domain of group theory. They concern more

general operations which are difficult to investigate, the totality

of which I have covered by the name "games".

The situation is quite different when one investigates problems

in which homotopic transformations are neglected. Here one can

greatly benefit from methods and results of group theory in many

cases. For manifolds, even those of higher dimension, the

* Uber kombinatorische Topologie, Acta math. 67 (1936), 123-168

(Translator's note).
 260

fundamental group of POINCAR~ opens a way to attack such problems.

For two-dimensional manifolds, the theory of mappings in particular

has been worked out with great success in recent work which

collects all mappings which differ only by homotopies into a class

(see §l).

In the present work the mapping classes are represented as

operations in an arithmetic field on the surface. The individuum

of such a field cannot be a point on the surface, because a point

can be sent to any other point by a homotopy. In fact we choose

the individuals to be arithmetic representations of certain curve

systems which are partitioned into sets which collect together all

systems related by homotopies. Each such set is represented by

one curve system, which is uniquely determined by a number of

integers. One can consider this number to be the dimension of the

arithmetic field. The sequence of integers which determines a

particular curve system or the set of systems it represents is the

individuum of the arithmetic field. The mapping classes induce

transformations of this arithmetic field which are linear transfor-

mations related to modular substitutions.

This arithmetic representation makes it possible to solve a

series of simple mapping problems. The present work uses these

solutions to seize hold of more general problems, for which the

arithmetic field and its transformations are still only used in a very

rudimentary way. In this way it is possible to give a finite set

 261

of generators for the mapping class group of an arbitrary surface.

These generators are all similar types of mapping of the surface

onto itself, namely twists along certain curves, which are simply

related to usual representation of surfaces of higher genus.

This is a settlement of sorts, but the theory of transformations

of the arithmetic field is very incomplete. A continuation in this

direction would be of great significance. Certainly, with the

presentation of the transformation formulae, the mapping class group

is "represented" by linear transformations. But perhaps this

representation will also make possible an attack on the conJugacy

problem for mapping class groups, which has been the main theme of

previous works in this field, from a new side.

Because of this possibility, the present work is to be considered

only of a preparatory character. This character is also expressed

in the detail devoted to simple things and in the fact that many known

results are derived, either as a necessary basis for the method, or as

useful examples.

§l.

Generalities on self-mappings of surfaces

a) The mapping ~ of a surface onto itself is given by a

one-to-one correspondence between the vertices, edges and cells of

one decomposition Ll of the surface and another decomposition L2

 262

such that corresponding vertices lie on corresponding edges and

the latter lie in the boundaries of corresponding cells. One can

However, ~ maps an arbitrary decomposition

E of the surface onto a determinate decomposition E'. This is

because E results from El by internal subdivision. The internal

subdivision is sent by ~ to a (topologically) determinate sub-

division of E2 . We can therefore give the mapping for an arbitrary

and thus obtain ~(E) = E'. In particular, we obtain ~(E2) = E3

in this way, and hence ~2(El) E3 , and by continuing we obtain all

powers of ~. If ~ is another mapping, then by the above we can

also apply ~ to E2 and thus obtain E4 = ~(E2) = ~~(EI). We

can therefore compose the mappings of the surface, and the mappings

of the surface form a group.

However, it would be impractical to consider this group directly,

since it cannot be generated by a finite number of elements. One

therefore considers a quotient group, which corresponds to a division

of the mappings into classes one reckons the mapping ~ to belong

to the identity class if ~(EI) = E2 results from El by a homotopy

(that is, by deformation). Correspondingly, two mappings belong

to the same class when one is the product of the other with a mapping

in the identity class. The mappings in the identity class constitute

a normal subgroup of the group of mappings. Because, if E is

such a mapping, then ~E~-l is the corresponding mapping on the

polyhedron ~(E) resulting from E, and hence also homotopic to the

 263

identity. The quotient by this normal subgroup, the mapping class

group is the subJect of our investigations.

b) A non-identity mapping can map some decomposition onto itself,

in case the decomposition admits a non-identity mapping onto itself.

Such a mapping is always of finite order, i.e. it has a power which

not only belongs to the identity class, but is itself the identity

mapping. Other mappings are not of this kind and have infinite

order, so that none of their powers even belong to the identity class.
th
Finally, there may also be a mapping of which a finite, say the n

power belongs to the identity class, without any mappings in its

class having an nth power equal to the identity. I.e. we would

have ~n = E, where E is a deformation, while (E,~)n is different

from the identity, no matter which deformation E' is chosen. We

shall give examples of the first two kinds of mapping in the §§

which follow. Whether there are any of the third kind is doubtful.*

c) For bounded surfaces one can choose the identity class in

different ways. One either lets the identity class consist of

mappings which are homotopic to the identity without being the identity

on the boundaries, or else one requires that each such mapping leave

the boundaries pointwise fixed. This second, smaller, identity class

is also a normal subgroup in the group of mappings. But when one

uses it as a basis, one is restricted to mappings which leave the

boundaries pointwise fixed. These mappings are subdivided more

finely by their identity class than the group of general mappings

* Nielsen showed there are none in Acta math. 75 (1942), 23-115.

(Translator's note.)
 264

is subdivided by the larger identity class.

If "punctured" surfaces are under consideration, i.e. if each

boundary is contractible to a point, then the two identity classes

coincide. Because each deformation leaves each boundary pointwise

fixed, since it consists of only a single point. In addition to

the mappings which leave the boundaries fixed one may also mention

the special mappings which permute the boundaries (punctures).

d) A mapping sends an oriented curve to another curve with a

determinate orientation. A mapping (not in the identity class)

either reverses or preserves the indicatrix. The group of mapping

classes which preserve the indicatrix is a normal subgroup of the

group of general mapping classes. When the contrary is not expressly

stated in what follows, mappings are assumed to be indicatrix

preserving (orientation preserving) .

§ 2.

Self-mappings of the one-, two- and three-holed sphere

We shall call a sphere with n holes an n-holed sphere and

denote it by Ln.

a) It is practical to begin with direct consideration of a

few quite simple mapping problems, before considering the arithmetic

field on surfaces. The mapping class group of the one-holed sphere

 265

(the disc) is the identity group. All mappings belong to the class

of the identity, whether the boundary is taken to be fixed or not.

The proof proceeds by successive homotopies from one subdivision to

another (topologically the same).

b) The mapping class group of the two-holed sphere (the cylinder)

with non-fixed boundaries is the group of order two (exchange of the

boundaries), but if boundaries are not to be exchanged (though their

points are not fixed) the group is the identity. with pointwise

fixed boundaries the mapping class group is the infinite cyclic

group. This happens because a point Yl of one boundary can be

connected to a point Y2 of the other boundary in infinitely many

different ways (see the connections Y1 YY 2 and Y1 ZY 2 in Fig. 1)

7
."
I--~--I

Fig. I.

Fig. 1 represents a mapping which generates this cyclic group. It

consists in the pointwise correspondence between the polygon

 266

This correspondence

leaves the boundary points fixed (only the points Y and Z are

exchanged) . The polygons form two topologically equal decompositions

of the cylinder. We shall call this mapping a twist. When the

boundaries are fixed the twist is not a homotopy. The proof

proceeds by first considering a simple arc YI YY 2 , connecting one

boundary to the other, and its image Yl 'Y'Y 2 '. If the boundaries

are allowed to move under homotopies, then all these connecting

lines are homotopic to each other. But if the boundaries are

pointwise fixed, in which case Y ,

1

infinitely many non-homotopic connecting lines, which are determined

up to homotopy by the number of times they loop around one boundary

in the positive or negative sense. This number is given by the

algebraic sum of the positive or negative crossings of a connecting

This number associated

with Y1 Y'Y 2 is invariant under homotopy, but the twist ~ shown

in Fig. 1 raises or lowers it by one. This is easy to see when

one deforms the mapping shown in Fig. 1, i.e. when one replaces it

by a mapping 6' in the same class. In fact (see Fig. 2) we take

a point Y , on in the neighbourhood of Y, and a point

1
 267

Fig. _.

Z ' on Zl Z2 in the neighbourhood of Z. We then replace the

I

connecting line YI ZY 2 by YI YI 'ZY 2 and the connecting line

We can use a homotopy to arrange that the

mapping remains the identity on the segments YIY I ' and ZIZI'.

If we now have an arbitrary connection between YI and Y2 , then

we move all its points of intersection with Zl Z2 and YI Y2 homo-

the connection between the last point of intersection on ZIZI' and

Y2 can then be deformed into a connection between this point of

intersection and a point P on YIY l ' without any other inter-

sections with Zl Z2 or Yl Y2 , together with the polygonal path

The mapping 6' then deforms the segment PY I 'YY 2 into

the curve PY l 'ZY 2 and we therefore get one more intersection point

on Zl Z2' as claimed, because the remaining part of the deformed

connecting line can be assumed fixed under the mapping.

 268

A line connecting Yl and Y2 which has the number zero is

By a suitable number of repetitions of the

mapping ~ or its inverse, YI Y2 can therefore be converted into

a line homotopic to an arbitrary line connecting Yl and Y2 .

But the correspondence between YI YY 2 and Yl 'Y'Y 2 ' or YI Y'Y 2

respectively completely determines the class of the mapping,

because cutting along YI YY 2 converts the two-holed sphere into

the one-holed sphere, for which all mappings belong to the class

of the identity. Thus the mapping class group is the identity in

the first case (movable but not exchangeable boundaries) and the

infinite cyclic group in the second (fixed boundaries), as asserted

at the beginning, and it is generated by the mapping ~. This group

induces the group of translations in the numbers representing the

lines connecting the boundaries.

I have presented the proof in detail because here we have the

simplest manifestation of curves being given (up to homotopy) by

numbers, and transformations of these numbers giving the group of

mapping classes.

c) For the three-holed sphere (L 3 ) the group of mapping classes

with movable boundaries is the symmetric group of permutations of the

three boundaries l , and the group with movable but not permutable

lIn agreement with MAGNUS, Math. Ann. 109.

 269

boundar~es
. .~s the ~' d '
ent~ty
1 The proof proceeds by considering

two (simple) connections between the same ~ boundaries and

showing that they are homotopic under the stated assumptions. We

use three connections AB, CD and EF to divide the sphere into

two discs (see Fig. 3). We shall prove that each connection V

Fig. 3.

between A and B is homotopic to AB on L3 . The connection

V is determined by its successive intersections with AB, CD and

EF. We can assume that no consecutive intersections lie on the same

one of these segments, because the corresponding piece of V

could then be deformed into a piece on AB, CD or EF, and then

a further displacement of V into V' would result in two

fewer intersection points. In the same way we can arrange that

first intersection point does not lie on AB. If it lies on

CD, say at X, then all other intersections lie on AB and XC,

because an intersection Y on EF would imply that all later

IThis is the basis for the works on curve systems on surfaces by

DEHN, Autogr. Vortrag, Breslau 1922 and R. Baer, Journ. f. Math.,
vols. 156, 160.
 270

intersections lay on YE and DX, and V then could never arrive

at the point B. Hence in this case V makes a multiple circuit

round the boundary BC, which can be reduced by deformation so

that V no longer has intersections with the three connecting

segments, and hence it bounds a disc on L3 in conJunction with

AB, i.e. it is homotopic to AB. The remaining case is where

the first intersection X of V lies on EF (see Fig. 4). If

the next intersection then lies on DC, the only possibility

r A

Fig. 4.

for V is to make circuits around the boundary CB, which can be

ruled out. We can therefore assume that the next intersection

Z lies on AB. Then the third intersection can again lie on DC,

and we have circuits around the boundary CB, which we can rule out.

Or else the next intersection U lies on XE. For the same

reason we can now assume again that all later intersections lie

alternately on AB and FE. We then have circuits around the

boundary AF, which can likewise be ruled out. This completes

the proof that V is homotopic to AB. But we also see from

the proof that, with fixed boundaries, V is always homotopic

 271

to a number of circuits of the boundary AF followed by circuits

of the boundary BC. We note that it also follows that thr~e

disJoint lines connecting pairs of the boundaries can be deformed

by a homotopy with movable boundaries into AB, CD and EF.

This follows immediately when one deforms V into AB and then

considers the L2 resulting from L3 by cutting along AB and

the boundary connections on it.

An arbitrary mapping ~ of L3 with movable boundaries may

carry AB into v. A homotopy H may likewise carry AB into

v. Then the given mapping ~ differs from H by a mapping 'I' of

the two-holed sphere which results from L3 by cutting along

AB, where the points A and B remain fixed. But each mapping

of a two-holed sphere, under which at least one boundary is movable,

is a homotopy. Thus 'I' is a homotopy of the two-holed sphere under

which the two segments corresponding to AB remain fixed.

Consequently, 'I' is also a homotopy of L3 . Thus ~ differs

from H only by a homotopy and it belongs to the class of the

identity. This completes the proof that the group of mapping

classes of L3 with movable boundaries is the identity.

The group of mapping classes of L3 with fixed boundaries

is the free abelian group with three generators. The three

generators are the twists along the three boundaries. A twist

along a boundary curve results from a twist of the two-holed sphere

 272

with fixed boundaries, on the surface itself the twist curve

lies between the boundary itself and a parallel curve. In our

case we shall choose the three parallel curves to be disJoint.

The proof of our assertion follows easily from the foregoing.

Because the mapping ~ of L3 may carry AB, the line connecting

the first and second boundaries, onto V. As we have seen above,

we can convert AB into a curve V homotopic to V (with

fixed boundaries) by twists ~l and ~2 along the first and

second boundaries. I.e., when n l twists ~l and twists

-n -n I
~2 are needed the rnapping l ~ = ~l l ~2 2H ~ carries AB into

itself, whe.re H' is a homotopy which carries V to V. If we

again cut the L3 along AB, the result is an L2 . But a mapping

~ of L3 which is the identity on AB and which by hypothesis

leaves the three boundaries of the L3 fixed is a mapping which

carries L2 onto itself with fixed boundaries, and hence by b)

it belongs to the class of a suitable power of the twist ~3 along

the third boundary. Thus

where H" is a homotopy of L2 with fixed boundaries, and hence

also a homotopy of L3 . Hence, since the homotopies form a

normal subgroup of the group of mappings,

n n n
H' -1 2 1 3H "
~2 ~l ~3

IThe order of operations is always the right to left order of the

symbols representing the operations.
 273

i.e. the twists along the three boundaries gene~ate the mapping

class group.

The fact that these twists commute with each other follows

from the assumption that the regions in which they occur are

disJoint. There are no further relations between these three

n l n 2 n3
twists. This follows easily. If, say, 61 6 2 6 3 1 in the

mapping class group, then we consider a connection v 12 between

the first and second boundaries. with a suitable choice of the

twist strip along the third boundary, v 12 is not changed at all

nl n 2
by 6 3 , and then 61 6 2 only carries i t into a homotopic curve

if nl and n2 are both zero. But since a mapping belongs to

the identity class only when each connecting line v 12 is carried

nl n2
to a homotopic line, 61 6 2 can only belong to the identity class

in case n l = n2 = o. We obtain the condition nl = n3 = 0

similarly. Thus 61 , 6 2 , 6 3 generate a free abelian group.

3.

Special mappings of the four-holed sphere

The four-holed sphere L4 possesses mappings with movable

but not permutable boundaries which do not belong to the identity

class. We denote the four boundaries by r l , r2, r 3, r 4 (see Fig. 5)

 274

Fig. ;.

and consider three closed curves YZ and ST and UV which

respectively separate and from and and

from r2 and r 3 , and r2 and r4 from and We now

consider (by §2) the twists along ZY, ST and UV. These are

mappings which do not belong to the class of the identity, even with

movable boundaries, because the twist along ZY, e.g., converts ST

to the curve f = SY 1 Z2TY 2Z1 (see Fig. 5) which is not homotopic

to ST. One sees this most easily by considering the fundamental

The latter is the free group on the three generators

and within it ST corresponds to a conJugate of r l r 4,

-1
corresponds to a conJugate of r l r 3r 4r 3 But the latter

element is not a conJugate of rl r4 in the free group. Thus f

is not homotopic to ST and the (twist) mapping which carries ST

to f is not a homotopy. One reads the representation of f in

the fundamental group from the course of the two curves in the

polygon formed by cutting L4 along connections to the four

 275

boundaries from a point 0 (see Fig. 6).

o
Fig. 6.

The ~ holds for the twists along ST and UV. We shall see

later (§7), that one can generate a mapping of L4 with movable

but not permutable boundaries in each class by means of the twists

along XY and ST, so that the mapping class group in question is

generated by these two mappings. It is obvious that the twist along

ST results from that along XY by conJugation with a suitable

permutation of the boundaries.

In the case where the boundaries are immovable, the mapping

class group results from the above by adding the twists along the

four boundaries. These commute with each other and with any mappings

which leave the boundaries fixed, in particular with the twists along

ST, YZ and UV, and there is a simple relation between the four

boundary twists and the three twists along XY, ST and uv (see

§7g) .
 276

§4.

Mappings of the torus and the one-holed torus

a) The mapping class group of the torus R has long been

known from function theory. However, we shall use our method to

treat it in connection with other problems. We first give two

special mappings of R, namely the twists along two closed curves

which cross at one point, say a and b (see Fig. 7).

Fig. 7. Fig. 8.

We denote the twists by ~a and ~b' ~a replaces the connection

XY by XZ I Z2Y, and bb replaces the connection XQ by XSQ. Then

if L is the decomposition of R shown in Fig. 7, generated by

a and b, one easily sees by successive application of the trans-

-1
formations in question that ~a~b ba(E) is given up to homotopy

by Fig. 8. But this figure represents a rotation of the topo-

logically regular torus polyhedron of Fig. 7, with edges a and b

and a vertex X, about its vertex, and hence

 277

a mapping class of finite, in fact fourth, order. We obtain

Of course We

shall see later that representatives of all mapping classes of R

can be generated by ~a and ~b.

~a and ~b are also mappings of the bounded torus Rl ,

because the twists affect only the neighbourhoods of a and b.

The mapping class group for the closed torus R is endowed

with an orientation by the mapping of a and b. Namely, i f ~

and ~' both send the (oriented) a and b to a' and b',

then ~,-l~ carries the curves a and b onto themselves with

preservation of orientation. We have to prove that ~,-l~ belongs

to the class of the identity. First we deform ~,-l~ so that it becomes

the identity on a and b. This is possible because ~' -l\f?

maps the curves a and b onto themselves with preservation of

orientation. If we now cut R along a and b, the result is a

disc which ~,-14? maps onto itself with fixed boundary. But such

a mapping is homotopic to the identity with fixed boundary, and a

homotopy of the disc with fixed boundary is also a homotopy of R.

Thus ~,-l~ is homotopic to the identity mapping and therefore

belongs to the identity class. One observes that a homotopy of

the disc with non-fixed boundary in general does not correspond to

a transformation of R at all, when the four points on the boundary

of the disc which correspond to the intersection of a and bare

not sent to themselves.

 278

b) It follows similarly that the mapping class group of the

one-holed torus Rl with movable boundary is the same as that

for the closed torus. Because, if we make exactly the same

operations as in the above section, then ~,-l~ maps a two-holed

sphere with one fixed and one movable boundary onto itself. But

the mapping class group for this figure is also the identity.

Because, as we saw in §2, each mapping of the two-holed sphere is

deformable with fixed boundary so that the neighbourhood of one

boundary is unmoved while the neighbourhood of the other is twisted.

If the second boundary is now taken to be movable, then the mapping

may be deformed into the identity. Consequently ~,-l~ belongs

to the identity class.

§5.

The arithmetic field on the two- and three-holed

sphere and on the torus

a) The arithmetic fields on the surfaces result from systems

of curves on the surfaces.

Our general principle of representation is to represent homo-

topic curves identically. Also, we shall take the points of the

arithmetic field to be, not single open or closed curves, but whole

systems of such curves without double points or intersections.

Moreover, we shall omit from such systems all curves which are
 279

contractible to a point or to a proper part of the boundary. It

is necessary to observe these conditions in order to make the

representation at all simple. We recall for a moment the usual

representation of surface curves by elements of the fundamental

group in order to see where it is different : the elements of the

fundamental group correspond to single closed curves with arbitrary

singularities. But the curves are represented uniquely only when

a fixed point on the surface is given as their initial point, and

when they are assigned orientations ; curves for which these extra

specifications are not made correspond to all the elements which result

from a single element by conJugation. In our representation, the

curves or curve systems are represented by a fixed number of integers,

corresponding to the dimension of the arithmetic field. On the

other hand, the element of the fundamental group representing a

curve is given by an arbitrarily long sequence of integers, the

exponents of the different generators. Our simpler representation

makes the action of a self-mapping of the surface on these number

complexes much easier to follow, and brings it into line with known

arithmetic transformations.

b) Curve systems on the two-holed sphere

When the boundaries are movable, a system of lines connecting

the boundaries is determined by their number, say n, alone.

Because any two connecting lines are homotopic when the boundaries
 280

are movable. Any two systems of n disJoint connecting lines

are always homotopic. Also, one can pair one line of one system

with an arbitrary line in the other system. However, these

connecting lines do not exhaust the possibilities for non-contractible

curves. There are still the closed curves parallel to the

boundaries. These become represented when we now fix the boundaries

under the homotopy on each boundary we choose n fixed points,

and we also choose a normal line v connecting the boundaries and

a positive side of v. Then we consider the algebraic sum 0 of

the positive and negative crossings of v as we traverse the n

connections from the "first" boundary to the "second". By means

of a homotopy with fixed boundaries we can arrange that all

connections, since they are diSJoint by hypothesis, cross v in the

same sense. If, first, 101 is smaller than n, then there are

exactly 101 connections which cross v. They result from

connections which do not cross v by means of a single twist of a

boundary through one turn (see Fig. 9). Hence by §2b there

Fig. ~.
 281

can be no homotopy with fixed boundary which changes this number

o. However, if 0 = v+ ~n with Ivl < n, where v has the same

sign as ~, then the system of connections results from a single

twist of v connections followed by a ~-tuple twist of all n

connections. Since, by §2b, the number of windings of each

individual connection is not changed by homotopy with fixed boundary,

the number 0 is independent of such homotopies.

The number pair (n) determines a system of connections up

o
to. homotopy with fixed boundary. We now associate the symbol

with a system of 101 oriented curves parallel to the boundaries,

with the sign of 0 describing whether they cross the normal connection

v in the positive or negative sense ; then the correspondence

between curve systems and symbols is complete, and unique up

to homotopy with fixed boundaries.

The mappings of the two-holed sphere onto itself which do not

exchange boundaries are the twists. They transform the symbol

n
(by §2b) into (o+kn)' where k is a positive or negative

integer. The collection of transformations of these symbols, when

n 1 0, makes up a group isomorphic to the mapping class group of

the two-holed sphere. When n = 0 the twist transforms the symbol

into itself, so the group of transformations of this symbol is the

trivial group. In this representation only the closed curves have an

orientation independent of the symbols.

 282

c) Curve systems on the torus

We divide the torus into two two-holed spheres and L2

by means of two disJoint non-separating curves and (Fig. 10).

Fig. 10.

An arbitrary curve system on the torus, apart from curves parallel

to aI' consists of connections between n points on al with n

points on a 2 , in both L2 and Z' with

L any self-connections of

or in or L' being removed by homotopies. By homo-

2
topies with movable boundary, say in L2 , these n points can be

carried to n given points on each of the boundaries, and likewise

the connections in L2 to given connections. Even then, the

association of the n fixed points and the n fixed connections

to the n given connections is determined only up to cyclic inter-

change. This cyclic interchange corresponds to a homotopy of the torus.

 283

If the connections in L2 are now fixed then L' is transformable

2

only by homotopies with fixed boundaries. The n connections

of the n fixed points on the two boundaries in L' are therefore,

2
n
by b) , represented by a number pair (0) , where 0 is the algebraic

sum of the number of crossings, in the positive or negative sense,

over a normal connecting al and a 2 , if we always take the curve

system inside L' to run from to We can take such a

2

normal to be, say, the piece in L' of a curve bl which cuts

2

once (see Fig. 10) and, when one side of bl is chosen to be positive,

a crossing of the normal from the negative to the positive side will

be reckoned positive. By means of a homotopy, it can be arranged

that all the crossings have the same sign, so that 101 represents

the number of crossings. We also take the piece of b2 in L2

parallel to the fixed connections in L 2 , so that the curves of

the system have exactly 101 points of intersection with bl .

By means of homotopies we can also arrange that the crossings of

curves in the system with a curve b2 parallel to bl all have

the same sign. Then the curve system can also be regarded as a

system of connections of 101 points on bl with 101 points on

in the two two-holed spheres L' , and L I ' , which result

2 2
when the torus is cut along bl and b2 . Finally, we shall

associate the symbol (0) with a system of curves parallel

o
to and We then have a number pair for each curve

system on the torus, with n ~ ° and arbitrary, positive,

negative or zero. With the exception of curves parallel to aI'

 284

we have given the curves ~ orientation, and this is also expressed

by our symbol (n is only a number) . This has the consequence

that a mapping of the torus onto itself is first determined by the

images of three curves, since e.g. al and bl can be mapped

onto themselves with reversed orientation. On surfaces of higher

genus the problem has not previously been solved, since the

orientation has not been represented arithmetically. However, on

the torus, the introduction of an orientation for curves can be

made without difficulty : on the torus, all curves of a system are

homotopic; they are parallel curves, as follows immediately from

topological considerations. Namely, cutting along a curve c

yields an L2 , and all simple, closed, non-contractible points on

the L2 are homotopic to the boundary curve c. We give the

curves of the system an orientation so that homotopic transformation

of one curve in the system to another preserves orientation. If

one comes from to by traversing a piece of the curve

system in L 2, then the same is true for any piece of the system

in Li' since we have assumed that each self-connection of al or

or L' has been removed by a homotopy. It is the

2

same for the pieces of all other curves in the system. In this

case we give n the positive sign. If, on the other hand, one

comes from to on traversing a curve piece in Li' then

n shall have the negative sign. In addition - and anomalously

with the above determination - 0 shall have the positive sign when
 285

the curve piece in a particular one of the pieces into which the

torus is divided by bl and b2 goes from bl to b2. If, in

Fig. 10, we take this part as the back side of the surface, then

the curve represented in the figure has the symbol If we

reverse the orientation, we obtain the symbol

curve. In general, reversal of orientation changes the symbol

to Finally, we shall associate the symbol

a system of 101 curves parallel to a l , with the sign of

corresponding to the orientation.

d) Now we consider the effect of the mappings b and bb

a
(§4) , the twists along al and bl in L' and Lit' respectively,
2 2
on the symbol (n) . As we have said in §5c) and §2 about these
0
mappings and the mappings of a two-holed sphere, ba changes the

number 0 in 0 + n, and bb changes n into n + o. Thus we can

use ba and bb' assuming n > 0, to effect the operation of the

euclidean algorithm on the symbol, and hence by a suitable sequence

±l
of applications of b and reduce to the symbol
a
1
( (n,o))
o ' h' h
w ~c represents (n,o) curves parallel to bl with

positive orientation, by the above prescription. As we have seen

in §4, we can also use ba and bb to reduce the system to (n,o)

curves parallel to bl with negative orientation, or (n,o) curves

parallel to al with positive or negative orientation. This

lWhere (n,c) denotes the greatest common divisor of nand c.

 286

corresponds to the arithmetic fact that we can reduce to

(±(n,o» as well as o
(±(n,o» by our arithmetic operations.
o

If the original curve system consists of a single curve, then

(n,o) = 1. If we have two systems which each consist of a single

curve, then we can reduce their symbols, by I!.a and I!.b or the

euclidean algorithm, simultaneously to (1)

0
and
(::) (1)
0
is

a curve parallel to b l , and hence it cuts the second curve in

1021 points. If the two given curves have only one point of

intersection, then O2 = ±l, and by further application of we

transform the symbols for the two systems into and

Now each mapping sends a certain oriented curve pair to the one

represented by (~) (~), and it sends the same pair, in which the
second curve has the opposite orientation, to the pair represented

By §4a) , each mapping which simultaneously sends

1 0
(0)' (-1) to themselves belongs to the class of

the identity. Thus two mappings which send the same curves to

respectively belong to the same class.

The transformations of the arithmetic field uniquely determine the

mapping classes. On the other hand, when there is a mapping which

1 0
carries a curve pair to (0)' (1)' then none of the mappings
1 0 .
considered here will carry this curve pair to (0)' ( -1)' otherw~se

one of the mappings would reverse orientation. We can therefore

transform each curve pair which is sent to (~), (~) by a mapping,

into Thus each mapping class

contains an element generated by I!.a and I!.b

the mapping class group of the torus.

 287

In §4 we have given two relations for ~a and ~b' The

following consideration gives us one more the mapping

-1 -1
~a ~b~a changes the curve a l , with a certain orientation, into

bl with a certain orientation, hence the twist ~a along al is

transformed by the operation into a twist along bl ,

Accordingly,

-1
~
a

is a relation, easily checked by the representation of ~a and

~b as linear transformations, If we set

-2 -1
~a E T

with the solution:

in these relations then we obtain

On the other hand, it follows from the relation

obtained in (§4) that

 288

Thus we have : the mapping class group is generated by two operations

Land T which satisfy the relations

1.

As is well-known, the group with these two relations between its

generators is the homogeneous modular group, and hence the latter

is the mapping class group of the torus. In fact, one easily

obtains a normal form for the elements on the basis of the two

relations, and then one proves, with the help of uniqueness of the

continued fraction expansion, that the two relations are sufficient

for the definition of the mapping class group, and hence this group

is identical with the homogeneous modular group.

e) Curve systems on the three-holed sphere

We first consider only the unclosed curves. These connect

either two different boundaries or else two different points on the

same boundary. In the latter case we call it a self-connection

of the boundary in question. Since we consider no curves which

are contractible on a segment of the boundary, a curve connecting

two points on the same boundary must separate the other two boun-

daries from each other, and the latter must therefore have no

connection with each other. If are the numbers of

endpoints of curves of the system on the three boundaries, then

Further, if vik = vki is the number of

connections between the ith and kth boundary, then

 289

n. and
~

if there are no self-connections, or, what is the same thing, if

n i + nk is always greater than n~. If, however, say

then

v ..
~~

is the number of self-connections of the ith boundary. Moreover,

v . n~
~~

in this case. Thus in each case the number of connections from

boundary to boundary is uniquely determined by n l ,n 2 ,n 3 and

the whole system is determined up to homotopy, provided the boundaries

are movable. I.e. for given values n. we can convert each system
~

into one we call a normal system, by homotopies with movable boundaries.

This is what we proved in §2c) for connections between different

boundaries. However, it also follows easily for self-connections.

Suppose that v ll is such a self-connection, say of the first

boundary. Then there are disJoint connections and of the

first boundary with the second and third which do not cut v ll ' and

a connection of the second boundary with the third which

cuts once. We similarly associate with a second self-connection

vi three connections vi2' vi3 and vi3 with the same properties.
 290

However, when we carry v 12 ' vl3 and v 23 simultaneously to

vi2' vi3 and vi3 by a homotopy as in §2c) , then v ll goes to

a v ll which does not cut vi2 and v\3 and which cuts vi3
once. Thus we can convert vII' and hence also vII' into v·
1
by a homotopy. However, i f v ll is fixed, then v 12 and vl3
are convertible to vi2 and vi3 by homotopies which move the

second and third boundaries.

In contrast to the situation on the two-holed sphere or the

closed torus, the association of the given connections with the

fixed connections can never be cyclically altered. Because in a

series of disJoint connections of, say, the first boundary with the

second (see Fig. 11) there are two distinguished connections

Fig-. JI.

and which cut the into two parts - an on

which the other connections lie, and an L2 . No other pair of

connections has this property of excluding all other connections from

the resulting L2 . Also, v 12 ,1 cannot be exchanged with v 12 ,m'

because this would reverse the orientation of the surface.

Similarly, there are two distinguished, non-exchangeable connections

 291

and Vll,t among the self-connections of a boundary (see Fig. 11).

All curve systems which contain no closed curves are represented

by the number triple n l ,n 2 ,n 3 relative to the three boundaries.

The systems with closed curves are obtained when we again fix the

boundaries. To investigate this we use the same methods as for

the two-holed sphere: we connect the three boundaries r l ,r 2 ,r 3 to

each other by disJoint normal connections v 12 ' v 23 ' v 31 and cut

the by curves r.l. ' and r ' parallel to r. and in

k l.

six points altogether (see Figs. 12, 13). The normal connections

and the boundary curves form two hexagons 81 and 8 2 , and similarly

the pieces of the normal connections and the parallel curves form

hexagons 8 ' and 8 ' which are parts of and respectively.

1 2
We now determine n. fixed points on r. and ri', and in fact,
l. l.

if

(in Fig. 12, i = 1, k = 2, t = 3) we place all points on r. and r. '

l. l.

in the hexagon 81 , However, if

(Fig. 13) then we put

points on r. and r.' in

l. l.
 292

points in 81 . Now in the first case we connect the ni points on

r.' with
l.

points on r ' and

k

points on r ' by the disJoint connections in 8 1 ',

t

(1S3)
1-10
~+-H(711)
10'101

Fil!. IJ.
ri::=:o 12.

in the second case we connect

n.-n -n
l. k Q,
2

suitable points on ri', which lie in 8 1 ', with the

 293

points on r. , which lie in 82

, by the disJoint lines which do not
~

cut v ik and vh and which cut Vkt only once, and the other

nk points on r. , with the nk points on

,
r k , nt points on
~

r.
, with the nt points on r t ' , by the lines in 81
, which do
~

not cut themselves and the In such a system of connections,

each system of connections between two boundaries, to the extent

that it lies in 81
, and 8 2 ', can be homotoped with fixed boundaries

r. but movable r.
, Only the connections in the three L2 ,i
~ ~

between the r. and r.' now remain. We treat these exactly as

~ ~

above and determine 0i as the algebraic sum of the number of

crossings of v ik in the positive and negative sense, where the

positive side of v ik is taken to be the one in 8 1 , say. The pro-

longation of connections from r. to r.' is fixed. This determines

(:1 :2 :3).
~ ~

a symbol We shall show that it is unchanged by

123
homotopies with fixed boundary.

If no self-connections occur, then we can apply exactly the same

considerations as we did above with the two-holed sphere: if, say,

°1 = ~l + ~lnl' this means that the system of connecting lines of

the ("undisturbed") normal system, which consists purely of v 12

disJoint connections, is made up of connections which circle

rl once before being accompanied by the rest of the system in ~l

circuits of in the same sense. This property, and with it

cannot be altered by homotopies, as we know from the investigation

of connections between different boundaries on the three-holed sphere

(§2c)) • However, we must still consider the case of self-connections,

 294

which was not investigated in §2c). We first remark that, as we saw

above, all self-connections of a boundary can be carried into

each other by homotopies with movable but not exchangeable boundaries,

however, the endpoints cannot be associated arbitrarily. Because

(see Fig. 14) the self-connection AB, which runs from A to B,

determines two different orientations for the two L2 's into which

i t cuts the L3 . Since both L2 's contain non-exchangeable

boundaries, the two L2 'S cannot be exchanged. Thus if A'B' is

another self-connection of the same boundary, then, assuming A'B'

gives the two L2 's the same orientations as AB, A'B' can be

carried to AB only by carrying A' to A and B' to B. When

A' goes to Band B' goes to A, then the orientations given by

AB and A'B' must be different.

Moreover, the connections between, say, the first and second

boundaries with the same endpoints can be converted into each other

by mappings with fixed.boundaries. The self-connections of the

first boundary, on the other hand, divide into two classes (see Fig. 15).

Fig. J~. Fig, I;.

 295

The connection AXCYB =v cannot be converted into the connection

AX'CY'B _ v' with the first boundary fixed, because the parts

ADB = r ll and AEB = r 12 of the first boundary remain fixed. The

closed curve made up of v and r ll is however homologous to r2

on the L3 , whereas the closed curve made up of v' and is

homologous to r 3 , hence no mapping with fixed boundary carries

v into v'. If we again argue as in §2b) , then we realise that

all connections between A and B (other than those homotopic to

a piece of the boundary, which are excluded), result from v or v'

by circling the boundary rl .

In the case that n l > n 2 + n3' suppose first that 1°11 < n l .

For the sake of convenience we shall replace the normal connection

v 3l by v 3l ', where all fixed points on r l , and the corresponding

points on r ' lie in one of the hexagons or S2' say Sl' made
1

as well as the r, (see Fig. 16). The

~

number °1 , since it is defined by crossings of v 12 , is not altered

by this. Now there are again exactly 01 connecting segments in

the L2 ,1 between and r ' which circle around once

1

relative to and As long as these segments belong to

connections with other boundaries, the circling, as we already know,

is not altered by homotopy. However, when one of the two segments

belonging to a self-connection circles r l , this self-connection

cannot be converted into a self-connection without circling segments

by mapping with fixed boundaries. Because (see Fig. 16) the

 296

undisturbed self-connection GM, together with the part of rl

belonging to 81 is homologous to r 3 , but the disturbed self-

connection G'M', together with the part of rl in 81 is

homologous to r2. However, when both segments in L2 ,l belonging

to a self-connection are replaced by circling connections, then

the disturbed self-connection results from an undisturbed self-

connection by a twist in L2 ,l (see Fig. 17). Here the word

"undisturbed" means that the connection in question has the same

M
Fig. 16. Fi:.!. 17.

endpoints on r l and r l ', but it does not cut the segments v 12

(and v 31 ') in (in Figure 17 the segments in which

convert M'G' into an undisturbed connection are shown dotted).

Altogether, we see that the number 01 of crossings of v12 is

also not changed in this case by a homotopy with fixed boundary,

because it represents a distinction between the given connections

of the first boundary with other boundaries or itself and the

undisturbed connections which cannot be changed by homotopy. And

 297

in fact there is a different distinction for positive and negative

0.

Finally, i f l0ll > n l , say \1 1 has

the same sign as 01' then we have to add \1 1 twists of the whole

system of connections in L2 ,1' with the same sign as 01' to the

alteration given by vl . Homotopy with fixed boundary does not

change Thus each system of connections uniquely yields the

symbol
n n)
02 0 3 independently of homotopy with fixed
2 3
boundaries.

.th
When n. = 0 (so that there is no connection with the ~
~

boundary) we associate the number o. ~

with curves parallel

to r i , with orientation corresponding to the sign of o..

~
Then

each such symbol is also associated uniquely, up to homotopy, with

a curve system connecting the n fixed points on the three boun-

daries, and conversely, each system of closed or non-closed curves

on the three-holed sphere is represented by such a symbol.

Incidentally, each such curve system divides into at most three

classes of curves which are mutually homotopic with movable boundaries.

The numbers of members of these classes are, for the different

cases,

or
n.-n -n
~ k R,
and or finally 0. (for n. 0) •
2 ~ ~

The mappings of L3 onto itself are known to us; they result

from twists along the three boundaries. They transform our symbol
 298

into

where the ki are integers, positive, negative or zero.

In conclusion, we remark that here we can give the n. no signs,

l

unlike the case of the torus.

§6.

The arithmetic field or the curve systems on the one-holed torus

a) We divide the one-holed torus as we did the closed

torus Rl (see §5c)), by two disjoint closed non-separating curves

and a 2 , into a two-holed sphere L2 ,a and a three-holed sphere

(see Fig. 18). A curve system in has say n points of

Fig. 18.

intersection with aI' as well as a 2 , and 2q endpoints on the

boundary r. By means of homotopies with movable boundaries on

 299

L3 ,a' we can convert the curve system in L3 ,a into one with n

fixed points on each of al and a2 and 2q fixed points on the

boundary r, and fixed connections. The latter may be curves

which do not cut the curves bl and b2 or, in the case of seH-

connections of r in L 3 ,a' they may have one point in cornmon

with each of bl and b2 · In fact, this association of the given

connections with the fixed ones is unique, as we saw for the three-

holed sphere in §5e). Self-connections of or in

cannot appear, since n < n+2q. However, there certainly can be

self-connections of the boundary r, if q > n.

There still remain the connections in L 2 ,a between the n

points on al and the n points on a2 · These are determined

by a positive or negative number <5 (see §5c» . As with the torus,

we now give curves on Rl which cut al and a2 equal orientations,

and thereby give n a sign. However, not all curves are homotopic

here. We must therefore arrive at agreement of sign somewhat

differently so that all curves which cut al cross it in the same

sense. We can arrange this, because a curve which crosses more

than once (apart from points removable by homotopies) always crosses

it in the same sense, by circling around the curve bl . (It is

here that the determination of sign for higher genus fails.) This

determination gives n a sign, positive for positive crossing and

negative for negative crossing. At the same time, the determination

gives all crossings of bl the same sense. For those of the latter
 300

which do not cut ai' namely the self-connections of the boundary

r in L3 ,a' we determine the orientation so that the crossing of

bl is in the same sense for all curves. In this way, is

n
definable analogously as for the closed torus. Thus (0 q), where

n and are arbitrary integers, and q is a natural number,

gives us an arbitrary curve system on Rl , when we disregard curves

parallel to the boundary. The only exception is the curve system

represented by the symbol In this case the orientation of

the q - Inl self-connections of the boundary is arbitrary.

In order to obtain the curves parallel to r as well, we must

consider homotopies with fixed boundary (just as with L3 , §5e))

and introduce a rotation number for the boundary. This proceeds

exactly as above, however we do not need it later and we shall

therefore leave these curves out of consideration. Later in this

paragraph, though, we shall give a theorem on the mappings with

fixed boundary.

The arithmetic field we have introduced for Rl is three-

dimensional and its points are all oriented curve systems on Rl

minus curves parallel to r.

b) Transformation of the arithmetic field or the symbols

by mappings of the one-holed torus onto itself.

The mappings are the same as for the torus (see §4b)) the
 301

mapping classes are generated by the twists, 6a along a curve a

parallel to a l , and 6b along a curve b which cuts a once.

These mappings induce the operations of the euclidean algorithm in

the symbols (n) of the torus R. How do these mappings act on

o
the symbols (~q)? The action of 6
a
is the same as on the
n
symbol (0)' because it affects only the two-holed sphere L2 ,a.
n n
Thus 6 a ((0 q)) = (o+n q), Next we shall consider not the action

of 6b , but that of This transformation carries a to

b- l and b to a (see §4a)). Now Inl is always the irreducible

number of intersections of the curve system with a, but 0 is

only the irreducible number of intersections of the system with b

in case Inl > q, i.e. when there are no self-connections of the

boundary in L3 ,a If we now assume that q is not greater than Inl

and 101, then 101 is the number of intersections of the system

-1
with b, and hence under the transformation 6a~ 6a it is the

number of intersections with a, and hence equal to In' I, where

Inl is the number of intersections of the

system with a, hence under the mapping it becomes the number of

intersections with b, and hence equal to I 0' I, because

In'l = 101 > q by hypothesis. The signs of 0 and n are easy

-1
to determine when one recalls that 6a ~ 6a can be regarded as a

rotation of- the system a,b. It follows from this, as with the

closed torus, that n' -0, 0' = n. By continued application of

-1
the operations 6a ' and the operation
 302

n'
convert the symbol into (0' q), where 10'1 is now < q.

As long as q is not greater than Inl or 101, the euclidean

algorithm holds. Thus we now consider the case In I > q > I 0 I .

If again, then In'l 101 still. Inl

is the number of intersections the transformed system has with b,

but since In'l < q now, this number is also equal to

10' I + q - In' I, because q - In'l is the number of self-connections

of the boundary in L 3 ,a in the transformed system. Thus

10' I = Inl + 101 - q. The sign is also not difficult to determine,

and it turns out that n' has the opposite sign to 0, and 0' has

the sign of n. However, Inl becomes <q by these new operations,

-1
and with a suitable number of operations t:.a or t:.a we also get

101 < Inl < q, so that we now have the case q > Inl > 101. We
-1 n n'
again denote t:.at:.b t:.a«o q)) by (0' q). The number of intersections

with b is now 101 + q - Inl, thus In'l = 101 + q Inl. The

number of self-connections of the boundary in L 3 ,a for the trans-

formed system is q - In' I, hence equal to Inl - 101. Thus

The signs are as above.

It is now easy to see that there are only 12 different curve

systems which satisfy the condition q > Inl > 101 and which are

transformable into each other. Because in this case we have (see Fig. 19)

Inl - 101 connections of two points on r which cut a once and

b not at all and which are homotopic to each other, 101 connections

which cut a and b once and which are homotopic, and finally
 303

q - Inl connections which cut only b, one time (the self-

connections of r in L 3 ,a)' and which are also homotopic to each other.

Fig. It).

The curve systems therefore divide into 3 divisions and, when

is a symbol transformable into which satisfies the same

inequalities, In'l - 10' I and 10' I must each be equal to one of

the 3 numbers for the 3 divisions. But choice of the two divisions

and the evaluation of In'l - 10' I and 10' I determines In'l and

10' I. Thus we have altogether 6 different possibilities for the

pair In' I, 10' I, and since double application of the mapping

-1
~a~b 6a simultaneously changes the signs of nand 0, we have the

12 different possibilities claimed. At the same time, we see that

these can be reached by application of and ~a' so that we

have the result that ~ and

a
 304

the perforated torus. This is only a confirmation of the earlier

result, since the generation from 6 and ~ immediately yields

a
the generation by 11 and At the same time, it follows
a
that each curve system on the one-holed torus divides into 3

divisions, whose numbers can be determined by our recursion process

(see the example).

An overview of the generation of the 12 different possibilities

for nand 0, when q > Inl > 101, is still lacking. We shall

deal with this simultaneously with a general geometric representation

of the arithmetic field and the transformations of it induced by mappings.

However, we confine ourselves to the representation of transformations

of Inl and 101. We construct an ordinary square lattice for

these numbers (see Figs. 20 and 21) and denote the square with

[. Ax; f ,.'

Fi~. ~O. Fig. 21.

 305

vertices (0,0), (q,O), (q,q), (O,q) by OQRS. In the sector

q ~ Inl, q ~ 101 the ordinary operations of the euclidean algo-

rithm,

( Inl) + (101) and ( Inl) (Inl)

101 Inl 101 + lo±nl

then represent such transformations. By means of these operations,

one gets from any point of the sector into the right-angled half strip

In turn, one can get from the latter, by means of

the operation,

( Inl) ( 101)
101 + Inl-q+lol

into the oblique-angled half-strip q > Inl < 101. Finally, when

n f 0, one can get from the latter into the triangle OQR,

q > Inl > 101 by suitable repetition of the operation ( Inl) ( Inl )
101 + 10Hni .

This triangle is subdivided into 6 triangles by the medians. By

suitable repetition of the two operations

( In I) ( I 0 I+q-I n I )
( Inl) ( Inl )
101 + 101 and 101 + Ini-iol '

a lattice point in one of the 6 triangles can be carried to any other.

These 6 triangles correspond to the 6 different permutations of the

3 divisions of curves, or to the 6 different distributions of divisions

among the divisions of curves which cut a. and b i , only

~
 306

only b .. On the medians the two division numbers are equal. On

l.

the sides of the big triangle a division number equals 0, hence

at the vertices two division numbers equal O. As we saw, each

lattice point can be brought into one of the 6 triangles by an

operation, provided it neither lies on (0,151) nor is brought

there by an operation. But (0,151) in the oblique-angled strip

results from (Inl,O) in the right-angled strip Inl > q > 151.

This lattice point results from the point with coordinates Inl and

(n,5) > q. All lattice points for which the greatest

common divisor of nand 5 is smaller than q can therefore be

sent into one of the 6 sUbtriangles by our operations. All curve

systems for which (n,5) > q are transformable into curve systems

consisting of (n,5)-q closed curves parallel to b. and q self-

l.

connections of the boundary in L 3 ,b, because such a system is

n
represented by (0 q) for Inl > q. Thus a subtriangle, together

with the half-line 5 0, Inl > q, represents a fundamental domain

on the arithmetic field or for the curve systems under transformations

with movable boundary. The fundamental domain for the closed torus

is given by the half line = 0, because one can reduce each

to «n,5)). We also remark that two different lattice points in

a
the same subtriangle give two different triples of division numbers.

Examples: 1) (25 7), Inl and 151 > q.

11

One obtains Le. by by

 307

-2
~a Now we are in the region n > q > 0 and we obtain

101 , Le. by by ~
-2
Inl-q+lol a

This brings us into a subtriangle and hence we obtain the division

numbers: 101 = 1, Inl-Iol = 2, q-Inl = 4. We confirm this by

25
direct consideration of the curve system (11 7) : we number the

points on al and from 1 to 25. The first 7 points on

al and a2 have connecting lines in L3 ,a with r, the rest are

connected in L3 ,a to a2 and al respectively. The passage

from al to a2 in L2 ,a occurs by addition of 11 modulo 25, the

passage from al to a2 in L3 ,a by connection of like-numbered

points. We go from r to the point 1 on and then obtain

the series:

1 (on a l ), 12, 23, 9, 20, 6 on a2 and, since this point is

connected to r in L 3 ,a ' 1 on is connected to r by a 4-tuple

circling of b, 2-tuple circling of a.

Leaving 2 on r, we obtain a parallel curve: 2, 13, 24, 10,

21, 7 ; but leaving 3 on r yields the series 3, 14, 25, 11, 22,

8, 19, 5, and thus a 6-tuple circling of b, 3-tuple circling of a;

leaving 4, 5, 6, 7 on r, we obtain the four similarly connected

series:

4, 15, 1

5, 16, 2

6, 17, 3

7, 18, 4, each of which is a l-tuple circling of b and a

 308

I-tuple circling of a. Thus in fact we obtain 3 divisions with

2, 1, 4 elements respectively, as we have already found by our

arithmetic process above. We also remark that the fact, resulting

here from topological considerations, that the direct process always

leads to only 3 divisions with series of different length, is a

purely arithmetic theorem. The purely arithmetic proof could

scarcely be carried out otherwise than by proving that the division

numbers are preserved by our "process".

2) (23 7) .. (11 7) .. (11 7) .. (1 7) .. (1 7)

11 23 1 5 O·

Thus we have only 2 divisions, with one member and 6 members

respectively. To check, we construct the series

1, 12, 23, 11, 22, 10, 21, 9, 20, 8, 19, 7

2, 13, 1

3, 14, 2

4, 15, 3

5, 16, 4

6, 17, 5

7, 18, 6,

which in fact divide into 2 different divisions with one member and

6 members respectively.

Finally, we remark that a common divisor of n, 0 and q

is not lost by the process and it also appears in the division

pn
numbers. Consequently, the symbol (po pq) represents the curves

with the symbol (~q), each provided with p-l parallel curves.
 309

When closed curves which are not parallel to r appear, then there

can be only 2 because i f we cut Rl along such a curve, say c,

1
the result is an L3 ,c· Curves on this L3 ,c which do not meet

two boundaries and which are not parallel to the third boundary

are either parallel to these boundaries, and hence homotopic to c

on Rl , or else self-connections of the third boundary. The symbols

corresponding to such curve systems are always of the kind for which

either n = 0 or (n,o) > q.

When no closed curves appear, then all curves are bounded by

2q points on r. They divide, as we know, into three pairs of

divisions. We see from Fig. 19 that each pair of corresponding

points in one division separates each pair of corresponding points

from another division on r. The pairs from the same division do

not separate each other. This corresponds precisely to the result l

for polyhedra with p = 1 and ~ vertex, with no interior vertices,

in an earlier work. The one vertex corresponds to the boundary here.

This result can be derived quite easily without use of the symbols

for the curve system, i.e. without passing through the arithmetic

field.

c} One holed torus Rl with fixed boundary

1
Rl T4 -1 4
On the mapping = ( ll a ll b lI a ) with movable boundary belongs
1
to the class of the identity. It is easy to see that with fixed

lsee Acta math. 67, p. 165 ff.

 310

boundary T4 represents a twist along the boundary. In fact a

mapping in the class of T transforms the connecting line A1BA2

(see Figs. 22 and 23) into A1CA 2 , a mapping of the class T2 trans-

forms A1BA2 into A1D1E2BE1D2A2' and finally a mapping of the

.B

Fig. __ .

class T4 transforms A1BA2 into A1D1E1BE2D2A2' which results from

by the twist lor along the boundary. Thus T4 6- l carries

r
the curves a and b, as well as the connecting line A1 BA 2 , into

themselves. If we cut along for

which the only mappings are twists along the boundary, i.e. twists

along b. But these twists do not map a onto itself, hence

T46- l belongs to the identity class of mappings of Rl onto itself.

r 1

Thus T4 belongs to the class of 6r and represents a twist along

the boundary.
 311

The mapping cl~ss group for with fixed boundary is

generated by the mappings and T with the relation z3 T2 1

(see §5d) . This relation holds with a fixed boundary, because

z3 T2 = expresses the fact that the twist
1
~ goes to the twist
lI- l -1
by a mapping (namely lIallb lI a )· This must also hold with a
a
fixed boundary, because a twist always goes to a twist under a

mapping, with fixed or movable boundary. Thus lib cannot go

to the twist lI- l connected with a twist along the boundary. On

a
the other hand, the second relation T4 I does not hold, because

and are not homotopic with fixed boundary, as

one sees immediately by consideration of the universal covering

surface of It then follows from the normal form of elements

generated by Z and T on the basis of the relation z3 T2 = I

that this relation alone defines the mapping class group of

and hence this group is identical to the group of the trefoil knot.

§7.

Arithmetic field and curve systems on the four-holed sphere

a) Systems of closed curves on the L4 (Fig. 24)

Fig. :q.
 312

We divide the L4 by two parallel curves al and a2 into

two three-holed spheres L and L and a two-holed sphere

3,a l 3,a 2
L2 ,a Since the curves of the system are to be closed, all curves

on the three-holed spheres which are not parallel to the boundaries

or to and are self-connections of the boundaries and

a2 respectively. We join the four boundaries of the L4 with normal

connections and cut each of and

a2 at one point. The curves of the system have, say, 2n points

in common with each of al and a2· We then place n of the

points on al and a2 on one of the two parts between V23 and V4l

and fix connections in L and L between any two such points,

3,a l 3,a 2
cutting V12 resp. V34 once and V23 and V4l not at all. Each

system of closed curves on L4 which includes no curves parallel to

the boundaries or and is homotopically transformable to

these connections inside and with movable boundaries.

Only the connections in the L2 ,a remain, and these can be determined

by a number 6 as before. Thus the curves of the system are given

n
by the symbol (6)' where is an integer, positive, negative or

zero, but n is a natural number. The symbol is associated

with curves parallel to a l , oriented according to the sign of

6. It is not possible to give an orientation to general curves

without something extra; we come back to this later. The symbol

(~) will represent all systems of closed curves on the L4 which

contain no curves parallel to the boundaries. And in fact nand 0

 313

are independent of homotopies with movable (but not exchangeable)

boundaries. One can prove this by direct topological considerations,

though perhaps more intricately. It is more convenient to again use

the representation by the fundamental group of L4 , which is the

free group generated by 3 of the operations which

represent circuits around the 4 boundary curves. In fact it is

easy to obtain the representation of the curve system in the funda-

mental group from our symbol. However, we shall first simplify this

problem by supposing that our curve system consists of a single curve.

We shall see later that, just as on the torus, the symbol

represents a single curve if and only if (n,a) = 1, and that the

general curve system consists ot (n,a) parallel curves. Thus,

n
for our proof of the claimed homotopy invariancE' of (a)' i t suffices

to consider the case (n,a) = 1. A] so, we first assume 0 < a < n.

It then follows by consideration of the path in the dissected surface

5
(cf. p. 275 above and Fig. 25, where (2))' that the curve

Fi~. ~;.
 314

is represented by a 2n-termed expression in the generators

±l ±l ±l
r3 , r 4 The generator r3 appears times in this expression,
±l ±l
the generator r4 appears n-o times, and rl appears n times.

Since a conjugation makes only a cyclic permutation of the terms in the

-1
expression (except that successive terms ee which occur under

cyclic permutation may be omitted), because of the freeness of the

generators, nand 0 are uniquely determined by this expression

or any of its conjugates, i.e. in this case nand remain

unchanged under homotopies. However, if 101 < n but 0 > 0,

then the same consideration applies when we introduce the generator

r2 in place of r3 and represent the fundamental group as the

free group on the generators r l , r 2, r 3.

The case 101 > n now remains. We reduce this to the preceding

by a "coordinate transformation", which is basic to our understanding

of the L4 : namely, instead of determining the curve system by

the n points of intersection on the parallels and and the

measure of twist 0, we determine it by the n' points of inter-

section with the curves bl and b2 (see Fig. 24.) and the measure

of twist 0' in the between and We place

and b2 so that all self-connections of the a. lie in the L2 ,b

1

Then we have exactly 101 self-connections of the b. in the L3 ,b ..

1
1
Thus n' = 101 , and all these self-connections lie in the L2 ,a
By the reverse argument we obtain n = 10' I. Finally, we give 0'

the sign of by a suitable determination of the positive side of al .

 315

The coordinate transformation then reads: n' = 101, 0' = lfln.

The contrast with the analogous transformation on the torus Rl

is clear : on the L4 we are dealing only with a coordinate trans-

formation, whereas on Rl this can also be regarded as a mapping.

This is out of the question for the L4 under the assumption of

non-exchangeable boundaries, because the b± separate pairs of

boundaries different from those separated by the a .. When

1.

boundaries can be exchanged, the coordinate transformation corres-

ponds to a quarter turn. The case 101 > n has now been reduced

to the case 101 < n.

Only the case n = 101 =1 now remains. A symbol with other

n or 0 cannot go to under homotopic transformation, because

we have proved the invariance of nand 101 for the other symbols.

But also cannot go to Because if we transform both

curves by a twist in L 2 ,a' then they go to curves with the symbols

n
and (-0+2n)' which, since the numbers 10 +2nl and l-o+2nl

are different for 0 F 0, represent two non-homotopic curves. Thus

(~) also does not go to (_~) by a homotopy, and we have now

proved the invariance of (n)

I)
under homotopy in general. Here it

is equally valid to consider the boundaries as fixed or movable,

since we are dealing purely with closed curves.

b) Mappings of the four-holed sphere onto itself and the action

on the arithmetic field. In §3 we have considered special mappings

of onto itself, the twists of and It follows

 316

from the early results and the above development that the first

twist carries n to n, 101 to 10+2knl. and the second carries

n to In+2k'ol, 101 to 101. Thus we again have (at first only

for the absolute values) a euclidean algorithm, but of such a

kind that we can only add or subtract an ~ multiple of one number

to or from the other. These operations constitute a (non-invariant)

subgroup of the operations of the general euclidean algorithm. We

shall call them even-euclidean operations. If 101 > n > 0, then

we can determine k so that 10+2knl ~ n, but if n > 101 > 0 we

can determine k' so that In+2k'ol < 101 . The process breaks off

in the following three cases : 1) n = 0, 2) 0= 0, 3) n = 101·

In the first case we obtain 101 curves parallel to a l , in the second

n curves parallel to b l , in the third case we have in.

However, the symbol ( n) goes to (n) by addition of 2n to -no

-n n
That is n curves parallel to a certain curve, say c, separating

and from and (see Fig. 24). Thus we have the

Result : each system of disjoint simple closed curves on the four-holed

sphere can be mapped, by means of the twists along a and b,

onto a number of curves parallel to one of three curves which separate

two of the boundaries from the others. This number is equal to

(n,o) when the curve system is represented by Here, curves

parallel to the boundaries are ignored. In this way we have derived

the fact used above, that when (n,o) = 1 the curve system consists

of only one curve. In doing so we have not used the theorem on the
 317

homotopy invariance of the symbol.

Now a mapping which sends, say, and to themselves

can be generated on and L by homotopy (with movable

3,a 2
boundaries) . In the it generates a twist, which corresponds

to the transformation A mapping ~ of the L4

which sends two curves and to and can then be

transformed by a suitable ~, composed of twists in L2 ,a and

to send and to curves parallel to or c. But

they must be curves parallel to aI' because al and ell separate

the same pairs of boundaries by hypothesis. Thus E~(ell) = aI'

EI~~-l(al) = a l · Consequently, ~~-l equals a twist in L 2 ,a' up

to homotopy. Thus, representatives of each mapping class of the

four-holed sphere can be generated by twists in and L2 ,b' which

correspond to the even euclidean operations. However, if an even

euclidean operation is not the identity transformation, i.e. if it

does not carryall symbols to themselves, then it represents a non-

homotopy transformation, since, as we have seen, homotopies carryall

symbols to themselves. On the other hand, a non-homotopy transfor-

mation sends at least one of the symbols or to another

symbol, and therefore corresponds to a non-identity transformation

in the group of even euclidean operations. We therefore have:

the even euclidean transformations generate a group isomorphic to the

mapping class group of the four-holed sphere. l

lW. MAGNUS informs me that it follows from this, by means of the

result of H. FRASCH, Math. Ann. 108, p. 245, that the mapping class
group for the L4 is the free group on two generators.
 318

c) It is easy to see which symbols represent curve systems

which can be mapped onto a particular one of the three systems of

parallel curves. Because, the even euclidean operations preserve

the parity of nand 5. We can assume that nand are not

both even, otherwise the greatest power of 2, say 2~, which

divides them both, could be taken out, and for that reason

would represent a curve system consisting of 2~ parallels to the

n
and are not both even. We then
2~

have three cases : 1) nand odd, (n) represents a curve system

5
which can be mapped onto (n,5) curves parallel to c.

2) n even, 5 represents a curve system which can be

mapped onto (n,5) curves parallel to a. 3) n odd, 5 even,

the curve system can be mapped onto (n,5) curves parallel to b.

In this way we have derived the invariants of a curve system under

mappings, from the symbols.

d) An orientation of the system (n) induces one on r1,

5
chosen so that the piece of L4 between r1 and a curve of the

system receives the same indicatrix from both sides. A mapping

sends rl to itself with orientation preserved, and hence the

orientation.of the system is carried over to its image by

means of r l , and hence onto the curve a, b, or c onto which

can be mapped. It is easy (see Fig. 26) to carryover orientation

 319

from to a,b, c on the one hand and - by the first self-

connection of a in L - to (n) on the other.

3,a l 6

We shall now take a particular orientation of rl as positive,

and thereby also take a particular orientation of a l , bl , c and

(~) as positive. We shall take to be the positively oriented

curve, or positively directed parallel system in case (n,6) > 1,

whkh was also denoted by (~) previously. The oppositely directed

curves will from now on be denoted by the symbol Our

coordinate transformation (a + b) is now seen as:

In this way the upper symbol of the pair, n, n, etc. always retains

the same sign under the operations of coordinate transformation and

the euclidean algorithm. Now we shall also, as we did earlier with

the symbols for curve systems on the torus, give the symbol a

different geometric meaning according as n is positive or negative,

and in fact change of orientation of the curves shall also change

the circling direction for or, in other words, when represents

a curve system, (-0)

-n shall represent the same curve system with the

opposite orientation. This convention is necessary in order to

represent one mapping by the same transformation formula

6' =6 + kn for arbitrary orientation (without this convention k

would have to change its sign with the sign of n for the same

mapping. )
 320

Examples 1)

by twist in L .. ( 1)
2,a -1

by coordinate transformation a" b

by coordinate transformation b.. a

i.e. a twist in and a twist in L 2 ,b yields a mapping which

sends c " to itself, hence it is a twist along c.

2)

-1 -1 -1
c "(-1)" (1) by coordinate transformation a" b

.. (-11) .. (-1) by coordinate transformation b .. a

-1
-1
.. (-1)·

The result of this section is : there is a one-to-one correspondence

between the systems of closed, like-oriented curves, disregarding

curves parallel to the boundaries, and the symbols

and are positive, negative or zero numbers, but not both zero.

e) Geometric representation of the symbols

n
Once again we shall consider the original symbol (8)' where

n is a positive number. The set of 2n symbols Sl' ... 'Sn'

Sn+l, ... ,S2n corresponding to the points on al and is mapped

onto itself by Si" SI* ' where I* i + 8 mod 2n, by a twist in

The self-connections in send to SI ' where

 321

i 2n + 1 - i*, hence i =1 - i - 0 mod 2n ; then another twist

in L2 ,a sends i to i'*, where i'* =i - 0 =1 - i - 20 mod 2n,

and then a self-connection in L sends it to

3,a l
i' = 2n + 1 - i'* _ i + 20 mod 2n.

Thus the symbol (n) yields a substitution

o

i' _ i + 20 mod 2n

under a single circling of b. This substitution is cyclic, and

hence all cycles into which it decomposes have the same number of

terms, and in fact there are exactly (n,o) such cycles. This

corresponds to the above result that the curves of a system are all

parallel, and (n,o) in number.

The connection between the symbol (n) and the above substitution
o
leads to a graphical representation of the curve system (see Fig. 27):

s" r"
n)
1 S, .. , 5~ SJ'-t

Fig'. 07·
 322

we use a square net and place the points Sl' ... 'S2n on all

horizontal sides in the same order, at distance E from each other,

and with the points Sl and S2n at distance from the vertices.
2

Then the curve system represented by is also represented by

all lines which connect a point S. with abscissa x on the lower

l

side of a horizontal strip with a point with abscissa x + 20E on

the upper side of the strip. Lines which are homologous in the

net, i.e. differing only by parallel displacement of the net, are

considered to be the same, i.e. lines are considered only "modulo

the net". The lines yield the corresponding substitutions by their

intersections with successive horizontal lines of the net. In place

of the square net we can use a torus to represent the curve system,

since the universal covering of the torus is the square net. However,

it is obviously impossible to regard the square net as universal

covering of the four-holed sphere.

As we have just seen, the curve system on L4 corresponds to a

substitution of 2n symbols, generated by a reflection I, namely

the index substitution i' = 2n + 1 - i, and a cyclic shift

Z : i' == i + mod 2n. The substitution I-1Z-1IZ is again a

shift: i' i + 20 mod 2n. This is merely the simplest case of a

very frequent phenomenon in topology. Let us consider, e.g., the

1
double torus which results from joining two one-holed tori Rl,l
and along the boundary r. As we saw in §6b) , the curve

systems on and map onto curves in three divisions.

 323

These divide the 2n intersection points on r into three pairs

of divisions. Suppose we have the intersection points

1
for Rl,l Here the points are ordered on r according to the

natural order of indices 1 to 2n. In the order written above,

1
the endpoints associated by connections on Rl,l stand in the same

column. The resulting correspondence is a permutation of the

naturally ordered 2n symbols of order 2, call it I. We have a

similar substitution I' of the 2n symbols corresponding to the

curves on with other division members, say The

curve system on the double torus results from a twist Z along r,

i.e. a substitution i' =i + 0 (mod 2n), and this curve system

-1 -1
therefore corresponds to the substitution ZIZ I' . For other

topological figures we have quite differently constructed substitutions

I of second order, e.g. for curve systems on the six-holed sphere,

where the three division pairs follow each other in the order

In the simplest case, which we

Fig. ~8.
 324

have considered above, where only one pair of divisions is present

-1 -1
and (quite unimportantly, however) I = I', ZIZ I was again a

cyclic substitution and the euclidean algorithm gave us the essential

properties. In other cases one at present knows only very unclear

processes based on topological considerations. Even a simple

transformation corresponding to the above "coordinate transformation",

the exchange of nand 0, is lacking. Perhaps the geometric

representation of the simplest case by a square net may be carried

over to more complicated cases by regular subdivisions of the non-

euclidean plane.

f) Curve systems on the four-holed sphere with endpoints on

one boundary

1) We shall proceed differently from our treatment of the one-holed

torus by first making a direct derivation, by adding some topological

considerations to the previous results. If we cut the L4 along

a self-connection 0 of the boundary rl which is not contractible

to (see Fig. 29), the result is an L 3 ,a and an

Fig. 09.
 325

On the there is a second boundary of the

The boundaries and are then the other two boundaries of

the The boundary of L2 ,a other than is a curve

consisting of the self-connection a and a piece r 1,2 of the

boundary r1 between the endpoints of a. Similarly, the boundary

of the other than and r4 is a curve made up of a

and the other piece r l ,34 of between the endpoints of a.

We now consider a curve a parallel to this latter boundary

in This is a closed curve on the which separates

and from and It then follows from previous results

that it can be mapped onto the curve al by a mapping of L4 onto

itself with movable, but not exchangeable, boundaries. Under this

mapping a goes to a self-connection s of rl which lies in

L and which, by results proved earlier (§3), is homotopic

3,a l
with movable boundaries to a certain self-connection s of the

boundary in L , and hence also on

3,a l

If we have several (disjoint) self-connections a. of

1.

which separate from and r 4 , then we cut along the connection

for which the boundary piece r l ,34 of r l , which belongs to

L3 ,a' contains no endpoints of other such self-connections. A

curve parallel to in can again be carried to

by a mapping. Now all these self-connections lie in L and

3,a l
they can be carried to fixed self-connections of rl by a homotopy
 326

with movable boundaries. We now consider the three-holed sphere

It contains any self-connections of which are not

homotopic to a O' because all those lying on the two-holed sphere

L2 ,a (and which are not homotopic to a piece of rl ) are homotopic

to The endpoints of every other connection are movable on

r l ,34 by a homotopy of the L4 , but naturally not across aO.

Thus we have here the case of the fixed boundary, and by the preceding

we obtain two different kinds of self-connection of r l , not homotopic

to each other or to a boundary piece, and especially not homotopic

a. ,
,
to aO' which differ from each other in the fact that one, say 1-

encloses r3 together with a piece of r l ,34' while the other,

say a i ' , , similarly encloses r4 (see p.294 and Fig. 29a) .

Fig. '9 H.

The mapping of a onto sends the first onto the three-holed

sphere Because the parallel curve a', which is constructed

 327

relative to 0i' as a is above to 0i' twice cuts a, or its

image al under the mapping a + a l , as we see from the figure

la' is shown dotted in both fi~ures of 29a), and therefore has

1
(after the mapping a + all the symbol (6) , and can be mapped
1 1
by a twist in L 2 ,a onto (0) , Le. b l , or (1) , Le. c. But

it can only be bl (Le. (1) and even) in case a' separates

0

and and r 3 , as supposed. Similarly, the

curve ai' parallel to 0. II is sent by the mapping to a curve

1
1
with the symbol (6) • However, since a" also has two points

in common with a', a" must go to a curve with the symbol

in case a' is mapped onto bl · Now, we can convert the curve

( 1) into (1) by a twist in or

-1 " cl * 1 - cl L2 ,a L2 ,b' but in
the process either bl or al goes to a non-homotopic curve.

When the curves al parallel to 5 and bl parallel to s' both

go to homotopic curves then only a homotopic mapping is possible,

and connections s" parallel to cl are not convertible into

parallels to c1* The connections 5' , lie either in the three-

holed sphere L (containing determined by

3,c l
in the three-holed sphere L3 ,c l *
(containing rl ) determined by

In summary we have:

Each system of self-connections of a boundary rl can be carried

by a single mapping onto a system of fixed connections consisting of

three divisions, which lie in or

 328

It is clear that one can pre5cr~be one of the three-holed

spheres L or but then one must admit L3 ,a l * as well

3,c l L3 ,c l *'
as L , or else L3 b * as well as L3 b ' according as one
3,a l
' 1 ' 1
converts c l into c * by a twist along bl or al ·
1

If our curve system consists, not only of self-connections of

the boundary r l , but also of closed curves, then we can convert

all the closed curves into one of either a l , bl or c l ' and the

self-connections of rl all go thereby into the division belonging

Thus in this case we have, as on the one-holed

torus, only two divisions : the closed curves and one division of

self-connections.

2) In 1) we have found a normal form for (unoriented) curve

systems with endpoints on ~ boundary, and we shall now see how this

normal form is expressed in symbols. AS before, the symbols are

introduced by dividing the L4 and Then the curves

in L , and likewise in L , can be carried to fixed connections

3,a 1 3,a 2
by homotopies with movable boundaries. All that now remain are

the connections in L2 ,a' which are determined by a twist number 0

relative to a fixed connection, say the piece of bl between and

a2. The curve system is then determined up to homotopy by the

number 2n of intersection points on a1 and a 2 , the number 2q

n
of endpoints on r l , and o. We denote the curve system by (0 q).

In the case where the curve system is made up of ~ closed curves

 329

parallel to al resp. b l , resp. c I resp. c l * and m self-

connections we have the symbols

(0)
i. m resp. (mo+i.m) resp. (m+i.
m+i. m) resp. ( -m-i.
m+i. m) .

In the case where there are no closed curves, but instead m self-
a
connections parallel to aI' ~ self-connections parallel to bl ,

and mc self-connections parallel to cI resp. c l *, we have the

symbols

ma+m c m +m. +m ) resp. (ma +mc m +m. +m )

( mao c -m abC
c c

We shall call this coordinate system the (a,b)-coordinate

system (the second coordinate is decisive for the number 6). If

we choose the corresponding (b,a)-coordinate system, then m and

a
are exchanged. We denote the symbol in the latter coordinate system
n'
by (6 ' q) . We see that when ~ "f 0, q > n, and when rna "f 0,

q > n' . However, i f m =~ 0, then we introduce the (c,a)-

a
coordinate system and obtain the symbol (~mc) in case the self-

connections are all parallel to c l ' which we can always arrange by

twisting in L2 ,a or L2 ,b In the (c,a)-coordinate system we

n' ,
denote the symbol by (6' , q) . We then have the result that for

each curve system either n, n' or n' , < q in a suitable coordinate

system and after a suitable mapping. The (c,a)-coordinate system is

necessary only when n = 161 ~ q, i.e. when n = 6~ q after a

possible twist in L 2 ,a In this case the curve system is parallel

to and contains n-q closed curves.

 330

We can avoid the topological considerations when we consider

n
transformations of the symbols (0 q) as we previously (§6)

considered the transformations of the analogous symbols for the

one-holed torus. It is little more than a repetition, with the

modification that now only the even euclidean operations appear:

n
the reduction of the symbol (0 q) proceeds by first using the

substitution °= °+ 2kn to make 151 < n. Then we go to the

nl
(b,a)-coordinate system and obtain the symbol (0 q) where
1
nl = 151 < n, and 01 is determined in exactly the same way as when

q = 0 (no boundary points) in case 101 and n are both ~ q.

In this case ° = _o_n.

By even euclidean operations we can
1 n 161 nt _l
therefore convert the symbol ( q) into a symbol (
° °t_l q) for

The symbol represents a curve system, which

results from the given one by mappings corresponding to even euclidean

operations. However, if nt_I> q > lOt-II then

(exactly as in the case of the one-holed torus) and in general we

nt
can proceed from ( q), when n f 0, by the operation
°t (0*
nt
to a symbol q) in which nt as well as 10*1
is < q.

Here we have considered the symbols only as far as is necessary

for handling the problem of the five-holed sphere. However, we also

remark that in the (n,o)-plane (for a particular coordinate system)

 331

a fundamental domain in this case is the isosceles right-angled

triangle with the vertices (0,0), (q,q) , (q,-q) together with the

half-lines n = 6 >q ; n = 0, > ° and n > q, 6 = 0.

g) L4 with fixed boundaries

1) We saw above (Examples 1 & 3) that a twist in L2a and a

twist in L2b generate a twist along c. Thus, with fixed boundaries,

suitable twists along a, band c yield a combination of twists

along the four boundaries. We shall now determine this combination.

We denote the twists along a, b, c and ri by 6a , 6b , 6c and

6 and fix the sense of twist (see Figs. 30, 30a) as follows:
r.
1

6a replaces XX O by XX I X2XO'

6b replaces YY O by YY I Y2YO'

6r replaces RR O by RR I R2RO'
1
-1
6c replaces ZZO by ZZ l Z2 Z3Z0

Fig. 30. Fig. 30:1.

The twists commute with each other and with the twists
 332

We now consider the line A1A3 connecting the first and

third boundaries (see Fig. 31)

Fit:. 31. Fi;:. 31 :1.

By our definition, AbA a (A1A3 ) is homotopic to the connecting line

The latter also results from A1A3 by the mapping

Ar Ar Thus A-lA-lAb'
u u u u maps th e connec t'~on A1A3 onto ~'t se lf •
l 2 r r3l a
The mapping A-1A- A A similarly carries the connection A A
r2 r4 b a 2 4
(see Fig. 31) to itself. However, since do not

alter the connection A2A4 , and Ar and do not alter the

-1 -1 -1 -1 2
connection A1A3 , Ar Ar Ar Ar AbAa carries both connections
1 2 3 4
to themselves. If we cut the L4 along these two

connections, the result is an L2 which is mapped onto itself by

Thus the latter is a mapping of the L2 , i.e. a

twist along c, and hence equal to In order to find m, we

consider the connection F1F2 (see Fig. 31a). We have

 333

as follows from the figure. Thus

This mapping

therefore belongs to the class of the identity, and in the mapping

class group,

2) We shall apply this result to the two-holed torus with fixed

boundaries. By cutting along a (see Fig. 32) we see from 1) that,

with suitable labelling,

Fig. 32.

Now ~c is generated by ~
a
and ~b' by §6c. Thus is

generated by ~
rl
,~,
a
~
a2
'~b and ~d: The twist along one boundary

is generated by twists along the other boundary and along the curves
 334

a, a 2 , band d which do not separate the torus.

3) Surfaces of higher genus with one boundary curve

The surface is divided by p-l separating curves Cl ,··· ,Cp _ l

Rl 1,
into an and p-l R2 s. (See Fig. 33. ) By §6c, the twist
1
1
along the boundary Cl of the Rl is generated by twists along

Fi,!!. 33.

1
curves which are not separating on the (single) Rl , and hence

certainly not on the By the result just proved in 2), the twist

along C2 is generated by twists along Cl and along curves which

Rl p
are non-separating on the and hence also on the Rl ·
2,c l c 2 '
Thus the twist along C2 may be generated by twists along curves
p
which do not separate the Rl ·

Continuing in this way, we obtain the result: the twist along

the boundary of a surface Ri may be generated purely by twists

along non-separating curves of the surface.

 335

4) Twists along separating curves of an or

A separating curve on an RP divides it into an Rq and an

0 1
p-q
Rl ' a separating curve on an RP divides it into an Rq and
1 1
an It follows from 3) that : the twist along a separating

curve of a surface with at most one boundary curve may be generated

purely by twists along non-separating curves on the surface.

since the non-separating curves can all be sent to each other

by mappings of the surface, i.e. they are all equivalent, and

since mappings send twists to twists, we have the result: each

mapping of a surface with at most one boundary curve generated by

twists may be generated purely by mappings which are conjugate in

the mapping class group.

§ 8

The five-holed sphere

a) Coordinate systems. On the L4 we have three closed

curves a, band c which intersect pairwise in two points and form

the basis for the determination of any curve system. Each closed

curve of L4 (which is not contractible nor homotopic to a boundary)

may be mapped onto one of them. One thinks analogously of introducing

ten closed curves on the LS ' each of which separates two of the

boundaries from the other three. If we denote such a curve by

(i 'I k) in case it separates the boundaries r. and

1

rk from the other boundaries, then the ten curves a ik must satisfy
 336

the condition that and intersect in two points, but

and a (9, ~ i,k ; m ~ i,k) have no point in common. How-

9,m

ever, this is impossible because, as is easily seen, the latter

condition will not be satisfied for at least one pair. This is

because five points on the sphere cannot be connected with each

other by segments which do not meet outside the five given points.

We therefore content ourselves with five closed curves as "coordinate

axes", say and (see Fig. 34).

Each of these curves divides the surface into an

and an L 3 ,i· The contains the two curves disjoint from

ai,i+l' namely ai+2,i+3 and a i +3 ,i+4· The indices here are

always reduced modulo 5.

b) Representation of a system of closed curves on the LS.

We divide the surface into an and an by a curve

ai,i+l' with i = 4, say. In and we draw curves

and a' I parallel to Then the curve system on the L4 ,4

4S
is given by the number n l2 of intersections with the curve a l2 ,

a twist number ~l2' and the number n 4S of boundary points on

a 4S ; n 4S is also the number of intersections of the system with

 337

a' , , r 4
a 45 · On the L3 ,4' with boundaries 45
and r 5 , the curve

system consists, apart from possible parallels to a 45 , of self-

connections of the boundary a 45 , and hence it is given by the

number (with movable boundary There still remains the

curve system on the L2 ,4 with the boundaries and a 45 , which

is determined by a twist number 045. The curve system is there-

fore fully determined by the pair of symbols

We can similarly divide the surface by a 12 . Then for the

same curve system we have the symbol pair

We shall denote this coordinate transformation, an exchange of a 12

If, on the other hand, we carry out an exchange

of coordinate systems on L4 ,4' from (a 12 ,a 23 ) to

(see the exchange of a- and b-axes on the four-holed sphere in

§7), then we obtain the pair

If, e.g., n 45 < n 12 and < 1°121, then

 338

and

Under a change of coordinate system in L4 ,4 the normal connections

of the boundary with the axis must be carried to normal

connections of with the axis The result is a twist

along the boundary a 45 and the relation 045 = 045 ± n 45 . We

shall not go into the sign determination for this relation or the

derivation of the various other coordinate transformations for the

five-holed sphere. They are not important for our present purpose.

However, the following is important: the coordinate system (a 12 ,a 23 )

in L4 ,4 corresponds to the coordinate system in L4 ,1

under W14 , but the coordinate system in corresponds

to the coordinate system in under

c) Reduction of symbols. By §7f) we can use a mapping of

L 4 ,4 with movable boundaries to carry the curve system to one with

the symbol pair

( ni2 n
0' 45
Xn45 ), where
0'
is smaller than In general the
12 45

coordinate system remains unchanged in this reduction, it can remain

as However, when the curve system in L4 ,4 consists

of only two divisions, then it may be necessary to use the (a 23 ,a 12 )-

coordinate system in place of (a 12 ,a 23 ) in order to achieve the

reduction to ni2 < n 45 · Finally, it is necessary to use the

 339

(a 13 ,a 12 )-coordinate system in order to achieve ni2 < n 45 when

We shall treat this exceptional case later.

Then after reduction and the exchange W14 of four-holed spheres

we have the symbol pair

where is now and where the three-termed symbol refers

to the and the (a 45 ,a 34 )-coordinate system or, for the

special case, to the L4 ,2 and the (a 45 ,a 5l )-coordinate system.

In each of these cases we can again reduce the symbol, so that

We then take the corresponding exchange of four-holed

spheres again, either W4l or W3l or W42 or W52 Jthe first

exchange corresponds to the "general" case) and reduce in the new

four-holed sphere. This process breaks off in case the number of

boundary points in the three-termed symbol is equal to zero. We

then have, say,

c::::: ~ Gk.J·
This symbol denotes curves parallel to ak,k+l and

(ni,i+l' 6i ,i+l) curves parallel to one of the three axes in Lk,k+l

Two of these three axes are of the form ag"g,+l and are already

determined. Still missing is the determination of the third, which

 340

is of the form a t ,t+2' where t, t+l, t+2, k, k+l are all

different from each other modulo 5. We must determine the five

axes in such a way that they do not intersect the axes

ak,k+l' This is done in Figure 35, in which the holes are reduced

Fig. 3;.

to points for the sake of clarity, the curves a ik are given by

doubly-traversed segments, and circuits around the holes are left out.

The fact that and intersect at four points in our set-up

is not important for these cases.

However, we still have to deal with the exceptional case, say

Suppose that is denoted by in

with the (a 12 ,a 23 )-coordinate system, and that n 12 = °12 , which

we can always arrange by a twist along a 12 . Then the symbol

 341

represents curves which do not cut a l3 . Among them, if n l2 > n 4S '

there are closed curves parallel to The remainder

are closed curves which are not parallel to boundaries of the L4

cut along a l3 , and hence capable of being mapped onto curves

parallel to a 24 , a S4 or a 2S by twisting along a 24 and a S4 ·

But the twist along a 24 in L4 ,S can be composed from twists

along a 23 and a 34 and the twist along the boundary a Sl · In the

exceptional case the fact that a S2 and a l4 intersect

comes into play, in case our reduction process brings us to curve

systems in L4 ,4 or Because the axis does not lie

in the L4 corresponding to a l4 (in fact it intersects a l4 ) , and

a l4 does not lie in the L4 corresponding to a 2S · Thus we choose

two more curves, ais which does not meet a l4 , and ai4 which does not

meet a 2S (see Fig. 3S) . We use these only in case we have to

represent the curves in the corresponding to or

The exceptional case always appears only at the end of the reduction

process.

A curve can be mapped onto no curve a for which one of

~m

the pair (i,k) equals one of the pair (~,m). On the other hand,

a mapping of the LS onto itself with movable boundaries which fixes

two non-parallel, disjoint curves and a~m is a product of

twists along and because the latter curves divide the

Thus in summary we have:

 342

1) Each mapping of the five-holed sphere with movable boundaries

may be composed from twists along and

and by means of such a mapping each system of closed curves, apart

from curves parallel to the boundaries, can be mapped onto closed curves

parallel to two disjoint curves among the twelve curves a ik , ai4 and

ais (there are IS such pairs) .

2) The mapping class group for the LS is generated by

linked even euclidean operations on the symbols and

§ 9.

Generation of the mapping class group for the

We shall prove the following theorem:

Each mapping of an n-holed sphere belongs to the same class as

a mapping composed from twists along fixed curves, to be chosen.

The proof is derived from Lemmas 1) to S):

1) Two closed curves which separate the same boundaries on the

Ln are equivalent, i.e. they can be mapped onto each other by a

mapping of the L with fixed or movable boundaries. Suppose, say,

n

IThe mapping class group for the L, when the boundaries are points
and can be exchanged, has been invgstigated by W. MAGNUS, Math. Ann.
109, p. 617 ff. Magnus finds not only a system of generators, but also
a system of defining relations. Use is made of the connection between
the mappings of L onto itself and the "braids" introduced by E. ARTIN.
W. BURAU, Abh. HamB. Math. Sem. 9, p. 117 ff. has given generators and
relations for a subgroup of the braid group which corresponds to mappings
of the punctured sphere with non-exchangeable boundaries. It may be
conjectured that his results are related to ours.
 343

that and both separate the boundaries from

the boundaries Then by cutting the L along

n

resp. c 2 we obtain the surfaces LI andL2 LI resp.

m+l m+l n-m+l
and L2 By a mapping ~, say, we can map the surface LI
n-m+l m+l
onto L2 in such a way that the boundaries r1,a .. ,rrn on LI
m+l m+l
go onto L2 onand so that goes onto
m+l'
Moreover, we can map the surface LI onto L2 by 'i' in
n-m+l n-m+l
such a way that rm+l, ... ,r n on the first surface go onto

rm+l, ... ,r n on the second surface, and likewise so that ci goes

onto c 2 ' in the same way that ~ maps ci pointwise onto c2.

Then ~ and 'i' together form a mapping (~,'i') of the whole L

n
onto itself with non-exchangeable boundaries, carrying ci to

One proves similarly that: two self-connections of the same

boundary which separate the same boundaries are equivalent. The

same holds for two connections between the same pair of boundaries.

2) Two closed curves which separate the same bou~dary from

all other boundaries are (not only equivalent, but also) homotopic.

Because each of them is homotopic to the separated boundary, since

each bounds jointly with it an L2 on the L.

n

3) Two closed curves which are disjoint and which separate

the same boundaries from each other are (not only equivalent, but

also) homotopic. Because the two curves together bound an L2 on

the L.
n
 344

4) When cl divides the boundaries into two groups, neither

of which includes one of the groups enclosed by c 2 ' then cl and

must intersect. In particular, we take c l to be a curve which

separates just two boundaries, r l and r 2 , from the others, and take

c2 to be an arbitrary curve which separates rl and r2 from

each other. By a homotopy of within or its neighbourhood,

c2 can be changed so that all parts of c2 which run in the L3

bounded by cl ' r l and r2 become self-connections of cl which

separate rl from r2 (see Fig. 36) • Since rl and r2 now

lie on different sides of c 2 ' the number of these self-connections

Q
Fij!·36. Fig. 37.

is odd and hence the number of intersections of cl with (the

transformed) c2 is not divisible by 4. Now we similarly consider

the L n _l (see Fig. 37) formed by the boundaries on

the other side of together with itself. Again we can

use a homotopy to remove pieces of which are self-connections

of not separating any of the boundaries r. from the others

1.
 345

on the Ln _l . Such a removal reduces the number of intersections

of with The result is possibly more self-connections

inside the L3 on the other side of cl which do not separate rl

from r 2 , and which we can therefore remove. We repeat this process

until each self-connection in the L3 as well as in the Ln _ l

separates at least one boundary from the others. Thus we obtain,

0
by homotopic transformation of c 2 ' a curve c2 with 4m+2 inter-

sections with cl and 2m+l self-connections of cl in L3 and

L of which each in L3 separates rl from r2 and each in

n-l'
L n _l separates at least one boundary from the others.

We now consider another curve c l ' disjoint from c 2 ' which

again separates just two boundaries, say r4 and r5 from the

others, but now suppose and lie on the same side of

We apply the same homotopic transformation as produced from

relative to L3 to L n _l , but now apply it to relative

and L n _l , i.e. we remove all pieces of which are not

boundary-separating self-connections in L3 and Ln _l . Removal

of self-connections in L3 does not alter the system of self-

connections in L3 and Ln _ l at all. We shall not at first

discuss the possibility that removal of self-connections in L n _l ,

i.e. replacement of parts of by segments which lie in L3 and

therefore outside L3 and disjoint from c l ' might result in the

creation of self-connections in L3 or If such changes

did occur, then the number of intersections of with could

 346

only be reduced, and hence the number of self-connections in L3

and Ln _l would be also. Then we could again apply the above

process for L3 and Ln _l and obtain a further reduction in the

number of intersections between and c l ' until we again came

0*
to a curve, say c 2 ' which had only boundary-separating self-

connections in L3 and Ln _l . Then we again operate with L3 and

Ln _ l and possibly reduce the number of intersections of the curve

with The process must terminate, and thus we finally

00
obtain a curve c2 which gives only boundary-separating self-

connections in L3 and Ln _l as well as in L3 and Ln _l .

Because of our hypothesis that rl and r2 lie on different sides

00
of c 2 ' while r4 and rS lie on the same side, c 2 has numbers of

intersections with both ci and ci which are not divisible by 4.

Now we shall convince ourselves of the fact, which was indeed

00
not necessary for the construction of c 2 ' that application of the

reduction process to relative to L3 and Ln _l cannot change

the system of self-connections in L3 and Ln _l . We have in fact

the general theorem : two simple closed curves on a sphere which cross

form at least two 2-gons on each side of each curve. The proof

of this theorem follows immediately by consideration of the self-

connections formed by one curve, say y, in one of the two regions

determined by the other curve; because the pairs of endpoints of such

self-connections cannot separate each other on y. Now a self-

connection of c~ in Ln _l which separates no boundary from the

 347

others and therefore must be removed, forms a curve y' in

combination with a piece of el , dividing the sphere into two regions,

one of which contains no boundary. If now each self-connection,

and hence y', were to intersect the curve c l ' then, by our general

theorem, the self-connections of cl in this region would form

with y' at least two 2-gons containing no boundaries. However,

a piece of y' which forms a 2-gon with a piece of cl is a self-

connection of or Ln _ l which separates no boundary

from the others, and hence does not separate from Thus

our process for the L3 and Ln _ l would not be terminated, contrary

to hypothesis. Thus no self-connection of in Ln _l which is

to be removed can meet the curve c l ' and hence the removal of such

a self-connection of the system cannot alter the self-connections

Sa) We consider a self-connection v of the boundary rand

an arbitrary closed curve c. Then there will always be a self-

connection v
o between the same points of the boundary r, separating

the same boundaries from each other, but with two or no points in

common with c. Because c divides the L into an L and an

n m
L and r lies on L
m'
say. If v separates none of the
n-m+2'
0
boundaries lying on L then there is a self-connection v
n-m+2'
of r in L which separates the same boundaries on L as v
m n
and has no intersection with c. If on the other hand v separates

boundaries on Ln - m+2 , then there are first of all two connections

00.
of c and r, v ll and v12 ' wh~ch together separate the same
 348

boundaries on L as v. Likewise, on L there is a connection

m n-m+2
o
v 2 ' between the endpoints of and on c, which separates

the same boundaries as v. In this case we obtain the desired

self-connection v
o of r as the polygonal path which

twice cuts c.

b) Let vi and be two self-connections of the same

boundary r whose endpoint pairs do not separate each other on r.

Then it is proved as in a) that can be carried by a mapping of

the L into a self-connection which has two or no points in

n
cornmon with

c) We consider a system v l ,v 2 , ... ,vm' of disjoint self-

connections of r, and an arbitrary curve c. In the part of r

between the endpoints vi there can be no endpoints of v 2 , ... ,vm'

in the part of r between the endpoints of v2 only the endpoints

of vi' etc. Some mapping of L takes to which has two

n
or no points in cornmon with c. This mapping takes v 2 '··· ,vrn to
0 0 0 0
v 2 '··· ,Vme We construct a closed curve r from vi and the
0 0
part of r on which all the endpoints of v 2 '··· ,vm lie (see Fig. 38,
0
in which r is mistakenly denoted by rl and vi by vi) .

Fig. 38, Fig. 3<).

 349

This closed curve bounds an L on the L which contains all

m n
self-connections This Lm either contains no part

of c, in which case do not cut c at all, or else

the whole curve c lies on Lm' or a part c

o of c is a self-

connection of r O with endpoints separating no endpoint pair of

0 0 rOo We now apply the considerations of a) resp. b)
v 2 '··· ,vm on
to c and
0
v2 resp. c
o and and obtain, by a mapping of L
m

with fixed boundary r O' and hence also by a mapping of Ln' a

0 00 00
mapping of onto v 2 , where v 2
v2 has two or no points in common
0 0 0
with C. In the process, v l is unaltered, v 3 '··· ,vro go to
00 00
V3 , ... ,vm . We now handle similarly and continue, finally

obtaining the result: the system v l ' v 2 , ... ,vm can be carried,

by a mapping of Ln' onto a system of self-connections, each of which

has at most two points in common with c, so that the system itself

has at most 2m points in common with C.

6) The theorem stated at the beginning has been proved for

n ~ 5 (§§5,7,B). We shall now prove it for each n (first without

determining the number of generators) with the help of complete

induction. We therefore assume that it has been proved for L

m
with m< n and from this assumption prove it for m n. We also

assume that n ~ 5. We consider three curves (see

Fig. 39) which separate the boundaries rl , r2 resp. r 2 , r3 resp.

r 4 , r5 from the remaining boundaries. We choose these curves so

that and have two points in common, and is disjoint from

 350

cl and c2° Moreover, we denote the Ln_l's which cl ' c2 and

Ll 3
c3 bound on the L by respo L2 respo Ln_lo An
n n-l n-l
arbitrary mapping ~ of L sends c2 to Y2 ; in general, Y2
n
has points in common with c3 0 We shall now construct a mapping
-1 -1
~ of Ln such that ~ (Y 2 ) has no point in common with c3'
3 -1
and therefore lies in Ln_lo Then, by 1), the curve ~ (Y 2) can
-1 3
be carried to c2 by a mapping ~3 of Ln _ l , so that
-1 -1 -1 -1
~3 ~ ~(c2) :
c2 ° Leo ~3 ~ ~ is a mapping (~2,!l2) composed

from a mapping ~2 of and a mapping of the three-holed

sphere By our induction

hypothesis, ~3' ~2 and !l2 can be composed from twists along finitely
3
many particular curves in Ln _ l , and It remains only

to compose the mapping ~-l from twists along finitely many particular

curves:

By 4), we can use homotopies to convert the curve Y2 into

00
the curve Y2 which forms no boundary-free 2-gons with cl respo
00
Then Y2 has intersections with and inter-

sections with c 3 ' where is odd and is eveno Thus

1
m3 ; suppose We then consider Ln _ l , in which

consists of ml self-connections of the boundary clo By 5),

we can move these self-connections (with fixed boundary c l ) by a mapping

of Ll so that they have at most 2m l points in common with

n-l
00 00
Also, we can deform the curve ~ll (Y 2 ) : Y21 homotopically

(by 4) so that it forms no boundary-free 2-gons with cl and

 351

In this way we obtain a curve which has points in

common with and points in common with c 3 , where

Thus after possible homotopies,

which we think of as combined into ~ll' the number of intersections

with is reduced, the number with is not increased, and

we still have mll f m31 · If mll < m31 again, then we change
00 Ll
Y21 by a mapping ~12 of However, i f m31 < mll , then
n-l
00 L3
we change Y21 by a mapping ~32 of In this way we
n-l
00
obtain a curve Y22 which has m12 -;:: mll points in common with cl

and m32 ~ m31 in common with c 3 ' where strict inequality holds
in at least one case. We continue to so reduce the number of

intersections with at least one of the two curves and until

000
we finally have a curve Y2 with no more points in common with c3.

This attains the goal of finding the mappings ~ik (i = 1 or 3)

-1
with which to compose the desired mapping ~ which makes into

a curve disjoint from c3. By our induction hypothesis, the

mappings 'I'ik result from twists along particular curves in

and Thus in summary we have: each mapping of the L results

n
by composition of mappings of L1n _l , L2 and L3 (The mappings
n-l n-l
L2 3
of onto itself can be regarded as mappings of Ln _l · )
3

7) As the first example of the general method we shall once

again treat the LS (see Fig. 40).

 352

..
.,. ,

"
...,
Fi.~. 40.

We denote the of the previous section more precisely

by c S2 ' c U ' c 3S ' and correspondingly denote the L4 which

results from cutting along By the results on the

L4 , the mappings of (with fixed boundaries) are generated by

twists along c S2 ' c 13 ; c 3S ' r l , r 2 , r3 ; the mappings of

(with fixed boundaries) by twists along c 3S ' c 24 ; c S2 ' r 3 , r 4 , rS

the mappings of (with fixed boundaries) by twists along

Thus we have newly obtained the result that

all mappings of the LS (with fixed boundaries) are generated by

8) In general we arrange the boundaries r 1 ,··· ,rn of the L

n'
as for L4 and LS' along a curve C which cuts each of them

twice (see Fig. 41) . The piece of C between r.~ and r i +l is

denoted by C .. We connect a piece C. with a piece Ck (i < k)

~ ~

by a curve c ik c ik separates the boundaries r k +l '··· ,r n ,

r 1 ,··· ,r i from the boundaries r i +l '··· ,r k · We choose the curves

so that and have two points in common if (i,k)

 353

separates (i' ,k'), and no point in common if noto The corres-

ponding curves on the L(k-i)+l and Ln-(k-i)+l cut off by

are those already constructed for the L • Now since, as we have

n
already seen, all mappings of an LS ' say, can be composed from

twists along curves of such a system, it follows from our con-

struction of mappings of the L that the latter corne from mappings

n
of three Ln_l'S and, by the induction step from n-l to n, all

mappings of the L can therefore be generated by twists along the

n
n(n-l)
--2- curves Here the mappings are considered to have

fixed boundaries. If one allows the boundaries to move, then the

number of generating twists is reduced by the number of boundaries,

i.e. by n. We then obtain generators. This result

holds for n? 3, while the result for the number of generators for

mappings with fixed boundaries holds for n ~ o.

no

The mapping class group for each orientable surface

1) By mapping an RP onto itself, an arbitrary simple connection

n
between two boundaries and can be carried to any

prescribed such connection

2) For any two systems of disjoint simple self-connections i

vl

and vj of a boundary rl and a boundary r2 on an Rn there is a

2 p
connection between and which does not meet the

and v~ : the self-connections vi, in conjunction with r l , divide

 354

the RP into various simply or multiply connected regions. How-

n
ever, none of these is bounded by lines alone, because the

latter are disjoint and not closed. The boundary lies in one

of these regions, say B. The self-connections of all

lie in this region and, in conjunction with r 2 , divide the region

B into subregions. The whole boundary of B belongs to one of

these subregions, say B. Consequently, the boundary of B contains

parts of both rl and r 2 , which can therefore be connected by a

i k
curve not meeting the VI and v2 •

3) We now suppose that we can already find generators for all

Rq when q <p (for q = 0 we have given this system in §9); and

m
on the basis of this assumption we construct the system for We

divide the given (see Fig. 42) by a curve C into an and

o o

an RP- l In the RI we consider a canonical cut system a, b.

n+I,C I,C
An arbitrary mapping ~ of the RP onto itself carries a to (1.
n
We now construct a suitable mapping qil such that ,!,-l~ (a) = a.
 355

Then is necessarily a mapping of the RP- l which results

n+2,a
from cutting RP along a, and hence a mapping whose generators
n
are known by hypothesis" Consequently, our problem is to compose

such a from mappings of the and the RP- l which

n+2,a'
can be generated in turn by our induction hypothesis. It may be

remarked that the mapping problem for an cannot be reduced

in this way, because the C-curve always cuts off an However,

the mappings of the have earlier been dealt with directly

(see §b)). 1

p-l
a) When the curve crosses a, it is represented in Rn+ 2 ,a
by a system of self-connections of the boundaries a and a of
i
RP- l and connections v - from one of these boundaries to the
n+2,a aa
1
other. We choose one of the latter, say v - By 1), we can
aa
send it onto the connection b of a and a in RP- l by a
n+2,a
mapping of RP- l Thus in this case there are (non-reducible)
n+2,a
1
self-connections of the curve a in Rl,C after the mapping, and

hence by §6b) the number of intersections between a and a can

be reduced by mapping the In this way we can reduce the

1
number of intersections as long as the connections v - are present.
aa
Since the number of intersections between a and a is finite,

this process must terminate, i.e. by alternate mappings of the

IFor the once and twice punctured torus, R. FRICKE has already given
generators and defining relations for the mapping class group. For
this and the multiply punctured torus cf. W. MAGNUS, Math. Ann. 109.
p. 634 ff. A ~ystem of generators for the mapping class group of the
double torus R has been given previously by the method of the
arithmetic fielg, see M. DEHN, Autographie Breslau 1922 and R. BAER,
J. f. Math. 160, p. 1 ff.
 356

p-l i
Rn+2 ,a and Rl we can remove all connections v -. Suppose
l,e aa
these mappings carry a to a .
o

b) If
0 i
a yields self-connections v of the boundary a and
a
k p-l
v- of the boundary a on R + ,a' then by 2) we can connect a to
a n 2
i k
a by a curve a which meets no v or v-. By 1), we then have
a a
a mapping of RP- l onto itself which sends a to b. Then the
n+2,a
self-connections of a and a no longer have points in common with

b. If we return from the RP- l to the RP then we have to join

n+2,a n'
the boundaries a and a together. This possibly results in

intersections of the self-connections with b, but by twisting along

a we can arrange that this number is fewer than the number of

intersections with a. By mapping Rl onto itself, namely by

l,e
exchange of a and b, we can then achieve a reduction in the inter-

sections between a and a also. We continue in this way as long

as the transformed a
o yields connections between a and a or

self-connections of a and a. This process must terminate, i.e.

we can reduce the number of intersections between a and a to zero.

Suppose a
o has now become a' .

c) Either a' is homotopic to a, and hence convertible to

a by a homotopy, in which case we have reached our goal, or else it

is non-homotopic to a, disjoint from a, and non-separating on

p-l
On Rn +2 ,a it is either non-separating or else it separates the

boundaries a and a, since i t is non-separating on RP . In both

n
 357

cases we use a mapping of RP- l to send a' to a curve a"

n+2,a
which cuts b once (see Figs. 42 and 43). A suitable neighbourhood

,,_ v

c,/;'
000
Fi.!.!. 4:;,

1
of the curve pair all, b on the RP constitutes an Rl · By
n
twisting along a" and b in this Rl we can (§6b) ) send a" to
1
b, and by twisting along b and a we can also send b to a

(in But in each case the twist along a" is a mapping

of the RP- l onto itself . Thus in fact we have sent a to a

n+2,a
by mappings in the Rl and RP- l i.e. by a mapping 1[1-1
with
1,C n+2,a'
the desired property. We have now proved: all mappings of a surface

of genus p and n holes can be generated by twists along a finite

number of particular closed curves. In 5) we shall derive such a

system of curves directly from the normal representation of the

4) Examples

a) R~ (see Fig. 44). With our process in mind, we consider

 358

the Rl on and the containing Each mapping

2,a~
of the RO onto itself is generated by mappings of these two surfaces

onto themselves. The mapping of the Rl is generated by twists

1,C
1
along al and
bl · In order to obtain the mappings of the R ,
2,a l
1*
we consider the Rl,C (with a 3 and b 2 as canonical cut system)

on and the All mappings of the onto itself

are generated by mappings of these two surfaces onto themselves. The

1*
mappings of Rl,C are generated by the twists along

the mappings of are generated by twists along (and

along a3 (and a3 ), along a2 and along C. However, (§6c), the twist

along C can be composed from twists along al and bl (or also from

Thus we have: all mappings of the

closed double torus are generated by twists along the five curves

(see Fig. 45). Cutting the along and

 359

yields an L6 whose boundaries may be denoted and ordered as follows

We consider, as in §9, the generating twists c ik ' but leave out

c 62 ' c 24 ' c 46 ' because they can be generated by twists along al

and b l , resp. a 5 and b 2 , resp. a 4 and b3 · We have c l3 = a2,

c 35 a 3 , c 51 and also c 14 ' c 25 ' c 36 = a6 · We claim that, on the

basis of the construction in 3), all mappings of the triple torus

are composed from twists along the 12 curves a l ,··· ,a 6 , b l , b 2 , b 3 ,

and In fact, we consider the Rl on the R3

l,c 62 0

resulting from a cut along c 62 ' and the R2 resulting from a

2,a l
cut along The mappings of Rl are generated by twists
l,c 62
2
along and In order to obtain the mappings of R2 ,a ,we
l
consider the Rl and on this surface. The mappings
l,c 46
of Rl onto itself are generated by twists along and
l,c 46
In order to obtain the mappings of Rl onto itself, we consider
4,a l a 4
the and on this surface. The mappings of

are generated by twists along and b 2 , the mappings of

the L6 are known by §9, 8 to be generated by twists along the

and Thus our assertion is proved.

5) In general, the closed surface is subject to the

following consideration : we choose p jointly non-separating curves

b. has one
l.

point in common with ai' and none in common with ak (k I i).

Cutting the along the a. yields an

l.
 360

By §9, 8, the mappings of this L2p are generated by twists along

2p(2p-l)
curves. But this counts the twists along the p curves
2
a.~ twice, since the twist along is also the twist along a ..
~

Moreover, there are another p twists, as we have previously remarked

several times, which can be generated from twists along the a. and
~

b. which on the other hand generate the mappings of the RI

~ l,aib i
onto themselves. The curves c2i ,2i+2 (see below) along which

these twists are made are all different for p > 2, whereas for

p = 2 they all coincide. Thus the formula for the number of

generating twists when p > 2 has to be reduced by I when p = 2.

For p > 2 we have: all mappings of the (which preserve orientation)

are generated by twists along

2p(2p-l) _ 2p + P 2p(p-l)
2

curves.

These curves consist, apart from the a. and b i , of curves

~

which can easily be given when we arrange the boundaries a. and a.

~ ~

with the p curves b. and another p connecting segments

~

forming a closed curve, as above for p = 3. If we number these

connecting segments successively by v l , ... ,v 2p ' then the other

curves for the generating twists are curves which meet each of

two segments v. and at one point, where and

~

also have two resp. no points in common according as (i,k) and

(i',k') separate each other or not.

 361

Thus for the R~ we obtain 24 generating twists in all.

Among these we have two twists along separating curves, namely

(see Fig. 46) along and By §7f)2) the latter can be

generated by twists along non-separating curves, which are given

in the figure as c 4568 and

6) We now add a short remark on the arithmetic field for the

One obtains the arithmetic field for the Ln by

dividing it into n-2 three-holed spheres L3 ,j by the n-3

closed curves c j , so that the three-holed spheres are bounded by

A curve system on the Ln with fixed boundaries is given by the

 362

intersection numbers and twist numbers for the n boundaries r.

~

and the n-3 curves c j ' and hence by 2n-3 number pairs in all.

With movable boundaries the n twist numbers for the boundaries

drop out. If we divide the jointly non-separating

curves, then we obtain an L2p A curve system on the RP yields

o
a curve system on the L2p for which the intersection numbers on any

two corresponding boundaries are equal and for which the twist

number along anyone boundary in each pair can be taken equal to

zero. Thus we obtain 3p-3 number pairs in all from the above

determination of curve systems on an Correspondingly, the

curve system on an is determined by 3p+2n-3 number pairs.

In §5e) we have proved that the three number pairs which

determine a curve system on the three-holed sphere are unchanged

by homotopies. One concludes from this that the system of number

pairs which we have just given for curve systems on the most general

surface is also unchanged by homotopies.

 APPENDIX THE DEHN-NIELSEN THEOREM

JOHN STI LLWELL

1. Introduction

One important theorem attributed to Dehn does not appear in his

published work, and his proof of it seems to be lost. This is the
theorem that every automorphism of the fundamental group nl (S) of a
closed orientable surface S can be induced by a homeomorphism of S.
The earliest known proof appears in Nielsen [1927], and consists of two
parts. The first, which is in §9 of Nielsen's paper and expounded
below as Theorem 5, is quite elegant and is attributed to Dehn. The
second, in §23, is agonising1y long and is apparently Nielsen's own,
though he still credited the theorem to Dehn. As was mentioned in the
introduction to Paper 7, Nielsen in 1931 found a replacement for §23
which was much simpler, but he did not publish it. We give a similar
proof below as Theorem 6.

It is a pity that Nielsen did not publish his 1931 proof, as his
[1927] proof was made obsolete by the much shorter one of Seifert [1937],
with the result that Dehn's contribution to the theorem was buried along
with Nielsen's. Moreover, since Seifert's proof was purely topological,
it became forgotten that the roots of the Dehn-Nielsen theorem were in
hyperbolic geometry.

With the recent revival of interest in hyperbolic geometry, it seems

timely to reconstruct the Dehn-Nie1sen proof and its geometric background.
This not only fills a gap in the record of Dehn's work, it also brings
to light some little known but influential work of Poincare. I t is

fairly well known that Poincare discovered the connection between hyper-
bolic geometry and surface topology in his work on fuchsian groups in the
early 1880's (for an English translation of these papers, see Poincare
[1985]). In a less well known (though often cited) paper, Poincare [1904]
 364

took up the idea again for purely topological motives. He obtained

results about simple curves on surfaces by a method which turned out to
be crucial to the Dehn-Nielsen proof - the use of geodesics as canonical
representatives for homotopy classes of curves. This method is explained
in sections 3 and 4 below.

It was Poincare [1904] which inspired Dehn to use hyperbolic geometry

in his solution of the word and conjugacy problems for surface groups,
Dehn [1912a] (Paper 4 in this volume). In the 1920's, Dehn was still a
little under the influence of Poincare, judging by the remarks at the
beginning of Dehn [1921], but he was gravitating towards purely topological
methods. The torch of hyperbolic geometry in topology was picked up by
Nielsen in his [1927], [1929], [1932], a monumental series of works which
have only recently been assimilated, in the reworking and extension of
Nielsen's theory by Thurston. In surveying the development of geometric
methods in topology, the Dehn-Nielsen proof seems an excellent vantage
point from which to look back to Poincare and forward to Thurston.

It should be mentioned that the hyperbolic metric used in the Dehn-

Nielsen proof can be replaced by a metric on elements of the fundamental
group, the purely combinatorial "word metric" introduced by Dehn [1912a]
(see the Introduction to Paper 4). A generalisation of Dehn's Theorem 5
is proved using the word metric by Floyd [1980], and it seems clear that
the concepts in the second part of the Dehn-Nielsen proof (Theorem 6)
could similarly be replaced by combinatorial ones. What this probably
means, however, is not that hyperbolic geometry is irrelevant, but that
the most relevant features of hyperbolic geometry are the combinatorial
properties of its regular tesselations.
 365

2. The special case of the torus

Among the closed orientable surfaces of genus ~ 1, the torus

T (genus 1) is distinguished by having a natural euclidean structure.

This is one reason for a separate discussion of T. The other is that

T enables us to review standard facts about canonical curves, univer-

sal covering and Cayley diagram in a context which is free of geometric

difficulties. When we move on, to a surface S of genus> 1, it will

then be possible to concentrate on the geometric difficulties which

arise from the hyperbolic structure on S.

The torus T contains a pair of canonical curves a,b with a

characteristic property: they are simple curves which meet at exactly

one point, 0, where they cross. Cutting T along a,b yields the

canonical polygon TO for T, which is topologically a square whose

edges originate from a and b as indicated in Fig. 1.

cut
~ b

a
Fig.l

This follows from the classification theorem for surfaces (Dehn and

Heegaard [1907]), and does not depend on the special position of

a,b, as Fig.l might suggest. Namely, the cut along b leaves a

 366

connected orientable surface with two boundary curves (the two

"sides" of the cut, which are connected by the curve a). The

classification theorem says that any such surface is topologically a

cylinder, and cutting a cylinder along a simple arc a joining its

boundaries yields the square shown.

Thus if a', b ' are any other curves with the characteristic

property of canonical curves, then cutting T along a', b ' yields a

I
polygon To homeomorphic to TO' and with correspondingly labelled

sides. For example, Fig. 2.

a'

a'
Fig. 2

It follows that there is a homeomorphism T: T + T which maps a onto

aI and b onto b I . One only needs to construct a homeomorphism

T'
of onto
o which maps equivalent points in the boundary of
to equivalent points in the boundary of T~. A more intuitive way

to construct T is to paste the equivalent sides of pI together

again to form a torus on which a', b ' "look like" a, b (Fig. 3).
 367

Fig 3.

Then the map ,: T ~ T which sends a, b onto a', b ' respectively

is defined by sending each point z to the similarly positioned Z'.

To summarise, we have

Theorem 1: If a, b are simple curves on T which meet at a single

point, where they cross, and if a', b ' is another pair with the same

property, then there is a homeomorphism ,: T ~ T which maps a onto

a' and b onto b' . o

It follows that if '*: ~l(T) ~ ~l(T) is the automorphism of the

fundamental group induced by

" then
'*
sends the homotopy class

[al of a to the homotopy class [a I I of a', and the homotopy class

[b] of b to the homotopy class [b'l of b' • We are therefore

able to induce any automorphism I: [al 1+ [a~ ] , [b] 1+ [b '] of ~l(T)
for which [a'], [b'l are representable by a pair of canonical curves.
To prove the Dehn-Nielsen theorem for the torus, then, it suffices
that any automorphism preserves the characteristic property of canonical
curves.

Theorem 2. Any automorphism I: ~l(T) ~ ~l(T) can be induced by a

homeomorphism ,: T ~ T.

Proof: The fundamental group ~l(T) is the free abelian group on two
generators (corresponding to a fixed pair of canonical curves a,b), so
we can identify it with the lattice of integer points (m,n) in the plane,
 368

under vector addition. An automorphism I: ~l(T) ~ ~l(T) is then a one-

to-one linear map of the lattice onto itself, hence given by an integer

matrix

with determinant + 1. I sends the generators (1,0) to (Pl,P2) and

(0,1) to (Ql,q2)' where PI' P2 are coprime, and so are Ql,Q2' since

det I =+ 1.
It can then be checked by brute force that the corresponding
PI P2 ql q2
homotopy classes [a bland [a b 1 can be represented by simple

curves which meet, and cross, at a single point, whence the theorem follows

from Theorem 1. 0

A much clearer view of the above proof can be obtained by looking

at the universal cover of T. I shall now sketch this approach, since

it points the way to go in the case of higher genus, where the auto-

morphisms of the fundamental group are no longer so accessible.

The universal cover T of T is a plane obtained by pasting

together copies of the canonical polygon TO for T so that the

neighbourhood of each vertex looks like the neighbourhood of the single

vertex 0 on T (Fig. 4).

a ~a
r

b b ~ ~ ~

a
T: ~ ,
a a

b~~ b ~ ~~

vertex neighbourhood -0 a
.~

'a
Fig.4
 369

A point P on T, relative to 0, represents the homotopy class

[cJ of an arc c on T with fixed endpoints 0 and P. , is

found by lifting c to an arc , on T with initial point 0, and

its association with the homotopy of class is due to the fact that a homotopy of

c with endpoints fixed lifts to a homotopy of C' with endpoints fixed.

Fig. 5 shows an example in which c is deformed into the straight line

segment between 0 and P.

T: T:
a

b b

o a
Fig.5

If we pull back the euclidean metric from the plane T to the torus T,

then line segments on T cover geodesics on T, and we can represent

each non-trivial homotopy class by a unique geodesic. If we imagine

that T is a self-contained world in which light rays travel along

geodesics, then T is the actual view of T seen by its inhabitants.

In particular, the vertices, which we take to be situated at the

integer lattice points, are the different images of the origin seen

along closed geodesics. Thus the lattice points represent the homo-

topy classes of closed curves based at 0, the elements of TIl(T),

with (m,n) corresponding to the element [a~nl. This recovers the

representation of TIl(T) as the integer lattice under vector

addition, and shows that the net of labelled edges on T is the

Cayley diagram of TIl(T).

An automorphism I: TIl(T) + TIl(T) can then be viewed as a one-to-

 370

one linear map f of the lattice onto itself, and we can extend f to

a one-to-one linear map of the whole plane T. This extended map

sends points (x,y) and (x,y)+( m,n) over the same point P of T to

points f(x,y) and f(x,y) + f(m,n) (by linearity of I) which again

lie over the same point, f(P) say, since f(m,n) is a lattice point by

hypothesis. Thus f covers a homeomorphism f: T + T, and f induces

I because

by definition of f.

Fig. 6 shows the 'f and f associated with the automorphism

I: [a)" lab), [b)>+ [ab 2 ).

a
b

-0 -'" a o
Fig.6

f
~

Although the above proof does not require an explicit

determination of automorphisms of TII(T) , it depends heavily on the

ability of the euclidean plane to admit linear maps which are not rigid

motions. This ability can be expressed algebraically by saying that

 371

any automorphism of the group of lattice translations extends to an

automorphism of the group of all translations. In the hyperbolic plane

there are no linear maps except rigid motions,* and hence there is no

obvious way to extend the automorphism of the "lattice" which represents

the fundamental group of a surface of genus >I to a map f of the whole

plane. In the absence of a map f we have to make a rather exacting

analysis of simple curves and intersections to show that an automorphism

sends one canonical geodesic curve system to another.

*This can be seen from the projective model. In this model, the

hyperbolic plane is the interior of the unit disc, "lines" are portions

of ordinary straight lines within the unit disc, and rigid motions are

all collineations of the ordinary plane which map the unit disc onto

itself. Hence there are no non-rigid maps of the hyperbolic plane

which map "lines" to "lines".

 372

3. Facts from group theory and hyperbolic geometry

Hyperbolic geometry enters the theory of surfaces when one attempts

to construct the universal cover 5 of a closed surface 5 of genus ~ 2

in a metrically regular way. For example, cutting the surface of genus 2

along its canonical curves yields an octagon 50 as canonical polygon

(Fig. 7).

and 5 is constructed by pasting copies of 50 together so that each

vertex has a neighbourhood like the neighbourhood of 0 on S,

namely, Fig. 8

Fig. 8

Thus eight octagons meet at each vertex on 5. This is not possible for

regular euclidean octagons, but it is possible for regular octagons

in the hyperbolic plane, where the angle sum of a polygon can take

any value between 0 and the euclidean value. Thus one has
1f
regular octagons with corner angle 4" ' and one can fit them together

to form 5 as a tessellation of the hyperbolic plane, the labelled

 373

edges of which form the Cayley diagram of ITl (S). The construction is

similar for any genus p ~ 2 - one has 4p 4p-gons meeting at each vertex.

Since ITl (S) is the automorphism group of its own Cayley diagram, this

gives a faithful representation of ITl (S) as a group of hyperbolic motions.

The motion associated with the element g E ITl (S) shifts the vertex

representing the identity of ITt (S) to the vertex representing g, and

hence the vertex representing h, for any hE ITt (S), to the vertex

representing gh.

The topological properties of Cayley diagrams are said by Dehn 1922

to underly the proof of Dehn's theorem. To reconstruct such a proof

I generalise the Cayley diagram construction to any curve system {ail

which cuts S down to a simply connected piece. The curves ai then

lift to a net which partitions S into cells congruent to P. Since

P is simply connected, each element of ITt (S) is represented by a

closed path through the union of the ai on S, hence by a vertex of

the net on S and by the corresponding motion of the net onto itself.

To avoid making continual pedantic distinctions between elements

g E ITt (S) and the vertices or motions which represent them, I shall

speak of "the element g", "the vertex g" or "the motion g" as the

occasion requires.

Of course, one can define S and its motions purely combinatorially,

but the lesson to be learned from Poincare, Dehn and Nielsen is that

hyperbolic metric concepts have great heuristic value, even if they can

theoretically be eliminated. We shall therefore assume the basic facts

of hyperbolic geometry, including the Poincare disc model, and develop only

the less familiar facts which are pertinent to surface theory. The best

sources for the facts we assume seem to be works on complex analysis, such

as Siegel [1971], and Zieschang [1981].

 374

Fact 1 (see Siegel [1971]). The hyperbolic plane can be modelled by

the open unit disc D C:IR? when "motions" are interpreted as conformal
(equivalently, circle-preserving) maps of R2 which map the unit circle,

3 D, onto itself. Hyperbolic straight lines are then circular arcs

in D orthogonal to 3D, hyperbolic angle equals euclidean angle,

and the motions are of three types corresponding to the following

families of euclidean circles in D (Fig. 9. (i), (ii), (iii»

Fig.9

(i) Circles orthogonal to the pencil of hyperbolic lines through a

point P E D are the hyperbolic circles with hyperbolic centre P.

A motion which leaves P fixed maps each circle with hyperbolic centre

P onto itself, permutes the hyperbolic lines through P, and is called

a rotation about P.

(ii) Circles orthogonal to the pencil of hyperbolic parallels

with limit point P E 3D are called lIyperbolic limit circles with

centre P. A motion which leaves P fixed maps each limit circle onto

itself, permutes the parallels with limit P, and is called a limit

rotation about P. Any two hyperbolic parallels approach each other

arbitrarily closely in the hyperbolic metric, and under a limit

rotation there are points which are moved through arbitrarily small
 375

hyperbolic distances.

(iii) Circles orthogonal to the hyperbolic perpendiculars to a

given line LCD are the curves at constant hyperbolic distance from L

(distance curves of L). A motion which fixes the end points of L

on dD maps each distance curve of L onto itself, permutes the

perpendiculars to L, and is called a translation, or displacement,

with axis L. Each point of L is sent a constant hyperbolic

distance, A, called the displacement length of the motion, and the

segment between any point of L and its image is called a displacement

segment. Each point on a distance curve C of L is also sent a

constant hyperbolic distance, but the constant is > A and increases

with the distance of C from L. o

Fact 2. Each non-identity motion g E nl(S) is a translation.

Proof. Since the motion g maps the Cayley diagram of nl(S) onto

itself non-trivially, it must move every point through at least the

diameter of a cell in the tessellation. Thus there is a non-zero

lower bound to the distance moved by each point under the motion g,

whence g is a translation by Fact 1. 0

As Dehn [1910] remarks, the non-existence of points on S which

move through arbitrarily small distances reflects the non-existence of

small curves on S which do not contract to a point. This is why

translations are important for topology, and we shall analyse them

in detail. Following Nielsen [1944], we call the fixed points of a

translation g its fundamental points; g moves points of D away

 376

from the negative fundamental point, called U(g), towards the positive

fundamental point, called V(g), along the "streamlines" which are the

axis of g, called A(g), and its distance curves.

The next fact, also observed by Dehn [1910], gives a way to

find U(g), V(g), and hence A(g), from the Cayley diagram.

-1 2
Fact 3. The vertices ..• , g , 1, g, g , ... of the Cayley diagram

lie on a distance curve through U(g), V(g). Moreover gn + V(g) and

g-n + U(g) (in the euclidean metric), as n + 00.

Proof. The motion g, by definition, maps each member of the

-1 2
sequence of vertices ..• , g 1, g, g , ... onto its successor. Hence

the vertices must lie along a streamline of the motion, i.e. along a

distance curve of A(g), by Fact 1. Since they are equally spaced

along the distance curve, by Fact 1, the hyperbolic distances of gn and

g-n from 1 tend to 00 as n + 00 , hence g-n + U(g) and gn + V(g) in

the euclidean metric. o

Fact 4. If f, g are any two translations, then fgf- l is a

translation of the same length as g, and

-1 -1 -1
A(fgf ) = A(g).f (the f -image of A(g))

-1 -1 -1
Proof. Consider the f -images, U(g)·f and V(g)·f , of U(g)

and V(g), and their images in turn under fgf- l

-1 -1 -1 -1
U(g)·f ·fgf U(g)·gf = U(g)·f

-1
since U(g) is fixed by g. Thus U(g)·f is a fixed point of
-1 -1 Hence these are the fundamental
fgf ,and similarly so is V(g)'i .
 377

-1 -1
points of and A(g)·f = A(fgf ). r'\oreover, a point

P E A(g) and its g-image p.g E A(g) are sent by f- l to

-1 -1 -1
and P·gf image of Pf under fgf ,hence

the translations on the two axes are of equal length. 0

The remaining facts are proved following Nielsen [1927).

Fact 5. In a group of translations, two non-commuting elements have

no common fundamental point.

Proof. Suppose on the contrary that f, g are non-commuting

elements with a common fundamental point. By replacing one of f, g

by its inverse, if necessary, we can assume that V(f) V(g). Now

-1
V(fgf- l ) = V(g) by Fact 4, thus fgf- l and g have parallel axes

(with limit V(fgf- l ) = V(g) = U(g-l» and displacement lengths which

are equal (by Fact 4) but opposite in direction. It then follows

from the continuity of motions that by choosing a point P on

A(fgf- l ) sufficiently close to A(g-l), the point P.fgf-l.g- l can

be made arbitrarily close to P. Thus the non-identity group

element fgf-lf- l moves points through arbitrarily small distances,

contrary to the properties of translations (Fact 1). 0

Fact 6. In a group of translations, two elements commute if and only

if they have the same axis.

Proof. If f, g have the same axis then

fg = [translation of lengtr. Af + Ag with axis A(f») = gf.

Conversely, . i f fg = gf then fgf- l = g, hence A(fgf -1 ) = A(g).
But A(fgf
-1
) = A(g).f -1 by Fact 4, hence f- l must map the
 378

hyperbolic line A(g) onto itself. This line is therefore the

axis of f- l , and hence of f, by Fact 1. o

For the last fact we use Nielsen's notation I for an

automorphism and gI for the I-image of g.

Fact 7. An automorphism of a group G of translations induces

a permutation of the fundamental points of G.

Proof. An automorphism sends ghg-lh- l to 1 if and only if

fg = gf. Since

I permutes the elements of G, it then follows from Fact 6 that I

induces a permutation of the axes of G. But no two distinct

axes have a common endpoint by Fact 5, hence the permutation of

axes is in fact a permutation of fundamental points. 0

 379

4. Axes, simple curves and intersections

If we attempt to generalise the analysis of the torus given in

section 2 to a surface S of higher genus, then several generalisations

of the canonical geodesic curve system suggest themselves. For reasons

which will appear later, it is good to take a system consisting of simple

closed geodesics with no multiple intersections, and such that cuts along

these geodesics render S simply connected. In fact, on the surface S

of genus p there are geodesics aI' a 2 , •.• , a 2p with the following

properties

(i) Each a i meets a i +1 at exactly one pOint, where they cross.

(ii) There are no other multiple points; in particular, each <Xi

is simple.

We shall see in Theorem 3 below that cutting along such curves ai renders

S simply connected, so this system of curves meets all our requirements.

Such a system was used by Dehn's student Baer, [1927], and a similar

one appears in Nielsen [1929, §4], though neither notes that the a. can be
~

realised by geodesics. Certainly it is clear that there are simple ~

a i with properties (i), (ii), e.g. as in Fig. 10, but to show that

Fig. 10

topologically realisable incidences of curves are also realised by

geodesics one needs the following ideas of Poincare [1904].

 380

The metric on S is inherited from the hyperbolic metric on

S, hence the geodesics on S are the curves covered by hyperbolic

straight lines on S, and intersections of geodesics lift to inter-

sections of hyperbolic lines. In turn, hyperbolic lines L1 , 12

intersect if and only if the end points of 11 on as separate

the endpoints of 12 , Poincar~ makes this trite remark topologically

potent by combining it with the observation that a deformation of S

lifts to a deformation of S through a bourtdedhyperbolicdistartce,

which therefore,deforms any line into a curve with the same endpoints

on as.

1emmas 1-5 below systematise Poincar~'s rather fragmentary appli-

cations of these remarks, and they suffice in particular to show the

existence of geodesics ~i with properties (i), (ii). In stating

the proofs it is convenient to extend the notation of section 3 to allow

closed curves on S in place of their homotopy classes, the elements of

11 1 (S) • Thus when a closed curve c on S is lifted to c on S

with initial point 1 we shall call the final point c. It is a

vertex of the Cayley diagram, and the translation which carries 1 to c

will be called the motion c, its axis A(c), and the closed curve on S

covered by a displacement segment 0 (c) of A(c) will be called ~ (c) •

(Of course, when we are discussing elements g E 111 (S) directly, the

old notation will still be used, and ~(g) will denote the curve on

S covered by a displacement segment of A(g).)

In what follows, "equivalent" sets on S are sets with the same

projection on S, i.e. translates of each other by motions in 111 (S).

 381

Lemma 1. The closed geodesics on S are precisely the curves ~(g),

Proof. A closed curve c on S lifts to an arc c on S with

endpoints equivalent under the motion c. If the curve c is a

geodesic then c is a hyperbolic line segment which is collinear with

its own translate under the motion c. That is, c is a displacement

segment of A(c) •

Conversely, a displacement segment 6(g) of A(g), g E TIl (S), has

a closed projection ~(g) on S, since the endpoints of 6(g) are

equivalent (under g). And every segment of a(g) is geodesic because

successive translates of 6(g) by g are collinear and equivalent

(It follows that ~ displacement segment of A(g) has the same projection

~(g), if we do not distinguish an initial point.) 0

In general, the curve on S covered by a hyperbolic line segment

with equivalent endpoints is a geodesic monogon with a corner where the

two endpoints project, rather than a closed geodesic. For example, the

curve covered by an edge labelled a1 in the Cayley diagram of TIl (S)

has a right angled corner at the basepoint 0 on S, since adjacent

edges labelled a 1 meet at right angles; thus segments of the curve which

contain 0 are not geodesic. This monogon does however extend to a

-1
closed "figure eight" geodesic covered by adjacent edges aI' a 2 ,since
-1
such edges are collinear. In particular, the segment labelled a 1a 2
-1
issuing from 1 on S is a displacement segment of A(a 1a 2 ).

(Cf. Figs. 7 and 8).

 382

Since successive edges in the Cayley diagram are not

collinear, the distance curve of A(a 1) through the vertices

is not a hyperbolic line, and hence the curve a 1 on S cannot be

deformed into a geodesic with 0 fixed. This leads us to consider

free homotopies on S, i.e. deformations without fixed basepoint, and

their lifts to S which are not fixed at the vertices of the Cayley

diagram.

Lemma 2 Any closed, non-contractible curve c on S is freely homotopic

to a unique closed geodesic, namely a(c).

Proof. Consider the lift c of c which runs betwen the vertices 1 and

c in the Cayley diagram. Since the curve c is non-contractible on S,

these vertices are distinct and, since they lie on a distance curve of

A(c), we can drop perpendiculars from them to the endpoints of a dis-

placement segment 6(c) of A(c). If we deform c into the distance

curve segment between land c, and then descend through the intervening

distance curve segments onto e(c) (Fig.ll), then the endpoints

Fig. 11

A(c)

remain equivalent throughout, and hence the whole deformation projects

to a free homotopy on S which converts c to a(c).

 383

The uniqueness of a(c) can be seen as follows. By Lemma 1,

any deformation of c to a closed geodesic lifts to a deformation

ct ' 0 ~ t ~ 1, from Co = c to c l = some o(g), in which the endpoints of

c t are equivalent for each t. Initially, when c t = c, the endpoints

are equivalent under the motion c, hence they must remain so (since ITl (5)

is a discontinuous group, e.g. by Fact 2). But the only axis which con-

tains points equivalent under the motion c is A(c), hence c 1 = o(c),

which projects to a(c) as required. 0

The next three lemmas exploit the separation properties of po~nts

on as and the fixture of these points under deformations on S.

Lemma 3. A curve c on 5 is freely homotopic to a simple curve

. . a(c) is simple.

Proof. (.) is immediate from Lemma 2. To prove (~), suppose that a(c)

is not simple. A self-intersection of a(c) is covered by an inter-

section of A(c) with one of its translates A(c)'g, g E TIl (5).

Case 1: A(c) crosses A(c).g. Then the endpoints

of A(c) separate those of A(c)·g on as. We know from Lemma 2 that there

is a deformation on S of a(c) to C, and this deformation lifts to a

deformation on 5 of A(c) to a curve Coo with the same endpoints on

as. The deformation similarly lifts to a deformation of A(c)·g to

Coo 'g, with fixed endpoints. The endpoints therefore continue to separate

each other on as, and hence Coo crosses coo·g (e.g., by the Jordan

curve theorem). This crossing covers a self intersection of c on S.

Case 2. A(c) = A(c)'g, where g necessarily has axis A(c)

although it is not a power of c, since a double point on c lifts to

points on 5 which are inequivalent under the motion c. The corresponding

 384

deformations then yield c 00 and with the same endpoints as

A(c). Since Coo is periodic under the translation c along A(c),

it has points at maximum distance to left and right of A(c), and

since coo.g is congruent to Coo by the translation g along A(c),

coo.g attains the same maximum distances to left and right of A(c).

It follows that coo.g cannot lie entirely to one side of coo' and

hence they meet. Since g is not a power of c, an intersection of

Coo and coo.g covers a double point of c. 0

Lemma 4. If c,d are disjoint simple curves on S, then a(c), a(d)

are either disjoint or identical.

Proof. Axes, and hence geodesics, cannot be tangential unless they

coincide, so if a(c). a(d) are neither disjoint nor identical they

must cross, and hence A(c) crosses A(d) on S. Thus the endpoints

of A(c) on as separate those of A(d). Deformations of a(c),

a(d) to c,d respectively lift to deformations of A(c), A(d) to

Coo ' d oo ,where Coo has the same endpoints as A(c) and d has the

same endpoints as A(d). Then Coo crosses doo ' and this crossing covers

a point on S where c meets d, which is a contradiction. 0

The reader may easily guess that a(c) = a(d) . . c, d are freely

homotopic, though we shall not need this result. It is worth pointing

out, however, that the criterion a(c) = a(d) then leads to an

algorithm for deciding when two given curves are freely homotopic

(Dehn [1910]), just as Lemma 3 leads to an algorithm for deciding

whether c is homotopic to a simple curve (Poincare [1904]).

 385

Lemma S. If c,d are simple curves which meet at a single point, where

they cross, then a(c), a(d) have the same property.

Proof. We know from Lemma 3 that a(c), a(d) are simple.

To establish the crossing property, let coo' doo denote the curves

on S which cover c,d and have the same endpoints as A(c), A(d).

The single crossing X of c by d lifts to a single crossing

X of c by d
n
together with a single crossing X·c of - c by
-
00 00 ' 00

n
each translate da: c of d 00
by a power of the motion c. The latter

correspond to the encounters with X which result from successive tra-

versals of c, and since c,d do not meet in any other way, Coo does

not meet any other translate da: g, where gE1fl (S). We want to show

that A(c) is similarly crossed - by A(d) and its translates A(d), c n ,

and no other translates A(d)·g - as this will imply a single crossing

of a(c) bya(d).

Since d is simple, all the translates -doo• c n are disjoint, hence

doo cannot lie along A(c) by the argument of Lemma 3, case 2. Thus

A(d) f A(c) and the two axes must cross, otherwise there could not be

a single crossing of Coo by doo • We therefore have a situation like

that shown in Fig. 12.

Fig. 12
 386

Translates of A(d) by the motion c then also cross A(d), but

translates by any other g E ITI (S) do not, since this would imply a

crossing of Coo by doo .g. 0

It follows easily from Lemmas 3-5 that the curves al, ••. ,a Zp

on S with the properties (i), (ii) stated at the beginning of

this section can be taken as geodesics. (The only worry is that

disjoint curves might have coincident geodesic representatives, e.g.

But this is ruled out because one of ai' a i +Z crosses

a curve which the other does not cross.) Any geodesics a l ,·.·, a Zp

with properties (i), (ii) will be said to constitute a canonical geodesic

system. The proofs of Lemmas 1-5 show that al, ..• ,a Zp can be represen-

ted as a(gl)"'" a(gzp) for some gl, .•• ,gzP E ITl (S) whose axes have

intersections which reflect the intersections of We combine

these results into the following.

Theorem 3. For the surface S of genus p there are elements gl, ••• ,gzP

E ITl (S) such that a(gl) ••..• a(gzp) is a canonical geodesic system on S.

This means

(i) For each crosses all translates

A(gi+l) by the motion gi' but no other translates A(gi+l).g where

(ii) Except for the trivial coincidence of A(gi) with its own

translates under the motions the different translates A(gi).g,

1 ~ i ~ Zp, g E ITl (S), have no other intersec~ions. o

 387

The above theorem does not appear explicitly in the literature. to

my knowledge. but it would surely have been clear to Poincar€. Since

Poincar€ [1904] also proves the existence of homeomorphisms mapping curve

systems with given intersections onto other systems with the same inter-

sections. it also seems reasonable to attribute the following theorem to

him.

Theorem 4. If and I
Cl l •···• Cl 2p
I
are canonical geodesic

systems on S. then there is a homeomorphism T : 5 ~ 5 such that

Cl.]. ' for each i.

Proof. It is easily seen by induction on i that cutting S along

Cl l •...• Cl 2i _l gives a connected surface. with two boundaries that are

joined by Cl 2i • and that cutting along Cl 2i gives a connected surface

with one boundary. An Euler characteristic calculation then shows that

when i = P the cut surface becomes simply connected. i.e. a polygon P.

The boundary OP of P is divided into 8p-4 segments identified in

pairs: one pair for each of Cl l .Cl 2p and two for each Cl i • 2 ~ i ~ 2p 1.

since each of the latter Cl i is subdivided at two points. by Cl i _1 and

Cl i + 1 • We label and orient these segments by the names of the curves from

which they originate. calling the pieces of Cl i • 2 ~ i ~ Zp - 1. Cl i l and

Cl iZ ' The latter pieces are determined by the orientation of Cl.]. if we

let Cl i l run from Cl i _1 to Cl i +1 and Cl i2 from Cli+l to Cl i _l

(See Fig. 14 below for the case p = 2.)

It can then also be checked by induction that the cyclic sequence of

labels on OP is unique. hence the polygon P can be mapped onto the

polygon pI which results from cutting S along the Cl ].. I in such a way
 388

that each segment in ap is mapped onto the corresponding primed

segment in ap'. Adjusting the map by a deformation i f necessary.

we can arrange that equivalent point pairs on ap are sent to

equivalent point pairs on ap'. so that we have a homeomorphism

Informally speaking. both P and p' can be pasted together to

"look like" the standard picture of Fig. 10. This shows that there

are homeomorphisms of 5 which send both the a i and ai' systems

to standard position. and hence there is a homeomorphism of one system

onto the other.

Let us now compare our situation to where we were with the torus

after Theorem 1. Then it was clear. because the basepoint for ~1 (T)

could be taken at the intersection of the two canonical curves a,b.

that an automorphism I of ~ 1 (T) was "given" by the images a'. b'

of a.b. and it only remained to show that the I-images of a.b were

canonical curves for any I.

Now, in avoiding multiple intersections. we seem to have lost a

basepoint for ~1 (5). since there is no point common to all of

a 1 ••••• a Zp • However. since the a i cut 5 down to a simply connected

piece. it turns out that an automorphism I of ~1 (5) is in fact

"given" by its effect on the a i • The main problem is showing that I

sends one canonical geodesic system to another. when we know nothing

about the automorphisms I.

This is where the avoidance of multiple intersections becomes

important. We only have to show that I preserves the single intersection

 389

for each i, and the non-intersection of u. with

~

Uj ' j f i - 1, i + 1, which in turn reduces to showing that I preserves

intersections and non-intersections of axes. Dehn proved this result

very elegantly by looking at the behaviour of TIl (8) at infinity. His

idea is developed in the next section.

 390

5. Mappings of as induced by automorphisms of ~li§l.

As just mentioned above, the key step towards the Dehn-Nielsen theorem is

to prove that an automorphism I of ~1 (5) preserves intersections and

non-intersections of axes. We let I act on the set of axes by sending

A(g) to A(gI) for each g E ~1 (5). Since A(g) is determined by

U(g) and V(g), it is equivalent to let I act on the set of fundamental

points by sending U(g) to U(gI) and V(g) to V(gI)' We know that this

map is well-defined (in fact a bijection) by Fact 7, and it is

appropriate to view the action of I on axes as really taking place on

as. In S itself, I naturally acts on the vertices of the Cayley

diagram by sending vertex h to vertex hI' and this can only be inter-

preted as an action on axes by going to infinity (letting h±n + U(h),V(h»

as in Fact 3, we get h~n + U(h I ), V(h I ».

It was Dehn's idea to investigate the action of I on axes by

approximating axes by edge paths in the Cayley diagram. One then needs

nothing but obvious properties of automorphisms in order to conclude

that the map vertex h t+ vertex hI is a "quasi-isometry" of the

Cayley diagram, and hence that nothing drastic can happen at as, where

the axes and their intersections are determined.

The approximation of axes by edge paths is implicit in Dehn's

[19l2a,bl solution of the conjugacy problem (see Nielsen [1927, §7(g) 1),

but Dehn's analysis of automorphisms seems to be lost, and the proof

which follows is extracted from Nielsen [1927, §91, where credit is given

to Dehn for the essential ideas.

 391

Theorem 5 (Dehn). U(g), V(g) separate U(h), V(h) on as

Proof. The first step of the proof is to approach U(h), V(h)

by a doubly infinite edge path p(h) through the vertices

hn , n = 0, ± 1, ± 2, ... We know from Fact 1 that these vertices

lie on a distance curve of A(h). The segments of p(h) between

them can be chosen with identical sequences of edge labels (spelling

out a word for h), and then p(h) lies within a bounded hyper-

bolic distance of A(h). We can similarly approach U(g), V(g)

by polygonal paths through the vertex sequences {g-nln ~ N} and

{gnln ~ N}, for any positive integer N. Assuming U(g), V(g),

U(h), V(h) are distinct, Fact 3 says we can choose N so that the

latter paths are arbitrarily far away from peg), and then if

U(g), V(g) separate U(h) , V(h) we can also join their ends

g-N, gN by a path through successive vertices g-N

-N N
which are all at least as far from p(h) as g , g themselves.

The resulting path peg) is then far from p(h).

We can also interpret "far" in terms of the word metric on the Cayley

diagram, which is the minimum number of edges connecting given vertices.

Vertices "far apart" in the hyperbolic metric are also "far apart" in the

word metric, since their hyperbolic distance is ~ (number of edges x maximum

length of an edge), and conversely, since only finitely many vertices lie

within a given hyperbolic radius.

 392

To simplify notation we denote the successive vertices of

peg) by ... , g-l' go' gl' gz"" and the successive vertices

of p(h) by ... , h-l,hO' hI' h 2 ,.... Fig. 13 shows how these

paths might look.

Fig.13
We now show that there are similar disjoint edge paths, p(h I )

between U(h I ) and V(h r ) , and p(gr) between U(gr) and V(gr)' by

showing that r cannot send vertices which are far apart to vertices

which are close together. (This will show that U(gr)' V(gr)

separate U(h r ), V(h r ) if U(g), V(g) separate U(h), V(h), and

the converse follows by considering r- l .)

rn fact, letting d(u,v) ~ length (u-lv) denote the minimum number

of edges in a path between the vertices u, v in the Cayley diagram

(the word metric) we shall show that there are constants k, K > 0

such that
 393

(1)

(2)

This says that 1 is a "quasi-isometry", while (2) in particular says

that for any distance d > 0 in the word metric there is a D >0

such that vertices > D apart are sent to vertices> d apart.

-1
To prove (1) , let w be a word of minimal length for u v. Then

d(u,v) = length (w)

while d(ur,v r ) = length

= length

~ K length (w) = Kd(u,v)

where K = max length {(al)r, ••• ,(bp)r}' since a word for wr is

obtained by replacing each generator ai by (ai)r and bi by (bi)r.

The inequality (2) now follows from

-1
which is proved by applying the same argument to r ,and taking

k = 11K'.

Now to obtain a path p(gr) through the vertices (gi)r' and a

disjoint path through the vertices (hi)r' we first notice, from (1),

that each vertex (gi)r can be connected to (gi+l)r by a path of

length ~ K, since gi+l is one edge away from gi by definition.

We construct p(gr)' and similarly p(h r ), as the union of such subpaths.

Then to ensure that p(gr) does not meet p(h r ) it is enough to

ensure that each (gi)r is distance ~ 2K + 1 from each (hj)r. By (2),

 394

this can be achieved if each gi is distance ~ D from each hi' for

a suitable D, and D in turn can be achieved by choosing a sufficiently

large N in the construction of p(g). 0

The Dehn-Nielsen theorem now follows from Theorems 3, 4, 5.

Theorem 6. (Dehn-Nielsen) For any automorphism I ~l (5) + ~l (5) there is a

homeomorphism T : 5 + 5 which induces I.

Proof. Theorem 3 gives canonical geodesics

specifies intersections of the axes A(gi),g, g E ~l (S), on S which

cover them. The axes A«gi)I.gI' gI E ~l (S), which cover the geodesics

(ai)I = a«gi)I)' are seen to have the same intersections, by Theorem 5,

when one notes that they result from the A(gi).g by the I action.

Namely A(gi)·g = A(g -1 gi g) by Fact 4, and

~ ~
A«g gi g)I) = A«gI) (gi)I gr) = A«gi)I)·gI·

Thus the (ai)I also form a canonical geodesic system on S, and there

is a homeomorphism 1

In fact, the proof of Theorem 4 tells us that the polygons P'P I

which result from cutting S along the a i and (ai)I systems respectively

have isomorphically labelled boundaries, hence the generalised Cayley

diagrams 0, Vr which result from lifting the a i and (ai)r systems

respectively to S are isomorphic. Collinear edges of 0 unite to form

the axes A(gi).g, g E TIl (S), and collinear edges of VI unite to form

the axes of A«gi)I).gr' gr E TIl (S).

Fig. 14 shows the surface of genus 2 being cut along the ai' and Fig.15
shows the cell of the Cayley diagram V which results.

Fig. 14.
 395

Fig. 15

If we let the identity vertex of V be 1:.. = A(gl) nA(g2) and let the

identity vertex of Vr be 1:..r = A«gl)r) n A«g2)r)' then the isomorphism

V-+Vr means that i f an axis A(h) is reached by following a certain sequence

of ~i' ~i1 or ~i2 edges from 1:.. in V, then A(h r ) is reached by following
the corresponding sequence of edges in Vr from 1:..r . But the vertex of V
representing g E TIl (5) is just
-1 -1
A(gl)·g n A(g2)·g = A(g glg) n A(g g2 g)

and hence g is determined by a common path to these two axes. The corres-

ponding path from 1:..r in Dr leads to

-1 -1 -1 -1
A«g gl g )r) n A«g g2g)r) = A«gr) (gl)r g r) n A«gr) (g2)r g r)
= A«gl)r) ·gr n A«g2)r) ·gr

which is the vertex gr of Vr.

rn other words, a closed path from ~1 n ~2 on 5 representing

g E TIl (5) is mapped by T onto a closed path from T(~l n ~2) =

(~l) I n (~2) I on T (5) representing gI E TIl (5). Thus T induces the auto-

morphism I: TI l (5)-+TI l (5). 0

 396

References

Baer, R. [1927] Kurventypen auf Flachen.

J. f. Reine und angew. Math. 156, 231-246.

Dehn, M. [1910] : Papers 1 and 2, this volume.

[1912a] : Paper 4, this volume.

[1912b]: Paper 5, this volume.

[1922] : Paper 7, this volume.

Dehn, M. & Heegaard, P. [1907] : Analysis situs,

Encykl. Math. Wiss., vol. III AB3, Leipzig, 153-220.

Floyd, W. [1980] Group completions and limit sets of kleinian groups.

Inv. Math. 57 (1980), 205-218.

Nielsen, J. [1927, 1929, 1932] : Untersuchungen zur Topologie der

geschlossenen zweiseitigen Flachen I, II, III. Acta

math. 50, 189-358; 53, 1-76; 58, 87-167.

[1944] Surface transformation classes of algebraically finite

type. Det. Kgl. Danske Vidensk. Selskab, Math-fys.

Medd. XXI, 1-89.

Poincare, H. [1904] Cinquieme complement a l'analysis situs.

Rend. circ. mat. Palermo 18, 45-110.

[1985] Papers on Fuchsian Functions. Springer-Verlag.

Seifert, H. [1937] Bemerkungen liber stetigen Abbildungen von Flachen.

Abh. math. Sem. Univ. Hamburg 12, 23-37.

Siegel, C.L. [1971] Topics in Complex Function Theory.

Vol. 2, Wiley, New York.

Zieschang, H. [1981] Finite Groups of Mapping Classes of Surfaces.

Springer-Verlag.